4th International Conference on Integrating GIS and Environmental Modeling (GIS/EM4):
Problems, Prospects and Research Needs. Banff, Alberta, Canada, September 2 - 8, 2000.

Strategies for implementing a multi-scale assessment tool for natural resource management:

a geographical information science perspective

GIS/EM4 No. 173

Chris S. Renschler
Bernard A. Engel
Dennis C. Flanagan

Abstract

Spatially distributed, multi-scale modeling assessing environmental processes for practical decision-making requires a holistic geographical information science perspective. The uncertainty of assessment results and decision-making at various scales of interest caused by uncertainties in data source, processing method and model application have to be at a minimum. The research presented outlines the procedures in GIS-based data processing for an effective simulation of fundamental environmental processes that are matched to realistic data availability, along with hydrologic and geomorphic settings. Procedures for implementing a multi-scale assessment tool for natural resource management with a focus on a process-based soil erosion model are discussed. To be useful for decision-makers, surface runoff generation and soil erosion prediction models must consider spatial and temporal variability in hydrological and water erosion processes and must be applicable to a variety of regions, with a minimum of calibration. The growing use of mathematical models in local to regional scale soil conservation requires the development and evaluation of data scaling methods. In particular it is important to assess how models function at management levels and scale of interest and with commonly available data. The process-based hillslope version of the Water Erosion Prediction Project (WEPP) model for simulating soil erosion along multiple flowpaths was applied for various scales of watersheds. The approach is based on the requirements of an ongoing implementation process of erosion models and utilizing standardized data that is commonly available in the United States. As an example the effect of different sources of commonly available Digital Elevation Models (DEM) on the model prediction results and the decision-making process of land users are investigated. The study demonstrates that data scale and quality of commonly available data sources can have a significant impact on model results. Geographical Information Systems (GIS) play a major role in the data preprocessing and visualization of model input and prediction results. These effects of data transformation within the GIS and the model have to be quantified and visualization techniques have to be integrated within a graphical user interface to make the user aware of model input and output reliability. If model results are used in management decisions, it is critical to assess how models behave with both research grade data and commonly available data, and the implications that these differences might have for management and policy decisions.

Keywords

Soil erosion, assessment, implementation, decision-making, WEPP, digital elevation model.

Introduction

Decision-makers operating at different scales of interest and responsibility have to deal with different types of environmental problems and seek solutions to handle the complexity of natural and human actions causing these problems. The expansion and concentration of cultivated acreage, for instance, led to a drastic change in land use as well as water and sediment fluxes in environmental systems. Catastrophic events such as the Dust Bowl during the 1930s or the Mississippi floods in 1993 caused excessive losses of fertile agricultural land and increased concerns about the economic and environmental impact of soil erosion at local, regional, national and global scales. Scale-dependent variations in soil erosion processes can be seen through sediment yield measurements and predictions, and well-illustrated in publicly available data (e.g. NRCS 1999, Trimble 1999). Methods used to extrapolate observations of plot studies or model approaches to determine average annual soil erosion rates at the regional, continental and global scale have proved controversial (Pimentel et al. 1995; Boardman 1998). Interpolated erosion rates over large areas, if measured or estimated by models, have limitations when used to guide management at local levels. This is because interpolated rates do not identify where the erosion problem occurs specifically, and do not provide sufficient information to guide or assess efforts to develop ways to make land use sustainable at a specific location. Thus research that aims to predict natural processes requires detailed study of isolated processes as the basis for gradually understanding complex multiple process interaction. In response to such hazards related to soil erosion, especially from agricultural land, the development and implementation of techniques for natural resource management were fostered by interest groups and policy-driving organizations to support decision-makers in countries worldwide (Troeh et al. 1999).

