A Hybrid Convolutional Attention Model for Tropical Cyclone Intensity Estimation Using Satellite Images and Scaler Features

Tropical cyclones are notable challenges in disaster management system due to their unpredictability and destructive nature. Accurate intensity calculation of TCs is very important to reduce the damages and to mitigate the damages. Existing methods for estimating cyclone intensity often rely on manual analysis of traditional method, which is lack of the precision and efficiency for real-time applications. In this research we propose the Tropical Cyclone Intensity Convolutional Multi head Attention (TCI-HCMA) model. The proposed model consists of five convolutional blocks for extracting spatial and contextual features followed by a multi head attention layer which capture long-range dependencies and strain feature representation. Additionally, the model incorporates four scaler features such as latitude, longitude, mean sea-level pressure (MSLP) and R35_4qAVG ( average gale-radius) which are concatenated with the extracted features of satellite images. The proposed TCI-HCMA model achieves notable performances with a RMSE of 4.30115, a coefficient of determination(R2) of 0.9751 and MAE of 3.0861. The precise predictions can significantly aid in early warning systems and helping to mitigate the cataclysmal impacts of tropical cyclones on societies an infrastructure