[A-CG43] Earth & Environmental Sciences and Artificial Intelligence/Machine Learning

[ACG43-07] Automated identification of tornadic vortex with Doppler radar data using deep learning approaches: Recent progress and challenges

★Invited Papers

*Kusunoki Kenichi¹, Naoki Ishitsu², Toru Adachi¹, Hanako Inoue¹, Osamu Suzuki¹, Ken-ichiro Arai², Chusei Fujiwara³, Hiroto Suzuki³ (1.Meteorological Research Institute, 2.Alpha-denshi Co., Ltd., 3.East Japan Railway Company)

Keywords:tornado, Doppler radar, deep learning

In recent years, deep learning (DL) techniques have been applied in many fields.The Convolutional Neural Network (CNN) is one of the most widely used DL techniques mainly for image processing tasks. The collaborative study between the Meteorological Research Institute and the East Japan Railway company has developed a CNN model to extract vortex pattern from Doppler velocity field while reducing false pattern.

The accuracies of the CNN model is applicable for detection of wintertime tornadic vortices over the coast of the Sea of Japan and has been in operational use since November 2020. In this presentation, we will introduce the basic concepts of our model and further approaches to extend to other regions and/or other seasons especially tornadoes of warm- season over the Pacific coast in Japan.

This study is partly supported by Cabinet Office, Government of Japan, Public/Private R&D Investment Strategic Expansion Program (PRISM).