Regionalization of Streamflow Characteristics for the Gulf-Atlantic Rolling Plains Using Leverage-Guided Region-of-Influence Regression

Multivariate regression models are often applied to hydrologic regions to estimate peak flows at ungauged basins in an area. Such regression models can be derived using a region of influence (RoI), which is a selected set of basins that are hydrologically similar to the ungauged basin for which peak flow estimates are required. These regions can be poorly defined resulting in unstable parameter estimates because observations at a few basins may disproportionately influence the parameters: Conventional treatment is to drop the basin, if the problem is even recognized. We propose a leverage-guided RoI regression approach that redefines the region of influence. This new procedure uses two newly defined RoI leverage and influence metrics. The proposed approach is applied to 996 streamflow gauging stations in the southeast United States to estimate the 50-year peak flow. The new leverage-guided RoI regression approach resulted in lower root-mean-square estimation errors, produced fewer observations with large leverage, and eliminated all influential observations.