A PREDICTION SCHEME FOR THE PRECIPITATION OF SPR BASED ON THE DATA MINING ALGORITHM AND CIRCULATION ANALYSIS

Abstract:

Based on the 74 circulation indexes provided by National Climate Center of China (hereinafter referred to as NCC) and the 24 indexes compiled by NOAA, the study used the C4.5 algorithm in data mining to establish a decision tree prediction model to predict whether the Spring Persistent Rains (hereinafter referred to as SPR) of 55 years (from 1961 to 2015) is more than the normal, and obtained 5 rules to determine whether the SPR is more than the normal. The accuracy rate of the test set, namely “whether the SPR is more than the normal”, is 98.18%. After evaluating the model by conducting ten 10-fold cross validations to take the average value, the test accuracy rate gained is 84%. There are differences between the three types of years with a SPR more than the normal when it comes to intensity and distribution. In spring, they have respective anomalous 850hPa monthly mean wind fields and water-vapor flux distribution, and 700hPa forms the zone where the vertical speed is anomalously negative. As indicated by the results, the SPR prediction model based on the C4.5 algorithm has a high prediction accuracy rate, the model is reasonably and effectively constructed, and the decision rules take comprehensive factors into consideration. The anomalous rainfall and circulation distribution characteristics obtained based on the decision classification results provide new ideas and methods for the climatic prediction of SPR.