| Summary: | An automated system for distributed hydrologic modeling was developed for use with the Soil and Water Assessment Tool (SWAT: Arnold et al., 1996) and the Kinematic Runoff and Erosion Model (KINEROS; Smith et al., 1995). This suite of programs, the Automated Geospatial Watershed Assessment (AGWA) tool, was used to investigate the impacts of land cover change in watersheds within the Catskill/Delaware watershed complex in upstate New York and Upper San Pedro River basin in southeastern Arizona. As shown by classified remotely sensed imagery, these watersheds have undergone opposing land cover transitions over the past 25 years. SWAT simulations illustrated minor improvements in hydrologic response within the Catskill/Delaware watershed but a moderate increase in the San Pedro. A small watershed within the San Pedro was modeled using KINEROS to demonstrate localized impacts of land cover transitions. Fifty watersheds ranging in size from 5 to 100 km - were prepared for input to the KINEROS model using a range of geometric complexities and rainfall events. A nonlinear response to watershed complexity as a function of watershed scale and rainfall magnitude was observed. Small watersheds were more sensitive to landscape variability when small rainfall inputs were used, but the impact of spatial representation became insignificant when large rainfall inputs were used. Conversely, large watersheds were sensitive to spatial complexity for all rainfall data, although the relative effects were mitigated by increased rainfall. Remotely sensed imagery served as the primary mechanism for determining land cover within the study areas. A technique was created in which the errors from a misclassification error matrix were systematically introduced to a classified image to produce 100 realizations of land cover within the San Pedro basin. These realizations were used by AGWA to prepare input to KINEROS at a range of basin scales and rainfall inputs. The propagation of error and resultant uncertainty in simulation results was found to vary with watershed scale and rainfall input. Larger watersheds are more sensitive to misclassification error, while uncertainty in model output is inversely related to rainfall magnitude.
|
|---|
