TMDL Data Requirements for Agricultural Watersheds

Transcription

1 This is not a peer-reviewed article. Pp in Total Maximum Daily Load (TMDL) Environmental Regulations: Proceedings of the March 11-13, 2002 Conference, (Fort Worth, Texas, USA) Publication Date March 11, ASAE Publication Number 701P0102, ed. Ali Saleh. TMDL Data Requirements for Agricultural Watersheds A.S. Cotter 1, I. Chaubey, T.A. Costello, M.A. Nelson, T. Soerens Abstract. This paper describes the effect of DEM data resolution on predictions from the SWAT model for total maximum daily load (TMDL) development. Measured hydrologic, meteorological, watershed characteristics, and water quality data from Lincoln Lake watershed (near Lincoln, AR) were used in the simulation. The effect of input data resolution was evaluated by running seven scenarios at increasing DEM grid sizes (30 x 30 m, 100 x 100 m, 150 x 150 m, 200 x 200 m, 300 x 300 m, 500 x 500 m, and 900 x 900 m). The model was calibrated on an annual basis for discharge, NO 3 -N, and total P using 30 x 30 m DEM data. The predicted output at the calibrated scale was used to evaluate output accuracy for the remaining input resolutions. Results of this study showed that DEM resolution affects the watershed delineation, stream network and sub-basin classification in the SWAT model. Lower values of discharge, NO 3 -N, and total P were predicted at the 5 coarsest DEM resolutions, probably due to decrease in computed total watershed area. The model predicted higher values of all three parameters at 100 m DEM resolution, when the total watershed area increased compared to the base 30 m resolution. Keywords. SWAT, DEM data resolution, Output uncertainty, model calibration 1 Respectively, Graduate Assistant, Assistant Professor, and Associate Professor, Department of Biological and Agricultural Engineering; Director, Water Quality Laboratory, Arkansas Water Resources Center; and Assistant Professor, Department of Civil Engineering, University of Arkansas, Fayetteville, AR Corresponding Author: I. Chaubey. chaubey@uark.edu

2 Introduction Nonpoint source (NPS) transport of nutrients, sediment and pathogens from agriculturally dominated watersheds is a major concern in Arkansas (Edwards et al., 1992; 1997). There is ample evidence to suggest that excess land application of animal manure and row crop agriculture have led to surface and ground water pollution (Edwards et al., 1996). Runoff losses of nutrients and sediment have resulted in excess algal blooms, eutrophication and turbidity of lakes and streams. The 303(d) list for Arkansas identifies sedimentation, mercury and nutrients as the top three pollutants of concern affecting more than 70% of total impaired water bodies in the state. In conjunction with the Clean Water Act, the US Environmental Protection Agency (EPA) requires the states to establish Total Maximum Daily Load (TMDL) estimates for pollutants in watersheds of concern. The TMDL program is identified as the method to resolve continuing water quality problems, including polluted runoff from nonpoint sources. Water quality models are frequently used to estimate NPS pollutant loads from watersheds and to predict stream response to various pollutant loading scenarios. Models are also used to estimate TMDLs from point and nonpoint sources that may result in desired optimum water quality improvement with minimum TMDL implementation cost. Because intensive monitoring of watersheds is very expensive, it is important that model estimates of effectiveness of various Best Management Practices (BMPs) are accurate so that costly mistakes of developing inaccurate water quality goals and implementing ineffective practices can be avoided. The Soil and Water Assessment Tool (SWAT) is one hydrologic model that has become widely accepted as an effective tool to predict the effects of management on water, sediment, nutrients and pesticide yields (Arnold et al., 1993; 1998). The US EPA is using the SWAT model as a component in its program (Better Assessment Science Integrating Pont and Non-point Sources, BASINS) to aid in achieving the goals of the Clean Water Act. However, this model has not been validated in Arkansas, and there is a critical need to evaluate the SWAT model for developing TMDLs in agricultural watersheds in the state. This paper reports the initial evaluation of the SWAT model for TMDL development in Arkansas by determining the optimum scale of Digital Elevation Model (DEM) data required to accurately develop TMDLs for agricultural watersheds. The optimum spatial scale of watershed characteristics for preparing model input was evaluated by running seven scenarios at increasing DEM grid sizes (30 x 30 m, 100 x 100 m, 150 x 150 m, 200 x 200 m, 300 x 300 m, 500 x 500 m, and 900 x 900 m) to predict discharge, nitrate, and phosphorus loads over a period of 4 years. The output accuracy for each input resolution was evaluated. As part of this exploratory evaluation, the model was tested using data obtained from the Lincoln Lake watershed, near Lincoln, AR. This watershed is representative of the Ozark Plateau Region of Arkansas. Lincoln Lake has been extensively monitored for over a decade and high quality data are available to critically evaluate TMDL model needs. The objective of this project is to evaluate the effect of input DEM data resolution on SWAT model predictions for TMDL development.

