Assessing water erosion in small watersheds using WEPP with GS and digital elevation models TA. Cocbrane and D. C. Flanagan ABSTRACT Three different approaches using geographical information systems (Gs) and digital elevation models (DEMs) are described and evaluated for applying the Water Erosion Prediction Project (WEPP) model to assess water erosion in small watersheds. The first approach describes a typical application of the watershed version of WEPP using Gs only as an aidfor construction of required input files. The second approach presents an automated method for the application of WPP through the extraction of hillslopes and channels fiom DEMs. The third approach uses WPP model simulations on all possible flowpaths within a watershed. The three method were applied to six research watersheds: one fiom Trqnor, la., two fiom Watkinsville, Ga., and three fiom Holb Springs, Miss. A statistical analysis for all methods and watershed compared the predicted us. measured runoff and sediment yiem fiom watershed outlets on an event- by-event basis for runoff and sediment loss. The results indicate that the automatic hillslope method (the second approach) peforms as well as the manual technique (the first approach) for all watershed. A comparison of erosion from onb hillslopes for all three methods indicates that the fiwpath method (the third approach) is statistically comparable to the other methods. Results of the analysis suggest that, given an accurate DEM and valid input aha for a simple watershed, the automatic hillslope method can be used to facilitate the application of the watershed version of WEPR and that predictions shoum be comparable to an expert user; application of WEP? Kq words: DEM, flowpaths, Geographic nformation System, hydrology, modeling, run08 sediment. soil erosion, Water Erosion Prediction Project. W atershed erosion is a cause of nonpoint source pollution that can have an adverse effect on downstream and ecosystem water quality. Unfortunately, it is very expensive and impractical to monitor erosion in all watersheds of interest, hence, the need to predict erosion with modeling. The Water Erosion Prediction Project (WEPP) model is a process-based continuous simulation erosion model (Flanagan and Nearing 1995). One advantage WEPP has over existing models, such as the popular Universal Soil Loss Equation (USLE), is that deposition of sediment also can be predicted (Wischmeier and Smith 1978). n other words, soil loss on a complete continuous hillslope profile can be calculated. The WEPP watershed model is an extension of the WEPP hillslope model that can be used to estimate watershed runoff and sediment yield (Ascough et al. 1997). The application of WEPP to a watershed requires that hillslopes be 4-1. GS Maps..... -... -... 12a. Manual method < delineated and channels identified (Baffaut et al. 1997). Each hillslope, represented as a rectangle in WEPP, must have a representative length, width, and slope profile as shown in Figure 1, part 3. Hillslopes drain into the top, left side, or right side of a channel, eventually leading to the watershed outlet. ntegration of WEPP with geographical information systems (GS) is desirable to facilitate and possibly improve application of the model. Savabi et al. (1995) conducted an initial application of WEPP on the Purdue University animal science watershed with a Gs. n this study, they used the GRASS GS (U.S. Army 1987) to obtain some of the physical parameters required by WEPP. However, they did not address the discretization of hillslopes, channels, and representative slope profiles from digital elevation models (DEMs). As illustrated in Figure 1, GS maps can be used to identify the watershed and extract hillslopes and channels. When dealing with small watersheds and relatively uniform soils and management practices, the primary map used to delineate hillslopes and channels is a topography map. Topography in GS usually is represented by a DEM or can be converted to a DEM from a triangular irregular network (TN). The most common type of DEM is grid-based, where each grid point represents a cell of a certain size or resolution. Extraction of features, such as hillslopes and channels, Manual discretization of \ ' hillslopes, channels, and,) sentative slope p r o f w G),yiEFGiq ~~~ ~ Thomas A. Corhrane is a graduate research assistant at Purdue University, and Dennis C. Hanagan is an agricultural engineer with the U.Si)A Agricultural Research Service (USDA- ARS) h'ationui Soil Erosion Research Laboratory, West Lafhyette, nd. Thgv thank scientists lit tlle National Soil Tilth Laboratory, esp e cia 1 ji Mi L' ha e R u r ka r t, Da v id /a rn es, Larty Kramer,,ind ]O/J? Layen, for providing valuable information and the digital elevation models (DEM,fir the Trrynor Watershed. The use oj-trade numes does not imply endorsement by Purdut> Uniiwxip or tbe USDA-ARS. DEM Soils ManagemdCrops 2b. Hillslope method Figure 1. Watershed modeling with GS and WEPP. / Automatic discretization of\ hillslopes. channels, and w t i v e slope2iv "
