IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING_Presentations... · ieee transactions on geoscience and remote sensing january 1986 volume ge-24 number 1 (issn 0196-2892) a publication

, ,'.+ IEEE TRANSACTIONS ON

GEOSCIENCE ANDREMOTE SENSINGJANUARY 1986 VOLUME GE-24 NUMBER 1 (ISSN 0196-2892)

A PUBLICATION OF THE IEEE GEOSCIENCE AND REMOTE SENSING SOCIETY

SPECIAL ISSUE ON AGRICULTURE AND RESOURCES INVENTORY SURVEYS THROUGH AEROSPACEREMOTE SENSING (AgRISTARS)

#;"'"(~) IEEE GEOSCIENCE AND REMOTE SENSING SOCIETY~

The Geoscience and Remote Sensing Society is an organization, within the framework of the IEEE, of members with principal professional interest Ingeoscience and remote sensing. All members of the IEEE are eligible for membership in the Society and will receive this TRANSACTIONS upon payment of theannual Society membership fee of $7.00. For information on joining, write to the IEEE at the address below.

ADMINISTRATIVE COMMITTEE

D. A. LANDGREBE, President

R. K. RANEY, Vice President J. A. REAGAN, Secretary-Treasurer

1986

K. R.CARVERJ. ECKERMA'\iA J. SIEBER

C. T. SWIF1K TOMIYASli

F. BECKERP. GLDMANDSENW. KEYDEL

1987

D. A. LANDGREBEF. 1. ULABY

R. BERNSTEINA. BLANCHARDW. CHEW

1988

V. KAUPPR. K. RANEY

Standing Committee Chairmen

C. A. BAI.ASISA. J. BL."""CHARDE. M. DIDWAII

J. ECKERMANV. H. KAUPPD. A. LANDGREBE

R. E. MciNTOSHK. TOMIYASUF. 1. UL.ABY

W. T. WAL.TONE. A. WOL.FF

IEEE TRANSACTlONS~ ON GEOSCIENCE AND REMOTE SENSING

Editor

FAWWAI T. Ul.ABYThe University of Michigan Radiation LaboratoryDepartment of Electrical and Computer Engineering4072 Easl Engineering BuildingAnn Arbor, MI 48109

Associate Editors

M. T. CHAHINE, Atmospheric Remote SemingW.-c. CHEW, /:;Iectromagnetic Subsurface Remote SemingR. E. MURPHY, Extraterrestrial Geophrsics and Remote SensingL. S. W ALTER, Geodl'namicsJ. M. MC:-.IDEL, Geophysical Signal Processing

C. H. CHE"", Information ProcessingE. NJOKU, Microwace Remote SensingR. K. RANEYJ. A. SMITH, Visible and Infrared Remote SensingP. N. SLATER

BRliNO O. WEI1'.SCllEI., PresidentHENRY L BACHMA1'., President-ElectEMFRSO,," W. PUGH. Execl1tiee Vice PresidentEDW"RD J. DOll.F. TreasurerMICHIYLKI tl'""OHARA. ,';ecrerary

THOMAS W. BARTLETT, ControllerDONALD CHRISTIANSEN, Editor, IEEE SpectrumIRVING ENGELSON, Staff Director. Technical Actil'itiesLEO FANN lNG, Staff Director. Professional ActieitiesSA VA SHERR, Staff Director. Standards

THE INSTITUTE OF ELE(TRICAL AND ELECTRONICS ENGINEERS, INC.

OfficersCYRIL J. TLNIS, Vice President, Educational ActidtiesCARLETON A. BAYLESS. Vice President. Professional ActieitiesCHARLES H. HOUSE, Vice President. Publication ActivitiesDENNIS BODSON, Vice President. Regional ActieitiesMERL.I:-I G. SMITH, Vice President, Technical Ac/id/ies

N. REX DIXON, Director. Dil'ision 1)(~Signals and Applicatiom

Headquarters StaffERIC HERZ, Executice Director and General Manager

EL.WOOD K. GANNETT, Deputy General ManagerDAVID L. STAIGER, Staff Director, Publishing ServicesCHARLES F. STEWART, JR., Staff Director. Administration ServicesDONALD L SUPPERS, Staff Director. Field ServicesTHOMAS C. WHITE, Staff Director. Public InformationJOHN F. WIUlELM, Staff Director. Educational Services

Publications DepartmentProduction Managers: ANN H. BURGMEYER, GAIL. S. FERENC, CAROLYNE TAMNEY

Associate Editor: RICHARD C. DODENHOFF, II

IEEE TRANSACTIONS 0"" GEOSCIENCE AND REMOTE SENSING is published bimonthly by The Institute of Electrical and Electronics Engineers. Inc.Headquarters: 345 East 47 Street. New York, NY 10017. Responsibility for the contents rests upon the authors and not upon the IEEE. the Society, or itsmembers. IEEE Senice Center (for orders, subscriptions, address changes, Region/Section/Student Services): 445 Hoes Lane, Piscataway, NJ 08854.Telephones: Headquarters 2 I 2-705 + extension: Information-7900, General Manager- 791 0, Controller-7748, Educational Services-7860, Publishing Ser-vices-7560, Standards-7960, Technical Services-7890. IEEE Service Center 201-981-0060. Professional Services: Washington Office 202-785-0017. NYTelecopier: 212-752-4929. Telex: 236-41 I (International messages only). Individual copies: IEEE members $6.00 (first copy only), nonmembers $1 2.00 percopy. Annual subscription price: IEEE members. dues plus Society fee. Price for nonmembers on request. Available in microfiche and microfilm. Copyrightand Reprint Permissions: Abstracting is permitted with credit to the source. Libraries are permitted to photocopy beyond the limits of U.S. Copyright law forprivate use of patrons: (I) those post- I 977 articles that carry a code at the bottom of the first page, provided the per-copy fee indicated in the code is paidthrough the Copyright Clearance Center. 29 Congress Street. Salem, MA 01970; (2) pre-1978 articles without fee. Instructors are permitted to photocopyisolated articles for noncommercial use without fee. For other copying, reprint or republication permission, write to Director, Publishing Services at IEEEHeadquarters. All rights reserved. Copyright © 1986 by The Institute of Electrical and Electronics Engineers, Inc. Printed in U.S.A. Second-class postagepaid at New York, NY and at additional mailing offices. Postmaster: Send address changes to IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTESENSING, 445 Hoes Lane, Piscataway, NJ 08854.

+ IEEE TRANSACTIONS ON

GEOSCIENCE ANDREMOTE SENSINGJANUARY 1986 VOLUME GE-24 NUMBER 1 (ISSN 0196-2892)

A PUBLICATION OF THE IEEE GEOSCIENCE AND REMOTE SENSING SOCIETY

SPECIAL ISSUE ON AGRICULTURE AND RESOURCES INVENTORY SURVEYS THROUGH AEROSPACEREMOTE SENSING (AgRIST ARS)

A Tribute to the Late R. Jeffrey Lytle F. T. Ulaby 2

Foreword H. C. Hogg 3

PAPERS

Hydrologic Research Before and After AgRIST ARS E. T. Engman 5Passive Microwave Soil Moisture Research T. J. Schmugge, P. E. O'Neill, and J. R. Wang 12Active Microwave Soil Moisture Research M. C. Dobson and F. T. Ulaby 23Soil Water Modeling and Remote Sensing T. J. Jackson 37Progress in Snow Hydrology Remote-Sensing Research A. Rango 47Early Warning and Crop Condition Assessment Research G. O. Boatwright and V. S. Whitehead 54Field Spectroscopy of Agricultural Crops M. E. Bauer, C. S. T. Daughtry, L. L. Biehl, E. T. Kanemasu, and F. G. Hall 65Light Interception and Leaf Area Estimates from Measurements of Grass Canopy Reflectance .

· G. Asrar, E. T. Kanemasu, G. P. Miller, R. L. Weiser 76Spectral Components Analysis: A Bridge Between Spectral Observations and Agrometeorological Crop Models .

· C. L. Weigand, A. J. Richardson, and P. R. Nixon 83Development of Agrometeorological Crop Model Inputs from Remotely Sensed Information .

C. L. Weigand, A. J. Richardson, R. D. Jackson, P. J. Pinter, Jr., J. K. Aase, D. E. Smika, L. F. Lautenschlager, andJ. E. McMurtrey, III 90

Detection and Evaluation of Plant Stresses for Crop Management Decisions .· R. D. Jackson, P. J. Pinter, Jr., R. J. Reginato, and S. B. Idso 99

Vegetation Assessment Using a Combination of Visible, Near-IR, and Thermal-IR A VHRR Data .· v. S. Whitehead, W. R. Johnson, and J. A. Boatright 107

Analysis of Forest Structure Using Thematic Mapper Data .· D. L. Peterson, W. E. Westman, N. L. Stephenson, V. G. Ambrosia, J. A. Brass, and M. A. Spanner 113

A Correlation and Regression Analysis of Percent Canopy Closure versus TMS Spectral Response for Selected ForestSites in the San Juan National Forest, Colorado M. K. Butera 122

Use of Remotely Sensed Data for Assessing Forest Stand Conditions in the Eastern United States .· D. L. Williams and R. F. Nelson 130

Coniferous Forest Classification and Inventory Using Landsat and Digital Terrain Data .· J. Franklin, T. L. Logan, C. E. Woodcock, and A. H. Strahler 139

On the Design of Classifiers for Crop Inventories R. P. Heydorn and H. C. Takacs 150Crop Acreage Estimation Using a Landsat-Based Estimator as an Auxiliary Variable .

· R. S. Chhikara, J. C. Lundgren, and A. G. Houston 157A Review of Three Discrete Multivariate Analysis Techniques Used in Assessing the Accuracy of Remotely Sensed

Data from Error Matrices R. G. Congalton and R. A. Mead 169Landsat Large-Area Estimates for Land Cover G. A. May, M. L. Holko, and N. Jones, Jr. 175

IEEE COPYRIGHT FORM 185

2IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. GE-24 NO. I, JANUARY 1'186

A Tribute to the Late R. Jeffrey Lytle

(February 10, 1941-September 4, 1985)

This issue is dedicated to the memory of R. Jeffrey Ly-tle, whose accomplishments were so widely known andrespected in the scientific community and who so ablyserved on the Administrative Committee of the IEEEGeoscience and Remote Sensing Society; we all share inthe loss of our esteemed colleague and friend.

Jeff was born in Columbus, Ohio, and passed away inWalnut Creek, California. He fought cancer valiantly and,true to his nature, was optimistic to the very last.

Jeff attended Purdue University and received his B.S.,M.S., and Ph.D. degrees in electrical engineering. Mostof his professional career was spent at the Lawrence Liv-ermore National Laboratory where he was Group Leaderof the Electromagnetic and Acoustic Sensing Group in theEngineering Research Division. As Group Leader, he en-couraged his team members to be innovative and bold intheir approach to research. Jeff was instrumental in de-veloping very advanced electromagnetic applications to

geophysical investigation. One technique developed by Jeff,geotomography, earned him a world-wide reputation.

Prior to his death, he was elected Fellow to the Instituteof Electrical and Electronic Engineers. Jeff was a Regis-tered Geophysicist in the state of California, an activemember in the Society of Exploration Geophysicists, andserved as Vice President of the IEEE Geoscience and Re-mote Sensing Society.

Jeff always had time to listen to new ideas, and his en-thusiasm for developing novel methods was infectious. Hewas a very trustworthy and honest person in all his inter-actions. A compassionate man, he was very active withyoung people, involved in church activities, and familyoriented. Jeff will be sorely missed by many of his closefriends and associates. Jeff is survived by his wife,Glenda, and three children, Ivan, Janette, and Bobby.

FAwwAz T. ULABYEditor

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING. VOL. GE-24 NO. I. JANUARY 1986 3

Foreword

SIGNIFICANT DATES LEADING UP TO AGRISTARS

OBJECTIVES AND PROGRAM STRUCTURE

truth cases; measure soil moisture and snowpack proper-ties, model snowmelt runoff, provide information on re-source status and condition; and to conduct forest inven-tories and condition assessments.

1) early warning and crop condition assessment;2) inventory technology development-originally for-

eign commodity production forecasting;3) yield model development;4) soil moisture;5) domestic crops and land cover;6) renewable resources inventory (forestry);7) conservation and pollution; and8) supporting research.

Development of multispectral scanners.Establishment of organized remote sensingresearch program in scientific communityto explore applications for agriculture.Development of first digital processingsystem to analyze CCT's from airbornescanners.First computer-aided classification ofwheat using airborne multispectral scannerdata and digital analysis system.Definition of spectral bands, etc. for firstearth resources satellite to be launched in1972.Apollo Multiband Camera Experiment.Corn Blight Watch Experiment.Landsat I launched.Joint Canadian study and initial LACIEproposal.LACIE tri-agency project officially autho-rized.LAClE ended.AgRISTARS program initiated.

1975

19781980

1969197119721973

1966

1967

1966

The AgRISTARS program was organized into eightprojects each with its own set of objectives, funding, andmanagement. The projects typically were staffed by morethan one participating agency and in most cases used acommon data system. Periodic progress reviews, con-ducted by program management, cut across all projects.The overall goal of the program was to determine the fea-sibility of integrating aerospace remote-sensing technol-ogy into existing and future remote-sensing systems. Theoverall approach was a balanced program of research, de-velopment. testing, and evaluation of techniques to im-prove information for USDA program needs. The projectsincluded:

Early 1960's1965

IEEE Log Number 8406233.

BACKGROUND

EARLY JOINT RESEARCH by the United States De-partment of Agriculture (USDA) and the National

Aeronautics Space Administration (NASA) pioneered theuse of remotely sensed multispectral data for agriculturalapplications. Initially these data were provided through anaircraft-based research program and multispectral cam-eras on manned spacecraft which were used to assess theutility of low-spatial-resolution imagery from space. In1970, the severe infestation of Southern Corn Blight in theU.S. Corn Belt provided the opportunity for the first real-time large-scale experiment involving multispectral anal-ysis and machine processing; the Corn Blight Watch Ex-periment conducted jointly by NASA and the USDA in1971.

In 1972, massive grain purchases in the U.S. disruptedU.S. commodity markets which, together with the highlymarginal (and apparently worsening) global food supplyof the early 1970's, created a strong perceived demand forbetter information on global crop production. This led tothe formation of the multi-agency Large Area Crop Inven-tory Experiment (LACIE) to evaluate Landsat data for thispurpose. The LACIE baseline approach used Landsat datato estimate crop acreage and meteorologically driven yieldmodels to estimate crop yield. Landsat data acquisitionand analysis was conducted by NASA, yield estimation bythe U. S. Department of Commerce (USDC), and the sam-pling and aggregation, which combined the acreage andyield estimates, was conducted jointly by NASA and theUSDA, as was the accuracy assessment process. Simul-taneously with LAC IE , the USDA Statistical ReportingService (SRS), which is responsible for domestic crop es-timates, developed an analysis approach that used Landsatdat.a to refine the accuracy and spatial detail of existingestImates based on ground data acquired by the SRS in itsannual June Enumerative Survey. In addition, a number ofresearch projects within NASA, the USDA, and their as-sociated university communities investigated applicationsof Landsat data to a wider variety of problems, includingforestry, range management, and agricultural hydrology.

Building on this research base, Secretary of AgricultureBergland initiated discussions with the USDC, the U.S.Department of Interior (USDI), and NASA in September1977 that led to the establishment of the AgRISTARS pro-gram in FY 1980; the so-called Secretary's Initiative iden-tified eight priority areas (projects) where remote sensingoffered potential for improved information. Under Ag-RISTARS, commodity forecasting research was expandedfrom the wheat emphasis of LACIE to include all majorgrains. The AgRISTARS program also funded research inthe utilization of remote sensing data to: produce crop andland use statistics in both the with and without ground

4 IEEE TRANSACTIONS ON (;EOSClENCE AND REMOTE SENSING, VOL. GE-24 1\0. I, JANUARY 1986

Participating agencies included: the USDA, NASA, theUSDC, the USDI, and the U.S. Agency for Internationa]Development (U SAI D), wh ich participated as an ex-officioobserver.

AgRISTARS rapidly evolved into an "umbrella" pro-gram tl)r remote-sensing research in renewable resourcesaccounting for about 75 percent of federally funded re-search in this area during the 1980-]983 period. The pro-gram included optical as well as passive and active micro-wave systems, and conducted research with field, aircraft,and space data sources.

RESUlTS

The early warning/crop condition assessment researchprovided empirical models fi)r detecting and estimating theyield impacts of stress conditions associated with mois-ture, flooding, insect damage, winterkill, and hot drywinds. The project also pioneered the use of the NOAA-7 AVHRR data for large-area assessments. Important re-sults obtained in crop identification/area estimation re-search included the separation of corn and soybeans andsmall grains as a class, greatly improving processing speedand cost by ful] automation of classification procedures.and further development of data compression techniquesto reduce data volume. In addition, significant basic rc-search was conducted in quantifying soil background ef-fects, the use of active microwave systems in crop iden-tification, and estimating the leaf area index from spectraldata f{)r input into process level productivity models. Re-suIts in domestic crops and land cover statistics researchincluded the development, testing, and evaluation of op_erational procedures fllr estimating crop acreages overlarge areas. These estimates were made for major cropsin seven states and are provided annually to the USDACrop Reporting Board for inclusion in their otIlcia] esti-mates. Work was initiated in the highly diversified irri-

gated crop lands of California with encouraging results.Full land-cover surveys were conducted in Kansas, Mis-souri, and Arkansas-the first applications of Landsat dataon this scale. Yield modeling research developed a seriesof new empirical models suitable for use with remote-sensing data (corn, soybeans, wheat, and barley). Plantprocess/simulation models were developed and/or testedfor wheat and barley. Conservation/pollution research de-veloped techniques li)r measuring snow pack properties toestimate water content, modeled snowmelt runoff for U.S.river basins, found an extreme sensitivity of runoff modelsto assumed soil moisture levels, and developed proceduresfllr monitoring high sediment loads in reservoirs and riv-ers.

The papers included in this issue do not provide com-prehensive coverage of program accomplishments a]-though all of the major research areas are represented.The interested reader should obtain a copy of one of theAgRISTARS Annual Reports which includes a full listingof all program publications and reports. A single referencethat provides a reasonable coverage of program high-lights. in abstract and sh0l1 paper format, is the Proceed-ings of the 1985 International Geoscience and RemoteSensing Symposium, held at Amherst, Massachusetts. AnAgRISTARS Annual Report can be obtained from the Re-mote Sensing Branch, Statistical Reporting Service. U.S.Department of Agriculture. Washington. DC 20250.

HOWARD C. Hoc;c;

Guest Editor

RLI+.RI'NCLS

III c. E, Caudill and R, E, Hatch. "Overview of the AgRISTARS pro-gram." plenary paper. presented at the Int. Cieosci, Renlllte SensingSymp" Amherst. MA. Oct. 7-9. 1985,

121 H. c. lIogg ;l11d 'vI. Triehel. "Utility of Landsat dala in llIeeting USDAinl(>rrnation requirements." Stair Rep" NASA Headquarters. Jan, 1984,

Howard C. Hogg received the B.S. and M.S. degrees in agricultural economics fromOregon State University in 1958 and 1959. respectively, and the Ph, D. degree in resourceeconomics from the University of Hawaii in ]965.

He is eurrent]y a Visiting Fellow at the World Resources Institute, Washington, DC.Prior to joining the staff at WRI. he spent 20 years with the Federal Government, firstin the U.S. Department of Agriculture (1965-1980) and then in the National Aeronauticsand Space Administration (1980-1985). His last assignment at NASA was DisciplineChief f()J' the AgRISTARS Program, Throughout his Federal career. which included as-signments in several foreign countries. he specialized in natural resource problems andissues including the development of large-scale data and analytical systems. He has pub-lished about 50 articles in journals. conference proceedings. and agency reports.

Dr. Hogg is a member of the American Agricultural Economics Association and theInternational Association of Agricultural Economists.

-----.--------

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING. VOL. GE-24 NO. 1. JANUARY 1986

Hydrologic Research Before and After AgRISTARSEDWIN T. ENGMAN

5

Abstmct-Hydrologic research prior to AgRISTARS had followed arather defined path in which the knowledge curve with time resemblesa staircase, rather than a constant incline. AgRISTARS did not intro-duce remote sensing to hydrology. Some aspects of land-cover analysis,snow area, and floodplain delineation were being studied. However, theconcentrated effort of remote-sensing applications to hydrology did helpadd another step to the knowledge curve.

I. INTRODUCTION

THIS PAPER has two objectives. First, I will try todescribe the status of hydrologic research and knowl-

edge when the AgRISTARS program was begun. The threepapers that follow this one will highlight some of the workthat has been accomplished in modeling, snow hydrology,and water quality. In essence, the first part of this paperis the yardstick by which to evaluate progress in Ag-RISTARS. The second goal of this paper is to projectwhere future steps in the knowledge curve may lead us.Remote sensing is a fairly new tool in hydrology. I willtry to convince the reader that this is not simply a fad butwill have a major role in future hydrologic applications.

II. BRIEF HISTORY

Hydrology's historical roots are based on ancient ob-servations and man's attempts to manage water for sur-vival. Chow [1] has broken the history of hydrology downinto eight epochs. This is in concert with most other de-scriptions of man's increase in knowledge-typically anexponential curve. In the Period of Speculation (ancient-1400), Plato, Homer, and Aristotle recognized some formof hydrologic cycle. Although these early philosophers andscientists did not have a quantitative understanding of hy-drology, a great number of practical hydraulic structures,such as aqueducts and irrigation systems, illustrated man'sdesire and need to control water resources as a prerequi-site for civilization as we know it. During the Renais-sance, in the Period of Observation (1400-1600), Palissyand Leonardo da Vinci described a hydrologic cycle inwhich water moved from the oceans to rain on the landand returned to the oceans. Quantitative hydrology prob-ably began during the Period of Measurement (1600-1700)in which scientists such as Perrault, Mariotte. and Halleymade measurements of different hydrologic components.It is interesting to note that Perrault and Mariotte werephysicists and Halley an astronomer (no, not an astrolo-ger!). The Period of Experimentation (1700-1800) gave us

Manuscript received April 26, 1985; revised June 25. 1985.The author is with the U.S. Department of Agriculture, Agricultural Re-

search Service, Beltsville, MD 20705.IEEE Log Number 8406232.

a long list of familiar names that includes Pitot, Bernoulli,0' Alembert, and Chezy with discoveries that bear theirnames. According to Chow [I], the nineteenth century,which he called the Period of Modernization (1800-1900),saw the establishment of the science of hydrology as wenow know it. Many significant advances to hydrology wereaccomplished in this century, especially in the areas ofgroundwater and surface water. The foray into quantita-tive hydrology was extended in the Period of Empiricism(1900-1930). A lack of good scientific understanding ofhydrology led to a large number of empirical formulas de-veloped to solve site-specific problems. It is interestingthat most references associated with this epoch are insti-tutions (Bureau of Reclamation, Miami Conservancy Dis-trict, etc.) rather than people. This is not the case for thePeriod of Rationalization (1930-1950). Here we find thetrue gurus of modern hydrology. People like Sherman,Horton, Theis, Gumble, Hazen, Bernad, and Einsteinpublished their research and developed procedures that arestill very much in use today.

Chow's last epoch is the Period of Theorization (1950to date, e.g., 1964). Here hydrologists attempted to usetheoretical approaches to solve hydrologic problems. Thiswork provided a great deal of insight into the complexitiesof hydrology but did not do a great deal for the practicinghydrologist.

Were this list to be updated to 1985, several additionalperiods could be added-the period of the computer, theperiod of multidisciplinary research, the period of systemsanalysis, the period of environmental quality, the periodof modeling, the period of stochasticism versus determin-ism, etc. I think we could also add a period of remote-sensing applications to this list.

The interesting thing to notice about the post-1964epochs is that they generally widen the breadth of hydrol-ogy as a science. Environmental quality research may bean example of this. However, other areas, such as the pe-riod of the computer, certainly have also increased thedepth of our knowledge. The period of remote sensing alsohas increased the depth of our knowledge.

Ill. EFFECTIVENESS OF NEW METHODS

One interesting aspect of recent hydrologic research isthat although we feel we know more about the physicalprocess, use sophisticated analysis techniques, and canproduce very elaborate output, we have not been able todemonstrate consistently improved accuracy or reprodu-cibility.

The volume of published techniques that are availablefor use (e.g., the rational formula and SCS technique) is

U.S. Government work not protected by U.S. copyright

6IEEE TRANSACTIONS OK GEOSCIENCE AND REMOTE SENSING, VOL. GE-24 NO, I, JANUARY 1986

not in proportion to their use [2], In addition, newer orcomplex (sophisticated?) methods have not been adoptedwidely by practicing engineers, Presumably, this is be-cause the newer techniques have not been shown to givedemonstrably better results and, in general, their use re-quires more data and usually a large computer.

An interagency work group of the Hydrology Commit-tee of the Water Resources Council [31 was assigned thetask of developing consistent national guidelines for defin-ing peak flow frequencies at ungaged st ream locations,The many differing procedures used in practice and thelack of agreement about their use precluded selecting pro-cedures to include in a national guide without developingobjective information about procedure performance,

In a recent paper, Naef [4] addressed the success ofmodels in reproducing measured discharge, His condu-sions are based on two projects: the World MeteorologicalOrganization Intercomparison of Conceptual Models usedin operation hydrological forecasting [51, and on a studyof rainfall runoff models using data from small basins inSwitzerland, The results show that simple models can givesatisfactory results: however, neither the simple nor themore complex models tested were free from failure in cer-tain cases because none of them adequately describe therainfall-runoff' process, In addition, it could not be proveJthat complex models give better results than simpler ones,

A third study by Loague and Freeze [61 presentedmodel-performance calculations for three event-basedrainfall-runoff models on three data sets involving 269events from small upland catchments, The models includea regression model, a unit-hydrograph model, and a quasi-physically based model, The results of the study show sur-prisingly poor model efficiencies for all models on all datasets on an event-by-event bases, The poor performance ofthe quasi-physically based model could probably be as-cribed to a combination of model error and input error.They speculated that the primary barrier to the successfulapplication of physically based models in the field may liein the scale problems that arc associated with the unmea-surable spatial variability of rainfall and soil hydraulicproperties, The fact that simpler less-data-intensive modelsprovided as goe,d or better predictions than a physicallybased model is food for thought.

If one accepts these studies as indicators of the efl'ec-tiveness of recent hydrologic research results, one shouldask, Why? Why is it that more complex and more physi-cally based models do not give us better results? There isperhaps no clear answer, but I would speculate that lackof the proper amounts and types of data may be a largepart of the answer. I will try to show how remote sensingmay provide some new types of data that will help makethe complex models easier to use as well as improve theirperformance, For example, remote sensing may be theonly viable approach to handle spatial variability of wa-tershed properties because the basic data are spatial in na-ture, In this paper I will try to show how the period ofremote sensing may develop into a very large and signif-icant step in the knowledge curve, This will be based on

the uniqueness of remote sensing to obtain spatially dis-tributed information as well as some entirely new formsof measurement.

IV DEVELOPMENT OF REMOTE SENSING

Remote sensing as it is generally known today is an out-growth of photogrammetry, Strictly defined, remote sens-ing involves the collection of data by systems which arenot in direct contact with the item being measured. Earlyremote sensing emphasized interpretation of photos andiescriptive analysis of the subject. The launching of the~RTS A (later renamed Landsat J) satellite in July 1972tarted the modern era of remote sensing, The availability

of so many data on a repetitive basis covering tl.Jur spectralbands, all areas of the Earth, and a relatively fine resolu-tion was the impetus for a great deal of research. It is withthis background that this paper addresses remote-sensingapplications to hydrology, Two major subjects are ad-dressed-I) current applications of remote sensing to hy-drology and modeling, and 2) future directions for remotesensing in hydrology,

A, Current Applications

Landsat data have become a common source of infor-mation in hydrology. These data, like other remote-sens-ing applications, are generally used in a fairly simpleextension of photogrammetry. However, the unique cha -acteristics of specific spectral bands and the temporal se-quence of the data extend their usefulness beyond photointerpretation.

B. Land Use and Runoff' Coefficients

Land use or cover is an important aspect of hydrologicprocesses, particularly infiltration, erosion, and evapo-transpiration, Because of this, any process-oriented model(as opposed to a "black-box model") incorporates someland-use data or parameters. Distributed models, in par-ticular, need specific data on land uses identifled by lo-cation within the watershed. Most of the work to date onadapting remote sensing to hydrologic modeling has beenwith the Soil Conservation Service (SCS) runoff' curvenumber [7], The runoff curve number (RCN) is a coeffi-cient developed from a hydrologic characterization of thesoil and the land cover. The RCN may be further adjustedby antecedent precipitation to account for very wet or dryconditions, The importance of land cover can be demon-strated by comparing predicted runoff f()r a conditionwhere only the land use cnanges. For example, consider aB soil and a 10-cm rain; the calculated runoff for goodpasture condition would be approximately 0.25 cm,whereas, if that same field were planted in a small grainwith straight rows, the runoff calculated by the SCS pro-cedure w0uld be approximately 2.8 cm.

A number of studies have demonstrated the feasibilityof developing the land use categories from Landsat. First,suburban and urban areas were studied because the great-est contrast would be available between the imperviousand other more pervious areas. In a study on the UpperAnaeostia River Basin in Maryland, Ragan and Jackson

ENGMAN: HYDROLOGIC RESEARCH BEFORE AND AFTER AgRISTARS

[8] demonstrated the suitability of using Landsat-derivedland use data for calculating synthetic flood-frequency re-lationships. The Landsat-derived results were comparedto relationships developed from a conventional approachusing air photos.

In early work with remote-sensing data, Jackson et al.[9] demonstrated that land cover (particularly percent im-perviousness) could be used effectively in the U.S. ArmyCorps of Engineers [10] STORM model. In connectionwith the same study, Jackson and Ragan [II] used Baye-sian decision theory to demonstrate that computer-aidedanalysis of Landsat data was highly cost effective.

Slack and Welch [12] demonstrated that SCS RCN'scould be developed in a cost-effective manner for a pri-marily agricultural watershed in Georgia. Ragan and Jack-son [8] modified the land-cover requirements for the SCSprocedure for suburban areas so that Landsat data couldbe used. The RCN's developed from the Landsat dataclosely matched those obtained from a conventional ap-proach based on air-photo analysis. Synthetic flood fre-quencies developed from the two procedures were essen-tially identical. Bondelid et al. [13] developed a softwarepackage and user's manual to estimate RCN's efficientlyfrom Landsat data.

C. Snow Hydrology and Water Supply ForecastingWater supply forecast models for the western United

States have typically been of the multiple-regression form.

Y = a + bixi -I- b2X2 + b3X3 + ... bnxn

where Y is the runoff volume for the forecast period, Xh

... , Xn are the snow water contents at each of n snowcourses. The coefficients a and bh ... , bn are developedfrom empirical data. Other variables, such as fall precip-itation, base flow, etc., have been included in specificmodels. The areal extent of snow cover has not generallyb~en used because data to define it have not been availableexcept in certain case studies.

Leaf [14] used aerial photographs to develop relation-ships between snow cover and accumulated runoff for someColorado watersheds. He also showed that sequential pho-tos showing snow-cover depletion relationships could beused to help estimate the timing and magnitude of snow-melt peaks.

Since about 1973, the National Oceanic and Atmo-spheric Administration (NOAA) weather and Landsat sat-ellites have provided a visible and infrared data base ofsnow cover. With these data available, procedures for ana-lyzing the data have been developed [15]. NOAA has beenusing satellite data to map mean monthly snow cover overthe Northern Hemisphere [16].

Some of the first applications of satellite data were doneby Rango et al. [17]. They developed a regression modelfor snow melt in the Indus River basin. This study dem-onstrated the utility of satellite snow-cover data for largeareas with little or no data base. Aircraft and Landsatsnow-cover data were combined to develop a long-termdata base in California. The addition of snow-cover area

7

considerably reduced the seasonal runoff forecast error forthe King's and Kern River Basins [17].

In the Pacific Northwest, satellite snow-cover data arepresently being used operationally in the Streamflow Syn-thesis and Reservoir Regulation (SSARR) model. In testcases for five basins over a six-year period, the addition ofsatellite snow-cover data to the model resulted in a defi-nite but statistically insignificant improvement [18].

Landsat imagery was used to calculate snow-cover <'xeasfor six basins in Colorado over the period of 1973-1978[19]. They concluded that forecast error can be reducedon the order of 10 percent by using snow-cover data de-rived from the satellite.

In California, two areas were studied by comparing sat-ellite-derived snow-cover areas with conventional snowdata and by incorporating snow-cover areas into the State'sforecasts [20]. Results indicated potential improvement inthe forecast accuracy by using snow-cover area, particu-larly in areas where conventional snow data were limited.

Martinec [21] developed a snowmelt runoff model thatuses snow-cover area and temperature as input data. Rangoand Martinec [22] have demonstrated that this model canbe successfully used on basins as large as 500 km2 by usingLandsat data to determine the snow-cover area. Using thisapproach they were able to simulate seasonal volumeswithin 5 percent of actual values and were able to explainapproximately 85 percent of the variation in daily runofffor basins in the Wind River Mountains in Wyoming.

D. Flood and Floodplain MappingThe area inundated by floods and floodplains can be ef-

fectively mapped with remotely sensed data. Satellite datasuch as that from Landsat can be used to define coverageof a large river basin but may have some limitations onsmall basins because of resolution. Infrared photography,thermal infrared data, and multispectral scanner data haveall been successfully used to map the areal extent of flood-ing. These applications depend upon measuring changesin reflectivity caused by standing water, high soil mois-ture, moisture-stressed vegetation, and changes from am-bient temperatures. Flooding effects last for some timeafter inundation and can be detected up to two weeks afterthe passage of a flood. A number of studies using Landsatdata and infrared photography have been reported in a se-ries of papers related to this subject which can be foundin the Water Resources Bulletin, vol. 10, no. 5, 1974. Inspite of the coarser resolution (900 m versus 80 m forLandsat), the NOAA satellite thermal infrared sensor hasproved effective in measuring areas of flood inundation[23], [24]. In addition, the NOAA satellites have the ad-vantage of more frequent coverage (twice daily averageversus 18-day coverage for Landsat).

Floodplains have been delineated using remotely senseddata and inferring the extent of the floodplain from vege-tation changes or some other features commonly associ-ated with floodplains. Rango and Anderson [25 j have de-veloped a list of indicators that can be used to inferfloodplains from Landsat data. In a more recent study,

8IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL, GE··24 NO. I, JANUARY I~86

Sollers et af. 126J examined multispectral aircraft and sat-ellite classifications of land cover features indicative offlood plain areas. They concluded that satellite data canbe used to delineate flood-prone areas in agricultural andlimited development areas but may not give good resultsin areas with a heavy forest canopy. The remotely senseddata may best be used for preliminary planning and formonitoring flood plain activities with time,

V. FUTURE DIRECTIONS FOR REMOTE SENSING IN

HYDROLOGY

To date, most remote-sensing applications have con-sisted of fairly direct extensions of photogrammetry. How-ever, using information from specific spectral bands to in-fer land-use properties is an example of remote-sensinginformation used as a unique data source or measurement.The spectral classification used by Bondelid et al. [13 J isan example of this. Fortunately, we appear to be on thethreshold of major new breakthroughs in the uses of re-mote sensing data. These fall into four areas:

I) Measuring System States: Use of electromagnetic ra-diation outside of the visible range such as thermal in-frared and microwave for their unique responses to prop-erties important to hydrology.

2) Area versus Point Data: The use of data representingan area in which the spatial variability of specific param-eters of the area have been integrated.

3) Temporal Data: The potential for frequent measure-ment to develop time series of changes in given parame-ters and to monitor the dynamic properties in hydrology.

4) New Data Forms: The merging of several data setsof different wavelengths, polarizations, look angles. etc.to provide specific measurements of hydrologic parame-ters developed from the unique characteristics of remotesensing.

Each of these areas presents a unique opportunity forhydrologists to apply remote sensing in ways other thansimple extensions of photogrammetry. Remote sensing canproduce a complex measurement that is simultaneouslyobserving several factors. It is also giving us a view thatis uncommon to our past thinking in that it looks at a rel-atively large area and somehow integrates informationfrom the entire scene. The challenge is to learn how touse this information and to understand it. To do this wemust develop new concepts and challenge our usual wayof conceptualizing hydrologic processes. Some areas ofcurrent research and areas of opportunity are discussedbelow.

A. Monitoring System StatesOne of the more exciting aspects of remote sensing for

hydrologists is the potential for measuring and monitoringthe state of the hydrologic system. The major state varia-bles that appear to be useful are the soil moisture, snowwater content, snowpack condition, frozen soils, and tem-perature. For the most part, hydrologists have modeled thehydrologic system pretty much as a "black box," usingonly input data (usually rainfall and maybe potential evap-

oration) and the output hydrograph. The unit hydrographis a good example of a hydrologic "black box." The de-velopment of the comprehensive hydrologic model such asthe Stanford Model 1271 exposed the interior of the blackbox and subdivided the rainfall-runoff process into a num-ber of physical processes. However. from a systems pointof view, this type of model was still pretty much a blackbox because there were no provisions for monitoring ormeasuring any system states.

Attempts to use ancillary data such as soil moisture toimprove model performance have not been very success-ful. For the most part this is because existing models rep-resent the soil in a way to make the model work and havenot considered the possibility of independent determina-tion of soil moisture or soil parameters. Morton [28J makesthe point that our models attempt to make reality conformto our own concepts. He presents the argument that thecommonly used simulation models require assumptionsthat cannot be supported by theory or empirical studies.For the most part, he claims, the simulation models stressmathematical tractability and attempt to make reality con-form to conventional wisdom. Morton 128] proposes ananalytical approach based on spatial averages of the majorwater balance components. He suggests independent es-timates of areal evapot ranspiration because spatial esti-mates of changes in storage (soils, swamps, rocks) are toocostly to contemplate even for a small basin. However. thismay not be the case with remote sensing inasmuch as spa-tial changes in basin storage in the form of soil moisturecan be measured.

Snow, the amount and its condition, are important in-puts to models that predict the timing and amount ofsnowmelt runoff. Like soil moisture, microwave data ap-pear very promising to the snow hydrologist. Not only cana microwave sensor be an all-weather instrument becauseit penetrates cloud cover, it can also penetrate the snowpack, which presents one with the opportunity of inferringmany of the properties of the snow pack and the under-lying soil. These include depth and water content as wellas the degree of ripeness, crystal size, and the presenceof liquid water in a melting snow pack. As with soil mois-ture, the microwave measurement reflects several charac-teristics at once.

Frozen soils are another system state that, if known,would be extremely useful to hydrologists. With micro-wave remote sensing we can differentiate between frozenand nonfrozen soils. For a given soil moisture, the dielec-tric constant changes dramatically when the soil waterchanges from a frozen to a liquid state. Thus we have thepotential tor determining remotely whether or not a soilis frozen. This should greatly benefit the people respon-sible for flood forecasting, particularly in the upper mid-western United States. However, I am not sure that weknow how to use this information. How many of our pre-diction models are truly distributed so that they can useinformation that tells us the soils in certain areas of a wa-tershed are frozen? How do we change our runoff coeffi-cients or infiltration models to account for frozen soils?


How can we treat areas that are partially frozen and par-tially frost free (like south-facing slopes)? Can we deter-mine what type of frost is present (concrete versus col-umnar) and assign infiltration rates to each? These arequestions that must be answered before we realize the ben-efits possible from this measurement.

Surface temperatures are an additional system state thatmay prove useful to hydrologists. Satellite measurementsof surface temperature fields offer the potential for energybudget studies over large complex areas. Dodd 1291usedHeat Capacity Mapping Mission (HCMM) data in com-bination with a numerical model of the boundary layerproposed by Carlson and Boland 130J to estimate the spa-tial distribution of thermal inertial, moisture availability,and the sensible and evaporative heat fluxes. The approachwas tested over two urban areas, Los Angeles and St.Louis. Recent research by Price 1311 suggests the poten-tial for using remotely sensed thermal data for assessingthe surface moisture budget. In this study analyticalexpressions were derived, with a diurnal correction, thatrelate mean evaporation rate and a soil moisture parameterto surface temperature of bare soils. The possibility of de-termining the spatial distribution of evaporative flux ormoisture availability for complex areas has many potentialuses in hydrology, agriculture, forestry, and climatology.Much work needs to be done, particularly in improvingthe boundary layer models and understanding the edge ef-fects caused by land use changes and the spatial variationof roughness length.

Using system-state data will require new models devel-oped to incorporate the new data types. Such modelswould structurally resemble contemporary simulationmodels but would be more capable of accounting for spa-tial variability and changes. Also, the subprocess algo-rithms (infiltration, evapotranspiration, etc.) would be de-signed to use remote-sensing data as well as the moretypical inputs. Research being done at the Remote Sens-ing Systems Laboratory at the University of Maryland indeveloping a remote-sensing-based hydrologic model [32]has demonstrated the potential. This model is similar instructure to other watershed models such as the StanfordModel, but more of its parameters are physically based inthe sense that data to describe them can be obtainedthrough remote sensing. Another feature of this model isthe use of a geographic information system (GIS) as a datamanagement tool to produce the input data in a useableformat. The GIS assimilates remote-sensing data and thehistorically more common point data and provides a spa-tially distributed framework for the model.

B. Area versus Point DataRemote sensing measures spatial information rather

than point data. To some, this is a deficiency because theywould like to reproduce the point data they are comfort-able with. I would suggest this is because our conceptsand models have been developed from a point concept,i.e., raingage, soil column, and soil moisture access tube.Apparently there is much more information in a remote-

9

sensing scene, and it may be much more valuable than apoint measurement. We simply have to learn what infor-mation is there and how to use it. This may require de-veloping new concepts and models to accommodate thistype of information. As a mental exercise, consider howyou would develop a hydrologic model if you had only re-motely sensed data, had no schooling in traditional hy-drology, and had no awareness of raingages, soil columnmodels, and similar point concepts or measurements. Thehigh degree of understanding we have developed for themovement of water down through a soil profile is a casein point. I suggest that by concentrating on details only inthe vertical direction, you have been looking at the wrongquestion. It seems to me that variability in the horizontalplane may be hydrologically made more significant thananything we have been studying in the last few decades.Remote sensing, and its ability to measure the responsefrom an area, is potentially one way to approach this prob-lem.

C. Temporal DataRemote-sensing data from a satellite platform can pro-

vide unique time series data for hydrologic use. The actualfrequency of observation can vary from continuous to onceevery two weeks or so, depending upon the sensors andtype of orbit. This approach is appealing because it maybe a very cost-effective method to monitor various hy-drologic states over very large areas.

Most continuous simulation models are mass balance-type models, taking rainfall (or snowmelt) as input andafter storage and losses, routing it to stream flow. Thestored water defines the state of the system and, as such,controls the rate of sequential processes and events. Sinceeach successive computation is based on the previous stateof the system, errors in the predicted output often getlarger with time. How well could we improve our predic-tion if we could check our system periodically and updateour predictions? Repetitive measures of soil moisture usedas feedback to the model could do this. Improved predic-tion accuracy may have large tangible benefits.

A recent study by Jackson et al. 1331 demonstrated howpossible applications of repetitive remote measurements ofsoil moisture might be used. They discussed how theseareal data may be used to calibrate soil and vegetationparameters and to correct errors resulting from point mea-surements of precipitation. In the study they demonstratedhow soil moisture observations are useful in calibrationand updating the state of the system. However, it was alsopointed out that the model structure itself may preclude avalid analysis of the value of soil-moisture measurementsor the frequency needed to improve the simulations. Onemust carefully choose the model to be used in this type ofstudy; it may be necessary to develop a new model or makesignificant modifications to existing models.

D. New Data FormsThe spatial and temporal possibilities of remote-sensing

data coupled with direct measurement of hydrologic state

10IfiEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSiNG, VOL. Gfi-24 NO. I, JANUARY 1986

variables may lead to an entirely new type of hydrologicdata or model parameters. The use of remote sensing todetermine land use for the SCS RCN and other runoff coef-ficients was reviewed in an earl ier section. Although usedas an extension of photogrammetry, these examples didillust rate the use of characteristics of specific spect ralbands to infer the land-use properties. We may be able todetermine runoff characteristics directly because a re-mote-sensing measurement potentially can integrate sev-eral features into one response.

The response of different wavelengths is determined bythe surface and near surface of the target. This measuredresponse is a composite response of several individual fea-tures. For example. microwave brightness temperature isaffected by surface roughness, grain size of the soil, veg-etation cover. and soil mois,ure. Each of these has a dif-ferent effect, and this effect varies with wavelength, angleof incidence, and polarization. Therefore, anyone micro-wave measurement is an integrated measure of these ef-fects as well as a spatial sample. These features are thesame watershed characteristics that are llsed to describethe hydrologic characteristics of a watershed, i.e., soilcover or management practice and antecedent moisture.

It is possible that remote sensing in the microwave areacan give us a direct measure of runoff potential or a runoffcoefficient. Blanchard et at. [34] have had some successin determining a SCS RCN for some watersheds in Texas.In their study using an airborne passive microwave imag-ing scanner, they investigated the relationship betweenRCN and the antenna temperature differences for twoflights over the same watersheds. Recent research by Zev-enbergen et at. [35] showed high correlations betweenLandsat-derived soils or vegetation indices and RCN forrangelands. This work suggests that a soil-cover complexmay be a good estimate of potential runoff in natural rangeareas. This type of study suggests that we should considerthe remote-sensing measurement as a direct measure ofrunoff potential in the same sense as an infiltrometer di-rectly measures infiltration. Use of several different wave-lengths, polarizations. etc., may provide all the informa-tion we need to predict runoff for a fairly large area. Todo so may require that hydrologists develop some new con-cepts or models to use this type of information.

The spatial and temporal measurement of soil moisturemay lead to an entirely new types of hydrologic data ormodel parameters. The key to this approach would be toexamine changes in a watershed state with time series re-motely sensed data. Observations of how a watershed soilmoisture changes during a drying cycle of several days (10or so) may provide new insight into the storage changesin a watershed. It may also give one insight into the hy-drologic performance, rather than hydrologic characteris-tics, of soils. The difference between hydrologic perfor-mance and characteristics would be that a hydrologiccharacteristic would be analogous to a lab measurementwhich has no context with spatial variability or topo-graphic location within a basin. The hydrologic perfor-mance of a soil, on the other hand, would be a quantifiable

characteristic that reflects not only the physical character-istics of the soil but also how it behaves hydrologicallywith the basin. Its spatial distribution with respect to ele-vation, flowing streams, and other soils would combine todefine its hydrologic performance.

VI. FUTURE CONSTRAiNTS

All these potentia] approaches to improving our hy-drologic knowledge exist. There have been enough data.truck experiments. aircraft flights. and space platforms towhet our appetites. Unfortunately, we apparently arecaught in a Catch-22 situation. We need to be able to dem-onstrate that these remote sensing data will indeed im-prove our hydrologic knowledge and improve hydrologicapplications in order to justify specific instrumentation onsome type of aircraft or space platform. However, we needthese platforms and instruments to demonstrate the utilityof the data.

What is needed is a commitment to put some of theseinstruments into space [36] so that the hydrologic proce-dures can be developed. Our work to date has highlightedthe potential for remote sensing in hydrology. Improvedperformance of our hydrologic procedures and models ispretty much at a standstill because of a lack of the propertypes and amounts of data. The integrated Earth obser-vation system being planned by NASA [37] would go along way to provide these types of data. We need to en-courage the development of this program and even speedup its development (p]anned for flight in the ]990's). Ithink we know enough about wavelengths. incident an-gles, and resolution to come up with a sound HYDROSATpackage. It may not be the optimal system, and undoubt-ably the characteristics of future systems will change fromwhat we learn. In the meantime we need to anticipate howwe could use such data with aircraft experiments and sim-ulated data.

REFERENCES

[11 v. T. Chow, Ed., Handbook of Applied Hydrologv. New York:McGraw-Hili, 1964.

[2] R. H. McCuen. and W. J. Rawls, "Classification of evaluation of floodflow frequency estimation techniques." Water Res, Bull., voL 15, no,L pp, 88-93, 1979,

[31 Water Resources Council, Hydrology Committee, "Estimating peakflow frequencies for natural uugagcd watersheds-A proposed nation-wide test," U,S. Water Resources Council. Washington, DC, 1981.

[4] F, Naef, "Can we model the rainfall-runoff process today?" Hydrol-o[{ie Sei. Bull,. voL 26, no. 3, pp, 281-289, 1981.

[5] World Meterorological Organization, "Intercomparison of conceptualmodels used in operational hydrological f()recasting," WMO Opera-tional Hydrology Rep. 7, 1975,

[6] K, M. Loague and R. A, Freeze, "A comparison of rainfall-run off-modeling techniques on small upland catchments," [¥iller ResourcesRes" voL 21, no, 2, pp, 229-248, 1985,

[7] U,S. Soil Conscrvation Service, SCS National Engineerin[{ Hand-IJlJOk, Section 4: Hvdrologv, U.S, Dep, of Agriculture, Washington,DC. 1972.

l8] R, M, Ragan and T J. Jackson, "Runoff synthesis using Landsat andthe SCS model," Proc. ASCE, paper 15387, vol. 106, no. HY5, pp.667-678, 1980,

[9J T. J, Jackson, R, M, Ragan, and W, N, Fitch, "Test of Landsat-hasedurhan hydrologic modeling," ASCE J Water Resources, Planning andM{l/w[{ement Division, voL 103, no. WRI, pp, 141-158. 1977,

[10] US Army Corps of Engineers, "Urhan storm water runoff STORM,"


Hydrologic Eng. Center. Davis, CA, Computer Program 723-58-L2520, 1976.

[II] T. J. Jackson and R. M. Ragan, "Value of Landsat in urban waterrcsourccs planning," ASCE J. !¥lIter Resources, Planning and Man-agement Division, vol. 103, no. WRI, pp. 33-46, 1977.

[12] R. B. Slack and R. Welch, "Soil conservation service runoff curvenumber estimates from Landsat data," Water Resources Bull., vol.16, no. 5, pp. 887-893, 1980.

[13] T. R. Bondelid, T. J. Jackson, and R. H. McCuen, "A computer basedapproach for estimating runoff curve numbers using Landsat data,"AgRISTARS Conservation and Pollution Tech. Rep. CR-R 1-04040,1981.

[14] C. F. Leaf, "Aerial photographs for operational streamflow forecast-ing in the Colorado Rockies," in Proc. 37th Western Snow ConI (Sa]tLake City, UT), 1969.

[15] A. Rango and K. l. Wen, "Satellite potentials in snowcover monitor-ing and runoff prediction," Nordic Hydrology, vol. 7, pp. 209-230,1976.

[16] D. R. Wiesnet and M. Matson, "Monthly winter snowline variationin the Northern Hemisphere from satellite records, 1966-1975," Nat.Environ. Satellite Service, Washington, DC, NOAA Tech. MemoNESS 74, 1975.

[17] A. Rango, J. F. Hannaford, R. L. Hall, M. Rosenzweig, and A. J.Brown, "The use of snow covered area in runoff forecasts," NASAGoddard Space Flight Center, Greenbelt, MD, Document X-9i3-77-48, 1977.

[18] J. P. Dillard and C. E. Orwig, "Use of satellite data in runoff fore-casting in the heavily forested, cloud covered Pacific Northwest," inProc. Workshop on Operational Applications of Satellite SnowcoverObservations, NASA Conf. Pub. 2116, pp. 127-150, 1979.

119] B. A. Shafer and C. F. Leaf, "Landsat derived snow cover as an inputvariable for snowmelt runoff forecasting in central Colorado," in Proc.Workshop on Operational Applications of Satellite Snowcover Obser-vations, NASA Conf. Pub. 2116, pp. 151-169, 1979.

[20] A. J. Brown, J. F. Hannaford, and R. L. Hall, "Application of snowcovered area to runoff forecasting in selected basins of the Sierras,Nevada, California," in Proc. Workshop on Operational Applicationsof Satellite Snow Cover Observations, NASA Conf. Pub. 2116, pp.185-200, 1979.

[21] J. Martinec, "Study of snowmelt-runoff process in two representativewatersheds with different elevation range," in Results of Research andExperimental Basins, Proc. Wellington Symp., publ. 96, pp. 29-39,1970.

[22] A. Rango and J. Martinec, "Application of a snowmelt-runoff modelusing Landsat data," Nordic Hydrology, vol. 10, pp. 225-238, 1979.

[23] C. P. Berg, M. Matson, and D. R. Wiesnet, "Assessing the Red Riverofthe north 1978 flooding from NOAA satellite data," in Proc. Pecora5 Symp. (Sioux Falls, SD), pp. 309-315, 1979.

[24] G. Tappan, N. C. Howath, P. C. Doraiswany, T. Engman, and D. W.Goss, "Use of NOAA-N satellites for land/water discrimination andflood monitoring," AgRISTARS Rep. EW-L3-04394, 1983.

[25] A. Rango and A. T. Anderson, "Flood hazard studies in the Missis-sippi River Basin using remote sensing," Water Resources Bull., vol.10, no. 5, pp. 1060-1081, 1974.

[26] S. C. Sollers, A. Rango, and D. L. Henninger, "Se]ecting reconnais-

11

sance strategies for flood plain surveys," Water Resources Bull., vol.14, no. 2, pp. 359-373, 1978.

[27] N. H. Crawford and R. K. Linsley, "Digital simulation in hydrology:Stanford watershed model IV," Stanford University, Tech. Rep. no.39, 1966.

[28J F. I. Morton, "Integrated basin response-A problem of synthesis ora problem of analysis," in Proc. Canadian Hydrology Symp. (Asso-ciate Committee on Hydrology, Natal Res. Council, Canada), pp. 361-363, 1982.

[29] J. K. Dodd, "Determination of surface characteristics and energybudget over an urban-rural area using satellite data and a boundarylayer model," Master's thesis, The Pennsylvania State Univ., 1979.

[30] T. N. Carlson and F. E. Boland, "Analysis of urban-rural canopyusing a surface heat f1uxltemperature model," J. Appl. Meteorol.,vol. 17, no. 7, pp. 998-1013, 1978.

[31] J. C. Price, "Use of remotely senscd infrared data for inferring en-vironmenta] conditions from surface characteristics and regional scalemeteorology," in Proc. 1981 11ll. Geosci. Remote Sensing Symp.(Washington, DC), pp. 1195-1201, ]981.

1321 J. R. Groves and R. M. Ragan, "Development of a remote sensingbased continuous streamflow model," in Proc. 17th 1m. Symp. RemoteSensing Environ. (Ann Arbor, MI), pp. 447-456, 1983.

[33] T. J. Jackson, T. J. Schmugge, A. D. Nicks, G. A. Coleman, and E.T. Engman, "Soil moisture updating and microwave remote sensingfor hydrologic simu]ation," Bull. Int. Assoc. Scientific Hydrology, vol.26, no. 3, pp. 305-319, 1981.

[34] B. J. Blanchard, J. W. Rouse, Jr., and T. J. Schmugge, "Classifyingstorm runoff potential with passive microwave measurements," WaferResources Bull., vol. II, no. 5, pp. 892-907, 1975.

[35] A. W. Zevenbergen, "Runoff curve numbers for rangeland fromLandsat data," Hydrology Lab. Tech. Rep. HL85-1, 1985.

[36] A. Rango, "Assessment of remote sensing input to hydrologicmodels," Water Resources Bull., 1985.

[37] NASA, "National Aeronautics and Space Administration, Earth ob-serving system," Goddard Space Flight Center, Greenbelt, MD,NASA Tech. Memo. 86129, 1984.

*

Edwin T. Engman received the B.E. and M.S.degrees in agricultural engineering from CornellUniversity, Ithaca, NY, and the Ph.D. degree incivil engineering from Pennsylvania State Univer-sity, University Park, PA.

Before assuming his current position in 1976,he spent 13 years with the Agricultural ResearchService as a Research Hydrologist in Vermont andsubsequently as the Director of the Northeast Wa-tershed Research Center in Pennsylvania. In ad-dition, he worked for two years as a Principal Hy-

drologist with the NUS Corporation in Maryland. Since 1976, his researchhas concentrated on remote-sensing applications in water resources, em-phasizing use of microwave data for soil-moisture studies.

12IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING. VOL. GE·24 1\0 I, JANUARY 1986

Passive Microwave Soil Moisture ResearchTHOMAS SCHMUGGE, MEMBER, IEEE, PEGGY E. O'NEILL, MEMBER.IEEE. AND JAMES R. WANG

Thermal microwave emission from soils is generatedwithin the soil volume. The amount of energy generatedat any point within the volume depends on the soil dielec-tric properties (or soil moisture) and the soil temperatureat that point. As the energy propagates upward throughthe soil from its origin, it is affected by the dielectric gra-dients along the path of propagation. In addition, as theenergy crosses the surface, it is reduced by the effectivetransmission coefficient (emissivity) of the surface, whichis determined by the average dielectric characteristics ofthe soil in a transition layer just below the surface. It isthe thickness of this layer which determines the actual soilmoisture sampling depth and it is the properties of thislayer which have the dominant effect on the emitted inten-sity. Theoretical studies [5], 16] have indicated that thistransition layer is a few tenths of a wavelength thick orabout 2 to 5 cm at a 21-cm wavelength. The field verifi-cation of this depth was done as part of the AgRISTARSprogram and will be described in Section II.

