FireDetXplainer: an explainable artificial intelligence framework for wildfire detection

Abstract

Wildfires pose significant environmental, ecological, and socioeconomic threats, necessitating rapid and reliable detection systems for timely emergency response and disaster mitigation. Recent advances in deep learning have substantially improved wildfire detection accuracy; however, most existing models operate as black-box systems, limiting transparency and reducing user trust in safety-critical applications. This study proposes FireDetXplainer (FDX), an explainable artificial intelligence (XAI) framework designed to enhance the interpretability of deep learning-based wildfire detection while maintaining high predictive performance. The proposed framework integrates convolutional neural network (CNN)-based image classification with explainability techniques to identify the visual regions that contribute most to wildfire detection decisions. By generating intuitive visual explanations, FDX enables users to understand, validate, and trust the model's predictions, thereby supporting transparent and accountable decision-making. Experimental evaluation demonstrates that the proposed framework effectively distinguishes wildfire images from non-fire scenes while providing meaningful visual interpretations that improve model transparency without compromising detection performance. The findings highlight the potential of explainable AI to strengthen the reliability, usability, and practical deployment of intelligent wildfire monitoring systems for environmental surveillance, disaster management, and early warning applications.