can be performed based on flow-routing algorithms that determine the steepest descent direction and gradient between cells. Examples of flow-routing algorithms are presented in the research of O Callaghan and Mark (1 984), Jenson and Domingue (1988), Martz and Garbrecht (1992), and Zevenbergen and Thorne (1 987). Some of these algorithms have been used to help integrate erosion models, such as the USLE, with GS (Desmet and Govers 1996). Other models that use flow-routing algorithms and GS maps to predict erosion or runoff include ANSWERS, AGNPS, and SWAT (Srinivasan and Arnold 1994). GS analysis using DEMs is an obvious tool for parameterization of hillslopes, channels, and representative slope profiles for WEPP simulations. The purpose of this study was to describe and evaluate three methods of integrating WEPP and Gs. These methods will facilitate the application of WEPP with available GS data; results then can be compared to each other and to observed runoff and sediment loss from monitored watersheds. Materials and methods Three methods have been developed to apply WEPP using Gs: manual, hillslope, and flowpath methods. All three techniques take advantage of tools provided in the Arc ViewTM 3.0 GS and its Spatial AnalystTM extension program (ESR 1998). Additional algorithms have been programmed in FORTRAN and Arc View s AvenueTM script language. The hillslope and flowpath methods also use the TOPographic evaluation, drainage identification, watershed segmentation and s u b c a t c h m e n t p aram e t e r i Za t i on (TOPAZ) automated digital landscape analysis tool (Garbrecht and Martz 1997). The three methods were evaluated using observed data from ARS experimental watersheds at Treynor, a. (W2), Holly Springs, Miss. (WC1, WC2 and WC3), and Watkinsville, Ga. (P andp2). Manlcal method. This method consists of using Arc ViewTM and Spatial AnalystTM to set up and describe watershed components (e.g., hillslopes and channels) for a WEPP watershed model application by using tools, such as graphics drawing and on-screen digitizing, with the computer s mouse. The advantage of this method is apparent only if the user has a DEM, as well as soil and management data represented in GS maps. 1 Watershed Figure 2. Flowpaths in watershed modeling. First, the user identifies channels in the watershed. Their location can either be represented by on-screen digitizing or automatically extracted from the DEM using Spatial AnalystTM or TOPAZ. Parameters such as width, shape, depth, and erodibility must be entered manually for each channel. Hillslopes are then defined by digitizing the hillslope boundaries on the watershed with on-screen digitizing tools available in Arc ViewTM. The user can divide the watershed into as many hillslopes and channels as permitted by the WEPP watershed model code. A representative profile for each hillslope is defined by drawing a line that represents the location of the profile. This line is overlaid on the DEM, using features in Spatial AnalystTM, to obtain an actual elevation profile. The length of the digitized profile line and the area of the hillslope are then used to calculate a width of the hillslope. Soils, management practices, and crops maps can be used to define the soil and management properties used for each hillslope. Technically, this method is the same as manually preparing the WEPP model from paper maps; however, it saves time in defining the components of a watershed and it allows the user to study several configurations more rapidly. t also allows the modeling of special situations and for indispensable human judgment (intervention) in defining hillslopes and channels. Hillslope method. The hillslope method consists of automatically defining the watershed components (hillslopes and channels). Hillslopes and representative slope profiles are extracted by using a DEM of the watershed and algorithms that simulate the flow of water between cells in the DEM. The coding for the hillslope and profile creation algorithms was done in the FORTRAN programming language; the TOPAZ program was used for the extraction of routing features in the watershed. Although GS programs (e.g., Arc ViewTM) have add-on programs, such as Spatial AnalystTM, that can handle grid-based flow routing. TOPAZ was chosen for its ability to overcome limitations of previous algorithms with respect to drainage identification in depressions and over flat surfaces (Garbrecht er al. 1996). Channel location and lengths are initially defined by selecting a critical source area (CSA). The CSA represents a drainage area whose concentrated water flow defines the beginning of a channel (Garbrecht and Martz 1997). When using a DEM, the CSA represents a certain number of cells flowing into one single cell (defined as the starting point of a channel). Correct identification of channels may be verified by aerial photography or field surveys. The critical source area for the two Watkinsville watersheds was 0.5 ha (1.2 ac). The CSA for the WC1 watershed (Holly Springs) also was 0.5 ha, but for WC2 and WC3 (also FOURTH QUARTER 1999 679
