Modeling monthly actual evapotranspiration: an application of geographically weighted regression technique in the Passaic River Basin

Abstract

Abstract Actual evapotranspiration (ET) is perhaps the most difficult quantity to directly measure among the major water balance components. Because of the high cost and labor constraints associated with the direct measurement of ET, empirical data-driven modeling has frequently been used to estimate ET. Beyond the widely used traditional type regression that has the effect of producing ‘global’ parameter estimates, assumed to be uniform throughout an area, we utilized a more localized spatially non-stationary technique – the geographically weighted regression (GWR) – to estimate mean monthly ET in the Passaic River Basin (PRB). We identified the key environmental controls of ET and developed new sets of spatially varying empirical ET models based on variable combinations that produced the best-fit model. The analysis showed that temporal and spatial variabilities in ET over the PRB are driven by climatic and biophysical factors. We found that the key controlling factors were different from month to month, with wind speed being dominant throughout the year in the study basin. A monthly mean ET index map was further generated from the model to illustrate areas where ET exceeds precipitation. This will among others enable water loss due to evapotranspiration to be accounted for in future water supply plans for the basin.