Setting Up a Computer Simulation Model in an Arkansas Watershed for the MRBI Program Setting Up a Computer Simulation Model in an Arkansas Watershed for the MRBI Program

The Mississippi River Basin Healthy Watersheds Initiative (MRBI) program launched by the USDA Natural Resources Conservation Service (NRCS) aims to improve the water quality within the Mississippi River Basin. Lake Conway Point Remove (LCPR) water- shed, being one of the MRBI watersheds, is a potential candidate for evaluating the effectiveness of MRBI program. Recommended best management practices (BMPs) for LCPR watershed are pond, wetland, pond and wetland, cover crops, vegetative filter strips, grassed waterways, and forage and biomass planting. Before simulating these practices, it is essential to prepare the data needed for model setup to avoid the issue of garbage in, garbage out. This chapter focuses on detailed steps of preparing the data for model setup along with the calibration and validation of the model. The calibration and validation results were within the acceptable bounds. The results from this study provide the data to help simulate the MRBI best management practices effectively and prioritize monitoring needs for collecting watershed response data in LCPR. The model for at model and satisfactory by studies. Results from this study can be used to evaluate the relative effectiveness of MRBI-recommended agricultural BMPs for analyzing and improving in similar data-limited watersheds.


Introduction
The Mississippi River Basin Healthy Watersheds Initiative (MRBI) program aims at implementing best management practices (BMPs) to control water quality. Quantifying the impacts of BMPs is important to demonstrate the worth of the MRBI program. Out of various MRBIselected watersheds, the Lake Conway Point Remove (LCPR) watershed is the one listed in the 2011-2016 priority watershed by the Arkansas Natural Resources Commission (ANRC) [1,2].
cropland. An increase in urbanization, in parts of the watershed, has occurred since 1999. The subwatersheds within LCPR along with the area and hydrological unit codes (HUC) can be seen in Table 1.

Data preparation
The objective of this task was to collect and organize all data needed for the SWAT model setup at a 12-digit hydrological unit code within the LCPR watershed. Geospatial, watershed management,

Subwatershed
Subwatershed name Area (km 2 ) HUC no. water quantity, and point source data that were available and usable at the time of modeling were collected and reorganized in a consistent format for use in the SWAT model.

Elevation
The elevation dataset was retrieved at a 5 m resolution from GeoStor. This 5 m dataset was resampled to a 10 m resolution to reduce the size of huge files and increase the computation efficiency. The elevation map for LCPR can be seen in Figure 2.

Soils
The soil data were acquired from the Soil Survey Geographic (SSURGO) database for all LCPR counties in Arkansas and combined to make a soil map for the entire watershed. The SSURGO is the most comprehensive and detailed soil dataset available for LCPR. The soil map for LCPR can be seen in Figure 3.

Land use/land cover
Land use and land cover data were acquired for 1999, 2004, and 2006 from GeoStor. Forest area was observed to be the most dominant land use and cover in the LCPR watershed. All land use and land covers were reclassified to make it compatible with the SWAT model. The land use and land cover map for LCPR can be seen in Figure 4.

Climate
Climatic data specifically daily precipitation and maximum and minimum temperature data were obtained from 90 climate stations from the NOAA's National Climatic Data Center (NCDC). Data are available from 1980 to 2012 for at least one of the climatic parameters. The procedure recommended by USDA-ARS in developing SWAT-formatted climate data were followed. Daily climate data were obtained using an inverse distance-weighted interpolation algorithm. The average data were calculated for each subwatershed using a pseudo-weather  station. NCDC validation results at each calibration station using leave-one-out cross-validation technique can be seen in Table 2. NEXRAD data were obtained from the Arkansas Basin River Forecasting Center (ABRFC).

Streamflow
The flow data are available for the West Fork Point Remove Creek near the Hattieville monitoring station from the US Geological Survey (USGS). This monitoring station is located in subwatershed 3 and covers approximately 20% of LCPR. The flow data were split between surface and baseflow using the baseflow filter program by [17].

