Quantitative Analysis of Primary Attribution Explainable Artificial Intelligence Methods for Remote Sensing Image Classification

XAI methods for image classification vary significantly in quality: The paper demonstrates that different explainability techniques produce different results when applied to the same AI models, suggesting that practitioners cannot assume all XAI methods are equally reliable for understanding how AI systems make decisions about images.

Quantitative evaluation of explanations is necessary: Rather than relying solely on visual inspection of explanations, the authors emphasize the importance of measuring XAI methods against "various categories of desired properties," indicating that responsible AI requires rigorous testing of whether explanations actually reflect what models are doing.

Context matters when choosing explanation methods: The paper provides "insights and recommendations for selecting the most appropriate XAI method(s)," suggesting that different remote sensing applications (such as land use classification or disaster response) may require different explanation approaches rather than a one-size-fits-all solution.

Transparency supports accountability in remote sensing AI: By making their code publicly available and systematically comparing XAI approaches, the authors enable other researchers and practitioners to verify findings and apply these methods responsibly—a key requirement for AI systems used in high-stakes applications like environmental monitoring or urban planning.