3 Methods and Materials The project was carried out in two phases. First the SWAT model was calibrated for the Ozark Plateau Region of Arkansas. Second, the optimum scale of input DEM data was identified that would predict NPS pollution and allow development of TMDLs with a high degree of accuracy. Description of Watershed Lincoln Lake is a small agricultural watershed located in western Washington County in Northwest Arkansas (Figure 1). It is a sub-basin of the Illinois River watershed. The drainage area of the watershed is 3240 ha. Major land uses within the watershed are pasture (55%) and mixed forests (39%). Animal production is prevalent, in the form of numerous poultry and beef operations located in the watershed. Water from Lincoln Lake serves as the secondary drinking water supply for the city of Lincoln. Water quality degradation of the lake is believed to be caused by runoff of nutrients from surface-applied animal manure in the watershed. Moores Creek, a tributary of Lincoln Lake, has been monitored for nutrients at two sites since Figure 1. Location and land use of the Lincoln Lake watershed Available Data Detailed hydrology and water quality data were available for this watershed from the Arkansas Water Research Center. Continuous stage and discharge data and water quality data of nitrate-nitrogen (NO 3 -N), ammonia-n (NH 4 -N), total Kjeldahl N (TKN), phosphate-phosphorus (PO 4 -P), total-p (TP), and total suspended solids (TSS) were available at 30-minute intervals during the rising limb and 60 minute intervals during the falling limb of storm hydrographs; and,

4 at 14 day intervals during base flow conditions. Available measured data from 1997 to 2000 are shown in Table 1. A detailed digital soil map for the watershed was obtained from the University of Arkansas Soil Physics Lab. The DEM for the watershed was downloaded from the United States Geological Survey. Both of these maps are scaled 1:24,000, which is approximately 30 x 30 meters in horizontal resolution. Land use and land cover data for 1999 at 30 x 30 m horizontal resolution, road network, and stream network data in GIS format were obtained from the Center for Advanced Spatial Technology (CAST) at the University of Arkansas. TABLE 1. Measured data (average annual) from the Lincoln Lake watershed Year Discharge (x 1000 m 3 ) NO 3 -N (kg) NH 3 -N (kg) TKN (kg) TP (kg) Description of the SWAT Model The SWAT model is a physically-based distributed-parameter watershed scale model. It divides the study watershed into sub-basins, or smaller homogeneous areas (Neitsch et al., 2000). SWAT considers all hydrologic processes within the sub-basins of the watershed. The EPA currently supports this model for developing TMDL in agricultural watersheds. SWAT consists of three major components: (1) sub-basin, (2) reservoir routing, and (3) channel routing. The sub-basin component consists of eight major divisions. These are hydrology, weather, sedimentation, soil temperature, crop growth, nutrients, agricultural management, and pesticides. Channel inputs include reach length, channel slope, channel depth, channel top width, channel side slope, flood plain slope, channel roughness factor, and flood plain roughness factor. GIS interfaces have been developed for the SWAT model to facilitate the aggregation of input data for simulating watersheds. The ArcView interface developed for the SWAT model was used to prepare input data files in this study. This interface requires a land cover map, soils map, and DEM as spatial inputs. Currently, a detailed description of the model can be found at the SWAT website: Model Calibration The SWAT model was calibrated for flow, NO 3 -N and TP loads in the Lincoln Lake watershed. Measured hydrologic and water quality data from (Table 1) were used. The 1997 data was used to calibrate the model on an average annual basis at the 30-meter DEM resolution. Monthly and daily calibrations are planned for future study. The model was calibrated first for annual flow, including total flow, storm-flow, and base-flow of the stream. Relative error (RE) between predicted flow (P) and observed flow (O) was calculated and minimized by adjusting the curve number (CN) in the model. An average CN value of 78 resulted in the minimum RE of 2% for predicted runoff. The RE was calculated as ( O P) PO 4 -P (kg) TOC (kg) * * * 456 RE (%) = 100 (1) O The procedure outlined in the model user s manual (Neitsch et al., 2000) was followed to adjust other model parameters during the calibration process. The parameters adjusted were the TSS (Mg)