Field and aircraft experiments up to 1980 had demon-strated the basic sensitivity of radiometric measurementsto surface layer soil moisture. These experiments also in-dicated that there are a number of factors other than soilmoisture which influence the intensity of the emission fromthe soil; these include, among others, surface roughness,vegetation cover, and soil texture. A review of the ap-proximate status of microwave sensing of soil moisture wasgiven in the paper by Schmugge P] presented at the ERIMSymposium of that year. Thus, at the beginning of theAgRISTARS program we were able to identify a numberof questions which should be addressed in the program.Those revelvant to passive microwave systems are:

The brightness temperature observed by a radiometer at aheight H above the ground is

where r is the surface reflectivity and T(H) is the atmo-spheric transmission. The first term is the reflected skybrightness temperature which depends on wavelength andatmospheric conditions; the second term is emission fromthe surface (1 - r = e, the emissivity); and the third termis the contribution from the atmosphere between the sur-face and the receiver. At the longer wavelengths best suitedfor soil moisture sensing, the atmospheric effects are min-imal. Thus, (1) can be written as

(2)

L INTRODUCTION

Abstract-During the four years of the AgRISTARS Program, sig-nificant progress was made in quantifying the capabilities of microwavesensors for the remote sensing of soil moisture. In this paper we discussthe results of numerous field and aircraft experiments, analysis ofspacecraft data, and modeling activities whkh examined the variousnoise factors such as roughness and vegetation that affect the inter-pretability of microwave emission measurements. While determiningthat a 21-cm wavelength radiometer was the best single sensor for soilmoisture research, these studies demonstrated that a multisensor ap-proach will provide more accurate soil moisture information for a widerrange of naturally occurrring conditions.

AT THE START of the AgRISTARS Program in 1980,much was known about the basic sensitivity of micro-

wave sensors to soil moisture variations, although the im-portance of other soil and scene parameters had not yetbeen determined. Therefore, the major thrust of theAgRISTARS Soil Moisture Program was to quantify thesenoise sources and to demonstrate the utility of microwavesensors to soil moisture research. The basis for microwaveremote sensing of soil moisture is the strong dependenceof the soil's dielectric properties on its moisture contentdue to the large contrast between the dielectric constant ofwater (- 80) and that of dry soil (- 3). The dependenceof the dielectric constant on moisture content was modeledas a function of soil texture by Wang and Schmugge [1].Subsequent improvements in our understanding of the di-electric properties of soils were made in this program andwere recently reported by the University of Kansas group[2], [3]. This dependence of the soil's dielectric propertieson moisture content can be observed with either passiveor active microwave sensors through its effect on the soil'semissivity and reflectivity. The progress made with activemicrowave sensors will be documented in another paperin this issue [4).

The change in the soil's dielectric constant based on itswater content produces a change in its emissivity fromabout 0.95 when dry to 0.6 or less when wet, which im-plies a change of about 30 percent in the natural emissionfrom the soil. At microwave wavelengths, the intensity ofthis emission is essentially proportional to the product ofthe temperature and the emissitivity of the surface (Ray-leigh-Jeans approximation). This product is commonlyreferred to as the microwave brightness temperature (Tn).

ManuscriptreceivedApril I. 1985; revisedAugust I. ]985.The authorsare with the Spaeeand EarthSciencesDirectorate.NASA

GoddardSpaceFlight Center.Greenbelt.MD20771.IEEE LogNumber8406231.

u.s. Government work not protected by U.S. copyright

SCHMUGGE et at.: PASSIVE MICROWAVE SOIL MOISTURE RESEARCH

What are the optimum wavelengths or combination ofwavelengths for microwave and thermal IR systems?

What is the soil moisture sampling depth as a functionof wavelength?

Can moisture gadient information be obtained bymultiwavelength systems?

What are the limits on the accuracy of the soil moistureestimation imposed by:

surface roughnessvegetative coversoil heterogeneityincident angle dependencesurface cover heterogeneityatmospheric effects for space systemsmixed scene in the large footprint of spaceborne sen-sors.

In the course of the four years that the AgRISTARS pro-gram was active, measurements and analyses were per-formed which addressed, at least indirectly, all these ques-tions except the last two. As discussed above, it was feltthat the atmospheric effects are not significant at the mi-crowave wavelengths relevant for soil moisture sensing.The studies of the effects of mixed scenes within aspaceborne radiometer footprint were scheduled for laterin the program and were not performed. There had beenone such study done prior to the AgRISTARS program[8]. In this paper, we will summarize the progress madein addressing these questions through the field and aircraftmeasurements, analysis of spacecraft data, and modelingstudies.

II. FIELD MEASUREMENTS

A. Bare SoilThe purpose of the field measurements program is to

perform detailed studies of the basic interaction of theelectromagnetic radiation with the soil surface and the de-pendence of the subsequent emission on various soil prop-erties. Field measurements have the advantage that only asmall plot (e.g., 20 X 20 m) is observed which can berelatively well documented. As a result, the effects ofchanges in the surface parameters can be studied undercalibrated conditions. Field programs have been per-formed by the Remote Sensing Center of Texas A&MUniversity during several years of the AgRISTARS pro-gram and as a cooperative effort of GSFC and the USDAHydrology Laboratory of the Beltsville Agricultural Re-search Center (BARC) during the four years 1979-1982.

One of the early significant results of these experimentswas the verification that the soil moisture sampling depthwas only a few tenths of a wavelength. This was done bythe group at TAMU by studying the drydown of a moistsoil with both radiometric and gravimetric (0-2, 0-5, 0-9-cm layers) measurements of the surface soil moisture[9]. Initially, the three layers dry at the same rate, butafter three days they begin to diverge with the surface layerdrying faster. These observations were compared with the

13

temporal variation of surface soil moisture estimated fromthe radiometric measurements made at 1.4, 5.0, and 10.7GHz. Again, results at the three frequencies indicatedrying at the same rate initially but then diverging afteronly about 2 days. The two higher frequency estimatesshow a dry value of about 8 percent on day 12, which thelowest frequency and the 0-2-cm gravimetric measure-ment did not reach until day 14. This result indicates thatthe sampling depth for the l.4-GHz radiometer is betweenthe 2- and 5-cm level.

Bare soil measurements were made at BARC as func-tions of angle between 10° and 70° at both L (1.4 GHz)and C (5 GHz) bands for several years over bare fieldshaving two different soil textures [10]-[12]. Examples ofthe data are given in Fig. 1 for L- and C-bands at 10°.These plots show emissivity and normalized brightnesstemperature (NTB) versus soil moisture in a 0-2.5-cmlayer. NTB is obtained by multiplying the emissivity by300, so that the plots can be compared more directly withthe observations. The data are from two fields with differ-ent soil textures. Most of the data were from a sandy loam(sand = 68 percent, clay = 11 percent) field which had arelatively smooth surface and the remainder were from aloam (sand = 34 percent, clay = 24 percent) plot whichhad a somewhat rougher surface. The data from the twosoils are represented by different symbols (* = sandy loamand x = loam). The dashed curves are the calculated em-issivities from the Fresnel equations for the two soils as-suming a smooth surface. The lower one is for the sandyloam. The solid line is the regression fit to these data. AtL-band, the calculated emissivities are at most 0.05 belowthe observed regression line, with much of the data lyingabove the calculated curves. We believe that the higherobserved values are primarily due to the surface rough-ness of these fields which tends to increase the emissivityof the surface [13]. The range of emissivities is about thesame for the calculations (0.6 to 0.9) and the observations(0.63 to 0.95). The C-band data behave similarly with sev-eral important differences. While the emissivities for thewet soils are about the same at the two wavelengths, theyare higher at C-band for dry soils. This results from thefact that the sampling depth at the shorter wavelength isshallower than the soil layer measured (0 to 2.5 em). Forexample, at a soil moisture of 10 percent in this layer, theC-band is responding to the drier soil closer to the surface.As a result, the slope and intercept of the regression lineare greater at C-band.

For a smooth surface the emissivity in the vertical po-larization will increase with increasing incidence angle outto Brewster's angle (OB) where the emissivity is one. Theeffect of soil moisture is to increase the value of OB (OB =tan -1 .JK, where K is the dielectric constant) from about62 ° for a dry soil to 79° for a wet soil. The net result ofthis is to decrease the sensitivity of the vertical polariza-tion to soil moisture variations at the off-nadir angles.

The results of the statistical analyses of these data arepresented in Table 1. The values presented are the slopeand intercept of the regression, the rms difference be-

14IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL GE-24 NO. I. JANUARY 19S6

BAR'> (-8A/{) DATA ANJ L1~AR REGl'lESSICl'<RE9-LTS

A • 3128 ••. 22 ANJ SID OCVIATlCl'< • 16,21<

BARe, L-BANJ DATA Ai\O LI"EAR REGRESS1()\j RESLlT5

A •• 2938" 3.17 At-£] sro OCV!ATION - J2.4K

L H£TA"]O OCGREE

I 10-2.5 CM)

140 L. __.1 __ 1.__ ~ --.l . L ---L-----L __ J Q.5

:0 20 30 40

VCt.lt£TR1C SOIL I"O[STL~ el)

BARE FIELD (SM:XJTHl

Ii • 0.7B

N • 6B

1.D

D.9

0.8

0.7

0.6

TH

0,5

10

Fig. 1. Field measurcmcnts of normalized brightness temperature (NTB)or emissivity for bare smooth fields at L- and C-bands !()r an incidenceangle of 10°, The values of the regression results are also givcn. A: in-tercept, B: slope and the rrns deviation from the regression.

TABLE I

ANGLE l-BAND (HORIZONTAL) l- BA N D (VERTICAL) C-BAND(HORIZONTAl) C-BAND(VERTICAl)AI 7 1RR4N -B~ STW N STD RR srn RR II A STD RR

10 71 293 3.17 12.4 0.76 71 294 3. 00 11. 3 0.78 68 312 4.22 16.2 0.78 68 312 4.06 15.6 0.7720 71 288 3.24 12.4 0.77 71 294 2.88 11.2 0.77 68 310 4.37 17.6 0.76 68 312 3.98 16.1 0.76:\0 71 281 :\.30 12.4 0.78 71 293 2.69 10.5 0.77 68 307 4.58 18.4 0.76 68 314 3.80 15.2 0.7640 69 271 3.:\0 12.8 0.77 71 291 2.36 10 I 0.73 68 .30I 4.70 19.6 0.75 68 315 3.40 13.4 0.7750 69 255 3.21 12.8 0.76 70 287 1.87 8.9 0.69 66 288 4.66 21.0 0.72 68 314 2.75 10.7 0.7760 69 228 3.02 13.1 0.73 71 277 1.15 6.8 0.59 68 266 4.52 22.8 0.67 68 305 1.81 6.9 0.7870 71 188 2.49 12.4 0.67 70 249 0.05 6.3 O. 00 68 232 3.92 23.1 0.59 68 276 0.30 4.7 0.17

BRIGHTIIESS TEMPERATURES NORMALIZED TO' 300. a K. THE SURFACE ROUGHNESS PARAMETER. H = 0.00

lA intercept/~B = slope

= standard error4RR = r-squared value

tween the data and the regression line, and the r-squaredvalue of the regression. There are several factors whichshould be noted:

1) the monotonic decrease in sensitivity for the verticalpolarization with increasing angle, which is due tothe Brewster angle effect;

2) the almost constant slope and r-squared values forthe horizontal polarization out to 50°; and

3) the similar behavior at C-band with the differencethat the slope and intercept are higher due to theshallower sampling depth at this frequency.

absorption is primarily due to the water content in the veg-etation. The precise sources for the scattering are not wellunderstood at the present time and are the subject of muchcurrent research, both experimental and theoretical.

The radiation measured by a radiometer can be ex-pressed as the sum of three terms: the emission from thesoil reduced by the vegetation and the emissions from thevegetation, both direct and that reflected from the soil sur-face. The strength of the emission from the vegetation willbe proportional to its absorption, which is described bythe optical depth (7):

where t is the canopy thickness and a is the volume ab-sorption coefficient which depends on the real and imag-

B. Vegtation-Covered SoilA vegetation layer covering the soil will absorb and

scatter some of the microwave radiation incident on it. The

T = at sec (8) (3)

SCHMUGGE el al.: PASSIVE MICROWAVE SOIL MOISTURE RESEARCH]5

o 234567

W, CANOPY WATER (kg/m')

Fig. 2. Variation of optical depth (7) determinations with vegetation watercontent for corn-, grass-, and soybean-covered fields [16],

• SOYBEAN0.5 x CORN

• GRASS"SORGHUM

CfJ "CfJ "wz 0.4><:uIf-...J<iu 0.3~0

0.2

0.1

inary parts of the canopy dielectric constant. Brunfeldt andUlaby [14] have shown that the dielectric constant for theleaf portion of the canopy can be expressed in terms of itsfractional volume and its dielectric constant. The frac-tional volume of the leaf portion of the canopy can be es-timated from the leaf thickness and the leaf area index(LAI) for the canopy, The leaf dielectric constant is astrong function of plant water content. The stalk and fruitportions of the canopy are more difficult to estimate be-cause their dimensions are comparable to the sensor wave-lengths. However, since water is the dominant dielectriccomponent of the vegetation, the optical thickness can beparameterized in terms of the canopy water content (W)given in units of kilograms per square meter to a first ap-proximation [15].

Values for the optical depth can be estimated by com-paring the observed radiation from a vegetated field withthat expected for a bare field either from a model or fromobservations of the field after the vegetation has beenstripped off. A summary of optical depth measurementsobtained by these approaches is given in Fig. 2 plottedversus canopy water content [16]. It is clear that there isa strong linear dependence of the optical depth on the can-opy water content.

In an analysis of the expected emission over a vegetatedcanopy, Jackson et al. [15] have shown that by assumingthat the plant and soil temperatures are approximatelyequal the observed emissivity can be written as

e = TalTs = 1 + (e, - 1) exp (-27) (4)

where es is the bare soil emissivity and Ts is the soil tem-perature. Thus, the sensitivity to soil moisture variations

(5)

C. Multiwavelength MeasurementsSeveral experiments involved microwave measurements

made at wavelengths greater than 21 em, i.e., at 50 cm atBARC [11] and at 40 em by a JPL group [17]. As ex-pected, the longer wavelengths showed a greater ability topenetrate vegetation and a deeper soil moisture samplingdepth. However, it was found in both experiments that thesensitivity to soil moisture variations was less at the 50-em wavelengths than at the shorter wavelengths. For theBARC data, the sensitivity of the 50-em brightness tem-peratures to soil moisture was approximately one-half thatobserved in the 21-cm data. For the wettest fields, the em-issivitiy was up to 0.1 higher than that observed at eitherthe 21- or 6-cm wavelengths. The reasons for this de-creased sensitivity have not been determined. Because ofthis result and the problems of increased interference fromman-made sources of radiation and poorer spatial resolu-tion for a given antenna size, further studies at the longerwavelengths were not pursued.

A. Bare FieldsMost of the aircraft experiments utilized the sensor pack-

age flown on NASA research aircraft (primarily the C-130), which included microwave radiometers at 1.4 and 5GHz, and microwave scatterometers at 13.3, 4.75, 1.6, and0.4 GHz. Cameras, a thermal infrared radiometer, and a

where sm is the volumetric soil moisture. An optical thick-ness of about 0.7 reduces the sensitivity of microwave ra-diometers to soil moisture to - 25 percent that of baresoil. In Table I the sensitivity was - 3.2K per unit soilmoisture; thus, the presence of vegetation would reducethat to 0.8K per unit soil moisture which is about the limitof useful sensitivity for soil moisture measurement. Thislevel of optical thickness is attained by a mature corn can-opy, making soil moisture detection very difficult througha crop cover of this density.

III. AIRCRAFT EXPERIMENTS

As discussed in the previous section, microwave sen-sors mounted on truck platforms are used to develop soilmoisture estimation algorithms based on remotely senseddata, since they allow the evaluation of factors such as soiltype, roughness, and vegetation under well-controlled con-ditions. At the same time, complementary research hasbeen conducted with aircraft sensors in order to examinethe effects of large area coverage, scene heterogeneity, andinstrument sensitivity at the lower resolutions typical ofaircraft systems. While results from these experimentsgenerally verify the basic relationships between sensormeasurements and soil moisture derived from theory andthe small-scale ground-based studies, they also demon-strate the limited utility of a single-sensor approach to soilmoisture prediction for a wide variety of naturally occur-ring conditions.

is reduced by this exponential factor

deldsm = exp (- 27) deJdsmL-BAND = 21 em

T = (0115 ± 0.0041 W" = 0.950.6

07

16IEEE TRANSACTIONS ON (iEOSCIENCE Ar\D RE\lOTE SEr\SING, VOL. (iE-24 NO. I, JAr\lJARY 1986

B D

\ 00

'"Q,

::<'" 850-

V'l

:flz0-rCJ 80ii'CD

ICJ

75

70

J I I0 10 20

I

30

D

I40

GD

G

50

B J Ii&'B J

BBB

B

3--.."'--K

"- "'--

B B

B

B

KEYJ = JPL TRUCK DATA 1979-808 = BARC TRUCK DATA 1979-81G = GUYMO,\j A/C DATA 1978D = DALHART A/C DATA 19BOK = CHICKASHA A/C DATA 1978, 1980

r'= .79SLOPE = -.0023

SOUTH DAKOTA ond

PHOENIX data

r2.O.70

SLOPE •• ',001 S

K

170

0-2CM VOLUMETRIC SOIL MOISTURE BY 96 OF FIELD CAPACITY

Fig, 3, Summary "rtruck and aircraft results at L-band !()r several differentmeasurcment calnpaigns comparing emissivity with volumetric soilmoisture 12HJ,

visiblelinfrared scanner were also available aboard this air-craft. Flights took place between 1976 and 1980 at a num-ber of agricultural sites across the country, such as HandCounty, South Dakota; Colby, Kansas; Chickasha, Okla-homa; Guymon, Oklahoma; Dalhart, Texas; and TaylorCreek, Florida, Field cover conditions ranged from bareand grassland pasture to crops such as wheat, milo, al-falfa, corn, and citrus trees.

In examining the data from bare and minimally vege-tated fields, researchers noted st rong correlations betweenthe microwave brightness temperature and near-surfacesoil moisture (typically 0-2 cm or 0-5 cm depth). An ex-ample of this relationship with data from several of theaircraft experiments is presented in Fig. 3. The solidregression line is the combined result of three aircraft andtwo ground-based experiments, while the dashed line rep-resents aircraft data from South Dakota and Phoenix.Normalized TB is calculated by dividing the observed T

Bby the soil temperature, and is an approximation to thesoil's emissivity. At the high soil moisture levels indi-cated, measurements from aircraft radiometers showedhigher emissivities than those from truck sensors, due tosoil texture, roughness, and vegetation cover differencesbetween the two types of test sites. (The truck experi-ments involved carefully prepared plots of sandy soil withsmooth, bare surfaces.) Even though these factors werenot corrected for in this comparison, the relationship be-tween microwave emissivity and surface soil moisture isstill quite strong.

As roughness and vegetation increased beyond minimallevels, the microwave sensitivity to soil moisture de-creased, especially at higher frequencies, Even smallamounts of vegetation masked underlying soil moisturevariations at frequencies higher than C-band [18J. Basedon an evaluation of coincident data from the different in-struments, the 1.4-GHz radiometer operating at a nadirlook angle was determined to be the best single sensor forsoil moisture determination in bare fields [ 181-120 I.

A problem with evaluating the accuracy of soil moistureestimation algorithms derived from remotely sensed datalies in the variability of point samples of soil moisturewhich are averaged together to determine mean moisturelevels on a per field basis. Using I.4-GHz passive data forbare and pasture fields, numerous researchers were ableto estimate volumetric soil moisture with a standard errorof 4-6 percent [18], 123J-[251. However, this level of erroris approximately the same as the variation in the groundsamples of soil moisture [24J. In fact. Mo and Schmugge[26] were able to reproduce the scatter in microwave ob-servations of soil moisture in South Dakota solely throughintroducing error into the soil moisture measurement (witha coefficient of variation of 0.25) though a Monte Carlosimulation technique. Both studies illustrate the uncer-tainty in assessing the accuracy of remote sensing tech-niques with "ground truth" that is inherently noisy. Al-though an L-band radiometer has a sensitivity to surfacesoil moisture of 2-3K per percent volumetric soil moisturefor minimally vegetated fields 127], [28], differences in

SCHMUGGE e( ai.: PASSIVE MICROWAVE SOIL MOISTURE RESEARCH 17

300

Fig. 4. L-hand aircraft results over the Hand County. South Dakota. testsite. The data were segregated into 3 qualitatively ditferent roughnesscategories hased on ground photographs.

200 30.0

SOIL MOISTURE CONTENT

50040_0

•••

CJ

ONE INCH HORIZON

S~~"R,\TION BY SURf-ACE ROUGHNESS

•

o ~fDIUM

o HIGH

••• LOW

• LOW Y D!I%Ol,lEOIUMy 2601. + 28149 1921O'''IIGH y~ 2Hf, •• nOJ3 0952

00

200

N

II~ 260

w0:=>I-..ffiCo

~<nif]zI-r

"ii'co 220

w>..~o0:(),.

B. RoughnessAlthough it is clear from the previous discussion that

microwave radiometers can estimate soil moisture contentfor smooth, bare fields, the presence of variations in sur-face cover conditions such as roughness and vegetationcan greatly reduce microwave sensitivity to soil moistureand introduce scatter into the emissivity/soil moisture re-lationship [27]. Since very few agriculturally or hydro-logically important areas are always smooth and bare, it isimperative to develop techniques for removing the effectsof roughness and vegetation from microwave estimates ofsoil moisture, preferably in a manner compatible with aremote sensing approach.

Choudhury et al. [131 determined that the rough sur-face reflectivity could be related to the smooth surfacereflectivity (Ra) in the following way:

R = Ro exp (-h cos2 0) (6)

soil properties such as soil type, bulk density, and surfaceroughness over the larger areas typical of aircraft and sat-e~~ite systems may permit quantification of only 4-5 levelsof soil moisture. Relative changes in moisture content forthe same site over time from repeat microwave measure-ments should have a better resolution.

where the h parameter is related to height variations of thesoil surface. The effect of increasing h values (i.e., rough-nesses) on the passive microwave response to soil moistureis to decrease the sensitivity of the microwave emissivityto the moisture variations by the "exp (~h)" factor in thesame fashion as the vegetation in (5). Jackson et al. [29Jused representative values of h based on the observedrange of bulk density and soil moisture in their test fieldsand were able to reduce the standard error of their soilmoisture estimate by a few percent. While this procedureis an improvement over ignoring the effects of surfaceroughness, it is of limited usefulness in an operationalsense.

Theis et at. [30] used one possible approach in analyz-ing their aircraft data, which contained 1.6-GHz scatter-ometer data taken coincidentally with 1.4-GHz radiometerdata. Noting that the influence of surface roughness wassubstantial at large incidence angles while the sensitivityof the radar backscatter (J0) to soil moisture remainedabout the same, the researchers plotted (J0 versus volu-metric soil moisture at a 40° look angle and determinedthe (J0 intercept for a O-percent soil moisture level. Theythen used this intercept as a "quantification" of the degreeof roughness in combination with the 1.4-GHz emissivityto estimate soil moisture directly. With the roughnesscompensation supplied by the L-band scatterometer, theywere able to improve the R2 value to 0.95, compared to R2

= 0.69 obtained when predicting soil moisture from thepassive data alone. Although conducted on a limited dataset, the encouraging results from this study clearly pointout the potential of multisensor techniques for improvingmicrowave estimates of soil moisture.

A comparison of the different experiments raises a ques-

tion about the actual magnitude of roughness effects undernormal agricultural conditions. Owe and Schmugge [191in the analysis of 1.4-GHz aircraft data obtained over anagricultural site found that on the average the roughnesseffects were not large (Fig. 4). They segregated the datainto three roughness categories based on ground photo-graphs of the individual fields and found that the regres-sion slopes for TB versus surface soil moisture were notsignificantly altered by the roughness but that the roughestfields were about 8K higher than the smoothest. Thesefields covered the range of roughnesses that might nor-mally occur in an area which is mostly planted in smallgrains or in pastures. These results point out that whileroughness can have a very significant effect, its naturallyoccuring range may not be as great as the extremes ob-served in prepared plots for truck measurements.

C. VegetationThe dominant factor affecting the accurate interpretation

of microwave data in terms of soil moisture information isthe presence of vegetation. A vegetation canopy over asoil volume attenuates the emission of the soil and adds tothe total radiative flux from the scene with its own emis-sion (assuming that scattering is minimal at low frequen-cies). As frequency and incidence angle increase, the mi-crowave sensitivity to the soil moisture underneath thevegetation is greatly reduced. depending on the canopytype and water content. Ulaby et at. [311 found that thevegetation loss factor was twice as great at 5 GHz than at1.4 GHz using aircraft radiometer data from Colby, Kan-sas; this result was confirmed by the flights over nativegrass pastures at Chickasha as shown in Fig. 5 r 18]. Theiset al. [20] noted that the effect of vegetation was much

18IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. GE-24 NO. I, JANUARY 1986

Fig. 5. Comparison of L- and C-band emissivities measured from an air-craft platform with ground measurcmcnts of sDil moisture for native grasspastures in Oklahoma 1181. Note the 50-percent decrease in sensitivityat C-band.

more significant on the microwave response to soil mois-ture than surface roughness, especially as the plant wetbiomass increased to the level of dcnse corn,

In an attempt to quantify the effect of vegetation on ra-diometric measurements of soil moisture and thus extendthe usefulness of the microwave approach to a wider rangeof agricultural conditions, Theis et ai, [20] calculated theperpendicular vegetation index (PYI) from visible/near-in-frared data acquired at the same time as the microwavedata, The PYI, which is the perpendicular distance ofmeasured values of reflectance from the bare soil reflec-tion line on a plot of MSS band S (0.6-0,7 j.tm) versus MSSband 7 (0,8-1.1 j.tm), is directly related to plant biomassin that higher PYI values correspond to higher biomass132]. They then used the PYI in combination with lA-GHz emissivity values to estimate surface soil moisturedirectly for every test field at Guymon and Dalhart exceptfor dense corn (PYI > 4.3). With only remotely sensedinformation, they were able to improve their estimates ofthe 0-2-cm soil moisture from R2 = 0.09 to R2 = 0.7S forvegetated fields. Jackson et al. [IS] also utilized a com-bination of emissivity and wet biomass values to predictsoil moisture under a variety of vegetation layers. Al-though the actual levels of biomass were known in thiscontrolled study, the investigators expressed their beliefthat the same Information could be obtained through vis-ible/infrared remote sensing. Since vegetation in a givenarea does not change as fast as surface soil moisture, somerepeat reflectance data could be used in conjunction withmore frequent passive microwave measurements to esti-mate soil moisture through vegetation. Although thetimeiy acquisition and registration of different types of datamight present some operational challenges, the multisen-sor technique appears to be significantly better for soilmoisture determination over a range of surface conditionsthan the use of a single sensor alone.

Ulaby et al. [33] arrived at a similar conclusion afterexamining active and passive microwave data from theColby experiment. They found that the 1.4-GHz radiome-ter produces lower estimation errors at low soil moisturelevels, while the 4.75-GHz scatterometer is more accurateat soil moistures greater then 70 percent of field capacity.

VOLUMETRIC SOIL MOISTUREQ·5CM (%)

wa:::>"'g O,8;~

J ~'0:"'~ 0.7 • -1978 DATAo c 1980DATA

Z l.' HIGH ALTITUDE DATA

Q.6 .6-S1015-rti2s3o35t04s-5o

IV. SPACECRAFT RESULTS

Taking advantage of the complementary nature of the twosystems, a combination of a radiometer and a radar willreduce soil moisture prediction errors to less than ± 30percent of true percent of field capacity, even under a lossycorn canopy.

In analyzing aircraft radiometer and radar data in thecontext of simple vegetation canopy emission and scatter-ing models, vegetation loss factors (which describe theamount of soil signal attentuation) were found to be greaterfor the passive data than the active at a given wavelength[33]. This is primarily due to radiation (scattered or emit-ted) from the vegetation which is reflected by the soil sur-face. As soil moisture increases, the amount of energyreflected from the surface increases. Thus, in the passivecase the increased reflectivity counteracts the decrease inemission for wet soils [341.

Most measurements by orbiting microwave radiometersfor soil moisture estimation were made at frequenciesgreater than S GHz, with a short Skylab mission with a1.4-GHz radiometer during the sLImmer of 1973 being theonly exception. Results from these measurements havebeen reported by Schmugge et al. [3S], Blanchard et al.[2]), Eag]eman and Lin [36], Allison et al. [37], andWang [38]. The work of Allison et al. was based onESMR-S (l9.3-GHz Electronically Scanning MicrowaveRadiometer on board Nimbus-S satellite) observations ofthe great eastern Australian floods during January-March1974. Because of the frequent, wide-swath coverage ofESMR-S, they were able to observe the development andsubsidence of the flooded region day and night over theentire two-month period. The studies by Schmugge et al.and Blanchard et al. were also based on the ESMR-S ob-servations, while those by Eagleman and Lin used theSkylab 1.4--GHz radiometer measurements. Since soilmoisture ground truth was difficult to acquire in large-area measurements, the Antecedent Precipitation Index(API) was used and determined to be a good indicator ofsoil moisture content [2]]. It was found that over a regionwith sparse vegetation cover like western Texas and Okla-homa, microwave emission strongly correlated with API[2]), [36]. Wang [38] has extended the analyses to ex-amine the effect of vegetation cover over a large scale.

In contrast to studies of the vegetation effect from truckand aircraft platforms, large-area observations and anal-yses of this effect with satellite sensors are quite limited.Eagleman and Lin [36] have pointed out the microwavesignature of vegetation cover in their analyses of the Sky-lab ] A-GHz radiometer data. More recently, Wang [38]has used the data obtained from both the Skylab radiome-ter and the two lowest frequency channels of the ScanningMultichannel Microwave Radiometer (SMMR) to studythe vegetation effect in more detail. Two areas, Region Ain western Texas and Region B in eastern Texas and Okla-homa, were chosen for the analyses. The biomass distri-bution of vegetation cover in Texas has been analysed by

o000

oo

ooo 0

e 0.95 Q,Q031SM

R' 0.77

,\ Gem1 01

VOLUMETRIC SOil MOISTUREQ-2.5CM ('.\;)

O'9f; 08

:J~ I'U-~ 07-

e l.-1978 DATA2:)-1980 DATA

o 6 ------'-----L~_-----L ___._L-------L-_L L _____L____J. 0 5101520253035404550

00067SM

0'·cC" ••

e 096

Rt 0.95

.\ 21 em

oo0'o

oo

1 C

0.9

SCHMUGGE et al.: PASSIVE MICROWAVE SOIL MOISTURE RESEARCH 19

o REGION A

• REGION B

(b)

'/L__ T~ 2/99 841APO

r2 06~N 127

" -471<!'055CMI J. 1- 1. .. I

2 4 6 B

(a)

l!' •.•.•..•~:it-~'" .•..•~~ ..• -

0° 000,: •.•. --;--- ......•......•..•0°0 r$! •••••-_

0° t \-00 \

TB·27S.4 1.BIAPIl ~r2 042N 63

:'5 I( 1- 1 9 CM)

24J230Lo

2BO q,.. •

mf-UIa:::Jf-«a::UI0-::;; 260UIf-if)if)UI

r= 250:r:'-"a'en

ANTECEDENT PRECIPITATION INDEX, CM

290

Fig. 7. Comparison of microwave brightness temperatures measured by theNimbus-7 SMMR instrument with API for the same two regions of Texasfrom June to September 1979; (a) 6.6-GHz results, (b) 1O.7-GHz results.

230 I0 REGIO'

~GIO~ B

2100 2 4 6 8

ANTECEDENT PRECIPITATION INDEX, CM

""oJf-UJa::::Jf-<l:a::UJa..:2:UJ

f- 250CflCflUJZf-I!:2a::a:l

290

SKYLAB RADIOMETRIC MEASUREMENTS OVER TEXASJUNE - SEPT. 1973

Fig. 6. Comparison of microwave brightness temperatures measured by theSkylab 1.4-GHz radiometer with the antecedent precipitation index (API)for two regions in Texas [38].

Newton et al. [8] using Landsat imagery and by Greegorand Norwine [39] using NOAA satellite imagery. Both ofthese studies have shown a gradual increase in vegetationbiomass from western to eastern Texas. Therefore, themicrowave signatures from regions A and B are expectedto be different. Both Sky]ab radiometer and SMMR mea-surements with ground resolution cells falling in these tworegions were used to correlate with API, which was cal-culated from the daily average total precipitation in cmfrom the weather stations falling within the resolution cellof each radiometric measurement.

Fig. 6 shows a scatter plot between the TB' s measuredfrom the Skylab 1.4-GHz radiometer and the correspond-ing API's for both regions A and B. Applying linearregressions to each of the two data groups results in r-squared values of 0.75 and 0.59 for regions A and B, re-spectively. Two main features of interest are observed inthis figure. First, the radiometric measurements over re-gion B with dense vegetation cover result in a shallowerslope compared to thzt of region A with sparse vegetationcover. This is in agreement with the results of measure-ments at ground level and aircraft altitudes described ear-lier. Secondly, the large API values of about 6 cm are theresults of heavy rainfalls on the days immediately beforethe microwave radiometric measurements. The moisturecontent of surface soils under this condition is generallyat field capacity. The radiometric measurements over suchsoils should give TB's of about 190-200K when there isno vegetation cover [12]. The extension of the region Aregression line to API = 6 cm in the figure gives a TB ofabout 210K. This is in a reasonable agreement with thevalues obtained from either the ground level or aircraftexperiments considering the fact that some areas withinthe ground resolution cell of the Skylab radiometer arecovered with vegetation. A vegetation-covered soil is gen-erally associated with a higher TB•

Fig. 7(a) and (b) show the scatter plot between TB's andAPI's for both the 6.6- and 1O.7-GHzSMMR frequencies.The microwave radiometric responses over the two re-gions are quite different at both frequencies. Linearregressions applied to the four data groups correspondingto two regions and two frequencies give the correlationcoefficients and regression slopes indicated in the figure.Clearly, the slopes derived from region B are much shal-lower compared to those from region A, showing the effectof vegetation. The presence of this vegetation effect intro-duces ambiguity in soil moisture determination. For a TB

measurement from either Fig. 6 or Fig. 7, the estimatedAPI value would depend on whether the land within theradiometer resolution cell is covered with vegetation ornot. To overcome this problem, it is necessary to havesome recent estimate on the biomass of vegetation cover[15] such as the approach used by Theis et al. [20] asdescribed earl ier.

Radiometric measurements generally can be made to anaccuracy of ± 3K or less [II]. The large scatter of the datapoints in Fig. 7 is largely caused by the inhomogeneity ofsurface cover and rainfall distribution within the resolu-tion cells of the sensors. The statistics of the measure-ments in both figures are characterized by the standarderrors of estimates (SEE) derived from the regressionanalyses using either TB or API as independent variablesin each data group. At 1.4-GHz, the derived SEE's of APIare 0.65 cm and 1.03 cm for regions A and B, respectively.For 6.6- and 1O.7-GHz measurements, the derived SEE'sare 0.56 and 0.55 cm, respectively, for region A. For regionB, the estimated SEE's are 1.90 cm at both frequencies.These results suggest that for measurements over sparselyvegetated terrain like region A, it is possible to determineat least 5 moisture levels within the API range of 0-6 cm.For measurements over terrain with dense vegetation coverlike region B, it is still possible to estimate 4-5 moisturelevels at the l.4-GHz frequency. At 6.6- or 1O.7-GHz fre-

20 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SEKSING. VOL. GI:-24 NO. I. JANUARY 1986

quencies, however, it will be difficult to determine morethan 3 moisture levels.

V. MODELING REsuus

A. Roughness

Experimentally, the effects of surface roughness on theradiometric response to soil moisture were observed asearly as 1974 in a field experiment done at Texas A&MUniversity 142J. They observed a significant increase inemissivity with increasing surface roughness and thus adecrease in the sensitivity to soil moisture. This effect wasalso observed in field measurements at BARC, althoughdata from the aircraft experiments indicate that the impactof roughness arising from normal agricultural practicesmay be less severe.

Attempts to model the effects of surface roughness havebeen slow in coming. As mentioned earlier, Choudhury etal. 113] related rough and smooth surface microwave re-flectivities by an exponential factor h. Using this expres-sion, they found that they could fit the observed data withan empirical value for h, but that h did not scale properlywith wavelength or the rms surface height variations. Thisdiscrepancy occurred because the model did not considerthe incohercnt part of the scattered field which depends onthe horizontal scale of the surface height variations. Work-ing independently, Tsang and Newton [431 and Fung andEom [44] developed models which include both the finssurface height and the horizontal correlation length (I).Both groups found that their f{)rmulations can give goodagreement with both active and passive microwave mea-surements.

B. Sensitivity Analysis

Based on the model of Mo et al. [16], Jack Paris at the1982 Soil Moisture Project Review presented an analysisof the effects that vegetation and surface roughness haveon the radiometric sensitivity to soil moisture. From (5)and (6) we have

tl.NTB/tl.SM = (tl.NTBlLiSMharc exp (-h) exp (-27)

= (LiNTBI LiSM)barc exp (- h - 27)

where the emissivity is expressed in terms of NTB forconvenience in comparing to radiometer noise levels. Ona h versus 7 plot the locus of points having equal sensitiv-ity reduction will be a straight line of the form h + 2T =constant. This result is illustrated in Fig. 8 showing therange of sensitivity reduction which can be expected as afunction of hand 7. The range of h values observed byChoudhury et al. [131 in their analysis of both field andaircraft data was from 0 to 0.6. Therefore. even for theroughest fields it should still be possible to sense moisturevariations through optical thicknesses up to 0.5, at whichlevel the sensitivity is reduced to 20 percent of the barefield value. Fortunately, for vegetated fields surface rough-ness tends to be reduced as the sharp edges are erodedwith time so that heavier vegetation canopies (up to 0.7)can be penetrated. At this reduction level the range of NTB

REDUCTION FACTOR1.0

ex;'::':r! 0.6<l:a:<l:0.UJ

t2 0.4zICO::>oa: 0.2

~2 OA ~6 ~8 1~VEGETATIVE OPTICAL THICKNESS. T

Fig. 8. Nomograph or the sensitivity reduction ractor as a runction or theroughness parameter (h) and the vegetative optical thickness (7). Thisractor when multiplied by the bare field sensitivity yields the reducedsensitivity under the conditions II and T.

from wet to dry is about 20K which is the range of bright-ness temperature that was observed by the Kansas groupin aircraft data over a corn field. This level of sensitivitywould be adequate if there were no other sources of un-certainty present; however. with additional uncertaintiesin soil temperature and soil texture, it may only be pos-sible to distinguish between wet and dry conditions undervery dense vegetation.

C. Other Research

In this paper we have only presented model ing resultswhich are directly related to some of the measurementprograms. There have been other theoretical activitieswhich are of a more fundamental and esoteric nature.These include the work of Tsang et al. 1451 showing thecorrespondence between active and passive microwave ob-servations and the work of Kong et al. 146] on the emis-sion from furrowed fields.

These modeling activities are important because theyprovide the physical framework for interpreting the exper-imental results and for extending these interpretations toa wider range of observational conditions. As noted ear-lier, this modeling involves the extensive use of empiricalparameters because of the ditliculty in determining therelevant quantities from first principles or direct measure-ments, e.g., the scattering of microwave radiation fromvegetation or from rough soil surfaces. In dealing withnatural systems such as soils and plants, we arc con-fronted with a high degree of naturally occurring varia-bility which makes the sampling of these quantities diffi-cult. Thus, the use of empirical parameters based onremotely sensed observations which tend to average outthis variability may be the most appropriate approach totake.

VI. CONCLUSIONS

Significant progress was made during the AgRISTARSSoil Moisture Project in quantifying the capabilities of mi-crowave sensors for the remote sensing of soil moisture.

SCHMUGGE et ai.: PASSIVE MICROWAVE SOIL MOISTURE RESEARCH

The use of the 21-cm wavelength as the best single channelfor radiometric observations of soil moisture was verified.There were no indications that radiometric observations atmultiple microwave wavelengths yield significant addi-tional information about the moisture conditions in the soilvolume beyond that obtained from a single wavelengthalone. In fact, other remote sensing approaches used inconjunction with L-band passive data are more successfulthan multiple wavelength microwave radiometry in provid-ing accurate soil moisture information over a wider rangeof conditions, e.g., the use of visible/near-infrared forvegetation estimates and active microwave for roughnessestimates.

In addition to examining combinations of different sen-sors for soil moisture research, the studies conducted un-der the AgRISTARS Program also improved our under-standing of the various noise factors which affect theinterpretability of microwave emission data. Researchersverified experimentally the magnitude of the soil moisturesampling depth, while surface roughness was explainedtheoretically in terms of two measurable parameters, thesurface height variance and the horizontal correlationlength. Absorption of soil emission by vegetation wasquantified, and although scattering of energy within veg-etation canopies was observed in some instances [47], thiseffect is less important than absorption effects for micro-wave radiometry. Our knowledge of the impact of soil tex-ture variations became more refined [3], with more recentwork [48 J confirming the importance of soil propertiessuch as density and soil structure. Based on a considera-tion of these results, we feel that it should be possible tomeasure the soil moisture of the surface layer (0-5 cm) toan accuracy of ± 5-percent absolute about 90 percent ofthe time where vegetation permits, the major difficultybeing when the soil surface has just been worked and isextremely rough and of low density.

It is important to realize that while remote sensing mea-surements will not provide as accurate or as deep a mea-surement of soil moisture as can be obtained by conven-tional in situ measurements, they do provide a means forgetting repetitive measurements over large areas of themoisture condition of the surface soil layer. This type ofinformation has not been conventionally available in thepast, and thus, a major task for the near future is the dem-onstration of the utility of surface soil moisture in deter-mining such things as the partitioning of energy at the landsurface and its effect on surface runoff. An opportunityfor this will be the field experiments planned in the Inter-national Satellite Land Surface Climatology Program(ISLSCP). Successful demonstration of the utility of mi-crowave remotely sensed soil moisture information in theseexperiments will be necessary before passive sensors re-quiring large antennae are flown in space as part of a pro-posed space platform.

REFERENCES

III J. R. Wang and T. J. Schmugge. "An empiricalmodcl for the complexdielectric permittivity of soils as a function of water contcnt."' IEEETrans. Ceosei. Remote Sensin/;. vol. GE-18, pp, 288-295. 1980.

2]

(2] M. T. Hallikainen, F. T. Ulaby. M. C. Dobson, M. A. EI-Rayes. andL. Wu, "Microwavc dielectric behavior of wet soil-Part I: Empiricalmodels and experimental observations." IEEE Trans. Ceosei. RemoteSensin/;, vol. GE-23. pp. 25-34. 1985.

[3] M. C. Dobson, F. T. Ulaby, M. T. Hallikainen and M. A. EI-Rayes."Microwave dielectric behavior of wet soil-Part II: Dielectric mixingmodels," IEEE 7i'ans. Ceosei. Remote Sensin/;. vol. GE-23, pp. 35-46. 1985.

[4] M. C. Dobson and F. T. Ulaby, "Active microwave soil moistureresearch." IEEE Trans. Ceosei. Remote Sensing, this issue, pp. 23-36.

(5] T. T. Wilheit. "Radiative transfer in a plane stratified dielectric." IEEETrans. Ceosei. Electron .. vol. GE-16, pp. 138-143, 1978.

16] E. G. Njoku and J. A. Kong. "Theory for passive microwave sensingof near-surface soil moisture." 1. Ceophys. Res .• vol. 82, pp. 3108-3118.1977.

17] T. J. Sehmugge. "Soil moisturc sensing with microwave techniques,"in Proc. 14th Int. Symp. Remote Sensing Environ. (San Jose. CostaRica, Apr. 1980, 23-30). pp. 487-505,

18] R. W. Newton. B. V. Clark. J. F. Paris, and W. M. Pritchard. "Or-biting passive microwave sensor simulation applied to soil moistureestimation," Texas A&M Remote Sensing Center, Final Rep. RSC-3753 on NASA Grant NSG-5266. p. 208, 1980.

191 R. W. Newton. Q. R. Black, S. Makanvand. A. J, Blanchard, and B.R. Jean, "Soil moisture inf(lrmation and thermal microwave emis-sion." IEEE 7i'ans. Ceosei. Remote Sensin/;, vol. GE-20. pp. 275-281. 1982.

110] J. R. Wang . .I. E. McMUl1rey, E. T. Engman. T. J. Jackson, and T. J,Sehmugge. "Radiometric measurements over bare and vegetated fieldsat 1.4 GHz and 5 GHz frequencies, "Remote Sensing Environ., vol.12, pp. 295-311, 1982.

III] J. R. Wang, T. .I. Schmugge. J. E, McMurtrey. W. I. Gould. W. S.Glazar. and.l. E. Fuchs, "A multifrequency radiometric measurementof soil moisture content over bare and vegetated fields," Ceophys.Res. Letl., vol. 9, pp. 416-419, 1982.

112] J. R. Wang. P. E. O·Neill. T..I, Jackson, and E. T. Engman, "Mul-tifrequency measurements of the ell"ects of soil moisture. soil textureand surface roughness," IEI-."!,·TrW1S. Ceosei. Remote Sensin/;, vol.GE-21. pp. 44-50. 1983.

113] B . .I. Choudhury, T. J. Schmugge. R. W. Newton. and A. Chang,"Etrect of surface roughness on the microwave emission from soils."1. Ceophys. Res., vol. 84, pp. 5699-5706. 1979.

114] D. R. Brunfcldt and F. T. Ulaby. "Measured microwave emission andscattering in vegetation canopies." IEEE Iiw/s. Ceosei. Remote Sens-ing, vol. GE-22. pp. 520-524. 1984.

115] T. .I. Jackson. T. .I. Schmugge, and .I. R. Wang, "Passive microwavesensing of soil moisture under vegetation canopies," Water Res. Res.,vol. 18. pp. 1137-1142, 1982.

116] T. Mo. B. J. Choudhury. T. .I. Schmugge, .I. R. Wang, and T. .I. Jack-son, "A modcl Itlr microwave emission from vegetation-coveredfields," 1. Ceophvs. Res., vol. 87. pp. 11229-11237. 1982.

1171 E. G. Njoku and P. E. O'Neill, "Multifrequency microwave radiome-ter measurements of soil moisturc." IEEE Trans. Ceosei, RemoteSensing, vol. GE-20. no. 4. pp. 468-475, 1982.

118] T. .I. Jackson, T..I. Sehmugge, and P. E. O'Ncill, "Passive microwaveremote sensing of soil moisturc from an aircraft platform," RemoteSensing Environ., vol. 14, pp. 135-151. 1984.

1191 M. Owe and T. J. Schmugge, "Microwave radiometer response to soilmoisture at thc 21 cm wavelcngth." presented at the Amer. Soc.Agronomy, Annual Mecting (Washington. DC). 1983.

1201 S. W. Theis, B. J. Blanchard, and R. W, Newton, "Utilization of veg-etation indices to improve microwave soil moisture estimates over ag-ricultural lands." IEEF Ii·ans. Ceosei. Remote Sensing, vol. GE-22,pp. 490-496. 1984.

(211 B. J. Blanchard, M. J. McFarland. T. J. Schmugge, and E. Rhoades,"Estimation of soil moisture with API algorithms and microwaveemission," Water Res. Bull .. vol. 17. pp. 767-774. 1981.

1221 T. J. Jackson, "Profilc soil moisture from surface measurcments," 1.Irrigation Drainage Di\' .• ASCE, vol. IR-2 pp. 81-92, 1980.

123J Q. R. Black and R. W. Newton. "Airborne microwave remote sensingof soil moisture." AgRISTARS Tech. rep. SM-T1-40457, 1981.

1241 M. Owe. E. B. Jones, and T. .I. Schmugge, "Soil moisture variationpatterns obscrved in Hand County, South Dakota," Water Res. Bail.,vol. 18. pp. 949-954, 1982.

(251 H. K. Burke and .I. H. Ho. "Analysis of soil moisture extraction al-gorithm using data from aircral"t experimcnts." ERT Inc .. Concord.MA, Final Rep. Cont. NAS5-26361, Doc. P-A826. 1981.

126] T. Mo and T. .I. Sehmugge. "Monte Carlo simulation of the efrect of

---------------------------------,22 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSlNG, VOL GE-24 NO, I. JANUARY 1986

•.

soil moisture variation on the microwave emission from soils," IEt't'Trans, Ceosei, Remote Sensing, vol. GE-21, pp, 473-479, 1983,

[27] T. Schmugge, "Remote sensing of soil moisturc with microwavc ra-diometers," Trans, ASAE, vol. 26, pp. 748-753, ]983.

[28J P. E. O'Ncill, "Microwave remote sensing of soil moisture: A com-parison of results from different truck and aircraft plathmns." Int. 1.Remote Sensing, vol. 6, no. 7, pp, 1125-]134, 1985.

[29] T. 1. Jackson, P. O'Neill, J. Wang, and J. Shiue, "Evaluation of apushbroom microwave radiometer aircraft soil moisture remote sens-ing system," paper 84-25]6, presented at the ASAE (Ncw Orleans,LA), Dec. ] 1-14, ]984.

[30] S. W. Theis, B. J. Blanchard, and A. J. Blanchard. "Utilization ofactive microwave roughness measurements to improve passive micro-wave soil moisture estimates over bare soils," in Pme. ICARSS'84(Strasbourg, France), Aug. 27-30, ]984.

[31] F. T. Ulaby, M. Razani, and M. C. Dobson, "Effects of vegetationcover on the microwave radiometric sensitivity to soil moisture," IEEETrans. Ceosei. Remote Sensin[;, vol. GE-2], pp. 51-6], ]983.

[32] A. J, Richardson and C. L. Wiegand, "Distinguishing vegetation fromsoil background inl()rmation," PhOlO[;rammetr. En[;., vol. 48, pp.]54]-1552, ]977.

[33J F. T. Ulaby, M. C. Dobson, and D. R, Brunfeldt, "Improvement ofmoisture estimation accuracy of vegetation-covered soil by combinedactive/passive microwave remote sensing," IEEfi Trans. Ceosei. Re-mote Sensing, vol. GE-21, pp. 300-307, 1983.

[34] A. K, Fung and H. J, Eom, "A comparison between active and passivesensing of soil moisture from vegetated terrains," IEEE Trans. Ceose;,Remote Sensing, vol. GE-23, pp, 768-775, ]985.

[3SJ T. J. Sehmugge, J. M. Meneely, A. Rango, and R. Neff, "Satellitemicrowave observations of soil moisture variations," Water Res. Bull.,v~. ]3, p~ 265-281, ]97~

[36] J. R. Eagleman and W. C. Lin, "Remote sensing of soil moisture bya 21 cm passive radiometer," J. Ceophvs, Res., vol. 81, pp. 3660-3666, 1976.

[37] L. J. Allison, T. J. Schmugge, and G. Byrne, "A hydrological analysisof east Australian floods using Nimbus-S electrically scanning radi-ometer data," Bull. Amer, Met. Soe.. vol. 60, pp. 1414-1427. 1979.

138J J. R. Wang, "The effect of vegetation on soil moisture sensing ob-served from orbiting microwave radiometers," Remote Sensin[; fin-viron., vol. II, pp. ]4]-]51, ]985.

[39] D. H. Greegor and J. Norwine, "A gradient model of vegetation andclimate utilizing NOAA satellite imagery," NASA Johnson SpaceFlight Center, Houston, TX, AgRISTARS FC-Jl-04176, JSC-17435.1981.

[40J C. J. Tuckcr, J. H. Elgin, and J. E. McMurtrey, "Relationship of cropirradiance to alfalfa agronomic values." Int. 1. Remote Sensing, vol.8, pp. 69-76, 1980.

[41J C. J. Tucker, J. R. G. Townshend. and T. E. Goff, "African land-cover classification using satellite data." Science, vol. 227, no. 4685,pp. 369-375, 1985.

[42] R. W, Newton and J. Rouse, "Microwave radiometer measurementsof moisture content," IEEE Trans. Antennas Prof'!Iga/ .. vol. AP-28,pp. 680-686, ]980.

[43] L. Tsang and R, W. Newton, "Microwave emission from soils withrough surfaces," 1. Ceof'hys. Res., vol. 87, pp. 9017-9024, 1982.

[44] A. K. Fung and H, J. Eom, "An approximate model for backscatter-ing and emission from land and sea." Remote Sensing Lab., Univ. ofKansas, AgRISTARS Rep. SM-K ]-04049, 1981.

[45] L. Tsang, A. J. Blanchard, R. W. Newton, andJ. A. Kong, "A simplerelation bctween active and passive microwave remote sensing rnea-surements of earth terrain," ItEE TI'Q/1.\'. Ceosci. Remo/e Sensing,vol. GE-20, pp, 482-484, 1982.

[46J J. A. Kong, S. L. Lin, and S. L. Chuang, "Microwave thermal emis-sion from periodic surfaces," IEEE Trans. Ceosci. Remote Sensing.vol. GE-22. pp. 377-382, 1984.

[47] P. E. O'Neill, T. J. Jackson, B. J. Blanchard, J. R. Wang, and W. I.Gould, "Effects of corn stalk orientation and waler content on passive

microwave sensing of soil moisture," Remote Sens. Environ., vol. 16.pp. 55-67, 1984.

[48J T. J. Jackson, P. E, O'Neill, G. van der KoltL G. J. Koopman, andJ. R. Wang, "Effects of soil properties on the microwave emission ofsoils," Pres. 4th lilt. Symp. Remote Sensing Soil Surv".\', Int. Soe.Soil Sci, (Wageningen and Enschede, The Netherlands. March 4-8,1985).

*

Thomas Schmugge (M'83) received the B.S. de-gree in physics from the llIinois Institute of Tech-nology in ]959 and the Ph.D. degree in physicsfrom the University of Calif()rnia. Berkeley, in]965.

From ]964 to 1970, he was a member of thePhysics Department at Trinity College, Hartford,CT. He began his work on the use of microwavetechniques f()r remote sensing in 1970 as a Na-tional Academy of Sciences Senior Research As-sociate at the NASA Goddard Space Flight Cen-

ter, Greenbelt, MD. Since that time, he has been at Goddard workingprimarily on the application of passive microwave techniques f(Jr the remotesensing of soil moisture and snow. He is also involved in the developmentof techniques for using remotely sensed data to estimate areal averages ofthe evapotranspiration flux. He was the Project Scientist fin the Soil Mois-ture Project in the AgRISTARS Program. Currently, he is a Senior Scientistin the Hydrological Sciences Branch at Goddard and is the Project Scientistlor the First 1SLSCP Field Experiment that is part of the International Sat-ellite Land Surface Climatology Project.

*

Peggy E. O'Neill received the B.S. degree in ge-ography from Northern llIinois University in ]976and the M.A. degree in geography from the Uni-versity of California, Santa Barbara, in 1979.

Since 1980. she has been employed as a Physi-cal Scientist in the Hydrological Sciences Branchat the NASA Goddard Space Flight Center.Greenhelt. MD. Her major areas of research in-clude analysis of microwave remote-sensing tech-niques for the determination of soil and vegetationpropcrties.

Ms. O'Neill i,; a member of the American Water Resourccs Association.the American Geophysical Union, and the Remote Sensing Society.

*

James R. Wang received the B.S. degree in physics from the University ofWashington, Seattle, and the M.S. and Ph.D. degrees in physics from theUniversity of Chicago, Chicago, IlL

Prior to joining the NASA Goddard Space Flight Center, Greenbelt, MD,in 1977, he was employed by Columbia University, New York, NY; Com-puter Sciences Corporation, Silver Spring, MD; and Lockheed ElectronicsCompany, Houston, TX, His research interests inelude the applications ofboth active and passive techniques for the remote sensing of geophysicalparameters such as soil moisture, water vapor, and precipitation.

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. GE·24 NO. J, JANUARY 1986

Active Microwave Soil Moisture ResearchM, CRAIG DOBSON, MEMBER, IEEE, AND FAWWAZ T, ULABY, FELLOW, IEEE

23

Abstract-This paper summarizes the progress achieved in the activemicrowave remote sensing of soil moisture during the four years of theAgRISTARS program. Within that time period, from about 1980 to1984, significant progress was made toward understanding 1) the fun-damental dielectric properties of moist soils, 2) the influence of surfaceboundary conditions, and 3) the effects of intervening vegetation can-opies. In addition, several simulation and image-analysis studies haveidentified potentially powerful approaches to implementing empiricalresults over large areas on a repetitive basis. This paper briefly de-scribes the results of laboratory, truck-based, airborne, and orbitalexperimentation and observations.

I. INTRODUCTION

THE OBJECTIVE of the AgRISTARS soil moistureproject was to develop and evaluate the technology to

make both remote and ground measurements of soil mois-ture, The attainment of this objective was viewed as a pre-cursor to using soil-moisture information in applicationmodels for predicting crop yield, plant stress, and wa-tershed runoff. This paper summarizes the advances madeduring the AgRISTARS program with respect to the sens-ing of soil moisture using active microwave techniques; acompanion paper in this issue deals with passive micro-wave techniques [I].

