Assessing a Regression-Based Regionalization Approach to Ungauged Sites with Various Hydrologic Models in a Forested Catchment in the Northeastern United States

Abstract

Analysis of daily streamflow is of interest to river restoration and conservation efforts in many regions in the world. However, the paucity of stream-gauging stations presents significant challenges in making flow predictions. In this study, two process-based rainfall-runoff models that differ in complexity were used to estimate a reliable simulation of the daily streamflow hydrograph using 15 subbasins in the Deerfield River Basin, a major tributary to the Connecticut River Watershed. Catchment characteristics were employed across the study area, and regional regression equations were developed that correlate these physical and climate characteristics with the parameters of the rainfall-runoff models. The regression-based regionalization approach had a higher degree of accuracy when compared with simpler regionalization approaches, with an average normalized RMS error (NRMSE) value of 0.26 compared with 0.42 and 0.32 for a spatial proximity method and a naïve-mean method, respectively. In addition, the more complex rainfall-runoff model performed better than the less complex model with Kling-Gupta efficiency (KGE) values of 0.78 and 0.68, respectively, suggesting that model discretization may play a significant role in hydrologic model accuracy. These findings support a viable framework for addressing water resource management at a small-catchment level. In addition, this study may contribute to regionalization of rainfall-runoff model parameters for ungauged basins in the northeastern U.S. region.