109

Chapter 11.

Forest Mensuration with Remote Sensing:A Retrospective and a Vision for the Future

Randolph H. Wynne1

Abstract—Remote sensing, while occasionallyoversold, has clear potential to reduce the overallcost of traditional forest inventories. Perhaps mostimportant, some of the information needed formore intensive, rather than extensive, forestmanagement is available from remote sensing.These new information needs may justify increaseduse—and the increased cost—of remote sensing.

Organization of the United Nations 1997). Muchof this loss resulted from the conversion of forestland to nonforest uses such as agriculture, pasture,or development. Environmental regulationsand the desire to preserve native forests tomaintain biodiversity further restrict harvestingof forest products. For example, harvest of timberfrom the national forests in the United Statesdecreased from 12.0 billion board feet in 1989 to3.5 billion board feet in 1997 as the focus of theU.S. Department of Agriculture Forest Serviceshifted from timber production toward wildernesspreservation, protection of habitat for threatenedand endangered species, watershed protection andrestoration, and recreation (U.S. Department ofAgriculture, Forest Service 1998).

Industry Trends Affecting Remote SensingMany forest industry trends will affect the

future of remote sensing in forestry. Some ofthe most important are as follows:

• Smith and others (2003) note that industrialforest landowners have moved from anexclusive emphasis on supplying fiber towarda view of forest land as a biological asset thatmust be managed financially. If this is true,additional information inputs, such as remotesensing, that increase financial returns mightwell be justified.

• As noted earlier, industrial forest managementis increasing in its intensity, partly becausethere has been an effective decrease in theamount of public land available for activemanagement. It can make economic senseto spend more for information about theforest resource if the additional expendituredecreases the overall cost of inputs, particularlyin the establishment and early growth phases.

• Information available from digital remotesensing can now be combined with otherdigital geospatial information to providea complete scheduling picture from sitepreparation to harvest scheduling withinan organization.

1 Associate Professor, Virginia Polytechnic Institute and StateUniversity, Department of Forestry, College of NaturalResources, Blacksburg, VA 24061.

INTRODUCTION

Forestry Information Needs of the 21st Century:Increasing Demand and a Changing Landscape

Demand for forest products is expected toincrease rapidly during the 21st century.Population growth and economic development

that increase the per capita consumption of forestproducts drive this trend. Global population, nowapproximately 6 billion, is estimated to increaseby 900 million each decade for the next 50 years(Food and Agriculture Organization of theUnited Nations 1997). Most of this increase willoccur in developing nations. Level of economicdevelopment strongly affects the demand forforest products. Worldwide income measured asgross domestic product increased by 109 percentbetween 1970 and 1994 (Food and AgricultureOrganization of the United Nations 1997). Grossdomestic product is estimated to rise from $20trillion in 1990 to $69 trillion in 2030, with the mostdramatic increases occurring in the developingnations (World Bank 1992).

Although the demand for forest productsis increasing, global forest area is decreasing.Between 1985 and 1995, the area of the world’sforests decreased by 180 million ha, an annualloss of 18 million ha (Food and Agriculture

Sout

hern

For

est

Scie

nce:

Past

, Pre

sent

, and

Fut

ure

Prod

ucti

vity

110

Chapter OverviewRemote sensing has been actively integrated

into forest inventory and management systemsfor more than half a century. The methods usednow are still effective, but there is much potentialfor increased use of remote sensing both fortraditional inventory purposes and to providethe information necessary to increase forestproductivity. This chapter starts with aretrospective view of photo mensuration, andthis is followed by a brief discussion of lidarremote sensing, one of the more promisingnew technologies for collecting data for forestinventory and monitoring. The discussion of lidarremote sensing is followed by an example thatshows how information obtained by remotesensing could increase productivity and, thus,justify increased expenditure for information.The chapter closes with a brief discussion ofbarriers that must be overcome before remotesensing data are transformed into informationdirectly useful to foresters.