Prior to the AgRISTARS program, the capability of ac-tive microwave techniques to sense near-surface soil mois-ture had been, for some years, an area of considerable re-search interest. A number of field experiments had beenconducted, most of which used truck-mounted FM-CWscatterometers. The systems had been used as spectrom-eters over the 1- to l8-GHz frequency band in order toinvestigate the spectral properties of radar response tofirst-order soil properties such as soil moisture, and to therandom component of soil roughness induced by agricul-tural tillage practices [2]-[4]. These efforts identified ra-dar sensitivity to near-surface soil moisture as a functionof surface roughness and soil texture for various combi-nations of the radar sensor parameters of frequency, po-larization, and angle of incidence with respect to nadir[5], [6]. In general, the studies concentrated on nonve-getated soil surfaces and approached the problem from ananalytical viewpoint, i.e., as an optimization problem,with the objective of identifying sensor parameters havingmaximal sensitivity to and correlation with near-surfacesoil moisture but also having minimal sensitivity to sur-

ManuscriptreceivedJuly 8. 1985. revisedSeptember12, 1985.Theauthorsare with theRadiationLaboratory,DepartmentofElectrical

Engineeringand ComputerScience, Universityof Michigan,Ann Arbor,MI 48109.

IEEE LogNumber8406230.

face roughness and agricultural canopy cover, The result-ing recommendations for a C-band radar (at about 5 GHz)operating at angles of incidence in the 100 to 200 rangehave not been substantively altered by the findings of sub-sequent investigations.

Research undertaken as part of the AgRISTARS pro-gram was directed at verifying these preliminary findingsand extending them via a parametric analysis of each ofthe scene variables expected to affect the radar backs cat-tering from an agricultural setting. The scene variablesexamined include soil-moisture profile and samplingdepth, soil bulk density, soil surface boundary conditions(such as random surface roughness, row direction effectsrelated to ridge/furrow tillage practices, and local slopeas related to local angle of incidence), vegetation cano-pies, and geographic conditions (such as variability in lo-cal topography, soil texture, field size and shape, and thepresence of nonagricultural features such as urban areas,forests, and water bodies). The preceding variables wereexamined (sometimes not definitively) through a series oflaboratory and field experiments generally coupled withconcurrent modeling efforts. A summary of the significantresults of these investigations is presented in the ensuingsections.

II. SOIL DIELECTRIC PROPERTIES

The dielectric properties of moist soils are quintessen-tial in determining the microwave scattering and absorp-tion by a soil medium. Whereas the relative permittivityof dry soil constituents is typically about 3 and dependsupon packing density, the permittivity of water is about80. Although naturally occurring soils are spatially andtemporally complex media, it has proved convenient to ex-amine the dielectric behavior of relatively simple and "ho-mogeneous" test soils in the laboratory. This simplifica-tion is justified when applying soil dielectric properties toscattering and emission models at the microscale level; itbreaks down at larger scales (related to sensor resolution)only because the true variance in the spatial and temporalproperties is exceedingly difficult to quantify.

In general, a soil medium can be treated as a volumeconsisting of variable fractions of soil solids, aqueousfluids, and air. Soil solids are characterized by the distri-bution of particle sizes (texture) and the mineralogy of theirconstituent particles (particularly the clay fraction). Sev-erallaboratory studies have been conducted to investigatethe effects of soil moisture, bulk density, and soil textureon the net dielectric behavior of the soil medium usingeither guided-wave or free-space transmission techniques

0196-2892/86/0100-0023$01.00 © 1986 IEEE

24 IEEE TRANSACTIONS ON GEOSCIEKCE AND REMOTE SENSING, VOL. GE-24 KO. L JANUARY 1986

Fig. I. Measured (a) real part and (b) imaginary part of the dielectric con-stant as a function of frequency with volumetric wetness as a parameter.

,30

6 r """:,,:1 ~ :"": ~ ": ~~,

o 2 4 6 8 lO 12 14 16 18Frequency f IGHzi

(a)

III. NONVEGETATED SOIL

The microwave energy incident upon the soil surfacemay be scattered, transmitted, or absorbed; the relativequantities of each of these processes and their directionalcharacteristics are determined by the intrinsic dielectricproperties of the soil medium and by the boundary con-ditions at the air-soil interface. Boundary conditions ofinterest include the small-scale random surface roughnessgenerated by agricultural tillage practices, azimuthallydependent ridge/furrow patterns, and the slope of a ter-rain element, which affects the local angle of incidence.

A radar measures that quantity of the incident powerwhich is backscattered, and, in the general case, thisquantity can consist of both a coherent component (from

-.w

40

80

60 ~coaU

50 ~

~

90

o18

~ 14co

~ 12

.~ 10'"

0.1580.079

0236 8 10 12 14 16

Frequency f IGHzl

(b)

10 Field 2, Lo~m

Temperalure, 23°CFree-Space Melhod, 3 - 18 GHzWaveguide Method, 1. 4 GHz

o

~ 6

.§ 5u

~ 40;o'0VI

[7]-[9]. In particular, these studies sought to quantify therole of dielectrically bound water (not necessarily chemi-cally bound water), whose quantity is strongly dependentupon soil texture and mineralogy. The results both of thestudies and of subsequent analyses [10] indicate the fol-lowing.

I) The dielectric constant of dry soil is independent offrequency over the microwave region and is primarily de-pendent upon soil bulk density.

2) The addition of water to a dry soil medium resultsin an increase in the dielectric constant that is smaller inmagnitude for initial increments of "bound water" thanfor subsequent additions of "bulk water."

3) The quantity of "bound water" is controlled by soiltexture and mineralogy (being roughly proportional to thesoil clay fraction), which results in profound differencesamong soil types with respect to the dielectric constant ata given moisture content.

4) The observed differences among soil types are fre-quency dependent and are greatest at the lower frequen-cies (those less than approximately 3 GHz), where the ef-fects of the effective salinity of soil fluids exert significantinfluence.

5) The frequency dependence of soil dielectric proper-ties is generally of the Debye type and is similar in formto that observed for water.

6) Because the dielectric constant of moist soils is pro-portional to the number of water dipoles per unit volume,the preferred measure for soil moisture is volumetric .

The study by Wang and Schmugge [7] at 1.4 and 5 GHzresulted in an empirical formulation for the calculation ofthe soil dielectric constant as a function of soil moistureand soil texture. A later study by Dobson et al. [9] overthe frequency range from I to 18 GHz resulted in bothmulti frequency empirical formulations and a physicallybased theoretical model that explicitly treats a number ofsoil physical properties including soil bulk density, spe-cific surface area, cation exchange capacity, volumetricsoil moisture, and the quantity and dielectric nature of"bound water." An example of the frequency response ofsoil dielectric properties is shown for silt in Fig. I.

The scientific rationale for conducting the dielectric in-vestigations was clearly twofold: first, to gain a funda-mental understanding of the basic property governing mi-crowave sensor response and, second, to provide anaccurate data base for the derivation of dielectric proper-ties as needed inputs to increasingly accurate and de-manding microwave emission and scattering models. Inparallel with the soil dielectric work, preliminary investi-gations have sought to determine the dielectric propertiesof common components of vegetation canopies such asfruit, stalks, and leaves [11]. These efforts have been com-plicated by the fact that the canopy elements are com-monly similar to a wavelength in size, they assume pre-ferred orientations in nature, and they may be irrevocablyaltered by the sample-preparation process.

DOBSON AND ULABY: ACTIVE MICROWAVE SOIL MOISTURE RESEARCH 25

0.6 SOIL' CLASS1 0 SandY LO(Jl12 4 LQ(JI1

3 0 Si I t Loam •0.5 4 '1 Silt Loom •5. Silty Cloy·

40.

0.40~

04 •.~.44

:i •'-~ 0.3 0 •.:: 0 •u

P.0.> 0

~ 0.2 400.1'- •

<5'o.

0.1 l

••

o

o

0.00.0 0.1 0.2 0.3 0.4 0.5 0.6 0 25 50 75 100 125

Volumetric lIoisture "v (g·cm') Percent Field Caoacity, M, (Xl

(a) (b)Fig. 2. Power reflection coefficient at nadir at 1.4 GHz versus (a) true vol-

umetric soil moisture Mv and (b) percent of field capacity Mf'

A. Soil PropertiesRecent investigations have yielded considerable insight

into the nature of the soil bulk properties that control theradar backscattering response. The studies have exploredthe role of soil moisture and profile shape, soil bulk den-sity, and soil texture [5], [6], [13]-[16]. However, the ef-fects of organic constituents, clay mineralogy, and stonysoil inclusions remain largely unexplored.

1) Near-Surface Moisture Profile: For the simplestcase, that of a semi-infinite internally homogeneous soillayer bounded by a smooth surface, the power reflectioncoefficient at nadir is determined from the dielectric con-stant by

specular reflection) and an incoherent component (fromscattering). Although both terms are strongly dependentupon the Fresnel power reflection coefficient determinedby the dielectric properties of the soil (as modified by sur-face roughness), the coherent component is more stronglydependent upon the angular properties of both the scene(roughness and local angle of incidence) and the sensor(beamwidth). Hence, the coherent component can domi-nate the integrated response at near-nadir angles, espe-cially for systems having large beamwidths. For applica-tions in which the purpose is to identify the backscatteredsignal that would be derived by an orbital synthetic-aper-ture radar (SAR) processed to have an effective pencilbeam, the effective weighting by the antenna pattern ofthe measurement system (truck-mounted or airborne scat-terometer) must be taken into account. A procedure wasimplemented to retrieve the "true" backscattering coef-ficient from the truck-mounted scatterometer data, whichwas experimentally obtained at angles near nadir [12].

r1

..fE - 112..fE + 1

(1)

where E = E' - jE. By applying (1) to dielectric data mea-sured at 1.4 GHz for several different soil textures, Dob-son et al. [to] found that values of r range between 0.04for dry soil and 0.52 for saturated soil, which correspondsto a difference of 11dB, as shown in Fig. 2. However, fieldexperimentation with both truck-mounted and airbornescatterometers at this frequency exhibited a dynamic rangeof 12 to 15 dB over the same moisture range [5].

A comparison of scattering-model calculations withscatterometer-measured data from plots of smooth, baresoil led to the postulation of two possible explanations forthe apparent discrepancy: 1) the existence of subsurfaceeffects and 2) an impedance-matching layer at the surface.Allen et al. [17] discounted subsurface effects due to soil-moisture profile shape (i.e., increasing soil moisture ver-sus depth for a dry surface layer) because these effectswould typically lead to an increase in the reflection coef-ficient calculated for a dry surface. Assuming the exis-tence of a transition zone (in which the upper millimetersof soil are considerably drier on a volumetric basis thanthe average of the top several centimeters) functioning asan impedance-matching layer, Allen et al. [17] comparedfield backscattering measurements to the solutions of aKirchhoff scattering model using both Wilheit's [18]method for calculating the reflection coefficient from alayered medium and an iterative solution to the Riccatiequation. This assumption yielded good fits to the mea-sured data and indicated that the thickness of the transitionlayer is inversely related both to near-surface soil moistureand to frequency. From the standpoint of field measure-ments, this result emphasizes the need to pay critical at-tention to the moisture profile at the surface (at a subcen-timeter level) in order to produce exact model calculationsof the backscattering from dry soils.

2) Soil Moisture Response: Before the advent of theAgRISTARS program, extensive measurement programs

26 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. GE-24 NO. I, JANUARY 1986

,were conducted by the University of Kansas using truck-mounted scatterometer systems to observe test plots ofnonvegetated soil with distinctive surface roughnesses andsoil textures [2]-[5]. For a given soil condition (roughnessor texture), radar backscattering was found to be linearlydependent upon the volumetric moisture Mu in the upper2 to 5 cm of soil and to have linear correlation coefficientsp typically on the order of 0.9

(T°(dB) = A + B Mu. (2)

For a given sensor combination of frequency, polariza-tion, and angle of incidence, the empirically derivedregression coefficients A and B were found to be depen-dent upon soil surface roughness and soil texture, whereinA is primarily controlled by surface roughness, and B isprimarily controlled by soil texture. For the prairie mol-lisols examined in these studies, polarization had no sta-tistically discernible effect on the sensitivity term B. Bothcombinations of like linear polarizations (HH and VV)yielded equivalent A, whereas cross-polarization produceda substantially lower value of A. In addition, the sensitiv-ity term B was observed to be dependent upon both fre-quency and angle of incidence, gently decreasing witheither increasing frequency or angle of incidence.

Subsequent studies by independent groups using ground-based scatterometers in the United States, France, andJapan [16], [19], [20] have verified many of these results.However, the results obtained by Hirosawa et al. [20]based on 9-GHz observations of Kanto loam represent anotable exception: they found that the cross-polarized sen-sitivity to near-surface volumetric soil moisture was fourtimes that of the like-polarized backscattering. This resultwas attributed to the effects of multiple surface scattering;however, a similar effect has not been observed for veryrough mollisols [5], [21].

Several investigators have reported the results of air-borne scatterometer observations designed to sense soilmoisture. These experiments were conducted during a se-ries of overflights in 1978 and 1980 [21], [22]. Typically,these experiments used fan-beam Doppler scatterometers,operating at P-, L-, C-, and Ku-bands, mounted aboard aNASA Johnson Space Center C-130 aircraft. Multitem-poral observations of test areas in Kansas, Oklahoma, andFlorida yielded fairly robust data sets in terms of soilmoisture, vegetation cover, and surface roughness condi-tions. Analyses of these data support the conclusionsreached on the basis of the more geographically limitedtruck-mounted scatterometer observations.

Calculations of the power reflection coefficient from themeasured dielectric data shown in Fig. 2 indicate that thebackscattering coefficient (in square meters times recip-rocal square meter) should be linearly dependent upon soilmoisture at moisture levels below saturation. Near satu-ration, the backscattering should level off, apparently be-coming less sensitive to added increments of water. Be-cause all of the field measurements of backscattering todate have reported (To in decibels, empirical regressionshave taken the form given by (2). The scattering typically

inherent in the field measurements makes if difficult tosubstantiate this expectation. However, field measure-ments have shown the saturation effect at high moisturecontents, and these studies demonstrate that supersatu-rated and flooded soils behave as specular surfaces, whichyield lower backscattering at off-nadir angles than non-saturated (but wet) soils [6], [23].

3) Soil Bulk Density: The dielectric studies of moistsoils show that, for a given gravimetric soil moisture (theratio of water mass to dry soil mass), the effect of in-creased soil bulk density should be to increase the reflec-tion coefficient due both to increased soil solids and towater dipoles per unit volume of soil. Because the contri-bution of the dry soil solids is small relative to that of thewater component, the effects of density on dielectric prop-erties (and hence on the reflection coefficient) are largelyaccounted for by expressing soil moisture on a volumetricbasis (the ratio of water volume to moist-soil volume). Thesignificance of soil bulk density effects has proved to bevery difficult to verify by field measurements, due to 1)spatial variance in bulk density, 2) the temporal dynamicsof bulk density, particularly for certain clay-rich soils, and3) the great difficulty in obtaining an accurate determi-nation of field bulk density for very thin layers of near-surface soil. The issue of soil bulk density effects has beenaddressed recently by several studies [10], [15], whichconclude that very careful attempts should be made toquantify soil bulk density in the field in order to avoid mis-taking the density effects on sensor response for soil tex-tural or roughness effects on sensor response.

4) Soil Texture: Because most of the ground- and air-craft-based scatterometer studies of moist soils were pur-posely chosen to cover test sites having lateral homoge-neity of soil type and texture, the effects of soil textureand mineralogy on radar backscattering are less under-stood than properties such as surface roughness. The ef-fects of soil texture are best inferred from dielectric stud-ies and, as observed by Dobson and Ulaby [6], duringtruck-mounted scatterometer measurements.

The dielectric data strongly suggest that the firstmonolayer of water surrounding the surface of a soil par-ticle is largely irrotational under an impressed microwavefield and hence is characterized by a relatively low dielec-tric constant that is dissimilar to either bulk water or ice[9]. The quantity of water that is dielectrically "bound"is determined by soil-particle size distribution (texture) andmineralogic composition via the specific surface area ofthe soil. The data also suggest that additional volumetricincrements of water (beyond the "bound" component) ex-hibit dielectric properties that appear to be independent ofsoil texture per se but are dependent upon the effectivesalinity of the soil solution (which may be controlled bytexture and mineralogy).

Attempts to compare early field investigations of differ-ent soils led to the development of a normalized soil-mois-ture index known as percent of field capacity, which isdefined as the ratio of the gravimetric soil moisture to themoisture at a soil's field capacity. In practice, field capac-


The coherent scattering coefficient is given by the ap-

1) the roughness itself is concurrently extracted viamulti frequency or multi polarized observation (like-and cross-polarized returns), or

2) soil moisture is estimated via a change-detection ap-proach because surface roughness varies slowly withtime for agricultural fields in the absence of tillageoperations.

The preceding approach seems to be tractable becausethe backscattering behavior of randomly rough surfaces isshown to be well described by current scattering models.These models take two general forms with some modifi-cations. The Kirchhoff model with the scalar approxima-tion, or physical optics model [25], is used to describe theexponentially decaying angular dependence characteristicof smooth surfaces. The Kirchhoff model with the station-ary-phase approximation, or geometric-optics model, isapplied to relatively rough surfaces that display a slowlyvarying angular dependence near nadir.

The like-polarized backscattering coefficient of an iso-tropically rough surface consists of both a coherent term,a~PC' which is important only at angles near normal inci-dence, and a noncoherent term a~pn, which is important atall angles

(3)p = v or h.

terometer. The angular dependence of the cross-polarizedreturn is far less than that shown for like polarization. Fora given frequency and polarization, the bias in aO causedby the variation in local angle of incidence related to to-pographic relief is seen to be a function of angle of inci-dence, local slope, and the random roughness of the soilsurface as seen in Fig. 4, whereby the smoother surfacesyield a greater bias per degree of angular uncertainty.

2) Random Surface Roughness: Examples of the mea-sured effects of small-scale random surface roughness onradar backscattering are shown in Figs. 3 and 4 for non-vegetated soil surfaces that were specifically prepared forthis purpose. The data used to derive Fig. 3 have beendeconvolved to remove the coherent portion of the netmeasured backscattering related to the antenna pattern ofthe measurement system [12]. The angular effects ofvariable surface roughness are seen to decrease rapidlywith frequency, as even the smoothest surface observed(rms height is 1.1 em) is no longer smooth by the Rayleighcriterion at frequencies above 3.4 GHz. The observedcrossover in the angular responses for the various surfacesover the angular range'from about 7 to 15° has been in-terpreted [5] as the optimal angular range for soil-mois-ture sensing with a minimal dependence on agronomicallyinduced random surface roughness (certain geologic sur-faces can be much rougher). Even within this angularrange, however, it can be seen that roughness effects canbe a significant source of error in soil-moisture determi-nation from like-polarized backscattering for a particularfield unless:

B. Boundary ConditionsThe nature of the boundary at the air-soil interface de-

termines both the amplitude and the phase properties ofthe reflection and transmission by the soil medium. Thenature of the effects upon radar backscattering caused bysoil surfaces can be subdivided into three categories:

1) effective local angle of incidence related to terrainslope,

2) small-scale surface roughness with laterally randomsize distributions, and

3) azimuthally dependent and generally periodicroughness patterns induced by agricultural tillagepractices.

For a given sensor combination of frequency, polariza-tion, and angle of incidence (relative to the mean surface),boundary conditions do not significantly affect the sensi-tivity to soil moisture but do add a bias term to the re-sponse.

l) Local Slope: The effects of a variable local angle ofincidence can be inferred from an examination of Fig. 3,which shows the angular behavior of aO for five nonvege-tated soil surfaces as measured by a truck-mounted scat-

ity is typically defined on the basis of laboratory measure-ments of soil water retention at an arbitrarily defined valueof ~-bar matric potential. This index was an attempt toaccount for the soil properties governing the apportion-ment of soil fluids into "bound" and "bulk" water. Theapplication of this index to empirical comparisons of air-borne radiometer [13] and truck-mounted scatterometerdata [5] with soil moisture as observed for two differentsoil types yielded relationships that were apparently in-dependent of soil type. Further work by Dobson and Ulaby[6], comparing the backscattering from three smooth soilsurfaces having distinctive soil textures yielded the sameresult, but also showed that the expression of soil moisturein terms of matric potential produced linear relationshipsthat were independent of soil texture. The linear relation-ship between matric potential and reflectivity [16] orbackscattering [14] has also been noted by more recentinvestigations, which, unfortunately, have dealt with sin-gle soil textures only.

In partial contradiction to the preceding observations,an analysis of the dielectric data brings into question thephysical basis of the percent-of-field-capacity index (andhence its geographical extensibility) on the basis that theindex functions as a surrogate for accurate volumetric soilmoisture information by partially accounting for the inter-soil variability in soil bulk density [10]. Hence, to someextent, the use of percent-of-field capacity may be useful,though not rigorously correct. As a consequence, it is be-lieved that the best physical descriptor of soil moisture isvolumetric, and the evidence to date indicates that the in-ter-soil variability in radar sensitivity to Mv is related tothe soil-specific nature of the characteristic curve relatingmatric potential to Mv. However, this should be examinedspecifically by additional experimentation.

,28 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. GE-24 NO. I, JANUARY 1986

:1

1:\

-5 !\:l_, '--10 ' '_"""'_""""'-.-.-.-.

-15L~""''''''L'~ --......---.......- -. -......-......-20 "'"

-'.- --.

-25 ' , , , , , . , I , , , , Io 10 20 30

Angle of Incidence e (Degrees)

--20

30

R"5 Soil Moisture II

RO~[SS (g.cm-3)

-- 4,1 0.382.2 0.353,0 0.301.8 0.35_m_ 1.1 0.35

:0 20A ngle of Incidence e (Degrees I

5

10

20~ " Frequency IGHzl: 4.25.,', PolarIZation: HH

15~\'"\,\,\,

",\<\.

'\\,.:,-,

'''''<.';<","''''''''~<,,>...j~l-15

-10

r"-25 L~.,.----L-~_.:. .. _l_..--'---___L__:_~_L_l __ L l__.....L__.L_L_"i.. ~~~

o~ 0 -

2I-'C

'">~ -5

'"

R"5 Soli Moisture~::im Ig.cm-3)-- 4,1 0.34

2,2 0.35--- 3,0 0.36-- 1.8 0.29_un 1.1 0.32

-""I

Frequency (GHzl: I. 5Polarization: HH

5

15

10

20

(a) (b)

20 ~uency (GHzl: 7.25Polarization: HH

15 r~

10 ~.i ',""5 '.: ,

"'-. '"

R"5 Soil Moisture~[SS Ig.cm-31

-- 4.1 0.362.2 0.353.0 0.35

- .1.8 0.34------ 1.1 0.30

ot:> 0'"2I-

~~ -5

'"-10

(c)Fig. 3. Angular pattern of retrieved true aO for five soil surfaces with dif-

ferent roughness scales (a from 1.1 to 4.6 em) at (a) 1.5 GHz, (b) 4.25GHz, and (c) 7.25 GHz (from [12]).

proximate expression [24]

r/8) 2 2 2 2(J~pc(8) :::::[j2 exp (-4K ()) exp (-8 IS )

where

rp(8) is the Fresnel reflectivity for polarization p at inci-dence angle 8, k = 27[1)"" (J is the surface rms height, Ro

(4) is the range from the antenna to the center of the illumi-nated area, and {3is the one-sided beamwidth of the an-tenna for a nonimaging scatterometer or its pixel-equiva-lent for an imaging system.

For the noncoherent component, the physical optics


~~ 5u

::: 0§~ -5 ...- __ !-.--.-.-",?. "~ -10 ".M'- ."" "~ -15 ,,"

/~-20 It" FreQuencv: 1.16Hz

-251 2 3

RMSHelght a (em)

(a)

20

'b 10

Polarization: HHKoJsture Content:

0.34 - 0.40 g.em-3Angle of Incidence 9:

_0'.-- 10-;--- 10-

•. ..•....--L.::-_-:=:A.-- - - ••.-

~,-,""--

Frequency: 4.256Hz

2 3RMSHeight a I ern)

(b)

...___ ..-_.A.- -- --.-------

L-,==,,"'"'1 2 3 4

RMSHeight a lem)

(c)

Fig. 4. Scattering coefficient dependence on nns height and angle of inci-dence at (a) l.l GHz, (b) 4.25 GHz. and (c) 7.25 GHz.

525

kG kL

20 - " 0.2 70.2 300.5 7

l'l 0.5 300.6 70.6 30

-400.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.6 0.9 1.0

ka

(b)

8 kL

" 20· 720· 1520° 30

l'l 50°50· 1550Q 30

a

-5

-35

-30

-10

'"'C--15oI:>owt!-20--'«'"'"oz-25

6010 20 30 40 50INCIDENCE ANGLE 8 (Degeee.)

(a)

-30

-35

-40o

-25

00 -5

okj-10

Ls'"oz-20

- 0'"'C

Fig. 5. Computed normalized backscallering using the scalar approxima-tion as a function of (a) incidence angle and (b) ka.

(6)

model gives [25]

u~PIl(8) = 2 k2 cos2 8 rp(8) exp r -(2ku cos 8)2]

00

.6 [(4k2 if cos2 8)"ln!]n=l

(5)

where Jo( ) is the zeroth-order Bessel function of the firstkind and p(O is the surface correlation function. Fig. 5shows plots of U~PIl( 8)/r p( 8) as a function of 8 and ku foran exponential surface correlation function p(O = e -UL,

where L is the correlation length of the surface. The geo-metric optics model gives the same expression for HH and

VV polarizations [25]o J

u~:(8) = reO) exp (-tan" 812m-)2m2 cos4 8

where m is the rms slope and reO) is the Fresnel reflectiv-ity evaluated at normal incidence. Plots of u?,(8)/r(O) ver-sus 8 are shown in Fig. 6.

3) Periodic Row Directional Effects: Agriculturalcrops are generally planted in parallel rows in either arectangular format or in concentric rings (as in the case ofsome center-pivot irrigation systems). Soil tillage is alsoconducted by parallel operations using farm implements,which typically produce nonrandom and periodic ridge Ifurrow boundary conditions that modulate the small-scaleand isotropic roughness components. The periodic com-

30 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING. VOL. GE-24 NO. I. JANUARY 1986

INCIDENCE ANGLE e (Degrec3)

Fig. 6. Computed normalized backscattering using the stationary-phaseapproximation (geometric-optics model) as a function of incidence angleand rms slope. (7)

IA

Scatterometer observations have shown that the azi-muthal effects of ridge/furrow patterns are also dependentupon the degree of isotropic small-scale surface roughnesspresent, angle of incidence with respect to the ridge/fur-row profile shape, frequency, and polarization. These ef-fects are accurately described by a modified form of thescattering models previously introduced, which treats theperiodic surface modulation as modifying the local angleof incidence for finite elements within the integrated il-lumination area [27J. In functional form, the backscatter-ing coefficient (J°(O, 1/;) of a periodic surface observed atan incidence angle 0 (relative to the mean surface) andazimuth angle 1/; (relative to the row direction) is relatedto (J°(O'), the backscattering coefficient at the local angleof incidence 0', by an integral of the form

605040302010o

Fig. 7. Scatterometer time response measured for a wheat-stubble fieldwhose row pattern is shown in the insert. The observation angle was 20°.

:::::~.:..::..~"-,,,'\

\>_._4'::'-Gc!LH~j/\.J._~_G~_\_i_H ~'~' ... ,/i-20 --- ~

~ P~rpendi(~I_a_~ Parallel ~_~_!:p!~_d!~ular

~ ::-:..=-:- 475GHzHV "'7"/.::...-.....-... .c'ri -30 "" ---7<::::r:-:,--;>'~ ' ,,/]6 GHz, HV '--

oc -48 L-----L- __:~~-L__L ,_.L_____L _ __'_____.JI 2 J 4 5 6 7

Time (s)

ponents of surface roughness are of particular interest tothe soil-moisture estimation problem because they havebeen observed to exert a considerable angular effect onradar backscattering [26]. Of major concern is the azi-muthal dependence of the radar backscattering from ridge/furrow patterns. Examples of this type of dependence in-clude the "bowtie" effect commonly seen on radar imagesof rectangularly tilled agricultural fields and in the air-borne Doppler scatterometer time traces shown in Fig. 7.In certain respects, this phenomenon is analogous to the"cardinal direction" effect observed in radar images ofurban scenes in which radar view angles orthogonal to cul-tural features yield very high levels of backscattering.

Observations of agricultural fields with truck-mountedand airborne scatterometers [21], [26], [27] and Seasat L-band SAR [28] suggest that radar is most sensitive to azi-muthal viewing geometry for angles within 15° of orthog-onal to the row direction and that this sensitivity decreasesin an exponential fashion as view angle becomes parallelto row direction. Because many, but not all, agronomi-cally important areas are planted in a rectangluar grid pat-tern with a north-south and east-west orientation, an op-erational orbital radar intended for soil moisture sensingshould have an orbital inclination greater than 15° frompolar orbit in order to minimize these effects at most lat-itudes.

IlluminatedArea

IV. VEGETATED SOIL

where A is the illuminated area. The preceding form isgiven here simply to indicate that (Jo(O, 1/;) depends on thefull angular range of uO(O'); the actual transformation in-volves the various polarization states of (J°(O'), which leadsto a more complicated integral [27] than that given in (7).Many of the row-direction effects can be summarizedusing the look-direction modulation function M(O) definedas the difference in (J0 (in decibels) between parallel andperpendicular observations with respect to row directionas follows:

I) M( 0) is greatest for fields having the least randomroughness and decreases rapidly as the surface becomeselectromagnetically rough at a given frequency.

2) Related to the preceding, M(O) decreases rapidlywith increasing frequency.

3) M(O) has a local maximum for local angles of inci-dence that are tangential to furrow slopes; this angle istypically in the 20° to 40° range and depends upon thefield-specific tillage practices in use.

4) Importantly, the cross-polarized scattering coeffi-cient is relatively insensitive to row direction effects andis typically found to be less than 2 dB for the reportedmeasurements and models. The larger variance seen in the1.6-GHz HV response shown in Fig. 7 has been attributedto poor polarization-isolation of the antennas.

Remotely sensing the moisture of the soil beneath a veg-etation canopy has been the subject of keen interest andmoderate experimental attention for the past 10 years.Early work, based largely on truck-mounted scatterome-ter measurements, sought to identify those sensor com-binations of frequency and angle of incidence least sensi-tive to the presence of agricultural canopies. The studiesconcluded that the optimum parameters for moisture sens-ing should be frequencies of less than 6 GHz and angiesof incidence of less than 20° in order to minimize boththe direct back scattering by the vegetation and the effec-

l3. 3 GHz, VVo

.~ olD

ou


This term is thought to become significant for sensor con-figurations for which the transmission loss is small andwhen there is significant scattering from either the vege-tation volume or the soil surface. The surface-vegetationinteraction term can be the dominant term in cross-polar-ized return.

Empirical evaluations of scatterometer data with re-spect to (8) have generally ignored the contribution of theinteraction term and have sought to define the remainingmodel coefficients based upon muItifrequency or multi an-gle curve-fitting solutions for single-target observations[34] or have sought to define crop averages for the vege-tation term and loss over a growing season [35]. Attemptsto estimate the canopy loss factor on the basis of mea-surements of the dielectric properties of the vegetationhave resulted in good fits to the measured data [11], [361.In addition, several recent attempts have been made to

(8) define the loss parameter directly from one-way canopytransmission measurements using scatterometers at C- andX-bands [36], [37] and radiometers at S- and C-bands[38] .

The one-way canopy loss as derived from radiometer(9) observations of test plots of wheat, corn, and soybeans is

shown in Fig. 8 as a function of crop-development stage.These values were obtained through a comparison of theapparent brightness temperatures of test plots in their nat-ural state with those of adjacent plots in which the under-lying soil surface was covered with a reflective material(wire mesh screens) or microwave absorber. Temporal be-havior is clearly related to changes in vegetation biomassand to the appearance of distinctive canopy structural ele-ments; the exact nature of these relationships remains tobe determined. In general, the studies conducted to dateshow that

For an isotropic canopy characterized by an opticaldepth 7, the surface term is given by

(12)

::::::~ (1 + cos2 20)Ks T2(fJ, T) rp(8)

1. exp [ - (2ka cos Otl.

oUintemction

1) both the canopy loss and the vegetation volume scat-tering coefficient are linked to the canopy's bio-physical properties, and especially, but not exclu-sively, to canopy type, canopy structure, and thewater volume fraction within the canopy;

2) the canopy loss and the volume scattering coefficientincrease with frequency;

3) the vegetation term in (8) tends to dominate the netreturn as either frequency or incidence-angle in-creases; and

4) the interaction term functions to enhance radar sen-sitivity to the moisture contained in the soil beneatha vegetation canopy.

For purposes of soil-moisture sensing, it is preferablethat the sensing systems exhibit no sensitivity to canopybiophysical parameters. If this cannot be the case, it ispertinent to define how much the canopy's effects reduceradar sensitivity to near-surface soil moisture for canopyconditions typical of an agricultural setting. The effects of

(11 )

(10)T(O, 7) = exp (-7 sec 0).

o 3 Ks cos 0 2a vegetation == ----[1 - T (0, 7)J

4 Ke

o 0 0 0a total = a surface + a vegetation + a interaction'

where Ks is the volume scattering coefficient and Ke is theextinction coefficient of the vegetation layer. A Rayleighscattering phase function was assumed in the derivationleading to (11). Both parameters, which are dependentupon the biophysical properties of the canopy, are assumedto be polarization- and direction-independent. The sur-face-vegetation interaction term is determined by multiplereflection between the canopy and the surface, and itsmagnitude can be estimated approximately by

The vegetation layer is treated as a uniform "cloud" ofidentical water particles, with the resulting scatteringbeing entirely due to volume scattering, in which casethere is no need to account for the scattering at the diffuseair-vegetation boundary. A full theoretical treatment ofthe vegetation term by Eom and Fung [32] permits mul-tiple scattering within the vegetation layer. However, dueto the small magnitude of the single-scattering albedo typ-ically ascertained for crop canopies at frequencies below6 GHz (on the order of 0.1), empirical model evaluationshave generally simplified the treatment of this term byconsidering only single scattering. This simplification re-sults in the" cloud model" developed by Attema and Ulaby[30], which is only applicable to like-polarized returns

where T (0, 7) is the one-way transmissivity of the vege-tation layer

A. Bulk Canopy Biophysical PropertiesIn general, the radar backscattering from a vegetated

soil surface consists of three components: 1) a soil surfacecomponent, 2) a vegetation component, and 3) a surface-vegetation interaction component.

tive attenuation loss related to the two-way transmissionthrough the canopy [29]. Subsequent studies using truck-mounted scatterometer data as well as data obtained bythe airborne Doppler scatterometers have provided bothsimple empirical models and more robust theoreticalmodels for the effects of agricultural canopies [30]-[32].To date, most of the work has treated the vegetation can-opy as an isotropic medium of disperse scattering ele-ments with properties linked to bulk biophysical parame-ters such as crop type and wet and dry biomass; somerigorous studies of the role of canopy structure (the size,shape, and orientation distributions of canopy elements)have been undertaken [33], but further work is needed.

(14)

I

JlI5

_0m ~.

~ T~

o 764 D.190.110 0.76II 054 0.880.070 0.61

Bare = ACOril = BBeans CMilo 0Wheat = E

o°soit,

O.88,oOldBI 0.133" 14.340,85 oOjdBI 0 148Mf - 15,960.92,oO(dBI 0 113M, - 13 84

Frequency fGHz): 4.5Angle of Incidence; 100Polarization; HH

(a)

---- Bare SoilI _I __1_1 L_l. J I 1 _L __1 __.1. J__1 1 J L~..L----L

~ ~ ~ m ~Soil Moisture Content of Top 5-cm Layer

"I IPercent of Field Capacity)

All, N 314, PBare: N - \81. P

Ve9: I. -143. P

(Ghz): 4_ 5of Incidence IDegreesl: 10

~H

0.066 + 0.75

~ _ ~ .l I _----L__ ----1. --1 L 1. 1 J ~_

a 25 50 75 100 125 150

Soil Moisture Content in Top 5-clll Laver. ",0:, of Field Capacityl

-24 -

14

18

12

ro

'b-6

-18 -

0-

-5-r

-10

In an anlysis of the prediction errors arising from theuse of generalized algorithms such as (13) for bare soil or(14) for vegetated soil, Ulaby et al. 135] concluded that itwould be difficult to estimate soil moisture with any gooddegree of accuracy for low soil-moisture conditions (lessthan about 50 percent of field capacity) from a single sen-sor observation unless the presence of a canopy cover isknown a priori or from other sensor observations (visible/IR or some other microwave frequency. polarization, orangle). On the other hand, for soil moisture conditionsgreater than 50 percent of field capacity, it is estimatedthat the 90-percent probability confidence interval wouldyield an uncertainty of ± 15 percent of true field capacity,which corresponds to an uncertainty in volumetric mois-ture of about ±0.02 and 0.05 cm' . cm -, for sands andsilty clay soils, respectively. It is interesting to note that

net backscattering of between 0.7 and 2.0 dB for corn andmilo, respectively, as compared to that from bare soilalone. Application of the regression procedure to all 143observations of the various crops at this sensor combina-tion yielded a general algorithm for estimating the mois-ture of soil beneath agricultural crop canopies with a lin-ear correlation coefficient of 0.91 139].

(b)

Fig, 9. (a) Linear regrcssion of a" (in dccibels) vcrsus M, for bare fields.vegetation-covered fields. and both typcs combincd; (b) variation of can-opy backscattering coclilcient with soil-moisture contcnt IClrbare soil andindividual crop lypes.

300

)110

--jl,'tlcat---- Corn-- Soybeans

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. GE-24 NO. I, JANUARY 1986

Apoeo ranceof Cob

1 Appearanceof F lowers /I

// ''---, )---/ \/ y-, \

j- '"'' \./ __._L_._1_' __ L\_-'-

220 260

Day of Yea r

2.7 Gllz

(a)

Frequency: 2.7 GllzPolarIzation: IIAngle of Inc!r1ence: 10·-- W~eot---- Corn- - SoYbeans

Day of Yenr

5.1 Gill

~ ADD~aranC~----F~~l;;-'enCY~-'-k-.~'-TJof lie ad . ~.

Polarization: f-IAngle of InCldencc: 10-

]2 -

aua~

'"~ 10 -

32

14

12

'":::10-'~~a 8-'>- Appearancea

~ of lIead6

>-a{

4:g0

0]00 140

(b)

Fig. 8, Comparison of canopy altenluation l()r various crops at (a) 2.7 GHzand (b) 5, I GHz.

canopies of corn, soybeans, wheat, and milo (sorghum)on radar sensitivity to soil moisture at the sensor config-uration deemed least sensitive to surface-boundary con-ditions (C-band at 10° to 200 angles of incidence) havebeen examined empirically using multiyear scatterometerobservations [35]. Data obtained over the period from cropemergence to crop harvest were used to define averagecanopy loss and {T8egctatiol1 for each crop type through a lin-ear regression approach that assumed (J2oil to be that cal-culated from measured soil moisture by

(J~oi1 = 0.025 exp (0.034 Mr). m2 m-2 (13)

where Mr is the 0-5-cm percent of field capacity. Equation(13) results from the linear regression of 181 observationsof bare soil plots with rms surface-height variations rang-ing from 0.7 to 4.3 cm at C-band with HH polarizationand a 10° angle of incidence; the linear correlation coef-ficient was found to be 0.85 [39]. Assuming a negligibleinteraction term (see 12), the average canopy effects yieldthe responses shown in Fig. 9. It is apparent from Fig. 9that at low soil-moisture levels (less than 50 percent offield capacity) the backscattering contributions from thecrop canopy itself dominates the total return, whereas athigher moistures, the canopy loss causes a reduction in the


of the 583 discrete moisture observations made in Kansasbetween May and November, 80 percent had 0- to 5-cmmoisture values in excess of 50 percent of field capacity.

15 30 45 60Angle of Incidence B lDegreesl

(b)Fig. 10. Effect of a mature corn canopy undergoing progressive stages of

defoliation on (a) emission and (b) backscattering at C-band.

B. Canopy StructureThe canopy structure is the complex spatial organiza-

tion of discrete canopy components such as stalks, leaves,and fruit. Each component has a characteristic size, shape,orientation, and location distribution. In part because can-opy structure is exceedingly difficult to quantify undernatural field conditions, very few experiments have beenconducted to examine its effects upon radar backscatteringwith respect to frequency, angle, and polarization. As aconsequence, the impact of the variability of canopy struc-ture upon radar sensitivity to soil moisture over time orbetween species cannot be specifically addressed. Never-theless, several very interesting studies have been con-ducted and can be classed into two groups: transmissionmeasurements [36], [37] and defoliation experiments.

The transmission measurements examined the verticalstructure of wheat (grain heads versus stalks and leaves)with respect to polarization and local angle of incidenceat X-band and demonstrated that in certain cases the lay-ered structure of the canopy is important because canopieshaving strong angular properties (such as vertical stalks)can couple differentially at HH and VV polarizations [36],[37].

The defoliation experiments examined the backscatter-ing and emission from canopies from which successivelayers or types of canopy components had been progres-sively removed (by cutting). Fig. 10 is an example of a

V. SOIL MOISTURE RETRIEVAL

The ultimate objective of the soil-moisture researchconducted under the auspices of the AgRISTARS programis to develop the algorithms and methodology necessaryfor retrieving soil-moisture estimates on an areal basis forapplications in hydrologic and agronomic monitoring andassessment. The sensitivity of radar backscattering toscene parameters including soil moisture, soil texture, soildensity, surface roughness, surface slope, crop-canopycover, and row-directional effects has been examined,either independently or in combination, by means of scat-terometer studies of individual test plots in which the scat-terometers were capable of relatively fine spatial resolu-tion. The next logical step, then, in the analysis is toexamine the combined effects of all scene variables on thecapacity of radar to accurately estimate soil moisture foran imaging system with a coarser resolution, such as anorbital SAR.

Experimental work in this area has been limited to theL-band and HH-polarized SAR systems carried by Seasatin 1978 [28] and SIR-B in October of 1984. The data pro-duced by the SIR-B mission are currently under investi-gation by several research groups. Although the scatter-ometer studies indicate that this frequency and polarizationcombination is less than satisfactory for purposes of soil-moisture sensing, primarily because of the pronounced de-pendence upon surface roughness and row-directional ef-fects, the Seasat data were found to be highly correlatedwith near-surface soil moisture (p = 0.84) for agriculturaltest field in the Great Plains for which concurrent groundtruth was available [28]. In addition, a qualitative analysisof several Seasat scenes over Iowa revealed a dramatic sen-sor sensitivity to antecedent rainfall events, although cor-responding ground truth was not available to resolve thequestion of whether the sensor response was driven by freewater present on the crop canopies or by soil moisture (orboth) [40].

Because orbital sensors with the scatterometer-definedoptimal sensor configuration (i.e., C-band at 10° to 20°angles of incidence) are not yet available, the expected

corn canopy monitored with a scatterometer and a radi-ometer (both at 5.1 GHz) through successive defoliationstages until only the bare soil surface remained. For theradiometer, the canopy brightness temperature is domi-nated by the vegetation contribution at all angles; the in-crement in brightness temperature over that observed forthe bare soil case is roughly proportional to the overlyingwater density of the canopy. In sharp contrast, the radarbackscattering is observed to be insensitive to the pres-ence of the corn canopy at incidence angles of less than150. At higher angles, the backscattering contribution ofthe canopy increases and is dominated by the return fromthe vertically aligned stalks and cobs, whereas the canopyloss component is apparently dominated by the leaves.Observations such as these strengthen the argument forusing incidence angles near nadir for radar sensing of soilmoisture.

Canopy WaterDensity (kg.m-J,

1.511.170.81

. 0.00

75

Frequency IGHz~ 5. I Canopy WaterPolarization, HH Density Ikg·m-3)

• Whole Plant 1.51• Stalks and Cobs (No Leaves!.. I. 27•• StalksOnly D.82• Soil D.002 Two Overlappi ng Poi nts3 Th ree Overlappi ng Poi nts

15 30 45 60Angle of Incidences (Degrees)

(a)

A-FrequencylGHH 5.1• Polarization: V

Soil Moisture Content19·cm-3): 0.35

Canopy Height (eml: 277

100

m 110

16

-16

-24, I

o

'b-8

34 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. GE-24 NO. L JANUARY 19X6

JOO

Fig, I L Cumulative percent area of all moisture-dependent pixels (ex-cludes cultural features, water, and woodland) in each subregion as afunction of absolute moisture classification error for (a) 100 m x 100 mradar resolution and (b) 1 kill x 1 kill radar resolution, Mrrepresents thetrue moisture and Mr the moisture estimated by the radar (from [42]),

performance of such systems has been tested via simula-tion studies that incorporate all known sensor and scenecharacteristics 141]-[431. These studies have sought to de-fine both sensor characteristics (i. e., resolution, antennasize, power requirements, and data rate) and the influenceof scene confusion factors (i.e., topographic effects, var-iable canopy cover, and the complex spatial distributionsof water bodies, forests, and urbanized areas) on the ac-curate retrieval or soil moisture from a SAR image.

The simulation studies are based upon digital terrainmodels in which each terrain element is characterized asto land-cover category, soil properties, and crop row di-rection. Typically, meteorologic events are used to simu-late dynamics in the near-surface soil-moisture distribu-tions over the test region as a function of time and localevapotranspiration demands. At selected time intervals,the backscattering properties of each subresolution ele-ment are defined by a Monte Carlo procedure based uponthe scatterometer studies and estimates of "true" naturalscene variability. The effects of signal scintillation (fad-ing), shadowing, and layover are also incorporated intothe image-formation model.

Early simulations of a 20 km X 20 km, largely agricul-tural test region indicated that 0- to 5-cm soil moisture(expressed as a percent of field capacity) could be re-trieved with an accuracy of ± 20 percent for 90 percent ofthe agricultural area using a C-band SAR with HH polar-ization at ]0 to 1]0 angles of incidence and resolutions of100 m X 100 m [4]], [42]. Thc simple retrieval algorithmrequired only the range position of the image pixel and themagnitude of the backscattered signal as input. A subse-quent simulation for a much larger area, "'" 100 km X 120km, which included more diverse topographic and land-cover conditions [43], reached much the same conclusionbut more fully addressed the effects of scene confusionfactors on the expected retrieval accuracy using a verysimple "blind" algorithm dependent only upon range. The

results are summarized as follows (Fig. II):

I) retrieval accuracy is optimized when radar resolu-tion is smaller than the expected field-size dimen-sions of agricultural fields,

2) retrieval accuracy is optimized when radar resolu-tion is coarser than local topographic variation inhilly areas,

3) the effects of row direction on retrieval accuracy aresmall, provided that the orbital trajectory yields azi-muth view angles not orthogonal to row direction,and

4) retrieval accuracy can be improved by about 10 per-cent when multidate change detection is used to pro-vide updates.

VI. CONCLUSIONS

During the period of the AgRISTARS program, signif-icant progress was made toward understanding the fun-damental processes of target/sensor interaction, quanti-fying the effects of bulk scene properties and air-soilboundary conditions, developing both empirical and the-oretical backscattering and emission models, and evalu-ating the potential performance of "optimal" orbital radarin terms both of sensor requirements and of possible soil-moisture retrieval methodologies. The research to date in-dicates that estimates of soil moisture in the 0- to 5-cmlayer can be retrieved with reasonable accuracy for mostrequirements over agricultural areas from multidate andsingle-sensor observations. Such a system should operateat C-band over angles of incidence from about 10° to 20°.There is strong-but not conclusive-evidence to indicatethat HV polarization will yield superior performance toHH polarization for soil-moisture retrieval.

There are, however, many pieces of the puzzle remain-ing to be fitted by means of further experimental investi-gation. For example, the precise physical role of soil tex-ture as related to volumetric water content and soil matricpotential with respect to radar backscattering is undefinedat present. The utility of high-quality, cross-polarized

The rationale for the postulated effectiveness of multi-date change detection is based upon a simple considera-tion of scene-confusion factors and scene dynamics. Forpractical purposes, topography is constant, surface rough-ness decays slowly with time (except at critical points inthe local crop calendar such as planting and harvest pe-riods), and the interfield variance in canopy cover variesover periods of weeks for most canopies.

A change-detection approach applied to Seasat imageryover a test site in southwestern Kansas shows the tech-nique to be effective for discriminating fields subjected toirrigation or tillage operations from larger spatial scalevariations related to antecedent rainfall events [441. Theuse of spatial filtering techniques on multidate "differ-ence" images permits the ready distinction of these twogeneral types of scene dynamics.

(b)

JOOm [IV 100m

1 I10 20 ll) 40

1M, - Mf I

(a)

80o~ /0'"!l-I' f)()X>:: f)Uo

~ 40~--

50

gO

DOBSON AND ULABY: ACTIVE MICROWAVE SOIL MOISTURE RESEARCH

backscattering in minimizing the effects of surface bound-ary conditions (i .e., topographic slope, small-scale rough-ness, and row direction) needs to be examined. The the-oretical backscattering models must be rigorously testedusing data sets that provide sufficient physical characteri-zation of the dielectric and roughness properties of soil.The effects of complex vegetation canopies on radar back-scattering are only marginally understood and need fur-ther study. Finally, the confusion effects of complex geo-graphical distributions of land-cover categories need to bebetter defined on the basis of the analysis of calibratedImagery.

REFERENCES

[I] T. J. Schmugge, P. E. O'Neill, and J. R. Wang, "Passive microwavesoil moisture rescarch," IEEE Trans. Geosei. Remote Sensing, thisissue, pp. 12-22.

[2] F. T. Ulaby, "Radar measurements of soil moisture content," IEEETrans. Antennas Propagat., vol. AP-22, pp. 257-265. 1974.

[3] F. T. Ulaby, J. Cihlar, and R. K. Moore, "Active microwave mea-surement of soil water content," Remote Sensing Environment, vol.3, pp. 185-203, 1974.

[4] F. T. Ulaby and P. P. Batlivala, "Optimum radar parameters for map-ping soil moisture," IEEE Trans. Geosei. Electron .. vol. GE-14, pp.81-93, 1976.

[5] F. T. Ulaby, P. P. Batlivala. and M. C. Dobson, "Microwave back-scatter dependence on surface roughness, soil moisture, and soil tex-ture: Part I-Bare soil," IEEE Trans. Geosei. Electron., vol. GE-16,pp. 286-295, 1978.

[6] M. C. Dobson and F. T. Ulaby, "Microwave backscatter dependenceon surface roughness, soil moisture, and soil texture: Part III-Soiltension," IEEE Trans. Gem·d. Remote Sensing. vol. GE-19, pp. 51-61, 1981.

[71 J. R. Wang and T. J. Schmugge, "An empirical model for the complexdielectric permittivity of soil as a function of water content," IEEETrans. Geosei. Remote Sem'ing, vol. GE-18, pp. 288-295, 1980.

[8] M. Hallikainen, F. T. Ulaby, M. C. Dobson, M. EI-Rayes, and L. K.Wu, "Microwave dielectric behavior of wet soil, Part 1: Empiricalmodels and experimental observations," IEEE frans. Geosei. RemoteSensing, vol. GE-23, pp. 25-34. 1985.

[9] M. C. Dobson, F. T. Ulaby, M. T. Hallikainen, and M. A. EI-Rayes,"Microwave dielectric behavior of wet soil, Part II: Dielectric mixingmodels." IEEE Trans. Geosei. Remote Sensing, vol. GE-23, pp. 35-46, 1985.

[10] M. C. Dobson, F. Kouyate, and F. T. Ulaby, "A reexamination ofsoil textural effects on microwave emission and backscattering," IEEETrans. Geosei. Remote Sensing, vol. GE-22 , pp. 530-535, 1984.

[II] F. T. Ulaby and R. P. Jedlicka, "Microwave dielectric properties ofplant materials," IEEE Trans. Geosci. Remote Sensing, vol. GE-22,pp. 406-414, 1984.

[121 F. T. Ulaby, C. T. Allen, and A. K. Fung, "Method for retrieving thetrue backscattering coefficient from measurements with a real an-tenna," IEEE Trans. Geosei. Remote Sensing, vol. GE-21, pp. 308-313, 1983.

[13] T. J. Schmugge, "Effect of texture on microwave emission from soils,"IEEE Trans. Geosci. Remote Sensing, vol. GE-18, pp. 353-361, 1980.

[14] R. Bernard, P. Martin, J. L. Thony. M. Vauclin, and D. Vidal-Mad-jar, "C-Band radar for determining surface soil moisture," RemoteSensing Environment, vol. 12, pp. 189-200, 1982.

[15] J. R. Wang, "Passive microwave sensing of soil moisture content: Soilbulk density and surface roughness," Remote Sensing Environment,vol. 13, no. 4, pp. 329-344, 1983.

[16] W. P. Waite, A. M. Sadeghi, and H. D. Scott, "Microwavc bistaticreflectivity dependence on the moisture contcnt and matric potcntialof bare soil," IEEE Trans. Geosei. Remote Sensing, vol. GE-22, pp.394-405, 1984.

[17] C. T. Allen, F. T. Ulaby, and A. K. Fung, "A model for the radarbackscattering coefficient of bare soil," Remote Sensing Lab., Uni-versity of Kansas Center for Research, Lawrcnce. KS, AgRISTARSSM-KI-04181, 1982.

35

118] T. T. Wilheit, Jr., "Radiative transfer in a plane stratified dielectric,"IEEE Trans. Geosei. Electron., vol. GE-16, pp. 138-143, 1978.

119] T. LcToan and M. Pausader, "Active microwave signaturcs of soil andvegetation-covered surfaces. results of measurement programs," inProe. ISP Int. Colloq. Spectral Signatures of Objects in Remote Sens-ing (Avignon, France, Sept. 8-11. 198]), pp. 303-314.

[20] H. Hirosawa, S. Komiyama, and Y. Matsuzaka, "Cross-polarized ra-dar backscatter from moist soil." Remote Sensing Environment, vol.7, pp. 211-217.1978.

[21] G. A. Bradley and F. T. Ulaby, "Aircraft radar response to soil mois-ture," Remote Sensing Environment, vol. II, pp. 419-438.1981.

[22J T. J. Jackson, A. Chang, and T. J. Schmuggc, "Aircraft active micro-wave measurcments for estimating soil moisturc." Photogrammetr.Eng., vol. 47, pp. 801-805, 1981.

[23] M. C. Dobson, H. J. Eom, F. T. Ulaby. and A. K. Fung, "Activc andpassive microwave sensitivity to ncar-surface soil moisture and fieldflooding conditions," presented before Microwave Signatures in Re-mote Sensing, URSI Commission F (Toulouse, France), Jan. 16-20,1984.

[241 A. K. Fung and H. J. Eom, "Coherent scattering of a spherical wavefrom an irregular surface," IEEE Trans. Antennas Propagat .. vol.AP-31, pp. 68-72, 1983.

[25] F. T. Ulaby, R. K. Moore, and A. K. Fung, Microwave Remote Sens-ing: Active and Passive, vol. Ill. Dedham, MA: Artech House, 1985.

[26] F. T. Ulaby and J. E. Bare, "Look-direction modulation function ofthe radar backscattering coefficient of agricultural fields," Photo-grammetr. Eng.. vol. 45, pp. 1495-1506, 1979.

[27] F. T. Ulaby, F. Kouyate. A. K. Fung, and A. J. Sieber, "Backscat-tering model for a randomly perturbed periodic surface," IEEE Trans.Geosei. Remote Sensing, vol. GE-20. pp. 518-528, 1982.

[28] A. J. Blanchard and A. T. C. Chang. "Estimation of soil moisturefrom Scasat SAR data," Wclter Res. Bull., vol. 19, pp. 803-810, 1983.

[29} T. F. Bush and F. T. Ulaby, "An evaluation of radar as a crop clas-sifier," Remote Sensing Environment, vol. 7, pp. 15-36, 1978.

[30] E. A. W. Attema and F. T. Ulaby, "Vegetation modeled as a watercloud," Radio Sei .. vol. 13, pp. 357-364, 1978.

[3]] F. T. Ulaby, C. T. Allen, G. Eger. III, and E. Kanemasu, "Relatingthe microwave backscattering coefficient to leaf area index," RemoteSensing Environment, vol. 14, pp. 113-133, 1984.

132] H. J. Eom and A. K. Fung, "A scatter model for vegetation up to Ku-band," Remote Sensing Environment, vol. 15, pp. 185-200, 1984.

133] R. H. Lang and J. S. Sidhu. "Electromagnetic backscattering from alayer of vegetation: A discrete approach," IEEE Trans. Geosei. Re-mote Sensing, vol. GE-21, no. I, pp. 62-71,1983.

134] T. Mo, T. J. Schmugge, and T. J. Jackson. "Calculations of radarbackscattering coefficient of vegetation-covered soils," Remote Sens-ing Environment, vol. 15, pp. 119-133, 1984.

[35] F. T. Ulaby, A. Aslam, and M. C. Dobson, "Effects of vegetationcover on the radar sensitivity to soil moisture," IEEE Trans. Geosei.Remote Sensing, vol. GE-20, pp. 476-481, 1982.

[36] F. T. Ulaby and E. A. Wilson. "Microwave attenuation properties ofvegetation canopies," IEEE Trans. Geosei. Remote Sensing, vol. GE-23, pp. 746-753, Sept. 1985.

137] C. T. Allen and F. T. Ulaby, "Modeling the polarization dependenceof the attenuation in vegetation canopies," in 1984 IEEE Int. Geosei.Remote Sensing Symp. (lGARSS'84) Dig. (Strasbourg, France), Aug.27-30. 1984.

