EXPLAINABLE AI APPROACHES FOR SATELLITE IMAGE INTERPRETATION

Abstract

This paper explores contemporary Explainable Artificial Intelligence (XAI) methods applied to the interpretation and analysis of satellite and aerospace imagery within remote sensing tasks. We present a comprehensive overview of primary categories of XAI techniques – including feature attribution, model distillation, intrinsically interpretable models, and contrastive explanations – and discuss their operational principles, strengths, limitations, and computational characteristics. The practical relevance of XAI for tasks such as natural disaster monitoring, agricultural assessment, socio-economic indicator estimation, and land use/land cover classification is highlighted with real-world examples. Emphasis is placed on how explainability enhances model transparency, reliability, and integration with domain expert knowledge. Challenges such as ambiguous interpretations, high computational costs, and the lack of standardized evaluation metrics for explanations are also discussed. The review underscores XAI’s growing importance in bridging the gap between black-box AI performance and human understanding in Earth observation applications.