Prediction of severe winds and rainfalls in Western Siberia using plane rotations

A statistical model for prediction of hazardous phenomena such as severe winds and heavy rainfalls is described. The pattern recognition method based on direct minimization of the errors of recognition is used as the main instrument for modeling and prediction. The construction of the hyperplanes based on error minimization is performed with simultaneous screening of the initial parameters using the same criterion. The algorithm elaborated takes into account different error prices for different kinds of errors and thus effectively distinguishes the “domain” in some multidimensional space of the situations determined as “dangerous.” The domain may be confined by several hyperplanes the number of which depends on the distribution of dangerous situations in multidimensional space. The model comprises the accommodation that takes into account the development of the phenomenon in time and space. The construction was carried out for prediction of winds and rainfall 24 and 36 h in advance for four administrative districts of Western Siberia.