138] D. R. Brunfeldt and F. T. Ulaby, "Measured microwave emission andscattering in vegetation canopies," in 1983 IEEE Int. Geosei. RemoteSensing Symp. (lGARSS '83) Dig .. vol. II (San Francisco, CA), Aug.31-Sept. 2, 1984.

[39] F. T. Ulaby, G. A. Bradley. and M. C. Dobson, "Microwave back-scatter dependence on surface roughness, soil moisture. and soil tex-ture, Part II: Vegetation-covered soil," IEEE TlmlS. Geosei. RemozeSensing, vol. GE-17, pp. 33-40. 1979.

[40] F. T. Ulaby, B. Brisco. and M. C. Dobson, "Improved spatial map-ping of rainfall events with spaceborne SAR imagery," IEEE Trans.Geosei. Remote Sensing, vol. GE-21, pp. 118-122, 1983.

[41] F. T. Ulaby, M. C. Dobson. J. Stiles, R. K. Moore, and J. C. Holtz-man, "A simulation study of soil moisture estimation by a space SAR,"PholOgrammetr. Eng .. vol. 48, pp. 6545-6560, 1982.

[42] M. C. Dobson, F. T. Ulaby, and S. Moezzi, "Assessment of radarresolution requirements for soil moisture estimation from simulatedsatellite imagery," Remote Sensing Lab., Univ. of Kansas Center forResearch, Lawrence, KS, 1982, RSL Tech. Rep. 551-2.

36 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING. VOL. GE-24 NO. I. JANUARY 1986

1431 M. C. Dobson, S. Moezzi, F. T. Ulaby, and E. Roth, "A simulationstudy of scene confusion factors in scnsing soil moisture from orbitalradar," Remote Sensing Lab., Univ. of Kansas Center for Research,Lawrence. KS, 1983, RSL Tech. Rep. 601-1.

{44] B. Brisco, F. T. Ulaby, and M. C. Dobson, "Spaeeborne SAR dalafor land-cover classification and change detection," in 1983 1f:f:E lilt.(;eosci. Remote Sellsing Symp. (lGARSS'83) Dig (San Francisco, CA),Aug. 31-Sept. 2, 1983.

*

M. Craig Dobson (M'83) was born in Rochester,NY, on October 25, 1951. He received the B.A.degree in anthropology and geology in 1973 fromthe University of Pennsylvania and the M.A. de-gree in geography in 1981 from the University ofKansas.

He has been a Senior Associate Research En-gineering with the Departmcnt of Electrical En-gineering and Computer Science at the Universityof Michigan sincc September 1984. Prior to that,hc was an Associate Research Scientist with the

Remote Sensing Laboratory at the University of Kansas Center for Re-search, Inc. For the past ten years, his research interests have been in thearea of microwave remote sensing of renewable resources and include thedielectric properties of natural media, baekscattering and emission fromterrain, and radar image analysis and simulation.

M r. Dobson is a member of the American Society of Photogrammetry.

l<'awwaz T, U1aby (M'68-SM'74-F'80) was bornin Damascus, Syria, on February 4, 1943. He re-ceived the B.S. degree in physics from the Amer-ican University of Beirut, Lebanon, in 1964 andthe M.S.E.E. and Ph.D. degrees in electrical en-gineering from the University of Texas, Austin, in1966 and 1968, respectively.

From 1968 to 1984, he was with the ElectricalEngineering Department at the University of Kan-sas. where he was the J. L. Constant Distin-guished Professor, and the University of Kansas

Center for Research, where he was Director of the Remote Sensing Labo-ratory. He is currently with the Radiation Laboratory and the Departmentof Electrical and Computer Engineering, University of Michigan, Ann Ar-bor. His current research interests involve 1l1iL'rowavc propagation and ac-tivc and passive microwave remote sensing. Along with R. K. Moore andA. K. Fung, he is a coauthor of the three-volume series Microwave RemoteSellsing: Active alld Passive, Reading, MA: Addison-Wesley. In addition,he is coeditor of the Manual of Remote Sensing, 2nd cd., vol. I, AmericanSociety of Photogrammetry.

Dr. Ulaby is a member of Eta i\.appa Nu, Tau Beta Pi, and Sigma Xi.He has been named the Executive Editor for JEEE TRANSACTIONSON GfO-SCIENCEANIlREMOtE SU';SING, 1984-1985, and was the Geoscience and Re-mote Sensing Society's Distinguished Lccturer for 1984. Hc was named anIEEE Fellow in 1980 "for contributions to tbc application of radar to remotcsensing for agriculture and hydrology," received the GRS Society's Out-standing Service Award in 1982, and its Distinguished Servicc Award in1983. In 1984. he also receivcd a Presidential Citation I,ll' Meritorious ser-vice from the American Service of Photogrammetry. He reccived the Uni-versity of Kansas Chancellor's Award J()r Excellence in Teaching in 1980,the University of Kansas Gould Award I'lf "distinguished service to Ilighcrcducation" in 1973, and the Eta Kappa Nu MacDonald Award as an "out-standing electrical enginecring professor in thc Unitcd States of America"in 1975.

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SEI\:SING. VOL. GE-24 NO. L JANUARY 1986 37

Soil Water Modeling and Remote SensingTHOMAS 1. JACKSON

II. SOIL WATER BUDGET MODELING

SM, = SM, _ 1 + P - R - L - E - T + C - Q

Soil water modeling was included as part of a surveyconducted by Schmugge et at. [2] at the outset of theAgRISTARS project. They reviewed all methods thatcould be used to estimate soil moisture including in situ,remote sensing, and modeling.

The basic conservation of mass equation describing soilmoisture is:

3) integration of remotely sensed data and soil watermodels;

4) profile soil moisture from surface layer measure-ments; and

5) estimating soil water properties.

Although the primary purpose of this review is to pre-sent the results of AgRISTARS funded research, it willinclude several studies that were conducted independentlybecause they clearly address the same objectives. In somecases these independent efforts have produced more rele-vant results than the AgRISTARS-supported research.

is the soil moisture volume at time t;is the soil moisture volume at previous time;is the precipitation;is the surface runoff;is the net lateral subsurface outflow;is the evaporation or condensation;is the transpiration;is the capillary rise from lower levels; andis the percolation.

SMrSMr-1PRLETCQ

where

Schmugge et at. [2] found that the published soil mois-ture models varied in the level of detail used to representthe physical system and the temporal definition of the driv-ing forces. The important differences between models theyidentified were the 1) method used for computing the po-tential evapotranspiration, 2) method used for computinginfiltration and runoff, 3) temporal definition of evapora-tive demand and precipitation, 4) consideration of satu-rated and unsaturated levels, 5) number of soil layers used,6) method used for computing soil evaporation and planttranspiration, and 7) consideration of the thermal prop-erties of the soil system.

The state of the art in soil water profile modeling at thetime of the Schmugge et at. [21 survey included models

Manuscript received June 1, 1985; revised August 20, 1985.Thc author is with the U.S. Department of Agriculture. ARS Hydrology

Laboratory, Beltsville, MD 20705.IEEE Log Number 8406229.

1. INTRODUCTION

SOIL WATER modeling research was one of the prin-cipal areas of the AgRISTARS Soil Moisture Project.

Much of this program was adopted from the Plan for Re-search for Integrated Soil Moisture Studies (PRISMS) de-veloped by the Soil Moisture Working Group [I]. Therewere three research directions related to soil moisturemodeling identified in this report:

I) Evaluate the models that are currently in use becausethese will help define the type of data that is needed.

2) Develop new models or adapt existing models to bet-ter utilize remotely sensed data.

3) Since remote sensing methods of estimating soilmoisture provide information on the surface layer,develop methods for extrapolating this informationthrough the root zone.

In addition, it was also recognized that remote-sensingsoil-moisture research could be conducted using simula-tion models and that there was a need for the developmentof sophisticated physically based models with this capa-bility. Another problem related to modeling was how todeal with the spatial variability of large sensor resolutionunits when modeling soil water.

Over the course of the AgRISTARS project, all of theproblems mentioned above were addressed. All of themodeling research was pertinent to the program and someresults have had impact on the science of soil water mod-eling. In this review, the results of soil water modelingstudies conducted during the AgRISTARS project will bedescribed. For the purpose of organization, I have subdi-vided the material as follows:

I) reviews and comparisons of soil water budgetmodels;

2) development of sophisticated models for simulationstudies;

Abstract-Soil water modeling research conducted as part of theAgRISTARS Soil :\loisture project was reviewed along with other rel-evant studies. Research was categorized as follows: reviews of models,development of simulation models, integrating remotely sensed data andmodels, surface versus profile soil moisture, and estimating soil waterproperties. This review and evaluation found that some of the majorobjectives of the program were satisfied and that several of the resultsrepresent significant contributions to the science.

u.S. Government work not protected by U.S. copyright

38IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSJi':G. VOL. GE-24 NO. I. JANUARY 1986

that could simulate water movement and heat transfer verywell using theoretical functions for bare soils. However,root water extraction was and still is very difficult to modeltheoretically and has only been treated empirically.

Kanemasu et at. [3] reviewed all aspects of soil mois-ture as related to crop yield modeling. They found thatmost crop yield models utilized a soil water component ora surrogate variable such as precipitation or API; however,the level of detail used was variable. They recommendedthat in order to evaluate the sensitivity of crop yield modelsto the soil moisture component, it is necessary to use amodel that includes a plant stress component.

The authors concluded from a review of the literaturethat crop yield sensitivity to soil moisture was related toclimate. Under drier climatic conditions, crop yield ismuch more sensitive to soil moisture, presumably becauseunder wetter conditions there is less variability in soilmoisture. They also noted that good estimates of soilproperties such as field capacity were also very important.

Next Kanemasu et at. [3] considered the use of two typesof soil moisture observation information, profile and sur-face. Profile soil moisture would have a variety of uses inmodels. First, it could be used to correct estimates of soilmoisture (updating). This would prevent the compoundingof errors in simulation that might result from errors in theinputs such as precipitation. Second, observed profile soilmoisture could be used to define model relationships. Es-timating parameters such as field capacity can be difficultdue to either the lack of data or spatial variability. By usingcomparisons between observed and simulated values ofsoil moisture, the best value of the parameter could bedetermined. A third use of soil moisture would be in es-timating initial conditions for simulation. Finally, frequentobservations of profile moisture could be used to estimateevapotranspiration (ET) directly. With the exception of ETestimation, the frequency of observation would be be-tween one to three days. In order to use surface moisturefor any of these same applications, much more frequentobservations would be required.

Kanemasu et at. [3] suggest that surface moisture, ifmonitored daily, could provide information for models of:

I) runoff problem areas,2) water erosion problem sites,3) spatial variability of rainfall and subsequent spatial

variations in surface evaporation,4) watershed management,5) mandatory minimum tillage and conservation mon-

itoring,6) remote resolution of irrigation frequency and rates

on larger systems (e.g., center pivots),7) remote resolution of irrigation frequency and rota-

tion patterns in large surface irrigation projects suchas in the Soviet Union and China,

8) pollution and saline deep reclamation,9) environmental impact data (strip mine seepage,

etc.) ,10) planting data models,

II) trafficabi]ity, and12) thermal inertia.

Hildreth [4 J compared the characteristics of eight soilwater profile models. His criteria for selecting these fromthe many described in the literature included: it must beadaptable for multiple locations and crops, some testingmust have been conducted using field observations, andthe computer program must be available.

The models Hildreth [4] evaluated fell into three cate-gories based upon the method used to determine soil waterchanges. These categories were budget, semidynamic, anddynamic. Budget models used an accounting procedurewith empirical functions which approximated the actualprocesses. The dynamic models attempted to utilize theprecise theoretical relationships as much as possible. Thesemidynamic approach falls somewhere between the twoextremes. A summary of the characteristics of thesemodels is presented in Table I. His evaluation also in-cluded a comparative summary of some of the significantfeatures and limiting factors of each model. These arepresented in Table II.

The principal recommendation of Hildreth's study [4]was that all of these models should be evaluated using thesame data set. It was intended that data collected in anexperiment conducted in Kansas would be used; however,Arya and Hi]dreth [5] found that the quality of these datawas very poor and the evaluation was never completed.

One model that has received significant attention as partof the AgRISTARS project is the Soil-Plant-Atmosphere-Water (SPAW) model originally developed by Saxton et al.[6]. Fig. 1 is a flow chart of this model. It incorporatessome physically based and some empirical components.SPAW was later modified to include a means for estimat-ing crop water stress and the effect of stress on the canopydevelopment [7]. This model was then tested over largeareas of several states.

In a related study Calder et at. [8] conducted an eval-uation of 35 different formulations of soil water deficitmodels. These were all evaluated using neutron probe ob-servations of soil moisture at six grassland sites. They be-gan by using a very simple formulation and then made im-provements in meteorological inputs and/or the soil waterfunctions to assess the effects on predictions. Fig. 2 isreproduced from their paper and shows the characteristicsof the models and the expected trend in the accuracy.

Calder et at. [8] found that successive improvements tothe soil characterization resulted in improved predictions,as might be expected. What was somewhat surprising wasthat increasing the detail of the meteorological input didnot always result in improved predictions. The authorssuggest that the daily time step and the low variability ofevaporation demand in the United Kingdom probably af-fect these results.

This study represents a very systematic approach to de-termining the incremental value of additional input dataand mode] complexity on the ultimate decision variable,the soil water deficit. However, part of the problem with

JACKSON: SOIL WATER MODELING AND REMOTE SENSING

TABLE [SUMMARY OF SOIL W liTER PROFILE MODELS (41

39

Type Name Characteristics Potential evapotrans-piration function

Crops So II Input

--------------------------.-----------------------------------~._-Budget CMl

(Hill version)two layers

oSM"P-ETTop layer evaporatesfirst at potential rate

Thornthwaites empiricalfunction; average tem-perature. day 1ength

,Il,veraqevegetation

Va"'; ablefield capacityand wi 1tinqpoi nt u<;erl

P"'ec iP i tilt ion.dailv temDpr-a turF~

(OlTloarps Il1PilSlJrprl d'lrl o"'e-(jicterl SM: (jpterm;nes rliLferencf', plot orf'dictpriSM 'IS. :ieptn. flutputsP:T anrl An.

-------------------~-----------------~-----------~-----------~-----_.•.....-------------------Budg et Baier and

Robert sonsix layers

115W==P-ET-R-Q converts1ayers to zones wi thstandard percent oftotal water

Empirical Spring wheat,soybe~ns ~fallow

Tot~l availablewater, eightdrawndown tahles,field capacity,wiltil1q point.

f:oll1plete mete-OrOl0qicaldata, initialSM, \"'ootrlist ....ibution

Doll V output flf ,l'I,n, DFT,watf'''' h(llanr:e (0mOilrativpdata pm of)sprvprj a"rlprprl icteo S~.

------~-----------------------------------~----------~--------------Budget Feyerherm

six layersSimilar to Baier andRotlert son

Similar to Baier andRobertson

Winter anrl Specific table Similar tospring wheat ann total water Baie'" and

Ro~e •..t son

Simili'lr to 13aiF'r ar>dRohertSOl"l.

------------~---------- ~~""- ------.-----------Budget Kanemasu

five layersSimilar to Baier andRobertson but moreflexible

Energy balance,empirical coefficients,and leaf area index

Wheat, cornsoybeans,sorgl-Jum

Similar toBaier andRohert son

romplete mete- Complete water halance,orol<]qica1 ,I\ET, PET, soil evaporationrlata, more ex- S"" hv layer.tensi ve soi 1anrl 01 ant rlataneerl~rl than Baierand qobertson

----.----------------------------.-----------------------------Budget SIMBAL (Stuff)

10 1ayers(needsupdating)

6.SM""P-ET±(+ti l edrainage

Pan evaporat ion Corn Field capacityand wiltingpoint. Neerlsspecial data

Precipitationand Dan evap-oration, neerlsctf'ot'15 to tileanrl "'latertahl~

Similar to Raier annRobert son witl, morpwater l1al anCf:i nfor'1latinrJ.

-------------------------.--------------------------------_. __._-~------------Semi-dynamic

Saxtonvariable

oSM~P-ET -R+O-CSoi 1 moistureredi stributionace amp 1ished bysimplified dynamicflow equation

Pan evaporation,empirical relationships

Spec i fi erlby tablesor funct ions

Depend s onmoisture-tension andmoisture con-rluctivityfunct ions

Complete mete- romoact rJili 1v outout ofO1"ol<]Q1cal water halance, croodata or pan stress, AET, PET anrlevap0ration, pan evaporation.extensive plantanrl canopy data,much of which isnot rearli1vavai lab1e

----~-~~-----~-~-------_.~----~--------_.~-------~----~--------------_._--------Dynaml.c Hanks

variableASM"P-ET -R<Q-CDynamic water flowand variable timestep, limitedplant capability

Arbitrary. PETor AET

Depenrls onroot andsoi 1 data

Depends on'l1oisture-tension andmoisture con-ductivityfunctions

Precipitation Limiterl w~tel" l1a1ance atand oan evapor- end of eacl, inout pel"lorl:atior}, limiterl plot S"" vs. rleptl-J.root distribu-tion and cropcovel" data

Dynami c Van Savelvariable

oSM~P-ET -R<O-CDynamic water flowequation, dynamicp 1ant-soi l-atmosphereinteraction; veryflexible, uses CSMPIII,shading output, SM 'Is.time and depth alsoavailable

Evaporation only,energy ba 1ance andleaf area index; usesempirical coefficients

Given bycanopy,root andsoi 1 data

Given bymoisture-tens i on andmoisture con-ductivityfunctions used

Complete mete-orolngica1data, detai le~plant soil an<1a tmo<;pner i cdata needed

Flexihlp time i.,crement<;fOr all outout: ea1cu-1ates eroo water use as afunctlon of Cl"OD chil.I1Qeand SM ehanqe

the apparent inconsistency of the conclusions of this studymay be that all of the models used fall under the budgetcategory described above and may not be able to benefitfrom detailed input data.

Jackson et at. [9] considered the impact of soil moistureobservations on hydrologic modeling. As described abovefor crop yield models, they recognized two primary uses-calibration and updating model predictions. They evalu-ated the sensitivity of one model by updating surface soilmoisture using observed data and found that under thetested conditions observed soil moisture was of limitedvalue. However, this study dealt only with the simulationof annual runoff. It is very likely that simulations of dailyrunoff would be much more sensitive to soil moisture.

Peck et at. [10] conducted a survey of the possible useof remote sensing in hydrologic modeling. One of the sub-jects they considered was soil moisture. Six hydrologicmodels were described and the soil moisture accountingcomponent was outlined. In a subsequent study, Peck et

at. [11] considered how remotely sensed data, includingsoil moisture, might be used in these models. Each modelwas evaluated to determine if it could use soil moistureobservations in its current configuration or by minor mod-ification 1) as an input, 2) an update, or 3) to calibrate.

None of the models calibrated by Peck et at. [11] couldcurrently utilize soil moisture observations nor could theybe adapted easily to use the data as input. However, mostof the models could utilize the data for update and cali-bration.

The studies summarized above constitute a fairly goodsurveyor review of soil water modeling and how soilmoisture observations might be used in models and/or ap-plications. However, I do not think that a true evaluationof these models has been completed. The study which hadbeen planned by Arya and Hildreth [5) would probablyhave answered many questions concerning model sensitiv-ity and accuracy as related to soil moisture. As men-tioned, the study conducted by Calder et at. [8] is along

---------------------------------140 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. GE-24 NO. '- JANUARY 1986

TABLE IISOIL WATER MODEL COMPARAIWE SUMMARY FROM [4] ; )'urn OliY"1

----------- -------------- ...------- '---.----._---------

Model Sign; ficilnt features

-------------------~-- --------~-------------~----------

Optlmleed

Fig, 2. Effects or increasing the meteorological and/or soil water functionson the accuracy in estimating the soil water deficit (SMD) from IHI.

tremely difficult to collect data at the level of spatial andtemporal resolution that is needed as input to physicallybased emission models, However, it is quite simple giventhe appropriate soil water model to simulate these data.

In order to be useful in this type of study, the modelshould be soundly based in theory and include all possiblefactors that affect both water movement and heat transfer.Such models should be extremely accurate if the physicalsystem and the input data can be described in enough de-tail.

Van Bavel and Lascano [12] proposed a model for baresoils that could simulate water and energy balance. Las-cano and Van Bavel [13] used a version of this model calledCONSERVB to simulate a set of detailed field observa-tions over a 3D-day period, Their results showed goodagreement between the predicted and observed soil mois-ture. The model was able to predict the measured valuewithin one standard deviation of the observed value. Themodel also predicted the temperature profile very well.

Camillo and Schmugge r 14] also developed a computermodel that simulated both moisture and heat flow in baresoils by solving the partial differential equations describ-ing the physical processes. Comparisons to analytical so-lutions and field observations were used to verify the mod-el's performance. The model was also tested using adetailed bare soil data set [15].

An example of the use of a soil water simulation in con-junction with a microwave emission model was presentedby Camillo and Schmugge [16]. Using their model de-scribed above, the authors simulated the soil and heat bal-ance for a series of soils, initial conditions, and rainfallamounts over time. These detailed descriptions of waterand heat were then used as input to a microwave emissionmodel so that a detailed simulation of emission could begenerated.

IV INTEGRATIONOF REMOTELYSENSED SOIL MOISTUREAND MODELS

All of the research described previously dealt with howthe remotely sensed data could be used to estimate soil

Penman

Constant

Priestley -Taylor

Increa!dng knowledge ot the SOil W8t8_' _

a vallab Iii ty/dral nage characteristic

No root Penman Layer Linear LayerConstant R.C. Exponential

MeanClimatological

'ri

TIUE OF YEAR

q "(.) .(~~Z) (At)

I

Neerl s crop ,1a":i1 not "'Pi!rl i '( ~~i! i 1.'!" 1 ••.•

UsPS pdn e'iilpnr'ijti('lr

SQil -l1r1 fielIsps ""la""nl 'IMflny f'mpi.-ic,ll constants

No prO"ll'r rop anrl sp,:>( if; ~

~ee1s hl'rlr01 'lg ic nil td for Sppci f icsoil

'JPf'rls estim'ltps"lr>ed<;. CilMpy rnO· rlppt_h

j.jot tpstp,1r:ompl(')(UsPs f'e~i hle ~n<1 I'dW

which may 'lot "'(".'1 <1i 1'1

SOIL+-- DEPTH.fut

ACTUAL

~

Uses readily avai 1ablemeteorological rlatil

Tested for several cropsFive layers

Uses readily avail ab'e meteoro-logical data

Requires minimum ;npu~ d.'ltaLong usage

For poorly dnined soilSma1l3mount of input dat.il

needed

Uses readily available meteoro-logical data wi'f:h options touse more if ilYilihble

Takes into account freezingsoil and snow

Used extensively in Cilnilr1a

Includes datil for wide rilngpof soil s

Cc'llculates stressHas feedbackRepresents milny processe~

Dynamic modelbtensi"ely tested ann usedPrints depth p •..ofi1e

ACTVAL ET SOIL AI:~'''"''.m"?l('"':~ACTtAL ET •••••• ~ h g'WITHDRAWAL INFI~RAitON 0-- 0

FROM SOIL .' ~ OIL MOl .•% ~ IL ••••MOISTURE ••••

: .•.----....•.

POTENTIAL ET

JINTERCEPTION

!E&~iwAT/ON_C~~~:J J~-\- TRANSPIRATION

J TIME OF YEAR 1'IT] UNUSED ENERGY

~POTEN. 10 CANOPY ~

o POTENTIALSOIc:"'MOIST2~-6".!.V 00 60 100 TRANSPIRATION," I CANOPY,% •••! 100 PHENOLOGICALSTATE

OTHER +-%0' CANOPY I r----":~;~~~A~~b'N ENERGY SINKS TRANSPIRIN:) ~

I A TIME OF YEAR

•••ooT DISTRIBUTION %

Bilie" andRobertson

CMI(HillVersion)

Feyerherm

Kanemasu

SIMBAL

Saxton

Hanks

Van Ba"el Dynamic anrl ""lexi bleEase of programing an1 lJ~e

Calr:ulates f'"aporation andtrilnspiration di rect1yfrom meteorologicill datil

Se"eral new i1pproaches

Fig. L The SPAW model from 161.

---------------------------------~--------------------------------

the lines of what should have been performed as part ofthe AgRISTARS program.

III. SOIL WATER SIMULATIONMODELS AS A RESEARCHTOOL

Soil water models can be more useful than field obser-vations as inputs to microwave emission models. It is ex-

JACKSON: SOIL WATER MODELING AND REMOTE SENSING 41

SURFACE SOilMOIS TURE

OBSEl1'.'ATrON

IINITIAL r

SURFACE SURf:ACESOil F~UX

MOISTURE

CONVENTIONALAPPROACH INITIAL SOil MOISTURE

PROFilEI

:~~C~61;:~i~ANl IEV APOTRANSPIRA TION I

I I, I

RELATIONSHIP BETWEEN INITIAL IACTUAL EVAPOTRANSPIRATION SURFACE I

AND POTENTIAL ~ __ . _ 4~_

SOil IEVAPOTRANSPIRATION MOISTURE I

AND SURFACE ~OIL MOISTURE 1

I I

, SOil WA\ R MODELACTUAL MOISTURE-TENSION AND_

EV APOTRAN SPI RA TION MOISTURE-CONDUCTIVITY

FUNCTION BY DEPTH

~~NEW SOil MOISTURE

PROFilE.EVAPORA TION

REMOTE SENSINGAPPFiOACH

it

ACTUAL EVAPORATIONOBSERVED SURFACE

SOlll/,CJISTURE

Fig. 3. The general approach used by Bernard £'1 al. 118] versus a conven-tional approach to eslimating bare soil evaporation.

moisture within the framework of existing soil water mod-eling approaches. An alternative to this approach is to de-velop new models that take advantage of the type of infor-mation that remotely sensed soil moisture can provide.Earlier work, such as that conducted by Idso et ai. [17],focused on the use of thermal infrared data in conjunctionwith energy budget models to estimate soil moisture andevapotranspiration. Bernard et al. [18] suggested that ifdirect estimates of surface soil moisture were available,these could be used to determine the actual evapotran-spiration through relationships between potential and ac-tual evapotranspiration or as the upper boundary layercondition in soil water models.

Meylan et ai. [19] point out that difficulties in specify-ing the upper boundary condition are a limiting factor insoil water modeling. They suggested that the use of fre-quent surface soil moisture measurements could improvethe model simulation. These authors outlined a physicallybased water and heat transfer model that would utilizethermal infrared and passive microwave data as inputs.The model would also require measured soil water char-acteristics for the soil and meteorological data (precipita-tion, insolation, temperature, humidity, and pan evapo-ration). From the soil and meteorological observations, thesoil temperature and moisture profiles would be com-puted. Remotely sensed data would then be used to adjustthe simulations and to independently estimate evapotran-spiration. To date, the authors have not published any sim-ulation or test results using this model.

Another French research group has conducted a muchmore comprehensive and thorough evaluation of the use ofmicrowave remote sensing in soil water simulation. Ber-nard et ai. [18] developed an approach for modeling soilwater and estimating evapotranspiration under bare soilconditions. Their method is based on Richards' equationfor one dimensional isothermal water movement. It re-quires as input the moisture-tension and moisture-conduc-tivity relationships and initial conditions, in addition toestimates of surface soil moisture. Fig. 3 illustrates the

general operation of a procedure employing remotelysensed surface soil moisture and compares it to a conven-tional approach. Here the conventional approach used issimilar to that employed in the SPAW model outlined inFig. I under bare soil conditions. The authors evaluatedthis approach by using field observations of all inputs, soilmoisture, and evaporation. Observed surface soil mois-ture was used to simulate microwave observations. Com-parisons between measured evaporation and simulatedevaporation indicated that the model was very accurateunder the tested conditions. These tests conducted on abare light clay indicated that the procedure would be ac-curate for estimating cumulative evaporation if the surfacemoisture was measured once every three days. Daily evap-oration could be determined accurately with two dailymeasurements of surface soil moisture, one in the earlymorning and the other in the early evening.

Prevot et at. [20) followed up on the approach describedabove by conducting a field experiment that would provideall the data required for testing the model. They collectedall meteorological data, soil moisture, and radar data re-quired over a one-month period. The results of this inves-tigation showed that the remote sensing approach couldbe used to determine the soil water balance and evapora-tion with the same accuracy as neutron probe methods.The authors found that this approach actually worked bet-ter if rainfall observations were not used. They attributethis to the structure of the model and the radar measure-ment frequency.

Smith and Newton [21] also developed a physically basedsoil water simulation model that utilizes remotely senseddata. Their general approach is similar to that of Bernardet at. [18] in that microwave emission is used to estimatesurface soil moisture which in turn is used to determinethe surface flux that drives the model. The focus of thiswork was to predict profile moisture; further details areprovided in the next section.

The general approach used by Bernard et at. [18] showsa great deal of original thinking. The basic concept in-

42 IEEE TRANSACTlOJ\S ON GEOSCIENCE AND REMOTE SENSING. VOL. GE-24 NO. I, JANUARY 1986

volved is not that complex; however, the linking of all thecomponents into a prediction process is. A major draw-back to all of the research conducted to date is that theuseful procedures are designed t()r bare soils. These tech-niques should be expanded to vegetated fields .

V. ESTIMATING PROFILE SOIL MOISTURE FROM SURFACE

LAYER MEASUREMENTS

Remote sensing of soil moisture usually provides ameasure of surface moisture. If these techniques are to beof value in applications, methods must be developed topredict profile soil moisture using the surface data. Thesolution to this problem was identitied as major objectivein AgRISTARS Soil Moisture project. Although some ef-forts have used simple procedures, most have involved theuse of soil water modeling.

Jackson [22] pointed out that the problem could besolved using several levels of data collection and modelsophistication. A single surface layer measurement everyn days represents the simplest sensor configuration and theleast costly alternative. From this option the frequency ofsurface measurements can be increased, ancillary mete-orological data can be used, and sophisticated models canbe employed. All of these alternatives for improvementwill, of course, involve greater costs.

The simplest approach to the problem is to develop aregression equation to predict profile soil moisture fromsurface layer measurements. Biswas and Dasgupta [23]presented the results of one investigation using data col-lected at depths of 7.5, 15, 30, 45, and 60 cm. They foundgood correlations in almost all cases. The coefficient ofdetermination decreased as the depth of the profile layerfrom the surface increased. One problem they detectedwas that there was no clear relationship between theregression parameters and site variables.

Blanchard [24] evaluated linear correlations betweensoil layers on 16 small watersheds located in Oklahoma.Data were available every two weeks for several years. Be-cause the data were collected using neutron probes, thefirst surface layer available was 0-22.8 cm. His resultsalso showed that this approach produced high correlationsbetween the surface layer and the profile moisture to adepth of 50 or 60 cm, although the decrease at deeperdepths was very small. He found better results on range-lands than on croplands.

Smith and Newton [21] used a data set generated by adetailed simulation model to evaluate the relationship be-tween different surface layer thickness soil moistures andthose in a 100-cm profile. They found that this approachworked well during a dry down and would be better ap-proximated using a nonlinear function.

Arya et al. [25] evaluated the regression approach usingfield observations of surface and profile soil moisture. Fig.4 shows the results for different field conditions that wereobserved for the 0-5 em and 5-45 em soil moisture val-ues. It is apparent in these figures that the surface soilmoisture exhibits a very large dynamic range while theprofile soil moisture exhibits much smaller variations. This

18

jFAllOW FIELDS

18

14

12 ,..10 e••• r . ! ,. ..

!\. ..

8 ..6 Y'7.6·2.5X

R'· 0.224

SV' 2.0cm2 N-811E

u0

ri 18w>- WHEAT STUBBLE FIELDS< 18oJ

E 140I

'" 12..0 10l-

I .. ,"'. '.

'" 8

~ .•.. ..8 ,..... -cr '..' .W Y-6.0+2.1XI- 4 '. R2- 0.28<

== 2 Sv- 1.8cmoJ • N-77< aI-0 18I-

18 IRRIGATED CORN FIELDS

14 ~.12

" ...-I,.

10

8, ,

6Y-8.1'3.QX

4 Rt -0.44

Sv-l.9cm2 N -46'

0, ,

0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.8 1.8 2.0

TOTAL WATER IN O-'TO 6-em lAYER. em

Fig. 4. Relationship between the 0-5-cm and 5-45-cm soil water as a func-tion offield cover from [25].

is more significant in fields that are dominated by soilevaporation, such as fallow and stubble fields. The cornfields show a much stronger relationship. Fig. 5 is adaptedfrom Arya et al. [25] and shows the correlation betweenthe surface layer of a given thickness and a varying profiledepth soil moisture. Results are presented separately forfallow and corn fields. This figures illustrates the follow-ing points:

1) Correlation decreases as the profile depth increases.2) The correlation between the surface and profile

moisture is larger for planted fields than for fallow fields.3) Increasing the thickness of the surface layer im-

proves the relationship between the surface and the profilemoisture.

The reason why simple regression relationships can beused, under some conditions, to predict profile moisturefrom surface layer measurements is that the laws of phys-ics link all layers of the soil together. If these relationshipsare approximately linear then the regression approach willwork. Kondratyev et al. [26] approached the problem with

JACKSON: SOIL WATER MODELING AND REMOTE SENSING 43

0.8o 4 8 12 18 20 24 28 32 38 40 44

PROFILE DEPTH, em

Fig. 5. Correlation between various surface depth soil moistures and pro-file depths from [25].

a more sophisticated yet simple approach. They used anequation that for a given soil moisture described the gra-dient of soil moisture with depth. This gradient is relatedto the wilting point profile of the soil. The authors reportthat the method has been tested and performed very well.A recent report [27] states that the method is being usedoperationally in conjunction with passive microwave re-mote sensing.

Jackson [22] developed and numerically evaluated amethod that was based on the assumption that the soil pro-file was in the state of hydraulic equilibrium. Under thisassumption, the laws of physics specify that all points inthe soil column must have the same hydraulic potential,which is made up of its matric potential and gravitationalpotential (i.e., depth below the surface). By specifying themoisture-tension (matric potential) relationship for eachsoil layer, the observed surface soil moisture can be usedto predict the profile soil moisture.

Using a detailed soil moisture simulation model andrepresentative meteorological and soils data, Jackson [22]simulated data to evaluate his approach. His results, basedupon the standard error of estimate in predicting the pro-file soil moisture, showed that the accuracy of the predic-tions increased at the thickness of the surface layer i.n-creased. However, beyond approximately 10 cm theimprovement was marginal. Other analysis showed thatthis approach worked better on soils with higher waterconductivities (i.e., sands). Another series of simulationsanalyzed the effect of time of day on the predictions. Fig.6 shows the results obtained. As we would expect, theapproach works best when the soil is most likely to be inhydraulic equilibrium (predawn).

Smith and Newton [21] evaluated the method proposedby Jackson [22] using a data set simulated by a detailedmodel and an actual set of observed meteorological datacollected over a 30-day period. They found that the soilmoisture in at least the surface 20 cm could be accuratelypredicted using the surface 0-5 cm average soil moisture.

25

------- 2400-0600-- - - - -- 0600-1-200

1200-1800--- --- 1600-2400

oo

.08

a: .06oa:a:w

~ .04<oz<t-en .02

Poor results were obtained for predictions of the 20-50 cmsoil moisture. It should be noted that Smith and Newton[21] did not consider the effects of time of day on the re-sults. Improved predictions would be found if predawnsurface soil moisture was used.

All of the methods discussed so far for predicting profilesoil moisture have been very simple and are based on re-lating the surface moisture at that instant in time to theprofile moisture. If the data are available on a frequentbasis, the change in surface soil moisture provides addi-tional information for assessing the profile moisture. Aryaet al. [25] developed a relatively simple algorithm thatcombines the rate of change of the surface moisture andflux at the bottom of the surface zone, estimated from soilhydraulic properties, to determine the profile soil mois-ture. The one drawback to this procedure may be that itrequires a moisture gradient in the surface zone (0-5 cm).The authors propose that if the 0-5 cm soil moisture wasavailable, the 0-2 cm moisture could be predicted usingan empirical equation. From the 0-2 and 0-5 cm soilmoistures, the gradient could be approximated. Compar-isons between predictions made using this approach andboth field and simulated data showed good agreement.

In a previous section, the model developed by Bernardet al. [18] was described. This method can also predictprofile moisture. It utilizes the change in surface moisturein conjunction with a soil water simulation model. Smithand Newton [21] utilized a simila~ but more sophisticatedsurface flux-modeling approach. They modified a physi-cally based soil water simulation model so that it wasdriven by the surface layer soil moisture. This approachwas based on hourly observations of surface moisture andit is unlikely that such a sophisticated approach wouldwork with less frequent observations.

Unfortunately, all of these more sophisticated methodshave only been evaluated for bare soils. Vegetation pre-sents a problem in soil water modeling because of the dif-ficulty in specifying root water extractIOn. Camillo andSchmugge [28] concluded a study to determine if profilemoisture could be predicted from surface measurementsin vegetated fields. They found that the matric potential

5 10 15 20SURFACE LAYER THICKNESS (em)

Fig. 6. Error of estimating profile soil moisture from surface soil moistureas a function of the time of day from [22].

------ FALLOW FIELDS

---- IRRIGATED CORN FIELDS

SURFACE ZONEDEPTH, em

A 0-1B 0-5C 0-15

a:....~ 0.9zwuu..u..wou 0.8zo...<...JWa:g; 0.7u

1.0

44 IEEE TRANSACTlOI\:S ON GEOSCIENCE AND REMOTE SENSING, VOL. GE-24 NO. I, JANUARY 14X6

0 100

51

10

20

30SIGl

40

>I';: 50<{

-'060

70

60

90

100

(a)

0 10 20 30 400

10

20

30SICl

>I'

>- 50<{-,U

C

----70-

80

90

100

0.5% ORGANIC MA TTER

0.0 '!bPOAOSITY CHANGE

0.5% ORGANIC MA TTER

0.0 % POROSITY CHANGE

(b)

Fig. 7. Estimating (a) O.3-bar and (b) IS-bar volumetric soil moisture asfrom the percent sand and clay from 1311.

profile could be predicted from the root density profile.Once vegetation reaches maturity, the root density profileis constant. Given the root density profile, the surface soilmoisture, and the profile hydraulic properties, the profilesoil moisture can be predicted, Limited tests with labo-ratory and field data indicated that the method has poten-tial.

VI. ESTIMATING SOIL WATER PROPERTIES FOR

MODELING

A recurring problem in all the research dealing with re-mote sensing of soil moisture is accounting for the effectsof soil properties, The most useful types of information inmost cases are the relationships between soil moisture and

matric potential (suction or tension) and soil moisture andhydraulic conductivity, Moisture-matric potential func-tions can be used to estimate widely used model parame-ters, such as field capacity and wilting point. Over thecourse of the AgRISTARS project a number of significantresults were published that describe methods for estimat-ing soil water characteristics without resorting to sam-pling.

Clapp and Hornberger [29) published one of the firstgeneralized approaches for determining soil water char-acteristics. Their approach only requires the soil texture.However, it does not consider the effects of factors suchas bulk density and it utilizes empirical equations.

Arya and Paris [30] developed a procedure for predict-

JACKSON: SOIL WATER MODELING AND REMOTE SENSING

ing the soil water characteristics that utilizes particle sizedistribution and bulk density as inputs. Their approach isfounded in theory but requires the acceptance of severalassumptions concerning the shape and size of particles,especially in the clay fraction. Tests against laboratorydata showed good agreement between the predictions andlab results.

Rawls et at. [31J developed a series of regression equa-tions for predicting the soil moisture at specific matric po-tential values. These equations can be used with 3 differ-ent levels of input data:

1) Percent sand, silt, and clay, organic matter, and bulkdensity.

2) Same as I, but it also uses a measured IS-bar watercontent.

3) Same as 2, but it also uses a measured 0.33-bar watercontent.

In addition, this publication also summarizes the averagevalues of the following soil properties by texture classbased on over 5000 samples: total porosity, 0.33- and 15-bar soil water values and the saturated hydraulic conduc-tivity. Fig. 7 from Rawls et at. [32] is a simple graphicalprocedure for estimating the 0.33- and IS-bar matric po-tential water contents based on the sand and clay fractionsand porosity. In this same paper the authors also presenta method for accounting for different tillage treatments onsoil water characteristics.

Another soil property of interest is the bulk density.Rawls [33] presented a method for predicting the bulkdensity from the particle size distribution and the organicmatter content for natural undisturbed soils. Average val-ues of organic matter content for different soil textureclasses were summarized.

VII. SUMMARY

Prior to the AgRISTARS Project, the basic questionsrelated to remote sensing of soil moisture and soil watermodeling were: 1) What type of data do we need? 2) Whatcan we do with the projected data products? 3) How cansurface measurements be extrapolated to develop profilesoil moisture?

The first question concerning what type of data do weneed was addressed through reviews of existing soil watermodels and soil water components of crop yield and hy-drologic models. These reviews are useful, however, theyare qualitative. We still need a qualification of the sensi-tivity of these models to their soil water component. Thestudy that was planned but never completed as part ofAgRISTARS [5J would have provided most of this infor-mation. This quantitative comparison should be completedand should take into consideration the approach used byCalder et at. [8].

What can we do with the projected data? The efforts byBernard et at. [18] show that frequent surface soil mois-ture observations could be extremely valuable if new ap-proaches to modeling are developed that take full advan-tage of the data. However, these methods still need

45

additional research to account for all the components ofthe water balance and verification under a wide range ofconditions.

The final questions concerning extrapolating surfaceobservations to the profile was one of the most critical. Ifwe could do this, that data could be used in almost everyapplication. Experimental and theoretical analyses haveshown that useful information on the upper portion of thesoil profile, approximately 40 cm, can be extracted fromsoil moisture measurements.

The research on soil water modeling conducted duringthe AgRISTARS project has brought us closer to the an-swers to the questions listed above. However, there areseveral additional steps that should be taken to finalizethese studies. These include a quantitative comparison ofsoil water models and further verifications of surface-pro-file extrapolation techniques. In addition, a significant ef-fort should be made to develop a model, or utilize the samemodel, as that presented by Bernard et at. [18]. This effortshould involve a greater amount of ground data collectionand wider range of conditions than has been evaluated todate.

REFERENCES

[II Soil Moisture Working Group, "Plan of research for integrated soilmoisture studies," NASA Scientific and Technical Information Office,Oct. 1980.

[2] T. J. Schmugge, T. J. Jackson, and H. L. McKim, "Survey of methodsfor soil moisture determination," Water Res. Research, vol. 16, no.6, pp. 961-979, Dec. 1980.

[3] E. T. Kanesmasu, A. Fcyerherm, J. Hanks, M. Keener, D. Lawlor,P. Rasmussen, H. Reetz, K. Saxton, and C. Wiegand, "Use of soilmoisture information in crop yield models," AgRISTARS Rep. SM-MO-00462, July 1980.

14] w. w. Hildreth, "Comparison of the characteristics of soil water pro-file models," AgRISTARS Rep. SM-LO-00490, Jan. 1981.

[5] L. M. Arya and W. w. Hildreth, "Agricultural soil moisture experi-ment: Evaluation of 1978 Colby data collected for comparative testingof soil moisture models," AgRISTARS Rep. SM-Ll-04047, May 1981.

[6] K. E. Saxton, H. P. Johnson, and R. H. Shaw, "Modeling evapo-transpiration and soil moisture," Trans. ASAE, vol. 17, no. 4, pp. 673-677, 1974.

(7] K. E. Saxton and G. C. Bluhm, "Predicting crop water stress by soilwater budgets and climatic demand," Trans. ASAE, vol. 24, no. I, pp.105-109, 1982.

[8] 1. R. Calder, R. J. Harding, and P. T. W. Rosier, "An objective as-sessment of soil-moisture deficit models," 1. Hydrol., vol. 60, pp.329-355, 1983.

[91 T. J. Jackson, T. J. Schmugge, A. D. Nicks, G. A. Coleman, and E.T. Engman, "Soil moisture updating and microwave remote sensingfor hydrologic simulation," Hydro. Sci. Bull., vol. 26, no. 3, pp. 305-319, 1981.

[10] E. L. Peck, R. S. McQuivey, T. N. Keefer, E. R. Johnson, and J. L.Erekson, "Review of hydrologic models for evaluating use of remotesensing capabilities," AgRISTARS Rep. GP-G 1-04102, Mar. 1981.

[II] E. L. Peck, T. N. Keefer, and E. R. Johnson, "Strategies for usingremotely sensed data in hydrologic models," AgRISTARS Rep. CP-G 1-04151, July 1981.

[12] C. H. M. Van Bavel and R. J. Lascano, "CONSERVB, A numericalmethod to compute the soil water content and temperature profilesunder a bare surface," Texas A&M Univ., College Station, TX, Re-mote Sensing Center Tech. Rep. RSC-\34, Dec. 1980.

[13] R. J. Lascano and C. H. M. Van Bavel, "Experimental verification ofa model to predict soil moisture and temperature profiles," Soil Sci.Soc. Amer. J., vol. 47, pp. 441-448, 1983.

[14] P. Camillo and T. J. Schmugge, "A computer program for the simu-lation of heat and moisture flow in soils," AgRISTARS Rep. SM-G 1-04086, May 1981.

[IS] P. J. Camillo, R. J. Gurney, and T. J. Schmugge, "A soil and atmo-

---------------------------~J46

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. GE-24 NO. I. JANUARY 19X6

spheric boundary layer model for evapotranspiration and soil moisturestudics," Water Res. Research, vol. ]9, no. 2, pp. 37[-380, Apr. 1983.

[16] P. J. Camillo and T. J. Schmugge, "Correlating rainfall with rcmotelysensed microwave radiation using physically based models," IEEETrans. Geosci. Remote Sensinf{, vol. GE-22, no. 4, pp. 415-423, July,1984.

[17] S. B. Idso, R. J. Reginato, and R. D. Jackson, "An equation for po-tential evaporation from soil water and crop surfaces adaptable to useby remote sensing," Geophys. Res. Lelt., vol. 4, pp. 187-188, 1977.

[18] R. Bernard, M. Vauclin, and D. Vidal-Madjar, "Possible use of activemicrowave remote sensing data for prediction of regional evaporationby numerical simulation of soil water movement in the unsaturatedzone," Water Res. Research, vol. 17, no. 6, pp. 1603-1610, Dec. 1981.

[]9] P. Meylan, C. Morzier, and A. Musy, "Determination of changes inthe hydrologic and thermal profiles of soil by simu lation and remotesensing," Tech. Trans., AgRISTARS Rep. SM-Jl-00800, Feb. 1981.

[20] L. Prevot, R. Bernard, O. Taconet, D. Vidal-Madjar, and 1. L. Thony,"Evaporation from a bare soil evaluated using a soil water transfermodel and remotely sensed surface soil moisture data," Wmer. Res.Research, vol. 20, no. 2, pp. 311-3]6, Feb. ]984.

[21J M. R. Smith and R. W. Newton, "The prediction of root zone soilmoisture with a water balance-microwave emission model," Ag-RISTARS Rep. SM-T3-04425, May ]983.

[22] T. J. Jackson, "Profile soil moisture from surface measurements, " 1.Jrrig. Drain. Div. ASCE, vol. 106, no. IR2, pp. 8[-92, June 1980.

[23] B. C. Biswas and S. K. Dasgupta, "Estimation of soil moisture atdeeper depth from surface layer data," Mausam, vol. 30, no. 4, pp.5/1-516, [979.

[24] B. J. Blanchard, "Correlation of spacecraft passive microwave systemdata with soil moisture indices (API), " Remote Sensing Center, TexasA&M Univ., College Station, TX, RCS-3622-2, 1979.

[25] L. M. Arya, J. C. Richter, and J. F. Paris, "Estimating profile waterstorage from surface zone soil moisture measurements under bare fieldeonditions," Water Res. Research, vol. 19, no. 2, pp. 403-412, Apr.1983.

[26] K. Ya. Kondratyev, V. V. Melentyev, Yu. I. Rabinovich, and E. M.Shulgina, "Passive microwave remote sensing of soil moisture," inProc. lIth Symp. Remote Sensing Environ. (Environmental ResearchInst. of Michigan, Ann Arbor), 1977.

127] N. A. Armand, V. N. Oleksich, V. G. Shinkaryuk, and A. M. Shutko,"Remote determination of the moisture content of soils in the irri-gated lands of Moldavic," Gidromekh. Melior (NASA Tech. Transl.),pp. 58-60, 1981.

[28] P. Camillo and T. J. Schmugge, "Estimating soil moisture storage inthe root zone from surface measurcments," Soil Sci., vol. 135, no. 4,pp. 245-264, Apr. 1983.

[29] R. B. Clapp and G. M. Hornberger, "Empirical equations for somesoil hydraulic properties," W,lter Res. Research, vol. 14, no. 4, pp.60[-604, Aug. 1978.

[301 L. M. Arya and J. F. Paris, "A physico empirical model to predictthe soil moisture characteristic from particle-size distribution and bulkdensity data," Soil Sci. Soc. Amer. 1., vol. 45, pp. lO23-1030, 1981.

[3]] W. J. Rawls, D. L. Brakensiek, and K. E. Saxton, "Estimation of soilwater properties," Trans. ASAE, vol. 25, no. 5, pp. 1316-1320, Oct.1982.

[32] W. J. Rawls, D. L. Brakensiek, and B. Semi, "Agricultural manage-ment effects on soil water processes Part I: Soil water retention andGreen and Ampt infiltration parameters," Trans. ASAE, vol. 26, no.6, pp. 1747-1752, Dee. 1983.

[33] W. J. Rawls, "Estimating soil bulk density from particle size analysisand organic matter content," Soil Sci., vol. 135, no. 2, pp. 123-125,Feb. 1983.

*

Thomas J. Jackson reeeived the Ph. D. degree incivil engineering from the University of Mary-land, College Park, in 1976.

He is a hydrologist with the Hydrology Labo-ratory of the U.S. Department of Agriculture, Ag-ricultural Research Scrvices, Beltsville, MD. Hisresearch focuses on the application of remotesensing to watcr management in hydrology and ag-riculture. He has conducted studies utilizingLandsat data in hydrologic modeling and micro-wave 01casurcments for soil moisture monitoring.

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. GE-24 NO. I, JANUARY 1986

Progress in Snow Hydrology Remote-SensingResearchALBERT RANGO

47

Abstrad-Snow hydrology research conducted as part of the Ag-RISTARS Conservation and Pollution Project was reviewed along withother relevant studies. The major areas of emphasis were visible snowcover analysis, snowmelt-runoff modeling, and microwave snow inves-tigations. Results from these areas of investigation were very positiveand contributed greatly to our scientific understanding. Based on theAgRISTARS results, specific components of additional snow researchhave been defined that will permit future operational applications.

I. INTRODUCTION

THE AGRICULTURE and Resources Inventory Sur-veys Through Aerospace Remote Sensing (Ag-

RISTARS) program provided major support for NASA andUSDA snow hydrology research related to remote sensingfor the period 1980-1984. AgRISTARS was the vehicle tocontinue work started in the 1970's that showed the po-tential for remote sensing of snow for hydrologic pur-poses.

Significant knowledge on snow-hydrology remote sens-ing started to be accumulated soon after the first Landsatand NOAA satellites were launched in 1972. The firstsnow parameter of interest to be extracted from satellitedata was the areal extent of snow cover using visible im-agery and photointerpretation methods. Several Federaland state water resources agencies participated in a proj-ect on the operational applications of satellite snow-coverobservations. The satellite snow-cover data were success-fully tested primarily in empirical seasonal runoff esti-mation methods. For example, three years of testing inCalifornia resulted in the reduction of seasonal streamflowforecast error from 15 to 10 percent on three study basins[1]. Potential benefits of these types of improved satellitesnow-cover based predictions across the 11western statestotal $10 million for hydropower and $28 million for irri-gation annually [1], assuming an operational remote-sens-ing capability. These results were based on empirical tech-niques and little testing of snow cover data in hydrologicmodels was performed.

Before AgRISTARS was begun, investigators began toexamine the potential use of snow cover in existing hy-drologic models because of the potential demonstrated inthe interagency project [1]. Only a few models like theStreamflow Synthesis and Reservoir Regulation (SSARR)

Manuscript received May 15, 1985.The author is with the U.S. Department of Agriculture, ARS Hydrology

Laboratory, Beltsville, MD 20705.IEEE Log Number 8406224.

model and the Snowmelt-Runoff Model (SRM) had pro-visions to make use of measurement of the portion of thebasin or elevation zone covered by snowpack. The majordifferences between the two models were simplicity (SRM)and that SSARR would simulate the snow-covered area ifit was not measured, whereas SRM required an actualsnow-cover measurement to operate. Investigators atNASA decided to test the applicability of using satellitedata in SRM and chose two remote basins in the WindRiver mountains of Wyoming. Landsat photointerpreta-tion was used to extract snow extent as a decimal fractionby elevation zone. Snow-cover depletion curves were thenused to derive daily snow-cover values for input to SRMfor simulation purposes. Both seasonal volume and dailystreamflows were simulated quite accurately (96 and 84percent, respectively) [2], [3].

During this early work with visible snow-cover data, ad-ditional research was being conducted to evaluate the util-ity of the microwave spectral region for snow measure-~ents .. All types of data collection were attemptedmcludmg the use of microwave sensors on trucks, air-planes, and satellites. Much of the early truck and aircraftwork was reported in [4], [5]. Most of these studies in-v~lved passive microwave techniques as opposed to activemIcrowave investigations. The utilization of satellite mi-crowave sensors for snow measurements was first at-tempted and reported by Rango et al. [6]. The Nimbus-6Electrically Scanning Microwave Radiometer was used toseparate snow-covered and snow-free areas and to mapsnow-covered area on a continental basis. Additionally, onthe Canadian high plains significant relationships betweendry snow depth or water equivalent and microwave bright-ness temperature were obtained, and the presence of melt-wat~~ in the snowpa~k was easily detected [6]. These earlyPOSItIveresults proVIded the impetus to include microwavestudies with the snow-cover and snowmelt runoff researchin the Conservation and Pollution Project (CPP) of Ag-RISTARS.

II. SNOW COVER AND SNOWMELT RUNOFF RESEARCH

A. Snow Cover DelineationT~e mapping of snow-cover extent was found to be very

effiCIent when photointerpretation was used in the previ-ously reported demonstration project [1]. As a result,photointerpretation of snow cover was employed forAgRISTARS using a zoom transfer scope whenever snow-cover data was needed for modeling. Because it was so


-------------------------------IIIIIIIIl.

48IEEE TRANSACTIONS ON GEOSCIEf'.iCE A'4D REMOTE SEKSING, VOL. GE-24 NO. L JANUARY 1986

-~_.---_._,___._----___._---_____,___"-- ..---_ ..-.-- 0

APR MA Y JUN JUL

1974---------

Fig. 1. Discharge simulation !()r the Kings River basin (3999 km2), Cali-I()mia, using the snowmeit-runoO' model.

C. Indirect Water Equivalent EstimatesAn indirect approach to estimating the snow water

equivalent in a basin is possible using visible imagery anda grid system superimposed over the basin. To achievethis, air temperatures are extrapolated to the mean eleva-tion of each grid unit in the basin, melting-degree days are

2833 :~

2766 ~

1::::"" ~

COMPUTED FLOW 339.9 ~

.1.60%

MEASURED FLOW

:>40COr~ASH_SUTCllrFE R2 0.9122()OC .'. VOLUMETRIC DIFFERENCE Dv

20000 -

1(\000

'''cor~ 14000·

~ 12000~I

~ 10000

o

In AgRISTARS it was decided that SRM utility shouldbe tested using satellite data in a wide variety of basins tosee if further model adaptation work was required. Severalwestern United States basins were selected (that were ofparticular interest to the Soil Conservation Service (SCS))as well as some internationally diverse basins. Three bas-ins were selected and tested by the SCS [121, [13J in theRio Grande basin of Colorado-South Fork of the RioGrande (559 km2

), Conejos River (730 km2), and the RioGrande above Del Norte (3419 km2

). Additionally, theKings River (3999 km2

) in California, the Dischma basin(43.3 km2

) in Switzerland, and the Okutadami River (422km2

) in Japan were selected for testing. Combined withbasins tested earlier the size (area) of the basins of SRMapplication ranged from 2.65 to 3999 km2•

Although the basins tested had a wide diversity of con-ditions and data quality, the original statistical results re-ported for Wyoming [2], [3J held consistent and were ac-tually somewhat improved. For the six basins tested inAgRISTARS, the seasonal volume simulation accuracywas 97 percent (volumetric difference) and the daily flowaccuracy was 86 percent (R 2 value). Fig. I shows the SRMsimulated hydrograph for the 1974 snowmelt season for theKings River.

Consideration of the AgRISTARS results indicates thatsnow-covered area from satellites can be used very effi-ciently by SRM to simulate flow on a wide variety ofmountain snowmelt basins. The results merit the expen-diture of effort to convert SRM from the simulation to theforecasting mode and to link the model to an operationalsnow cover data stream. Finally from the results reported,it is apparent that other watershed models could advan-tageously incorporate snow cover data for improved sim-ulation or prediction.

easy on the basins used, not much effort in the way ofdevelopment of improved snow-mapping techniques wasconducted in AgRISTARS. The digital snow-mapping ap-proach, although potentially useable, was not employed.