WHERE WE HAVE BEEN: PHOTOMENSURATION

Vertical aerial photographs, while used inforestry since the late 1920s (particularly inQuebec and Ontario, Canada) (Spurr 1960),

have been commonly used by foresters in theUnited States since the 1940s (e.g., Lund andothers 1997). Forestry applications of aerialphotography have been diverse, covering mostaspects of the private and public goods providedby forests. However, a primary driver for forestphotogrammetry has been forest mensuration,classically defined as “the determination ofdimensions, form, weight, growth, volume, andage of trees, individually or collectively, and ofthe dimensions of their products” (Helms 1998).Substantial effort has gone into ways of usingphotographs as a means of determining volumeby species accurately, precisely, and at the lowestpossible cost. This effort has been reasonablysuccessful, but applications of aerial cruising(e.g., Avery 1978) are becoming significantlyless common in the United States, althoughquite common elsewhere (e.g., Canada).

There are several possible reasons for thedecline of photographic mensuration. Thesecan be summarized as follows:

1. The precision of photographically derivedvolume estimates is not as high as that offield-derived volume estimates.

2. Photographic interpretation andphotogrammetry require specialized skillsthat are increasingly being supplanted byother ones, such as skills in the use ofGeographic Information Systems (GIS), inaccredited forestry programs (Sader andVermillion 2000). This is occurring despiteresults from the most recent survey of desiredentry-level competency and skill requirementswhich show that more entry-level forestrypositions require knowledge of aerial photos(68 percent) than GIS (43 percent) (Brownand Lassoie 1998).

3. In much of the United States, forest landparcel size is steadily decreasing andaccessibility increasing, thus decreasingthe economic justification for (or even thefeasibility of) photo mensuration.

4. Research in forestry remote sensing nowfocuses on actual and potential applications ofnewer airborne and spaceborne sensors, suchas radar, lidar, and high-resolution digitaloptical sensors.

5. Wide use of medium-resolution spacebornesensors such as the Landsat MultispectralScanner and Thematic Mapper that areinherently digital has led to an expectationof ever more automated approaches toinformation extraction. Most algorithms inoperational use are based almost entirely onthe use of discriminant functions to categorizethe brightness value vector from each pixel,even though this approach makes use only ofhue. Hue is just one of the nine commonlyidentified elements of image interpretation,the others being shape, size, pattern, texture,association, shadows, resolution, and site(Olson 1960).

6. The actual or perceived benefits of usingphotos for inventory may be less than thecosts incurred.

The last of these points is the most important,since we can assume that forest managers andscientists are obtaining the information neededwithout widespread use of ordinary photography.However, demands on forests are increasing, andthe information required to sustainably manageforests in the face of this demand must alsoincrease. Foresters are being asked to increaseproduction of wood and fiber on an ever-decreasing land base while maintaining theimportant supplies of public goods (viable fish

111

and wildlife populations, clean water, andrecreational opportunities) that well-managedforests have always provided. To meet thischallenge, forest managers will require newtypes of information, and remote sensing willhelp to supply these. If remote sensing fills ourrequirements, we will need to evaluate previoussuccesses and failures and work to improve thematch between information that can be objectivelyand accurately derived from remotely senseddata and the information needed for forestmanagement (Wynne and others 2000).

Aerial CruisingAerial (photo) volume tables have been

constructed for both individual trees and stands(Avery 1978). All are based on total tree heightand visible crown diameter for individual trees.For stands, volume tables use stand height andpercent crown closure at a minimum; many alsoinclude visible crown diameter classes. Thesetables use visible crown diameter as a surrogatefor stem diameter and percent crown closure asa surrogate for basal area or stem density.

Tree heights are typically measured bystereoscopic parallax methods that employa parallax bar or wedge and large-scalephotographs. Shadow lengths can be used whereterrain is level and stands are relatively open.While the accuracy of these measurements varies,on 1 inch = 20 chains photography, the averagedifference between ground- and photo-measuredtree heights (with well-trained interpreters)is typically about 1 foot (Spurr 1960).