Holly Springs), the CSA was 0.25 ha (0.62 ac). The CSA value for the Treynor watershed was 4 ha (9.9 ac). For these watersheds, there seemed to be a direct correlation between the size of the watershed and the CSA value representing the watershed channels. However, the exact point of channel initiation can be influenced by other factors, such as ground slope, soil, management, and climatic factors (Montgomery and Dietrich 1989; Martz and Garbrecht 1992). Other WEPP channel input parameters, such as actual width, shape, depth, and erodibility, must be provided by the user. Hillslopes are defined as draining to the left, right, and top of the channels created from the [>EM (Figure 1). A representative slope profile is obtained by weighting all possible flowpaths in the hillslopes. Flowpaths are defined as the route water travels flowing from one cell to the next, starting from a cell where no other cells flow into it and ending at the point where a channel is reached (Figure 2). Each hillslope may have a large number of flowpaths, some that start at the edge of the watershed and others that start at points inside the hillslope. Many flowpaths may eventually intersect as they reach the channel. ndividual flowpaths on hillslopes were extracted by analyzing output files of the TOPAZ program. A representative slope profile for each hillslope was computed from all the individual flowpaths within the hillslope. This was done by weighting flowpaths according to their area and length, then comparing each cell length of a flowpath to all other matching cell lengths from flowpaths that start from the channel and move up the hillslope. t was assumed that flowpaths with greater area and longer lengths contributed proportionally more to the representative profile than the smaller and shorter flowpaths. Because cells are square, flowpaths that tnove diagonally across cells are longer than flowpaths that reach an equal number of cells, but move horizontally or vertically. This is why both the area (number of cells) and flowpath length were considered in the slope profile. The following equation illustrates how a representative slope profile was computed for each hillslope: p - 1 E, =' [1 Where Ei is the weighted slope value for all flowpaths at a distance i from the channel; zpi is slope of flowpath p at distance i from channel; and k is a weighting factor for flowpath p. Tke weighting was done by multiplying the upstream drainage area (ai, area of cells in the flowpath) by the flowpath length (ki=ai*li). Because this equation calculates a profile with a length equal to the longest flowpath, it also is necessary to determine the appropriate length of the profile. For hillslopes draining laterally into channels, the width of the hillslope is set equal to the length of the channel. The length of the hillslope is then calculated by dividing the total hillslope area by the width. The representative profile is then truncated so that it is equal to the calculated length, starting at the bottom of the hillslope. The length of hillslopes that drain into the top of the channel were determined by a similar method of weighting all flowpaths. This is illustrated by the following equation, presented in Garbrecht et al. ( 1996): E ar p= 1 Where L is the hillslope length;, is the flowpath length; ap is the area represented by the cells in the flowpath, and n is the number of flowpaths in the hillslope. The width of the representative hillslope profile is then easily calculated by dividing the total hillslope area by its length. The two main assumptions for this case are that the flowpaths are the routes traveled by water, and that larger and longer flowpaths contribute more than smaller and shorter flowpaths (Garbrecht et al. 1996). t is important to note that additional variations of this method may create slightly different representative hillslope profiles that have not been studied. These include using average values instead of weighted values, using different flowrouting algorithms, and others. However, it is believed that the method presented in this paper is the most theoretically sound and is an adequate representative of all these possibilities. Flowpath method. The third method of applying WEPP to a watershed using GS and DEMs is the flowpath method. n this procedure, WEPP is applied to all possible flowpaths in the watershed (Figure 2). The problem of interaction between flowpaths is handled in a unique way. All points where flowpaths drain into a channel are identified. Then, for every one of these points, an average sediment and runoff discharging into the channel is calculated. For example, Figure.3 shows three flowpaths (1, 2, and 3). At point C, flowpaths 2 and 3 intersect; at point B, all three flowpaths intersect. Point A shows the location where all runoff and sediment is discharged to the channel from the three flowpaths. nitially, sediment loss and runoff from each rainfall event is calculated for each flowpath independently by applying WEPP to the actual slope profile of the flowpath. The width used for the WEPP application of the individual flowpaths is calculated by dividing the total area of all three flowpaths (represented in yellow in Figure 3) by the length of the individual flowpath. The results of the WEPP application to each of the three flowpaths are averaged to obtain sediment and runoff discharge into the channel at point A. The maximum number of points of discharge to the channels is the number of cells that make up the channel multiplied by 2 (left and right sides). For each of these points, a channel segment could be defined to route the water to the outlet of the channel. However, since this would be done on a cell-by-cell basis down the channel, the current channel processes used in WEPP may not correctly handle this situation. Another alternative would be to add the contribution for all points Figure 3. ntersecting flowpaths draining into channel in watershed.
