SYMPOSIUM PAPERS

Advanced Information Technologies for Assessing Nonpoint Source Pollutionin the Vadose Zone: Conference Overview

D. L. Corwin,* K. Loague, and T. R. Ellsworth

ABSTRACTThe information age has ushered in an awareness of and concern

for global environmental problems such as climatic change, ozonedepletion, deforestation, desertification, and nonpoint source (NPS)pollution. Nonpoint source pollution is the single greatest threat tosurface and subsurface drinking water resources. Nonpoint sourcepollutants also pose a threat to sustainable agriculture, which is viewedas the most viable means of meeting the food demands of a worldpopulation that is expected to reach 9.4 billion by the middle of thenext century. The ability to accurately assess present and future NPSpollution impacts on ecosystems ranging from local to global scaleswould provide a powerful tool for environmental stewardship andguiding future human activities. Assessing NPS pollutant is a multidis-ciplinary problem. To address the problem, advanced informationtechnologies and methodologies are needed that draw from all areasof science and are applied in a spatial context. It was from this settingthat the 1997 Joint AGU Chapman/SSSA Outreach Conference Ap-plication of GIS, Remote Sensing, Geostatistics, and Solute Trans-port Modeling for Assessing Nonpoint Source Pollutants in the Va-dose Zone (19-24 Oct. 1997, Riverside, CA) materialized. Theobjective of the conference was to examine current muitidisciplinarytechnologies and methodologies for assessing NPS pollutants in thevadose zone, and to explore new conceptual approaches. It was theconference’s goal to provide a forum to stimulate multidisciplinaryinteraction to enhance the development of techniques for the real-time measurement and modeling of NPS pollution in the vadose zoneand subsurface waters.

TtHE INFORMATION-TECHNOLOGY AGE of the 1990s is aime of global environmental consciousness where

the ramifications of environmental stewardship stretcheven into political arenas. Historically, science and tech-nology have taken a back seat to political concerns inthe decision-making process. Science and technologyare no longer merely tools of national development, butrather they are becoming integral parts of the interna-tional political landscape as evidenced by the 1992 EarthSummit Conference in Rio de Janeiro and the 1997Kyoto Climate Summit. A knowledge and understand-ing of how resource utilization in individual countriesimpacts the global environment and associated political

D.L. Corwin, U.S. Salinity Lab., USDA-ARS, 450 West Big SpringsRoad, Riverside, CA 92507-4617; K. Loague, Dep. of Geological andEnvironmental Sciences, Stanford Univ., Stanford, CA 94305-2115;and T.R. Ellsworth, Dep. of Natural Resources and EnvironmentalSciences, Univ. of Illinois, Urbana, IL 61801. Received 17 Dec. 1997.*Corresponding author (dcorwin@ussl.ars.usda.gov).

Published in J. Environ. Oual. 28:357-365 (1999).

357

ramifications is essential to guide sound national andinternational policies.

Environmental issues such as climatic change, ozonedepletion, erosion, deforestation, desertification, andNPS pollution are of growing concern to industrializedand even nonindustrialized countries of the world.These problems are exacerbated by the trend in worldpopulation, which has doubled since 1950 and is ex-pected to reach 9.4 billion by the middle of the nextcentury (World Resources Institute, 1996). The world’sindustrialized countries now recognize that environ-mental issues are no longer parochial concerns definedby community, city, state, national, or even continentalboundary lines, but rather are global in perspective.

Because of the potential global impact of mankind’sactivities, it is no longer efficacious to address environ-mental issues as isolated or compartmentalized prob-lems. Concomitantly, the scientific weaponry necessaryto combat these global issues must be technologicallysophisticated to deal with the spatial scale and complex-ity of the problems. Technologies and methodologiesare needed that draw from all areas of science and crossover boundaries of scientific disciplines and subdisci-plines just as the problems themselves are interdisciplin-ary. It is against this backdrop that scientists addressingenvironmental problems find themselves, and fromwhich the 1997 Joint American Geophysical Union(AGU) Chapman/Soil Science Society of America(SSSA) Outreach Conference Application of GIS, Re-mote Sensing, Geostatistics, and Solute Transport Model-ing for Assessing Nonpoint Source Pollutants in theVadose Zone (19-24 Oct., 1997; Riverside, CA) materi-alized.

It is the objective of this paper to bring the NPSpollution problem, specific to the vadose zone (i.e., theunsaturated region from the soil surface to the ground-water table), into perspective and to discuss the multidis-ciplinary approach needed in the assessment of NPSpollutants. This paper serves as a preface to integrate thedisparate papers presented at the Joint AGU Chapman/SSSA Outreach Conference, some of which are pre-sented in this special symposium section of the Journalof Environmental Quality and others (i.e., keynote and

Abbreviations: NPS, nonpoint source pollution; AGU, AmericanGeophysical Union; GPS, global positioning system; NCGIA, Na-tional Center for Geographic Information and Analysis; REV, repre-sentative element volume; EM, electromagnetic induction.

Published March, 1999

358 J. ENVIRON. QUAL., VOL. 28, MARCH-APRIL 1999

invited papers) currently in press (see Corwin et al.,1999b).

EXTENT AND SIGNIFICANCE OF THENONPOINT SOURCE POLLUTION PROBLEM

Among the foremost global issues is satisfying the ever-growing need for natural resources to meet food and living-standard demands, while minimizing impacts upon an environ-ment that already shows signs of serious levels of biodegrada-tion (Corwin and Wagenet, 1996). There is concern for thefuture availability of limited natural resources such as waterand productive agricultural soil. Not only the availability offinite resources, but the condition of those resources as ina-pacted by domestic, industrial, and agricultural activities is ofconcern. The condition of soil and water resources is primarilyaffected by point source and NPS pollutants. Point source andNPS pollutants differ solely in the scale and the areal extent oftheir source. Nonpoint source pollutants in terrestrial systemsrefer to "those contaminants in surface and subsurface soiland water resources that are diffuse, or rather, are spreadover large areas" (Corwin, 1996). Nonpoint source pollutantsof the vadose zone primarily include pesticides, fertilizers,trace elements (such as B, Mo, and As), and salinity.