Point sources
Point source data were obtained from the Arkansas Department of Environmental Quality (ADEQ) and was processed in the SWAT-compatible format. Point source data were available for flow, total suspended solids, organic nitrogen, organic and mineral phosphorus, nitrate nitrogen, ammonia nitrogen, and carbonaceous biochemical oxygen demand (CBOD). Locations for active point source facility that was incorporated in the SWAT model can be seen in Table 3.

Cattle grazing, manure deposition, and poultry litter application
The detailed method for estimating pastures that should be receiving litter applications can be seen below. NEXRAD detects no rainfall event (DNO_RAIN). 3 Mean error (ME). 4 Index of agreement (d). 5 Percent bias (PBIAS). 6 Coefficient of determination (R2). 7 Nash-Sutcliffe efficiency (NSE). 8 Mean absolute error (MAE). 9 Root-mean-square error (RMSE). Detailed methods for estimating pastures that received litter application: 1. Create buffer of a random radius around the active poultry houses.

2.
Extract pasture areas under the buffer.
3. Assuming a grazing density of 1 cow/0.8 ha of litter amended pasture, calculate the number of cows that can fit the buffer.

4.
Compare the calculated number of cows to the number of cows in the subwatershed.

5.
Repeat steps 1-4 to obtain the best agreement between estimated numbers of cows.
6. Apply litter to pasture HRUs that fall under the best buffer radius.
The SWAT compatible data for cattle grazing, manure deposition, and poultry litter application can be seen in Table 4.

Urban pasture management
The pasture management schedule relating to specific operation and crop can be seen in Table 5.

Calibration and validation
Before calibrating a model, sensitivity analysis is usually performed to reduce the number of parameters. Latin hypercube (LH) one-at-a-time (OAT) method [20] was used to identify the sensitive parameters that might affect the output results. A total of 22 flow parameters were tested, and the following 12 were found sensitive: SOL_AWC, CN2, ALPHA_BF, SOL_K, CH_N2, CH_K2, CANMX, RCHRG_DP, SURLAG, GW_DELAY, OV_N, and GW_REVAP.
The model calibration period was from 1987 to 2006 and the validation period was from 2007 to 2012. The first 3 years of calibration period were selected as a warm-up period so that the model parameters can be initialized. The calibration started with baseflow followed by surface flow adjusting related parameters affecting baseflow and surface flow. The SWAT Check tool [21] was used before calibration to make sure that the simulated outputs were within the reasonable ranges. The Load Estimator (LOADEST) tool [22] was used on a water quality dataset available from Sept 2011 to Dec. 2012 at Hattieville and Apr. 2012 to Dec. 2012 at Morrilton. The regression coefficients were found to be statistically significant (p < 0.05) at Hattieville and Morrilton for sediment, total phosphorus, and nitrate nitrogen. The performance of the model was determined mainly using the coefficient of determination (R 2 ).

Calibration and validation results
Various SWAT parameters that were calibrated along with their parameter ranges and final calibrated values can be seen in was 0.84, 0.78, and 0.76 for the total, surface, and baseflow, respectively. The calibration and validation scatter plots for total flow, surface flow, and baseflow can be seen in Figure 5. The validated R 2 for water quality was 0.5-0.7 at Hattieville and 0.7-0.87 at Morrilton. The results are within acceptable limits of other modeling studies relating to limited data availability [24,25].

Conclusions
Modeling studies are gaining popularity due to rapidness of insight generation before actually performing field experiments. The initiative led by the Mississippi River Basin focused on analyzing the water quality benefits from intended best management practices with the help of modeling studies. However, merely simulating best management practices will not be able to provide reliable results unless the model has been set up correctly and robust. This chapter focused on the detailed discussion for setting up the model to a point where the model setup procedure can be replicated. The model was set up with all relevant information, and each data preparation step has been explained in detail. The model was calibrated and validated for flow at Hattieville. Due to limited water quality data, the model was validated for sediment, total phosphorus, and nitrate nitrogen at Hattieville and Morrilton. The results were satisfactory and within the ranges reported by previous studies. Results from this study can be used to evaluate the relative effectiveness of MRBI-recommended agricultural BMPs for analyzing pollutant load reductions and improving water quality in similar data-limited watersheds.

Conflict of interest
The authors declare no conflict of interest.