Despite the exclusive use of photointerpretation, severalresearch and/or technical problems remain for snow-covermapping. Although generally of minimal importance formost basins studied in the CPP (the exception is theDischma basin), photointerpretation is very difficult toemploy effectively in small basins or in basins with snow-packs that ablate in a discontinuous fashion as opposed toa regular ablation that allows a contiguous snow line to bedrawn, The digital approach would be much preferred insuch situations, The potential for digital snow-mappingtechniques was established in the 1970's (e.g., [7]); how-ever, the approach is not used operationally except in Nor-way with NOAA polar orbiting data [81, There is a definiteneed to develop the digital approach for widespread ap-plication to problem basins as well as for use when a largenumber of basins are studied,

Additionally, cloud cover remains a problem for snowmapping, In order to avoid this hinderance, the optimumapproach would be to develop an active microwave systemfor snow mapping under all-weather conditions. Foster [9]discovered a way to increase the number of useable visibleimage snow scenes (while waiting for development of anall-weather system) as part of the CPP, Because snow hassuch a high reflectivity, it can reflect enough moonlight atnight to allow snow mapping if the satellite sensors arcsensitive enough to low light levels. Nighttime snow map-ping would increase the chances of obtaining cloud-freeimagery, Although Landsat sensors are not sensitiveenough to map snow at night, the Defense MeteorologicalSatellite Program (DMSP) satellites have more sensitivesensors allowing mapping by moonlight as demonstratedby Foster [9J, Using the DMSP capability, it is projectedthat five additional images per month can be obtained thatare suitable for snow mapping, At present, however, spe-cial arrangements are necessary in order to obtain DMSPdata.

The final snow-cover mapping problem is that no com-plete operational delivery system exists for a type of in-formation that has been shown to have numerous provenoperational applications. Much more effort should be de-voted to this seemingly trivial task which has many com-plex aspects.

B, Snowmelt-Runoff ModelingThe SRM was developed by Martinec [toJ and utilized

with visual ground observations and aircraft photographyto obtain the required snow-cover input data, The avail-ability of satellite snow-cover observations permitted ap-plication on much larger and a greater number of basinsthan previously tested. These new applications entailedadapting the model to accept more data input and to runon a larger computer than previously, As a result a usermanual was developed to assist users in widespread appli-cation of the model with remote-sensing data [II],

RANGO: PROGRESS IN SNOW HYDROLOGY REMOTE-SENSING RESEARCH 49

)N--n '-f---'

<-' -~ ~i!f~'I~/.1"--- ~ '),- ~ 1u

- .. ~--

''we :): • \ l\- i If ~-k ""t~ /- ,-J\ ,( -<1 l ~f\

-~ .•.. V !7 .I -..' ( ~ \7 !I!l35 05cm

V OINWOODY

I • i--' SNOW COURSE.? ,t )

~

'"

,L~j t/' I} V.I- V /

.+-: I \J:::<f"" ( /. /. "/ )~ll\ "() v

"" =-: A'rT+ .A,;W_~ J-~ ~j AVERAGE BASIN SNOW WATER~ EQUIVALENT = 27.02cm

(a)

"

AVERAGE BASIN SNOW WATEREQUIVALENT = 29.45cm

(b)

Fig. 2. Isopleths of snow water equivalent in the Dinwoody Creek basin(228 km'), Wyoming for (a) April I and (b) May 1, 1976, obtained froma snowmelt model and Ladsat observations from [14].

calculated, and daily grid snowmelt values are calculatedand summed from the date of estimated maximum snow-pack water equivalent accumulation (near the beginningof the snowmelt period). The end date for summing thesnowmelt totals is the day snow cover leaves the grid basedupon satellite observations. The total snowmelt is cor-rected by subtracting snowfall amounts during the periodto yield the maximum snow water equivalent for each gridunit [14].

This approach was developed and applied on the Din-woody Creek basin (228 km2

) in Wyoming with a I-kmgrid and Landsat imagery. The temperature lapse rate anddegree day factor values were taken from the values usedin SRM runs on the basin. Fig. 2 shows the resulting arealdistribution of snow water equivalent values for April 1and May I, 1976 [14]. Similar maps were produced forApril 1 and May 1, 1974.

The utilization of this method provides areal snow waterequivalent maps that can be used only in retrospectiveanalysis because the final product in only available afterthe snow has disappeared. The product can be used forseveral purposes. Because winter precipitation measure-ments are notoriously in error, the maps could be used tocorrect the measurements for use in water balance studies.Analysis over several years will identify presistent areasof heavy snow accumulation in the basin which could beused to focus watershed management treatment activitiesfor improving water yield. The location of recurring snowaccumulation patterns can also be used as an aid in theestablishment of new point measurement sites, e.g., in theSnoTel network, so that they can be more representativeof conditions in a particular area of a basin. Finally, theareal water equivalent estimates available in this approachcan be used in conjunction with methods under develop-ment for remote measurement of water equivalent using

microwave techniques that will be described in Section IIIof this paper. Because of the spatial capabilities of themicrowave remote sensing measurements, this grid-basedmethod could be used to improve the required ground truthdata by adding appropriate areal information to the scarcepoint measurements.

Ill. MICROWAVE ANALYSIS OF SNOW CHARACTERISTICS

In the microwave wavelength region, several very sig-nificant and important advantages are evident for the snowhydrologist. Emission and reflection of microwave radia-tion from snow and ice surfaces is strongly affected bysubsurface properties, thereby permitting the possibilityof inferring information with depth. The other importantadvantage of the microwave region is that, depending onwavelength, microwave radiation will penetrate clouds andmost precipitation, thus providing an all-weather obser-vational capability. This is very significant in snow re-gions where clouds frequently obscure the surface.

Passive microwave resolution from space is inherentlypoor because of the large antenna sizes required, but im-provements in the next few years are foreseen. Using ac-tive microwave techniques, resolution from space can beas good as 10 m, which is more than sufficient for detailedanalysis of snow and ice properties. Microwave interactionwith snow is extremely complex, especially in the activemicrowave case, and the result is that data interpretationis extremely difficult. This basic complexity is further con-fused by the rapid changes in snow characteristics, suchas crystal size and liquid-water content, that are possibleunder varying climatic conditions. The dielectric con-stants of water and ice are so drastically different that evena little melting will cause a strong microwave response.Because of the uncertainties in the microwave interac-

50 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. GE-24 NO. I. JANUARY 19Kh

,----~-~-----------~---"----_._----------

TABLE [RESPONSE OE MICROWAVE EMISSION OF SNOW DlE TO VARIOUS PHYSICAL

CONDITIONS FROM 1161

tions, significantly more ground information is needed formicrowave snow studies than for comparable visible, nearinfrared, and thermal infrared studies,

Prior to the start of AgRISTARS enough microwavesnow data had been collected and analyzed to establishthat prospects were good for acquiring hydrologicallymeaningful snow information. During AgRISTARS, in-creased effort was devoted to additional data collectionfrom truck, aircraft, and satellite platforms for analysisand use with models. A good survey of truck, aircraft,and satellite microwave snow experiments is given by Fos-ter et al. [15]. Table I from Burke et al. [16] summarizesthe general response of microwave emission due to variousphysical conditions associated with the snowpack. Thetruck and aircraft data collected in AgRISTARS have beenused primarily to develop and verify radiative transfermodels. The passive microwave satellite data from Nim-bus-5, -6, and -7 have been used in empirical studies ofsnow over large areas.

The empirical studies over large areas that were begunon the Canadian high plains before AgRISTARS were ex-tended to similar areas with low vegetation cover in theu.s. and Russia during the project. The Nimbus satellitemicrowave data continued to show significant relation-ships between snow depth and 0.81-cm brightness tem-peratures with R 2 values of the same order (0.70-0.80) asfound in the Canadian study area. The relationships werespecific to each study area and could not be transferredfrom one area to another. In examination of these data, itwas found that where the snowpack undergoes considera-ble freezing and thawing that the horizontal polarizationis better suited than the vertical polarization for detectingvariations in snow depth. The ice lenses, layers, and sur-face crusts associated with the freeze/thaw cycles aretransparent to the horizontally polarized data but tend todampen the vertical polarization response [15 J.

A limiting factor to these empirical studies is the effectof high vegetation on the snow microwave response. Themicrowave brightness temperature of a snow scene in-creases as the vegetation cover density over the snowpackincreases. The emissivity of the vegetation tends to over-whelm the scattering effect of the underlying snow. Hallet al. [17] used a simple model to remove the effects offorest trees from the typical brightness temperature-snow

Pf-)ysical condition

Phy:;ical temperature

510'1'>' particle radius

Snow depth

Snow wetness

Background surface

Microwave emission response

Increases 1inearly as temperature increases.

Decreases as rarlids increases.

Decreases as depth increases uoti 1 saturationdepth which is a function of wavelength.

Increases rapidly i" presence of free water.

Affects the signature for thin and dry snowconditions.

depth regression relationship. The simple model was basedon the percent forest cover and the brightness temperatureof the forest trees in the absence of snow. The residualbrightness temperature after correction was found to becorrelated with the snow water equivalent under the can-opy.

A microscopic scattering model was developed by Changet al. [18] and used to simulate the effect of varying snowwater equivalent on microwave brightness temperature[19]. In the model the intensity of microwave radiationemitted from a snowpack depends upon physical temper-ature, grain size, density, depth or snow water equivalent,and underlying surface conditions beneath the snow. Fig.3 shows model-generated curves for 0.81-cm wavelengthrelating brightness temperature to snow water equivalentas a function of snow grain size [19]. These particularcurves are for a dry snow condition, unfrozen ground, andan incidence angle of 50°. Fig. 4 is a plot of model pre-dictions of snow water equivalent using truck-acquiredpassive microwave data versus measured snow waterequivalent. The agreement is quite good when typical grainsizes from the study site are known. Fig. 5 shows the scat-tering of the Nimbus-? Scanning Multichannel MicrowaveRadiometer (SMMR) 0.81-cm brightness temperature ver-sus snow depth for the Russian test site previously men-tioned [19]. A linear regression technique for relatingbrightness temperature to snow depth yields R2 = 0.75.The data display considerable scatter which is probablydue to inhomogeneity within each footprint and assump-tions made in interpreting the data. The microwave modelwas also used to generate a depth versus brightness tem-perature curve which fits well with the observations andthe empirical relationship in Fig. 5.

The shallow snow depths in the study shown in Fig. 5are indicative of the depths necessary for insulating winterwheat seedlings from extreme winter temperatures prev-alent in wheat growing areas like this one in Russia. Earlyknowledge of the snow depth during the winter months is,therefore, critical in predicting the forthcoming winterwheat yield. Testing of the model in deeper mountainsnowpacks with aircraft and truck (see Fig. 4) data indi-cates that it can also be used in situations to improvesnowmelt runoff forecasts.

Several other results from AgRISTARS are significantfor microwave snow applications. Snow boundaries cangenerally be defined in the 0.81-cm data because of thesharp decrease in brightness temperature when going froma land to snow surface [16]. In estimating snow waterequivalent, the radiation from dry snow is strongly influ-enced by grain size so that some independent means forobtaining grain size estimates needs to be developed. Re-garding the state of the underlying soil, the presence offrozen ground can be detected by using the polarizationratio at about 3-cm wavelength [19]. Because liquid waterin the snowpack coats the snow grains and causes a sig-nificant increase in internal absorption of the microwaveradiation and a decrease in volume scattering or an in-

RANGO: PROGRESS IN SNOW HYDROLOGY REMOTE~SENSING RESEARCH

VERTICAL POLARIZATION250 HORIZONTAL POLARIZATION

51

200 200

:z '2w wa: a:::> ::>>- >-<t <ta: a:w w0.. 0..

::;; 3 mm ::;;w w>- 150

>- 150fj) fj)

fj) fj)

w w 3 mmZ 35mm z>- >-I I<.:J <.:Ja: a: .35 mm

'" 4mm '"

100 100 .4 mm

5 mm

.5 mm

a 50 100SNOW WATER EQUIVALENT lemj

a 50 100SNOW WATER EQUIVALENT lem)

Fig. 3. Microscopic scattering model~generated calculations relating 0.81~cm brightness temperature to snow water equivalent over unfrozen soilfor several crystal radii (incidence angle is 50°) from [19].

50

w~ 30C/)

<:)wG 20

<:)W0:0.. 10

STANDARD DEVIATlON=3.4cm

SNOW GRAIN MEAN RADIUS

.0.35 mrr,

.0.25 mm00.20 mm

30

25

~ 20 ~

'='I>-~ 15<:)

S0zVl 10

,,,,,, .,.. ,,,",'< _-~ REGRESSION LINE" ..

MODELGENERATED---'Ci...;RVE " ,." ., .,

"

---.L... ....L-210 220 230

T81Kl

Fig. 5. Comparison between empirical relationship of snow depth versusbrightness temperature from Nimbus-7 SMMR 0.81-cm data and modelcalculations over Russia from [19].

garding snow water equivalent, empirical relationships canbe derived for specific large, flat, and low vegetation areasand likely used to estimate water equivalent (or depth) withpassive microwave data. The use of radiative transfermodels make the use of the microwave data more widelyapplicable but additional and difficult to obtain input in-formation is required, e.g., snow grain size. Again, theactive microwave region shows significant promise, how-ever, it has not been exploited to any great degree. Thesensitivity of both active and passive microwaves to liquidwater in the snowpack has been shown in numerous ex-

o 10 20 30 40 50MEASURED SWE (em)

Fig. 4. Comparison between microwave model predictions of snow waterequivalent at 0.81 cm and field measurements.

crease in the snow emissivity, the onset of snowmelt IS

easily detected [15].To summarize our capabilities for collecting snow data

with microwave techniques it should be helpful to considerthe major snow characteristics. Also the capabilities areevaluated with a space platform in mind, but it is impor-tant to remember that there are certain advantages to air-craft data collection, primarily improved resolution. Themeasurement of snow areal extent is feasible using passivemicrowaves on a global, regional, or even large catchmentbasis and really needs very little development to be oper-ational. For measurement of snow extent on small basins,active microwave techniques should work equally well,but, thus far, there have been few attempts to acquire rel-evant O.8l-cm radar data from aircraft or spacecraft. Re-

L __200

::.'-......',~. ~J __

240 250 260

52 IEEE TRAT\SACTIONS ON GEOSCIENCE AND REMOTE SENSING. VOL. GE-24 T\(l. I, JANUARY 1986

periments [201. The onset of snowmelt should be observedequally well with either technique. It remains to be seenif quantitative measures of the liquid water in the snow-pack can be extracted. Finally, using multiple wavelengthbands, inference of the condition of the underlying soilseems possible but further work will be required.

So despite many positive results, much snow researchneeds to be done in the passive microwave area but partic-ularly in the active region. How should this important re-search proceed? This topic was comprehensively ad-dressed during AgRISTARS, and a detailed research planwas completed [21]. Starting on this plan of research nowshould permit completion of an operational approach tomicrowave snow measurements by 1995,

IV. CONCLUSION

The results of research on snow hydrology completedduring AgRISTARS have important applications for theorganizations involved, particularly USDA and NOAA.One major application is seasonal and short-term stream-flow estimation in areas with a significant snowmelt-runoffcomponent. Remote sensing can be used now to obtain theareal extent of snow in a basin. and models exist to esti-mate runoff based on the input of this snow cover data.Because of very positive research results in AgRISTARS,it seems that researchers will be able to develop microwaveremote-sensing techniques to measure or detect snowwater equivalent or depth, the presence of liquid water inthe snowpack, and the hydrologic condition of the soil be-neath the snowpack. In order to make forecasts, the re-mote-sensing information will have to be input to an ap-propriate hydrologic model, such as SRM or SSARR,capable of producing seasonal and daily flow estimates.These estimates would be used as input f()r water man-agement in the areas of irrigation, flooding, water supply,and hydropower.

Aside from the uses of remote sensing for crop-condi-tion assessment developed in other AgRISTARS projects,the advances in snow hydrology have direct applicationsfor crop-yield estimates. First, the predicted water supplyfor irrigated areas has a direct bearing on the forecast cropyield in irrigated regions. Second, measurement of onsitesnow water equivalent in agricultural fields can be used tofigure the potential soil moisture recharge resulting fromthe melting snow. This will have a direct effect on cropgrowth in both irrigated and nonirrigated fields. Finally,the depth of snow and associated insulating effect will beimportant for winterkill estimates which comprise part ofcrop-yield estimation.

The results presented in this paper document a verypositive AgRISTARS effort, yet there are real problems ingetting these advances implemented in an operational way.This problem persists despite the fact that an extremelypositive benefit/cost ratio of 75: I for using just snow coverextent in snowmelt-runoff forecasts was shown before thestart of AgRISTARS in a conservative study [11. It ap-pears that no government or private agency is currently

capable of making snow extent data available in an oper-ational time frame or format to capitalize on the benefit/cost ratio. Secondly, no government research organizationseems amenable to mounting the solid research programnecessary to bring the microwave snow capabilities up toan operational level. A modest undertaking over a 5-10-year period would result in a very positive return even ifonly parts of the effort were successful. As stated before,the plan to direct this research already exists. Ag-RISTARS should not be the completion point, but only asuccessful beginning for snow-hydrology research.

REFERENCES

III A. Rango. "Operational applications of satellite snow cover observa-tions." Waler Res. Bull .. vol. 16, no. 6. pp. 1066-1073, ]980.

121 A. Rango and 1. Martinec, "Application of a snowmelt-runoff modelusing Landsat data." Nordic Hydrologv. vol. 10. pp. 225-238. 1979.

131 A. Rango. "Remote sensing of snow covered area for runotf model-ling," in Proc. (hjiml Symp .. IAHS-AISH Pub. 129. pp. 291-297.]980.

141 A. T. C. Chang. A. Rango. and J. C. Shiue. "Remote-sensing of snowproperties by passive microwave radiometry: GSFC truck experi-ment." NASA Conf. Pub. 2]53. pp. 169-]85. 1980.

IS] A. T. C. Chang. J. L. Foster. D. K. Hall. and A. Rango. "Monitoringsnow pack properties by passive microwave sensors on board aircraftand satellites," NASA Conf. Pub. 2153. pp. 235-248. 1980.

16] A. Rango. A. T. C. Chang, and J. L. Foster. "The utilization ofspaccbornc microwave radiometers for monitoring snowpack proper--ties. ,. Nordic Hydrology, vol. 10, pp. 25 ..40. 1979.

171 A. Rango and K. I. ltten, "Satellite potentials in snowcover monitor-ing and runoff prediction," Nordic Hydrologv. vol. 7, pp. 209-230,]976.

18J T. Anderson and H. Odegaard. "Application of satellite data ]()r snowmapping," Norwegian National Committee /(11' Hydrology. Rep.3. 1980.

191 J. L. Foster, "Night-time observations of snow using visible im-agery," /nt. J. Remote Sensing. vol. 4. no. 4. pp. 785-791.1983.

1101 J. Martinec. "Snowmelt·runotf model I(lr stream flow I(>recasts."Nordic Hvdrologv, vol. 6. no. 3. pp. ]45 .. 154. 1975.

III] J. Martinec. A. Rango. and E. Major. "The snowmelt-runofl' model(SRM) user's manual," NASA Ref. Pub. 1100. 1983.

1121 E. B. Jones. B. A. Shafer, A. Rango. and O. M. Frick. "Applicationof a snowmelt model to two drainage basins in Colorado," in Proe.49th Ann. WeslNn Snow Conf'., pp. 43-54. 1981.

I] 3] B. A. Shafer. E. B. Jones. and O. M. Frick. "Snowmelt runolf mod·eling in simulation and !()recasting modes with the Martinec·Rangomodel." NASA Contractor Rep. 170452. 1'182.

1]4] J. Martinec and A. Rango, "Areal distribution of snow water equiv-alent evaluated by snow cover monitoring." Waler Res. Researeh, vo!.17. 110. 5. pp. 1480 .. 1488, 1'181.

1151 J. L. Foster, D. K. Hal!. A. T. C. Chang. and A. Rango, "An over-view of passive microwave snow research and results," Rel'. Geophvs.Space Phvs., vol. 22. no. 2. pp. 195-208. 1'184.

116] H. K. Burke, C. J. Bowley, and J. C. Barnes. "Determination ofsnowpack properties from satellite passive microwave measure·ments." Remole Sensing A·IlI'iron .. vo!. 15, pp. 1-20. ]984.

1171 D. K. Hall, J. L. Foster. and A. T. C. Chang, "Measurement andmodeling of microwave emission rrom forested snowtields in Michi-gan," Nordic Hydrology. vol. 13. pp. 129-138. 1982.

118] A. T. C. Chang, P. Gloersen. T. Schmugge, T. T. Wilheit. and H. 1.Zwally, "Microwave emission from snow and glacier ice," ./. Gla-ciology, vol. 16. pp. 23-29. 1976.

119] A. T. C. Chang, 1. L. Foster, O. K. Hall. A. Rango, and B. K. Hart·line. "Snow water equivalent estimation by microwave radiometry."Cold Regions Sci. 7i'ch .. vol. 5. pp. 259 ..267. 1982.

120] w. H. Stiles. F. T. Ulaby. and A. Rango, "Microwave measurementsof snowpack properties," Nordic Hwlrolog-", 1'01. /2. no. 3. Pl'. /43-/66, 198/.

12]] NASA "Plan of research 1'01' snowpack properties remote sensing(PRS)2." Recommcndations of the Snowpack Properties WorkingGroup. Goddard Space Flight Center, Greenbelt. MO. 1982.

RANGO: PROGRESS IN SNOW HYDROLOGY REMOTE-SENSING RESEARCH 53

Albert Rango received the B.S. and M.S. degreesin meteorology from Pennsylvania State Univer-sity and the Ph.D. degree in watershed manage-ment from Colorado State University.

He has served on the faculty at PennsylvaniaState University and has been a private water re-sources consultant. From 1972 to 1982, he wasemployed by the NASA Goddard Space FlightCcnter in Greenbelt, MD, whcre he was Head ofthe Hydrological Sciences Branch. In 1983, he wasappointed Chief of the Hydrology Laboratory of

the Agricultural Research Service in Beltsville, MD. His special areas ofresearch interest are in the applications of remote sensing to snow hydrol-ogy, soil moisture, and hydrological modeling. His experience includesserving as project leader for two interagency remote-sensing demonstrationprojects on the operational applications of satellite snow-cover data and theutilization of hydrologic land use data in flood flow frequency simulation.At present, he is president-elect of the American Water Resources Asso-ciation and U.S. National Representative to both the International Com-mission on Snow and Ice and the International Committee on Remote Sens-ing and Data Transmission of the International Association of HydrologicalSciences.

54 IEEE TRANSACTIONS O!\ GEOSCIENCE AND REMOTE SENSING. VOL. GE-24. NO. I, JANLARY 19X6

Early Warning and Crop Condition AssessmentResearch

GLENN O. BOATWRIGHT AND VICTOR S. WHITEHEAD

r

Abstract- The Early Warning Crop Condition Assessment Project ofAgRISTARS was a multiagency and multidisciplinary effort. Its mis-sion and objectives were centered around development and testing ofremote-sensing techniques that enhance operational methodologies forglobal crop-condition assessments. The project developed crop stressindicator models that provide data filter and alert capabilities for mon-itoring global agricultural conditions. The project developed a tech-ni1lue for using NOAA-n satellite advanced very-high-resolution radi-ometer (AVHRR) data for operational crop-condition assessments. Thistechnology was transferred to the I<'oreign Agricultural Service of theUSDA. The project developed a U.S. Great Plains data base that con-tains various meteorological parameters and vegetative index numbers(VIN) derived from AVHRR satellite data. It developed cloud screeningtechniques and scan angle correction models for AVHRR data. It alsodeveloped technology for using remotely acquired thermal data for cropwater stress indicator modeling. The project provided basic technologyincluding spectral characteristics of soils, water, stressed and non-stressed crop and range vegetation, solar zenith angle, and atmosphericand canopy structure effects.

1. INTRODUCTION

THE EARLY WARNING and Crop Condition Assess-ment (EW /CCA) Project was one of eight projects in

the AgRISTARS' program. Its mission and objectives (de-velop and test remote-sensing techniques that enhance op-erational methodologies for crop-condition assessment)were in response to the initiatives issued by the Secretaryof Agriculture. The project was managed by the USDA-ARS but was a multiagency multidisciplinary effort inwhich funds and personnel were provided by the USDA-ARS, the USDA-SRS, the NASA-JSC, and the USDC-NOAA. The EW/CCA project conducted basic researchat various ARS research locations and provided resourcesto industry, universities, and other government agenciesfor early warning and crop-condition assessment technol-ogy development.

Early in the program the project concentrated on crop-stress indicator models for monitoring large areas. Laterin the program stress models were evaluated and verified,field experiments were conducted to relate plant stressesto remotely sensed characteristics, studies were pursuedto better understand and utilize NOAA-AVHRR satellitedata, and the transfer of technology to potential users wasemphasized.

Manuscript received May 20. 1985; revised August 20, 1985,The authors arc with the U,S, Department of Agriculture, Agricultural

Research Services. and the NASA Johnson Space Center. Houston. TX77058.


Many of the tasks undertaken by the Houston EW/CCAunit involved bringing existing research technology to apoint where it could be used in an operational environ-ment. In most cases the technology had to be modified andadapted before it could be transferred and implementedfor operational appl ications.

Early warning/crop condition assessment imp]ementa-tion plans were developed each year in anticipation of re-source increases that never materialized. In fact, yearlyresource reductions resulted in the termination of sometasks prior to their completion.

A]though many aspects of the EW /CCA project couldbe emphasized, this paper emphasizes crop-stress indica-tor models, environmental satellite studies, and conditionassessments. Other individual EW /CCA studies are beingdocumented.

II. CROP STRESS INDICATOR MODEL DEVELOPMENT

Meteorologically driven crop-stress indicator modelswere developed or modified [6], [12], r 16], [17], for wheat,maize, grain sorghum, and soybeans. These models pro-vide early warning alerts of potential or actual cropstresses due to water deficits, adverse temperatures, andwater excess that could delay planting and harvesting op-erations. The stress indicator models were intended to bedata filters and alert mechanisms for large-area monitor-ing rather than stand-alone systems. All stress indicatormodels require daily precipitation, maximum and mini-mum temperature, and evapotranspiration estimates as in-puts. The stress indicator models require accurate cropphenology and water budget subroutines because param-eter thresholds are crop specific and crop stage depen-dent. Consequently, they are best executed in conjunctionwith agrometeorological crop models.

The maize stress-indicator model is used to describe pa-rameter and threshold values used by the stress-indicatormodels. Fig. I illustrates specific threshold values for soilmoisture and air temperature. Hazardous alerts are pro-vided when soil water is deficient or in excess and whenhigh or low air temperature exceeds the threshold values,The threshold values are crop stage dependent for bothsoil water and temperature parameters. The models pro-vide running sums for both hazardous and optimum grow-ing conditions. Fig. 2 summarizes the maize stress modeloutput of optimum, adequate, and hazard days during ]979and 1980 seasons for a crop reporting district in Missouri.


BOATWRIGHT AND WHITEHEAD: EARLY WARNIt\G AND CROP CONDITION ASSESSMENT RESEARCH

50

HAZARD40

uo

55

30

28

I

L OPTr:<~M CONDIEONS

SURFACE ZONE HAZARD SURF>lCE ZONE HAZARD

I DOI "';-- h ___ :1 1---\.---HAZARD ,,~-- - - - --.,; 180

w

IOPTIMUM CO:,OITIONS~ 60

co

Ie-~,., I40 Iw

aHAZARD~ 20 HAZARD

1------- -- --,, SU~FACE ZONE ,

I- +-' HAZARD -1'-----.---....,...----,----...---- •....----.1Pre-plantlng Pldrrti;-g -IE~e~g;c; 11-10 Leaf 110-16+ Leaf Sl1k TassellBllster- Cent-Rlpe ?ostrlpe

Dent

CROP GROWTH STAGES

Fig. 1. Parameters and threshold values for a maize stress indicator modelthat proves daily hazardous and/or optimum flags for moisture and tem~perature conditions at various crop phenology stages.

80 D OPT r ~'UM e')

0~ l- E] HAZARD 79u 5Cc 600u 79'i 79 79 79.~

~'"

20~ 40 400cocw 80u 80 2Qc 20=>

79 80

20 2D

c0 79 80

u 40 Total days 132 128 4C

.3 Optimurl 52 20

uAdequate 74 75

w 50 Hazard 6 30.~ 60w

Yi el d (bu/ac) 106 58c~c 80 80wuc

co

PlantingSi 1 k r"at'Jre

Tassel

Fig. 2. Comparison of 1979 and 1980 strcss alcrts from thc maize strcssindicator model for a crop reporting district in Missouri.

In 1979, only 6 hazardous days were recorded whereas, in1980, the model alerted 30 hazardous days. Maize yieldswere 106 and 58 bushels per acre for 1979 and 1980, re-spectively. These results show that there is a strong rela-tionship between model results and maize yield as esti-mated by SRS for the crop reporting district.

The phenology r 16] of the maize model was evaluatedusing ground-based plant-growth observations. The re-

suits (Fig. 3) show that predicted growth stage lagged byabout 8 days during the early part of the growing season.However, during the critical tassel-silk stage the model wasonly about 4 days slow. One should remember that groundobservations were made at the crop reporting district(CRD) level and one would expect that phenology withina CRD would vary up to 7-8 days.

Test results for the phenology component [12] of the

--------------------------------;56 IEEE TRANSACTJONS ON GEOSCIENCE AND REMOTE SENSING, VOL. GE-24, NO. I, JANUARY 1986

~

101E z~NoLo6ICRl

STR6E -2 1r=""'T~'"'T"~nr=~T='=T'~"'T"'~.~,~~~--

o 10 12 14 16 18 20 22

WEEK

Fig. 3. Test results of the maize phenology model. S. W. Crop ReportingDistrict, Missouri.

..

1979 - SIDNEY, MONTANA

+ •

308~

258

112081~ 150j, 1T

1 OO~ER

J5

5BJ'j, ~

o 1,22MRR IIRPR OIMRY 21MRY 10JUN 30JUN 28JUL 09RUG 29RUG 1ZSEP05RUG15JUL25JUN06JUN

1972 - CARRINGTON,NORTH DAKOTA

17MRY27RPR

50

150-:

108

308

1250

28 Jj

13RPR 03MRY 23MAY 12JUN 02JUI. 22JUL 11AUG OIMRY 21MAY 10JUN

SUBSURFACE ~ SURFACE •• REAL. MOOEl - PRECIPITRTJOII 0

09RUG20JUL30JUN

1979 -- MANDAN, NORTH DAKOTA

...250

300

50

M 200!LL! 150MET

• E 100P5

1971 - OA.KES, NORTl-I DAKOTA380

250

M 208T

CL! ISO

1001J<

5]

24MRR

Fig. 4. Solid line shows ground truth, dashed line model results. Soil waterbudget model test results for !(lUr experimental field sites.

wheat stress indicator model suggest that adjustments areneeded at spring greenup. These adjustments can be madeif soil temperature data are available to establish whenspring regrowth begins [4].

The soil water budget model [4J was tested using groundtruth from experimental field plots in North Dakota andMontana. The results (Fig. 4) show that the model tracksprofile soil water fairly well when initializing parameters(surface and subsurface available water holding capacityand initial soil water) were known and accurately set.Other data sets, not shown in this paper, indicate that dur-

ing the hot dry summer periods, the model may overesti-mate the amount of available profile water.

Indicator models for monitoring wheat winterkill [ I] andfor assessing potential wheat yield reductions [13] weremodified and/or developed. Considerable field researchwas conducted to establish the parameters and thresholdvalues.

All models discussed in this section have been trans-ferred to USDA-FAS and are currently being used by theForeign Crop Condition Assessment Division.

Parameter and threshold values were established for po-

BOATWRIGHT AND WHITEHEAD: EARLY WARNING AND CROP CONDITION ASSESSMENT RESEARCH 57

Fig. 5. CorrelationanalysesbetweenLandsatMSSand NOAA-6AVHRRdata.

210

40

180

35

170

30

SOUTHEASTSOUTH DAKOTA

'50

25

DA Y OF YEAR

'30

20

LANDSAT CHANNEL 7 - CHANNEL 5

110

'5

/•...••.../ '/ ...•.......

/ ' .•...

/ "/ ,/ '

/ "/ ,

/ " SOUTH·CENTRAL'r SOUTH DAKOTA. ,,,,•

20

'60

'40

120

'00

~ 80

60

40

20

070 90

Fig. 6. Environmentalsatellite vegetativeindex (EVil trajectoriesduringthe summerof 1981 for twogeographicgrid cells in SouthDakota.

~-'Z 40Z

"xUI 35

N

-'WZZ" 30XU•...~•... 25

~

50

45

coefficients supplied in the header of the digital tapes tochannels I and 2. All pixels with values greater than 25-percent albedo in either channel were screened out and notused in the VIN computation. A second cloud screeningtechnique [11] utilized data from channels 3 and 4.Threshold values computed from these channels improvedthe ability to screen and remove cloud contamination pix-els. A third approach involved the development of a hier-archical classification of NOAA-7 AVHRR data intoclouds, haze, water, bare soil, and vegetation. Channels1 and 2 were used to detect clouds and to classify cloud-free data into water, bare soil, and vegetation. Channels3 and 4 were used to detect the presence of haze. Al-though these techniques improved the ability to screen andremove cloud-contaminated pixels, they were not imple-mented because channel 3 became unstable as the satelliteaged.

In the most recent cloud-screening technique, devel-oped by EW/CCA, threshold values for channels 1 and 2depend on solar zenith angle. Consideration of sun angle

III. ENVIRONMENTALSATELLITEAND SPECTRALSTUDIES

Perhaps the most significant contribution made by theEW /CCA project was the early research associated withNOAA-n environmental satellite data. In 1980, NOAA-6and NOAA-7 Advanced Very High Resolution Radiometer(AVHRR) data were analyzed to relate vegetation char-acteristics and satellite-derived vegetative index numbers.Correlation analysis [7] indicated a good relation betweenLandsat and NOAA-6 data, r = 0.86 (Fig. 5). Environ-mental satellite vegetative index number (VIN = channel2-channel 1) trajectories computed for two geographicgrids in southeast and south-central South Dakota areshown in Fig. 6. Changes in greenness between the twoI,J grids, during the summer of 1981, correspond quitewell to amounts of precipitation recorded at nearbyweather stations. Results like these and problems that de-veloped with Landsat 3, promoted the USDA-FAS to re-quest that the EW /CCA project develop a NOAA-nAVHRR data processor for operational crop-condition as-sessments.

Since 1982, NOAA has provided a weekly worldwidedepiction of a vegetative index based upon research con-ducted by the EW /CCA project.

The EW /CCA research unit developed a four-year database for an established geographical grid covering the U.S.Great Plains. Each cell represents an area of approxi-mately 25 x 25 nautical miles. The grid cells contain veg-etative index numbers (VIN's), daily precipitation,maximum temperature, minimum temperature, evapo-transpiration, and solar radiation. Historical and seasonalVIN trajectories for individual grid cells (Fig. 7) and forcountry and/or crop reporting districts (Fig. 8) suggestthat NOAA-AVHRR satellite data provide valuable infor-mation for local, regional, and global crop condition as-sessment purposes. Data presented in Table I show thatwheat and corn yields for Burleigh, Morton, and Olivercounties in North Dakota were greater in 1982 than in1983. VIN's shown in Fig. 7 indicate that greenness withinthe counties was much greater in 1982 than in 1983. Inthree Nebraska counties, wheat yields were greater in 1983than in 1982. In contrast, yields for corn and soybeanswere greater in 1982 than in 1983. VIN trajectories forgrids within these three counties show that greenness dur-ing the growing season for wheat was greater in 1983 thanin 1982, but during the corn and soybean growing seasonthe reverse was true.

Considerable research was conducted to develop a prac-tical method for automatically screening cloud-contami-nated pixels from NOAA-AVHRR satellite data. The ini-tial cloud screening technique applied calibration

tential wheat stripe rust epidemics and for losses causedby harvest delays [2], [3]. Although these models have notbeen completed, additional information can be obtainedfrom the Pathology Department, Montana State Univer-sity, and from the USDA-ARS Northern Plains ResearchCenter, Mandan, ND.

,58 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. GE~24, NO. I, JANUARY 1986

VIN TRAJECTORIES FOR IJ GRID CEll. 220-349JlIIJ1lIIaJI •••• _1lUCI11I.

VIN TRAJECTORIES FOR IJ GRID CEll. 221-349_0IftUL_1lWJU

VIN TRAJECTORIES FOR IJ GRID CEll. 222-349_CDrJa.oI.._~

VIN TRAJECTORIES FOR IJ GRID CEll. 223-349

'" '"

II' lrt ZI. lJ. inAIUMUT '" II

~

: '., ', ,, ', ,... .. ".., .

I" HI 11. i'1' lJ. mJIlUMOn '"

"II ,. "II

VIN TRAJECTOR~~~G~ CElL 219-350 VIN TRAJECTOR~~~G~ CElL 220-350 VIN TRAJECTORIES FOR IJ GRID CELL 221-350 VIN TRAJECTORIES FOR IJ GRID CELL 222-350'iI:IUJll1DrlWM...-JMlIO.GI1oI _aliTUl.~1lIOoW'-'

"

~

'

, '. ,. ,, ,- ", ,- ,~ .- .

In n. 211

.AlUM DIY

'ltM -_uINl

~

:,~'", ", ,, ,, ,, ,, ,

- .. ,

III 211

MI.UT

n-ltIl

;/". ,, ,. ', ,, '

-' '---

noAlUM DIY

u--ll1~

VIN TRAJECTORIES FOR IJ GRID CEll. 221-366 VIN TRAJECTORIES FOR IJ GRID CEll. 222-366-.uu.~~ ~1IIIm£UItB

VIN TRAJECTORIES FOR IJ GRID CELL 223-366~.......,~ VIN TRAJECTORIES FOR IJ GRID CELL 225-367~.......,~'"

VIN TRAJECTORIES FOR IJ GRID CELL 225-369..•.... .......,~

, "

~: "'"

--' \~------ ,. ,~~,," \

";i--~-~~~~-~-~-~It III It' 11. lit D' iU 2"

.Iltl""DU

VIN TRAJECTORIES FOR IJ GRID CEll. 221-367~.......,~~

/ \ ..•,,- -----.' ...

25-"-'---

It 11. lit m m 2MJIIUM DAr

VIN TRAJECTORIES FOR IJ GRID CEll. 222-367~IUo:JU.IfMU.R(B

,-,~

II

/\\\II III In 1/1 211 ill Z~ 291

VIN TRAJECTORIES FOR IJ GRID CELL 223-367~u~~

III 17' ZII DI '" ,.

II

~. -'". ,, ,. ,, ,

,,' . ,, ,, .

".i---_- ~-~-~~•• H' 111 ll' 2" 2•

.JIUM Ilfn~_ 1112

11. 211 III

.&lUMIA'

--- liIl

,. ~

.'" ... ,,- - -~- ---- . "" ,, ,, ,

Fig. 7. Comparison of 1982 and 1983 vegetative index numbers (V1N's)for individual 25 x 25 nautical mile grid cells in North Dakota and Ne~braska.

VEGETATIVE INDEX TRAJECTORIES VEGETATIVE INDEX TRAJECTORIESHEllRA51O, SOiIIHE4ST CllD

Z9B

196

,,,,,,,,

266

,,,

138 ~68

-- 1983

13B188

,,,,,,,,

178148liB

,,

78

55

88

25

85

78

55

8S

115

188

19B268238288I7a14B

, ,, ,,, ,,,,,,,,

,

liB

---~2S 'r-~~""""'-"""r-'~--r---r~-"'--"'~~T

118 14B 178 188 118 268 298 88 liB 1\8 176 168

JULIAN DAY JULIAN DRY

LEGE1iD: YEA~ --- 1932 -- 1933 LEGE1iD: YER~ ---- 198Z

Fig. 8. Comparison of 1982 and 1983 vegetative index numbers (VIN's)for counties and crop reporting districts in North Dakota and Nebraska.

115

188

85

7!

55

\8

25

88

115

188

8S

78

55

\8

25

88

O~ 100

J~ ~ ~ JUL ~p N~ J~ ~ ~ JUl ~p NOV1981 1982

Fig. 10. Prevailingnadiratmospherictransmission,ground-measured,andNOAAAVHRRminimumdigitalcountsbeforeandafter the El ChichonVolcanoeruptions.

S9

.45

.50z0

.55 Uj'!2

'".61 (I)z«CL

.67 f-

Uno.74 w

ICL(I)

.82 0

~«.90

~5 ~.~

~§

, CWRD aouDY

.8

.6

If-CLW0

.., .4«ue=CL0

.2

COUNTY YIELDS DIFFERENCE GRID CELLBu./Ac. Percent

STATE COUNTY CROP 1982 1983 1982-1983

North Dakota Burleigh Wheat 39.2 33.9 -14% 223-349North Dakota Burleigh Corn 55.0 40.5 -26% 223-350North Dakota Mort on Wheat 26.5 23.1 -13% 220- 350North Dakota Morton Corn 52.2 50.5 -03% 221-350North Dakota 01 i ver Wheat 30.6 24.5 -20% 222- 349North Dakota Oliver Corn 50.0 51.9 +04% 221- 349

Nebraska Nemaha Wheat 29.1 40.2 +27% 225- 367Nebraska Nemaha Corn 96.5 42.7 -56% 225-367Nebraska Nemaha Soybeans 35.1 24.3 -31% 225- 367Nebraska Sa li ne Wheat 27.3 39.7 +31% 223-366Nebraska Sal i ne Corn 117.2 100.3 -15% 223- 366Nebraska Saline Soybeans 34.3 29.3 -15% 223- 366Nebraska Fillmore Whe at 28.3 47.2 +40% 222-366Nebraska Fi llmore Corn 126.6 116.7 -08% 222- 366Nebraska fillmore Soybeans 41.4 38.2 -08% 222- 366

TABLEISTATISTICALREPORTINGSERVICECOUNTYYIELDESTIMATES,DIFFERENCES

INYIELDS(1982-1983), ANDIDENTIFICATIONOFI,J GRIDSLOCATED]I\VARIOUSCOUNTIES11\ NEBRASKAAI\DNORTHDAKOTA

BOATWRIGHT AND WHITEHEAD: EARLY WARNING AND CROP CONDITION ASSESSMENT RESEARCH

Fig. 9. Comparisonbetween the old and the new cloud screening tech-niquesdevelopedby EW/CCA.

automatically corrects for latitude and time of year or sea-sonal changes. This new cloud screening technique sta-bilized I,J grid VIN trajectories and permitted the use ofVIN's from grids when only a few pixels within the gridare retained and used. We recommend that vegetative in-dex numbers not be computed when more than 80 percentof the pixels within the grid are screened out.

Fig. 9 illustrates the improvement achieved when usingthe new technology compared to the original techniquedelivered to the USDA-FAS. This new cloud screeningtechnology has not been published. However, it has beentransferred to the USDA-FAS and is currently being usedin their operational system.

The increase of atmospheric haze caused by volcanicparticulates and reaction products, as measured by theNOAA-7 AVHRR data, is shown in Fig. 10. Ground-ob-served prevailing nadir atmospheric transmission de-creased approximately 11 percent after the El Chicon vol-cano eruption [14]. The decrease in atmospheric

transmission agreed with increases of NOAA-7 AVHRRdigital count minimum values in the visible and infraredbands obtained over the Gulf of Mexico. These resultsdemonstrate the importance of transient atmospheric hazeeffects relative to early warning crop stress monitoring.

The effects of solar illumination, view angle, and non-Lambertian surfaces [9] on NOAA-AVHRR sensor data aresummarized in Fig. 11. Results ofthis study indicated thatuseful greenness information can be derived using up toapproximately 512 pixels either side of nadir. The findingsalso show that solar zenith corrections are not necessarywhen computing VIN's and that increased shadowing withincreasing view angles playa significant role in the inter-pretation of data from non-Lambertian surfaces.

The effects of scan angle were studied using consecutiveday NOAA-AVHRR data. To assure that information col-lected on consecutive days came from the same groundlocation, the data were rectified [5]. A very large data setwas assembled to include data collected throughout theseason from locations across the U.S. Fig. 12 demon-strates the change in channel and VIN values betweenconsecutive day satellite passes, as the scan angle in-creases both east and west of nadir. Data depicted in thelower half of Fig. 12 were used to develop equations thatmodel scan angle effects for the center 1024 pixels of thescan. Individual channel data or computed VIN's can becorrected for scan angle effects. The models are more ef-fective when channel values are corrected before VIN'sare computed.

IV. CROP CONDITION ASSESSMENT

Soil water, when in limited supply or in excess, is amajor factor in crop condition and production. The EW ICCA project was involved in numerous tasks to detect,monitor, and determine the degree of stress associatedwith various soil water conditions for several crops. Re-search tasks ranged in scope from basic research to tech-nology transfer for operational applications.

Canopy temperatures, obtained by infrared thermome-try along with wet- and dry-bulb air temperatures and anestimate of net radiation were used in equations derivedfrom energy balance considerations to calculate a crop

60 IEEE TRANSACTIONS ON GEOSCIEt\CE AND REMOTE SENSIt\G, VOL. GE-24. NO. I. JANLARY 1986

water stress index (CWSI). The CWSI (Fig. 13), closelyparalleled a plot of the extractable soil water in the O-to1.1-m soil zone. Although the CWSI [8] was developedusing wheat plot data. the concepts led to the developmentof an index using NOAA-AVHRR data for large-area ap-pi ication.

The satellite-derivd stress index (SDSI) requires day!night canopy temperature measurements for individualsatellite pixels and air temperature measurements ob-tained from meteorological stations. The stress index iscomputed using the following equation:

DT - ATDT - NT

SDSI(a)

...• --

,,'..

,."w· 0 ~."

Fig. 11. Geometric and solar correction of the NOAA AVHRR data. (a)Radiance values as influenced by scan angle (pixel numher) for channelsI and 2 of the NOAA-6 AVHRR sensor. (b) Geometric relationship be-tween the Sun the AVHRR scanning limits with respect to the Earth. (c)Modeled EVI's (Lambertian surface) and computed NOAA-6 derivedEVl's illustrate the atmospheric and illumination geometry effects. Sim-ulated data were derived using Dave's data set.

//

//

/

e

(b)

e 8 e

{_"lie If!

'''~ ,-'" 1"_1

M(I lllOIlI 1600PUIl_' _

__ ., JYt' 1_

(c)

where

DT is the satellite acquired daytime maXJn1Um tem-perature,

AT is the meteorological station daytime maximumtemperature. and

NT is the satellite-acquired nighttime minimum tem-perature.

The numerator is a function of evapotranspiration. Thedenominator is a function of vapor pressure deficit, andindicator of potential evapotranspiration. The SDSI pa-rameter approximates the pattern of crop water stress in-dex [8]. Where high values of SDSI co-exist with largevalues of VIN's, it is interpreted as an indicator of cropwater stress. At this time the relationship is subjective ininterpretation, but an effort is in progress to quantify theparameter. Fig. 14 illustrates, in image form, the relation-ship between VIN's and the SDSI. Dark portions in thestress image indicate areas of low stress, light portions inthe VIN image indicate areas of high vegetation. The scat-ter plot in Fig. 15 represents a multitude of land uses, suchas water, bare soil, pasture, stream bed vegetation, andvarious crops. Although considerable scatter of data pointsexist, there is a strong correlation between VIN values andthe SDSI. The stress index would provide more consistentresults if the spatial distribution of meterological air tem-perature data were better.

Hot dry winds, such as a sukovey in the U.S.S.R., cancause a significant decrease in yield of spring and winterwheat. The Yield Reduction Model 113] was used overthree major winter wheat producing provinces in the NorthChina Plains. Results from thc model (Fig. 16) suggestthat the potential for yield reduction was greater in 1982than in 1983. Reports from China also verify that pooreryields were obtained in 1982 than in 1983.

Landsat MSS studies were conducted, across the U.S.Great Plains, to determine the feasibility of monitoringrangelands to predict the potential for water stress in ad-jacent croplands. The hypothesis was that water often be-comes limiting in rangeland areas before adjacent crop-lands show symptoms of water deficit. Results from thesestudies suggest that Landsat acquisitions are too infre-quent for reliable prestress indicators for adjacent crops.However, vegetation greenness indexes [10] computed for

BOATWRIGHT A1'D WHITEHEAD: EARLY WARNIN(, AND CROP CONDITION ASSESSMF:-iT RESEARCH 61

lHE. [nlUIl SUllt TI1E Enn~£ SCllnTHE EnmE SCAn

3' 3'l' ,,

18 18

18D[ DL " [ l'T L, T•(

H , V ,• [

" "1 -18

" -18 -18

"

-i" -2' -i"

-55.4 -C7.? B_B 21.7 55.4 -5U -27 .7 '-' 21.7 55.4 -55.4 --27.7 '.B ~7.7 55.'

THE([11m 58 z: Tll[ CErHn~58 1 TllE ITIfTil 58 %

3' l' 3B

18 " 18- 18

D ~Ifft~.[ 0L 18- E 18T

:- ~ - + If- .t-" ~ L

• +•••..•• T.. .... .. .. "'". •( .• .• + .•

H ,-• .. ri~ ""-18

.:~{.r:'-IB.,

-18- -2' -2B-

-21.78 -13.IS 8.118 11.8) ~7. 78 -21.78 -13.85 ILIlIJ !l.8S Z7.7I -27.78 -13.8S 8.1l11 13.115 17.78

mo-sciln Rrl'U MIO-SClln AnGl[ nlO-SCRn AnGLE

Fig. 12. Changes in NOAA AVHRR channel values and vegetative indexnumbers between consecutive day satellite passes as the scan angle in-creases both cast and west of nadir: Center 1024 pixels to be modeled f()rscan angle corrections.

y •.4290" 3,97X - 291X2

r2 _ 0,86

Fig. 13. The CWS[ as a function of Julian days for wheat plot A. whichreceived a single-post emergence irrigation. Circles represent the cal-culated CWSI data points and the hand drawn solid line show data pointtrends. The plus symbols represent the extractable water used from the0- to 1. [om depth. Ordinate values also represent extractable water used. Fig. 14. Comparison of satellite derived vegetative index image and ther-

mal stress indicator image for Grady County. Oklahoma.

xx

SUM OF CW$I

x

x

15

ooo

all Landsat MSS pixels within a 5 x 6 mile segment par-allel the greenness index of only rangeland pixels withinthe same segment. Fig. 17 illustrates the changes ingreenness within and between years and that the overallgreenness of the segment is never significantly differentfrom the greenness of rangeland. Similar results were ob-tained for all segments studied within the U.S. GreatPlains. These results combined with vegetative indicescomputed from AYHRR data for I,.! grid cells (Fig. 7)

suggest that AYHRR data could provide useful informa-tion about overall crop conditions for large areas.

Handheld radiometer studies (15) suggest that plant dis-eases may be monitored from satellite platforms if mois-ture and temperature conditions could be tracked usingmeteorologically driven disease indicator models. De-creases in vegetative index numbers can occur because ofdrought and/or plant disease conditions. Knowledge of

62

18

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. GE,24, NO, I, JANUARY 1986

" 1

68

158

148

T3B

EDU IBCTIa IeN

---* 198.3

r;

I;

;;

II.

;I

I;

II

;.

188148 168

THERMAL STRESS IIIDICATOR

Fig, 15, Scatter plot of satellite derived vegetative index numbers and ther,mal stress indicator values for Grady County, OK,

SF1NK ~OUNTV, SCJiH DilC"JA

B6APR 16APR 26APR 86NAY 16NAY 26NAY 85JUN 15JUN 25JUN

DATE

Fig, 16, Comparison of 1982 and 1983 results of the wheat yield reductionmodel for major wheat growing provinces of China,

4; 90 13; 181 225 270 J15 3bC 45

.• 1 T

; I 1~1:dIll II! lITr1

25j

~::J A ._~1 =/ '~-~

~'JIl:'l' III (rUlm jiI:;!1iI;i1hIhlllJllIu""-1 e ~m' '_~ __ ~~m'

45 9B

1977

18B 125 11e 315 ]sD 45 III

1978~:j p;18~ 1,

15~ iI:j iIiIr I i 1r IT,:fmmn;~tllil I 11ItIILitIlr Tn1~'-

-tB~TT'""""~"""'-'-"-'-'-~T""T~~~~~"'r-T"""""--"'''''-~~~'~

45 9B 135 18B 215 21B 315 JoB

JUlI RN DRY

GREENNESS 'RRrIGElRND (l)

PRECIPITATIUN X 10 (IN).GREENNESS-ENTIRE 5CENE (5 X 6 MIJ ,

DEPRRTURE fROM NDRMRL X Ii (IN) 0

Fig, 17, Seasonal and yearly trajeclories depicting rangeland and total scenevegetative indices for 5 x 6 mile segments in South Dakota and Colo-rado,

BOATWRIGHT AND WHITEHEAD: EARLY WARNING AKD CROP CONDITION ASSESSMENT RESEARCH 63

V. CONCLUSIONS AND RECOMMENDATIONS

1~JUN 2~JUN 04JU~ 14JUl ~4J~'I_ 03RJG 13RUG 23f1'JG

DAE 1982

ACKNOWLEDGMENT

The authors take this opportunity to express their grat-itude to the many government agencies, research scien-tists, agricultural research centers, universities, industry,and the AgRISTARS management team who supported theEarly Warning and Crop Condition Assessment (EW/CCA) Project. The joint cooperation and support fromthese various groups make it difficult to acknowledge spe-cific contributions made by each organization.

However, resources from the USDC-NOAA were di-rected toward satellite-derived products such as solar in-solation, maximum/minimum temperatures, precipita-tion, snow cover, basic research using NOAA-AVHRRdata, and the development of vegetative index numbers de-rived from AVHRR data.

The NASA Johnson Space Center provided resourcesthat were directed toward satellite sensor research, soft-ware development, studies to enhance the util ization ofNOAA-AVHRR data, parameter identification and devel-

FAS as data filters and as an alert tool for monitoring globalcrop conditions.

Environmental satellite research led to the developmentof a NOAA-AVHRR data processor, making it possible forthe USDA-FAS to use the data in their operational crop-condition assessment program. Research increased ourknowledge of the spectral characteristics of soils, water,vegetation, crop stresses, and crop types. Research in-volving the effects of cloud contamination, atmosphere,geometry, canopy structure, solar zenith angle, and scanangle improved our ability to utilize satellite data for ag-ricultural purposes.

Crop-condition assessment studies resulted in the de-velopment of crop water stress indexes based on canopyand air temperature differences obtained from eitherground or space platforms. A yield-reduction model wasdeveloped and transferred to an operational user. LandsatMSS and NOAA AVHRR studies indicate that large-areacrop condition monitoring can be achieved without ex-plicit knowledge of vegetative cover. Widespread plantdisease (stem rust) monitoring is feasible in conjunctionwith a disease susceptibility model that uses meteorolog-ical data inputs.

The authors recommend that research studies continuewhich meet the requirements supplied by users. Effortsmust be continued to enhance and improve the use of thegridded vegetative indices for the U.S. Great Plains. Thisimplies the development of geographically referenced in-formation systems. The scan angle model should be im-plemented so that more of the scan swath can be used. Weemphasize that participants in the EWICCA project weretruly interdisciplinary scientists and that excellent coop-eration was received from all researchers. Continuation ofa multidisciplinary team that works in conjunction withthe user will significantly enhance the research results andwill speed the transfer of technology to operational units.More emphasis and resources should be placed on tech-nology transfer.

2416o

o

.JW

ZZ<l::r:u

CHANNEL 2 ICOUNTSj

Fig. 19. Flood classifier for delineation of water, vegetation, and soils usingNOAA-6 spectral data.

moisture conditions will define which condition exist. Fig,18 indicates that the vegetative index (normalized differ-ence) began to decrease soon after wheat plants were in-oculated with rust spores on June 7 and again on June 18,1982. After July 14, the normalized difference (ND) of thediseased plants decreased almost linearly (from 0.78 to0.40), whereas the ND of nondiseased plants remainedabove 0.80 until August 13 when natural senescence wasin progress. Although yield data are not shown in this pa-per, diseased plots yielded significantly less grain than non-diseased plots.

Plant stress caused by flooding also results in significantcrop damage. Thus, a classifier (Fig, 19) and crop damageestimator was developed for monitoring flooded areas.Large river basin flooding is easily detected and moni-tored using this classifier, however small stream floodingis more difficult because of the AVHRR pixel resolution.

VlI-- 16Z~ou

24

Fig. 18. The effect of stem rust on vegetative index numbers during thegrowing season for wheat, Montana Rust Disease Study, Exotech Radi-ometer.

In the authors' view, the EW/CCA project made majorcontributions to science and to operational users of remotesensing. Crop-stress indicator models were developed thatprovide hazardous and optimal alerts for water and tem-perature factors. These models are used by the USDA-

64 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. GE-24, 1\0, I. JANUARY 1986

opment for modeling winterkill, wheat yield reduction,harvest loss, and potential evapotranspiration,

Resources provided by the USDA-SRS were utilized toestablish individual relations between spectral data andplant stresses, spectral characteristics for stressed andnonstressed crops, stress model development, improvedunderstanding and utilization of AYHRR data, softwaredevelopment, and statistical analyses for many of the re-search tasks.