As pointed out by Duda (1993), the magnitude of NPSpollution is so complex and difficult to characterize that itdefies proper evaluation with respect to its global environmen-tal impact. Nonpoint source pollutants are recognized as themajor contributors to surface and groundwater contaminationworldwide (Duda, 1993) and agriculture remains as the singlegreatest contributor of NPS pollutants (Humenik et al., 1987).Globally, 30 to 50% of the earth’s surface is believed to beaffected by NPS pollutants (Pimental, 1993). Even within thehighly industrialized USA where point source industrial pol-lutants abound, NPS pollutants impair far more rivers andlakes than point sources (USEPA, 1994). About 80% of theassessed rivers and lakes in the USA are impaired by NPSpollutants, and only 20% by point sources (USEPA, 1994).Most of the reason for this is because developed countrieshave concentrated their effort on controlling point-source pol-lutants because of the difficulty in regulating diffuse sourcesof pollution.

Maintaining a delicate balance between agricultural produc-tivity and minimizing agriculture’s impact on the environmentis crucial to sustaining a projected global population range ofbetween 7.9 to 11.9 billion by 2050 in lieu of agriculture’sprominent role as a contributor of NPS pollutants to soil andwater resources (World Resources Institute, 1996). It is self-evident that the ability to assess the environmental impact ofNPS pollutants at local, regional, and global scales on a real-time and predictive basis is a key component to achievingsustainability of the environment and agriculture.

Historically, public attention to the NPS pollution of waterresources has focused upon the pollution of surface watersupplies. Limited surface water supplies and increased demandby agricultural and domestic sectors have placed greater de-pendency upon groundwater supplies. Currently, half of ourdrinking water needs and 40% of the irrigation water needsin the USA are met by groundwater supplies. Because of thisincreased dependency upon groundwater resources, publicconsciousness has shifted to the protection of groundwaterresources and of the overlying soil that serves as a conduitfor the movement of pollutants from the soil surface.

The significance of the NPS pollution problem lies in theramifications to human health. The protection of groundwaterresources from NPS pollutants has become a primary concernbecause of the public’s apprehension over long-term health

effects of drinking water containing small concentrations oftoxic chemicals. A secondary concern is the accumulation ofsignificant levels of inorganic (e.g., salinity and trace elements)and organic chemicals (e.g., pesticides) in soil that detrimen-tally impact agricultural productivity and crop quality. Theextent of the impact of particularly dangerous contaminantsis only now becoming apparent to developed countries. Radio-nuclides, pesticides, solvents, heavy metals, and petroleumcompounds are well known, but a new generation of carcino-gens, mutagens, and teratogens are found to cycle from onemedium or form to another including various classes of PCBs,furans, dioxins, and organochlorines (Duda and Nawar, 1996).Some chemicals (e.g., Cd, Pb, Hg, DDT and degradation prod-ucts, methoxychlor, triazine herbicides, synthetic pyrethroids,lindane, PCBs, dioxins, and others) are known to disrupt theendrocrine system causing metabolic, neurological, and im-mune system abnormalities (Duda and Nawar, 1996).

Ostensibly, the point has been reached where the awarenessof NPS pollution is less of a concern than the voluntary andregulatory actions that must follow. Before decision makerscan formulate effective regulatory actions, the ability to reli-ably and cost-effectively assess NPS pollutants on a real-timeand predictive basis is of paramount importance. It is throughreal-time assessment that a continued inventory of the NPSpollution problem can be maintained to determine the extentof the problem and evaluate changes, whether for better orworse, that gauge the effect of regulatory and intended amelio-rative actions. In addition, predictive assessments set the stagefor posing "what if" scenarios that serve a preventative role bysuggesting actions that will alter the occurrence of detrimentalconditions before they manifest.

MULTIDISCIPLINARY NATUREOF THE PROBLEM

Solutions to complex global environmental problems standon the shoulders of technological and interdisciplinary scien-tific achievements. To solve the problems of climatic change,ozone layer depletion, deforestation, desertification, and NPSpollution, it is necessary to examine these issues from a multi-disciplinary, systems-based approach, and to look at theseproblems in a spatial context.

The formidable barriers to the NPS pollution problem arecomprised of the complexities of scale and position; the com-plexities of the physical, chemical, and biological processes ofsolute transport in porous media; and the spatial complexitiesof the soil media’s heterogeneity. The knowledge, information,and technology required to address each of these complexissues crosses several subdisciplinary lines: classical and spatialstatistics, remote sensing, geographic information systems(GIS), surface and subsurface hydrology, soil science and evenspace science. Spatial statistics is essential in dealing withthe uncertainty and variability of spatial information; remotesensing is needed to measure physical, chemical, and biologicalproperties of soil or measure real-time environmental impactsover vast areas in a cost effective and timely manner; GIS isneeded to manipulate, store, retrieve, and display the tremen-dous volumes of spatial data; and water flow and solute trans-port models are needed to simulate future scenarios to assesspotential temporal and spatial changes. Neural networks andtransfer functions provide a means of deriving complex solutemodel parameters from easily measured data. Hierarchicaltheory establishes an organizational hierarchy of pedogeneticmodeling approaches and their appropriate scale of applica-tion. Fuzzy logic provides a means of handling vague andimprecise data whether as a means to characterize map unitsor transitional boundaries between map units. Uncertainty

CORWIN ET AL.: ASSESSING NONPOINT SOURCE POLLUTION IN THE VADOSE ZONE 359

analysis serves as a measure of the reliability of simulatedmodel results. Precise geographic location and areal extentare captured with the space science technology of a globalpositioning system (GPS).

