DEEP-LEARNING MODEL FOR PREDICTION OF PRECIPITATION USING MSM -BASIC STUDY ON SPATIAL RANGE AND LEADING TIME-

Abstract

 Frequency of heavy rains is increasing in Japan and therefore it is important to improve the accuracy of precipitation prediction for dam management and appropriate provision of information. In this study, we tried deep learning models using MSMGPV and in-situ precipitation data to predict the precipitation several hours later and studied the possibility of prediction in terms of spatial range of MSMGPV and leading time. The followings are obtained. 1) The accuracy of prediction was generally improved, except for some precipitation patterns, compared to Fujimori et al. (2019) who constructed a deep learning model using only AMeDAS data around the target point. 2) The accuracy of precipitation prediction did not drop with leading time for prediction. 3) The accuracy can be more improved by learning both AMeDAS data and GPV together.