Visible crown diameter is measured with eithera micrometer wedge or a dot-type scale. It can beargued that crown diameter is more accuratelymeasured on large-scale photographs than on theground, but measurements made on photographsare not directly comparable to those made onthe ground because in photographs (1) only thedominant overstory trees are visible, and (2) theedges of any particular crown are obscured bythe crowns of adjacent trees. For these reasons,photo-derived visible crown diameters are alwaysunderestimates of actual crown diameter. Evengiven these limitations, however, photo-measuredvisible crown diameter is often better correlatedwith actual tree and stand volume than field-measured crown diameter, because it is a measureof the tree’s functional growing space (Spurr1960). Measurement consistency varies widelywith conditions, but can be expected to be onthe order of 3 to 4 feet two times out of three on

1:12,000 photos (Paine 1981), which is whymost volume tables are based on 3- to 5-footdiameter classes.

Percent (overstory) crown cover is the mostsubjective of the three direct forest measurementsmade from aerial photographs. It can simply beocularly estimated or (more commonly) ocularlyestimated with the aid of crown density scales. Itusually is an overestimate of actual crown cover,because small canopy gaps are often not visibleand shadows are often treated as trees. Whentypical forestry photo scales are used, standarderrors do not exceed 10 percent, but the bias of anindividual interpreter commonly ranges from 5 to10 percent (Spurr 1960).

However, volume estimates derived by usingstand photo volume tables are too imprecise formany uses, as standard errors of the estimate arelikely to exceed 25 percent (Spurr 1960). Whilestandard error can be reduced by increasing thenumber of samples, stand photo volume tablesare also biased, requiring double sampling withregression using matched field- and photo-measured plots (e.g., Paine 1981). This castsdoubt on the economic feasibility of photomensuration for the smaller tracts that areincreasingly common.

To summarize, timber volume can be estimatedfrom aerial photographs. Bias exists because(1) a vertical aerial photographs image-onlyportions of the crowns of dominant overstorytrees; and (2) subjectivity, particularly in crownclosure estimation, leads to interpreter-specificbias. The latter bias also leads to unacceptablyhigh standard errors of the estimate. Substantialtraining that is increasingly hard to obtain isrequired to make accurate direct tree and standmeasurements based on aerial photographs. Allthese factors combined make the use of aerialinventory cost-effective primarily for large,relatively inaccessible areas.

StratificationAt this point one might reasonably ask why

acquisition of photography is still so routine inmany organizations charged with managing forestlands. The answer is, in part, that photos are usedfor more than just aerial timber cruising. Otheruses include forest mapping for managementplanning, stress detection, forest area estimation,and land navigation, particularly in remote andinfrequently mapped areas. These uses, however,are often secondary in comparison to the routine

Cha

pter

11.

For

est

Men

sura

tion

wit

h Re

mot

e Se

nsin

g

Sout

hern

For

est

Scie

nce:

Past

, Pre

sent

, and

Fut

ure

Prod

ucti

vity

112

use of photos to stratify timber cruises.Stratification refers to “the subdivision of apopulation into strata (subpopulations) beforesampling, each of which is more homogeneousfor the variable being measured than thepopulation as a whole” (Helms 1998).Theadvantages of stratification include (1) moreprecise estimation of the population mean(given properly constructed strata), (2) separateestimates for each subpopulation, and (3) reducedcosts (Avery 1978, Avery and Burkhart 2002). AsAvery and Burkhart note (2002), photographs arecommonly used in stratified sampling to measurearea, allocate field samples by volume classes, andplan fieldwork. For many organizations, photoacquisition can be justified by stratified samplingalone. This stratified sampling not only improvesprecision and reduces cost, but also changesthe flow of information within an organization,with the result that there is a two-way flowbetween field personnel and the organization’sinformation systems.