CONFERENCE GOAL AND OVERVIEW

Methodologies integrating all of the aforementioned ad-vanced-information technologies are the linchpin to effectivelyassessing NPS pollutants. In particular, GIS has burgeonedas a data management and visualization tool for addressingspatially-related environmental problems. In the past 5 yr, theNational Center for Geographic Information and Analysis(NCGIA) has presented three conferences focusing on GISapplications to general environmental issues (see the NCGIAWeb Site for details http://www.ncgia.ucsb.edu). The 1995American Society of Agronomy-Crop Science Society ofAmerica-Soil Science Society of America (ASA-CSSA-SSSA) Bouyoucos Conference was the first conference tofocus on the application of GIS to the modeling of NPS pollut-ants specifically in the vadose zone (Corwin and Loague, 1996;Corwin and Wagenet, 1996). The Bouyoucos Conferenceserved as the predecessor to the 1997 Joint AGU Chapman/SSSA Outreach Conference. Broadening the focus beyondGIS, the Joint AGU Chapman/SSSA Outreach Conferenceaddressed the full spectrum of the multidisciplinary applica-tion of GIS, remote sensing, geostatistics, and solute transportmodeling to the assessment of NPS pollutants in the vadosezone. The objective of the jointly-sponsored conference wasto explore current multidisciplinary technologies and method-ologies for assessing the impact of NPS pollution upon soiland groundwater resources. Assessment, as used in this con-text, refers to the in-situ quantification of a NPS pollutantand/or the determination of change of some NPS pollutantover time. This change can be either measured in real timeor predicted with a model. It was the conference’s goal toprovide a forum to stimulate interaction between the subdisci-plines of spatial statistics, remote sensing, GIS, and solutetransport modeling to enhance the development and evalua-tion of techniques for the measurement, assessment and mod-eling of NPS pollution in the vadose zone and subsurfacewaters.

Thirty keynote/invited papers and 51 volunteered paperswere presented during the 4 1/2 d conference. The 30 keynoteand invited papers concerned assigned topics designed to pro-vide general reference information about the subdisciplinesinvolved with assessing NPS pollution in the vadose zone. SeeCorwin et al. (1999b) for a compendium of the conferencekeynote and invited papers. The volunteered papers providedspecific examples of state-of-the-art research that either con-centrated on a single aspect or component of the NPS pollutionproblem, or applied an integrated approach to the NPS prob-lem. The papers in this special symposium section of the Jour-nal of Environmental Quality are comprised of selected volun-teered papers.

Organizationally, the conference consisted of eight techni-cal sessions plus an introductory session. The introductorysession focused on (i) politics, technology, and environmentalpolicy-making; (ii) future trends in commercial GIS softwarespecific to environmental assessment; (iii) the societal valueof environmental information; and (iv) the extent of the NPSpollution problem. The subsequent eight technical sessionsconsisted of (i) integrated information technology approachesto NPS pollution problems; (ii) subsurface solute transportmodeling; (iii) spatial statistics, geostatistics, and uncertaintyanalysis; (iv) transfer functions, estimation methods, and pa-rameterization; (v) remote sensing and noninvasive tech-

niques; (vi) problems of scale and scaling; (vii) quantificationof spatial variability and soil-landscape models; and (viii)case studies.

The conference opened with an "ivory-tower shattering"keynote address concerning the role of environmental model-ing in the decision-making process. John King noted the grad-ual convergence of policy-makers and environmental model-ers, but warned that modelers must be aware that the policyprocess is necessarily about politics and that models used inthat realm are applied as political weapons--not as unbiasedtools. The notion that science and technology will mitigateenvironmental problems on a "truth wins" basis is illusionary.The success of GIS-linked models in policy-making simplydepends on the model’s ability to serve the analytical needsof the greatest number of policy makers even if they are inpolitical opposition. Furthermore, even though scientific andtechnological advances are viewed as critical for the successfulmanagement of environmental issues, it is unlikely that scien-tists will play central roles of power and authority in thedecision-making process. Subsequent keynote addresses dealtwith the advanced information technologies needed to assessNPS pollutants in the vadose zone. John Wilson focused oncurrent and future trends in the development of integratedmethodologies for assessing NPS pollution. He suggested thatthe explosive growth in data capture technologies has causeda data-rich, but information-poor condition that has led to thefragmentation of GIS applications because commercial GISsoftware cannot keep pace with the growth in niche implemen-tations. Geographic information sytems is in serious need oflinkage to software that improves data- and process-modelingcapabilities to provide a knowledge-rich system.

The session on subsurface solute transport modeling beganby Graham Fogg expounding the virtues of an integratedanalysis of both vadose zone and groundwater processes toaddress aquifer vulnerability. This analysis may be accom-plished by applying geomorphological modeling and hydro-stratigraphic structure to account for the spatial variability offlow within and below the vadose zone. William Jury notedthat stochastic approaches to modeling NPS pollutants in thevadose zone at regional scales are the only viable approachbecause extensive spatial variability of water flow and solutetransport properties make deterministic modeling unfeasibleat this scale, necessitating some form of approximate stochasticapproach using a local model representation that extrapolatesfrom limited sample data. Donald Myers surveyed the roleof spatio-temporal statistical modeling. Remote sensing andnoninvasive techniques were addressed both in general andin a specific application (i.e., geophysical measurements of soilelectrical conductivity) by Marion Baumgardner and JamesRhoades, respectively.

