DeepTC: ConvLSTM Network for Trajectory Prediction of Tropical Cyclone using Spatiotemporal Atmospheric Simulation Data

Seongchan Kim, Ji-Sun Kang, Minho Lee, Sa-kwang Song

Keywords: Tropical Cyclone Path Prediction, Convolutional LSTM, Weather Research and Forecasting model

TL;DR: We introduces a tropical cyclone path prediction model called DeepTC.

Abstract: Accurate forecasting of tropical cyclone trajectory is important because it can have a great impact on the safety of people and infrastructure. This paper introduces a novel data-driven tropical cyclone path prediction model called DeepTC. The proposed model makes use of data generated using the Weather Research and Forecasting model, which simulates spatiotemporal atmospheric conditions. Additionally, the proposed model utilizes convolutional long short-term memory, which is effective when operating on spatial data over time. Experimental results demonstrate that our methodology is promising, confirming that DeepTC learns the spatiotemporal dynamics of the atmosphere by the WRF effectively.