Implementation of soil erosion assessment tools

There is a long history of developing erosion models for soil and water conservation in U.S. research. The most known and applied approach for estimating long-term average annual soil loss is the Universal Soil Loss Equation (USLE) (Wischmeier and Smith 1978) and the Revised Universal Soil Loss Equation (RUSLE) (Renard et al. 1997). Both are simple empirical equations based on factors representing the main processes causing soil erosion. They have proved to be practical, accessible prediction tools and were therefore implemented in the U.S. soil and water conservation legislation. But these model approaches have been used and misused widely at various scales worldwide (see also Wischmeier 1976). Efforts in erosion process research lead to the development of the process-based hillslope soil erosion model WEPP (Flanagan and Nearing 1995). WEPP simulates climate, infiltration, water balance, plant growth and residue decomposition, tillage and consolidation to predict surface runoff, soil loss, deposition and sediment delivery over a range of time scales, including individual storm events, monthly totals, yearly totals or an average annual value based on data for several decades. The WEPP model is a continuous distributed-parameter soil erosion assessment tool that can be applied for representative hillslopes and a channel network at small watershed scales (Ascough II et al. 1997). A comparison of the performance of WEPP with other state-of-the-art erosion models using common data sets showed that data quality is an important consideration and primarily process-based models not requiring calibration have a competitive edge to those in need of calibration (Favis-Mortlock 1998).

Multi-scale assessment tools - the role of GIS

Geographical information science develops techniques for spatial analysis and decision-support in spatial environmental analysis. Geographical Information Systems (GIS) are dominantly used for data preprocessing and visualization of available data sources as well as the handling of data to apply environmental assessment models. A GIS-driven graphical user interface is a user-friendly approach to combine the decision-support of an environmental prediction model and the spatial capabilities of a GIS for practical assessment purposes (Figure 1). A successful implementation of an environmental model assessment approach requires the use of widely available data sets, reliable model predictions and analysis of the sequence of information transformation steps involved. Each of these data transformation steps considers a transformation of originally observed data at a certain scale for its optimum use in an assessment approach for a particular scale of interest. These transformation steps are called scaling steps.

The users' interest - scaling of geo-spatial information

The need for the best possible, accurate and precise assessment at the users' scale of interest requires designing a sequence of data transformations resulting in useful information with defined uncertainty. Therefore the final design of an optimum scaling sequence for decision-making has to consider the following: the user's spatial and temporal scales of interest, the best choice of a model to support decision-making at that particular scales, and the available data to allow a valid model application fulfilling the model's requirements. In the case of data availability it has to be questioned if the available data was measured at an appropriate scale to represent a natural process or property to fulfill the requirements and detail for modeling and decision-making. A separation of this sequence of scaling steps allows focusing on a particular scaling step in transforming spatial and temporal information in terms of dependency and requirements. The user's scales of interests have the highest priority in designing the overall assessment approach to meet their needs for decision support. These scales define the group of models capable of supporting this goal. Depending on the availability of appropriate model input data the final choice for one or a group of models can be made. Thus, the user's scales of interest, the model scale and the data scale available build the foundation to start the design of an environmental assessment approach. The following case study discusses the transformation of topographical data for each of the scaling steps within the sequence to predict spatially distributed sediment balance processes to support practical decision-making. The parameters on the weather, land use and soil are assumed to be independent from the topography.

Commonly available topographical data

One of the most fundamental information for landscape processes is the representation of topography. The U.S. Geological Survey (USGS) provides a nationwide set in 7.5 min-quadrangles through their publicly accessible data server (USGS 1999). Topographic information can also be derived by using contour lines of paper-based topographic maps, digital raster graphs (DRG) or occasionally available hypsographic Digital Line Graphs (DLG). DLG contour lines at the 1:24,000 scale are in 10 ft (~3.0 m) or 20 ft (~6.1 m) steps mapped. The U.S. National Map Accuracy Standard allows not more than 10% of randomly tested elevation points in a topographic map with errors of more than 1.5 times the distance between contours (U.S. Bureau of the Budget 1947). Digital Elevation Models (DEM) at the 1: 24,000 map scale are available nationwide with a 30 m pixel size and currently partially available at a 10 m pixel size scale. USGS DEMs have variable resolution and accuracy depending on the method used to derive them. Level 1 DEMs are derived from high altitude photogrammetry with a vertical resolution of 1m and an average vertical accuracy of 7 m (USGS 1995). Level 2 DEMs are interpolated from DLGs with a vertical accuracy depending on the distance between contour lines and a bias through the chosen interpolation algorithm.