In a pivotal keynote paper concerning scaling spatial pre-dictability, Philippe Baveye served as the conference’s "devil’sadvocate" by questioning whether GIS was the answer to thespatial problems associated with NPS pollution or, indeed, ifit was even part of the answer, and by stabbing directly intothe heart of distributed modeling as a viable NPS pollutionapproach. Philippe Baveye’s paper tweaked a cerebral nervethat placed it apart from all other conference presentationsmaking it a centerpiece worthy of greater discussion and ofpause for thought. Professor Baveye drew an analogy betweenthe present state of knowledge of modeling NPS pollutionand development of thermodynamic gas laws from physicaltheories of particle dynamics. Newton originally derived thefundamental equations describing the motion of a particleabout the central potential field resulting from another parti-cle. These equations are uniquely soluble and descrilzre themotion of conic orbits. The introduction of a third particle,

360 J. ENVIRON. QUAL., VOL. 28, MARCH-APRIL 1999

however, results in a situation where no solution can be foundand as more particles are introduced, the analytical intransi-gence continues. Despite the inability to predict the individualmotions of gas particles in a container from Newton’s funda-mental equations, an independent line of theory, initiallydriven by the need to improve the efficiency of the steamengine, led to the formulation of the thermodynamic gas laws.The thermodynamic gas laws are extremely simple statementsconcerning how properties of an ideal gas system change sub-ject to alterations in applied constants (i.e., how pressurechanges when the temperature or volume is altered). Thus,despite the inability to predict larger-scale behavior fromsmaller-scale conceptual understandings, there are coarser-scale descriptions that allow accurate predictions of the sys-tem. Analogously, scientists may be studying NPS pollutionat a level of understanding and detail dictated by measurementinstruments such as remote sensing and information technol-ogy tools such as GIS rather than at a level dictated by theappropriate conceptual framework. Furthermore, the funda-mental unit for landscape study is often specified as a catch-ment or watershed, or even more arbitrarily by survey orgeopolitical boundaries (i.e., quarter section lines or waterdistrict boundaries), without knowing if these units are mean-ingful or whether they are the appropriate levels of aggrega-tion for subsurface solute transport. Potentially, scientists arebeing asked by policy makers to predict at a level of detailsomething that is inherently unpredictable. This suggests theneed for exploration of the scales, or ranges of scales, at whichthe dynamics of processes are drastically simplified, which fallsin line with a quote from Einstein concerning the parsimony ofmodel design, "Models need to be as simple as possible, butnot any simpler."

The invited presentations were intended to be more paro-chial by focusing on specific issues concerning the assessmentof NPS pollutants in the vadose zone. From an economicsperspective, Richard Bernknopf provided an approach for theestimation of the societal value of information for assessingNPS pollution. A series of subsurface solute transport model-ing papers covered applications and philosophical issues. Don-ald Suarez suggested a reevaluation of the generally acceptednotion that sophisticated mechanistic models may be unrealis-tic for field- or regional-scale applications. This reevaluationis based on advances in remote data collection, geostatisticaltechniques, and advances in the understanding of soil chemicalprocesses (i.e., the substitution of kinetic models for equilib-rium assumptions and representation of adsorption processeswith defined surface species rather than site-specific empiricalparameters). Jin-Ping Gwo performed a numeric examinationof the multiscale effects of mass transfer processes on contami-nant transport at a watershed scale and quantified the uncer-tainties when small-scale processes are upscaled to a water-shed. Jets~ Stoorvogel and Johan Bouma conducted a riskassessment in space and time for nematicide leaching in aCosta Rican banana plantation using the LEACHP model.John Norman provided a broader perspective by addressingthe role of soil-plant-atmosphere models in NPS pollution.Finally, David Mulla tackled the problem of model validationby outlining factors to consider in validating NPS pollutionmodels at moderate to large scales. Invited papers concerningtransfer functions and parameterization involved (i) the appli-cation of artificial neural networks for developing pedo-trans-fer functions of soil hydraulic properties by Henk W6sten andS. Tamari; (ii) the accuracy and reliability of pedo-transferfunctions for use in regional-scale modeling by Yakov Pachep-sky and his colleagues; and (iii) an impressive study by MarkRockhold using geostatistical methods, similar-media scaling,and conditional simulation with soft data for the parameteriza-tion of flow and transport models.

The contingent of invited speakers concerning remote sens-

ing and noninvasive techniques remained true to their profes-sional stature by providing informative reviews of remote sens-ing to surface and subsurface hydrologic measurements.NASA’s Ted Engman provided an overview of remote sensingin hydrology. More specifically, Thomas Schmugge addressedthe potential for deriving soil hydraulic properties in the va-dose zone with remote sensing. Finally, Wire Bastiaanssenreviewed the current state of spatially delineating actual andrelative evapotranspiration from remote sensing.

The entire session for the quantification of spatial variabilityconsisted of invited speakers because the ability to delineateand quantitatively represent spatial variability is a key to theapplication of GIS to the modeling of NPS pollutants in thevadose zone. Richard Ferguson began with a survey of thesampling and spatial analysis techniques useful in quantifyingthe composition of soil map units. The timely and practicaltopic of incorporating spatial variability into existing soil data-bases was discussed by Russ Yost. Fuzzy logic and inferencetechniques were used by A.-Xing Zhu to derive more accurateand detailed soil spatial information that allow the realisticcharacterization of the spatial joint distribution of landscapeparameters for distributed modeling at the watershed scale.A particularly scholarly and informative discussion of fractalapplications to soil properties and solute transport in porousmedia was presented by John Crawford. The final conferencepaper presented by John Letey dealt with the controversialquestion, "Is modeling supplanting observation rather thancomplementing it?" Professor Letey indicated that the answerto this question is intuitively obvious from the plethora ofunvalidated models, and that modeling should not serve as asubstitute for experimentation and observation. Rather, mod-eling and experimentation should be ongoing processes thatreciprocally enhance one another.

The volunteered papers were dominated by integrated in-formation technology approaches and case studies. A particu-larly noteworthy and encouraging trend, in contrast to earlierGIS-linked NPS pollution modeling papers presented at theBouyoucos Conference, was the significant increase in theapplication of uncertainty analysis to augment the modelingstudies as a source of reliability information as recommendedfrom the work by Loague and his colleagues (Loague et al.,1996).