WHERE WE ARE GOING: THEEXAMPLE OF LIDAR

L idar, or light detecting and ranging, sensorsare the optical equivalent of radar. They usea light (laser) beam, rather than a microwave

radar beam, to obtain measurements of the speed,altitude, and range of a target (Helms 1998).Most of the current small-footprint (< 1 m) laseraltimeters can record the time (and sometimesintensity) of at least two returns, which oftencorrespond to the top of the canopy and theground. Many times, however, only one returnis recorded, and it may correspond to vegetation,the ground, or some cultural feature. Sophisticatedprocessing algorithms utilizing neighborhoodapproaches can usually identify the bulk of thenonground returns, thus making it possible tocreate a bare-earth digital elevation model (DEM)with suitable interpolation techniques. Once aDEM has been created, the first returns can beinterpolated to produce a canopy height model, arepresentation of the vertical distance from anyarbitrary point on the forest floor to the topmostpart of the canopy above that point. Over forestedareas, the canopy height model provides canopyheight at any point imaged. The canopy heightmodel differs from photogrammetrically derivedtree height in one important way—the canopyheight model is a continuous representation ofthe canopy surface, rather than the height at anyone point. It should be noted that canopy height

models can be derived photogrammetrically, butmany automated image-correlation algorithmsfunction comparatively poorly over tree canopies.

From this canopy height model, and sometimesin conjunction with coregistered optical data(Gougeon and others 2001, McCombs and others2003, Popescu and others 2004) tree, or moretypically stand, volume can be determined usingdirect measurements corresponding to those madefrom photographs, namely total height, visiblecrown diameter, and percent crown closureand or stem counts. Some of the same problemsassociated with direct measurements fromphotographs also pertain to lidar. These are asfollows: (1) lidar sensors measure the distanceonly to the crowns of overstory vegetation;(2) direct measurements from lidar tend to bebiased (e.g., Nilsson 1996); and (3) lidar data arevery expensive unless some economy of scale isrealized. However, there are two substantialbenefits. Firstly, the data seem to be veryamenable to processing by automated techniques(at least for conifers), which increases objectivityand thus precision. Secondly, when directmeasurements are used in empirical models toestimate either field-measured, e.g., total height,or derived parameters, e.g., volume or basal area,stand-level predictions are unbiased (Means andothers 2000, Naesset 2002).

Again, stand volume tables require measures ofheight, percent crown cover, and sometimes crowndiameter. Lidar-based estimates of volume requiresimilar measurements. Determination of individualtree heights using lidar data requires identifyingindividual overstory stems, usually through a localmaximum approach that presumes that the highestpoint in a local neighborhood corresponds to thetop of a tree. Although this technique is effective,errors of omission or commission can occur withimproper window size. Popescu and others (2002)used the height of each cell in the canopy heightmodel to set the size of a variable window basedon tree height, making possible the successfulprediction (R2 = 90 percent and 85 percent,respectively) of maximum height and mean heightof dominant stems (diameter at breast height> 5 inches) even on very small 0.017-ha (0.04-acre)plots of common southeastern conifer species. Asplot size increases, the need to measure individualtrees decreases, and the percentage of varianceexplained increases. For example, Means andothers (2000) used lidar-derived variables—without identifying individual trees—to explain

113

93 percent of the variance in height on 0.25-ha(0.6-acre) stands of Douglas-fir [Pseudotsugamenziesii (Mirb.) Franco].

Percent crown cover is particularly easy tocalculate using lidar data; it is simply the numberof vegetation (nonground) returns above a certainheight divided by the total number of returns.Crown diameter, however, has been a little moredifficult to determine, as it requires accurate stemidentification as well as a way of distinguishing onecrown from the adjacent one. There is substantialongoing work in this area, but Popescu (2002)determined crown diameter for each identifiedtree by (1) fitting a 4-degree polynomial with leastsquares using singular value decomposition inboth the horizontal and vertical dimensions ofthe canopy height model, (2) identifying criticalpoints for each of the two fitted functions basedon the first and second derivatives of a three-point Lagrangian interpolation, and (3) averagingthe distance between critical points on the twoperpendicular profiles. For southern pines, thistechnique explained 62 to 63 percent of thevariance in crown diameter for the dominanttrees on 0.017-ha plots (Popescu and others 2002).

