GeoAI Explanation Alignment with Domain Knowledge in Flood Mapping
A new framework called ADAGE (Alignment between Domain Knowledge And GeoAI Explanation Evaluation) is introduced to systematically assess whether deep learning model explanations for satellite-based flood mapping align with established remote sensing domain knowledge. The increasing number of satellites has improved temporal resolution, making satellite-based flood mapping promising for operational monitoring. Deep learning approaches, a key GeoAI application, show improved predictive performance by learning complex patterns from large remote sensing datasets. However, opaque decision-making remains a major barrier to integrating these models into critical scientific and operational workflows. The ADAGE framework addresses this research gap by evaluating explanation alignment with domain knowledge.
Key facts
- arXiv paper 2604.26051 introduces the ADAGE framework
- ADAGE stands for Alignment between Domain Knowledge And GeoAI Explanation Evaluation
- The framework assesses alignment of GeoAI explanations with remote sensing domain knowledge
- Increasing satellite numbers improve temporal resolution for flood mapping
- Deep learning models show improved predictive performance in flood mapping
- Opaque decision-making is a barrier to integration into critical workflows
- The study focuses on satellite-based flood mapping as a GeoAI application
- The framework systematically evaluates explanation alignment
Entities
Institutions
- arXiv