Spatial accuracy and model efficiency

In a comparison between raster data sets the absolute error is the total difference between the 'true' or observed value (O), and the 'representative' or predicted model value (P) for a certain raster pixel. In contrast to the one-dimensional approach for Model Efficiency (ME) (Nash and Sutcliffe 1970) as a measure for the performance of a series of model results in comparison to observed values, a new filter - is suggested to evaluate spatially distributed ME:

(Notation 1)

MEFV_x,y is the Model Efficiency Filter Value for a central pixel with x,y-coordinates of a n*m-pixel-sized filter area, O and P are the observed and predicted values at the i,j-coordinates within this filter area, and Ô is the mean of all observed values in this filter area. The MEFV is recommended as a filter with n=m=7 or larger to assure a sufficient number of samples. MEFV can range from negative infinity to 1. The closer the value is to 1 (or 100%), the better is the model representation. Negative MEFV indicate that the fit within this point's neighboring area is unacceptable poor and the average of the point's observed values of the neighboring area is an better estimator than the model. An acceptable threshold given by a decision-maker will indicate an distinct answer in accepting or rejecting the results in an area rather than obtaining absolute values.

Results for soil erosion assessment approach

The soil erosion assessment approach was applied for the Agricultural Research Service (ARS) experimental watersheds W-2, W-1, and W-11 at Treynor, Iowa. The assessment uses detailed observed climate data from 1985-1990 to simulate continuous corn with conventional tillage on a silt loam soil series (Cochrane and Flanagan 1999, Flanagan et al. 2000).

Pre-processing commonly available topographical data

The following available topographical data sources with supposedly decreasing accuracy in elevation data were used: (1) a detailed point measurement in the format of a Triangular Irregular Network (TIN) from ARS aerial photogrammetry, (2) USGS 10 ft-contour lines from DLG and (3) USGS 10 m-raster DEM Level 2. All data were converted into a 10 m-raster through interpolation (TOPOGRID procedure in Arc/Info). The analysis is based on a pixel size of 100 m² and a MEFV area of 0.49 ha that may represent a scale land use decision-makers are interested in. Note that only those watershed areas covered by all three data layers were considered in the analysis. To evaluate the effect of different elevation data sources and model results depending on them, data layers were compared with each other by considering the one with the supposedly higher accuracy as the true 'observed'. The correlation coefficients show a close similarity between all three data sources (Figure 2a). The visualization of the spatial distribution of the absolute error with an acceptance range of 1 m indicates differences for some hillslope areas. The MEFV indicate that all three watersheds are accepted with a 99% tolerance.

Absolute error Decision area: 100 m2 (1 pixel)       Model Efficiency Filter Value Decision area: 0.49 ha (7*7 pixels)       a) Elevation: r2 = 0.989 r2 = 0.989 r2 = 0.990             AE < 1 m     MEFV > 99% b) Slope: r2 = 0.761 r2 = 0.698 r2 = 0.806             AE < 2.5%     MEFV > 75% c) Detachment: r2 = 0.542 r2 = 0.498 r2 = 0.578             AE < 5 t ha-1 yr-1     MEFV > 0%   TIN-CNT TIN-RAS CNT-RAS       TIN-CNT TIN-RAS CNT-RAS

Figure 2. Absolute error and model efficiency filter value to evaluate effect of different data sources on topographic a) representation, b) model parameterization and model results.

