Statistical Models For Predicting Summer Rainfall And Its Extremes In China

In this paper we used the reanalysis data and the observation data of precipitation to predict the East Asian summer monsoon index based on multiple linear regression. After increasing the Qinghai-Tibet Plateau snow items, we established models to predict the expected values of the summer rainfall in stations, analysised the distribution of the percentage of precipitation anomaly, then used the generalized extreme value distribution to infer the distribution of extreme precipitation in summer and got the following conclusions.From 1979 to 2010, statistical regression tests show that the East Asian summer monsoon index had significant correlation with a small part of the summer rainfall in stations, therefore we tried to skip the East Asian summer monsoon index and directly used the factors to predict site precipitation. Considering precipitation in single station existed noise signal, so we classified all the sites to 16 partitions and established models that firstly predict the regional precipitation then estimate the rainfall in site. Comparison the result by all models from 2005 to 2009, we found the regression models that factors directly predicted precipitation in sites better than the models that used the summer monsoon rainfall index to indirectly predict the site rainfall; and the models factors predict regional precipitation then estimated station rainfall better than factors forecast precipitation in sites.The results of precipitation data processing using different approaches show that, in the same conditions, standardizing the logarithmic data in model is better than directly standardizing the data. Comparison of prediction score and anomaly correlation coefficient, we got the "best single model". From 2005 to 2009, its the average of prediction score is 76.2 and the mean value of anomaly correlation coefficient is 0.261; while it got better effect of prediction in Northeast, North China, Huang-Huai, South China and northern Xinjiang.Only used one model to predict summer rainfall for all sites, there are still some limitations. In practice, we try to establish assemble model to predict summer rainfall. In 2010, correct symbol rate of the percentage of precipitation anomaly is 55%, P score is 71, they are higher than average value of 16 single-model results, while higher than the "best single model" whose correct rate is 51.9% and P score is 67.6.Combinations of the above results, in general there are better effect of prediction in the Northeast, North China, the Yangtze River and South China, these areas are almost in eastern china. In practice, do not only use one model in the process of forecasting distribution of summer rainfall anomaly. And we can incorporate the results of all models to get the distribution of flood-drought in the predictive year.Using summer precipitation values obtained by "best single model" to infer the distribution of extreme precipitation in flood season from 2005 to 2009. Generally speaking, the effect of prediction in the northern region better than in the South, but there are many underreporting sites in all stations. Only in 2005 and 2008, the effects of prediction are better, and the right stations often appear in the northeast and the Yangtze River. For the case of 2010, the results of assemble model are better than the "best single model", the underreporting sites are also less. IV...