Characteristic Analysis Of Precipitation In Rainy Season In Fujian Province And Research Of Artificial Neural Network Forecasting Model

In this paper, we studied the spatial and temporal variable characteristics of precipitation in rainy season in Fujian Province and builded forecasting model of the flood and drought in both first stage and second stage of rainy season by using monthly rainfall data of 25 meteorological stations in Fujian Province, the global 500hpa height data , the Pacific SST data, 74 circulation characteristic values during 1961-2000 and the ways of statistic analysis and artificial neural networks. The results are as follows:i. The precipitation is relatively more and the long-term variable trend is decreasing in the first stage of rainy season; however the precipitation is relatively less and the long-term variable trend is increasing in the second stage of rainy season in Fujian Province. Owing to the effect of landform, the precipitation in the second stage of rain season is decreasing from the line of Jingnan-Yongchun-Fuding to the northwest and southeast, the spatial distribution characteristics of other three rainy seasons and all year precipitation are decreasing from northwest to southeast.ii. The main factors affecting the precipitation in the first stage of rainy season in Fujian Province are Pacific SST in preceding June , the western Pacific subtropical high area index in preceding May, the Asia polar vortex area index in preceding September, the Eurasian meridional circulation index in preceding April to June, the 500hpa height in northwest Asia in preceding spring . The main factors affecting the precipitation in the second stage of rainy season in Fujian Province are Pacific SST in preceding July and August, the Northern Hemisphere polar vortex area index in January in same term, the Pacific polar vortex intensity index in preceding September, the 500hpa height in south Europe in preceding summer.iii. The flood and drought in both first stage and second stage of rainy season in Fujian Province were forecasted by the ways of line regression prediction equation and three kinds of artificial neural network models, the results show: the Backpropagation (BP), Radial basis function (RBF) and Elman neural network models are much better than line regression prediction equation in historical sample fittings , independent sample test, actual prediction ability . The actual prediction ability of BP network model is improved and training velocity is increased by adding momentum factor and using the weight control calculation in the BP network model. Because RBF have advantages of localized approach and fastertraining , its training velocity is fastest in the three neural networks models and actual prediction ability is better too. Elman neural network model is a kind of dynamic neural network and have the localized feedback characters, so Elman's fiting precision is high and the actual prediction ability is better than feedforward neural network model and traditional statistic method.