Topographical discretization for model input

The commonly available Topography Analysis Software System (TOPAZ) (Garbrecht and Martz 2000) was used for the watershed discretization and flowpaths draining into channels. To achieve detailed spatially distributed results, the flowpath method was used to simulate all possible flow paths (Cochrane and Flanagan 1999). The method considers topographical information of all possible flowpaths on a certain hillslope as a function of width, length and complex slope shape. The hillslope topography depends on slopes in the flow direction along a flowpath for hillslope shape discretization and critical source areas (CSA) for initiation of a channel and watershed discretization. Slopes in the flow direction derived by TOPAZ, based on different data sources, indicated that the correlation between CNT and TIN is better than RAS and TIN (Figure 2b). The two commonly available sources RAS and TIN show the highest agreement in correlation and MEFV efficiency, but the distributed pattern of accepted areas between contour and TIN data is more consistent. The MEFV excludes some hillslope areas not represented with a 75% tolerance limit. Considering that slopes measurements based on a 10 m-grid in the field is very labor intensive, a decision-maker may rely on the common data sources in comparison to the detailed TIN data which is expensive to obtain.

Model application and mapping of results

Correlation coefficients of model output layers of long-term average annual erosion rate predictions with the WEPP flowpath method show differences between the topographical data sources (Figure 2c). The correlation between the common available CNT and RAS is higher than between CNT and TIN, while RAS and TIN have the lowest correlation coefficients. The spatial distribution of absolute and relative error show no large differences in the acceptance pattern if the decision-maker is willing to set this fairly high acceptance limits (Figure 2c). While absolute error patterns are equally distributed, the MEFV value shows a pattern concluding that RAS and CNT data are very similar in their performance in most of the areas due to the same base of data they were created from (Figure 2c). The low MEFV acceptance limit indicates that model performance was not very comparable for many areas when using the RAS or CNT data, but in an acceptable manner for the up-slope areas. These are the areas where surface runoff is initiated and the right management decision has a major impact on the mid- and downhill soil loss.

Post-processing and assessment evaluation

The ultimate test for a model application or assessment is the comparison of simulated with observed values. The comparison for 10 m-raster discretizations of the three different data sources TIN, CNT and RAS shows that not every observed event was actually simulated and vice versa (Table 1). The total of 70 common dates showed that event-based runoff discharges have acceptable correlation coefficients between observed and simulated runoff discharges and sediment yields for all data sets. The model efficiencies for sediment yields indicate acceptable correlation coefficients, but unacceptable model efficiencies. The flowpath method does not consider erosion processes in channels and therefore the comparison between observed sediment yields at the outlet and sediment yield (sediment delivery) from all hillslopes is better represented through correllation than model efficiency. Nevertheless, the correlation coefficients and model efficiencies between different topographical data sources indicate no major difference in event-based runoff discharge predictions. The correlation between observed and simulated sediment yields are also comparably high, while model efficiencies between commonly available data sources and detailed topographical data sets are lower, but still in an acceptable range.

Discussion

User-friendly and easy-to-use approaches with visual effects considering spatial uncertainties are very important to increase the use of spatially distributed process models. The spatially distributed analysis presented here showed that model analysis needs to undergo a critical reality check in evaluating data accuracy and uncertainty in assessment results. But one has to remember that if physical data from a large set of observations of 15 replicate erosion plots for individual cases produces correlation coefficients of 0.76, it should not be expected that an erosion model can perform better than that (Nearing 2000). In addition to absolute error or coefficients of correlation, the Model Efficiency Filter Value (MEFV) is a helpful method to understand and estimate spatially distributed data uncertainties of different data sources. In future model approaches, the user needs to be provided with such methods and tools to visualize data, pre-processed model input data as well as output results with the expected effect of combined uncertainties on the model predictions. The uncertainty assessment has to be tailor-made for a specific GIS-model framework and has to be capable of handling a large variety of user requests and commonly available environmental data input. Uncertainties of model results are important in addressing risk assessment, policy change and in developing strategies to reduce risks.

