Submitted to: Transactions of the ASABE
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: July 5, 2010
Publication Date: November 3, 2010
Citation: Veith, T.L., Van Liew, M.W., Bosch, D.D., Arnold, J.G. 2010. Parameter sensitivity and uncertainty in SWAT: A comparison across five USDA-ARS watersheds. Transactions of the ASABE. 53(3):1477-1486. Interpretive Summary: Using simulation modeling to evaluate management practices in watersheds that do not have measured data or are vastly different from previously modeled watersheds is difficult because there is limited information on how to change the calibration parameters. Five different USDA ARS watersheds were modeled using SWAT, a common watershed-level water quality model, and sixteen calibration parameters were studied. The parameter ranges needed to provide well calibrated solutions for each watershed were identified, and the uncertainty ranges of the all solutions were compared across watersheds. These findings provide SWAT streamflow calibration guidelines for a wider range of watershed locations than previously available and provide users some insight about the uncertainty of the simulation for their location.
Technical Abstract: The USDA-ARS Conservation Effects Assessment Project (CEAP) calls for improved understanding of the strengths and weaknesses of watershed-scale, water quality models under a range of climatic, soil, topographic, and land use conditions. Assessing simulation model parameter sensitivity helps establish feasible parameter ranges, distinguish among parameters having regional versus universal interactions, and ensure that one model process does not compensate for another due to poor parameter settings. The Soil and Water Assessment Tool (SWAT) parameter sensitivity and autocalibration module was tested on two northern and three southern USDA ARS experimental watersheds. These watersheds represent a range of climatic, physiographic, and land use conditions present in the United States. Sixteen parameters that govern basin, snow accumulation/melt, surface, and subsurface response in the model were evaluated. Average monthly streamflow over a 3-5yr period was calibrated against measured data for each watershed using mean square error, percent bias, and visual comparison. Parameters governing surface runoff due to rainfall were found most sensitive overall while parameters governing ground water were the least sensitive. Surface runoff parameters were found most sensitive for areas with high evaporation rates and localized thunderstorms. Parameters from all categories were important when precipitation includes both rainfall and snowfall. Differences in model performance were noticeable on a climatic basis; SWAT generally predicted streamflow more accurately and precisely, i.e. with less uncertainty, in humid climates than in desert or semi-desert climates. Results of these analyses will be especially valuable in CEAP investigations that involve quantifying probability distribution functions related to downstream hydrologic response as a result of conservation practice implementation.