A Deep Learning Framework for Spatiotemporal Heatwave Prediction in Oman

In recent years, Oman has witnessed a significant rise in temperatures, leading to abrupt heat flashes across various regions. This escalating climate trend underscores the urgency of accurately forecasting heatwave occurrences to mitigate their impact. Although numerous statistical techniques and Machine Learning (ML) and Deep Learning (DL) algorithms have been proposed, the search for the most effective predictive methods remains ongoing.To address this challenge, the present study introduces a hybrid DL framework that integrates Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM) networks, and an Attention mechanism to predict heatwaves in Oman. The framework processes hourly weather data using Conv1D and Max Pooling layers, capturing short-term temporal patterns. These are then fed into an LSTM layer to learn long-range dependencies, while the Attention mechanism highlights the most influential time steps contributing to heatwave events.Both the proposed CNN–LSTM–Attention model and a standalone CNN-1D model were trained and evaluated. Experimental results demonstrate that the hybrid model significantly outperforms its standalone counterpart, achieving superior performance metrics: RMSE of 0.01283, MSE of 0.00016, and an exceptional accuracy of 99.985%.