Conclusion

Process-based models represent our most detailed scientific knowledge at smaller spatial and temporal scales, but are limited by extensive data requirements. Empirical models are more applicable for large scale problems and where less data is available, but do not take advantage of our understanding of process mechanics. Increasingly attempts are being made to scale process models to address larger scale issues and make use of more generalized data. The research described here follows these trends in describing methods to expand the use of process-based models. The goal of using process-based prediction models to support decision-making for the management of natural resources cannot be met with models that can only be run with a level and quality of data that is only found for research sites. Commonly available data sources are widely available, but rarely precisely fit the requirements for applying sophisticated process-based models. Solving environmental problems associated with degradation of natural resources such as water and soil, often cannot be delayed by the time and expense of additional data collection to satisfy model-input requirements. Thus assessment tools with a minimum of calibration requirements should be evaluated based on their potential to provide sufficient information for decision making based solely on current commonly available data sources. Therefore this research contributes to the overall goal in evaluating quality and usability of commonly available information for practical decision-making.

Recommendations for implementing assessment tools

There is considerable need for more practical approaches to assess the risk associated with certain levels of data accuracy including methods to generate, interpret and use these risks in decision-making. The strategies below should be followed to successfully implement useful, process-based environmental assessment tools for natural resources management.

1. Development, application and implementation of environmental assessment tools involve scientists, engineers as well as practical specialists from a wide variety of backgrounds. An interdisciplinary assessment approach requires the willingness of scientists and engineers to open their field of expertise to allow the exchange of ideas and methods.
2. There has to be an effort in describing uncertainty related to the actual existing information available in an appropriate and effective way in interpreting model and assessment results. An assessment tool has to make users aware of the assessment uncertainty and reliability as well as their responsibility in making a certain decison.
3. The assessment should also be supported by educational tools encouraging a user's basic qualitative understanding of the process pattern. Additional effort has to be undertaken in educating the users in understanding basic concepts of the problem, defining potential solutions and the appropriate application of an assessment approach.

Acknowledgements

The National Soil Erosion Research Laboratory of the United States Dept. of Agriculture - Agricultural Research Service and the Dept. of Agricultural and Biological Engineering at Purdue University, West Lafayette, Indiana, are gratefully acknowledged for the support of this research. The authors want to thank especially Bernd Diekkrüger from the Geographical Institutes at the University of Bonn, Germany, and Jon Harbor from the Geomorphology Group at Purdue University, Indiana, for their advice and insightful comments provided to reach the goals of the project.

References used

Ascought II JC, Baffaut C, Nearing MA, Liu, BY. 1997. The WEPP Watershed Model: I. Hydrology and erosion. Transactions of the American Society of Agricultural Engineers 40 (4): 921-933.

Boardman J. 1998. An average soil erosion rate for Europe: Myth or reality? Journal of Soil and Water Conservation 53(1): 46-50.

Cochrane T, Flanagan DC. 1999. Assessing water erosion in small watersheds using WEPP with GIS and digital elevation models. Journal of Soil and Water Conservation 54(4): 678-685.

Favis-Mortlock D. 1998. Validation of field scale soil erosion models using common datasets. In: Boardman J, Favis-Mortlock D, editors. Modelling Soil Erosion by Water. NATO ASI Series I., Vol. 55. p 89-127.

Flanagan DC, Nearing MA, editors. 1995. USDA-Water Erosion Prediction Project Hillslope Profile and Watershed Model Documentation. NSERL Report No. 10, USDA-ARS National Soil Erosion Research Laboratory, West Lafayette, Indiana. (See also: <http://topsoil.nserl.purdue.edu/nserlweb/weppmain/wepp.html>. Accessed 2000 June 29.)

