IMAGE PREDICTION FOR TORRENTIAL RAIN AREAS

A new method is presented for the prediction of a torrential rain (TR) area, where some atmophysical parameters are used with their given values as criteria most favorable for the occurrence of TR; an over-all examination is done of the relative favorabilities of these factors for the TR production in other regions which are then composited and numeralized by means of the theory of nonlinear mapping with the results plotted (‘reflected') on a weather chart, allowing to make an objective forecast of the TR area. The preliminary results indicate that the technique is able to objectivelycomposite and clearly exhibit principal distribution features of the parameters on the map, thus showing a certain amount of effectiveness for the diagnosis and prediction of a TR area. Regression analysis is used for factor selection to automatically discriminate and locate the jetstream axis with the aid of a computer. In the calculation of the mapping the gradient method is adopted, in the light of the functional properties of the optimized index K, in place of the simple relaxational iteration now in general use, thus getting rid of the non-convergence by the iteration method because of the increased number of the samples used. The square of the gradient modulus 10-4 is set to be the criterion for the iteration convergence. The improved method can meet the requirements of operational forecasts generally with 50 or less iterations.