Investigation Of Rainfall Fusion And Stream Flow Prediction By Wavelet Neural Networks

There is a difficulty in describing the real rainfall by any single observational equipment due to the complex mechanism of cloud dynamics.Therefore,the fusion of multiple rainfall information has become a popular research topic in hydrometeorological field.However,most of the researches directly merge the rainfall series and ignore the complex time-frequency composition of the rainfall.In this study,three novel precipitation fusion methods were proposed by investigating the time-frequency component of different rainfall information using wavelet analysis.Meanwhile,the reliability and effectiveness of the rainfall fusion was compared with non-fusion and arithmetic average through streamflow forecasting.The results show that(1)the wavelet-based rainfall fusion model is superior to arithmetic average method,indicating the time-frequency component of rainfall provide better information on rainfall fusion.(2)As compared to arithmetic average and non-fusion scenario,the best fusion model provides an improvement of 42.4%and 48.5-52.6%in terms of streamflow forecasting,respectively.(3)The contribution of gauge measurements and satellite-derived precipitation on rainfall fusion is about 40-85%and 15-30%,respectively.(4)The weights of these two rainfall is different on 4 bands that means the effective time-frequency component observed by different multiplatform(Satellite and Gauge)is distinguishing.