Flanagan DC, Renschler CS, Cochrane TA. 2000. Application of the WEPP model with digital geographic information. In this issue: Problems, Prospects and Research Needs. Proceedings of the 4th International Conference on Integrating GIS and Environmental Modeling (GIS/EM4); 2000 Sep 2-8; Banff, Alberta, Canada.

Garbrecht J, Martz LW. 2000. TOPAZ: An automated digital landscape analysis tool for topographic evaluation, drainage identification, watershed segmentation and subcatchment parameterization: Overview. ARS-NAWQL 95-1, USDA-ARS, Durant, Oklahoma. <http://grl.ars.usda.gov/topaz/TOPAZ1.HTM> Accessed 2000 June 289.

Nash JE, Sutcliffe JV. 1970. River flow forecasting through conceptual models Part I - A discussion of principles. Journal of Hydrology 10: 282-290.

Nearing MA. 2000. Evaluating Soil Erosion Models using Measured Plot Data: Accounting for Variability in the Data. Earth Surface Processes and Landforms. (accepted).

NRCS. 1999. National Resources Inventory - 1997 State of the Land Update. <http://www.nhq.nrcs.usda.gov/NRI/. Accessed 1999 May 23.

Pimentel D, Harvey C, Resusodarmo P, Sinclair K, Kurz D, McNair M, Crist S, Shpritz L, Fitton L, Saffouri R, Blair R. 1995. Environmental and economic costs of soil erosion and conservation benefits. Science 267: 1117-1123.

Renard KG, Foster GR, Weesies GA, McCool DK, Yoder DC. 1997. Predicting Rainfall Erosion Losses - A Guide to Conservation Planning with the Revised Universal Soil Loss Equation (RUSLE). U.S. Dept. of Agriculture, Agricultural Handbook 703. 404 p.

Trimble SW. 1999. Decreased Rates of Alluvial Sediment Storage in the Coon Creek Basin, Wisconsin. 1975-93. Science 185: 1244-1246.

Troeh FR, Hobbs JA, Donahue RL. 1999. Soil and water conservation: productivity and environmental protection. Third Edition, Prentice-Hall Publishers. Upper Saddle River, New Jersey. 610 p.

U.S. Bureau of the Budget. 1947. United States National Map Accuracy Standards. Revised Version. 1 p.

USGS. 1999. U.S. Geodata Server. <http://edc.usgs.gov/doc/edchome/ndcdb/ndcdb.html>. Accessed 1999 May 10.

USGS. 1995. Standards for Digital Elevation Models - Part 2: Specifications. U.S. Dept. of the Interior., U.S. Geological Survey. 70 p.

Wischmeier WH. 1976. Use and misuse of the universal soil loss equation. Journal of Soil and Water Conservation 31 (1): 5-9.

Wischmeier WH, Smith DD. 1978. Predicting Rainfall Erosion Losses - A Guide to Conservation Planning. U.S. Dept. of Agriculture, Agricultural Handbook 537. 85 p.

Authors

Chris S. Renschler, Assistant Professor, University at Buffalo - The State
University of New York, Department of Geography, 116 Wilkeson Quad, Buffalo,
New York, 14261, USA.
Email: rensch@buffalo.edu, Tel: +1-716-645-2722 ext. 23, Fax:+1-716-645-2329.

Bernard A.Engel, Professor, Dept. Agricultural and Biological Engineering
Purdue University, 1146 ABE Building, West Lafayette, Indiana IN 47907-1146, USA.
Email: engelb@ecn.purdue.edu, Tel: +1-765-494-1162, Fax: +1-765-496-115.

Dennis C. Flanagan, Agricultural Engineer, USDA-Agricultural Research Service
National Soil Erosion Research Laboratory, 1196 Building SOIL, West Lafayette, Indiana, 47907-1196 USA.
Email: flanagan@purdue.edu, Tel: +1-765-494-7748, Fax: +1-765-494-5948.