In flood control, the decision support information made by short-term severe rainfall prediction methods and flood control support systems are indispensable for quick and appropriate operation. On the other hand, it is difficult to forecast the weather of meso-β to γ scales numerically because of limited computational resources, stability of calculation, time and cost for computer and exactness of calculation. In this paper, based on these background, we develop the qualitative short-term severe rainfall prediction method using artificial intelligence for practical use in order to apply to river site by forecasting the time series variation of distribution of severe rainfall having a lead time of six hours, automatically.