draining laterally in the channel. This alternative would require modification of the current WEPP code to handle channel routing with unevenly distributed lateral flow to the channel. Since it was not the intention of the authors to change the WEPP code, these modifications were not further pursued. This implies that a comparison of the flowpath method to observed soil loss results from the watershed outlets would only be appropriate if no scouring or deposition occurred in the channels. However, erosion from hillslopes can be compared between all methods. For visual assessments of water erosion from small watersheds, detachment or deposition values for each grid cell within a watershed can be estimated by using the flowpath method in a slightly different manner. n this case, the WEPP hillslope model is applied to each flowpath using the flowpath's length, width, and slope profile. Values of detachment and deposition for each cell along each flowpath are then extracted from the WEPP output. nteractions between flowpaths are treated by weighting the results for each cell along intersecting flowpaths based on length and drainage area of the flowpaths. Combining all flowpaths creates a visual representation of the watershed with detachment or deposition values for each grid cell. Application The three methods were applied to six research watersheds. The years of simulation, watershed area, and soil and crop types are presented in Table 1. The WEPP data files for the Watkinsville and Holly Springs watersheds are available via the nternet at: http://topsoil.nserl.purdue. edu/ wep p main/ we p p. h t ml. Additional information and measured data from the Watkinsville watersheds were provided by Smith et al. (1 978). Descriptions of the Holly Springs watersheds were provided by McGregor et al. (1969). Validation of the WEPP model for these watersheds was performed by Liu et al. (1997). These watershed applications of WEPP were used as templates for the simulations with the manual method. Management and climate data used for the Treynor watershed simulation was obtained from a previous application of WEPP to this watershed by Kramer (1 993). However, the optimized soils and channel parameters used by Kramer (1993) were found to be uncharacteristic for that region and Table 1. Research watersheds used for validation of methods. Watershed (location) P1 (Watkinsville, Ga.) P2 (Watkinsville, Ga.) WCl (Holly Springs, Miss.) wc2 (Holly Springs, Miss.) wc3 (Holly Springs, Miss.) Years of Area Soils simulation (ha) Crops 29 Monona-da-Napier Corn (Treynor, w2 la.) 1985-90 series (Silt loam) 2 1.8 h 3 1.6 aj aj 5 1.4 c 2 cn 1.2-0.d c 1.- E 3 0.8 cn W 0.6.- W 0.4 0.2 1972-75 2.70 Cecil (Sandy loam Wheat, sorghum, and SCL) barley, soybean, clover 1973-75 1.29 Cecil (Sandy loam Corn, bermuda and SCL) grass 1970-77 1.57 Grenada (Silt loam) Soybean, meadow 1970-77 0.59 Grenada (Silt loam) Corn, wheat, soybean 1970-77 0.65 Grenada (Silt loam) Corn, wheat, soybean hillslope method - - Linear (manual method) y=a79x+027 - ~ - ---- ~ --.-- - " ~ -. - 0 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 Observed sediment loss (tlhalevent) 1 /',', ' y=o73x+u22 R- = o 78 Figure 4. sediment loss from each watershed (normalized by watershed area and nu m be r of events). those conditions. Hence, default soil parameters from the WEPP soils database were used for the simulations, and appropriate channel parameters were chosen to represent the permanent grass waterways. The three methods were applied using WEPP Version 98.4. For these watersheds, each hillslope was represented by one soil type and crop/management description. A 1-m resolution was used for the Watkinsville and Holly Springs watersheds, and a 5-m resolution was used for the Treynor watershed. DEMs from Watkinsville and Holly Springs were created from field survey contour maps (0- and 1.524-m contours, respectively,) which were determined to be precise enough for a 1-m grid. Vertical accuracy of these maps is believed to comply with U.S. Geological Survey (USGS) map accuracy standards, which states that not more that 10% of points tested shall be in error by more than one-half the contour interval. The Treynor watershed was aerial surveyed with ground controls to create a DEM. Even though a smaller resolution could be used, the quality of the data and the size of the Treynor watershed suggested that a 5-m grid was appropriate for this modeling exercise. Analyses All watershed simulations were conducted with WEPP running in a continuous mode. The WEPP event-by-event **/ FOURTH QUARTER 999 681
