HSPF Modeling of a Forested Headwater in North-Central Minnesota

Abstract:

The Hydrologic Simulation Program Fortran (HSPF) model is a conceptual based model, in contrast to the alternative grid-based hydrologic models. In its application, the HSPF model has been applied almost exclusively to watersheds scales larger than HUC 14. We applied the HSPF model to a small forested watershed located at the Marcell experimental forest in north-central Minnesota. The watershed has an area of 9.7 hectares and is one of six experimental watersheds managed by the U.S. Forest Service. These experimental watersheds have been in operation since 1955. While some of the experimental watersheds have experienced environmental disturbance (e.g., clear-cutting), the S2 watershed has been in continuous forested condition for the last century. The S2 watershed is composed of forested upland hillslopes that connect to a raised bog area. The forest is aspen (Populus sp.) and birch (Betula papyrifera) on uplands and spruce (Picea mariana) on a 3.2 ha central bog. The upland soils are composed of a shallow sandy loam overlying clay loam which impedes vertical water movement. Water input to the peatland is either from direct precipitation or from the upland hillslopes. Water flux from the bog is either through ET, groundwater seepage losses, or surface discharge to a stream network. Streamflow is measured at a v-notch weir. Within the watershed two upland hillslopes have been isolated to monitor surface runoff and subsurface flow since the 1980’s. The HSPF model was applied to the watershed using precipitation and other meteorological data collected within the watershed or nearby on the Experimental Forest. The model representation of the watershed included one element for the upland hillslope area, and one element for the bog area. Daily flows from water years 1991 to 1995 were used to calibrate the model. An equal period of time was used to validate the model from 1996 to 2000. Nash-Sutcliffe values were very good for both calibration and validation. Calibration monthly and daily NS were 0.86 and 0.80, respectively. Validation monthly and daily NS were 0.85 and 0.71, respectively. The calibration time period overestimated flow by 0.33% while the validation period overestimated flow by 2.69%. Reduction of model accuracy was mainly related to the intermittency of the upland runoff and bog outflows.