Development of XAI Based Model for Prediction of Heavy Impact Rain Using Satellite Data Using Machine Learning

Abstract

This paper develops an Explainable AI (XAI) model to predict heavy rainfall using satellite data.  Forecasting significant rainfall occurrences is crucial for alleviating the detrimental impacts of severe weather phenomena, including floods, landslides, and infrastructure damage. Traditional weather forecasting models, although effective, often lack the fine resolution and adaptability needed for precise predictions, especially in localized areas. Moreover, many machine learning (ML) models, though promising in their predictive power, operate as "black boxes," providing limited interpretability of the underlying processes that lead to predictions. This poses challenges when stakeholders, such as meteorologists, disaster management agencies, and policymakers, require clear explanations to trust and act upon the predictions. This seminar explores the development of a novel Explainable AI (XAI)-based model designed to predict heavy impact rain events using satellite data in conjunction with machine learning techniques. The model leverages a comprehensive array of satellite observations, including atmospheric parameters (e.g., temperature, humidity, pressure), cloud properties (e.g., cloud density, type, and altitude), and surface conditions (e.g., sea surface temperature, vegetation indices). These factors are included into sophisticated machine learning algorithms, such as Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs), to elucidate spatial-temporal correlations essential for predicting rainfall events. This work's primary contribution is the application of XAI techniques to demystify the decision-making process of the ML models. By utilizing methods such as SHAP (Shapley Additive Explanations), LIME (Local Interpretable Model-agnostic Explanations), and attention mechanisms, the approach delivers not only great accuracy in forecasting heavy rainfall but also interpretable insights into the principal elements influencing the predictions. For instance, XAI techniques can emphasize the significance of specific cloud forms, humidity levels, or temperature gradients in forecasting extreme rain events, enabling experts to understand why and how certain weather phenomena are likely to occur.