output from each simulation was compared to runoff and sediment loss for each measured event. runoff and sediment loss from the watershed outlet were defined to be the sum of all events where measurements were not zero. For comparisons between methods for hillslope erosion, WEPP total simulation results were used. Several analyses were conducted on all six watersheds to compare the different methods using predicted vs. measured results. Event-by-event predictions are important for assessing a model's ability to predict the accurate distribution of events over time. runoff and sediment loss predictions are important for assessing the long-term effect of watershed soil loss. This can include the effects of sedimentation and water quality in downstream reservoirs and streams, and associated watershed conservation planning. Figure 4 shows the results of a comparison between manual and hillslope methods. The predicted vs. measured runoff and sediment yield were plotted using a best-fit regression line as an indicator of fir. The coefficient of determination, r', indicates variance from the best-fit line. The slope and intercept of each regression line show the bias between predicted and measured data for the two methods. Tabular data with statistical analysis of all simulations are presented in Table 3 for runoff, and Table 5 for sediment loss. Tabular data showing hillslope and channel erosion are provided in Table 6. 'T h e N a s h - S u t c i ffe C o e ffi c i e n t o f model efficiency, NS, also was used as a statistical criterion for evaluating hydrologic goodness ot tit between measured and predicted values for each method. This coefficient is calculated as follows (Nash and Sutcliffe 1970):.i Where Q are the ne'iwred values (e.g., sediment leaving wcitershed) on an eventby-went basi \. Q' &ire iiiodel-predicted values on an even t-by-even t bdsis, Q is the 'iverdge of measured values (average of dl1 events), and? i4 the number of values. An NS valuc of indicarcs a perfect fit between Table 2. Nash-Sutcliffe model efficiency coefficients for event runoff predicted by the three methods. Watkinsville, P1 Watershed Watkinsville, P2 Holly Springs, WC1 Holly Springs, WC2 Holly Springs, WC3 Treynor, W2 measured and predicted values for all events. A value of 0 indicates that the fit is as good as using the average value of all the measured data for each event. The NS values for each watershed are presented in Tables 2 and 4 for runoff and sediment loss, respectively. For long-term comparisons, predicted vs. measured total runoff or sediment yield normalized by number of events were plotted together with a regression line fit. Another criterion for long-term assessments of predicted vs. measured data is the percent deviation of runoff volume, D,,. This is computed with the following equation: Where Vis the measured runoff volume, and V" is the predicted runoff volume. The percent deviation of total sediment loss also can be computed in a similar fashion. These values are presented as part of Tables 3 and 5 for total runoff and sediment loss. Results and discussion Run08 Runoff values for all watersheds were comparable in all three methods. The predicted watershed outlet runoff from the manual and hillslope methods included contributions from both hillslopes and channels. Simulations using the flowpath method assumed that the runoff leaving the watershed was the summation of all runoff entering the channel at specific points along the channel. n other words, we assumed that the channel was merely a conduit to route the runoff from the hillslopes to the watershed outlet. The Nash-Sutcliffe model efficiency coefficient was calculated for all measured runoff events for each of the six watersheds (Table 2). Nash-Sutcliffe coefficient ~~ ~ Nash-Sutcliffe Coefflclent (NS) Manual Hillslope Flowpath 0.76 0.82 0.64 0.68 0.77 0.57 0.77 0.83 0.70 0.70 0.76 0.57 0.76 0.80 0.69 0.68 0.76 0.58 values ranged from 0.57 to 0.82, indicating that all three methods produced WE P P s i mu la t ions that sat is fac t o r i 1 y predicted runoff for events over time. There was little variation in Nash- Sutcliffe coefficient values between the manual, hillslope, and flowpath methods. The percent deviation of runoff volume also was calculated for all watersheds using the total runoff volume leaving each watershed for all measured events (Table 3). Comparisons of means between the three methods showed no statistical difference, but comparisons to the observed measurement showed statistical differences for the Holly Springs (WC 1 and WC2) watersheds. Differences between observed and WEPP-predicted runoff for the Holly Springs (WC2) watershed also were reported by Liu et al. (1 997). For the Holly Springs (WC) watershed, Liu et al. (1 997) did not find differences between the observed and predicted means for runoff, but only 237 selected events from the 284 observed runoff events were analyzed. The presence of some differences in predicted and observed runoff doesn't necessarily suggest that WEPP performs worse than other models, but is rather a reinforcement of the knowledge that erosion predictions in general contain large factors of error (Liu et al. 1997). Sediment yield. The Nash-Sutcliffe model efficiency coefficients for sediment loss for all watersheds ranged from -0.63 to 0.87 (Table 4). sediment loss and percent deviation are presented in Table 5. As expected, the flowpath method shows great variability in predicting sediment loss from the watersheds as a result of not simulating erosion in the channels as seen by low NS values and high percent deviations of some watersheds. This effect can either increase or decrease prediction rates, depending on whether the channel is erodible or acts as a sediment trap. Table 6 shows predicted hillslope erosion for the three methods and percent channel deposition for the manual and
