천리안 위성 자료를 활용한 합성곱 순환 신경망 기반 태풍 최대풍속 산출

Abstract

It is crucial to predict the intensity of typhoons since they cause massive casualties and damage in property. Wepropose a model for estimating the maximum wind speed of typhoons using Convolutional Recurrent NeuralNetwork (CRNN). Compared to the current method in investigating typhoons which is fully subjected to themeteorologist’s analyzing skill and domain knowledge, the proposed model assists meteorologists to obtain theobjective analysis of typhoons. In previous studies, they construct the model utilizing only CNN. However, oursuggested model is built with CNN followed by LSTM to consider the fact that the typhoons occur sequentially. We train the model by using each single channel in COMS satellite data composed of IR1, IR2, WV, and SWIR. As a result, the CRNN model trained on WV shows the lowest RMSE error, which is .