[A-AS10] 成層圏・対流圏過程とその気候への影響

[AAS10-04] Application of Deep Learning to Estimate Atmospheric Gravity Wave Parameters in Reanalysis Data Sets

★Invited Papers

松岡 大祐¹、*渡辺 真吾¹、佐藤 薫²、川添 祥³、Yu Wei⁴、Easterbrook Steve⁴ (1.国立研究開発法人海洋研究開発機構、2.東京大学大学院理学研究科地球惑星科学専攻、3.北海道大学大学院理学研究院地球惑星科学部門、4.Department of Computer Science, University of Toronto)

キーワード：Deep learning、Gravity waves、Reanalysis dataset

Gravity waves play an essential role in driving and maintaining the global circulation. In order to understand their contribution in the atmosphere, it is important to reproduce their distribution accurately. In this paper, we propose a deep learning method for estimating the momentum flux of gravity waves, and validate its performance at 100 hPa using low-resolution zonal wind, meridional wind, air temperature, and specific humidity (300, 700, and 850 hPa) data over the Hokkaido region. For this purpose, a deep convolutional neural network was trained on 29 years of reanalysis data sets (JRA-55 and DSJRA-55), and final 5 years of data were reserved for evaluation. The results show that the fine momentum flux distribution of gravity waves can be estimated with reasonable computational cost. In particular, the median root mean square error (RMSE) of the maximum momentum flux and the characteristic zonal wavenumber were 0.06-0.13 mPa and 1.0×10-5, respectively, during the winter season when gravity waves are stronger.