Resources from the USDA-ARS were divided amongseveral research locations. The unit in Weslaco, Texas,provided basic technology including spectral characteris-tics of soils, water, and crop and range vegetation, ther-mal and aridity relationships, atmospheric and cloud ef-fects, spectral inputs to crop models, and spectralcomponents analysis.

The ARS unit in Phoenix, Arizona, conducted basicspectral and thermal studies that contributed to improvedunderstanding of vegetative indices and crop water stress.They established relations of spectral and/or canopy tem-peratures with yield components, plant stresses, and finalyield. These studies provided insight and knowledge intothe effects of surface geometry, canopy structure, solarzenith angle, and atmosphere on both spectral and thermalcharacteristics.

The ARS Hydrology Research unit in Beltsville, Mary-land, developed a model for monitoring floods and as-sessed other hydrologic parametcrs.

The Crops Research unit in Beltsville and ARS person-nel at Bushland and Lubbock, Texas, provided groundbased spectral information for major crops under stressedand nonst ressed cond itions.

The research unit in Mandan, North Dakota, definedparameters and established threshold values for harvestloss modeling, and measured spectral responses of grass-lands under different management practices.

Research conducted at Sidney, Montana, and at Akron,Colorado, provided basic research needed to establish pa-rameter and threshold values for the wheat winterkill andwheat yield reduction models.

The EW /CCA unit in Houston coordinated and man-aged the project. This unit, composed of scientists fromARS, SRS, NASA, NOAA, and Lockheed Engineeringand Management Services Company (LEMSCO), di-rected resources toward crop stress indicator model de-velopment, development and implementation of basic andapplied remote-sensing technology, and the transfer ofproven technology to users. The Houston unit was colo-cated with the Foreign Crop Condition Assessment Divi-sion of the Foreign Agricultural Service (FAS). The FASdivision identified needed technology, provided computerresources, and assisted in technology transfer from theEW/CCA unit in Houston to FAS in Washington.

REFERENCESII] A, C. Aaronson, "The large area operational application of the win-

terkill model using real time data and evaluation of the results,"USDA-FAS Crop Condition Assessment Div" Tech, Memo, 13, 1980,

[2] A. Bauer, and A, L. Black, "Rain-Induced spring wheat harvestlosses," AgRISTARS EW/CCA Rep, EW-U3-04405, Feb, 1983,

13] -. "The water factor in harvest-sprouting of hard red spring wheat."

AgRISTARS EW/CCA Rep, EW-U3-04406. Feb, 1983,14] G, 0, Boatwright, F, W, Ravet, and T W, Taylor, "Development of

early warning models," Ch, 12, ARS Wheat Yield Project, USDA-ARR, to be published,

15J J. A, Boatwright and W, M, Bradley, "Image correction and registra-tion utility system (ICARUS) design document." Unnumbered Tech,Rep" LEMSCO; 1984,

16] C, J, A, Gay and R, p, Gay, "Early warning soybean stress predictionmodel," Ph,D, dissertation, Biosystems Research Group Departmentof Industrial Engineering, Texas A&M Univ" College Station, TE,Oct. ]983,

\71 T, L Gray and D, G, McCrary, "Meteorologieal satellite data: A toolto describe the health of the world's agriculture," AgRISTARS EWICCA Rep,. EW-NI-04042, JSC-17112, Feb, 1981.

18] R, D, Jackson, E, B, Idso, R. J, Reginato, and P. J. Pinter, Jr., "Can-opy temperature as a crop water stress indicator," Water Res, Re-search, vol. 17, pp. 1\33-1138, 1981.

191 W. R. Johnson, "Atmospheric elrects on metsat data," AgRISTARSEW/CCA Rep" EW-L2-04387, JSC-18589, Jan. 1983,

1101 R, J, Kauth, and G, S, Thomas, "The tasselled cap-A graphic de-scription of the spectral temporal development of agricultural crops asseen by Landsat." in Proc. Symp, Machine Process;n!! RemotelySensed Data (West Lafayette, IN), vol. 4B, pp, 41-5\, 1976,

1II1 M, L. Mathews, "Cloud screening and solar correction investigationson the influence on NOAA-6 advanced very high resolution radiome-ter derived vegetation assessment," AgRISTARS EW ICCA Rep., EW-L3-04402, JSC-18606, Mar. 1983.

1121 p, W, Ravet and J. R, Hickman," A meteorologically driven wheatstress indicator model," USDA-FAS Crop Condition Assessment Div,Tech, Memo, 8, 1979,

113] F, W Ravet, W J, Cremins, T W Taylor, p, Ashburn, D, Smika, andA, Aaronson, "A mctcorologically driven yield reduction model forspring and winter wheat," AgRISTARS EW ICC A Rep. EW-U3-04397, JSC-1860l, Feb, 1983,

114J A, J, Richardson, "EI Chicon volcanic ash effects on atmosphcric hazemeasured by NOAA-7 AYHRR data," Remote Sensing t.'nviroll ' , vol.16, pp. 157~164, 1984.

1151 E, L. Sharpe, C. R. Perry, A, L Scharen, G, O. Boatwright. D. C.Sands, L. F, Lautenschlagcr, C. M, Yahyaoui, and F. W, Ravet,"Monitoring ccreal rust development with a spcctral radiometer,"Ph)'topath" vol, 75, 1985,

1161 T. W. Taylor and F, W, Ravet, "A meteorologically driven maize strcssindicator model," AgRISTARS EW/CCA Rcp, EW-UI-04lI9, JSC-17399, Apr. 1981.

1171 -----, "A metcorologically driven grain sorghum stress indicatormodel," AgRlSTARS EW/CCA Rep, EW-UI-042ll8, JSC-I7797, Nov,1981.

*Glenn 0, Boatwright received the B,S, and M ,S,degrees in soils from Oklahoma A&M in 1954 and1956, respectively, and the Ph,D. degree in plantand soil science from Montana State University in1970.

From 1956 to 1972, he was a Research Scientistwith the Agricultural Research Service in Man-dan, ND, Bozeman, MT, and Gunnison, CO, Hewas a rancher in Chickasha, OK, from 1972 to1976, He returned to the Agricultural ResearchScrviee in 1976 as a Research Scientist where he

is currently the Early Warning Crop Condition Assessment Project Man-ager.

*Victor S. Whitehead was born on September 18,in Taylor, AR, Hc received the B,A, degree inphysics and math from Baylor University, the M,S,degree in meteorology from Texas A&M Univer-sity, and thc Ph.D, degree in engineering sciencesfrom thc University of Oklahoma.

Since 1968, hc has been employed by the NASAJohnson Spacc Ccnter in a variety of Eal1h Obser-vations Program and Remote Sensing Program as-signments, including LACIE and AgRISTARS,

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSrNG, VOL. GE-24, NO. I, JANUARY 1986

Field Spectroscopy of Agricultural CropsMARVIN E. BAUER, MEMBER, IEEE, CRAIG S, T. DAUGHTRY, LARRY L. BIEHL, MEMBER, IEEE,

EDWARD T. KANEMASU, AND FORREST G. HALL

65

Abstract-The design, implementation, and results of multisite mul-tiyear experiments to measure and model the multispectral reflectanceof agricultural crops in relation to their biophysical characteristics aredescribed. The experimental approach involved multitemporal reflec-tance measurements together with detailed measurements of the ag-ronomic characteristics of crop canopies. One result of the field mea-surements and analyses was a quantitative description of the complexrelationships among crop canopy, soil, atmosphere, and illuminationand sensor geometries. Leaf area index was identified as a key bio-physical parameter linking crop physiology and multispectral remotesensing. Quantitative understanding and models of this relationship ledto the development of spectral-temporal profile models for crop speciesidentification and development stage estimation. A second key devel-opment has been the development of conceptual approaches and modelsfor spectral estimation of leaf area index and light interception of cropcanopies as inputs to crop growth and yield models. Other results in-clude quantification of the effects of soil background, cultural prac-tices, moisture stress, and nutrient deficiencies on crop reflectance, andthe effects of sun angle and sensor view angle on measured canopy re-flectance. The field measurements of canopy reflectance and geometryalso provided data bases to test and validate canopy radiation models.In summary, the AgRISTARS field research on agricultural crops hasprovided a critical link between satellite and leaf spectral data.

Key Words-Multispectral, remote sensing, reflectance, canopy ra-diation, leaf area index, crop identification, crop-condition assessment,spectral inputs to crop models.

I. INTRODUCTION

To DEVELOP the full potential of multispectral dataacquired from satellites, quantitative knowledge, and

physical models of the spectral properties of specific Earthsurface features are required. Knowledge of the relation-ships between spectral-radiometric characteristics andimportant biophysical parameters of agricultural crops andsoils can best be obtained by carefully controlled studiesof fields or plots where complete data describing the ag-ronomic-biophysical properties of the crop canopies andsoil background are attainable and where frequent timelycalibrated spectral measurements can be made fl]. Theseattributes distinguish field research from other remote-sensing research activities. Although the term field spec-troscopy has generally not been applied to this research,

Manuscript received June 16. 1985; revised August 29, 1985.M. E. Bauer was with the Laboratory for Applications of Rcmote Sens-

ing, Purdue University, West Lafayette, IN 47907. He is now with the Re-mote Sensing Laboratory, University of Minnesota, SI. Paul, MN 55108.

C. S. T. Daughtry and L. L. Biehl are with the Laboratory for Appli-cations of Remote Sensing, Purdue University, West Lafayette, IN 47907.

E. T. Kanemasu is with the Evapotranspiration Laboratory, Kansas StateUniversity, Manhattan, KS 66506.

F. G. Hall was with the NASA Johnson Space Center, Houston, TX77058. He is now with the NASA Goddard Space Flight Center, Greenbelt,MD 20771.


in retrospect it seems to clearly and concisely describe theapproach. Definitions of spectroscopy include study ofspectra, especially the experimental observation of spec-tra; production and investigation of spectra; and physicsthat deals with theory and interpretation of interactionsbetween matter and electromagnetic radiation.

Satellite sensors will ultimately provide the data formost agricultural remote-sensing applications [2], [3], butthe spatial and temporal resolution of current satellite datais not well suited for efficiently determining the cause-effect relationships between spectral response and othercrop variables. And, while laboratory measurements ofleaf and soil reflectance spectra are important elements ofa balanced research program, these data cannot be di-rectly extrapolated to crop canopies where there are manyinteracting variables such as environment and crop ge-ometry, In situ spectral measurements of crop canopiesprovide an essential bridge between the macro observa-tions of agricultural fields by aircraft and satellite sensorsand micro observations in the laboratory of leaves and soilsamples.

A second key role of field research is in the developmentand verification of canopy radiation models. These modelsprovide a theoretical basis for remote-sensing experi-ments, and can greatly enhance the remote-sensing re-search by extending the field mesurements to a wider setof environmental conditions and sensor viewing and illu-mination geometries than can be obtained by direct mea-surements [4]. The possibility of inverting such models toestimate agronomically important canopy parameters suchas LAI adds to their importance. Accurate measurementsof canopy reflectance, as well as the measurements of themodel inputs, are required for model verification.

This paper describes field research that was sponsoredby the NASA Johnson Space Center, Houston, TX, as partof the AgRISTARS Supporting Research Project. Other,related research has been conducted by the USDA Agri-cultural Research Service and the NASA Goddard SpaceFlight Center. The remainder of the paper, which sum-marizes the field research accomplishments since the LA-CIE field research project [1], is divided into four mainsections describing the field research objectives (SectionII), development of capability (Section III), experimentdesign and approach (Section IV), and experiment results(Section V).

II. FIELD RESEARCH OBJECTIVESThe overall objectives of the AgRISTARS Field Re-

search project were to: 1) conduct analyses and develop

0196-2892/86/0100-0065$01.00 © 1986 IEEE

66 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. GE-24, NO. I. JANUARY 1986

physical models of the spectral properties of crops and soilsin relation to agronomic and physical properties of thescene, 2) provide candidate models and analysis tech-niques to other supporting research experiments andAgRISTARS projects, and 3) assess the capability of cur-rent, planned, and possible future satellite sensor systemsto capture available information for identification and as-sessment of crops and soils,

Specific objectives included:

I) Model the relationship of agronomically importantcanopy parameters (e.g. leaf area index and devel-opment stage) to reflectance properties of crop can-OpIes.

2) Determine the effects of plant stresses, particularlymoisture and nutrient deficiencies, on the spectralreflectance of crop canopies,

3) Quantify the effects of cultural, soil, and environ-mental factors, on the spectral characteristics ofvegetation.

4) Quantify the effects of sensor and illumination ge-ometry, and its interaction wih canopy geometry, onthe spectral reflectance of crop canopies.

5) Acquire canopy reflectance measurements, togetherwith measurements of the model input variables, forevaluation and inversion of canopy radiation models.

III. DEVELOPMENT OF CAPABILITY

During the past decade a great deal of capability to ac-quire meaningful spectral measurements of agriculturalcrops and soils has been developed, and a substantial num-ber of results are appearing in the scientific literature. Thecapability and results described in this paper are in largepart due to sustained support of the NASA Johnson SpaceCenter over an extended period beginning in the 1970's.The relevancy and critical role of field research in the de-velopment of satellite applications of remote sensing haslong been recognized [5], but a significant amount of de-velopment and testing of instrumentation and measure-ment procedures for remote-sensing field research hasbeen required.

A. Field Research Instrumentation

Although the importance of quantitative knowledge ofcrop spectral characteristics was recognized early in thedevelopment of contemporary remote sensing, the lack ofappropriate field instrumentation severely restricted de-velopment of a comprehensive research program. Labo-ratory measurements of leaves and soil did not representthe complexity and spatial variability of canopies in thefield, nor include the effects of illumination and viewinggeometry. Aircraft multispectral scanners, while provid-ing excellent data for development of digital image anal-ysis techniques, have not been widely used in field re-search because of the difficulties of operation on an on-call basis, calibration problems, inflexibility of wavelengthband configuration, and large costs of operation and dataprocessing. As a result, researchers requiring in situ mea-

surements have turned to multiband and continuous wave-length instruments.

Initial efforts involved extension of laboratory spectro-radiometers to field applications. Further efforts led to de-velopment of rugged high-resolution field spectrometersystems such as the Exotech 20C operated by Purdue Uni-versity/LARS and the S-191H operated by the NASAJohnson Space Center. These instruments are capable ofaccurately measuring spectral reflectance and emittedspectral radiance, and have been used on truck-mountedtowers and helicopters. Using these instruments the LA-CIE Field Measurements project [I] produced spectraldata which were calibrated (and, therefore, comparablefrom time to time and place to place). A large number ofspectra for experiments with both controlled plot and com-mercial fields were acquired, processed, archived, andanalyzed over the three-year project, but it became clearthat these systems could not economically satisfy the needto acquire data at the number of sites needed to representthe variability in crop, soil, and weather conditions asso-ciated with production of major crops.

At the same time it was equally clear that while havingthe advantage of simplicity and low cost, that the thenavailable multiband radiometers were not adequate. Inparticular, these instruments were characterized by re-stricted wavelength coverage and did not include bands inthe middle and thermal infrared. In summary, availableinstruments were either inadequate or too costly to obtainthe necessary data. Additionally, there was a critical needfor standardized acquisition and calibration procedures toinsure the validity and comparability of data.

In response to the need to increase the number of re-searchers and locations conducting remote sensing fieldresearch, researchers at Purdue University began to de-sign a multiband radiometer system especially for agri-cultural field research. The LACIE Field Research projecthad demonstrated that a practical means to obtain spectraldata from a wide variety of subjects and to increase thenumber of investigators who could afford to acquire andanalyze data was to simplify the instrumentation and re-duce the amount of data obtained for each observation. Toachieve this, a field-rated multiband radiometer with alimited, but sufficient, number of wavelength bands sam-pling all important parts of the reflective spectrum from0.4 to 2.4 I-tm, plus a thermal infrared band (lOA to 12.5I-tm) was specified. Other design criteria included: com-paratively inexpensive to acquire, maintain, and operate;simple to operate, calibrate, and service; rugged, lightweight, and portable; complete with data recording andhandling hardware and software; and well-documented foruse by researchers [6].

The NASA Johnson Space Center sponsored the devel-opment of the system and in 1981 purchased 15 of theradiometer units for use at NASA-sponsored researchsites. These instruments, together wih solid-state datalogger, camera, truck or helicopter platform, and reflec-tance calibration standard, became the primary spectraldata acquisition systems of the Field Research segment of

BAUER ,,1 al. FIELD SPECTROSCOPY OF AGRICULTURAL CROPS

(a)

(b)

Fig. I. (a) Truck-mounted multiband radiometer and (b) hclicopter-mounted spectrometer systems. The truck-mounted system includes ra-diometer, 35-mm camera, data logger, and calibration standard. Mea-surements of 40 to 60 plots/h can be made with this system. The heli-copter system, shown hovering over a canvas calibration standard, fliestransects, typically 6 miles long, over commercial fields at an altitude of75 m. With average field sizes, 10 to 30 spectra arc acquired of eachfield.

TABLE IAgRISTARS FIELD RESEARCH COMMERCIAL FIELD TEST SITES

Locat i on Major Crops Years Sensor s~':

Webster County. Corn 1979-81 FSSIowa Soybean NS-OO 1

Cass County, Spring wheat 1980-82 FSS ( '80, '81 )North Dakota Barley NS-OO 1

5unf lower MRS ( '82)

Wharton County, Co t ton 1980 NS-OO 1Texas Ri ce

Soybean

;':FS5. hel i copter-mounted field spectrometer system; NS-001.thematic mapper simulator; MRS. hel icopter-mounted multibandradiometer system.

the AgRISTARS Supporting Research Project (Fig. I).The locations and crop species for which spectral mea-surements have been made using these systems are listedin Table I.

B. Measurement Procedures1) Calibration of Spectral Data: One of the key as-

pects of the approach has been to follow procedures, such

67

that measurements acquired at different locations andtimes can be compared and/or combined. Although datafrom an individual experiment are certainly valuable, it isexpected that they will have greater value when data fromseveral sites are combined or when a model is testedagainst an independent data set. It therefore follows thatthe spectral and agronomic measurement procedures needto be carefully designed and executed.

The spectral data obtained by the spectroradiometer andmultiband radiometer systems are processed into compa-rable units, reflectance factor. A reflectance factor is de-fined as the ratio of the radiant flux actually reflected bya sample surface to that which would be reflected into thesame sensor geometry by an ideal, perfectly diffuse sur-face irradiated in exactly the same way as the sample [7].

The field calibration procedure consists of the compar-ison of the response of the instrument viewing the crop orsoil to the response of the instrument viewing a level ref-erence surface. The reference surfaces are 1.2-m2 paintedbarium sulfate panels for truck-mounted systems and a 6x 12 m white canvas panel for the helicopter-mountedsensors. For small fields of view (less than 20° full angle)the term bidirectional reflectance factor has been used todescribe the measurement: one direction being associatedwith the viewing angle (usually 0° from normal) and theother direction being the solar zenith and azimuth angles.

The spectral data are obtained following well-definedfield procedures [7]. Key components of the procedure are:

frequent observations of the reflectance reference panel(at least every 10 to 20 min),

instrument aperature is sufficiently distant from thescene (at least 3 m above the top of the canopy),

collect data when solar elevation angles are above 45 ° ,except for modeling experiments,

no clouds are in the vicinity of the sun, andthe reflectance reference surface is viewed in the same

manner as the scene.In addition, the non-Lambertian properties of the ref-

erence panels must be taken into account. The reflectanceof the barium sulfate panels are compared to that ofpressed barium sulfate in the laboratory at Purdue/LARSusing a bidirectional reflectometer for illumination zenithangles of 10° to 85 0. An extensive set of calibration mea-surements of reference surfaces are kept.

2) Sensor Altitude: Use of portable ground-based sen-sors for measuring crop reflectance has dictated a need forreliable measurement procedures capable of providing cal-ibrated and reproducible canopy reflectance data. An im-portant aspect of acquiring reproducible data for canopiesconcerns sensor altitude in relation to canopy type androw spacing. Daughtry et al. [8] measured the variationin reflectance of corn and soybean canopies as functionsof horizontal distance across rows and vertical distanceabove the soil. Variation as the sensor was moved acrossthe canopy disappeared as sensor altitude increased andintegrated across several rows. Coefficients of variationdecreased exponentially as sensor altitude increased.

6X IEEE TRANSACTIOt\S ON GEOSCIENCE AND REMOTE SE"lSINC" VOL. (;E-24, NO. I, JANUARY 1986

Sampling schemes employing prior knowledge of rowspacing were determined to be more efficient (requiredfewer measurements f<H a given level of precision) thanrandom sampling schemes. At altitudes where several rowsare included in the field of view, as few as two measure-ments were required to detect 10-percent differences (aspercent of the mean) in reflectance, whereas at lower al-titudes 20 to more than 100 measurements would beneeded.

3) ARronomic Measurements: Equally important to ac-quiring calibrated spectral measurements is the acquisi-tion of accurate agronomic measurements of the canopies.Particular emphasis has been given to measurements ofleaf area index because of its central importance to spec-tra] reflectance, photosynthesis and evapotranspiration ofcrop canopies. There are a number of satisfactory meth-ods to measure the leaf area of individual leaves and plants;however, to estimate the LAI of canopies, the variabilityin leaf area among plants within plots is an additionalsource of experimental error. In support of remote-sensingresearch, Daughtry and Hollinger 191 examined the magu

nitude of within plot variability and evaluated severalmethods for measuring LAI with known precision andprobability of detecting differences. The approximate er-rors, number of plants required, and relative costs of eachmethod were determined.

The results provided direction f()r the sampling andmeasurement procedures, plus strong evidence that spec-tral-agronomic relationships are best developed from con-trolled plots where plant to plant variability is relativelysmall (IO-percent CV) compared to that in fields, and thatif a parameter such as LAI is to ever be utilized in cropgrowth and yield models, it will have to be estimated fromremotely sensed spectral data. For example, direct areameasurements of all leaves on plants using an electronicarea meter (the method resulting in the greatest precisionwith time requirements comparable to other methods) re-quired measurements of 2\, 7, and 2 plants to detect truedifferences in LAI among treatments of 10, 20, and 50percent at 0.05 ]evel of significance and 90-percent prob-ability of success. These measurements required 168, 56,and 16 melD-minutes, respectively.

IV. EXPERIMENT DESIGN AND ApPROACH

Following the approach initiated in the LACIE FieldMeasurements project, a multistage approach to data ac-quisition was taken, including areal, vertical, and tem-poral staging. Areal sampling was accomplished with testsites at multiple locations in the U.S. Great Plains andCorn Belt, plus sites at Corvallis, Oregon, and Obregon,Mexico. The sites were selected to sample a wide range ofconditions under which corn, soybeans, and wheat areproduced. Vertical staging, or collection of data by differ-ent sensor systems and at different altitudes, ranged frommobile (truck) towers to Landsat. Temporally, data werecollected, depending on sensor system and location at 5-20 day intervals to sample all important developmentstages and during 5 years (1980-1984) to obtain a measure

of year-to-year variation in growing conditions and its in-fluence on spect ral response.

As established during LACI E [I], two major types ofexperiments were utilized: I) controlled experiments in-volving research plots at agricultural experiment stationsand 2) noncontrolled experiments in commercia] fields ]0-

cated in AgRISTARS segments. The controlled experi-ments enable detai led agronomic and frequent spectralmeasurements to be made of plots with known sourees ofvariation (agronomic treatments). The measurements incommercia] fields, although less detailed and frequent,provide a measure of natura] variation in the spectra]-spa-tial-temporal characteristics of the crops. Past experiencehas shown that there are generally too many interactingvariables in commercia] fields to determine exact causesof observed differences in spectral response. With datafrom plots where only two to four factors are varied undercontrolled conditions, it is possible to determine more ex-actly and understand more fully the relationships of agro-physical and spectral characteristics of the crop canopies.

A. Description of Etperiments and Measurements1) Controlled Plot Experimenls: With the availability

of the multiband radiometer systems described above, nu-merous experiments involving controlled plots have beenconducted during AgRISTARS at the locations listed inTable II. Although space does not permit describing, oreven listing all of the experiments, it may be helpful tosummarize as examples the objectives, experiment de-signs, and measurements for two experiments. The first isan example of what has been referred to as a cultural prac-tices experiment, while the second is an experiment in-volving moisture stress.

0) Corn cultural practices: This experiment wasconducted during the 1980-1983 growing seasons at thePurdue University Agronomy Farm, West Lafayette, IN.The objectives were to I) identify the threshold of earlyseason spectral detection of corn, 2) determine the rela-tionship of development stage and amount of vegetation(e.g., LA!) to spectral response, and 3) determine the ef-fect of soil background and cultural practices on spectralresponse. The treatments, with 1981 planting dates, were:four planting dates (May 8, 29, and June II, 29); threeplant populations (25 000, 50 000, and 75 000 plants/ha);and two soil types (Chalmers, dark color: Fincastle. lightcolor). A split piot factorial experiment design with tworeplications of each treatment was used. Spectral mea-surements, along with agronomic characterizations of thecanopies and surface soi], were made at approximatelyweekly intervals throughout the growing season. Thespectral measurements were made with an Exotech 100radiometer in all years: in 1982 and 1983 a Barnes mul-tiband radiometer was added. Radiant temperatures andoverhead color photographs of the canopies were obtainedsimultaneously with the reflectance measurements. Theprimary agronomic measurements included developmentstage, percent soil cover, LA/, fresh and dry biomass, andsurface soil moisture. Beginning in 1982, canopy trans-

BAUER ('( at.: FIELD SPECTROSCOPY OF AGRICULTURAL CROPS

TABLE IISUMMARY OF CONTROI.I.ED PLOT EXPF.RIME~TS Sf AGRIClIUURAL

EXPERIMENT STATIONS USING MI'UIB;\ND RADIOMFTERS

69

Location

Purdue Univ.'.I. Lafayette. IN

Univ. NebraskaL ineo 1n. NE

Univ. MinnesotaSt. Paul. HN

Kansas State Univ.Manhattan. KS

s. Dakota State Univ.Brook ings, SD

Oregon State Univ.Corvall is, OR

Texas A&M Univ.College Station, TX

C I ~\HYT

Obregon, Mexico

Univ. KansasLawrence, KS

NASA/GSFCGreenbelt. MD

Crop(s)

CornSoybeanWinter wheat

CornSoybean

CornSoybean

Winter wheatCorn

Spring wheatBar leyOats

Winter wheatBarley

Ri ceSorghum

Spr ing wheatWinter wheatSorghum

Corn

CornSoybeanWinter wheat

Years

1979-84

1981-84

1982-84

1981-84

1982-84

1982-83

1982-84

1982-84

1982-84

1982-83

Primary Objectives/Topics Addressed

Spectral estimation of LAI, light interception, and development stageEffects of cultural. soil. and environmental factorsCanopy model lng/sensor. illumination and canopy geometry effects

Moisture stress effects on crop reflectance and radiant temperatureSpectral estimation of LAl and grain yield

Effects of ti Ilage methods and residue on crop reflectance

Spectral estimation of LAI, Iight interception, and development stageEffects of cultural, soil, and environmental factors

Spectral separabi Iity of small grainsEstimation of LAI

Spectral reflectance characteristics of small grains

Effects of cultural practices on crop reflectance

Effects of moisture stress and cultural practices on crop growth,yield and reflectance

Synergistic effects of optical plus microwave measurements

Canopy modeling/sensor. illumination and canopy geometry effects

mittance was measured at the soil surface to calculate so-lar radiation interception. Grain yields were measured atharvest time. Similar experiments have been conductedfor soy beans.

b) Corn moisture stress: Moisture stress experi-ments were conducted on corn and soybeans by the Uni-versity of Nebraska at the Sandhills Agricultural Labora-tory near Tryon, NE, in 1981-1984. Irrigation facilities atthis site permit the application of water during prescribedgrowth stages on either whole plots or on gradient plots.For example, in one experiment gradient irrigation treat-ments were applied across 24 rows of a plot (row 1 re-ceives full irrigation, row 24 receives no water). One setof plots received gradient irrigation during vegetative,pollination, and grain filling stages. A second set of plotsreceived gradient irrigation during vegetative stages andfull irrigation thereafter, while a third set of plots receivedfull irrigation at all stages. This experimental setup pro-vided a wide range of stress conditions during the season.The sandy soils and the relatively low amounts of rainfallreceived at this site make it an ideal location to studymoisture stress effects on the growth and spectral char-acteristics of agronomic crops without the necessity of in-stalling expensive rainout shelters. Spectral measure-ments were made with Barnes and Exotech 100Amultiband radiometers, supplemented by additional ra-diant temperature measurements at oblique view angles.Agronomic measurements included development stage,LAI, biomass, canopy structure, le~f water potential, leafphotosynthesis, stomatal resistance, and grain yield. Me-teorological data routinely collected at this site include soil

moisture, soil temperature, solar radiation, windspeed anddirection, air temperature, relative humidity, and rainfall.

2) Canopy Modeling Experiments: Canopy radiationmodels have an important role in remote sensing research.However, the models require validation before widespreaduse in simulation and estimation. Providing the data formodel verification has been one of the objectives of theAgRISTARS field research project.

An efficient method to validate the models is to com-pare model estimates and canopy measurements of canopyreflectance as a function of sun angle and view angle fora variety of canopy types at different development stages.And, with the development of satellite sensor systems withoff-nadir viewing, there is current major interest in thedirectional reflectance properties of vegetation. With thisbackground, researchers at Purdue University have devel-oped an approach to efficiently acquire canopy reflectancedata as a function of view angle and sun angle.

The approach consists of making measurements from atower placed in the center of a uniform field. Both a truck-mounted tower and a stationary tower built of constructionscaffolding (Fig. 2) have been used. The latter method uti-lizes a 3-m boom mounted on a center pivot. The boomcan be rotated on both its horizontal and vertical axes.Rotating the boom about the pivot provides selection ofazimuth positions of 0°, 450

, 90°, 135°. 1800, 225°,

270°, and 315°. At each azimuth, measurements are madeat view zenith angles of 70°, 60°, 45 0, 30°, 22 0, 150

•

7°, and 0° followed by measurements in the opposite di-rection of r, 15° , ., 70 0. Two full hemispheres ofmeasurements can be obtained in 12 min. Calibration

70 IEEE TRANSACTIONS ON GEOSCIENCE AND RE'.10TE SENSING. VOL. <iE-24. NO. L JANUARY In6

(a)

(b)

Fig. 2. (a) Tower and (b) radiometer syslem for making rel1cctance mea-surements as a function of view and sun angle. The radiometer and cam-era mounted on the boom are pointed at zenith view angle of 0°. Thcboom is on a pivot to enable measurcments at different view azimuthsand zcnith angles.

measurements are made before and after each circuit. Tomeasure the effects of changing sun angle measurementsare made at 30- to 60-min intervals throughout the day.

To date measurements have been made on three cano-pies, soybean, corn, and winter wheat, at major develop-ment stages during 1980, 1982, and 1983, respectively.The canopy reflectance measurements have been accom-panied by detailed measurements describing the biophys-ical characteristics of the canopies. These data include leafarea index, total biomass, development stage, percent can-opy cover, canopy profile shape, leaf angle distribution,and leaf spectral reflectance and transmittance. Copies ofdata sets for selected dates have been provided to severalinvestigators for the purpose of evaluating canopy models[10]. The data have been used by Goel et al. [III andBadhwar and Shen [12] to test the inversion of canopymodels to predict leaf area index, while Ranson et al. [131have used it to investigate the angular reflectance prop-erties of corn and soybean canopies.

3) Commercial Field Test Sites: Measurements weremade during AgRISTARS at three commercial field sites(5 x 6 mile segments) as summarized in Table I usingspectrometer or multiband radiometer and multispectralscanner sensors. Each sensor system has unique capabil-ities for acquiring spectral data. The spectrometer sys-tems produce the highest quality reflectance measure-ments, but provides only limited measurements of spatialvariability. On the other hand, a multispectral scanner inan aircraft provides spatial sampling of the scene and can

obtain data at multiple altitudes, but its spectral coverageis limited to a fixed set of spectral bands. Both systemshave the advantage of flexible scheduling and, therefore,provide greater opportunity to obtain cloud-free data atcritical crop stages than Landsat provides. However, theprocessing of data from both systems is relatively expen-sive, a disadvantage which the multiband radiometer sys-tem does not have.

The data collection missions were scheduled at two- tothree-week intervals, when possible, coinciding withLandsat overpass dates. Spectral measurements were sup-plemented by aerial photography and field observationsand measurements, including development stage, plantheight, presence of stress, plant counts, row width, andgrain yield.

B. Data LibraryDevelopment of a crops and soils data base for scene

radiation research was initiated in 1972 at Purdue Univer-sity; it has continued to grow and develop as a part of theLACIE and AgRISTARS field research projects 114]. Itspurpose is to provide fully annotated and cal ibrated setsof spectral and agronomic data agricultural remote sens-ing research.

The data base presently includes more than 300 datesand 180 000 observations of spectroradiometer data, 250dates and 70 000 observations of multiband radiometerdata, and 70 dates and 400 flight lines of multispectralscanner data. These data are supplemented by an extensiveset of agronomic and meteorological data acquired duringeach mission. In addition, the library includes laboratorymeasurements of over 250 soils from 39 states. With theexception of photography, the data are available on com-puter tapes.

The data form one of the most complete and best doc-umented data sets acquired for remote-sensing research.It is unique in the comprehensiveness of sensors and mis-sions throughout several growing seasons, and in the cal-ibration of all multispectral data to a common standard.Copies of selected parts of the data have been provided tomore than 50 different investigators at universities andgovernment agencies over the past 5 years for research onspectral-agronomic relationships, definition of future sen-sor parameters, and development of advanced analysistechniques.

Y. EXPERIMENT RESLILTS

There has been a growing number of published field re-search results since the beginning of AgRISTARS Sup-porting Research Project, and although the Project hasformally ended, we expect results based on the extensiveand comprehensive data sets acquired during the projectwill continue to be generated. Since it would not be pos-sible to describe in one paper all of the research resultsbased on the data, we have chosen to summarize the re-sults of several key experiments. Complete descriptions ofthese and other results may be found in the cited refer-ences and in agronomic and remote sensing journals.

BAUER e/ al.: FIELD SPECTROSCOPY OF AGRICULTURAL CROPS 71

100

Fig. 4. Relation between absorbed photosynthetically active radiation(APAR) and greenness index for growth (planting to silking) and senes-cing (silking to maturity) periods of corn development.

for research plots. However, if these parameters could beestimated from remotely sensed spectral data it would en-able crop growth and yield models to be implemented overlarge geographic areas [19]. The objective of experimentson corn [20] and soybeans at Purdue University and win-ter wheat at Kansas State University [21] has been tomodel the relationship between photosynthetically activeradiation (PAR) absorbed by crop canopies and the spec-tral reflectance of the canopies.

Absorption of PAR was measured near solar noon incorn canopies planted in a field experiment conducted atthe Purdue University Agronomy Farm, at densities of50 000 and 100 000 plants/ha. Reflectance factor datawere acquired with a Landsat MSS band radiometer. Fromplanting to silking, the three spectrally predicted vegeta-tion indices (IR/red ratio, normalized difference, andgreenness) examined were associated with more than 95percent of the variability in absorbed PAR (Fig. 4). Therelationships developed between absorbed PAR and thethree indices were evaluated with reflectance factor dataacquired from corn canopies planted in 1979 through 1982that excluded those canopies from which the equationswere developed. Treatments included in these data weretwo hybrids, four planting densities (25 000, 50 000,75 000, and 100 000 plants/ha), three soil types (TypicArgiaquol, Udollic Ochraqualf, and Aeric Ochraqualf),and several planting dates.

Seasonal cumulations of measured LAI and each of thethree spectral indices were associated with more than 50percent of the variation in final grain yields from the testyears. Seasonal cumulations of daily absorbed PAR wereassociated with up to 73 percent of the variation in finalgrain yields. Absorbed PAR, cumulated through thegrowing season, was a better indicator of yield than cu-mulated LA!.

These results, as well as those from the wheat experi-ments by Kansas State University [21], [22], suggest thatAPAR may be estimated from canopy spectral reflectancefor large areas where direct measurements of LAI wouldbe prohibitive. Thus, estimates of absorbed PAR may beused directly in simple plant productivity models to esti-mate above ground phytomass production [23] or in more

30

TO SILKINGTO MATURITY

20

Greenness Index10

80

ou

t~ 60a.

"~ 40

(;<II.c<t 20

Cardinal Development StagePoint Corn Soybean Spring Wheat

t2 full dent beginning maturity dough

TABLE IIIRELATIONSHIP OF CORN, SOYBEA:-.l, AND SPRING WHEAT DEVELOPMENT

STAGES TO CARDINAL POINTS OE TEMPORAL PROFILE

where Go is the soil greenness, ex and (3are crop-specificconstants, to is the date of spectral emergence, and Gm isthe maximum greenness at time tp' The model has twoinflection points, t[ and t2, which are related to rates ofchange in greenness. The features Gm, tp, and a accountfor more than 95 percent of the information in the originaldata, while substantially reducing the dimensionality of thespectral-temporal data. And equally important, the modelparameters are related to agrophysical parameters. Appli-cation of the model to Landsat MSS [15], [16] and TM[17] data has resulted in accurate crop identification andarea estimation. In addition to crop identification the samemodel form has been used to estimate crop developmentstages (Table III) [18].

tp bl ister beginning seed heading

B. Spectral Estimation of Radiation Absorbed by CropCanopies

Most models of crop growth and yield require an esti-mate of canopy leaf area index or absorption of solar ra-diation. Direct measurements of LAlor light absorptioncan be tedious and time consuming, and are possible only

<II<IICIl

""CIlCIl(;

A, Spectral-Temporal Profile ModelingA major breakthrough in capability for crop identifica-

tion has been development of multitemporal profile models(Fig. 3). Much of the research from initial development torefinement and verification of the models has been withfield research data. The present form of the model devel-oped by Badhwar [15] is:

G(t) = Go + (Gm - Go) (2(3e1ex)"'/2

. (t - to)'" exp [- (3(t - to)2]

Ip

tl 12-14 leaves beginning bloom jointing

from Badhwar [15, 16J,

Time

Fig. 3. Temporal profile model of greenness. Key parameters include:spectral emergence date, 10; time of peak greenness, I{'; maximum green-ness, G",; and width of the profile, a.

72IEEE TRAt»iSACTIONS ON GEOSCIENCE AND REMOTE SENSING, YOLo (iE-24. t»iO I. JANUARY IYX6

elaborate crop growth and development simulation modelsto estimate final crop yields [24],

Relative L.af Araa Duration

Fig. 5. Influence of canopy temperature on the relationship hctwcen rela-tive grain yield and relative Icaf area duration for Pioneer 3901 andB73xM017 corn hyhrids at Sandhills Agricultural Laboratory in 1982.Temperature rcgions arc accumulated dift'crences (incrcases) in canopyradiant tcmpcratures bctwccn region x and rcgion 1 (full irrigation/nostress) on seven datcs during the growing scason.

D. Effects of Cultural and Environmental Factors onCrop Reflectance

Understanding the relationship between spectral reflec-tance and cultural and environmental factors is one of thekeys to development and use of remote sensing as a toolfor crop monitoring. Multiyear field experiments wereconducted by Purdue University to study the effects of cul-tural practices and soil type on the reflectance character-istics of corn, soybean, and wheat canopies [26 J. Treat-ments included: cultivar, row width, and planting date forsoybeans; hybrid, plant population, and planting date forcorn; and cultivar, N fertilization, and planting date forwheat. Soil type or soil moisture was an additional factorincluded in the experiments for both crops. Agronomic

measurements included development stage, LAI, percentsoil cover. and biomass. Reflectance factor measurementsof the canopies were made with multiband radiometerswith wavebands corresponding to the Landsat MSS andTM bands at weekly intervals throughout the growing sea-son.

The results of these experiments [26] indicate that thevarious cultural practices produced differences in LAI andpercent soil cover, which in turn are manifested in thespectral reflectance characteristics of the canopies. Soilcolor and moisture influenced visible and infrared reflec-tance early in the growing season. The near infrared/redreflectance ratio and the greenness transformation wereuseful for predicting LAI and were less sensitive to vari-ations in soil color and moisture than reflectances in singlebands. Variations in spectral response were strongly as-sociated with planting date during early to mid-season,with row width and plant population during mid-season tonear maturity, and with cultivar and hybrid at maturity(Table IV). Crist [27] has examined the effects of the cul-tural practices on temporal profiles of greenness and re-flectance.

E. Sun Angle- View Angle Effects

The bidirectional reflectance characteristics of vegeta-tion canopies vary with changing sun angle through theday and over the growing season. The measured reflec-tance is also a function of the view angle and direction.In an investigation at Purdue University, using the mea-surements approach described above, the effects of sunand view angles on bidirectional reflectance factors of corncanopies ranging in development from the six leaf stage toharvest maturity were examined [28], For nadir-view an-gles, there was a strong effect of solar zenith angle onreflectance factor in all spectral bands for canopies withlow LA!. A decrease in contrast between bare soil andvegetation due to shadows as solar zenith angle increasedappeared to be the major contributor to this change in re-flectance factor. Effects of sun angle on reflectance weresmall for well-developed canopies with high LA!. A mod-erate increase in reflectance factor was observed at thelarger solar zenith angles and was attributed to the pres-ence of specular reflectance.

A strong dependence of reflectance factor on view anglewas noted for all of the canopies considered (Fig. 6). Forcanopies with low LAI, reflectance decreased as view ze-nith angle increased for visible and middle-infrared wave-length bands which are absorbed by vegetation and in-creased with view angle for near-infrared wavelengthswhich are multiply scattered. For higher LAI canopies,reflectance increased as view zenith angle increased. Anincrease in reflectance at large view zenith angles for someview azimuths indicated a specular component in the re-flectance data.

Trends of reflectance with changing sun angle at differ-ent view azimuth angles illustrate that the position of thesensor relative to the sun is an important factor for deter-

'LI0.9 1.0

1.0 I~--~---~-

a ~ L---"-~ ~ _~0.5 0.6 0.7 0.8

"0 0'8f''i>: 0.6c';

"• 0.4 [~••'ia: 0.2

C. Evaluation of Crop Moisture Stress Effects

The overall objective of field research conducted by theUniversity of Nebraska at the Sandhi lis Agricultural Lab-oratory has been to measure and model the effects of mois-ture stress on the growth, yield, and spectral character-istics of corn and soybeans. In one experiment twohybrids, Pioneer 3901 and B73xM073, were planted at76 000 plants/ha in 76-cm-wide rows. A gradient irriga-tion system was used to provide 100, 66, 33, and 0 percentof the maximum water requirements of the crop. A Barnes12-1000 multi band radiometer was used to make reflec-tance measurements of the canopies; emitted thermal ra-diation was measured with this instrument and with aninfrared thermometer.

Relative leaf area duration was estimated with an equa-tion incorporating reflectances in TM bands 3, 4, and 5,and explained approximately 50 percent of the variationin grain yields. Periodic canopy temperature measure-ments accumulated over time accounted for additionalvariation of grain yields (Fig. 5), with stress causing anelevation in canopy temperatures and reduction in yield[25] .

BAUER ct al.: FIELD SPECTROSCOPY OF AGRICULTURAL CROPS

TABLE IV

PERCENT OF VARIATION IN REIJ (0.6-0.7 i<m) AND NEAR l~rRARED (0.8-1.1i<m) REFLECTANCE AND THE GREENNESS TRANSFORMATION OF CORN.

SOYBEAN. AND SPRI~G WHEAT CANOPIES ASSOCIATED WITH

SOIL TYPE AND ClJUURAL PRACTICES ON SEVERAL DATES

(DEVELOPMENT STAGES)

Spectral Variable

Agronomic Red Reflectance Infrared Reflectance Greenness TransformationF ae tor

Corn

6/11 7/15 8/22 9/26 6/11 7/15 8/22 9/26 6/11 7/15 8/22 9/26

Soil type 56 25 3 51 21 5 16 2Plant population 8 22 1 33 36 4 1 31 47 2Planting date 12 38 7 53 24 8 14 76 39 61 9 82

Soybean

6/18 7/17 8/22 9/26 6/18 7/17 8/22 9/26 6/18 7/17 8/22 9/26

50i 1 type 13 15 15 11 2Planting date 13 63 52 87 44 69 10 85 82 90 12 87Row width 8 9 2 5 29 1 4 29Cultivar 1 6 2 16 10 16

Spring Wheat

6/1 6/23 717 7/20 6/1 6/23 717 7/20 6/1 6/23 717 7/20

So i 1 Moisture 16 73 52 9 28 69 36 8 41 87 63Cultivar 1 2 10 4 4 9 2 3Nfertilizer 9 1 1 3 4 6 3 3Planting date 36 5 27 85 35 4 6 42 12 21

,':Hode 15 include variation due to treatments. Total variation is due to blocks, treatments. and exper i menta'error. Interaction terms as well as percentages less than 1. 0, are omitted for clarity, but were included

in model.-- .. --".- -- --,.._---,

73

VIEW ZENITH ANGLE, Degrees

Fig. 6. Effect of view zenith angle on reflectancc factors of corn canopies:(a) and (b) sparse. (c) fully-developed. and (d) senescent. Data were ac-quired near solar noon. Negative and positive view zenith angles indicatethat thc radiometer was looking east (1), = 90°) and west (1),. = 270°).respectively. TM spectral band numbers are indicated in the legend.

60 (a) Jllne 13, LAI~O.4

40

20

Ig6

;f. 4a: 2of-U I«LL 60 (c) August 12, LAI:44

~ 40z;:! 20u

~ IgW 60::

4

I-60 -30 0 30 60

(bLllne 24, LAI= 1.2

-60 -30 0 30 60

workers have developed the methodology [11] for invertingcanopy reflectance models. In their original form, withinputs of leaf reflectance and transmittances, soil reflec-tance, illumination and view angles, and canopy LAI andleaf area distribution, the models predict canopy reflec-tance. Goel has shown that the models developed by Suits[29], Verhoef and Bunnik [30], and Norman and Wells[311 can be inverted, i.e., measurements of canopy reflec-tance can be used to predict agrophysical canopy varia-bles. Spectral prediction of LAI would be a particularlyvaluable capability. Goel showed that canopy reflectancemeasurements for a set of several view and sun angles,together with measured (or assumed) leaf and soil reflec-tances and transmittances are required to achieve accurateLAI estimates. Using a somewhat different approach,Badhwar and Shen [12] have inverted the SAIL model [30]to predict LAI using only nadir view angles.

VI. CONCLUSIONS

mining the angular reflectance characteristics of corn can-opies. Reflectances were greatest for coincident sun andview angles and minimized when the sensor view direc-tion was towards the sun. View direction relative to roworientation also contributes to the variation in reflectance.

F. Inversion of Canopy Reflectance ModelsUsing data from the sun angle-view angle/canopy mod-

eling experiments at Purdue University [13J, Goel and co-

The AgRISTARS field research on agricultural cropsdemonstrated the key role of field spectroscopy in bridg-ing the gap between satellite and leaf spectral data. Oneresult of the field measurements and analyses was a morecomplete and quantitative description of the complex re-lationships among plant, soil, and atmospheric variables,and the effects of varying illumination and sensor geom-etries. Leaf area index was identified as a key biophysicalparameter linking crop physiology and multispectral re-mote sensing. Multitemporal field measurements of can-

74 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. GE-24, NO. I. JANUARY i986

opy reflectance were an integral part of the developmentof spectral-temporal modeling for crop species identifi-cation and development stage estimation, Subsequent re-search led to development of techniques and models to usespectral estimates of leaf area index and light interceptionby crop canopies as inputs to crop growth and yieldmodels, In addition, the field measurements of canopy re-flectance as a function of varying canopy, illumination,and viewing geometry has provided important data for theverification and further development of canopy radiationmodels,

In conclusion, field spectroscopy is an essential com-ponent of the development of remote sensing for monitor-ing agricultural and natural resources. A sound field re-search program can provide the basis on which larger scalesatellite experiments and operational systems are con-structed. Although the research summarized in this paperhas contributed substantially to our knowledge and under-standing of how biophysical properties are manifested inthe spectral characteristics of crop canopies, and howthese relationships can be utilized for crop identificationand condition assessment, there are still many unan-swered questions and problems. The complex and diversenature of agricultural and natural resources suggests theneed for continuing effort to measure and model the spec-tral-biophysical of vegetation canopies.

ACKNOWLEDGMENT

We are especially grateful to R. B. MacDonald, NASAJohnson Space Center, for the stimulation, challenge, andsupport he gave to this research, and to B. Robinson, Pur-due University, for his unyielding devotion to the devel-opment and use of sound measurement procedures and in-strumentation. A few of the many other individuals whohave contributed to the experiments and results describedin this paper include K. Gallo and 1. Ranson, Purdue Uni-versity; G. Asrar, Kansas State University; G. Badhwar,K. Henderson, and D. Pitts, NASA Johnson Space Cen-ter; and B. Blad and B. Gardner, University of Nebraska.

REFERENCES

[I] M. E. Bauer. M. C. McEwen. W. A. Malila, and J. C. Harlan, "De-sign. implementation. and results of LACIE field research." in Proc.LarKe Area Crop Inventory Equipment, vol. III (NASA Johnson SpaceCenter, Houston, TX), pp. 1037-1066. 1978.

[2] R. D. Jackson, "Remote sensing of vegetation characteristics for farmmanagement," Proc. Soc. PholO-optical Instr. EnKr., vol. 475, pp,81-96, 1984.

13] F. G. Hall, "Remotc sensing vegctation at regional scales," Proc.Soc. Photo-optical Instr. Engr., vol. 475, pp. 75-80, 1984.

[4] J. A. Smith, "Rolc of scene radiation modcls in remote sensing," inProc. 8th In/. Svmp. Machine Processing of Remotely Sensed Data(Purdue Univ., W. Layfayettc, IN), pp. 546-549, 1982.

[5] R. B. MacDonald, "A summary of thc history of the development ofautomated rcmote scnsing for agricultural applications," IEEE Trans.Geosci. Remote Sensing, vol. GE-22. pp. 473-481. 1984.

[6] B. F. Robinson, M. E. Bauer, D. P. DcWitt, L. F. Silva, and V. C.Vanderbilt, "Multiband radiometer for field researcb," Proc. Soc.Photo-opticallnstr. Engr .. vol. 196. pp. 8-15, 1979.

171 B. F. Robinson and L. L. Bicbl, "Calibration procedures for mea-surement of reflectance factor in remote sensing field rcsearcb," Proc.SOl'. Photo-optical/nstr. Engr., vol. 196, pp. 16-26, 1979.

[8J C. S. T. Daugbtry, V. C. Vanderbilt, and V. J. Pollara, "Variabilityof reflectance measurements with scnsor altitude and canopy type,"Agron. 1., vol. 74, pp. 744-751, 1982.

19J C. S. T. Daughtry and S. E. Hollinger, "Costs of measuring leaf areaindcx of corn," Agron. 1., vol. 76, pp. 836-841, 1984.

110] G. D. Badbwar. W. Verboef, and N. J. J. Bunnik, "Comparative studyof Suits and SAIL canopy reflectancc models," Remote Sensing En-,·iron., vol. 17, pp. 179-195. 1985.

[II] N. S. Goel and R, L. Thompson, "Inversion of vegetation canopyreflectancc models for estimating agronomic variables. V. Estimationof leaf area index and average leaf angle using mcasured canopy re-flcctances." Remote Sensing Environ., vol. 16, pp. 69-85, 1984.

[121 G. D. Badhwar and S. S, Shen. "Techniqucs f(lr estimation of leafindex using spectral data," in Proc. lOth Int, Symp. Machine Pro-cessing Remotely Sensed Data (Purduc Univ .. W. Lafayette. IN), pp.333-338. 1984.

113] K. J. Ranson. L. L. Biehl. and M. E, Bauer. "Variation in spectralrcsponse of soybeans with respect to illumination, view. and canopygeometry," /111. 1. Remote Sensing. 1985.

[14] L. L. Biehl. M. E. Bauer. B, F. Robinson, C. S. T. Daughtry, L. F.Silva. and D, E. Pitts. "A crops and soils data base for scenc radiationrescarch," in Proc. 8th Int. Symp. Machine Processing (if RemotelySensed Data (Purdue Univ., W. Lafayette, IN), pp. 169-177,1982.

115] G. D. Badhwar, "Automatic corn-soybean classification using Land-sat MSS data. I. Near-harvest crop proportion estimation." RemoteSensing Environ .. vol. 14, pp. 15-29. ]984,

[16] -, "Use of Landsat-derived profile features f(lr spring small grainsclassification," Inl. 1. Remote Sensing. vol. 5, pp. 783-799. 1984.

[17] -, "Classification of corn and soybcans using multitemporal the-matic mapper data," Remote Sensing Environ., vol. 16, pp. 175-181,1985.

118] G. D. Badhwar and K. E. Henderson, "Estimating development stagesof corn from spectral data-An initial model," AKron. J., vol. 73, pp.748-755, 1981.

[19] C. S. T. Daughtry. K. P. Gallo, and M. E. Bauer, "Spectral estimatesof solar radiation intercepted by corn canopies. ,. Agron. 1., vol. 75.pp. 527-531, 1983.

[20] K. P. Gallo. C. S. T. Daughtry, and M. E. Bauer, "Spectral estimationof absorbed photosynthetically active radiation in corn canopies," Re-mOle Sensing Environ., vol. 17, pp. 221-232. 1985.

[21] G. Asrar, M, Fuchs, E. T. Kanemasu, and J. L. Hatfield. "Estimatingabsorbed photosynthetic radiation and leaf area index from spectralreflectance in wheat," Agron. J" vol. 76, pp. 300-306, 1984.

[22] J. L. Hatfield. G. Asrar. and E. T. Kanemasu, "Intercepted photo-synthetically active radiation estimated by spectral reflectance, ,. Re-mote Sensing Environ., vol. 14. pp. 65-75. ]984.

[23] G, Asrar, E. T. Kanemasu. and M. Yoshida, "Estimatcs of leaf areaindex from spectral reflectance of wheat under different cultural prac-

,tices and solar angle," Remote Sensing Environ .. vol. 17, pp. I-II,1985.

124] G. Asrar, E. T. Kanemasu. R. D. Jackson, and P. J. Pinter, "Esti-mation of total above-ground phytomass production using remotelysensed data." Remote Sensing Environ., ]985.

[25] B. R. Gardner and B. L. Blad, "Techniques for remotely monitoringcanopy development and estimating grain yield of moisture stressedcorn," Center Agric. Meteor. and Climatology, Univ, of Nebraska,Lincoln. NE. Tech. Rep. 83-9. pp, 1l ]-116, 1983.

[26] M. E. Bauer, C. S. T. Daughtry, and V. C. Vanderbilt, "Spectral-agronomic relationships of maize, soybean, and wheat canopies," inProc. Int. Colloquium Spectral Signa/ures of Objects Remote Sensing(Avignon, France), pp. 261-272, 1981.

127] E. P. Crist, "Effects of cultural and environmental factors on cornand soybean spectral development patterns," Remote Sensing Envi-ron .. vol. ]4, pp. 3-13, ]984.

[28] K. J. Ranson. C. S. T. Daughtry. L. L. Biehl. and M. E. Bauer. "Sun-view angle effects on reflectance factors of corn canopies." RemoteSensinf( Environ., vol. 18. pp. 147-161. 1985.

129] G. H, Suits, "The calculation of the directional reflectance of a veg-etative canopy," Remote Sensing Environ .. vol. 2. pp. 117-125, 1972.

[30] W. Verhoef, "Light scattering by leaf layers with application to can-opy reflectance modeling: The SAIL model," Remote Sensing Envi-ron,. vol. ]6, pp. 125-]42. 1984.

[31] J. M, Norman and J. M. Welles, "Radiative transfer in an array ofcanopies," Agron. 1., vol. 75, pp. 48]-488. 1983.

BAUER et al.: FIELD SPECTROSCOPY OF AGRICULTURAL CROPS

Marvin E. Bauer (M'86) received the B.S.A. and M.S. degrees from Pur-due University and the Ph.D. degree in agronomy from the University ofIllinois.

From 1970 to 1983, he was a Research Agronomist at the Laboratory forApplications of Remote Sensing, Purdue University, where he had key rolesin the design, implementation, and analysis of data from several major ag-ricultural remote-sensing experiments. He is currently Director of the Re-mote Sensing Laboratory and a Professor of Remote Sensing at the Uni-versity of Minnesota.

Dr. Bauer serves as Editor-in-Chief of Remote Sensing Environment. Heis a member of the American Society of Agronomy, the IEEE Geoscienceand Remote Sensing Society, and the American Society of Photogrammetryand Remote Sensing.

*Craig S. T. Daughtry received the B.S.A. and M.S. degrees from the Uni-versity of Georgia and the Ph.D. degree in agronomy from Purdue Univer-sity.

He has been a Research Agronomist at Purdue University's Laboratoryfor Applications of Remote Sensing since 1976. He has a research back-ground in crop physiology which includes both field and laboratory exper-iments to evaluate the effects of the environment on yield, photosynthesis,water relations, and chemical composition of several crops. Since joiningthe LARS staff, he has actively participated in the development of computer-simulation models of corn and soybean growth and yield. His recent re-search has concentrated on measurements and modeling of the spectral-agronomic relationships of croP. canopies and incorporating spectrally de-rived estimates of canopy leaf areil index and light interception into cropgrowth and yield models.

*Larry L. Biehl (S'71-M'80) received the B.S. and M.S. degrees in elec-trical engineering from Purdue University.

