Based on 15-station flood season precipitation for 1952～2001 over Guizhou,preceding SST of Pacific ocean,and three-factor optimum subset regression analysis method,the relationship between early predictors of different time steps and different time intervals which are composed by Pacific SST field,and rainfall field in flood season is studied.

Baced on the general circulation background,synoptic-climatological cause and external forcing predictor of rainfall of Yunnan in May,the composite model of the long-term weather process of Yunnan in May for rainfall and the optimum subset regression model of rank statistics applied for dryness and wetness forecasting are presented.

Aiming at the shortage of Cobweb and according to some correlative theories, the theory of rough set was imported to solve a best reduced set of attribute-value pairs, and then it was combined with Cobweb algorithm to construct a hierarchical tree.

This paper studies 7 variable selections frequently used, selection of best regression model, and SAS analytic program and then attests its reasonableness and highly valuable application of the best regression model selection offered here by a positive case analysis.