Study Of Flood Season Precipitation Prediction Based On Statistical Model

As most of the area in China are in the monsoon climate zone, interannual variability is very strong, which may lead to meteorological disasters, including severe cycles of droughts and floods that have caused serious economic construction and social development in our country. Thus the forecasting of flood season precipitation is urgent and one of the important measures to prevent the damage. On the one hand precipitation is the product of the interaction between small scale system and large scale circulation; on the other hand it has a very close connection with local features and terrain. There are all kinds of nonlinear complex physical process in precipitation forecast, so it is difficult to predict and its development is relatively slow. The complexity of the precipitation and nonlinear determines the importance of flood season precipitation forecast methods.In terms of the rainfall forecast methods, this article choose the neural network prediction model and threshold autoregressive prediction models in numerous dynamic method and statistical method to solve these problems. Both methods have unique information processing ability and calculating ability. Neural network prediction model is good at processing random large amounts of information whose knowledge background which is not clear. Its nonlinear network belongs to nonlinear dynamic system which has very strong self-learning ability. Threshold autoregressive model in the forecast, will take into account the hydrological evolution process itself, and also the role of the early stage of the main influence factors. Through the piecewise linearization method, it can solve the nonlinear problem of hydrological systems. So neural network forecast model and threshold autoregressive model are the more effective models to describe complicated hydrological processes.Based on the flood season precipitation of yiwu as the research object, this paper attempts to use neural network prediction method and threshold autoregressive model respectively to study the flood season precipitation of yiwu.This paper chose the el nino events and the subtropical high ridge line position as antecedent influenced factors which have good response of flood season precipitation of yiwu,49 years of 1959-2007 flood season precipitation as the training sample, and 4 years of flood season precipitation of 2008-2011 as test samples. The two models were analyzed by three aspects:the historical sample fitting precision, the forecast results of independent sample and actual forecast ability. The final result is used to verify the possibility to use the neural network model and the threshold autoregressive model for flood season precipitation forecast in yiwu, and also provides the basis of the theory and practice of flood season precipitation forecast model in the coming years.The main conclusions of this study are summarized as follows:(1)This paper studies the characteristics of flood season precipitation of yiwu.and also the main factors influencing precipitation, proved the flood season precipitation of yiwu has a good response to solar activity, el nino events, and position of subtropical high ridge line. On the basis of many researchers to study conclusion, this paper chose el nino events and the subtropical high ridge line position as impact factors to establish models. The final data results confirmed that it can ignificantly reduce the precipitation forecast error while selecting the el nino events and position of subtropical high ridge line as the early stage of the impact factor.(2)Neural network forecast model and the threshold autoregressive model are effective processing nonlinear hydrologic system model of flood season precipitation forecast of yiwu city. BP neural network forecasting model is better when it comes to the historical sample fitting precision, the forecast results of independent sample and actual prediction ability. However, the data of neural network forecast model and threshold prediction model are insufficient, so the precision of their results needs improving.