Influence Of Precipitation Characteristics And Model Structure On Flood Simulation And Its Improvement In Xiang River Basin,China

Accurate flood events simulation and prediction,enabled by effective models and reliable data,are critical for flood prevention and mitigation.Satellite-based precipitation data is an important source of hydrological simulation,especially for ungauged basins,which effectively compensates for gauged precipitation data.In this dissertation,we firstly evaluate the accuracy of four satellite-based precipitation products with high spatio-temporal resolutions,namely IMERG-E,IMERG-L,IMERG-F and TMPA 3B42 RT,with CMA as reference.And their applicability in runoff and flood simulation in the Xiang River Basin is evaluated based on the HBV hydrological model.In addition to the bias of precipitation data,the resolution of precipitation,the model structure,and the calibration strategy also have impacts on the flood simulation.This study uses the high spatio-temporal resolutions IMERG products and a gaugebased product as precipitation forcings for hydrologic simulation.Three hydrological models(HBV,SWAT,and DHSVM)and a data-driven model(Long Short-Term Memory(LSTM)network)are utilized for flood events simulation.Two calibration strategies are carried out,one of which targets at matching the flood events and the other one is the conventional strategy to match continuous streamflow,for exploring the influencing factors of flood simulation.Finally,we combine the LSTM-based spatial interpolation method and dynamic Bayesian Model Averaging to merge the IMERG-E satellite-based precipitation data and gauge precipitation data by exploiting their spatial and temporal correlations simultaneously.The merged precipitation product is used in flood simulation in order to improve the accuracy of flood simulation.The major conclusions of the dissertation are as follows:(1)The statistical assessment about the accuracy of selected hourly satellite-based precipitation products(IMERG-E,IMERG-L,IMERG-F,and TMPA 3B42RT)shows IMERGF outperforms the others,and the performance of TMPA 3B42 RT is generally better than IMERG-E and IMERG-L in terms of BIAS and CC.Furthermore,IMERG-F performs the best in hourly streamflow simulation,as well as in flood events simulation with the mean NSE of0.71,while IMERG-E,IMERG-L,and TMPA 3B42 RT present overall acceptable performance.The simulated peak flows tend to be smaller than the observations,and the simulated time lags of occurrence are changeable.This study finds that IMERG and TRMM are suitable for hydrological simulations at hourly scale,and IMERG-E is an appropriate replacement for TMPA 3B42 RT in this case.(2)The event-based calibration strategy improves the performance of flood events simulation,compared with conventional calibration strategy,except for DHSVM.Both hydrological models and LSTM yield better flood events simulation at finer temporal resolution,especially in flood peaks simulation.Furthermore,SWAT and DHSVM is less sensitive to the spatial resolutions of IMERG,while the performance of LSTM obtains improvement from the scale transformation of spatial resolution.Generally,the LSTM outperforms the hydrological models in most flood events,which implies the usefulness of the deep learning algorithms for flood events simulation.(3)The merged precipitation product obtained by spatio-temporal dynamic fusion method proposed in this dissertation can well capture the spatial distribution of precipitation in the Xiang River basin,which shows a high correlation with the gauge observation precipitation.The accuracy and spatial representation are significantly improved compared with IMERG-E.Furthermore,the merged precipitation product significantly improved the performance of flood simulation based on hydrological model,which proves the effectiveness of proposed spatiotemporal dynamic fusion method.