A time series image prediction method combining a CNN and LSTM and its application in typhoon track prediction

Peng Lü, Ao Sun, Mingyu Xu, Zhenhua Wang, Zongsheng Zheng, Yating Xie, Wenjuan Wang

Mathematical Biosciences & Engineering · 01 January 2022

In one sentence. DeepTyphoon couples hybrid dilated CNN and LSTM on satellite image sequences to predict typhoon tracks, achieving 64.17 km mean error at 6 hours.

Abstract

Typhoon forecasting has always been a vital function of the meteorological department. Accurate typhoon forecasts can provide a priori information for the relevant meteorological departments and help make more scientific decisions to reduce the losses caused by typhoons. However, current mainstream typhoon forecast methods are very challenging and expensive due to the complexity of typhoon motion and the scarcity of ocean observation stations. In this paper, we propose a typhoon track prediction model, DeepTyphoon, which integrates convolutional neural networks and long short-term memory (LSTM). To establish the relationship between the satellite image and the typhoon center, we mark the typhoon center on the satellite image. Then, we use hybrid dilated convolution to extract the cloud features of the typhoon from satellite images and use LSTM to predict these features. Finally, we detect the location of the typhoon according to the predictive markers in the output image. Experiments are conducted using 13, 400 satellite images of time series of the Northwest Pacific from 1980 to 2020 and 8420 satellite images of time series of the Southwest Pacific released by the Japan Meteorological Agency. From the experimentation, the mean average error of the 6-hour typhoon prediction result is 64.17 km, which shows that the DeepTyphoon prediction model significantly outperforms existing deep learning approaches. It achieves successful typhoon track prediction based on satellite images.

Labels

Labels and summary are generated by Claude and reviewed weekly. Wrong? Edit this entry.