He is currently a Research Engineer in the Measurements Program areaat the Laboratory for Applications of Remote Sensing, Purdue University.He has participated in Skylab and Landsat MSS and Thematic Mapper Re-

75

search, and has had major responsibilities in NASA-sponsored field re-search. These roles have included development and testing of field spectraldata acquisition and calibration procedures, data preprocessing, and devel-oping software for effective analysis of spectral data.

*

Edward T. Kanemasu received degrees from Montana State Universityand the University of Wisconsin.

He joined Kansas State University in 1969 where he is currently a Pro-fessor of Agronomy and leader of the Evapotranspiration Laboratory. Hisprimary research interests have been in plant-water relations, evapotran-spiration, crop-yield modeling, and remote sensing. He has been an invitedlecturer at numerous conferences and symposiums in the U.S. and abroad,and has served on national committees for the USDA, NASA, and severalother agencies.

Dr. Kanemasu is a fellow of the American Society of Agronomy and hasserved as an Associate Editor and Technical Editor, Agroclimatology andCrop Modeling, of the Agronomy Journal.

*

Forrest G. Hall received the B.S. degree in mechanical engineering in1963 from the University of Texas, and the M.S. and Ph.D. degrees inphysics in 1968 and 1970, respectively, from the University of Houston.

A Mathematical Physicist, he has published models of the lunar atmo-sphere, microwave sea surface emission properties, and has conducted andpublished theoretical investigations in stochastic theory. In 1973, he joinedthe Earth Observations Division at the NASA Johnson Space Center, Hous-ton, TX. He has served as Project Scientist for the Large Area Crop Inven-tory Experiment (1974-1978) for which he received NASA's medal for Ex-ceptional Scientific Achievement, and as Chief Scientist for the EarthObservations Division (1979-1980). Since 1980, he has served as Managerof the AgRISTARS Supporting Research Project and as Ch ief of the EarthSciences Research Branch. In mid-1985, he joined the Laboratory for Ter-restrial Physics at the NASA Goddard Space Flight Center, Greenbelt, MD.

76tH'EI RANSAC'1](INS ON GEOSCIENCE AJ\D REMOTE SENSIN(;. VOL. (;E-24. NO. I. JAI\lIARY 1986

Light Interception and Leaf Area Estimates fromMeasurements of Grass Canopy Reflectance

GHASSEM ASRAR, EDWARD 1'. KANEMASU, GEORGE P. MILLER. AND R. L. WEISER

Abstract-Grassland is a major component of the E~II·th's availahleland. The vast area and "emokness of this ecusystem makes it difficultto assess its condition and monitor productivity hy traditional methods.Remote sensing potentially olfers a rapid nondestructive approach formonitoring snch e('osystems. A study was carried out in a tallgrassprairie site 11(,31' Manhattan, Kansas, during the 1983 and 1984 St'asonsto investigate the feasibility of estimating light intel'ception and greenleaf area index (LA I) from measun'ments of canopy multispel'lral re-flectance. GI'eenness (GfI) index was found to be strongly correlaledwith inten'epted photosynthetically active radiation (PAR). Two melh-ods, a direct regression (RGR) and an indirect allproadl (IN!», wereused to estimate LAI from Gfl index. The LAI valnes estimated by R(;Rmethod were consistently lower than the measul'ed ones; however, goodagreement was obtained between the LAI values (,stimated by INDmethod and the measured LA!. This suggests that Gfl transl.,rmalionof canopy spectral reflectan('e is more closely related to the fraction of'inlernpted PAR by green foliage than lhe qnantity of g"een LAI.

I. INTRODlJCTlON

THE EARTH'S surface includes a wide rangc of eco-systems and human land-use patterns with varying

vegetation covers (cropland, desert, f{)rest, and range-land). The differences in type and amount of vegetationof these ecosystems are caused primarily hy climate andsoil. Grassland covers ahout 17 percent of the Earth's landsurface [ I ] and 28 percent of the land in the United States[21· The largc extent and remoteness of these lands makeit difficult to assess their condition and monitor productiv-ity by the traditional methods. Tucker [31 evaluated sev-eral techniques f{)r the nondestructive estimation of grass-land biomass as an alternative to conventional point-sampled hand clipping. He concluded the choice of whichbiomass estimation method to use depends on the specificresearch application. Remotely sensed spectral measure-ments of reflected and emitted radiation, however, canprovide a rapid and nondestructive method f{)r monitoringthe changes in grassland ecosystems. This has been dem-onstrated by Tucker ct (II. 141 who f()UncI a strong corre-lation between integrated normalized difference indexbased on NOAA-7 satellite imagery and end-of-seasonabove-ground biomass.

Solar radiation provides the energy for photosyntheticactivities and hence plant growth. Intraleaf scattering of

Manuscript received June 16, 1')85; revised August 2'). 1')85. This workwas supported by NASA under Contract NAS-167457.

The authors are with the Evapotranspiration Lahorat<>ry, DcpariJIlCnl ofAgronomy. Agricultllre Experiment Station, Kansas Statc Ullivl~rsity, Manhattan. KS 66506,

IEEE Log Number 1\406222.

solar radiation is important in photosynthesis, since it in-creases the mean path length of absorbed energy thus fa-cilitating electron capture by plant pigments (chlorophylland carotenoids). The interception of phytosyntheticallyactive radiation (0.4-0.7 /fm) is correlated strongly withphytomass production [5]. Since light is primarily inter-cepted by plant leaves, the duration and magnitude ofgreen LAI is also correlated strongly with phytomass pro-duction [6 J.

Senescence affects the plant pigments by a selectivebreakdown of chlorophyll and carotenoids 17J. If the leafand canopy scattering characteristics remain intact as theabsorbing pigments are reduced, one would expect a di-rect relationship between plant senescence and canopy re-flectance at wavelengths of st rong pigment absorption [8].This would be opposite to the relationship existing be-tween reflectance and green f()1iage area during the growthperiod.

Previous studies [9[, [101 have shown the existence ofrelationships between PAR and canopy spectral reflec-tance for monocultural crops. Asrar et al. [I I I developeda procedure f{)r estimating both the PAR absorption andgreen LAI from measurements of recI and near-infraredreflectance in homogeneous plant canopies. This proce-dure was then evaluated under different management prac-tices and solar illumination angles for different geograph-ical locations 112]. The objective of the current study wasto develop a procedure for estimating PAR and LAI frommeasurements of grass canopy spectral reflectance.

II. SITE DESCRIPTION

The study was conducted during 1983 and 1984 at theKonza Prairie Research Natural Area (KPRNA) locatednear Manhattan, Kansas (39°9' N, 96°40' W). The pre-dominant soil at this site is a silty clay loam (Udic ustoll),typical of the Flint Hills uplands of Kansas. KPRNA is3487 ha of unplowed bluest em prairie with big bluest em(Andropogon gerardii Vitman), little bluestem (Andro-pogon scoparius M ichx.), and Indian grass (Sorghastrumnutans (L.) Nash) as the dominant species. Thirty-six otherspecies have been observed in a detailed vegetation com-position study [13]. The large number of species and thediversity in their spatial distribution results in a morecomplex canopy than monoculture crops.

Climate of the prairie uplands is humid subtropical, withtemperature ranging from ~35° to 4rc annually. Mean

o 196-2H92/86/0 100-0076S0 1.00 1986 IEEE

ASRAR eI al.: UGHT It\TERCEPTlON AND LEAF AREA ESTIMATES77

IV. DATA ANALYSIS

A simple analytical radiation transfer model was usedto relate the visible and near infrared canopy reflectanceto PAR interception (p) based on the measurements of leafscattering properties [11]; p is defined by Monsi and Sacki[ 14] as

where K I is the leaf angle shape coefficient and LAI is theleaf area index. The functional relationship between can-opy reflectance (pJ and p was expressed as

(pP,) piA) - 1) + (1 - p,(A)/Pu(A)) (1 - p)-2KPc(A) = A -7K(piA) - 1/pi ) + (Pu(A) - piA)) (l - p) -

(2)

Gne = -0.3974 PC! - 0.6849 PC2

+ 0.2564 PC} + 0.5543 PC4 (3)

where Pcl, ••• , Pc4 correspond to wavelength bands ofExotech radiometer, respectively. The values of Pc ob-tained by direct measurements with an Exotech radiome-ter or simulated using (2) can be used to compute Gne·

where Ps and Pu are soil and leaf reflectance at a givenwavelength (A), respectively, and K is a radiation extinc-tion coefficient that depends on foliage scattering propertyand canopy geometry. K values can be computed from theequations given in [11]. Thus, a set of measured Ps and Puvalues can be used to compute Pc from (2) for a range ofp values from 0 to 1. A greenness (Gn) transformationgiven by

(1)p = 1 - exp ( - K I • LAI)

ing downward to record the reflected PAR from the can-opy. The line quantum sensor was placed underneath thecanopy, below the last layer of green leaves, to measurePAR transmitted through the canopy. All three sensorswere wired into a data logger (Polycorder model 516A) forsimultaneous data acquisition. The PAR-sensor assemblywas placed at three different locations in each treatmenttransect. Five sets of measurements were made at eachlocation by transfering the line quantum sensor to differ-ent spots in the vicinity of the tripod-assembly. This se-quence of measurements were repeated until mid-after-noon for both treatments.

In 1983, four plant samples each 0.1 m2, were obtained

from three sampling locations that were established eachday of reflectance measurements on each treatment tran-sect (total of 12 samples per treatment per date). Thesesites were then marked to avoid measurements on the samesites later in the season. In 1984, one 0.1 m2 plant samplewas obtained from nine sampling locations on each of thetwo treatment transects (total of 9 samples per treatmentper date). In the laboratory, the plant samples were sepa-rated into green grass leaves, green nongrass leaves, andsenescent material. Total green leaf area of both grass andnongrass species was determined using a LiCor model LI-3100 optical area meter.

III. DATA ACQUISITION

Spectral measurements were conducted on separate sitesin 1983 and 1984. Each site was composed of two adjacenttransects of the prairie, which were separated by an accessroad (fire guard). The transects were initially covered withsenescent vegetation from previous year(s). Prior to theresumption of growth in early spring, the senescent veg-etation was removed from one of the transects by con-trolled burning. This resulted in two types of surfaces, abare soil surface that was exposed (burned treatment), anda surface that was covered with senescent vegetation (un-burned treatment).

Spectral measurements were continued throughout theentire season on days with clear sky conditions using atruck-mounted assembly equipped with Exotech model100-A and Barnes Modular Multispectral (MMR) model12-1000 radiometers. The Exotech radiometer has twowavelength bands in the visible (0.5-0.6 {tm and 0.6-0.7{tm) and two in the near infrared (0.7-0.8 {tm and 0.8-1.1{tm) regions of the spectrum each with a 150 field of view(FOV). These wavelength bands correspond to the multi-spectral scanner (MSS) on board Landsat satellites. TheBarnes MMR has three wavelength bands in the visibleregion (0.45-0.52 {tm, 0.52-0.60 {tm, and 0.63-0.69 {tm),two in the near infrared (0.76-0.9 {tm and 1.15-1.30 {tm),two in the middle infrared (1.55-1.75 {tm and 2.08-2.35{tm), and one in the thermal infrared (10.4-12.5 {tm) re-gions each with a 150 FOY. The thermal infrared wave-length band will not be considered in this report. TheBarnes MMR wavelength bands 1-4 and 6-7 correspondto wavelength bands 1-5 and 7 of Thematic Mapper (TM)sensor on board Landsats 4 and 5. Both radiometers weremounted in the nadir viewing position 8 m above the soilsurface.

Canopy reflectance measurements were replicated 20times on each transect mostly during midday and refer-enced to a painted BaS04 calibration panel approximatelyevery 15 min. Spectral reflectance and transmittance ofindividual green and senescent leaves of grass and non-grass vegetation were measured in 0.4-1.1 {tm at 0.01- {tmintervals with a LiCor model LI-1800 spectroradiometerin 1984.

The components of PAR were measured by two quan-tum sensors (LiCor model LI-190S) and a quantum linesensor (LiCor model LI-1915) during the 1984 season inboth burned and unburned treatments. The measurementsstarted during mid-morning on each date by pointing allthree sensors upward, leveling them, and sampling the in-coming PAR 10 times. These data were used for intercal-ibration of the sensors to avoid the error due to variabilityamong the sensors. The two quantum sensors weremounted on a tripod assembly and positioned above thecanopy. One sensor was facing upward to record the total(direct + diffuse) incoming PAR, and one sensor was fac-

precipitation is 750 mm/year, with variable seasonal dis-tribution resulting in many wet-dry cycles in a normalgrowing season of 180 days.

78IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING. VOL. GE-24. NO. I. JANUARY 1986

where Pn' ... , Pn correspond to wavelength bands ofBarnes MMR radiometer, respectively. Other spectraltransformations such as the near infrared to red ratio (RO)and normalized difference (ND) indices can be computedfrom the simulated (2) or measured canopy reflectancevalues by

The coefficients in (3) were derived from a data set ob-tained from measurements of the grass canopy in 1983 byan Exotech radiometer, and analyzed using a constrainedprincipal component analysis (CPCA) described in [15].A similar procedure was used to derive a Gnb index froma data set in which a Barnes MMR radiometer was used

Gnb = -0.044 PCI - 0.024 PC2 - 0.1747 Po

+ 0.7916 PC4 + 0.3875 Pcs - 0.2310 Pc"

- 0.3699 Pn (4)

GRASS CANOPYSPHERICAL,\"300

0---0 NORMALIZED DIFFERENCEe---e NIR·RED RATIOIf-...j( GREENNESS

0001020304050607080.910INTERCEPTION

Fig. l. Relationship bctween light interception and the spectral indices ofRO, ND, and Gn simulated for a grass canopy with sphcrical leaf angledistribution.

and

RO = Per'! PerTABLE I

(5) REGRESSIO"l PARAMETERS FOR THE LINEAR RELATIONSHIP BETWEEN

MEASURED AI\D SIMULArED (1) INTERCEPHD PHOTOSY'ITHETICALLY ACTIVE

RADIATION (IPAR) fOR DIFFERENT LEAF ANGLE DISTRIBUTIONS

V. RESULTS AND DISCUSSION

[PARo - (PARr + PARt)]

PARo

where PARo, PAR" and PARt are the incoming, reflected,and transmitted components, respectively.

where Pcn and Pcr are near infrared (0.8-1.1 Ilm or 0.76-0.90 Ilm) and red (0.6-0.7 Ilm or 0.63-0.69 Ilm) canopyreflectance, respectively.

Measured PAR data can be used to compute a meandaily p value as

t.!!2Lear Angl e ~1 ~

Fixed

15° 1.292 0.954 0.16730° 1.232 0.958 0.15445° 1.120 0.963 0.13060° 1. 051 0.973 0.10475° 1.313 0.965 0.148

Spheri cal 0.998 0.974 0.097

t Simulated IPAR = B1 t (Measured IPAR)

however, a 1:I relationship is observed between p and ROfor p > 0.5. This resulted in an overall nonlinear relation-ship between p and RO that was also dependent on 17 [11].ND changed proportionally with a variation in p for a widerange of p values. The sensitivity of this relationship de-creased for p > 0.75, but the relationship between p andND was found to be least sensitive to changes in 17 [11].

Gn was linearly related to p for p < 0.9, but the mag-nitude of this change was about 0.5 unit of Gn per 1.0 unitof p over this range. The p versus Gn relationship alsodepended on canopy geometry and solar angle, Fig. 2. Thestrongest relationship between Gn and p was obtainedwhen it was assumed that leaves form a horizontal greenlayer with no angular distribution. If a homogeneous anduniform canopy with spherical leaf angle distribution wasassumed, then the p versus Gn relationship was also foundto depend on 17. An increase in 17 resulted in an increasein the slope of p versus Gn line, due to changes in red andnear-infrared reflectance with 17. The diurnal variation ofthe red reflectance was found to be more dependent onsolar illumination and canopy geometry than the near in-frared reflectance [16], [17]. This is due to an increaseddiffusive component of the incoming radiation as a resultof longer path length, and increased absorption of shortwavelength radiation by vegetation at high solar zenith an-gles. In spite of the dependency of Gn on 17 and the canopy

(6)

(7)p=

A. Light InterceptionThe relationship between measured p values (7) and

those computed from (1) depends upon the leaf angle shapecoefficient (K ') for a given value of LA!. Since directmeasurements of leaf angle distribution were not avail-able, we assumed several fixed leaf inclination classes(150, 30°, 45 0, 60°, and 75 0) and spherical leaf angledistribution in computing p from (1). Table I presents thelinear regression parameters between the measured andcomputed p values for assumed leaf angle distributions.The highest R2 and lowest RMSE values were obtained forspherical distribution and 60° fixed leaf inclination angle.Therefore, either one of these leaf angle distributions canadequately characterize the grass canopy. We limit the restof our discussion to spherical distribution.

The relationships between PAR interception p, RO,ND and Gn transformations for a uniform grass canopywith spherical leaf angle distribution and a solar zenithangle (YJ) of 30°, computed from (2), (3), (5), and (6) arepresented in Fig. I. The change in RO was not propor-tional to variation of p over the complete range of p values.RO changed 0.4 unit for one unit change in p for p < 0.5;

ASRAR et al.: LIGHT INTERCEPTION AND LEAF AREA ESTIMATES 79

TABLE IIREGRESSION PARAMETERS FOR THE LINEAR RELATIONSHIPS BETWEEN

GREENESS (Gn) A"D LIGHT INTERCEPTION (p) BASED ON MEASURED ANDSIMULATED CANOPY REFLECTANCE AT DIFFERENT SOLAR ZENITH ANGLES

(The Gn values were computed using (3) and (4).)

MethodSolar ZenithAngle (Deg.) Gn Bot R!1SE

Exotech:1easured 2B-62 0.lB-0.85 0.057-0.240 -0.121 3.468 0.732 0.114

Simulated 30-60 0.00-1. 00 0.041-0.335 -0.125 3.853 0.915 0.088

BarnesMeasured 28-62 0.18-0.85 0.032-0.357 -0.275 2.748 0.817 0.095

t p = Bo + B,*Gn

00 01 0.2 03 0.4 0.5 0.6 0.7 0.8 0.9 1.0INTERCEPTION

Fig. 2. Relationship between light interception and Gn index simulated forgrass canopies with horizontal and spherical leaf angle distributions.

O.O'---'---'-----'-~--'------'----'-----'~-'--~----'----'-'100 120 140 160 180 200 220 240 260 280 300 320 340

DAY OF THE YEAR

Fig. 3. Seasonal trends in leaf area index and Gn values obtained fromspectral measurements on a tallgrass prairie in 1984.

KONZA PRAIRIE3 5 MANHATTAN, KS 0

1984 00

• LEAF AREA INOEXGREENNESS

, EXOTECHo 8ARNES

oo 0 00

00

,.:,:,: '

,Xu x

....

Q~3.0Ulwz~ 2.5w

'"to><""2.0woz

1.5<lw'"<l 1.0u.<lw...J 0.5

GRASS CANOPY.....---,c HORIZONTAL

SPHERICAL•........• 1'\=20"

0- --0 "1l= 40"0· .... ·0 T\ = 60"

025

010

015

020

030

0.35

040

where Gne and Gnb are the greenness values computedfrom (3) and (4), respectively. These relationships werethen used to estimate the green LAI from Gn data of the1984 season. The results are presented in Fig. 4. The LAIestimates based on both empirical equations underesti-mate the measured ones, in spite of relatively high R2 val-ues. This is illustrated by the slopes of the linear regres-sion lines between the estimated and measured LAIs thatwere significantly (0: = 0.0001) smaller than one. This isprobably due to the limitation of empirical relationships

curves illustrate that Gn peaks at the same time that max-imum green LAI occurs and then Gn declines due to de-crease in canopy reflectance, as a result of changes in leafcolor and structure. These data suggested that Gn indicesare correlated with photosynthetically active green vege-tation; therefore, Gn could be used to estimate green LA!.We used two different approaches to test this hypothesis.

In method 1 (RGR), empirical linear regression rela-tionships were established between measured green LAIsmoothed by a cubic spline procedure [18] and Gn valuesbased on the 1983 data set

geometry, Gn was selected as an index for estimating pfrom measurements of canopy reflectance because of theirlinear relationship to one another.

The measured p values were linearly regressed againstthe mean daily Gn values that were computed from (3)and (4) based on the canopy reflectance measurements.This subset of data then was eliminated from further anal-ysis to reduce the dependency of the empirical relation-ships on the data set. The intercepts and slopes of linearregression equations that were established between mea-sured and simulated p versus Gn values for a wide rangeof 1/ were found to be in good agreement (Table II). LowerR2 values for the measured data were due to the smallernumber of observations and inherent variability in ~hefieldmeasurements, as indicated by larger root mean squareerror (RMSE) values. A simulated relationship between pand Gn for the Barnes MMR was not obtained due to un-availability of grass leaf spectral properties for the Barnesmiddle infrared wavelength bands. It was concluded thatGn transformation of spectral reflectance measurementscan be used to estimate the fraction of intercepted PARby grass canopies.

B. Leaf Area IndexTemporal profiles of Gn computed from (3) and (4) par-

alleled the growth of green leaves in grassland, Fig. 3. The

and

LAI = -0.369 + 6.585 Gne

LAI = -0.502 + 4.541 Gnb

(8)

(9)

80IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING. VOL. GE·24. NO. I. JANUARY 1986

KONZA PRAIRIE 19843.0 MANHATTAN, KS

EXOTECH2.5 EST~ 0.061'0504 (MEAS)

Rl =0.862RMSE =0135

20

4o.--------------,KONZA PRAIRIE 1984MANHATTAN, KSEXOTECH

3.0 EST.~-0025+0.821(MEAS)R2~0.876RMSE ~0.197

Fig. 4. Relationship between measured and estimated leaf area index basedon the RGR method using measured canopy reflectance by (a) the Exo-tech and (b) Barnes MMR.

........ . .. - ..:._ e.:. e. e... _....0.0. e•••

1 5

a

b

oo

oo 0

o

oo

•• 0•• 0.... -.o 0 0o 0 _.o:....

o 0

8ARNES MMREST~·0.137+0.912 (MEASJR2~0.899RMSE ~0.195

1.0

2.0

x 2.0wo~<t I 0wa::<t

~ 00W...J 40owf-

~ 30~(/)

w

000.0 0.5 10 15 2.0 2.5 3.0

MEASURED LEAF AREA INDEX

Fig. 6. Relationship between measured and estimated leaf area index basedan the IND procedure and the measured relatianship between p and Gnfor the (a) Exotcch and (b) Barnes MMR.

3.0

a

b

:....... ..- ..: .05 1.0 15 2.0 2.5

MEASURED LEAF AREA INDEX

BARNES MMREST =0,043+ OA94(MEAS.lR2= 0.861RMSE ~0127

... .....0.5

1.0

0000

~ 1.0oz

1.5

~ 3.0{~ 25'"" .5; 2.0w

10 20 30MEASURED LEAF AREA INDEX

Fig. 5. Relationship between measured and estimated leaf area index basedon the IND procedure and the simulated relationship between p and Gn.

where In (1 - p) is the natural logarithm of the arithmaticmean of PAR interception estim~ted from grass canopyreflectance measurements, and K' is a mean leaf angleshape coefficient. In homogeneous canopies with spheri-calleaf angle distribution, K' is defined as

that are usually data-set specific. Therefore, the effects ofatmosphere, canopy architecture, soil background, andsun angle are not taken into account.

In method 2 (IND), the estimatedp values from Gn wereused in an indirect approach to compute green LAI as

and for canopies with fixed leaf angle, K' can be com-puted from the equations given in [II].

Fig. 5 shows the estimated and measured LArs for thep versus Gn relationship, which was derived from simu-lated data (2). Good agreements (R2 = 0.869) were ob-tained between the estimated and measured LAI values;

however, the estimated LAI values were slightly largerthan the measured ones for LAI > 2.0. This is illustratedby the slope of the linear regression relationship betweenthe estimated and measured LAI values, which was sig-nificantly (a = 0.0001) greater than one. This was prob-ably due to the wide range of canopy conditions that wereassumed in deriving the p versus Gn equation (Table II).However, the magnitude of RMSE of estimated LAI's(0.314) was comparable with the standard error of meanmeasured values. A similar procedure was used to com-pute LAI indirectly from Gn values using the measured pversus Gn relationships (Table II). The results are pre-sented in Fig. 6. The estimated LAI values based on can-opy reflectance measurements with the two different sen-sor systems (Exotech and Barnes MMR), were in closeagreement with the measured ones as indicated by the R2

and RMSE presented in Fig. 6. The intercepts for the lin-ear regression relationships between the estimated andmeasured LArs were not different from zero, but theslopes were significantly (a = 0.0001) different from one.The magnitude of the RMSE values were smaller than theone for LAI estimates presented in Fig. 5. In general, theIND method resulted in a better overall agreement be-tween estimated and measured LAI values than the RGRmethod. This suggests that Gn transformation of canopyspectral reflectance measurements is closely related to thefraction of intercepted PAR by green foliage rather thanthe quantity of green LA!. Although green LAI and PARinterception are strongly correlated, the PAR interceptionalso depends on the canopy geometry and the solar angle.These factors also affect the quality and quantity of theradiation that is reflected from the plant canopy. There-fore, the IND procedure accounted for some of these fac-tors, as demonstrated in (2), and resulted in a more reli-able estimate of green LAI in grass canopies.

(10)

(11)

4.0

.:oo

o 0o •o 0

o 0oo 0

K' = 0.5/cos 11

LAI = -In (1 - p)1 K '

. .\. ..000

o 0 0- .....

5.0 KONZA PRAIRIEMANHATTAN, KSSIMULATEDEST~-0.169+ 1198(MEASJR2 ~ 0.869RMSE ~0.314

oo 0"0 0

O.o~· ••••00

xwo": 4.0<twa::<t 3.0...<tw...J

20owf-<t

~ 10(/)w

ASRAR <'I al.: LIGHT INTERCEPTION AND LEAF AREA ESTIMATES 81

*

*

Ghassem Asrar received the M.S. degree in soiland fluid mechanics and the Ph.D. degree in mi-eroclimatology from Michigan State University.

He is currently an Assistant Professor of Ag-ricultural Physics at the Evapotranspiration Labo-ratory, Kansas State University. His primary re-search interests have been in radiative transfer inplant canopies, fluid flow through porous media,microclimatology, remote sensing, and modelingof soil-plant-atmosphere continuum.

Dr. Asrar is a member and referee !()r severalnational and international professional societies.

Edward T. Kanemasu received degrees fromMontana State University and the University ofWisconsin.

He joined Kansas State University in 1969where he is currently a Professor of Agronomy andleader of the Evapotranspiration Laboratory. Hisprimary research interests have been in plant-waterrelations, evapotranspiration, crop-yield model-ing, and remote sensing. He has been an invitedlecturer at numerous conferences and symposiumsin the U.S. and abroad. and has served on national

committees f()r the USDA, NASA, and several other agencies.Dr. Kanemasu is a fellow of the American Society of Agronomy and has

served as an Associate Editor and Technical Editor, Agroclimatology andCrop Modeling, of the Agronomy Journal.

synthetically active radiation estimated by spectral reflectance," Re-mote Sensing Environ., vol. 14, pp. 65-75, Jan. 1984.

[II] G. Asrar, M. Fuchs, E. T. Kanemasu, and J. L. Hatfield, "Estimatingabsorbed photosynthetic radiation and leaf area index from spectralreflectance measurements in wheat," AR'on. 1. , vol. 76, pp. 300-306,Mar.-Apr. 1984.

[12J G. Asrar, E. T. Kanemasu, and M. Yoshida, "Estimates of leaf areaindex from spectral reflectance of wheat under difl'erent cultural prac-tices and solar angle," Remote Sensillg Enviroll., va\. 17, pp. I-II,Jan. 1985.

[13[ R. L. Weiser, G. Asrar, G. P. Miller, and E. T. Kanemasu, "Assess-ing biophysical characteristics of grassland from spectral measure-ments," in Proc. 10th 1111.Svmp. Mach. Process. Remot. Sens. Data(Purdue Univ., West Lafayette, IN), pp. 357-36], June 1984.

[141 M. Monsi and T. Sacki, "Veber dem liehtfaktor in den pflanzenge-sellsehaften und seine bedeutung fuer die stoffproduktion," Japan. 1.Bot., vo\. 52, pp. 22-52, 1952.

[15] G. P. Miller, M. Fuchs, M. J. Hall, G. Asrar, E. T. Kanemasu, andD. E. Johnson, "Analysis of seasonal multispectral reflectanees ofsmall grains," Remote Sensing Environ .. vol. 14, pp. 153-167, Jan.1984.

[161 J. D. Kollenkark, C. S. T. Daughtry, M. E. Bauer, and T. L. Housely,"Influence of cultural practices on the reflectance characteristics ofsoybean canopies," Agron. 1., vo\. 74, pp. 751-758, July-Aug. 1982.

[17] J. C. Kollenkark, V. C. Vanderbilt, C. S. T. Daughtry, and M. E.Bauer, "Influence of solar illumination angle on soybean canopy re-flection." ApI'/. Opt .. vo\. 21, pp. 1179-1184. 1982.

[181 B. A. Kimball, "Smoothing data with cubic splines." Agron J.• vol.63, pp. 126-129, Jan.-Feb. 1976.

REFERENCES

VI. SUMMARY AND CONCLUSIONS

This study was conducted to assess the feasibility of es-

timating intercepted PAR and green LAI from measure-

ments of grass canopy multispectral reflectance.

PAR interception was found to be more linearly related

to Gn index than to RO or ND indices. The p versus Gnrelationship, however, was found to depend on canopy ge-

ometry and solar angle due to variation of both canopy

reflectance and p. Empirical linear regression equations

were derived, based on a simple radiative transfer model

and direct field measurements, that can be used to esti-

mate p from measurements of canopy reflectance by the

radiometers that mimic the MSS and TM sensor systems

on board Landsat satellites.

An RGR was used to estimate the green LAI from the

Gn index. The estimated LAI values by the RGR method

were consistently smaller than the measured ones. In an

IND, p values estimated from Gn index were used to es-

timate LAI from an exponential relationship between LAI

and p, by assuming a spherical leaf angle distribution. The

estimated LAI values based on the IND method were in

good agreement with the measured ones. These results

suggested that the Gn index is more closely related to the

quantity of intercepted PAR by green foliage than green

LAI. Although green LAI and PAR interception are

strongly correlated, the interception of PAR also depends

on the canopy geometry and solar angle. These factors

affect the quantity and quality of radiation that is reflected

from the plant canopy. The IND procedure accounted for

some of these factors and, hence, resulted in a more re-

liable estimate of green LAI in grass canopies.

ACKNOWLEDGMENT

We thank D. W. Reed, P. Chapman, and J. M. Kileen

for their technical assistance.

[I] G. L. Ajtay, P. Ketner, and P. Durigneaud, "Terrestrial primary pro-duction and phytomass," in The Glohal Carhon Cycle, B. Bolin. E.T. Degens, S. Kempe, and P. Ketner Eds. New York: Wiley, 1977,pp. 129-181.

[21 R. G. Bailey, "Description of the ecoregions of the United States,"U. S. Dep. of Agriculture, Intermtn. Forest Service, Ogden, UT, 1978.

[3] C. J. Tucker, "A critical review of remote sensing and other methodsfor non-destructive estimation of standing crop biomass," Grass For-aRe Sci., va\. 35, pp. 177-182, 1980.

[4] C. J. Tucker, C. L. Vanpraet, M. J. Sharman, and G. VanIttersum,"Satellite remote sensing of total herbaceous biomass production inthe Senegalese Sahel: 1980-1984," Remote SensinR Environ., 1985.

[51 J. L. Monteith, "Solar radiation and productivity in tropical ecosys-tems," J. Appl. Ecol., vo\. 9. pp. 747-766, 1970.

16] G. Asrar, E. T. Kanemasu, R. D. Jackson, and P. J. Pinter, Jr., "Es-timation of total above-ground phytomass production using remotelysensed data," Remote SensinR Environ., vo\. 17. pp. 211-220, 1985.

[7] J. E. Sanger, "Quantitative investigations of leaf pigments from theirinception in buds through autumn coloration to decomposition in fail-ing leaves." Ecology, vo\. 52, pp. 1075-1089, 197\.

[8] C. J. Tucker, "Post senescent grass canopy remote sensing." RemoteSensinR Environ., va\. 7, pp. 203-2\0, 1978.

[9] C. S. T. Daughtry, K. P. Gallo, and M. E. Bauer, "Spectral estimatesof solar radiation intercepted by corn canopies," ARron. J., vol. 75,pp. 527-731 May-June 1983.

[10] J. L. Hatfield. G. Asrar, and E. T. Kanemasu, "Intercepted photo-

------------------------ ••••••• 182 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. GE-24, NO, I, JANUARY 1986

George P. Miller received the B,S, degree fromColorado State University in 1973 and the M,S,degree in soil science from Washington State Uni-versity in 1981.

He is currently a Research Assistant with theEvapotranspiration Laboratory at Kansas State Uni-versity,

Mr. Miller is a member of the American Soci-ety of Agronomy,

R. L. Weiser received the B,S, degree in crop sci-ence from Oregon State University in 1983, He isworking toward the M,S, degree in agronomy atKansas State University where he is a GraduateResearch Assistant studying under Prof. E, T. Ka-nemasu in the Evapotranspiration Laboratory,

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. GE-24, NO. I, JANUARY 1986

Spectral Components A.nalysis: A Bridge BetweenSpectral Observations and Agrometeorological

Crop ModelsCRAIG L. WIEGAND, ARTHUR 1. RICHARDSON, AND PAUL R, NIXON

83

LAIIYI x YIELD/LAI = YIELDIYI (2)

where VI denotes anyone of several spectral vegetationindices computed from remote spectral observations; LAIis leaf area index, APAR is absorbed photosyntheticallyactive radiation, and YIELD is the salable plant part, e.g.,grain of cereals, lint of cotton, root of beet, harvestedphytomass of grasses or legumes for hay or silage, to in-terpret the information conveyed by canopies.

One objective of this paper is to report the results ofexperiments conducted under the AgRISTARS programthat were specifically designed to obtain the relationshipsfor each of the terms of (l). One experiment was con-ducted in the spring and summer of 1983 on cotton (Gos-sypium hirsutum, L.) and the other was conducted in thefall and winter, 1983-1984, on spring wheat (Triticumaestivum, L.). The only previous experiments in which allthe necessary measurements had been made were con-ducted at Purdue University using corn (Zea mays, L.)and soybeans (Glycine max, L.) as the test crops [6], [5],and Kansas State University using winter wheat [9], alsounder AgRISTARS sponsorship. Hatfield et at. [8], usedthe APAR versus LAI relation from the Hipps et at. [9]study in connection with LAI and VI data from the U.S.Water Conservation Laboratory's serial cereal experimentto estimate APAR from VI. Similarly, Wiegand and Rich-ardson [19] used the intercepted PAR (IPAR) versus LAIrelation for sorghum (Sorghum bicolor, L. Moench) fromMaas and Arkin [131 to complete the spectral componentsanalysis identity expressed by (l),

The second objective of this paper is to document andbriefly discuss how the information conveyed by canopyobservations helps form a bridge between spectral obser-vations and agrometeorological crop development andyield models, especially in connection with large-area cropcondition and yield estimates,

Itbslract-Spectral observations have been acknowledged to indicategeneral plant conditions over large areas but have yet to be exploited inconnection with agrometeorological crop models. One reason is that itis not yet appreciated how periodic spectral observations of row-croppedand natural plant canopies, as expressed by vegetation indices (VI), canprovide information on important crop model parameters, such as leafarea index (LAI) and absorbed photosynthetically active radiation(APAR).

Two experiments were conducted under AgRISTARS sponsorship,one with colton and one with spring wheat, specifically to determinethe relationships for each term in the "spectral components analysis"identity

LAI/VI x APAR/LAI = APAR/VI.

LAI and APAR could, indeed, be well estimated from vegetation indicessuch as normalized difference(ND) and perpendicular vegetation index(PVI)-apparently because of the close relation between tbe VI andamount of photosynthetically active tissue in the canopy. APAR and VImeasurements are similarly affected by solar zenith angle (SZA), andLAI can be divided by cos SZA at the time of the VI and APAR mea-surements to achieve correspondence. APAR, ND, and PVI plottedagainst LAI all asymptote to limiting values in the same way yield doesas LAI exceeds 5, further linking canopy development to yield capabil-ity. In summary, the spectral components analysis results presented addcredence to the information conveyed by spectral canopy observationsabout plant development and yield, and establish a bridge between re-mote observations and agrometeorological crop modeling through thevariables of mutual concern, LAI, biomass, and yield.

1. INTRODUCTION

AGRICULTURAL scientists and engineers have de-veloped crop growth, development, and yield models

that use weather and soils data as inputs [3 J, [101. But theplant canopies that develop will integrate the soil and aer-ial environments and also express development, stress re-sponse, and yield capabilities [22], [24J. Thus, there isopportunity to use direct canopy observations obtained re-motely in conjunction with the agrometeorological modelsto either provide selected model inputs or as feedback tothe agrometeorological models to keep them tracking ac-tual crop performance [19], [23], and [251.

Spectral components analysis (24) is a bridge betweenthe spectral observations and crop modeling. In its present

form, it uses the identities

LAIIYI X AP AR/LAI

and

AP ARIYI (1)

Manuscript received May 10, 1985; revised August 20, 1985.The authors are with the Agricultural Research Service, U.S. Depart-

ment of Agriculture, Weslaco. TX 78596.IEEE Log Number 8406223.

II. METHODS

The cotton cultivar, McNair 220, was planted March 10,1983, in north-south oriented rows spaced 1.02 m apart.

U.S, Government work not protected by U.S. copyright

,84 IFEE TRANSACTIONS ON <OEOSCIE:-<CE MW REMOTE SENSI:-<Ci. VOL. GE·c4. NO. I. JAMJARY Il)X6

and

I ProJuct namcs arc mcntioncd !(If thc convcnience of the reader anJ Jonot infer prefcrential treatmcnt or cnJorsclllent by thc U.S. Departmcnt ofAgriculturc ovcr other proJucts that may be available.

PVI = 0.580(RIR) - 0.815(VIS) - 0.410. (4)

The PVI equation is specific for the soil, a Raymondvilleclay loam (Vertic Calciustolls), of this studv and is basedon the soil line "

using a Mark-lI radiometer [211 which has a visible (VIS)band in the 0.63-0.69-p,m-wavelength interval, and a re-flective infrared (RIR) band in the 0.76-0.90-p,m interval.The spectral band responses plus incident solar radiationand time of observations were electronically logged con-currently I()r each of I()ur observations per replicate andreduced by the procedures described by Richardson [161.The vegetation indices employed were the normalized dif-ference (ND) and perpendicular vegetation index (PV!)117] for which the equations are

ND = (RIR - VIS)/(RIR + VIS) (3)

(5)VIS = -1.92 + 0.789(RIR)

as determined from bare soil reflectance factor observa-tions taken in the uncropped, interplot alleys each time thecotton and wheat canopy observations were taken.

Photosynthetically active radiation (PAR) incident (I)on, transmitted (T) through, and reflected (R) from thecotton and wheat canopies, as measured with Ll-COR®Iline quantum and irradiance sensors, provided the data torabsorbed PAR, (I-T-R)//. termed APAR. For the fieldmeasurements, an upward looking line quantum sensorwas inserted perpendicular to the rows below the canopiesto measure T, and a line quantum sensor was inverted 30cm above the canopies and perpendicular to the rows tomeasure R from the canopy plus soil. The irradiance sen-sor was moved from plot to plot on a 2-m-tall stand thatcould be leveled. All three sensors were calibrated to yieldthe same value of irradiance when simultaneously exposedto the sun.

At the time of each sampling, up to two days were re-quired to determine LAI, one day to make the PAR sensorobservations. and about I h to make the reflectance factormeasurements. Thus it was impossible to make all the nec-essary observations on the same day. Table I summarizesthe dates on which the variou~; measurements were madefor each experiment. When data from various sources weremerged, the sample date was considered to be that of thePAR or light absorption data and it was used as measured.The PAR data were paired with the reflectance factor datataken the same work week, except in January 1984, whena siege of extremely cloudy weather prevented our acquir-ing the reflectance factor data within three days of the PARmeasurements. For that date we plotted the VI versus timeand interpolated. LAl data for the eight measurementdates lor cotton and six I()r wheat were plotted versus timeand the LAI was interpolated to the dates of the PAR mea-surements.

Three treatments were thinned to 99 000 plants/ha, whilea fourth was thinned to 52 000 plants/ha. One of thedensely planted treatments (MC I) received an applicationof the growth regulator mepiquat chloride at the rate of 74g/ha active ingredient applied at pinhead-size square (April21). Another of the densely planted treatments (MC2) re-ceived ~ split application of the growth regulator (49 g/haon AprIl 21, and 25 g/ha at first bloom on May 19). Theremaining densely planted treatment (NT) and the treat-ment thinned to 52 000 plants/ha « NT) did not receivethe growth regulator and were maintained as controls. Thetreatments were each replicated three times in a random-ized block design. Individual treatment plots were 10 mX 15 m in size.

The spring wheat cultivars Aim, Nadadores, and Ya-varos were seeded on November 17, 1983, at the rate of80 kg/ha with a commercial drill in rows spaced 0.2 mapart. Individual plots were 12 m X 18 m in size. In halfof each plot, 100 kg . N/ha, 33 kg . P/ha. and 33 kg .K/ha had been applied prior to planting while the otherhalf received no fertilizer. Populations achieved as countedtwo weeks after emergence were 302,313, and 197 plants/m2 for Aim, Nadadores, and Yavaros, respectively. TheAim and Nadadores cultivars were planted in randomizedcomplete blocks with three replicates for each of two rowdirections, north-south and east-west. The Yavaros was asingle unreplicated planting in east-west rows adjacent tothe other plots. For both cotton and wheat the culturalpractices were typical for the area.

The measurements needed to determine each term in(1) and (2) were made. Leaf area index was obtained pe-riodically by harvesting I-m row segments of cotton and0.24 mC (0.4 m X 0.6 m) areas of wheat. Leaf blades weredetached at the petiole (cotton) and at the ligule (wheat)after Feekes growth stage 5 [12]. The \caves from twocotton plants within the I-m row segment samples wereoptically planimetered. Then the leaves from the two plantsand the leaves from the rest of the plants were oven dried.The relation between leaf dry weight and leaf area for thetwo plants was used t(H the whole samples. The equationfor the nontreated (NT) controls was LA = -389.9 +324.3 (LW) - 3.376 (LW)C and for the mepiquat chloridetreated (MC) cotton plants and LA = - 260.2 + 279.1(LW) - 3.135 (LW)C where LA is leaf area in square cen-timeters per plant and LW is leaf dry weight in grams. Thecoefficient of determination (RC

) in both cases exceeded0.99. LAI was determined on eight dates during the sea-son.

For wheat, the equations published by Aase [II, wereused to estimate leaf area in square centimeters from leafdry weight after Feeks [121 stage 5 and from above groundplant weight up to stage 5. Nongreen leaf and plant ma-terial was removed from the samples so that our LAI areeffectively green leaf area indices (GLAl). LAI was thencalculated f()r the known sample area. LAI of wheat wasdetermined on six dates.

The vegetation indices used were calculated from re-flectance factor observations made approximately weekly

WEIGAND el al.: SPECTRAL COMPONENTS ANALYSIS 85

APAR°1.18-8.1l/PVI .875

~-~--------- -----

TABLE II

EQUATIONS FOR EACH TERM OF THE SPECTRAL COMPONENTS ANALYSIS

IDENTITY OF (I) FOR BOTH COTTOI' AND WHEAT

TABLE I

DATES IN 1983 AND 1984 ON WIIICH MEASl:REMENTS FOR VARIABLES IN (I)AND (2) WERE TAKEN

(Days of year. in parentheses. follow calendar date.)

III. RESULTSA. Data Presentation

The equations that expressed the relations obtained foreach of the terms in (1) for the two crops are summarizedin Table II. Equation (5') (given in the Table) representsthe product (l - R)I I times (I - T)I I in which Rand Tare nonlinearly regressed onto LAI. Therein, (I - R)II isinterpreted as the net downward flux of PAR and (I - T)II as that fraction of the downward flux that is interceptedby (not transmitted through) the canopy.

There is a close relation between leaf area index and thespectral vegetation indices (see (3), (3a), (4), and (4a) inTable II). Thus the vegetation indices are a good measureof the size of the photosynthetic apparatus of the crop can-opies. As generally accepted in the literature, (5) and (Sa)demonstrate that APAR can be estimated very well fromLAI. The relation between APAR and the vegetation in-dices is closer for cotton than for the pooled wheat culti-vars (see (6) and (7) versus (6a) and (7a) and Fig. I). Thedata indicate that APAR can be satisfactorily estimatedfrom vegetation indices as has been demonstrated also by[6] and [8].

Because the vegetation indices can respond to non leafphotosynthetically active tissues such as heads and leafsheaths of cereals, they may be more accurate monitors ofthe photosynthetic capacity of standing canopies than isgreen leaf area index (GLAI), per se. This distinction maypermit the vegetation indices to characterize canopiesmore accurately for light interception or absorption thandoes GLAI during crop senescence, when leaves must beportioned into living and nonliving tissue parts and non-leafy photosynthetic tissue is excluded.

The RIR reflectance factors respond to the solar zenithangle due to the increasing optical path length through thecanopy as solar zenith angle (SZA) deviates from nadir[Ill. Thus, the vegetation indices, which are dominatedby the RIR reflectance factors, also respond to solar zenithangle [20]. This response agrees with the expected in-crease in APAR as expressed in light interception equa-tions by dividing LAI by cos SZA [14 J. This means thatfor measurements of APAR and VI made at the same time,the cas SZA effect cancels out since it occurs in both thenumerator and denominator of the right -hand side of (1).For APAR inferred from VI, the SZA effect is included inVI.

Cotton canopies are planophile (horizontal leaf dis-

The program is a least squares procedure that fits the datato each of the following equation forms: Y = A + (B*X),Y = A* EXP (B*X), Y = A*(X**B), Y = A + (B/X) , Y= I/(A + B*X), and Y = XI(A + B*X). Applied to dataherein, Y is the numerator variable and X the denominatorvariable of the terms in (1) and A and B are arbitrary coef-ficients. When more than one equation yielded almost thesame coefficient of determination, we plotted the data andthe curve for the contending equations and selected theequation form that fit the data best, as judged visually,over its entire range.

----yTfiT-

.914

.789

.973

.979

.979

LAI

Caeff. 02Det~~

20 Jun (171)

I Jul ( 186)

3 _,uq (211)17 ,~ug (229)

27 Apr (118)

28 Mar (088)

S Mar ~OfiH)

28 Apr (:18)

6 May~------16,17 May (136,137)23 May (143)

I Jun (112)2 Jun (159)

6 ~ec (340)13 Dee (347)

11, 12 Jan (011, 012)

13, 14, Feb (044, D45)

Measurement s

Equat ion

IJ Feb (04·11

29 ~'ar (089)

28 Feb (059)

15 [)ee (349)

11 Jan (all)

Crop

Cotton: LAI",-.006e6.54(ND)

LAlo .043e·13'( PVI)

',heat: LAlotiD/(3.26-2.94(NOij .941

L'I°.131e·I!l8(PV:) .904

Reflectance,~actor

l1'pr (101)IR Apr (108)25 Apr (lIS)

2 r~ay (122)/fMdY~~124)16 :'1ay !136123 May (143) 23 May (143)

1 JUIl (152) 2 Jun (IS3)9 3un (160)

IS Jun (166) 14 Jun : 165)20 Jun (171)27 Jun (178) 27 Jun (178)6 Jul (187)

14 Jul (195:22 Jul (203)28 Jul (209)

3 Auq (111,

5 Dee (339)12 Dee (346)lq Dee (363)6 Jar! {006)3 Feb (034)

16 Feb (047124 FeD (855)29 Feb (060)

S Mar ~(]6J.l16 Mar:(76)22 Mar ':G82l18 Mar (088)3l\PfT094 )

I~ ;~; !ig~l?4 Apr (115)

(I"Op

Cotton( 1983)

'vJheat

LAI vs VI (1st term)

APAR vs VI (right hann side)

Cotton: APARol.07NO(3.85)

APARo.OIO PVI(1.25)

Wh,'at: APAR.o.032e3.64(NO)

APAR vs l.t\.I (2nd term)

Cotton: AP,Ro.418(LAI·115) .924

APARo( 1_ .048LAI- .186)( I-I .Ole-· 79': LPd): (',')

',heat: AP,~Rol.01-.296/LAI .943

Yield of seed plus lint (cotton) and grain' (wheat) wereobtained from 5-m row segments per replication and three0.24-m2 samples per replication for cotton and wheat, re-spectively.

The data for each of the terms of (1) were submitted toa program called W 1105, from the statistical package pro-vided by International Business Machines for the prede-cessor local IBM 1800 computer as adapted to the currentin-house Hewlett-Packard Model 1000 mini-computer.

86 IEEE TRANSACTIONS ON GEOSCIEr-iCE AND REMOTE SENSIMi, VOL. GE24, NO. L JANUARY 19X6

COTION '83 WHEAT'83-'84

1.0 1.0

x <NT 8+ ,",u.8 ~ ,8 <>0 I<T <> Y"'VAROS

'P A ue II •• NAOAOORES * .•I 0 UC2 0 ••'" ,6 o*'A .6 ~.•I.:::. A ••Pi

,. .4 ••..•"-

-lJ- •• ",..•,2 ° ,2 <>

Ii" ••0,0 0,0

,5 ,6 ,7 ,8 .9 ',0 .5 ,6 ,7 .8 ,9 1.0

ND ND1.0 1.0

dt\io !'J< ••••.8 A ,8 <>

'P xo *;0:I 0 ••

'" .6 I/"

l~I

A.:::.Pi ,4..•

""~,2 fl'0 .2 'i

0,0 0,0a '0 20 30 40 '0 20 30 40

PVI PVI

Fig, ~ APAR versus ND and PVI, respcctively, Illl' hoth colton and wheat(right side or (1)).

Fig, L Diurnal (solar zenith angle) etfect on light interception for non-thinned NT treatment colton on three dates corresponding to plant groundcover percentages of 23, 72, and 86 percent.

There are both seasonal (time of year) and diurnal (timeof day) SZA effects on APAR and VI measurements IfAPAR and VI measurements are taken within 2 or 3 h ofsolar noon in summer the effect is quite small because thesun is closest to nadir and Ilcos SZA changes slowlyaround solar noon (cas 10°, 26°, and 41 ° are 0.98, 0.90.and 0.75, respectively). In winter the sun is quite low inthe sky so that the SZA is larger and the diurnal effect on

play), especially during midday since they are heliotropic,Wheat canopies generally are more erectophile (verticalelements including leaves). Diurnal light interceptionmeasurements were made on three dates in the NT treat-ment cotton canopy of this study, and on seven dates inthe Yavaros cultivar of spring wheat of this study as re-ported by Richardson and Wiegand r 18J, Light intercep-tion was very weakly affected by time of day for cotton(Fig. I) especailly for the lowest (23 percent) and highest(86 percent) plant cover on May 6 and June 28, respec-tively. Plants covered 72 percent of the ground area onJune 2. For the wheat, both the RIR reflectance factor andlight transmittance through the canopy (interception = I- T) were linear versus l/cos SZA over a twofold rangein l/cos SZA [18], that is for a SZA within 60° of nadir(cas 60° = 0.50).

Based on these results, we now recommend that LAI in(I) be normalized by cas SZA at the time of the VI andAPAR measurements. Measured values of VI and APARalready contain the effects of SZA. Thus, (I) becomes

LAI/cos SZAVI

APAR APARx -----~ - ---LAI/cos SZA VI

(I ')

APAR and VI measurements is more serious. Our wheatmeasurements were made from December through Marchand our cotton measurements from April through July. Ittook us about 5 h, beginning usually at 10:30 local clocktime (local standard time except daylight savings time afterthe last Sunday in April) to make the light transmissionmeasurements in both the wheat and the cotton experi-ments on clear days and longer if we had to wait for cloudsto pass. In contrast. the reflectance factor measurementsto calculate NO and PVI could be made in at most 45 min,beginning usually at 1300 h. Both the seasonal and diurnaleffects of SZA on APAR and VI were greater for the wheatthan for the cotton data and probably resulted in less ofthe variation being accounted for in Table II in (6a) and(7a) for wheat than by (6) and (7) for cotton.

In Fig. 2. where all treatment mean data points not ob-scured by others are shown. the scatter in the observationsof APAR over the season is greater for the three wheatcultivars than for the one cotton cultivar treated withgrowth regulator. For the wheats of Fig. 2. it is evidentthat APAR was greater for Aim for a given value of thevegetation index than for the other two cultivars. Statisti-cal analysis [IS J has shown that Aim was more efficient inintercepting light than was Nadadores, i.e .. that equationcoefficients difl'ered statistically significantly.

The other characteristic of the wheat data apparent inFig. I is that NO reaches a maximum value of approxi-mately 0.9 while APAR is still increasing. whereas APARbecomes asymptotic to a limit of 0.95 while PVI is stillincreasing. The saturation of NO at relatively low vege-tation greenness is one of its disadvantages 120].

In summary of the above discussion. seasonal effects onsolar zenith angle. time of day differences in determiningAPAR and the vegetation indices, and cultivar differencesall contribute to scatter in the APAR versus VI data of

WEIGAND et al.: SPECTRAL COMPONENTS ANALYSIS 87

~o

Fig. 5. Relation between leaf area index (LAI) and each of the variables,grain yield (YIELD), intercepted photosynthetically active radiation({PAR), and perpendicular vegetation index (PVI) for sorghum. (see [23]for data.)

I

1.0 [ +-+ 40--------~: ________o-O

.8 + 0 30/0/ ~D--------~ .8

+ /N'-'R 0-----o ./

~ r/ 20 l::ea .4

10.2 0

20 Ii:

10

30

2

L.ll

0.0o

Fig. 4. The same relations as in Fig. 3, but for wheat.

8 JOO Zl5A_A-A--:80- IP~--- x ............~;...--c:> If 80 20c:>-~. eo g J5Ul

~~

~ p.,

~2

.v 10

jl/ 20 5

D0 0

0 2 4 8 8

LAI

gested the spectral components analysis approach in thefirst place.

The right-hand sides of (1) and (2) indicate that cumu-lative APAR during the reproductive part of a crop's lifecycle should relate to yield providing yield is limited bythe size of the plant canopies achieved. For cultivars ofadapted crops grown commercially in subhumid and drierclimates, water availability often limits the LAI achievedby the canopies and yields are reduced proportionately.Even if high LAI's are achieved, low yields can occur dueto catastrophic events, such as epsiodic weather, e.g., frostduring anthesis, or outbreaks of insects that consume orcause abscission of fruiting forms. Thus an adequate can-opy for effective light interception is a necessary, but nota sufficient, condition for high yields. The weather data

0.0o 3

L.ll

Fig. 3. Relation between leaf area index (LA!) and each of the variables,absorbed photosynthetically active radiation (APAR), and two vegetationindices, normalized difference (ND), and perpendicular vegetation index(PVI) for colton.

B. Implications of FindingsThe findings have important implications for applying

agrometeorological crop models to many fields (largeareas) since the spectral observations can be made fromaircraft and satellite platforms with sensors that view largeareas. Such observations can adequately estimate APARand can closely follow canopy LAI development. Thus theycan be responsive to soil conditions, disease, water, andother stresses that affect development of the photosyn-thetically active tissue area of the canopies. The findingsare particularly useful in integrating the general under-standing among plant productivity, plant canopy develop-ment, and light absorption. Fig. 3 presents the relationbetween LAI and the variables ND, PVI, and APAR forthe cotton as expressed by (3), (4), and (5'), respectively.Likewise, Fig. 4 presents the same relations for wheat asexpressed by (3a), (4a), and (Sa). The similarity in shapeof the APAR and vegetation index graphs is apparent;expressions such as (6), (6a), (7), and (7a) illustrate thatAPAR may be generally estimable from vegetation in-dices.

Wiegand and Richardson [17] discussed how APAR (orIPAR) and various vegetation indices become asymptoticto a limiting value as LAI increases. Fig. 5 developed fromtheir data illustrates that YIELD and the VI during thereproductive stage of crop development (early grain fillingof cereals, boll growth of cotton) versus LAI have the sameshape. Thus light absorption, vegetation index, and yieldbehavior of crops versus LAI are mutually consistent andin agreement with field observations. This internal con-sistency strengthens the probability that correct infer-ences about present crop condition and yield capabilitycan be reached from periodic canopy spectral characteri-zation as expressed by vegetation indices. This contentionis strengthened for large area application since it was theLandsat spectral data for sorghum obtained for commer-cial fields in south Texas over multiple years that sug-

Fig. 2 for the wheat cultivars, yet the relationships aresatisfactory. For cotton, the APAR versus VI relations areimpeccable.

