深層学習によるメソアンサンブル降雨予測の実況補正手法の開発

抄録

In this study, we developed nowcast correction method for ensemble rainfall prediction using deep learning. We adopted convolutional lstm neural network for rainfall prediction model. Our model’s parameter has updated by past 4 years observed radar and MSM (Meso-scale Model) datasets . Furthermore, we applied our model to all ensemble member of MEPS (Meso-scale Ensemble Prediction System) rainfall prediction provided by JMA (Japan Meteorological Agency).

As a result, corrected rainfall prediction's MAE (Mean Absolute Error) has improved. Moreover, ensemble member's range of corrected MEPS can hold observed rainfall.