Interactive Convolution Based Multi-Source Data Fusion for Typhoon Intensity Prediction

Combining meteorological data from different sources can significantly improve the accuracy of typhoon intensity prediction. However, existing methods still have limitations in mining the correlation of multi-source data such as satellite images, meteorological observation data, and historical typhoon records. To address this challenge, we propose an interactive convolution based multi-source data fusion method IC-MDF in this paper for typhoon intensity prediction. Firstly, a dual-branch spatial-temporal feature extraction network (Du-STN) is proposed to extract the hidden features of meteorological sequences from multiple perspectives, including time, space and wind direction. Secondly, an Interactive Convolution (IC) module is constructed to fully capture the complex associations and fuse different data sources. The experimental results demonstrate that compared with the existing methods, IC-MDF has improvement in prediction accuracy, robustness and generalization ability, providing an effective new scheme for typhoon intensity prediction.