Table 3. Runoff and percent deviations of measured events.* Runoff in m3, ( 7 0 Deviation from observed) Watkinsville, P1 36 measured events Watkinsville, P2 55 measured events Holly Springs, WC1 284 measured events Holly Springs, WC2 257 measured events Holly Springs, WC3 255 measured events Treynor, W2 40 measured events Observed Manual Hillslope Flowpath 13530 11050( 18) 12170(10) 11420(15) 376 307 338 317 482 406 466 430 29-2202 0-1457 0-1663 0-1557 6670 4620(31) 4730(29) 441 O(34) 121 84 86 80 228 169 184 169 16-1151 0-757 0-875 0-815 58780 34860(41) 41960(29) 40970(30) 207t 123 148 144 257 195 223 215 2-1538 0-1465 0-1674 0-1624 14760(35) 22900 89t 58 59 57 108 1-787 84 0-611 87 0-636 1 a4 0-61 6 17370 68 94 1-667 59620 1490 21 39 93-1 1031 14500( 17) 57 92 0-669 50880( 15) 1272 2578 0-11935 15130(13) 59 93 0-679 54580( 8) 1365 2665 0-12033 * s were tested using student t-test, Duncan s, and Tukey s comparisions of means, all of which gave the same decision results for hypothesis testing at a = 0.05. The means of the observed and methods are not statistically different at the 95% confidence level unless otherwise marked. t Observed and method means were statistically different at the 95% confidence level. Table 4. Nash-Sutcliffe model efficiency coefficients for event sediment loss predicted by the three methods. Nash-Sutcliffe coefficient (NS) Watkinsville, P1 Watkinsville, P2 HoUy Springs, WC1 Holly Springs, WC2 Holly Springs, WC3 Treynor, W2 Watersheds Manual Hillslope Flowpath 0.50 0.67 0.84 0.79 0.23 0.36 0.39 0.87 0.54 0.79 0.42 0.40 manual method P1 P2 wc 1 WCL wc3 w2 watershed 0.48 0.74-0.63 0.60 0.13 0.47 Figure 5. Comparison of hillslope sediment yield between methods (channels excluded). 1451 O( 16) 57 89 0-652 60790( 2) 1520 2584 0-11695 hillslope method. t is apparent that the three Holly Springs watersheds have channels that are in an advanced depositional mode, whereas, channels in the Watkinsville watersheds may not be in such a dramatic depositional mode. Comparisons of hillslope sediment yield between the three methods suggest that they predict similar results; however, a visual observation of Figure 5 suggests that predictions from the flowpath method are consistently lower than the hillslope method. This may be due to the large amount of computational weighting involved in the flowpath method. Sediment yield results are comparable between the manual and hillslope methods because channels were modeled in both methods. Figure 4 shows that both manual and hillslope methods performed similarly for total sediment loss in all watersheds. The regression lines indicate both a good fit and low variance for the manual and hillslope methods. The slope and intercept of the regression fit show that the hillslope method predicts slightly higher values than the manual method. The figure also shows that for lower observed sediment losses, both methods over-predict; whereas, for large sediment losses, the methods underpredict observed values. From the NS values and percent deviations of watershed FOURTH QUARTER 1999 683
Table 5. Sediment yield and percent deviations for measured events.* Watkinsville, P1 36 measured events Watkinsville, P2 48 measured events" Holly Springs, WC1 207 measured events" Holly Springs, WC2 199 measured events** Holly Springs, WC3 187 measured events** Treynor, W2 40 measured events Sediment yield in kg, (% Deviation from observed) Observed Manual Hillslope Flowpath 1 84490 51 25 10645 53-50000 28640 597 2051 0-1 281 0 104870 507 3803 3-53300 40150 202 1601 1-22152 29520 158 821 1-7476 1181 150 29529 70345 907-333800 137430(26) 3817 8779 0-41100 21690 (24) 452 1043 0-5860 108200 (3) 523 2830 0-35600 6261 O(56) 315 1415 0-16899 43900( 49) 235 1133 0-13672 1674140(42) 41853 92080 0-375753 161210(13) 4478 11314 0-54300 28500 (4) 594 1688 0-10651 21 7770( 1 16) 1052 5192 0-59900 64490 (60) 324 1486 0-17820 37630 (27) 201 929 0-10871 1568820(33) 39220 86491 0-363200 100060(46) 2779 6676 0-32600 17050 (40) 355 1167 0-7717 306270( 192) 1480 7805 0-89700 82930( 106) 417 2132 0-25646 43770 (48) 234 1170 0-13679 1469590 (24) 36740 83222 0-366100 s were tested using student t-test, Duncan's, and Tukey's comparisions of means, all of which gave the same decision results for hypothesis testing