DEVELOPMENT OF CORRECTION METHOD OF METEOLOGICAL RADAR FOR PREFECTURAL RESERVOIR OPERATION BY USING A DEEP LEARNING

Abstract

 High-frequency and high-accuracy spatial distribution information of precipitation is required for disaster prevention and water resource management especially in small river basins due to its short runoff response time in hydrology. However, many small reservoirs are suffering the lack of such information because of poor surface gauging system and it seemed hard to grasp precipitation over the basin area. C-band radar rainfall gauging system provided by the Ministry of Land, Infrastructure, Transport and Tourism (MLIT), which is delivered in short intervals of five minutes, might be a solution for it because of its spatial and temporal resolution, but there is still room for improvement of accuracy. On the other hand, Japan Meteorological Agency (JMA) radar-raingauge analyzed precipitation data is more accurate but less frequent for example once in thirty minutes. In this study, the deep learning method was tested for improving accuracy of C-band radar-based rainfall by using JMA radar-raingauge analyzed precipitation data as supervised information. It was shown that driving a distributed hydrological model with the corrected rainfall reproduced well the reservoir discharge.