Deep Learning Approaches for Operational Tropical Cyclone Intensity and Wind Radius Estimation Using Geostationary Satellite Data

Tropical cyclones (TCs) are among the most severe and destructive natural disasters, having a significant negative impact on both the economy and society. In this study, we utilized data from the Geo-KOMPSAT-2A (GK2A) satellite to estimate TC intensity or TC radius of 15 m s^-1 wind (R15) in the western North Pacific using a convolutional neural network (CNN) model. Given that the GK2A data cover only the period since 2019, we applied transfer learning to the model using information learned from previous Communication, Ocean, and Meteorological Satellite (COMS) data, which cover a considerably longer period (2011–2019). The CNN model primarily focuses on learning spatial patterns, which limits its ability to capture temporal patterns. To overcome this limitation, a CNN-LSTM model was implemented, enabling the learning of both spatial and temporal patterns. This model is used for operational purposes and utilizes data from 3 hours earlier, including satellite images, TC intensity, and wind radius. The test results for 2023 showed that the proposed model achieved an RMSE of 2.93 knots for intensity estimation and 13.91 km for R15 estimation, outperforming both official subjective predictions and other advanced deep learning methods. This study suggests the potential application of the model in real-time TC forecasting, contributing to more precise predictions of TC risks and enhancing response strategies.

Acknowledgement. This research was supported by the Korea Institute of Marine Science & Technology Promotion (KIMST) funded by the Korea Coast Guard (RS-2023-00238652, Integrated Satellite-based Applications Development for Korea Coast Guard) and the Ministry of Oceans and Fisheries (RS-2022-KS221667), as well as by the National Research Foundation of Korea (NRF) grant funded by the Korea government (Ministry of Science and ICT) (No. RS-2022-00144325).