at = 0.05. The means of the observed and methods are not statistically different at the 95% confidence level unless otherwise marked. "Number of observed sediment yield events from watershed P2, WC1, WC2, and WC3 were less than measured runoff events for the same watersheds. Table 6. WEPP-predicted erosion from hillslopes and channels (kg).' t Manual method Hillslope method Watershed Hillslopes Channels %Channel Hillslopes 1 Channels 71 Flowpath name deposition deposition method P1 P2 wc1 wc2 wc3 w2 160900 37300 300600 136700 11 1000 2173100 156300 30600 164800 92800 64800 1738800 3 22 82 47 71 25 1 18500 35700 546300 146700 81 100 19461 00 182400 37500 3 1 6800 96000 55200 1635000 s of hillslopes for the three methods were tested using student t-test, Duncan's, and Tukey's comparisions of means, all of which gave the same decision results for hypothesis testing at a = 0.05. The means were not statistically different at the 95% confidence level. t values may be higher than in Table 5 because these include predicted values that were not measured. WC 1, the manual method seemed to perform best; for watersheds P2, WC3, and W2, the hillslope method appeared to be best. n other watersheds, both methods performed equally well. r is possible that the differences in the accuracy of the prediction5 for WC can be attributed to its coin pl icated topography and channel structure, which the manual method modeled in greater detail. The other watersheds are either smaller, such as WC2 and WC3, or have less topographical or channel structure complications. The Holly Springs watersheds show an over-prediction of sediment yield (Table 5). This is consistent with observations made by Liu et al. (19!17), who suggest that this is due to an over-prediction bias for low-ma~nitudt. erosion events, proba- bly from problems in representing plant and cover relationships. The channels of these watersheds also acted as sediment traps that, together with the over-prediction of small events, would help to explain large percent deviations of total sediment loss obtained. One of the biggest advantages of the flowpath method is that hillslope boundaries need no delineation, which removes the burden of defining hillslopes. Another advantage is the ability to obtain spatiallydistributed results of erosion or deposition within a watershed, as Figure 6 shows in the Treynor watershed (W2). These results can be visually displayed on an event-by-event or on a cumulative basis. The visualization approach in the flowpath method can help identify areas -35-5 72 53 47 19 107700 21 900 4491 00 122200 64800 1547200 of severe erosion within a watershed. However, caution should be used when interpreting quantitative results because spatially-distributed results within watersheds have not been validated (spatiallydistributed erosion measurements for watersheds over time are not available or are limited to small data sets). Application of the flowpath method has other limitations, too. One is its computational intensity. Due to the relatively high resolution of DEMS of all watersheds (1 m for smaller watersheds and 5 m for the large Treynor watershed), the number of flowpaths in each watershed also is very high. For example, the Watkinsville P2 watershed required 4,671 runs of the hillslope version of WEPP Other watersheds required fewer runs, but the number of
Detacirment (kgt/ma -6 - (-4) -4-1-31 rn m-3 - (-2) -2 - (-1) -1-0 l----10-1 Figure 6. Sediment detachment and deposition from flowpath application of WEPP on Treynor watershed W2 over a 6-yr period. Conclusions T h e manual and hillslope methods were comparable and produced reasonable results for most of the watersheds studied. Simulations of the Holly Springs (WC1 and WCZ) watersheds produced mean event runoff values that were not statistically similar to the observed means; however, there was no significant difference between t h e two prediction methods. No significant differences were observed between the measured sediment loss values and those predicted by the two methods. The Treynor and Watkinsville watersheds were well simulated with both methods for runoff and sediment loss. The main advantage of the hillslope method over the manual method is that major components (e.g., hillslopes, channels, and slope profiles), required for a WEPP model application, are extracted automatically from a DEM. This saves the user valuable time when assessing water erosion in watersheds and can allow the user to quickly assess conditions with different management practices. However, for topographically-complicated conditions or watersheds with complicated channel structures, the manual method has the flexibility to model these situations. The main advantages of the hillslope method over the flowpath method are that it is less computationally intensive and it can handle channel erosion. The flowpath method is comparable to both the manual and