Regional Heatwave Prediction Using Deep Learning Models In Bangladesh

With the growing threat of heatwaves in Bangladesh due to climate change, predicting heatwave days has become vital for effective mitigation measures. In this study, a robust dataset of 25 years, acquired through the Bangladesh Meteorological Department, is capitalized to explore Long-Short Term Memory (LSTM), Bi-directional LSTM (Bi-LSTM), and Convolutional Neural Network (CNN) models for predicting heatwave in four heatwave-prone districts of Bangladesh: Dhaka, Rajshahi, Bogra and Dinajpur. The selected features are maximum temperature, humidity, dew point temperature, pressure, wind speed and sunshine. The dataset was divided into 80% for training, 10% for validation and 10% for testing. The predictive performance of Bi- LSTM, Conv1D, and LSTM models was evaluated with Accuracy, Fl-score, R2 score. Bi-LSTM consistently showed high accuracy, with Dhaka and Rajshahi achieving 96.96% and 97.40%, respec-tively, and Bogra recorded the highest accuracy at 99.57%. In Dhaka, Bi-LSTM had the highest accuracy at 96.96%, followed by Conv1D at 95.65%, and LSTM at 93.04%. For the F1-score, Bi-LSTM led with 86.27%, and in R2 score, it performed best with 77.40%. These results demonstrate the superior performance of the Bi-LSTM model in predicting heatwave days, making it a beneficial tool for early warning and preparedness efforts. Despite the predominant use of statistical models on Bangladeshi datasets, this study explores the novel application of deep learning models like Bi-LSTM and CNNs for heatwave classification, which offer better identification of complex spatiotemporal patterns and are more resistant to anomalies.