Like photos, then, lidar canopy height modelscan serve as the base data from which importantvariables can be measured; namely, visible crowndiameter, percent crown closure or stem count,and total height. In addition, variables unique tocanopy height models, particularly those relatingto the distribution of heights within any one gridcell or plot perimeter, have been successfully usedas independent variables in models employed toestimate plot- or stand-level parameters (e.g.,Means and others 2000, Popescu and others 2002).Examples of this type of variable include thepercent crown cover or height at a specificheight percentile.

Volume has been successfully calculated forconifers by use of lidar-derived measures ofheight, crown closure, stem density, and/orcrown diameter, or of variables relating to thedistribution of heights within a particular grid cell.Popescu and others (2004) was able to explain > 80percent of the variance in volume on small (0.017-ha, 0.04-acre), heterogeneous southern pine plotsin Virginia’s Appomattox-Buckingham StateForest using average (per plot) crown diameterobtained by applying a lidar-derived canopy heightmodel as the only independent variable. Meansand others (2000) were able to explain 97 percentof the variance in 0.25-ha (0.6-acre) Douglas-firplots in the H.J. Andrews Experimental Forest

using the 80th and 0th percentile of height (theheight greater than the given percentage of lidarfirst returns) and the 20th percentile of crown cover(proportion of first returns below the givenpercentage of total height). Many other studieshave been similarly successful in estimatingvolume for coniferous plots or stands on the basisof lidar data.

Lidar-based forest measurements, while basedon many of the same principles as photo-basedmeasurements, such as using crown diameteras a surrogate for stem diameter, are typicallymore accurate and less biased than photo-basedmeasurements. As with aerial photographs, biasexists because laser altimeters see only portionsof the crowns of dominant overstory trees.However, increased levels of automation haveled to substantial reductions in both interpreter-specific bias and the need for specialized training.Furthermore, lidar canopy height models affordthe characterization of the whole population,rather than just a sample, of the dominantoverstory trees in a specific area of interest.However, lidar data are still quite expensive forsmall areas on a per-unit basis, so lidar-basedinventory, like photo analog inventory, is cost-effective primarily for large and/or relativelyinaccessible areas. This description hardlycharacterizes the typical southern forestlandscape, which is dominated by the holdingsof nonindustrial private forest landowners.

THE LIKELY FUTURE NEED FOR ADDITIONALREMOTELY SENSED DATA

The most commonly used standard for allremotely estimated forest measurementsis field inventory, which typically employs

remotely sensed data only for stratification. Thusin order to be widely accepted and used, remotelyderived estimates of important forest biophysicalparameters must have the same level of precisionand accuracy as field-derived measurements,and be less expensive than they are. The lack ofwidespread adoption of these new technologiesis de facto proof that this standard has not yetbeen met. As with all technological innovation,however, the cost per unit of information derivedfrom remotely sensed data will continue todecrease, thus increasing their potential use fortraditional inventory needs as a tradeoff withfield costs. Another factor to be considered is thatincreasing forest productivity will require moreremotely sensed data. Organizations will be willingto spend more for information if this expenditureresults in increased productivity.

Cha

pter

11.

For

est

Men

sura

tion

wit

h Re

mot

e Se

nsin

g

Sout

hern

For

est

Scie

nce:

Past

, Pre

sent

, and

Fut

ure

Prod

ucti

vity

114

JUSTIFYING INCREASED USE OFREMOTELY SENSED DATA: THE EXAMPLEOF INTENSIVE (AND/OR SITE-SPECIFIC)FOREST MANAGEMENT

Most of the preceding discussion has focusedon traditional assessment and inventory,an ongoing need whose importance is not

likely to diminish in the near future. However,inventories using existing remote sensingtechnologies suffer, on the whole, from beingmore expensive and less accurate than fieldinventory for the small tracts that are comingto dominate the southern forest landscape.Thus they are not widely used in most instances.However, traditional assessment and inventoryis only part of a larger picture of forestryinformation needs—needs that are likely tobecome urgent as management for increasedproduction intensifies on some private tractsand the demands for multiple uses continue todrive public forest management. The followingdiscussion addresses the potential effect of moreintensive management of portions of the privateland base on the future of forestry remote sensing.

