Varying Coefficient Regression Models And Application In Hydrological Forecasting

Hydrologic forecasting is very important parts of Hydrology,Reliable hydrological forecasting can provide a powerful guarantee for the usage of water resource and information for reduction of flood and drough disaster losser.Therefore,hydrological forecasting is a very important basic work.Statistic method is widely used in hydrological forecasting.The hydrological process is a nonlinear process.The functional autoregressive model is a powerful method of coping nonlinear time series.This paper introduces four kinds of varying coefficient autoregressive models FCARSE_d(p),FCARSEd(p,q),?-FCAR(p),?-FCARSO(p,q).Based on FCARSE_d(p),?-FCAR(p),?-FCARSO(p,q)models,we establish four hydrologic forecasting models for Xiangtan hydrologic station.We select the optimal model by cross-validation or AICc and estimate the varying coefficient autoregressive model by kernel polynomial smoothing.Four one-step models forecasting result show that the varying coefficient autoregressive model has a good prediction effect.Therefore,The varying coefficient regressive model has a high application value in hydrological forecastingThere may be heteroscedasticity in modeling stage-discharge curve.Introducing based on Box-Cox transformationmodeling stage-discharge curve,apply to Hengyang station in Xiangjiang River Basin.The results show that the Box-Cox transformation model can stabilize the variance better.