A Deep Learning-Based Study for Cyclone Track Forecasting: Comparative Analysis Using Historical Data from the Bay of Bengal

Accurate cyclone track forecasting is important for disaster management and mitigation. In our research, we have done a comparative study on cyclone track forecasting using deep learning models focusing on the Bay of Bengal. We employ LSTM, Convolutional LSTM, GRU architectures on datasets to train and evaluate the models. We achieve significant accuracy in predicting cyclone trajectories through extensive hyper-parameter tuning and experimentation. This research shows the efficiency and limitations of deep learning models to predict complex dynamics of cyclone tracks. The outcomes of our research focus on the effectiveness of deep learning models predicting complex problems like different dynamics of cyclone trajectory prediction. Multiple training methodologies and orientations are used in this research to demonstrate different insights. Though we use conventional evaluation matrix of machine learning like MSE, MAE, RMSE, R2, we use another evaluation matrix that is average distance error. Average distance error calculates the distance error of prediction plots and actual plots in kilometers. This matrix helps to accurately demonstrate the models’ efficiency in a different dimension. This experiment contributes to the ongoing research of cyclone trajectory forecasting using deep learning models and its practical implications. The findings of the work show the potential of deep learning architectures for accurate cyclone track forecasting that contributes to disaster management, preparedness and loss mitigation.