深層学習による降雨予測の時空間方向へのダウンスケーリング手法の開発
キーワード:
Rainfall prediction,
deep learning,
convolutional neural network,
spatio-temporal downscaling
2019 年 25 巻 p. 97-102
詳細
2019 年 25 巻 p. 97-102
In this study, we developed spatio-temporal downscaling method for rainfall prediction using deep learning. We adopted convolutional neural network for statistical downscaling model, and our model can convert 3hours/20km resolution data into 1hour/5km. Our model’s parameter has updated by past 11 years observed radar datasets. Furthermore, we applied our model to GSM (Global Spectral Model) Guidance’s rainfall prediction provided by JMA (Japan Meteorological Agency).
As a result, statistical downscaled rainfall prediction can smoothly describe rainfall area move and quantity change. Moreover, spatio-temporal downscaled rainfall prediction can generate peak time and quantity that is not given by GSM Guidance.
