Submitted to: Transactions of the ASABE
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: February 4, 2007
Publication Date: June 22, 2007
Citation: Feyereisen, G.W., Strickland, T.C., Bosch, D.D., Sullivan, D.G. 2007. Evaluation of SWAT manual calibration and input parameter sensitivity in the Little River Watershed. Transactions of the American Society of Agricultural and Biological Engineers. 50(3):843-855. Interpretive Summary: The loss of sediment, nutrients, and pesticides into surface waters of the Southeast Coastal Plain is being monitored within the Little River Experimental Watershed, an area of about 130 square miles just northwest of Tifton, Georgia. There is a need to know how various agronomic and conservation practices, and climate affect the water quality in the region. In order to extend analysis to longer time frames, computer simulation models are utilized to mimic the response of the natural landscape to these management and climate changes. Questions such as: “What if 50% of the cropped land were in conservation tillage?” or “What if 100% of the cropped land were in conservation tillage?” or “What if 25% of the cropped land was converted to permanent plantings?” can be analyzed. The first step in being able to answer questions like those above is to be able to accurately model the water cycle within a watershed. In order to accurately model the water cycle, it is important to have an intensively-measured watershed with actual results against which to compare the simulated results. Computer simulations do not perfectly model what happens in the field and on the watershed. The person doing the modeling needs to know how accurate and how precise the model being used is, and which inputs to the model cause the largest changes to the model output. When the most influential inputs are identified, more time and energy can be focused on measuring or estimating these inputs with a higher degree of accuracy. This paper identifies the input values to a widely-used model, the Soil Water Assessment Tool (SWAT), that are most sensitive to predicting annual total water yield and surface runoff for the 1692 ha (4181 ac) subwatershed known as “K” of the Little River Experimental Watershed. The most sensitive model inputs are the SCS soil curve number, the available water content of the soil, and a soil evaporation coefficient.
Technical Abstract: The watershed-scale effects of agricultural conservation practices are not well understood. A baseline calibration and input parameter sensitivity analysis were conducted for simulation of watershed-scale hydrology in the Little River Experimental Watershed (LREW) in the Coastal Plain near Tifton, GA, USA. The Soil and Water Assessment Tool (SWAT) was manually calibrated to simulate the hydrologic budget components measured for the 16.9 km2 subwatershed K of the LREW from 1995 to 2004. A local sensitivity analysis was performed on sixteen input variables. The sum of squares of the differences between observed and simulated annual averages for baseflow, stormflow, evapotranspiration, and deep recharge was 19 mm2; average annual precipitation was 1136 mm. The monthly Nash-Sutcliffe model efficiency (NSE) for total water yield (TWYLD) was 0.79 for the ten-year period. Daily NSE was 0.42. The NSE’s for three years with above average rainfall was 0.89 while it was 0.59 for seven years with below annual average rainfall. Monthly average TWYLD for the winter-early spring season was underpredicted while late summer-autumn TWYLD was overpredicted. Results were negatively influenced when seasonal tropical storms occurred during a dry year. The most sensitive parameters for TWYLD were curve number for crop land, soil available water content, and soil evaporation compensation factor. The most sensitive parameters for stormflow were curve number for crop land, curve number for forested land, soil bulk density, and soil available water content. Correcting the seasonal flow discrepancies and trimming the high peak predictions after long dry spells would improve the hydrologic modeling efficiency and enhance the model for future water quality simulation.