Partially as a result of increased demand, andpartially as a consequence of effective reductionsin the land base resulting from (1) permanentland use conversion, (2) changes in public landmanagement priorities, and (3) changes inthe motivations and attitudes of nonindustrialprivate forest landowners, forest managers areincreasingly being called upon to produce morewood or fiber from less land with shorter rotations.This challenge is being met with silviculturaltools that have agricultural analogs, such as sitepreparation, nutrient management, release fromcompetition, and improved genetic stocks. Theagricultural model can be pushed even fartheras management intensifies; it can be argued thatintensive forest management is as suited forprecision forestry as intensive farm managementis suited for precision agriculture. Mulla’s (1997)definition of precision agriculture is as follows:

Precision agriculture is an approach forsubdividing fields into small homogeneousmanagement zones where fertilizer,herbicide, seed, irrigation, drainage, ortillage are custom-managed according tothe unique mean characteristics of themanagement zone.

Precision forestry is analogous to precisionagriculture in that traditional management units(forest stands are analogous to agricultural fields)

must be subdivided for specific prescription ofsilvicultural treatments; e.g., site preparation,fertilization, and release from competition.

However, there are some important differencesbetween precision agriculture and precisionforestry. Forest stands are typified by the(1) important array of public goods provided,such as clean water, wildlife habitat, carbonstorage, and recreational opportunities; (2) muchlonger rotations required; (3) widespread use ofhelicopters for spraying; (4) species; (5) minimaluse of irrigation; and (5) general lack of effectiveyield monitoring at the time of harvest. Thislast difference cannot be seen as problematicfor precision forestry, as many experts (e.g.,Pierce and Nowak 1999) agree that yield-baseddetermination of preharvest spatial variabilityis a temporary solution, to be replaced by the useof remote sensing technologies. Given the relativeimportance and degree of development of forestryapplications of remote sensing, precision forestryshould not need to evolve through the yield-monitoring stage. Furthermore, precision forestrysubsumes precision silviculture, precisioninventory, precision growth and yield, andprecision harvest scheduling, creating a completeinformation pathway. Remote sensing and relatedgeospatial information technologies provide theinventory information necessary to (1) define thetreatment unit for each silvicultural prescriptionand (2) provide the within-stand measurementsof forest biophysical parameters necessary forgrowth-and-yield models and related harvestscheduling models.

Landowner records usually provide informationon stand type, age, initial stocking density, andsite preparation. Required parameters suited toremote estimation include leaf area index, currentstem density, crown diameter, height, and speciesor species group. The last is especially helpfulfor timing and/or locating need for release fromcompetition. Foliar nutrients can also be assessedusing airborne hyperspectral, or, potentially,tailored handheld instruments (e.g., Bortolot andWynne 2003), though such approaches have so farnot been cost-effective when compared with labanalysis of foliar samples.

The relative need for and cost:benefit ratio ofremotely sensed and precisely located in situ datamust drive both the research in and the adoptionof appropriate technologies. Furthermore withinthe wide variety of remotely sensed data that arewell suited for precision forestry applications,researchers must find the best combination of

115

spectral resolution, spatial resolution, and canopyheight information for estimating each requiredparameter. Data types include but are not limitedto (1) canopy height models derived from lidar ordigital photogrammetry, (2) high spatial-resolutionoptical data, (3) moderate-resolution multispectraldata, and (4) hyperspectral data at a variety ofspatial resolutions. Research being carried outby Government, industry, nongovernmentalorganizations, and universities is providing thebase for improved integration of remote sensingin forest management. However, the gap betweenremote sensing research and accessible, usefulinformation is still too large.