PIECING THE NONPOINT SOURCE-POLLUTION ASSESSMENT PUZZLE

TOGETHER

Integrated systems for the prediction and real-time mea-surement of NPS pollutants are more clearly discussed in thecontext of the individual components that comprise the piecesof the NPS-pollution puzzle (Corwin et al., 1997). These com-ponents are dependent upon the advanced information tech-nologies of GPS, GIS, geostatistics, remote sensing, solutetransport modeling, neural networks, transfer functions, fuzzylogic, fractals, scale and scaling, hierarchical theory, and uncer-tainty analysis. Corwin et al. (1997) discussed in detail therole and the interrelationship of these advanced informationtechnologies within an integrated GIS-linked NPS pollutionmodel.

Geographic Information System-Linked Modelingof Nonpoint Source Pollution in the Vadose Zone

Ever since the classic paper by Nielsen et al. (1973) concern-ing the variability of field-measured soil water properties, sci-entists have been aware of the spatial complexity of soil, andthe complexity of solute transport in soil. Because of the spatialcomplexity of soil, modeling the fate and movement of NPSpollutants in the vadose zone is a spatial problem well suited

CORWIN ET AL.: ASSESSING NONPOINT SOURCE POLLUTION IN THE VADOSE ZONE 361

for the coupling of a deterministic or stochastic solute trans-port model with a GIS (Corwin et al., 1997). Corwin et al.(1997) defined GIS-tinked NPS pollution models as an inte-grated system of three basic components: model, GIS, anddata. The subcomponents of geostatistics, fuzzy logic, anduncertainty analysis provide enhanced, higher-order informa-tion that augments and refines the system.

Primary Components

Solute Transport Model

A model seeks to extract the essence from experimentaldata through the formulation of mathematical representationsof the relevant processes constituting a system (Corwin etal., 1999a). In the case of NPS pollutants, the mathematicalformulations describe the appropriate chemical, physical, andbiological processes involved in the transport of a solutethrough the vadose zone within the context of a GIS thatcontains model input variables and parameters that are spa-tially referenced (Corwin et al., 1999a). Corwin and Loague(1996) and Corwin et al. (1997, 1999b) provided in-depth cussions of modeling specific to NPS pollutants in the va-dose zone.

Model conceptualization is governed by the intended appli-cation of the model (Corwin et al., t999a). The intended appli-cation determines the scale; the scale determines the predomi-nant processes, and the level of spatial variability that mustbe accounted for in the model; the predominant processesdetermine the type of modeling approach (i.e., functional ormechanistic, qualitative, or quantitative) that should be used(Corwin et al., 1999a). Addiscott and Wagenet (1985) categorized solute transport models based upon two mainconceptual approaches: deterministic and stochastic models.Deterministic models "presume that a system or process oper-ates such that the occurrence of a given set of events leadsto a uniquely definable outcome" while stochastic models"presuppose the outcome to be uncertain" (Addiscott andWagenet, 1985). Arguably, mechanistic-deterministic modelsare generally better suited as research models that help tounderstand the interrelationship between the physical, chemi-cal, and biological processes influencing solute transport, whilefunctional-deterministic and stochastic models are more effec-tive as practical applications to real-world problems.

Historically, the inventory of developed deterministic mod-els of solute transport has been far greater than stochasticmodels. This is no surprise because deterministic models dateback over half a century, while stochastic models first madean appearance only 15 yr ago with Jury’s introduction of thetransfer function model (Jury, 1982). Even though stochasticmodeling has gained momentum, the development of deter-ministic models still has out paced stochastic models over thepast 10 and even 5 yr (Corwin et al., 1999a). The greatestbarrier to stochastic modeling probably lies in the largeamount of data needed to formulate the statistical momentsor probability distributions.

Even though the number of developed deterministic modelsfar outweighs the number of stochastic models, there arestrong and almost incontrovertible arguments for the use ofa stochastic over a deterministic model with respect to thesimulation of NPS pollutants. The most compelling argumentis that stochastic modeling is "virtually the only viable optionto pursue in making large-scale simulations of chemical move-ment" because "a fully deterministic model would require soilproperty data for every point in three-dimensional space overwhich the simulation is to be run" (Jury, 1996). Jury (1996)argued that because no one will ever be able to characterizeevery point in three-dimensional space, then modeling waterflow and solute transport at scales greater than a small plot

must use a stochastic approach of one kind or another by ne-cessity.

A review of GIS-linked NPS pollution models of the vadosezone has shown that nearly all are one-dimensional determin-istic models coupled to a GIS (Corwin et al., 1997). Corwinet al. (1997) categorized GIS-linked NPS pollution modelsinto three categories: regression models, overlay and indexmodels, and transient-state models. Deterministic GIS-linkedmodels have dealt with spatial complexities by coupling a one-dimensional, deterministic model of solute transport to a GISconsisting of a collection of discrete map units where eachmap unit represents a contiguous area of homogenous solutetransport model parameters (such as, adsorption-desorptioncoefficients, diffusion-dispersion coefficients, and hydraulicconductivity) and variables (such as, irrigation and precipita-tion amounts, and evapotranspiration). These "homogeneoustransport units" represent noninteracting, spatial domainswhere the lower limit of measurement scale at which variabilitybetween measurements is least. This is analogous to the repre-sentative element volume (REV) defined by Bouma (1990)and Mayer et al. (1999), and the "stream-tube" defined Jury and Roth (1990). Simulations performed on each discrete"homogeneous transport unit" rendered predictions of soluteprofile distributions and solute loads to the groundwater.However, the assumption of a homogeneous transport unitmight be an oversimplification because solute transport prop-erties within each of these units are still spatially variable andshould be statistically represented by probability distributions.

