Evaluating the Alignment Between GeoAI Explanations and Domain Knowledge in Satellite-Based Flood Mapping

Researchers Hyunho Lee and Wenwen Li have introduced the ADAGE (Alignment between Domain Knowledge And GeoAI Explanation Evaluation) framework, designed to systematically evaluate how well deep learning model explanations for satellite-based flood mapping align with established remote sensing knowledge. The framework addresses a critical barrier to integrating deep learning into operational flood monitoring: the opaque decision-making processes of these models. ADAGE employs a Channel-Group SHAP (SHapley Additive exPlanations) method to estimate the contributions of grouped input channels to pixel-level predictions, focusing on the distinctive spectral properties of Earth's surface.

Experiments on two satellite-based flood mapping tasks demonstrated that ADAGE can quantitatively assess alignment between model explanations and reference explanations derived from domain knowledge. It also helps domain experts identify misaligned explanations through alignment scores. This work contributes to bridging the gap between explainability and domain knowledge in Geospatial Artificial Intelligence (GeoAI) for Earth observation, enhancing the applicability of GeoAI models in scientific and operational workflows. The paper includes 21 pages, 6 figures, and 5 tables, and is available on arXiv (2604.26051).