THE TRIALS AND TRIBULATIONS IN APPLYING THE SWAT WATERSHED MODEL TO THE WILLOW RIVER, WESTERN WISCONSIN

Marylee S. Murphy, University of Minnesota, Dept. of Geology and Geophysics
James E. Almendinger, St. Croix Watershed Research Station, Science Museum of Minnesota

We are applying the Soil and Water Assessment Tool (SWAT) to the Willow River watershed in western Wisconsin to assess the effects of land use and management changes. The Willow River drains 292 mi2 of predominately agricultural land in St. Croix County and enters the St. Croix River at city of Hudson. Current trends in land use include conversion of agriculture to rural residential, urban expansion, conversion of dairy farming to cash cropping of corn and soybeans, and increased conservation tillage. SWAT is a spatially-based, continuous watershed modeling program with a daily time step developed by the Agricultural Research Service. SWAT is a well-used and highly regarded modeling program; nonetheless, a number of problems had to be corrected or circumvented before the Willow model could be calibrated.

Two SWAT models of the Willow have been calibrated to the available data sets, which include flow and water-quality data from water year (WY) 1999, crop yield data from 1992-2001, estimates of gross field erosion rates, and estimates of sediment and phosphorus trapped in reservoirs. Model hydrology was first calibrated to mean daily flows for WY 1999 by adjusting curve number (infiltration) and snow-melt parameters. Crop yields were then calibrated by adjusting crop bio-efficiency, fertilization routines, and in-field denitrification. By summing the monitored sediment load in WY 1999 with the estimated sediment trapped in reservoirs, we calculated the target amount of sediment transported by the channel. Even after accounting for sediment trapped by closed depressions and riparian wetlands, preliminary model runs still overestimated this target amount. From this point, sediment calibration could proceed in at least two different directions. First, the model could trap all the excess sediment on the landscape by reducing the soil-loss equation P-factor (cropping practice) from a default value of 1.0 to 0.45. This model version is called the “passive channel,” because erosion or deposition in the channel is disallowed; all eroded soil that reaches the channel is routed directly to the reservoirs and outlet. Alternatively, when the P-factor was doubled to 0.9, sediment delivery to the channel also roughly doubled, and the model had to be adjusted to trap about half this sediment load in the channel, with the remainder reaching the reservoirs and outlet. This model version is called the “active channel,” because channel (or floodplain) processes play a role in the sediment budget. Final sediment calibrations for both the passive and active channel versions were attained by parameterizing the reservoirs to trap 75% of the sediment delivered to them. Total phosphorus loads were calibrated by adjusting the modeled amounts delivered from the uplands, trapped in the channel (active-channel version only), and trapped in the reservoirs. Which version of the model is more realistic is unknown, yet having both versions is instructive because they handle sediment and phosphorus transport differently and are thus expected to give different results from simulated changes in land use or agricultural practices. Currently, these models are being used to simulate the effect of selected agricultural management practices, including installation of buffer strips, conversion of all crops to conservation tillage (mulch or no-till), conversion of all crops to a simple corn-soybean rotation (both conventional and conservation tillage), reduction of dietary phosphorus in dairy feed, and reduction of phosphorus in inorganic fertilizer. Other scenarios planned include increased conversion of agricultural land to rural residential and a base-level run with native vegetation.