Applicability of the Neural Network Based Radar Rainfall Estimation in Streamflow Modeling

Abstract

This study evaluates performance of the neural network based radar rainfall estimation and examines the applicability of the radar rainfall input estimated from the neural network (R_NN) in streamflow modeling. The Uono River basin is selected as the study basin. A distributed hydrologic model is driven by using R_NN, the radar rainfall input obtained from the Z-R relationship (R_Z-R), and the gauge rainfall (R_G) respectively. The statistical results of radar rainfall estimation indicate that the radar rainfall product using the neural network is more accurate than that using the operational Z-R relationship. In addition, the streamflow simulation results show that the simulated hydrographs obtained from R_NN are more accurate than those obtained from R_Z-R. The study concluded that the neural network technique outperforms the existing operational Z-R relationship. The appropriately trained network at the rain gauge sites is accurate, stable, and robust for estimating radar rainfall over the whole basin. The study also suggested that the R_NN is an alternative input for hydrologic modeling when the gauge rainfall data is unavailable or insufficient.