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

2026-04-30 · Source: cs.CV updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Emerging Technologies & Innovation · Depth: Expert, extended

Summary

This study introduces the ADAGE (Alignment between Domain Knowledge And GeoAI Explanation Evaluation) framework, designed to systematically evaluate the alignment between deep learning model explanations and established remote sensing domain knowledge in satellite-based flood mapping. The framework addresses the opacity of deep learning models by employing a Channel-Group SHAP (SHapley Additive exPlanations) method. This method estimates the contributions of grouped input channels to pixel-level predictions, aligning the explanatory level with domain knowledge. Experiments on two satellite-based flood mapping tasks, including multimodal post-flood water extent mapping and SAR-based open/urban area flood mapping, demonstrate ADAGE's ability to quantitatively assess alignment scores. It also helps domain experts identify misaligned explanations, which could indicate novel pattern discovery or spurious correlations, thereby enhancing the trustworthiness and applicability of GeoAI models in operational workflows.

Key takeaway

For Computer Vision Engineers deploying deep learning models for satellite-based flood mapping, you should integrate the ADAGE framework into your model selection process. This allows you to evaluate not only predictive accuracy (IoU) but also the alignment of model explanations with established remote sensing domain knowledge. Prioritize models demonstrating both high performance and strong alignment to ensure trustworthiness and avoid deploying models that might rely on spurious correlations, requiring further investigation if misalignment is detected.

Key insights

The ADAGE framework quantifies alignment between GeoAI model explanations and remote sensing domain knowledge for flood mapping.

Principles

• Model explanations should align with established domain knowledge.
• Channel grouping enables comparable explanation levels.
• High alignment scores indicate trustworthy model reasoning.

Method

The ADAGE framework uses Channel-Group SHAP to estimate grouped input channel contributions to pixel-level predictions. It then calculates an mAP@k alignment score against domain knowledge-derived reference explanations.

In practice

• Use ADAGE to evaluate GeoAI model trustworthiness.
• Identify misaligned explanations for further investigation.
• Select models with high IoU and high alignment scores.

Topics

• Explainable GeoAI
• Satellite Flood Mapping
• ADAGE Framework
• Channel-Group SHAP
• Domain Knowledge Alignment

Best for: Computer Vision Engineer, AI Scientist, Research Scientist, MLOps Engineer

Related on AIssential

• Visualizing Uncertainty: Spatial Maps of Missing and Conflicting Evidence in Deep Learning — UAM visualizes deep learning uncertainty spatially, distinguishing between missing and conflicting evidence for enhanced interpretability.
• Advanced Flood Prediction with Physics-Guided Deep Learning: Combining UNet, FNO, and SAR/Optical Imagery — Embedding hydrodynamic principles into deep learning models significantly improves flood prediction accuracy and physical consistency.
• GeoDisaster: Benchmarking Orchestrated Agents for Operational Disaster Geo-Intelligence — GeoDisaster benchmarks operational geo-intelligence using a multi-agent framework with RCEA for improved disaster reasoning and evidence grounding.
• Textual Supervision Enhances Geospatial Representations in Vision-Language Models — Textual supervision significantly improves geospatial understanding in vision-language models, making language a key complementary modality.
• Land cover and flood type govern the detection limits of satellite-based flood mapping across diverse global flood events — Satellite flood mapping accuracy, even with advanced models like Prithvi-EO-2.0, is critically governed by land cover, flood type, and reference data consistency.
• Graph-based Semantic Calibration Network for Unaligned UAV RGBT Image Semantic Segmentation and A Large-scale Benchmark — GSCNet improves UAV RGBT semantic segmentation by addressing cross-modal misalignment and semantic confusion via graph-based calibration.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.