Temporal and spatial scale dictate the general type ofmodel. Models of solute transport in the vadose zone exist atall scales with hierarchical theory indicating the appropriatemodel for the scale of interest (Wagenet and Hutson, 1996).An important consideration in model conceptualization is forthe model to account for the predominant processes occurringat the spatial and temporal scales of interest. Qualitativelyspeaking, as spatial scale increases, the effects of complexlocal patterns of solute transport are attenuated and becomedominated by macroscale characteristics; so, as scale changes,different factors and processes may become dominant. A com-plete discussion of the application of models at different spatialand temporal scales was given by Wagenet (1993) and Wa-genet and Hutson (1996). Fractals provide a potentially usefultool for solute transport modeling because they are mathemat-ical constructs with complex geometries that have no charac-teristic scale; consequently, they have been used to modelprocesses across a range of scales. As such, fractals can serveas "thought tools" to avoid naive modeling assumptions andas components of predictive models (Crawford et al., 1999).

The magnitude and variability of a soil property often varieswith measurement scale. Wagenet and Hutson (1996) dis-cussed this issue of upscaling in greater detail and concludedthat sampling and measurement methods for both input pa-rameters and observations to evaluate model performanceneed to be consistent with the modeling scale. They also notedthat models that are developed for a specific spatial scaleare often calibrated with data from a much different scale,resulting in model parameters that have little physical signifi-cance, and a modeling approach that is questionable.

Geographic Information Systems

Geographic information systems have been succinctly de-fined by Parker (1988) as "an information technology thatstores, analyzes, and displays both spatial and nonspatialdata." Since the 1970s, GIS has burgeoned from a computer-based cartographic tool to a sophisticated information technol-ogy that can be applied to any georeferenced property orattribute, or even nonspatial information. The GIS has devel-oped into an extremely beneficial tool for the organization,

362 J. ENVIRON. QUAL., VOL. 28, MARCH-APRIL 1999

manipulation, and display of information associated with spa-tial environmental problems such as NPS pollution whetheratmospheric, terrestrial, or aquatic.

Even though tremendous advances in GIS have occurredduring the short span of 20 yr, commercially-available andpublic-domain GIS software still do not contain the cadreof statistical routines nor the internal software developmentflexibility needed to make the modeling of NPS pollutants aneasy task. This has prompted a plethora of GIS-linked solutetransport models "loosely" coupled to a GIS rather than em-bedded in the GIS (see the review by Corwin et al., 1997).

The GIS serves as a viable means for representing thecomplex spatial heterogeneity of the real world through inte-grated layers of constituent spatial information. To modelNPS pollution within the context of a GIS, each parameteror variable of the deterministic transport model is representedby three-dimensional layers of spatial information. The three-dimensional spatial distribution of each transport parameter/variable must be measured or estimated. This creates a tre-mendous volume of spatial information resulting from thecomplex spatial heterogeneity exhibited by the numerousphysical, chemical, and biological processes involved in solutetransport through the vadose zone. The GIS serves as thetool for organizing, manipulating, and visually displaying thisinformation efficiently.

The basis of the application of GIS to NPS pollution is theassociation of property attributes with spatial location. Globalpositioning system is an invaluable tool for quick and accurategeoreferencing of soil sample sites, thereby providing a carto-graphic location for soil physical and chemical property data,plant property data, and/or NPS pollutant concentration data.

Wilkinson (1996) noted that as the demand for spatial infor-mation grows, there is an ever-increasing complimentarity be-tween remote sensing and GIS. For instance, remote sensingcan be used to gather datasets for use in a GIS, and thesedatasets can be used as ancillary information to improve re-motely-sensed imagery, or GIS and remote sensing data canbe used in conjunction with each other for environmentalmodeling and analysis. Artificial neural networks have beenparticularly useful in extracting geographic information fromremote sensing imagery (Yoshida and Omatu, 1994; Chen etal., 1995). The simplest way to use preexisting GIS data inthe interpretation of remotely-sensed data is in broadly divid-ing the landscape into major areas with very different charac-teristics, that is, topographic, phenological, geological, andclimatological areas.

Data

Data is the fuel that drives a model. From this perspective,data are the most fundamental component of a NPS pollutionmodel linked to a GIS. Yet, obtaining data in a useful form tomeet the input requirements of current NPS pollution modelsremains a barrier (Corwin and Wagenet, 1996).

Corwin et al. (1997) pointed out three sources of data forGIS-linked NPS pollutant models: (i) existing data from soilsand meteorologic databases, (ii) estimated parameter and in-put data from transfer functions, and (iii) measured data.Examples of existing soils and meteorological databases in-clude: (i) SOTER--a world soils and terrain digital databasecompiled by the International Soil Reference and InformationCenter; (ii) SSURGO, STATSGO, and NATSGO--thecounty-, state-, and national-level soils databases, respectively,compiled by the USDA National Resource Conservation Ser-vice; (iii) UNSODA--the soil hydraulic properties databasecompiled by the USDA-ARS; (iv) NOAA’s weather stationdatabase; (v) SNOTEL--USDA/NRCS’s daily snow and pre-

cipitation database; and (vi) CIMIS--the crop evapotranspira-tion database for California compiled by the California De-partment of Water Resources. However, most existingdatabases, particularly soils databases, do not meet even theminimum requirements for many of the distributed-parametermodels used for NPS pollutants in the vadose zone; conse-quently, modelers’ thirst for data has driven them to the devel-opment of estimation methods such as transfer functions andartificial neural networks. Transfer functions relate readily-available and easy-to-measure soil properties to complex soilparameters and variables. A table of cited references for trans-fer functions of various solute transport model parametershas been compiled by Corwin et al. (1997). Artificial neuralnetworks are useful in solving multivariate nonlinear problemssuch that an output variable is explained by a set of inputvariables (Hush and Horne, 1993). Neural networks are alternative to predict soil parameters and variables from otherquantitative or qualitative soil variables (Pachepsky et al.,1996; Tamari et al., 1996). Remote sensing and noninvasivetechniques offer considerable promise for the measurementof some of the parameters and input variables used in model-ing NPS pollutants (Corwin et al., 1997). Corwin et al. (1997)reviewed the current status of remote sensing and its applica-tion specifically to modeling NPS pollutants in the vadosezone. Even though remote sensing holds great future promisefor fulfilling the input data needs of NPS pollutant models, ithas so far fallen short of expectations because of the initialoverselling of remote sensing’s environmental utility withoutthe fundamental research to backup initial boastful claims.Similarly, the real-time measurement of NPS pollutants in thevadose with remote sensing instrumentation is limited.

