Japan Geoscience Union Meeting 2019

Presentation information

[J] Oral

A (Atmospheric and Hydrospheric Sciences ) » A-CG Complex & General

[A-CG36] Earth & Environmental Sciences and Artificial Intelligence

Thu. May 30, 2019 10:45 AM - 12:15 PM 106 (1F)

convener:Tomohiko Tomita(Faculty of Advanced Science and Technology, Kumamoto University), Ken-ichi Fukui(Osaka University), Daisuke Matsuoka(Japan Agency for Marine-Earth Science and Technology), Satoshi Ono(Kagoshima University), Chairperson:Tomohiko Tomita(Kumamoto University, Faculty of Advanced Science and Technology)

11:50 AM - 12:05 PM

[ACG36-04] Deep-Learning Based Short-Term Prediction Method for Sea Ice Concentration Using Explanatory Variables

*Issei Kawashima1, Toru Kouyama1, Ryu Sugimoto1, Ryosuke Nakamura1 (1.National Institute of Advanced Industrial Science and Technology)

Keywords:Sea Ice Concentration, Deep Learning, LSTM

Importance of short-term future and middle-term future prediction techniques of sea ice concentration (SIC) has been significantly increasing for navigation of vessels which uses the Northern Sea Route. High frequent updating of the short-term prediction for adopting to SIC is required and thus its calculation load is an important issue same as the accuracy of the prediction. We have been developing a deep-learning based sea ice prediction method and adopting it to Sea of Okhotsk where sea ice coverage dramatically changes during a season. The proposed method can predict SIC immediately once we have sufficient observation or simulation data and the calculation time is very short. In addition, we have confirmed that the predicted SIC has basically good correlation with actual SIC, even though this method only uses previous and current observed SIC by AMSR2. However, considering actual applications, such as navigation for vessels, it is necessary to improve the accuracy of the proposed method as much as possible.

In this study, we proposed a new short-term prediction method for temporal variation of SIC. The new method uses not only SIC but also other variables which can be obtained from observations and/or simulations. The new method can improve the prediction accuracy than that in the previous method.