Interactive web-based Geospatial eXplainable Artificial Intelligence for AI model output exploration
Keywords: Wildfire Susceptibility, Explainable Artificial Intelligence (XAI), GeoXAI, GIS, Random Forest
Abstract. This case study presents a web-based Geospatial eXplainable Artificial Intelligence (GeoXAI) system demonstrated through a case study for wildfire susceptibility assessment. Addressing limitations in traditional GeoXAI tools, the system integrates XAI methods with open-source geospatial technologies. Using a Random Forest model, the system combines environmental, topographic, and meteorological features to provide global and local insights. SHAP values offer feature-level explanations, while the interactive platform enables users to visualize wildfire susceptibility, examine feature contributions, and correlate predictions with spatial patterns and distribution of feature values. This approach tries to enhance transparency in AI-driven environmental decision support systems, with a specific focus on the interpretability of model output.