hillslope methods for hillslope erosion predictions. However, application of the flowpath method to watersheds requires development of an adequate channel-routing mechanism or modifications of the original WEPP code to enable distributed lateral inflow to channels. Currently, the advantages of this method over the others are that discretization of hillslopes is not necessary, and the flowpath method can more easily create visual representations of the results within the watershed than can other methods. deally, the best system would be to provide the user with a combination of the three methods. The hillslope method facilitates t h e application o f WEPP through an automated simulation. T h e manual method allows the user to verify t h e simulation a n d make p e r t i n e n t changes. T h e flowpath method can be used in situations where hillslopes are difficult to delineate, or when severe erosion or deposition locations must be identified inside the watershed. We believe that with the future availability of more accurate DEMs using advanced satellites, airborne laser technology, or high-accuracy GPS surveying, the advantages of using these methods will be apparent not only for research purposes, but for routine land management and erosion assessments. REFERENCES CTED Ascough, J., C. Baffaut, M.A. Nearing, and B.Y. Liu. 1997. The WEPP watershed model:. Hydrology and erosion. Transactions of the ASAE 40(4):921-933. Baffaut, C., M.A. Nearing, J.C. Ascough, and B.Y. Liu. 1997. The WEPP watershed model: 11. Sensitivity and discretization. Transactions of the ASAE 40(4):935-943. Desmet, P.J.J., and G. Govers. 1996. A GS procedure for automatically calculating the USLE-LS factor on topographically complex landscape FOURTH QUARTER 999 685 flowpaths still were several thousand. units. Journal of Soil and Water Conservation 5 1(5):427-433. ESFU. 1998. Arc View 3.0a Users Manual. Environmental Systems Research nstitute: Redlands CA. Flanagan, D.C., and M.A. Nearing. 1995. USDAWater Erosion Prediction Project: Hillslope profile and watershed model documentation. NSERL Report No. 10. USDA-ARS National Soil Erosion Research Laboratory: West Lafayette, N. Garbrecht J., and L.W. Martz. 1997. 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, OK. Garbrecht J., P.J. Starks, and L.W. Martz. 1996. New digit a1 1an ds c a p e par am e t e r i z R t i o n methodologies. n: Proceedings of the 32nd Annual Conference and Symposium on CS and Water Resources, TPS-96-3. American Water Resources Association: Herndon, VA. Jenson, S.K., and J.O. Dominique. 1988. Extracring topographic structure from digital elevation data for geographical information system analysis. Photogrammetric Engineering and Remote Sensing 54(11):1593-1600. Kramer, L. 1993. Application of WEPP 93.005 to HEL watershed. Presented a t the 1993 ASAE Meeting, Paper No. 93-2501. American Sociev of Agricultural Engineers: St. Joseph, M. Liu, B.Y., M.A. Nearing, C. Baffaut, and J.C. Ascough. 1997. The WEPP Watershed model: 111. Comparisons to measured data from small watersheds. Transactions of the ASAE 40(4):945-952. Martz, L.W., and J. Garbrecht. 1992. Numerical definition of drainage network and subcatchment areas from digital elevation models. Computers and Geoscience 18(6):747-761. McGregor, K.C., J.D. Greer, G.E. Gurley, and G.C. Bolton. 1969. Runoff and sediment production from North Mississippi loessial soils. Mississippi State University Agricultural Experiment Station Bulletin 777. Montgomery, D.R., and W.E. Dietrich. 198'). Source areas, drainage density, and channel initiation. Water Resource Research 25(8):1907-1918. Nash, J.E., and J.V. Sutcliffe. 1970. River flow forecasting through conceptual models, Part -A, discussion of principles. Journal of Hydrology 10(3):282-290. O'Callaghan, J.F., and D.M. Mark. 1984. The extraction of drainage networks from digital elevation data. Computer Vision, Graphics, and mage Processing 28:323-344. Savabi, M.R., D.C. Flanagan, B. Hebel, and B.A. Engel. 1995. Application of WEPP and CSGRASS to a small watershed in ndiana. Journal of Soil and Water Conservation 50(5):477-483. Smith, C.N., Bailey, G.W., Leonard, R.A., a n d G.W. Langdale, 1978. Transport of agricultural chemicals from small upland piedmont watersheds. U.S. Environmental Protection Agency Report 600/3-78-056. Srinivasan, R., and J.G. Arnold. 1994. ntegration of a basin-scale water quality model with CS. Water Resources Bulletin 30(3):453-462. U.S. Army. 1987. GRASS-GS software and reference manual. U.S. Army Corps of Engineers, Construction Engineering Research Laboratory: Champaign, L. Wischmeier, W.H., and D.D. Smith. 1978. Predicting rainfall erosion losses, a guide to conservation planning. U.S. Department of Agriculture, Agricultural Handbook No. 537. Zevenbergen, L.W., and C.R. Thorne. 1987. Quantitative analysis of land surface topography. Earth Surface Processes and Landforms 1247-56.