As we anticipated (2) did not apply well in either of ourexperiments because a limited range in LAI was achievedamong treatments (and consequently in APAR and VI),the Nadadores wheat cultivar is adapted for forage pro-duction but not grain production (its harvest index, theratio of grain to above ground phytomass, was 0.22 com-pared with 0.34 for Aim and 0.47 for Yavaros), and themepiquat chloride treated cotton yielded somewhat betterthan the control cotton even though it had slightly lowerLA!. (Mepiquat chloride shortens both main stem andbranch internodes and thickens leaves slightly [7], butyield responses have been mixed [26]). Nonetheless, webelieve, (1) and (2) are applicable to commercial fields thatvary in soil, tilth and tillage, natural stresses, and qualityof agronomic management, and that the periodic remoteobservations of crop canopies can provide valuable inputand feedback to agrometeorological models.

88 IEEE TRANSACTIONS ON GEOSCIENCE AKD REMOTE SENSING. VOL. GE-24. NO. L JANUARY 19R6

inputs to the agrometeorological models themselves canalert or warn that an episodic weather event may have oc-curred and branching within the model can change modelcomputations to address the particular situation. Like-wise, the plant phenological stages predicted by the agro-meteorological model's plant development subroutine in-dicate whether or not the crop is at a susceptible stage fora given stress. The symptoms of stresses as diverse asdrought, nematodes, and restricted rooting by tight soilswill be revealed by the spectral data through effects oncanopy size achieved. In short, agronomic and physiolog-ical information aid in determining. for a given situation,whether reproductive performance was consistent withvegetative development. However, the work of Aase andSiddoway 12J and Barnctt and Thompson [41 indicate that,in commercial fields over large areas, yields are closelyassociated with the LAI of the canopies achieved.

IV. SUMMARY

In summary, the spectral indicators of crop develop-ment and canopy size available through vegetation indiceshelp determine how the crops are actually doing comparedwith how weather- and soil-property-driven agrometeoro-logical models predict they are doing. The individual spec-tral components terms provide information on leaf areaindex, biomass, and light absorption that can be llsedeither as input or as feedback to the models. Finally, therelations expressed by (I) and (2) agree, in general, withcanopy characteristics as they relate to yield. Thus, theyadd credence to the information conveyed by plant canopyobservations and form a bridge between spectral obser-vations and agrometeorological models. The joint use ofspectral observations and agrometeorological modelsshould improve crop condition and yield estimation ef-forts.

ACKNOWLEDGME'\T

The authors thank 1. Cuellar, W. Swanson, and M. Shi-bayama for assistance in making measurements, data rc-duction, and figure preparation.

REFERENCES

II] J. K. Aase. "Relationship between leaf area and dry matter in winterwheat." AKron . .I.. vol. 70. pp. 563-565. 1978.

12] J. K. Aase. and F. H. Siddoway. "Spring wheat yield estimates fromspectral reflectance measurements." /I-.'EI-,·Trans. Geosci. Rell/oteSensing. vol. GE-I'I. no. 2. pp. 77-84. 1982.

13] J. F. Arkin. C. L. Wicgand. and H. Huddleston. "The future role ofa crop model in large area yield estimating." in Proc. Crop WeatherModeling Workshop. pp. 88-103. 1979.

14] T. L. Barnett and D. R. Thompson. "Large arca relation of satellitespectral data to wheat yields." in Proc. SYIi/p. Machine Proc. or Re·motely Sensed Data (West Lafayette. IN). pp. 213-219. 1982.

15] C. S. T. Daughtry. K. P. Gallo. and M. E. Bauer. "Spectral estimatesof solar radiation intcrcepted by (orn eanopics." Agron . .I.. vol. 75.pp. 527-531. 1983.

(6] K. P. Gallo. C. C. Brooks. C. S. T. Daughtry. M. E. Bauer. and V.C. Vanderbilt. "Spectral estimates of intercepted solar radiation bycorn and soybean canopies." in Proc. Svmp. Maehine Proe. or Re-motelv Sensed Data (West Lafayette. IN), pp. 190-198. 1982.

(7] H. W. Gausman. L. N. Namken. M. D. Heilman. H. Walker. and F.R. Rittig. in Proe. Beltwide Colton Prod. Res. Conf (Phoenix. ALl.pp. 51-52. 1979.

18] 1. L. Hatfield, G. Asrar, and E. T. Kanemasu, "Intercepted photo-synthetically active radiation in wheat canopies estimated by spectralreflectance." Rell/ote Sen.1'. oj' f-.'11\·iron., vol. 14. pp. 65-75. 1984.

(91 L. E. Hipps. G. Asrar. and E. T. Kanemasu, "Assessing the inter-ception of photosynthetically active radiation in wintcr wheat." Agric.Meteorol .• vol. 28. pp. 253-259. 1983.

1101 T. Hodges. "Sccond gcneration crop yield models review," NASAJohnson Space Center. Houston, TX. AgRISTARS Rep. YM-12-04306.JSC-18245. 1982.

I I liD. S. Kimes. "Modcling the directional retieetance from completehomogeneous vegetation canopies with various leaf-oricntation distri-butions.1. Opt. Soe. AII/er. A. vol. I. pp. 725-738, 1984.

112] E. C. Large. "Growth stages in cereals-Illustrations of the Feekesscale." Pia II! Patholo .• vol. 3, pp. 128-129. 1954.

1131 S. J. Maas and G. F. Arkin. "User's guide to SORGF: A dynamicgrain sorghum growth model with feedback capacity." Research Cen-ter Program and Model Documentation no. 78-1 (Blackland Res. Cen-ter at Temple). TAES. Collcge Station. TX. 1978.

114] J. M. Norman and J. J. Welles, "Radiative transfer in an array ofcanopies." Agron. 1., vol. 75. pp. 481-488. 1983.

1151 C. R. Perry Jr. and C. L. Wiegand. "Modeling photosyntheticallyactive radiation intercepted or absorbcd by crops." Rell/ole SellsingEnviron. submitted j(lr publication.

1161 A. J. Richardson, "Measurement of rellcctance factors under dailyand intermittent irradianee variations," Appl. Opt .. vol. 20. pp. 3336-3340. 1'I8\.

117] A. J. Richardson and C. L. Wicgand, "Distinguishing vegetation fromsoil background infllflllation." PllOtogl'llll/l//etl'. Fng .. vol. 43. pp.1541- 1552. 1977.

1181 A. J. Richardson and C. L. Wiegand. "Diurnal· seasonal light intcr-ception. leaf area index. and vegetation index interrelations in a wheatcanopy." Phtotogl'llmmetr. t.'IIg.. submitted Illl' publication.

1191 A. J. Richardson. C. L. Wiegand. G. F. Arkin. P. R. Nixon. and A.H. Gerbermann, "Remotely-sensed spectral indicators of sorghum de-velopment and their use in growth modcling." Agrie. Meteor .. vol.26, pp. 11-23. 1982.

1201 M. Shibayama. C. L. Wiegand, and A. J. Richardson, "Diurnal pat-terns of bidirectional vegetation indiccs." /111 . .I. Remote SellSillg.. tobe publ ished.

121] C. J. Tucker. W. H. Jones. W. A. Kley, and G. J. Sundstorm. "Athree-band hand-held radiometer I(lr field use." Science. vol. 21 I. pp.281-283. 198\.

1221 C. L. Wiegand. "Candidate spectral inputs to agrometeorological cropgrowth/yield models." in Proe. 2nd /nl. Collo'l. Spectral Signalareso( Objects in Remole Sensillg (Monfavet. France). pub. 23. pp. 865-872. 1983.

1231 c. L. Wiegand. A. J. Richardson. and E. T. Kanemasu. "Leaf areaindex estimates for wheat from Landsat and their implications forevapotranspiration and crop modeling." Agron . .I.. no. 7\. pp. 336-342. 1979.

1241 C. L. Wiegand and A. J. Richardson. "Leaf area. light interccption.and yield estimates from spcctral componcnts analysis." Agron . .I.,76. pp. 543-548. 1984.

125] C. L. Wiegand. "The value of direct observations of crop canopiesIll[ indicating growth conditions and yield." in Proc. /8th lilt. Svmp.Remote Sensing Environ. (Paris. Francc. Oct. 1-5. 1984). vol. 2. pp.1551-1560.

(26] A. C. York, "Response of cotton to mepiquat chloride with varyingN rates and plant populations." Agron. 1.. 75. pp. 667-672. 1983.

*

Craig L, Wiegand received the BS. and M.S. de-grees in agronomy from Texas A&M Universityand the Ph.D degree in soil physics from UtahState Univcrsity.

He is a Research Scientist in the Remote Sens-ing Research Unit. Agricultural Research Service.U.S. Department of Agriculture. Weslaco. TX. Hehas been involved in agricultural applications ofremotc-scnsing rescarch since the mid-1960's andis currently emphasizing spectral inputs to cropmodels and large-area crop condition assess-ments.

WEIGAND el al.: SPECTRAL COMPONENTS ANALYSIS

Arthur J. Richardson received the B.S. degreein secondary education from Texas A&I Univer-sity, Kingsville, TX, in 1965.

He is currently a Rcsearch Physicist in the Re-mote Sensing Research Unit, Agricultural Re-search Service, U.S. Departmcnt of Agriculture,Weslaco, TX, whcre he has been involved in rc-mote-sensing research for 18 years. He has testedagricultural applications of Landsat, Skylab,HCMM, and NOAA data under NASA and Ag-RISTARS programs.

89

Paul R. Nixon received the B.S. and M.S. de-grees in agricultural engineering from Iowa StateUniversity and the M.S. degree in hydrology,Stanford University.

For seventeen years he worked in California insoil and water engineering and research of naturalground water recharge processes and use of waterby vegetation. Since 1971, his research in Texashas involved thermal data obtained by aircraft andsatellite. His current work includes use of multi-spectral video imaging for assessment and man-

agement of natural resources.

90 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. GE-24, NO. L JANUARY 1986

Development of Agrometeorological Crop ModelInputs from Remotely Sensed Information

CRAIG L WIEGAND, ARTHUR J. RICHARDSON, RAY D. JACKSON, PAUL J. PINTER, JR.,J. KRIS AASE, DARRYL E. SMIKA, LYLE F. LAUTENSCHLAGER, AND J. E. McMURTREY, III

Abstract- The goal of developing agrometeoi"Ological crop model in-puts from remotely sensed information (AgRISTARS Early Warning/Crop Condition Assessment Project Subtask 5 within the u.s. Depart-ment of Agriculture (USDA)) provided a focus and a mission for cropspectral investigations that would have been lacking otherwise. Becausethe task had never been attempted before, much effort has gone intodeveloping measurement and interpretation skill, convincing the sci-entific community of the validity and information content of the spec-tral measurements, and providing new understanding of the crop scenesviewed as affected by bidirectional, atmospheric, and soil backgroundvariations. Nonetheless, experiments conducted demonstrate that spec-tral vegetation indices (VI) a) are an excellent measure of the amountof green photosynthetically active tissue present in plant stands at anytime during the season, and b) can reliably estimate leaf area index(LAI) and intercepted photosynthetically active radiation (IPAR)-twoof the inputs needed in agrometeorological models. Progress was alsomade on using VI to quantify the effects of yield-detracting stresses oncrop canopy development. In a historical perspective, these are signif-icant accomplishments in a short time span.

Spectral observations of fields from aircraft and satellite make di-rect checks on LAI and IPAR predicted by the agrometeorologicalmodels feasible and help extend the models to large areas. However,newness of the spectral interpretations, plus continual revisions inagrometeorological models and lack of feedback capability in them, haveprevented the benefits of spectral inputs to agrometeorological modelsfrom being fully realized.

I. BACKGROUND

IN 1976, a decision was made in the AgriculturalResearch Service (ARS) of the U.S. Department of Ag-

riculture (USDA) to launch an effort to develop an agro-meteorological model for forecasting wheat (Triticum aes-tivum L) yields. At the first meetings of the scientists andadministrators to define and plan the project, there waslimited awareness and even skepticism about the possibil-ities of using remote spectral observations in crop models,

ManuscriptreceivedApril 26, 1985; revisedJune 27, 1985. This workwassupportedby the AgriculturalResearchServiceand the StatisticalRe·portingService.U.S. Departmentof Agriculture.

C. L. Wiegandand A. J. Richardsonare withthe AgriculturalResearchService,U.S. Departmentof Agriculture,Weslaco,TX 78596.

R. D. Jacksonand P. J. Pinter, Jr., are with the AgriculturalResearchServices,U.S. Departmentof Agriculture,Phoenix,AZ 85040.

J.K. Aase is with the AgriculturalResearchService, U.S. Depal1mentof Agriculture,Sidney,MT 59270.

D. E. Smikais withtheAgriculturalResearchService,U.S.Dcpartmentof Agriculture,Akron, CO 80720.

L. F. Lautenschlageris withthe StatisticalReportingService,U.S.Dc·partmelltof Agriculture,Washington,DC 20250.

J. E. McMurtrey,III, is with the AgriculturalResearchService, U.S.Departmentof Agriculture,Beltsville,MD 20705.

IEEE Log Number8406219.

except for one or two individuals who had been exposedto the Large Area Crop Inventory Experiment (LACIE)[34]. However, information such as the flow chart of Fig,1 illustrated and interrelated spectral data and model in-puts [53], [54] and provided evidence of technical feasi-bility.

Once remote observations were accepted as a legitimatepart of the effort, the Wheat Yield Project gave a sense ofmission and direction to the utilization of spectral mea-

. surements. The project also exposed additional scientiststo spectral observations. The early decision of the proj-ect's leadership to acquire and disperse handheld radi-ometers [51] and data loggers (Polycorders®) I to the proj-ect's participants, and the workshop held on their use [15]were important contributors to the experiments that havebeen conducted under the impetus of and with at least par-tial funding from the ARS Wheat Yield Project.

When the multi-agency AgRISTARS effort began in1979, the wheat modeling effort became part of the ARS'scontribution to it. A subtask with the title of this paperwas established within the Early Warning/Crop ConditionAssessment Project managed by Glennis Boatwright of theARS at Houston. The spectral research was concentratedat Weslaco, Texas; Phoenix, Arizona; and Beltsville,Maryland. Related additional experiments were con-ducted at Sidney, Montana; Mandan, North Dakota; Ak-ron and Ft. Collins, Colorado; and Bushland and Lub-bock, Texas. Personnel of the Statistical Reporting Service(SRS) of the USDA at Fort Collins, Houston, and Wash-ington participated in various studies [28], [29], [35]. Thecrop modeling effort was centered at Fort Collins, Colo-rado, and Temple, Texas, The scope of the effort was todevelop and test spectral data products for crop responseto management variables, early warning and crop condi-tion alarms and assessments, and crop growth and yieldmodel inputs.

The purpose of this paper is to overview the researchconducted within the USDA relevant to developing spec-tral inputs to agrometeorological crop models and to high-light some of the progress. Similar work conducted at Pur-due University, Kansas State University, University ofNebraska, and at the Johnson and Goddard Space FlightCenters under NASA sponsorship will generally not becovered.

'Product names arc given for informationpurposesand do not implyconsentor endorsementby the USDA.

u.S. Government work not protected by U. S. copyright

WEIGAND et al.: AGROMETEOROLOGICAL CROP MODEL INPUTS 91

SOLARRADIATION

- -- L!~clYWI!..oiDFig. 1. Information sources, inputs, and plant processes for agrometeorol-

ogical plant growth and yield models (after [59], [61]).

DAILY MAX. & MIN AIR TEMP

PRECIPITATION

SOLAR RAD.

W N

INTEGRATORPROCESSES

CANOPY TEMP

SOIL TEMP.

FERTILITY

ROOTING DEPTH

PLANT AVAIL.WATER

INPUTS

SURFACE

INFORMATIONSOURCES

ANCILLARY(SOIL MAP.TOPOGRAPHY •GROWERPRACTICES,LAND USEPATTERNS.CROP CALENDAR;ETC. )

\lEATHERDATA NET

EARTHOBSERVATIONSATELLITE

METEOROLOGICSATELLITES

II. INTRODUCTION

The use of remotely sensed information in agrometeo-rological models depends on its availability compared withtraditional data sources, and on the expertise and biasesof the individual or group applying the model(s). Herein,we define remotely sensed information as noncontact ob-servations in one or more wavelengths in the range 0.35J.tm (lower limit of visible light) through 14 J.tm (thermallyemitted electromagnetic radiation). Microwave (1-30 cmwavelength observations would be useful but are not gen-erally available. The agrometeorological crop models inmind are those that: a) use soil properties. (rooting depthand plant available water) and daily increments of weatherdata (temperature, precipitation, and insolation) as inputsto subroutines that simulate various plant processes (phen-ological or ontogenetic development, photosynthesis, res-piration, evapotranspiration, dry matter accumulation); b)are designed to describe crop behavior on a field scale; c)are capable of simulating the crop from planting to ma-turity; and, d) estimate yield of the salable plant parts.Models in this category include TAMW [33], CERES[46], and SORGF [32].

Remotely sensed information can be used in two prin-cipal ways in conjunction with an agrometeorologicalmodel. One way is to provide surrogate estimates of oneor more specific inputs that "drive" the model, e. g., leafarea index (LAl)2 or intercepted3 photosynthetically ac-

2The ratio of the area of green leaves to the ground area occupied on thewhole field basis.

'Typically, sensors sensitive to the PAR wavelength interval are used tomeasure the light incident (f) on the canopy, the light transmitted (T)through the canopy to the ground, the light reflected (R) from the plantsand soil, and the soil (R,). Intercepted PAR is defined as (f-T)/l and ab-sorbed PAR (APAR) as (f-T-R + TR,)/l. Sometimes investigators reportIPAR and sometimes APAR, but they differ by only a few percent for acanopy that fully covers the ground. They can differ more at low vegetativecover where surface wetness and organic matter and mineral content of thesoil affect albedo in the PAR wavelengths.

tive (0.4-0.7 J.tm) radiation (IPAR). The other way is toprovide independent feedback to override and reset themodel simulated canopy development or yield estimates[43], [55], [59], [60], [61]. In the first approach the spec-tral data provide an alternative way of acquiring the nec-essary inputs for the model. In the second approach, forexample, the LAI simulated by the model could be re-placed with LAI estimated by handheld, aircraft- orspacecraft-mounted sensors viewing the same field(s).Since such feedback capability is lacking in most agro-meteorological models at present, there is interest in athird way of using spectral data-as an independent directassessment of crop condition and probable yield.

The information needed for any of the above spectralapproaches is acquired by directly observing the plantcanopies. Thus, the spectral or remote sensing approachtakes advantage of the fact that the plants integrate theirsoil and aerial environments and express their develop-ment; stress response, and yield capabilities through thecanopies achieved [60], [61]. Vegetation indices [15], [17],[19], [24], [29], [35], [39], [43], [49], [57] calculatedfrom the spectral observations capture information oncanopy development and condition; respond to past andcurrent management (residual fertility, tillage, crop resi-due management, and cultural practices) and soil profiledifferences within and among fields that are not easily in-cluded in agrometeorological models; and provide a meansof quantifying canopy development in response to stresses(current nutritional level, nematodes, diseases, herbicideresidue, atmospheric pollutants, drought) [59], [61]. Thus,the use of spectral observations in conjunction with anagrometeorological model increases confidence that themodel is tracking the actual behavior of plants in individ-ual fields [62]. This confidence factor is extremely impor-tant. Crop models will not be applied for real world de-cisions unless consistently reasonable outputs can beexpected.

92IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING. VOL. GF~4. NO. I. JAJ\UARY 19Hh

III. PROGRESS UNDER THE ARS WHEAT ANDAgRISTARS PROJECTS

The experiments have dealt with a large set of issuesthat contribute directly or indirectly to use of spectral datain models by documenting relationships that exist, provid-ing new understanding of scene and atmospheric behavior,acquiring data sets for testing hypotheses and relation-ships, convincing the scientific community of the validityand information content of the spectral measurements, de-veloping interpretation skill and meaning, and providinginsights to support integration of spectral observations intocrop models. This whole spectrum of activities was nec-essary to a) establish the scientific validity of new mea-surements and concepts, b) acquire the necessary exper-tise and equipment to use the technology, and c) changetraditional or institutionalized procedures.

By 1981 the AgRISTARS effort was well underway. InOctober of that year the senior author suggested the fol-lowing as viable research objectives in a memo to col-leagues.

1) "Calibration of LAI, percent cover, and other ag-ronomic characteristics versus vegetation indices;checking their geographic generality; and, deter-mining the 'best' equation forms.

2) "Testing the above relations within and among cropspecies to determine, for example, whether the'calibrations' are the same for the temperate cer-eals, soybeans and cotton, sorghum and corn ...and pinpointing canopy 'architecture' and otherreasons for differences.

3) "Understanding the properties of vegetation'greenness' as expressed by the vegetation indices.

4) "Testing whether LAI is a necessary characterizerof canopies for light interception, or whether per-cent cover (PC) is adequate.

5) "Deveiopinguspectral measures of stress and com-paring them with traditional ones, or defining andeXplaining new spectral ones.

6) "Developing spectral surrogates of LAI, biomass,or genetic canopy coefficients for use in growth andyield models, or to reinitiate or override thesemodels.

7) "Compiling data sets to test whether we can go di-rectly from spectral measurements to interceptedlight.

8) "Testing spectral models of yield versus those fromagrometeorological and ecological-physiologicalmodels.

9) "Determining the effect of atmospheric corrections(sun angle, path radiance, haze) on the vegetationindices.

10) "Developing procedures to achieve agreement be-tween space and ground-observed vegetation andsoil indices."

Again, the list illustrates the diversity of activities thatneeded to proceed simultaneously to develop, understand,and use vegetation indices, the main vehicle for providing

spectral inputs to models. Despite this great diversity, andlack of a coordinated plan for research on such objectivessignificant progress was made.

Subject matter areas that were researched and docu-mented included:

I) Spectral-agronomic relations []], [4], [51, [10],[20], [43], [49], [52], [55].

2) Spectral-temporal and spectral-phenological rela-tions [2], [30], [371, [45], [48], 149].

3) Spectral transforms, vegetation and soil indices,their relation with canopy characteristics (LAI,green biomass, percent cover, chlorophyll content,phytomass), and interpretation techniques [11],[15], [19], [35], [37], 139], [49]. [52], [57], [59],[60]-[62] .

4) Wavelengths in addition to the Landsat wavelengthintervals (0.5-0.6,0.6-0.7,0.7-0.8, and 0.8-1.1 j.(m)and their utility and information content []5 I, [17],1301.

5) Procedures to achieve agreement between space(top of atmosphere) and ground-observed reflec-tance factors [171, []81, [40]-[42], [44].

6) Scene spectral modeling including effects of atmo-sphere, sun and view angles, and planting config-urations on observations []4], [231, 126], 127],[37], [38], [47], [48].

7) Spectra] measures of stress [16], [17], [21], [221,[36], 158].

8) Spectral estimates of yield [3], 15], [121, [13], 136],[50], [52], [60].

9) Spectral inputs or surrogates for agrometeorologi-cal models 19], [10[. 1431, 145], [55], [59], [60]-[62J.

10) Plant development scale comparison 16].

In addition, USDA researchers made their field plots avail-able to other scientists for experimental measurements[25]-[27], 131], [49].

Selected exemplary figures, tables, and equations fromthese publications illustrate the progress that has beenmade. Fig. 2 (after [43]) relates the LAI of grain sorghumon five dates to above-ground phytomass. Since the spec-tral observations are responding to the chlorophyll con-taining parts of the crop canopy [60], there is a close re-lation between LAI and above ground phytomass as longas the "stems" consist of leaf sheaths. But, when a truestem and then a head and grain develop, the latter containmost of the phytomass, and the relation between LAI andphytomass deteriorates. Whereas spectral vegetation in-dices relate less and less well to wet and dry phytomass ascrops approach maturity, the VI relate well to LAIthroughout the life cycle of the crop. In the agrometeoro-logical models, LAI is used to characterize crops for pen-etration and interception of photosynthetically active ra-diation in the photosynthesis and growth subroutines, andto partition insolation between evaporation of water fromthe soil and transpiration from the plants [55]. Thus, re-mote estimates of LAI can be direct inputs to the models.

__ ~ __ "_---'- __ ~ __ '....J-

r --.- 1;--- r··-~--- --....,----·-·.·--·-r·-- --- ----,-- -----'T--

WEIGAND fl a/.: AGROMETEOROLOGICAL CROP MODEL [NPUTS 93

1200"G/1'A IOJ ' 54.99(P\':~)<- c' 81".-'"

~n,--1'

BOO~a::......

~ 400~>-0:::Cl

00 B 12 16 20

1200

c"

N~ BOO..••..(J)0:::......-'-' 400i=

00 4 B 12 16 20

PVIFig. 4. Relation between p~rpendicular vegetation index (PV!) and two plant

characteristics (tillers per square meter and dry matter. kilograms perhectare) during the fall growth period of the ARS Wheat Yield Projectfields in the 1977-1978 season. Fields are located in Keith (K) Co .. NB;Grant (G) Co., OK; Washington (W) Co., CO; Finney (F) Co .. KS;Greeley (E) Co .. KS; and Jewel (1) Co .. KS.

14(.0QHQOD

A

l'. A

•~ .•

6'):~ BOCO lOOCO

Biomass (kCJ ha-')

Fig. 2. Plot of sorghum leaf area index measurements versus above groundbiomass measurements for five Landsat overpass dates during 1976 grow-ing season in Bell County. TX. The correlation coetlicients by date were:0, May 3, 1976,0.996**; •. May 21. 1976,0.990**; [ I, June 8. 1976,0.946**; _, June 26, 1976, -0.247; !'o. August 1. 1976,0.459 (**, sta-tistically significant at the 0.01 probability level) (after 1431l.

O. 1

O. 00 100 200 300 400 500 600 700 800

LEAF BIOMASS. kg/heFig. 3. Normalized dif!'crence vegetation index (ND7) versus leaf biomass

f(lr six spring wheat stand densities, expressed as a percentage of normalseeding rates (after 141l.

accomplishment of the AgRISTARS effort should not beunderrated. It was a necessary step in getting a new tech-nology accepted; without the acceptance there would beflO use. It is an accomplishment shared jointly by theNASA-funded and the USDA-funded research highlightedherein.

The value of the vegetation indices is that they condenseobservations in two or more wavelengths to a single num-ber that relates well to the amount of photosyntheticallyactive tissue [60], [62]. Thus the VI relate well to LAI,percent cover by green vegetation, leaf weight, plant pop-ulation, green or dry biomass of nonstemmy vegetation,chlorophyll content per unit area, and consequently, to thecrop's light interception capacity. Lautenschlager andPerry [29] and Perry and Lautenschlager [35] showed thata number of the vegetation indices are mathematicallyequivalent. Jackson et al. (17) described their sensitivityto atmospheric effects.

The relation between the perpendicular vegetation index(PVI) derived from Landsat-2 observations adjusted forsolar zenith angle and atmospheric haze and two ground-truthed plant parameters, tillers per square meter and drymatter (kilograms per hectare) collected in fields of theARS Wheat Project is shown in Fig. 4. The data are forthe fall growth period preceding winter dormancy. Thereare two observation dates for five of the fields and oneobservation date for the Jewel Co., Kansas, field.

The coefficient of determination between PVI and till-ers per square meter for the sites except Washington Co.,Colorado, is 0.96 whereas it is 0.82 between PVI and drymatter including the Washington Co., Colorado, field. Theslopes of the regression equations indicate there are 67

:,*.~

Y=0.24+7.23E-4XR' = 0.91

O. 8* lOO~

O. 7 x 80~/I 60~

b 0.60 40~+ 20~

Z D lO~

• O. 5 "*xw00.4z

00

Many researchers have verified that vegetation in-dices-differences, ratios, and linear transformations ofspectral reflectance or radiance observations [24], [39],[49], [15], [17], [19], [35], [48]-relate to crop canopy"greenness" [1], [4], [10], [17], [30], [37], [39], [43],[49], [52], [55], [60]. As an example of one vegetationindex, Fig. 3 (after [3]) shows that leaf biomass is relatedto the normalized difference (ND) defined by (MSS7 -MSS5)/(MSS7 + MSS5) where MSS5 and MSS7 denotethe reflectance in visible red (0.6-0.7 1J.m) and reflectiveinfrared (0.8-1.1 1J.m) wavelengths. These two wavelengthscorrespond to Landsat multispectral scanner (MSS) bands5 and 7, respectively. Data such as those in Fig. 3 hayebeen influential in convincing the scientific and user com-munities that spectral observations can be used to esti-mate important agronomic characteristics of crops. This

,94 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. GF-24, NO. I, JANUARY 1986

TABLE ICOEFFICIENTS OE DETERMINATlO'l BETWEEN 9 SPECTRAL MEASURES AND In

LAI, GRAIN YIELD, AND [PAR (PART A) A'ID AMONG THE 3 DEPENDFI\T

V ARIABLES (PART B)

(AfIer Wiegand and Richardson 160].)~_._.,---_.~._--~

2-r ---------

170

----2A---3A----4A

'••......•--'.~

APR MAYMAR

DAYS AFTER PLANTING

FEB

POST SENESCENCE BASELINE

o70

1.20

Fig. 5. The normalized difference (NO) versus time after planting for fourProdura wheat fields with widely varying yields and a hare soil plot. Theshaded portion under the curve for plot 2,4 is a graphic representation ofthe integration technique deserihed in the text (after 1361).

lPAR%

Yield_1kg ha

1. 000.773** 1.000

.676** .526**

.665** .524**

.661** .487**

.617** .504**

.670** .565**(- ) .447** (- ) .479**(- ) .521** (- ) .548**

.018 .005

.387** .203*

1n LA!

1. 000.827**.962**

.601**

.570**

.551**

.573**

.61 g**(- )~/.446**(-) .543**

.000

.284**

PVIGRGRwRVrNOMSS4MSS5MSS6MSS7

(B)1n LAIYieldr PAR

(A)Veg. indexor MSS band

** Significant at P=.Ol.* Significant at P=.05.

al Negative signs designate variable pairs that wereinversely related.

tillers/m2 per unit PVI and 55 kg/ha dry matter per unitPVI. Such relations between spectral and agronomic datamay prove useful for monitoring crop growth. The factthat the relation between tillers per square meter and drymatter, r2 = 0.83 (not shown) was no better than betweenthe top of the atmosphere Landsat observations and theseparameters individually indicates that the spectral sam-ples represented these fields as well as the plant samplesdid.

Ability to estimate tiller population spectrally is usefulfor establishing the plant population needed as initial inputto the models. (For a short time after emergence only pri-mary tillers exist, so the tiller population is the plant pop-ulation.) Also, the number of tillers estimated soon afterspring greenup compared with the number prior to winterdormancy indicates the number that survived the winter.The tiller estimates may also be of value in checking onthe number estimated by the agrometeorological modelused; number of tillers has not been easy to mimic accu-rately in agrometeorological models.

Wiegand and Richardson [60] summarized the coefli-cients of determination among nine spectral measures (fiveVI and the four MSS bands) and LAI, grain yield, andintercepted photosynthetically active radiation (lPAR) forgrain sorghum during the grain filling stage (Table I, partA). The coefficients of determination among the depen-dent variables LAI, YIELD, and IPAR are also presented(part B). The vegetation indices as they appear in the ta-ble, are the perpendicular vegetation index (PVI), thegreenness (GR) using universal coefficients, a greennessderived using local (Weslaco) scenes (GRw), the ratio veg-etation index (RVI), and the normalized difference (ND).The vegetation indices are superior to the individual Land-

sat MSS bands, and for the data set presented, ND andPVI related more closely to LAI and grain yield than didthe other vegetation indices. Coeflicients of determinationbetween yield and LAI (0.827) and between yield andIPAR (0.773) illustrate the predictability of yield throughspectral observations of crop canopies.

On most of the Great Plains, water deficits and otherconstraints usually prevent rain fed wheat from achievinga canopy dense enough to fully intercept the light. Butsince seeding rates and management practices are tunedto location specific climate and soil constraints, the har-vest index of wheat is remarkably constant even on thewestern Great Plains [3]. Because high yields cannot beachieved unless the crop canopy development is suflicientto intercept most of the incident insolation during the re-productive phase, the spectral measurements frequentlycorrelate well with yield [60]-[62]. For example, Tuckeret al. [50] reported that there was a five-week period, fromstem elongation through anthesis, over which the ND ex-plained approximately 64 percent of the grain yield vari-ation of wheat. Aase and Siddoway [3J reported that thehighest correlations between spectral indices and yield forwheat were obtained from stem elongation through wateryripeness of the grain. The reason the relations are bestthrough early grain filling is that the rgreenlleaf area in-dex reaches a maximum at about boot stage and declinesthroughout grain filling. Consequently. the later in grainfilling the spectral observations are made, the more thephotosynthetically inactive tissue dominates the observa-tions and the relationship degrades.

Pinter et at. [361 used a somewhat different approach(see Fig. 5). They summed the normalized differencesdaily for the period from heading to full senescence forall ND above the base value for harvest-ready (fully se-nescent) crops of wheat and barley. Thus, they took intoaccount not only the greenness of the canopy but also itspersistence. For Produra wheat whose canopy develop-ment had been affected by timing and amount of irrigationwater applied, the summed ND accounted for 88 percentof the yield variation. However, because the duration of

WEIGAND et al.: AGROMETEOROLOGICAL CROP MODEL INPUTS 95

al VI = spectral vegetation indices GR, PVI, "lD, etc. {see text)

TABLE IIREMOTELY SENSED INPUTS OR FEEDBACK TO AGROMETEOROLOGICAL MODELS

GROUPED BY MODEL SUBROUTINES

(After Wiegand [59], 1611.)

grain filling in temperate cereals, including wheat, is tem-perature dependent [56], any method analogous to leafarea duration cannot hold across environments. [8].

Temporal spectral measurements, such as those shownin Fig. 5, are valuable for following the pattern of canopydevelopment. For example, a leveling off in vegetation in-dex during the period of normal, rapid development of thecanopy may well correspond to a gradual depletion of soilwater, especially if a rapid rise in the VI is observed fol-lowing a known rainfall event. The vegetation index be-havior would correspond to a decrease in growth (produc-tion of leaf and canopy) during a water stress period and"boom" growth upon relief by the rain. For wheat andother temperate cereals, the decline in VI following an-thesis can be quantified into a senescence rate (VI day -])that can be related to agronomic and environmental con-ditions. Idso et at. [13] have even proposed that yields beestimated from senescence rates.

Wiegand and Richardson [60], [62] have proposedequations that interrelate the information conveyed byplant canopies about their development (or restraint fromdevelopment by stresses), light interception capability, andyield performance. The equations are

Mode 1 Sub rout ines

Growth or dry rratteraccur1Ul at ion

Photosynthes is

E vapot r an s p irat ion

Pheno logy

Stress

Yi eld

Remotely Sensed Input or Check

VI a/ --spectral surrogate ofgreen biomass

--s;Jectral profilehj--growth rate

VI--spectral surrogate of LAI for' lightabsorption estimate

Spectral estimates of IPI'\R

8R or SLIC/--albedo, surface wetness--ground cover for partitioll-

ing evaporation andtranspiration

Tc_Tad/-_as related to ratio ofactual to potential evapotrans-pirat ion, E/Ep

Spectral profile--e!1lergence orgreen-up date, mdX imumgreenness date

TCujn lieu of air temperatureto pace ontogenet i C even t 5

VI --Canopy fI greennes 5" and magn itude',Is. norllal; senescence l"ate

Tc- Ta--stress orin crop watel"( 1-E/Ep)

VI--near maximum canopy develoonent orearly in gl"ain fi~ling; spectl"alprofile integrals

In LAIVI

In LAIVI

IPARX

In LAI

Yieldx

In LAI

IPARVI

YieldVI

(1)

(2)

bl Spectral profile = vegetation index ',Is. time (see fig. 5. e.g.)

cl BR. SLI = brightness and the soil line index, spectral indicesdominated by soil background. (see Kauth atid Thomas [24]; 'vJiegand andRichardson, [57]).

dl Tc is canopy temperature; Ta is ail" tempel"ature

where VI denotes anyone of several spectral vegetationindices available, IPAR is intercepted photosyntheticallyactive radiation, and yield is grain yield.

Essentially, the integral VI are estimates of integral in-tercepted solar radiation which Daughtry et at. [7] andHatfield et at. [9] have shown can be estimated spectrally.Since the IPAR versus In LAI relations, available in theliterature and already in use in the agrometeorologicalmodels, can be transferred directly to (1) it becomes pos-sible to estimate IPAR remotely. This means in effect thatIPAR generated by the models can be checked by directspectral observations. Where the relation between LAI andVI is known from previous studies, such as it is for wheat,the VI's can also serve to check on the model's estimatesof LA!.

From historical Landsat or the currently availableNOAA meteorological satellite data, the relation betweenyield and VI can be established on field (Landsat) orcounty or crop reporting district synoptic scales (NOAA)from the VI observations those sensors provide and theyield data reported annually by the Statistical ReportingService. Wiegand et at. [58] and Wiegand [59] reportedsuch a relation for grain sorghum (Sorghum bicotor L.,Moench) in South Texas, established during grain fillingof the crop. By definition the difference between the spec-tral estimate for the current year and the long term aver-age is the production deviation from the average. Suchinformation when available in advance is useful in prepar-ing to harvest, transport, store, and market the crop.

The possibility of quantifying stress effects on yieldsthrough the canopy manifestations is an exciting one. Al-though the literature on crop stresses is voluminous, waysto relate stresses meaningfully to yields have been lacking[58]. Spectral observations to quantify stresses and relatethem to yield merit further emphasis.

Table II summarizes additional opportunities to aug-ment agrometeorological models with remote spectral ob-servations. The table is organized by the subroutines (pho-tosynthesis, growth or dry matter accumulation,evapotranspiration, phenology, stress, and yield) usuallyfound in the growth/yield models. A number of the pos-sibilities are hypothetical in that there is no known test inthe literature, although tests are technically feasible. Oth-ers depend, for acceptance, on the outcome of tests of therelations expressed by (1) and (2). Still others depend onthe availability of suitable data sets.

A point worth making is that there is no past experienceon using and incorporating remotely sensed observationsinto crop growth/yield models because such observationshave not been previously available, their usefulness hadnot been demonstrated, or operational products were notproduced. For example, NOAA can provide surface tem-perature (canopy temperature when the canopies are welldeveloped) and is developing precipitation estimates fromthe Advanced Very High Resolution Radiometer (AVHRR)aboard the operational meteorological satellites [63]. Thusthe models and operational products wi II evol ve graduallywith experience.

96 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. GE-24, NO. I, JANUARY 1986

An important aspect of any successful effort will be databank and data base management. Current research ongeographic information systems will be vital to successfuloperational application of crop models, As shown in Fig,1 there are myriad sources of relevant information thatcould be acquired, archived, merged and processed to ex-tract that needcd to exccute models, With current pres-sures on food, fuel, fiber, and forage vegctation resources,the scientists involved are working on projects with globalconsequences, In general, we feel that many of the can-didate spectral inputs for Table II are now ready for testingand adaptation for incorporation into crop growth/yieldmodels,

High priority needs to be placed on producing algo-rithms for resetting and continuing the execution of agro-meteorological models when remotely sensed canopy ob-servations are used as feedback to the models and ondevelopment of workable geograhic information systems,

IV. SUMMARY

The progress made in developing and using spectral in-formation promises to augment and enhance agrometeo-rological models by providing direct evidence of canopycondition that can be interpreted in terms of plant popu-lation, LAI, or IPAR for direct use in the models, or asfeedback to them, Thus, use of spectral observations inconjunction with agrometeorological models increasesconfidence that the correct deductions are being made, Inseveral instances the spectral data appear to he a mean-ingful way to quantify strcsses~through their effects onthe canopies the crops achieve. As a consequence of theconstancy of the harvest index of wheat and cnvironmentalconstraints on the canopies achieved over most of theWheat Belt, grain yield of wheat relates well to spectralvegetation indices during the period latc stcm extensionto early grain filling. Collectively these findings help de-termine whether or not agrometeorological model esti-mates of plant canopy characteristics, that in turn, affectthe model's photosynthesis, evapotranspiration, stress re-sponse, and yield subroutines, are being correctly pre-dicted for particular production areas, The understandingof plant canopies represented by these advances have beenincorporated into the subjective operational yield predic-tions of the Foreign Agricultural Service (FAS) of theUSDA, while agrometeorological models that would LIsespectral inputs are still being revised.

REFERENCES

II] J. K. Aase. "Relationship between leaf area and dry matter in winterwheat," Argon. J., vol. 70, pp. 56:1-565, 1978.

12] J. K. Aase and F. H. Siddoway, "Determining winter wheat standdensities using spectral reflectance measurements," Argon . ./ .. vol.72, pp. 149-152, 1980.

1:1] J. K. Aase and F. H. Siddoway, "Spring wheat yield estimates fromspectral reflectance measurements," IFEE liw1.I. Geosci. and Remo{eSensing., vol. GE-19, pp. 78-84, 1981.

[4] J. K. Aase and F. H. Siddoway, "Assessing winter wheat dry matterproduction via spectral reflectance measurements," Remo/e SensingEnviron., vol. II, pp. 267-277,1981.

[5] J. K. Aase, F. H. Siddoway, and J. P. Millard, "Spring wheat-leafphytomass and yield eslimates from airborne scanners and hand-held

radiometer measurements," 1m. 1. Remo/e SensinJ;, vol. 5, pp. 771-781, 1984.

16J A. Bauer. D. Smika, and A. Black, "Correlation of live wheat growthstage scales used in the Great Plains," Pub. AAT-NC-7, ARS-USDA.Peoria, IL, 198:1.

171 c. S. T. Daughtry, K. P. Gallo, and M. E. Bauer. "Spectral estimatesof solar radiation intercepted by corn canopies," Agron. J., vol. 75,pp. 527-531, 198:1.

[8] L. C. Evans and 1. F. Wardlaw, "Aspects of the comparative physi-ology of grain yield in cereals," Adv. Agron .• vol. 28, pp. 301-:150,1976.

19] 1. L. Hatfield, G. Asrar, and E. 1'. Kanemasu, "Intercepted photo-synthetically active radiation in wheat canopies estimated by spectralrc11cctance," Remo/e Sensing Environ" vol. 14, pp. 65-75, 1984.

1101 J. L. Hatfield, E. T. Kanemasu, G. Asrar, R. D. Jackson. P. J. Pinter,Jr., R. G. Reginato, and S. B. Idso, "Leaf area estimates from spec-tral measurements over various planting dates of wheat." In/. 1. Re-mOIl' Sensing, vol. 6, pp. 167-175. 1985.

[11] A. R. Huete, D. F. Post, and R. D. Jackson, "Soil spectral effects on4-space vegetation discrimination." Rell/ol" Sensing Em'iron., vol.15, pp. 155-165, 198:1.

1121 S. B. Idso, P. J. Pinter, Jr., J. L. Hatlield. R. D. Jackson. and R. J.Reginato, "A remote sensing model t<lf the prediction of wheat yieldprior to harvest," 1. Theor. Bio/ .. vol. 77, pp. 217-228, 1979.

1/:11 S. B. lelso, P. J. Pinter, Jr.. R. D. Jackson, and R. 1. Reginato, "Es-timation of grain yields by remote sensing of senescence rates," Re-mo/e Sensing Fnl'iron" vol. 9, pp. 87-91, 1980.

1141 R. D. Jackson, R. J. Reginato, P. J. Pinter, Jr., and S. B. Idso, "Plantcanopy in!(lflllation extraction from composite scene rellectance of rowcrops," App/. 0f'1., vol. 18. pp. 3775-:1782, 1979.

1151 R. D. Jackson, P. J. Pinter, Jr., J. R. Reginahl, and S. B. Ie!so, "Hand-held Radiometry," USDA-SEA, Agricultural Reviews and Manuals,ARM-W-19, 1980.

1161 R. D. Jackson, "Canopy temperature and water stress," in Advallcesilllrrigalion, D. Hillel. Ed. New York: Academic, 1982. pp. 43-85.

1171 R. D. Jackson, P. N. Slater, and P. J. Pinter, Jr .. "Discrimination ofgro\vth and water stress in wheat by various vegetation indices througha e1ear and a turbid atmosphere," Rell1o/e Sellsillg Fnviroll., vol. n.pp. 187-208, 1983

1181 R. D. Jackson, P. N. Slater, and P. J. Pinier, Jr., "Adjusting the tas-selled-cap brightness and greenness factors for atmospheric patb ra-diance and absorption on a pixel by pixel basis." 11//. 1. Rell10te Sl'IlS-ing., vol. 4. pp. 3n-.n:l, 1983.

119J R. D. Jackson, "Sp~ctral indices in N-Space," Remole Semillg EIl-l'iroll .. vol. 13, pp. 409-421. 198:1.

120] R. D. Jackson. "Remote sensing of vegetation characteristics J(lr farmmangagement." in Proc. Soc. PhOlo-Opl. II/slru. /:'I/g., vol. 475, pp.81-96, 1984.

121] R. D. Jackson and B. F. Robinson, "Field evaluation of the temper-ature stability of a multispectral radiomeler," ReII/O/I' Sensil/g FI/I"i-roll .. vol. 17, pp. Im-108, 1985.

1221 R. D. Jackson and C. E. Ezra. "Spectral response of cotton to sud-denly induce water stress," 1111. 1. Rell1o/e Sensil/g, vol. 6. pp. 177-185, 1985.

Inl R. D. Jackson, P. J. Pinter, Jr., and R. J. Reginato, "Net Radiationcalculaled from remote multispectral and ground station meteorolog-ical data," Agricu/. and rim's/ Me/eor. vol. 35, pp. 153-164. 1985.

1241 R.1. Kauth, G. S. Thomas "The tasseled cap-A graphic descriptionof the spectral temporal developmenl of agricultural crops as seen byLANDSAT," in Proc. SViI/f'. Machil/e Prof". Rell1o/e/v SeilSI'd Data,pp. 41-49, 1976.

{251 D. S. Kimes and J. A. Kirchner, "Diurnal variations of vegetationcanopy structure," 1m. 1. RClI1o/e Sell.\"ing, vol. 4, pp. 257-271. 1983.

1261 D. S. Kimes, W. W. Newcomb, J. B. Schutt, P. J. Pinter, Jr., andR. D. Jackson, "Directional reHectance factor distributions for a cot-ton row crop," 1111. 1. Rell1o/" Sel/sillg, vol. 5, pp. 26:1-277, 1984.

1271 J. A. Kirchner, D. S. Kimes, and J. E. McMultrey, ilL "Variation ofdirectional rellectance factors with structural changes of a developingalfalfa canopy," App/. 0f'1 .. vol. 21, pp. 3766-:1771,1982.

1281 G. A. Larsen, "Progress report on the evaluation of plant growthmodels," Stan' Rep. AGESS810716, Statistical Rep. Ser., USDA.Washington, DC, 1981.

1291 L. F. Lautenschlager and C. R. Perry. Jr., "An empirical. grapbical,and analytic study of the relationship between vegetation indices,"AgRISTARS Rep. EW-JI-04150, JSC-17424, NASA Johnson SpaceCenter, Houston, TX.

1:10] R. W. Leamer, J. R. Noriega, and C. L. Wiegand, "Seasonal changes

WEIGAND et a/.: AGROMETEOROLOGICAL CROP MODEL INPUTS 97

*

*

*Arthur J, Richardson received the B.S. degreein secondary education from Texas A&I Univer-sity, Kingsville, TX, in 1965.

He is currently a Research Physicist in the Re-mote Sensing Research Unit, Agricultural Re-search Service, U.S. Department of Agriculture,Weslaco, TX, where he has been involved in re-mote-sensing research for 18 years. He has testedagricultural applications of Landsat, Skylab,HCMM, and NOAA data under NASA and Ag-RISTARS programs.

Ray D. Jackson is a Research Physicist in the Soil-Plant-Atmosphere Sys-tems Research Unit, U.S. Water Conservation Laboratory, ARS-USDA,Phoenix, AZ. He has been involved in developing remote-sensing tech-niques to measure evapotranspiration and to quantify water and nutrientstresses in crops.

Paul J. Pinter, Jr. is a Biologist in the Soil-Plant-Atmosphere SystemsResearch Unit, U.S. Water Conservation Laboratory, ARS-USDA, Phoe-nix, AZ. He is currently studying the diurnal effects of water stress andcanopy architecture on the reflectance factors of crops.

*Craig L. Wiegand received the B.S. and M.S. de-grees in agronomy from Texas A&M Universityand the Ph.D. degree in soil physics from UtahState University.

He is a Research Scientist in the Remote Sens-ing Research Unit, Agricultural Research Service,U.S. Department of Agriculture, Weslaco, TX. Hehas been involved in agricultural applications ofremote-sensing research since the mid-1960's andis currently emphasizing spectral inputs to cropmodels and large-area crop condition assess-ments.

Workshop (Dept. Agric. Eng., Clemson University, Clemson, SCJ, p.15, 1977.

155] C. L. Wiegand, A. J. Richardson, and E. T. Kanemasu, "Leaf areaindex estimates for wheat from LANDSAT and their implications forevapotranspiration and crop modeling," Agron. J., vol. 71, pp. 336~342, 1979.

[56] C. L. Wiegand and A. J. Cuellar, "Duration of grain filling and kernelweight of wheat as affected by temperature," Crop Sci., vol. 21, pp.95-101, 1981.

[571 C. L. Wiegand and A. J. Richardson, "Comparisons among a newsoil index and other two- and four-dimensional vegetation indices,"in Proc. n,ch. Papers ACSM-ASP Convelltion (Amer. Soc. Photo-gram., Falls Church, VA), pp. 210-227, 1982.

158] C. L. Wiegand, P. R. Nixon, and R. D. Jackson, "Drought detectionand quantification by reflectance and thermal responses," Agric. WaterMgt., vol. 7, pp. 303-321, 1983.

159] C. L. Wiegand, "Candidate spectral inputs to agrometeorological cropgrowth/yield models," in Proc. 2nd Int. Colloq. Spectral Signaturesof Objects in Remote Sens. (Montfavet, France), INRA Pub. 23, pp.865~872, 1983.

[60] C. L. Wiegand and A. J. Richardson, "Leaf area, light interception,and yield estimates from spectral components analysis," Agron. J.,vol. 76, pp. 543-548, 1984.

161] C. L. Wiegand, "The value of direct observations of crop canopiesfor indicating growth conditions and yield," in Proc. 18th Int. Symp.Remote Sensing Environ. (Environ. Res. Inst. of Michigan, Ann Ar-bor, MI), pp. l551-1560, 1984.

[62] C. L. Wiegand, A. J. Richardson, and P. R. Nixon, "Spectral com-ponents analysis: A bridge between spectral observations and agro-meteorological crop models," IEEE Trans. Geosci., and Remote Sens-ing, this issue, pp. 83~89.

163] H. W. Yates, J. D. Tarpley, S. R. Schneider, D. F. McGinnis, andR. A. Scofield, "The role of meteorological satellites in agriculturalremote sensing," Remote Sensing Eviron., vol. 14, pp. 2l9-233, 1984.

in reflectance of two wheat cultivars," Agron. 1., vol. 70, pp. ll3~l18, 1978.

131] E. W. LeMaster, J. E. Chance and C. L. Wiegand, "A seasonal ver-ification of the Suits speetral reflectance model for wheat," Photo-grammetr. Eng., vol. 46, pp. 107~1l4, 1984.

[32] S. J. Maas and G. F. Arkin, "User's Guide to SORFG: A dynamicgrain sorghum growth model with feedback capacity," Research Cen-ter Program and Documentation No. 78-1 (Blackland Res. Center atTcmple), TAES, College Station, TX, 1978.

133] S. J. Maas and G. F. Arkin, "TAMW: A wheat growth and develop-ment simulation model," Res. Center Program and DocumentationNo. 80-3 (Blackland Res. Center at Temple), TAEX, College Station,TX, 1980.

[34] R. B. MacDonald and F. G. Hall, "Global crop forecasting," Sci-ence, vol. 208, pp. 670~679, 1980.

135] C. R. Perry and L. F. Lautenschlager, "Functional equivalence ofspectral vegetation indices," Remote Sensing Environ., vol. 14, pp.169-182, 1984.

[36] P. J. Pinter, Jr., R. D. Jackson, S. B. Idso, and R. J. Reginato, "Mul-tidate spectral reflectance as predictors of yield in water stressed wheatand barley," In/. J. Remote Sensing, vol. 2, pp. 43-48, 1981.

[37] P. J. Pinter, Jr., R. D. Jackson, S. B. Idso, and R. J. Reginato, "Diur-nal patterns of wheat spectral reflectances," IEEE Trans. Geosci. Re-mote Sensing, vol. GE-21, pp. 156-163, 1983.

[38] P. J. Pinter, Jr., R. D. Jackson, C. E. Ezra, and H. W. Gausman,"Sun angle and canopy architecture effects on the spectral reflectanccof six wheat cultivars," Int. 1. Remote Sensing, to be publishcd.

[391 A. J. Richardson and C. L. Wiegand, "Distinguishing vegetation fromsoil background information," Photogrammetr. Eng., vol. 43, pp.1541-l552, 1977.

[40] A. J. Richardson, D. E. Escobar, H. W. Gausman, and J. H. Everitt,"Comparison of Landsat-2 and field spectrometer reflectance signa-tures of South Texas rangeland plant communities," in Machine Pro-cessing of Remotely Sensed Data. New York: IEEE, 1980, pp. 88-96.

[41] A. J. Richardson, "Measurement of reflectance factors under dailyand intcrmittent irradiance variation, " Appl. Opt., vol. 20, pp. 3336~3340, 1981.

[42] A. J. Richardson, "Relating Landsat digital count values to groundreflectance tiJr optically thin atmosphcric conditions," Appl. Opt., vol.21, pp. 1457-1464, 1982.

143] A. J. Richardson, C. L. Wiegand, G. F. Arkin, P. R. Nixon, andA. H. Gcrbermann, "Remotely-sensed spectral indicators of sorghumdevelopment and their use in growth modeling," Agric. Meteor., vol.26, pp. Il-23, 1982.

[44] A. J. Richardson, "The equivalence or three techniques for estimatingground reflectance from Landsat digital count data," AgRISTARSRep. EW-U3-04407, NASA Johnson Space Center, Houston, TX,1983.

[45] A. J. Richardson, C. L. Wiegand, "Diurnal-seasonal light intercep-tion, leaf area index, and vegetation index interrelations in a wheatcanopy," unpublished.

[46] J. T. Ritchie, 1983, "CERES Wheat," in Wheat Model WorkshopNotebook (Temple, TX, Feb. 22~25, 1983).

[47] M. Shibayama and C. L. Wiegand, "View azimuth and zenith, andsolar zenith angle effects on wheat canopy reflectance," Remote Sens-ing Environ., vol. 18, pp. 91~103, 1985.

[48] M. Shibayama, C. L. Wiegand, and A. J. Richardson, "Diurnal pat-tern of bidirectional vcgetation indices for wheat canopies," Int. 1.Remote Sensing, to be published.

[49] C. J. Tucker, J. H. Elgin, Jr., J. E. McMurtrey, III, and C. J. Fan,"Monitoring corn and soybean crop development with hand-held ra-diometer spectral data," Remote Sensing Environ., vol. 8, pp. 237-248, 1979.

[50] C. J. Tucker, B. N. Holben, J. H. Elgin, Jr., and J. E. McMurtrey,III, "Relationship of spectral data to grain yield variation," Photo-grammetr. Eng., vol. 46, pp. 657-666,1980.

[51] C. J. Tucker, W. W. Jones, W. A. Kley, and G. J. Sundstrom, "Athree-band radiometer for field use," Science, vol. 221, pp. 281-283,1981.

[52] D. F. Wanjura and J. L. Hatfield, "Spectral procedures for estimatingcrop biomass," Trans. Amer. Soc. Agric. Eng., vol. 28, pp. 922-928, 1985.

[53] C. L. Wiegand, G. F. Arkin, A. H. Gerbermann, and A. J. Richard-son, "Complementary nature of spectral and physiological models ofplant growth," Agron. AI1St., p. 14, 1977.

[54] C. L. Wiegand, "Using LANDSAT LAI and biomass estimates inconjunction with plant growth models," in Abstr. Crop Modeling

98 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING. VOL. GE-24. NO. 1. JANUARY 1986

J. Kris Aase is a Soil Scientist and Research Leader for the Soil and WaterManagemcnt Research Unit. the Northern Plains Soil And Water ResearchCenter. ARS-USDA. Sidney. MT. He has been active in remote sensingsince 1977. His work has been on assessing stand. winter kill, phytomass.and yield of small grains spectrally.

*Darryl E. Smika is a Soil Scientist and Research Leader at thc CentralGreat Plains Research Station, ARS-USDA, Akron. CO. He has studiedthe effects of hot dry winds on wheat production. has used spectral inputsto assess crop conditions and growth in relation to production, and obtainedverification data for the ARS Whcat Modeling Project.

Lyle F. Lautenschlager is a Mathematical Statistician in the Remote Sens-ing Branch, Applications Section, Statistical Reporting Service (SRS).USDA, South Building, Washington, DC. Hc worked on the multiagencyLACIE and AgRISTARS Projects at the NASA Johnson Space Ccnter,Houston, TX, from 1977 to 1984. He has bcen stationed in Washington.DC, since May 1984.

*J. E. McMurtrey, Ill, is a Research Agronomist in the Field Crops Lab-oratory, Beltsville Agricultural Research Center, ARS-USDA, Beltsville.MD. He has worked cooperatively with personnel of the NASA GoddardSpace Flight Center on the seasonal agronomic and spectral characteriza-tion of corn, alfalfa, and soybcans.

Documents

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING_Presentations... · ieee transactions on geoscience and remote sensing january 1986 volume ge-24 number 1 (issn 0196-2892) a publication