Subcomponents

Geostatistics

Corwin et al. (1997) listed the multiple support role geostatistics in a GIS-linked NPS pollutant model including(i) the determination of the suitability of data as model input,(ii) the estimation of data at locations where no data exist,and (iii) the post-processing of model simulation results. Thepreprocessing of data with preliminary geostatistical analysescan discover spatial trends, spatial correlation, and erratic data(i.e., specific data values that lie outside the general group).Kriging-type geostatistical techniques are useful as an interpo-lation tool to estimate data values at unsampled points. Exam-ples of kriging-type techniques and other spatial statisticaltechniques include kriging, co-kriging, disjunctive kriging, dis-junctive co-kriging, sequential indicator simulation, and sto-chastic simulation. Kriging is also useful as an interpolationtechniques for the post-processing of model simulation results.Semivariograms of computed results and initial conditions canbe used to indicate changes in spatial variability. From a regu-latory standpoint, geostatistics can determine the conditionalprobability that a particular NPS pollutant concentration valueis exceeded.

Fuzzy Set Theory

The variation seen in soil is generally continuous ratherthan discrete because of geomorphologic processes influencingits genesis and continued alteration. This suggests the needfor a continuous rather than a discrete approach to model soilvariability. Fuzzy set theory offers a means of modeling theimprecision or vagueness of a soil by quantifying a propertyaccording to its membership on a continuous scale rather thanon a Boolean binary or an integer scale. Central to fuzzy settheory is the membership function, which is a mathematical

CORWIN ET AL.: ASSESSING NONPOINT SOURCE POLLUTION IN THE VADOSE ZONE 363

relationship that defines the grade of membership with 1 rep-resenting full membership, 0 representing nonmembership,and an appropriate function defining intermediate grades be-tween 0 and 1. Aside from representing imprecise or vaguedata, fuzzy set theory can be applied to transitional boundariesto reflect the gradual changes that actually occur between mapunit boundaries on thematic maps. Fuzzy set theory allowsfor generalization of zones of transition rather than sharpboundaries. Fuzzy sets are a representational and a reasoningdevice that allow a database to hold information concerningthe indistinctness of spatial features (Altman, 1994).

Applications of fuzzy set theory and fuzzy logic in NPSpollutant assessment have included (i) the quantification andclassification of map units of soil properties or pollutants,(ii) modeling and simulation of soil processes, (iii) fuzzy geostatistics, (iv) soil quality indices, and (v) fuzzy measuresof imprecisely defined soil phenomena (see the review ofMcBratney and Odeh, 1997).

Uncertainty AnalysisThe reliability of model predictions is determined by the

error associated with its simulated output and intended useof the output. There are three sources of error inherent to allNPS pollutant models: (i) model error, (ii) input error, (iii) parameter error.

Uncertainty analysis pertaining to risk assessment, riskmanagement, and modeling of NPS pollutants was compre-hensively reviewed by Loague and his colleagues (Loague andCorwin, 1996; Loague et al., 1996). In a series of papers byLoague and colleagues, the indispensability of uncertaintyanalysis as a means of evaluating the reliability of predictionsfrom GIS-linked NPS pollution models was substantiated(Loague et al., 1989; Loague, 1991).

Real-Time Measurement of NonpointSource Pollutants

Real-time measurements of NPS pollutants in the vadosezone rely primarily on noninvasive geophysical techniquessuch as electromagnetic induction (EM). Though generallyless desirable because of its disruptive nature and greater laborrequirements, soil sampling is considered of greatest value asa means of determining ground truth for the calibration ofremote sensing or noninvasive techniques. Electrical resistivitytechniques are also quite useful particularly when mobilized(Rhoades, 1999). Electrical resistivity approaches are not par-ticularly invasive, but do require contact with the soil.

Some of the most comprehensive research conducted forthe real-time measurement of a NPS pollutant has been forthe measurement of soil salinity. Rhoades and his colleagueshave made the greatest scientific strides in the developmentand testing of instrumentation for the real-time measurementand mapping of soil salinity, specifically the measurement ofbulk soil electrical conductivity. Rhoades and Corwin firstdeveloped empirical, site-specific equations that determinedthe real-time soil salinity profile from aboveground EM mea-surements collected at the soil surface in the horizontal andvertical coil configuration (Rhoades and Corwin, 1981; Corwinand Rhoades, 1982, 1990). Lesch et al. (1995) and Hendrickxet al. (1997) have made significant improvements to this earlyEM work. Real-time mapping of bulk soil electrical conductiv-ity has been demonstrated and used to derive soil salinitymanagement strategies (Rhoades, 1996; Rhoades et al., 1999).

SUMMARY

It is premature to determine whether the 1997 JointAGU Chapman/SSSA Outreach Conference achieved

its goal of stimulating interaction between the attendingscientists of the disparate disciplines considered essen-tial to address the problem of assessing NPS pollutionof the vadose zone. Rather, the direction of future scien-tific publications will provide the best indication of theconference’s impact. However, the fruit of the laborexpended on the conference’s predecessor, that is, the1995 SSSA Bouyoucos Conference, was self-evident atthe 1997 Chapman Conference as indicated by the in-crease in the number of volunteered presentations con-cerning uncertainty analysis. The association of uncer-tainties with visualized results generated from transportmodels coupled to a GIS was cited in the conclffdingremarks of the Bouyoucos Conference as an area ofneeded intensified study. Renewed interest in the poten-tial of remote sensing to provide cost-effective and effi-cient methods of measuring solute transport model in-put data at an increased resolution was another areacited for intensified study. Evidence of increasing inter-est in remote sensing was shown by the near capacityattendance at the special sessions on remote sensingheld at the 1997 ASA-CSSA-SSSA Annual Meetings(Anaheim, CA; 27-31 Oct. 1997), which followed theChapman Conference.