5 threshold depths of water in the aquifer for revaporation to occur, and soil available water capacities. Additionally the alpha base flow factor, an index of ground water flow response to changes in aquifer recharge, was increased from 0.05 to Groundwater delay time was decreased to 10 days. To account for karst topography, common in this area, the crack potential of the soil was increased to 0.75 (Dr. D. E. Storm, Oklahoma State University, personal communication). Also Manning s n was changed to reflect land cover in the watershed, using 0.04 for channel flow and 0.4 for overland flow (Haan et al., 1994). After flow calibration, nutrient loads were calibrated. The location of broiler operations in the watershed were obtained from 2000 CAST data, and were verified through groundtruthing. This information was used to estimate poultry litter applications and initial soil nutrient content. It was assumed that poultry litter was applied in all existing pasture areas. The litter application rate was estimated to be 5600 kg/ha/year. This total load was partitioned into two applications at six-month intervals. Calibration of nutrient component of the stream flow was more difficult because changing one parameter affected all simulated loads. During the model calibration, the sum of RE for NO 3 -N and TP predictions was calculated and minimized. The model parameters that were adjusted to calibrate nutrient output were average slope length, N and P percolation coefficients, and P soil partitioning coefficient. The calibrated model parameters resulted in a total RE of 8% for NO 3 -N and TP predictions. Results and Discussion Results of this study show that DEM resolution affected the watershed delineation, stream network and sub-basin classification in the SWAT model. The 30 x 30 m DEM resulted in 7 sub-basins and total watershed area of 1612 ha. As the DEM scale increased total computed watershed area and the number of sub-basins decreased for all resolutions except 100 x100 m. The 100 x 100 m delineation resulted in a 14% increase in total computed watershed area and included 8 sub-basins. The modeled stream network became consistently less accurate at coarser resolutions as illustrated in Figure 2 for three DEM resolutions. Figure 2. SWAT watershed delineation shown at 30 m, 200 m, and 900 m (from left to right).

6 Model runs were performed at the seven DEM resolutions for all storm events and base flow conditions for the period of Discharge, NO 3 -N and TP were the model outputs of interest. Discharge and NO 3 -N are directly calculated by SWAT. Total phosphorus was estimated as the sum of organic P and mineral P. The model predictions for discharge, NO 3 -N and TP loads were compared to the predictions at the calibrated DEM resolution (30 x 30 m). Table 2 shows average of total annual predicted values for each output. The effect of DEM data resolution on model output was measured using RE from Eqn. 1. Figure 3 shows how RE for each parameter was affected by data resolution for TABLE 2. Average SWAT model predictions at varying DEM resolutions Scale Average Discharge (x 1000 m 3 ) RE Discharge (%) Average NO 3 -N (kg) RE NO 3 -N (%) Average TP (kg) 30 x 30 m x 100 m x 150 m x 200 m x 300 m x 500 m x 900 m In general the results show that as the DEM horizontal resolution became coarser, the predicted discharge volume decreased. Coarser resolutions reduce the heterogeneity of the topography. The overall slope of the stream was reduced which resulted in a less flow. The same trend was seen in the NO 3 -N and TP predictions. This trend was seen at all resolutions with the exception of the 100 x 100 m, where the predicted values increased, perhaps due to increase in total computed watershed area. Predicted values for each individual year also showed discharge, NO 3 -N and TP were overestimated at the 100 x 100 m scale and underestimated at the other 5 scales. In a study done using 71 different watersheds and TOPMODEL, Wolock and Price (1994) found that increasing DEM coarseness from 30 x 30 m to 90 x 90 m increased runoff prediction by decreasing slope and increasing slope area per unit contour length. In a different study involving the SWAT model, Cho and Lee (2001) found that simulated runoff was higher for 30 x 30 m data resolution compared to runoff volume predicted using 90 x 90 m DEM. The authors surmised that a finer data resolution resulted in higher slope and hence the higher simulated runoff volume. Increase in predicted runoff volume, NO3-N and TP loads at 100 x 100 m DEM resolution needs further investigation to see why the trend turns there. We are currently working with 50 x 50 m, 75 x 75 m, and 125 x 125 m DEM data to see how computed watershed area, subwatershed delineation, watershed slope and watershed contributing area changes with change in input DEM data resolution. Future study will also involve working with other watersheds in Arkansas to test the generality of the results obtained in this watershed. RE TP (%)

