Study On Methods Of Mid-and Long-term Hydrological Statistical Prediction

Mid-term prediction of regional rainfall and long-term prediction of monthly charge are studied in this paper, and the author tries to establish a simple and applicable method with high precision of prediction.In mid-term prediction with a 3-day lead time, Panjiakou Reservoir and Yuecheng Reservoir are adopted as studied areas, and five meteorological elements on 850hpa, 700hpa,500hpa of ten meteorological stations around the country are chosen as pre-correlative factors. The paper adopts some statistical methods to set up the function to predict daily rainfall level within the area. The key method is multi-discriminant method based on Fisher-rule, which can establish a function to predict daily rainfall level through the data of the chosen factors. According to the result for testing, the method has a relatively high precision of prediction, and therefore can be taken into practice.As for study of long-term prediction for monthly charge, Datong Gauge of Yangtze River is chosen as the example, and two kinds of schemes are adopted to set up prediction functions, which are statistical method and artificial neural network model respectively. The correlative factors are three meteorological elements, including sea surface temperature of North Pacific, monthly average value at 500hpa and lOOhpa height in the Northern Hemisphere. Six methods are included in the statistical scheme, which are multi-regression, stepwise regression, stepwise-discriminant analysis, maximum entropy composite analysis, automatic adaptive and auto-regression method, and auto-regression model. Backward propagation is adopted in artificial neural network model. This paper uses data of monthly charge and relative factors from 1964 to 1995 to establish two prediction models. Predictions of these two models are verified by data from 1996 to 2000 at Datong gauge. According to the prediction results, it can be known that both of two schemes can provide reasonable predictions, while the artificial neural network model has a relatively high precision.