Assessing Impacts of Precipitation and Parameter Uncertainty on Distributed Hydrologic Modeling

Abstract

Distributed hydrologic models, based on conservation laws, simulate the flow of water over and through the land surface in response to forcing from precipitation, transpiration, and evaporation. Conservation laws provide a physical basis for runoff generation that are dependent on the accurate specification of initial conditions, boundary conditions, and representative parameter estimates which control the model's performance. The main objective of this dissertation is to develop and test a method for calibration of a distributed hydrologic model in the presence of rainfall input uncertainty that utilizes the physics of runoff generation processes.

The main hypothesis tested is that a model calibrated using spatially distributed (SD) parameter adjustments will have less prediction error than a model calibrated by a spatially averaged (SA) parameter adjustment. A Mann Whitney Wilcoxon (MWW) rank sum hypothesis test is used to test the statistical significance. The results of the MWW rank sum hypothesis test show the mean of RMSE from the model calibrated by SD adjustments is less than the RMSE from the model calibrated using the SA parameter adjustment. The Nash Sutcliffe Efficiency of the SD calibrated model is also consistently higher than the SA calibrated model. These results are consistent at both the calibration gauge and at the interior gauge point. Thus, a spatially distributed parameter adjustment technique leads to a reduction in prediction error compared with the spatially averaged parameter adjustment technique.