FROM DATA TO APPLICATIONS

I t has been said that production forestryorganizations make or lose money near thebottom rungs of the organizational ladder,

not at the top (Smith and others 2003). Most fieldforesters have substantial experience with aerialphotographs but have neither the time for northe interest in processing images from digitalremotely sensed data. In many cases, given thewidespread use of and familiarity with aerialphotographs in forestry, digital images can besubjectively analyzed for the wide variety ofapplications mentioned in this chapter. However,it can be argued that this model limits the potentialutility of remote sensing to forestry, as it may notprovide any net increase in information. The kindsof sensors that will help facilitate a net increase inforestry information derived from remotely senseddata include laser altimeters and hyperspectralscanners. The data these sensors yield will only besuited, in their raw form, for use by image analystsor other experts in the same field. Field forestersgenerally do not have the experience to qualifyas experts in image analysis. They “requireinformation, not images . . . .” (Oderwald andWynne 2000).

The National Research Council, in theirrecent study on “Transforming Remotely SensedData into Information and Applications” (2001),identified three gaps that must be bridgedto develop effective civilian applications ofremote sensing:

1. The gap between raw data and theinformation needed by end users

2. The gap in communication and understandingbetween end users and those with remotesensing technical experience and training

3. The financial gap between data acquisitionand usable applications

While these identified gaps apply across thespectrum of end users, they are particularlyrelevant for forestry. The report goes on to makethe following recommendations, which are alsoquite pertinent to forestry:

1. Publicly available studies identifying the fullrange of short-term and long-term costs andbenefits of remote sensing applications shouldbe carried out by a full range of public andprivate stakeholders.

2. Cognate Federal Agencies should help fundand foster a wide variety of remote sensingtraining materials and courses.

3. Staff exchanges should occur betweenremote sensing users and producers.

4. Graduate fellowships and researchassistantships should be sponsored by landgrant, sea grant, and agricultural extensionprograms to encourage work at agencies thatuse remote sensing data.

5. Federal agencies need to expand their supportfor applied remote sensing research.

6. Formal mechanisms should be established toenable applications users to advise private andpublic sector providers on their requirements.

7. Data preservation is important and should beroutinely addressed by all data providers.

8. Internationally recognized formats, standards,and protocols should be used wheneverpossible to facilitate data exchange and easeof use.

It is evident from the foregoing that forestryis not the only application area where there is alarge gap between obtaining remote sensing dataand translating the data into useful information.It is evident that concrete recommendations forbridging this gap have been made. As a communityof forest scientists and managers, it is ourresponsibility to identify information needs of thefuture, and to make sure that adequate methodsof meeting these needs can be made available.

LITERATURE CITEDAvery, T.E. 1978. Forester’s guide to aerial photo

interpretation. Washington, DC: U.S. Department ofAgriculture, Forest Service. 41 p.

Avery, T.E.; Burkhart, H.E. 2002. Forest measurements.Boston: McGraw-Hill. 456 p.

Bortolot, Z.J.; Wynne, R.H. 2003. A means of spectroscopicallydetermining the nitrogen content of green tree leavesmeasured at the canopy level that requires no in situnitrogen data. International Journal of Remote Sensing.24(3): 619–624.

Cha

pter

11 .

For

est

Men

sura

tion

wit

h Re

mot

e Se

nsin

g

Sout

hern

For

est

Scie

nce:

Past

, Pre

sent

, and

Fut

ure

Prod

ucti

vity

116

Brown, T.L.; Lassoie, J.P. 1998. Entry-level competency andskill requirements of foresters: what do employers want?Journal of Forestry. 96(2): 8–14.

Food and Agriculture Organization of the United Nations.1997. State of the world’s forests—1997. Rome. [Numberof pages unknown].

Gougeon, F.A.;. St-Onge, B.A; Wulder, M.; Leckie, D.G. 2001.Synergy of airborne laser altimetry and digital videographyfor individual tree crown delineation. Canadian symposiumon remote sensing. Sainte-Foy, Quebec, Canada: [Publisherunknown]. 7 p.