Even though there have been tremendous advancesin assessing NPS pollution in the vadose zone duringthe last 10 yr (NRC, 1993; Corwin and Loague, 1996;Corwin and Wagenet, 1996; Corwin et al., 1997), thereis much yet to be done. Three obvious limitations to thecurrent use of GIS in regional-scale NPS vulnerabilityassessments are: (i) the lion’s share of GIS-driven mod-eling applications are only loosely coupled; (ii) in almostall cases data sparsity in space and time over large scales,not computer speed or storage capacity, limits our abil-ity to produce reliable vulnerability assessments; and(iii) most attempts at characterizing model performancehave been qualitative in nature.

Not unexpectedly, some of the same research needscited at the 1995 Bouyoucos Conference still remainbecause of the short 2-yr time span between the confer-ences. For example, continued basic and applied re-search into preferential flow is a high priority becauseit provides a pathway for the rapid and direct movementof NPS pollutants from the soil surface to groundwaterat their original applied concentrations. Intensified de-velopment and testing of remote sensing and noninva-sive instrumentation and methods are needed to reducethe labor-intensive burden of meeting the data-hungryappetites of NPS pollution models. The continued de-velopment of spatial statistical techniques are neededto reduce soil sampling requirements without reducingthe characterization of the soil’s spatial variability andstructure.

Additional research needs became evident at theJoint AGU Chapman/SSSA Outreach Conference in-cluding: (i) instrumentation and methodology for thegeospatial establishment of spatial domains of statisti-cally homogeneous properties of solute transport andthe fuzzy boundaries between these domains; (ii) clearer understanding of the issues of upscaling of spa-tial data and its aggregation, as well as downscaling and

364 J. ENVIRON. QUAL., VOL. 28, MARCH-APRIL 1999

disaggregation; and (iii) ultimately, the integration ofscientific, economic, and political considerations tomake NFS pollution assessments with advanced infor-mation technologies a decision-maker's, rather than apurely scientific, tool. Economists, information special-ists, policy-makers, entrepreneurs and even politiciansmust become part of the "loop" because these are theindividuals who will bring the "ivory tower" technolo-gies to fruition by bridging the sociopolitical and eco-nomic barriers that will place them into decision-mak-ers' hands. It is an illusion for scientists to think thattheir knowledge of technology and pursuit of truth willwin out over the practicalities of business and politics,and the public's capricious perceptions. Environmentalissues, and especially global environmental issues, arenot solely scientific issues nor are they problems thatcan be solved solely by scientists.

Despite the remaining gaps in knowledge and practi-cal application, positive strides have been made in theshort span of 2 yr. The GIS has gained increased legiti-macy in the eyes of even the most rigorous of scientistsas a tool that, when used knowledgeably, is viable andworthwhile for attacking NFS pollution problems. Simi-larly, the limitations of GIS have been brought intoperspective to the point where it is not misunderstoodas a panacea for solving spatial environmental problems,but merely one piece of the puzzle. Rather, an array ofadvanced information technologies is needed along withthe technical expertise and scientific guidance to collec-tively unravel complicated environmental problems.

Perhaps the forthcoming greatest technological ad-vance for modeling and assessing NFS pollution in thevadose zone will be the ability to perform regional-scalesimulations with a system that integrates GIS; solutetransport modeling; pedo-transfer functions, neural net-works and remote sensing for estimating and/or measur-ing transport properties; geostatistics for characterizingvariability structures and spatial interpolation; fuzzy settheory for handling vague and imprecise data; spatialand temporal analysis; and uncertainty analysis to allowdecision-makers to evaluate the reliability of infor-mation.

ACKNOWLEDGMENTSThe authors wish to acknowledge the American Geophysi-

cal Union and the Soil Science Society of America for provid-ing the primary funds for convening the 1997 Joint AGUChapman/SSSA Outreach Conference Applications of GIS tothe Modeling of Non-Point Source Pollutants in the VadoseZone (19-24 Oct. 1997, Riverside, CA). The authors also wishto acknowledge the National Aeronautics and Space Adminis-tration, National Science Foundation, University of Illinois,and Center for Earth Science Information Research (CESIR)at Stanford University for generously providing additionalfunds to cover travel grants for participating students andinvited speakers. The USDA-ARS U.S. Salinity Laboratoryis acknowledged for providing the senior author the time toparticipate as the organizer of the conference. The co-sponsorsare acknowledged for their assistance in promoting the con-ference: American Society for Photogrammetry & RemoteSensing (ASPRS), American Water Resources Association

(AWRA), American Society of Agricultural Engineers(ASAE), and Soil and Water Conservation Society (SWCS).

The authors also thank Brenda Weaver, Kristen Hansen,Brenda Glenn, the entire Meetings Staff at AGU, and espe-cially Ellyn Grossman-Terry for their diligent work and efforton the 1997 Joint AGU Chapman/SSSA Outreach Confer-ence. On a technical note, the authors wish to express theirgratitude to Dr. John Crawford for his eloquent and perceptiveinsights into the subtleties and implications of Dr. PhilippeBaveye's presentation concerning scaling spatial predictabil-ity. Finally, the 1997 Joint AGU Chapman/SSSA OutreachConference and all publications derived from the conferenceare dedicated in honor and memory of Professor Robert "Jeff"Wagenet for his significant contributions to soil physics andenvironmental science.

CORWIN ET AL.: ASSESSING NONPOINT SOURCE POLLUTION IN THE VADOSE ZONE 365