Abstract: The rising rate and strength of extreme temperature occurrences, especially heat waves, have left society at risk to health, farming, and infrastructure. All the consequences of the mentioned events are aggravated by climate change, which further promotes heat waves, destroys ecosystems, human health, and economic systems, particularly in cities where the heat island effect is active. They are emerging trends in Artificial Intelligence (AI) in response to these challenges. The inclusion of XAI methods will add Explainability to such models, which will build trust in stakeholders and solidify the practicality of AI-based forecasting systems in designing climate models. Performance indicators of different models show that the LSTM model performs better than other models in terms of capturing delayed and seasonal patterns of the heatwaves with their Mean Absolute Error (MAE) of 1.38 °C, Root Mean Square Error (RMSE) of 1.9800C °C, and R2 value of 0.94. This highlights the effectiveness of deep learning solutions in handling the realities of extreme temperature forecasts. Besides, introducing Distribution-Informed Graph Neural Networks (DI-GNN) has improved the ability to predict rare extreme events further by incorporating Generalized Pareto Distribution-based descriptors. This new method has proved to be better in terms of performance, especially where the distribution of the data is not uniform, and the conventional metrics can be rather deceptive. Advanced AI approaches to machine literacy and resolvable AI are crucial for perfecting models of extreme temperature vaporization. These advances are sensitive to vapour delicacy and understanding climatic factors, enabling effective responses to heat swells.

Keywords: Heatwave prediction, LSTM, Explainable AI, SHAP, Climate modelling, Extreme temperature forecasting

Pages: 233-247