Helms, J.A., ed. 1998. The dictionary of forestry. Bethesda,MD: Society of American Foresters. 210 p.

Lund, H.G.; Befort, W.A.; Brickell, J.E. [and others]. 1997.Forestry. Manual of photographic interpretation. Bethesda,MD: American Society for Photogrammetry and RemoteSensing: 399–440.

McCombs, J.W.; Roberts, S.D.; Evans, D.L. 2003. Influence offusing lidar and multispectral imagery on remotely sensedestimates of stand density and mean tree height in amanaged loblolly plantation. Forest Science. 49(3): 457–466.

Means, J.E.; Acker, S.A.; Fitt, B.J. 2000. Predicting foreststand characteristics with airborne scanning lidar.Photogrammetric Engineering & Remote Sensing.66(11): 1367–1371.

Mulla, D.J. 1997. Geostatistics, remote sensing, and precisionfarming. In: Lake, J.V.; Bock, G.R.; Goode, J.A., eds.Precision agriculture: spatial and temporal variability inenvironmental quality. [Location of publisher unknown]:John Wiley. [Number of pages unknown].

Naesset, E. 2002. Predicting forest stand characteristicswith airborne scanning laser using a practical two-stageprocedure and field data. Remote Sensing of Environment.80: 88–99.

National Research Council. 2001. Transforming remote sensingdata into information and applications. Washington, DC:National Academy Press. 75 p.

Nilsson, M. 1996. Estimation of tree heights and stand volumeusing an airborne lidar system. Remote Sensing of theEnvironment. 56: 1–7.

Oderwald, R.G.; Wynne, R.H. 2000. Field applications forstatistical data and techniques. Journal of Forestry. 98(6):58–60.

Olson, C.E., Jr. 1960. Elements of photographic interpretationcommon to several sensors. Photogrammetric Engineering.26(4): 651–656.

Paine, D.P. 1981. Aerial photography and image interpretationfor resource management. New York: John Wiley. 571 p.

Pierce, F.J.; Nowak, P. 1999. Aspects of precision agriculture.Advances in Agronomy. 67: 1–85.

Popescu, S.C. 2002. Estimating plot-level forest biophysicalparameters using small-footprint airborne lidarmeasurements. Blacksburg, VA: Virginia PolytechnicInstitute and State University, College of NaturalResources. 155 p.

Popescu, S.C.; Wynne, R.H.; Nelson, R.F. 2002. Estimatingplot-level tree heights with lidar: local filtering with acanopy-height based variable window size. Computersand Electronics in Agriculture. 37: 71–95.

Popescu, S.C.; Wynne, R.H.; Nelson, R.F. 2003. Measuringindividual tree crown diameter with lidar and assessing itsinfluence on estimating forest volume and biomass. CanadianJournal of Remote Sensing. 29(5): 564–577.

Popescu, S.C.; Wynne, R.H.; Scriviani, J.A. 2004. Fusion ofsmall footprint lidar and multispectral data to estimate plot-level volume and biomass in deciduous and pine forests inVirginia, U.S.A. Forest Science. 50(4): 551–565.

Sader, S.A; Vermillion, S. 2000. Remote sensing education:an updated survey. Journal of Forestry. 98(4): 31–37.

Smith, J.L.; Clutter, M.; Keefer, B.; Ma, Z. 2003. Specialcommentary: the future of digital remote sensing forproduction forestry organizations. Forest Science. 49(3):455–456.

Spurr, S.H. 1960. Photogrammetry and photo-interpretationwith a section on applications to forestry. New York:Ronald Press. 472 p.

U.S. Department of Agriculture, Forest Service. 1998.Changing economics of the national forest timber saleprogram. Washington, DC. [Number of pages unknown].

World Bank. 1992. World development report 1992:development and the environment. Oxford: OxfordUniversity Press. [Number of pages unknown].

Wynne, R.H.; Oderwald, R.G.; Reams, G.A.; Scriviani,J.A. 2000. Optical remote sensing for forest areaestimation. Journal of Forestry. 8–14.