7 200 RE (%) Runoff RE NO3 RE TP RE Total RE 0 100m 150m 200m 300m 500m 900m DEM Data Resolution Figure RE for Runoff, NO 3 -N, TP, and Total RE for all model outputs. Summary and Conclusions Watershed topography is an important factor in determining watershed response to runoff and water quality parameters. In this study, the effect of DEM data resolution on SWAT model predictions for TMDL development was evaluated. Measured hydrologic, meteorological, watershed characteristics, and water quality data from Lincoln Lake watershed were used in the simulation. The effect of spatial scale input was evaluated by running seven scenarios at increasing scales (30 x 30 m, 100 x 100 m, 150 x 150 m, 200 x 200 m, 300 x 300 m, 500 x 500 m, and 900 x 900 m). A 30 x 30 m input DEM data resolution was used to calibrate the model on an annual basis for stream flow, NO 3 -N, and total P. The predicted output at the calibrated scale was used to evaluate output accuracy for the remaining input resolutions. Input DEM data resolution affected SWAT model predictions by affecting total area of the delineated watershed, predicted stream network, and sub-basin classification. Lower values of discharge, NO 3 -N, and total P were predicted at the 5 coarsest DEM resolutions, probably due to decrease in total watershed area, and watershed slope. The model predicted higher values of all three outputs at the 100 m DEM resolution, corresponding to an increase in predicted total watershed area. Similar studies are planned to investigate the effects of land cover and soil input data resolution on SWAT output. Currently we are working with 50, 75, and 125 m DEM data resolution to test the effect of input data coarseness on model outputs. Future work will also involve applying the SWAT model to other watersheds in Arkansas to test the generality of the results obtained here. Literature Cited Arnold, J. G., R. Srinivasan, R. S. Muttiah, and J. R. Williams Large area hydrologic modeling and assessment, Part1: Model Development. J. Amer. Water Resour. Assoc. 34 (1):1-17. Arnold, J. G., P. N. Allen, and G. Bernhardt A comprehensive surface-groundwater flow Model. J. Hydrology 142:47.

8 Cho, S. M. and M. W. Lee Sensitivity considerations when modeling hydrologic processes with digital elevation model. J. Amer. Water Resour. Assoc. 37 (4): Edwards, D. R., T. C. Daniel, H. D. Scott, P. A. Moore, Jr., J. F. Murdoch, and P. F. Vendrell Effect of BMP implementation on storm flow quality of two Northwest Arkansas streams. Trans. ASAE 40(5): Edwards, D. R., T. C. Daniel, H. D. Scott, J. F. Murdoch, M. J. Habiger, and H. M. Burks Stream quality impact of best management practices in a Northwest Arkansas Basin. J. Amer. Water Resour. Assoc. 32(3): Edwards, D. R. and T. C. Daniel Potential runoff quality effects of poultry manure slurry applied to fescue plots. Trans. ASAE 35(6): Haan, C. T., B. J. Barfield, and J. C. Hayes Design hydrology and sedimentology for small catchments. Academic Press. San Diego, CA. Neitsch, S. L., J. G. Arnold, J. R. Kiniry, J. R. Williams Soil and Water Assessment Tool user s manual. Blackland Research Center. Temple, TX. Wolock, D. M., and C. V. Price Effects of digital elevation model map scale and data resolution on topography-based watershed model. Water Resour. Research. 30(11):