Study On Performances Of Xin'anjiang Model In Forecasting Streamflow

Hydrological models are effective and convenient tools for streamflow forecasting,among which the Xin'anjiang model,due to its high accuracy and strong adaptability,is extensively used in engineering practice.The previous applications reveal that the key and also the difficulty with the Xin'anjiang model when applied to streamflow forecasting are to estimate its parameters,which highlights certain forecasting performances.With application to the Yuxiakou watershed in the Qing River,this work finishes the Xin'anjiang modelling and the parameter calibration,and their performances in forecasting streamflow are analyzed in detail with two years(1990-1991)of observed six-hourly stream flows.The main works and contributions are summarized as below.(1)The initial soil and river states are regarded as parameters to be optimized,which decreases the man-made bias in setting their values and helps improve the calibration precision by better fitting the simulated to the observed streamflow during the earlier stage of the calibration period.(2)The parameters are optimized by using the cyclic coordinate method(CCM),which though easy to trap into a local optimum is improved by selecting the one with the largest certainty coefficient(CC)among a sample of optimal parameters derived by starting with different initial values generated randomly within their rational bounds.(3)A method is proposed to calculate the steady warm-up period(SWUP),which is determined as the early period before the convergence of a group of CC processes simulated by starting with different initial soil and river states,and the possibility of replacing the SWUP with optimized soil and river states is also discussed.(4)The sensitivity of the final calibrated parameters to their initial values is analyzed by using a large sample of experiments,from which statistical values including the minimum,maximum,expectation,variance,variance coefficient and skew coefficient are calculated to investigate the sensitivity of the initially set values to the final optimal parameters.(5)The consistency of model parameters is analyzed by comparing the CCs between the calibration and forecasting periods in a scatter diagram showing their correlation.(6)A modelling idea of combining the Xin'anjiang with Sacramento is proposed and applied to analyze the complementarity between two models,providing a new thinking in forecasting research.