Simulating Typhoon Floods with Gauge Data and Mesoscale-Modeled Rainfall in a Mountainous Watershed

Abstract

A physically based distributed hydrological model was applied to simulate typhoon floods over a mountainous watershed in Taiwan. The meteorological forcings include the observed gauge rainfall data and the predicted rainfall data from a mesoscale meteorological model, the fifth-generation Pennsylvania State University?National Center for Atmospheric Research (PSU?NCAR) Mesoscale Model (MM5). This study investigates the flood responses of three Typhoons: Zeb (1998), Nari (2001), and Herb (1996), which possessed unique meteorological features and that all produced severe floods. The predicted basin-averaged rainfall hydrographs by the MM5 are compared with that interpreted by rain gauge data to reveal the discrepancies in rainfall peak amounts and time lags, and to explore their subsequent effects on flood generation. The simulated flood hydrographs at the Hsia-Yun station, which is upstream of the Shihmen Reservoir, are compared with observed flood discharges in terms of the amount and time lag of flood peaks. It is shown that the small discrepancy in rainfall peaks and phase lags could be significantly amplified in simulated flood responses of a mountainous watershed. The overall predictive skill of the distributed hydrological model with different rainfall inputs is examined with three parameters, which include the runoff ratio (RR), root-mean-square error (rmse), and goodness of fit (GOF). Although the runoff ratio for the MM5-predicted rainfall is superior to that for the observed gauge rainfall, the simulated hydrographs with observed gauge rainfall have smaller rmse and GOF values for three events. This study shows that the error in flood prediction with the mesoscale-modeled rainfall is mainly caused by the rainfall?peak difference, which arises from the inherent uncertainties in the mesoscale-modeled rainfalls over a mountainous terrain during the typhoon landfall periods.