A Multi-Layer Framework for Explainable and Accountable GeoAI in GIScience

Abstract

Recent advances in Geospatial Artificial Intelligence (GeoAI) have expanded the analytical capabilities of geographic information science (GIScience), enabling data-driven decision-making in disaster response, healthcare accessibility, and environmental justice. However, the growing use of complex spatial machine learning models raises critical challenges related to data representativeness, spatial bias, interpretability, and societal accountability. Despite increasing interest in explainable artificial intelligence, GIScience still lacks integrated frameworks connecting algorithmic behavior with spatial justice and decision-making contexts. This study proposes a multi-layer Ethical and Explainable GeoAI framework comprising four interconnected layers: ethical data representation, algorithmic transparency and explainability, societal impact, and governance. The framework is discussed through three use cases—post-earthquake building damage classification, healthcare accessibility, and green space accessibility—with a small-scale synthetic demonstration for disaster management. Global and local explainability, fairness analysis, and spatial error mapping illustrate that predictive accuracy alone may not capture spatial inequalities, emphasizing the importance of ethically grounded and explainable GeoAI for trustworthy spatial decision-making.