Modeling Study of Long-term Forecast of Annual Runoff on the Yingluoxia Station of the Heihe River

Through analyzing the flow changeable regularity on Yingluoxia station of the Heihe River, the annual runoff of the station is divided into three kinds according to frequency 25% and 75% : plenty water year x(i) > 55.0, medium water 42.1 <= x(i) <= 55.0, less water year x(i) < 42.1. By figuring out the shifting frequency, it is found that the annual runoff will have all sorts of possible changes from a state to another state, but the dataset of shifting frequency will be 81.8%. Thereby changeable process of annual runoff are random and mutuality. To forecast the process, the pre - atmosphere cyclic factor and average annual flow forecasting model as well as the time series composition model of average annual flow are set up. It has a highl forecasting precision for annual runoff. Through error evaluation of the two models, it can be found that both models are top level ones. Tests show they are rather steady.