Estimation of Tropical Cyclone Intensity from Satellite Data Using Self Attentive – TCNN-BiGRU Approach

Predicting a tropical cyclone's intensity is difficult since it need for human involvement at every stage of the process, from feature extraction through pre-processing to the integration of data from many satellites. There are many reasons why intensity estimation could be difficult, including inconsistent results, a huge quantity of data preprocessing, a difficult problem area, and concerns regarding generalizability. For the purpose of developing a hurricane classification system, this approach presents a GPU-based TCNN-BiGRU architecture. The proposed model achieves better accuracy and root-mean-square error compared to state-of-the-art methods when only satellite photos are available. To further elucidate the learning procedure, they also provide visual representations of learned features and related deconvolutions at many depths. The suggested method utilizes normalization for preprocessing, allowing for faster computations and higher quality results. PCA is used to lessen the number of dimensions and prevent model overfitting. The SA-TCNN-BiGRU Model is used to analyze the results. When compared to the CNN and GRU models, it fares exceptionally well.