Prediction of Precipitation by Aneural Network Method

概要

論文の詳細を見る
A neural network method to reduce natural disasters (particularly, avalanches, slush flow, and melting snow in snow hazards) was used to predict precipitation on the ground. Based on computation cost and similar pattern two types of simulations were made. Two methods were used to evaluate precipitation, mean square error which is superior for evaluating the strength of precipitation over the prediction range and Critical Success Index (CSI) evaluation used to judge the existence of precipitation. Results were compared with prediction results from the short-time precipitation forecast method used by the Japan Meteorological Agency. Mean square error was about 0.1 to 0.2 mm/h toward the term when there was much precipitation whether rain or snow. In the CSI evaluation, the neural network method gave high values as compared with the short time precipitation forecast method for the strong winter characteristics. Results of these two evaluations show that our method can adequately predict for the subsequent hour and is a practical tool for reducing snow hazards.
日本自然災害学会の論文