Flood Risk Follows Valleys, Not Grids: Graph Neural Networks for Flash Flood Susceptibility Mapping in Himachal Pradesh with Conformal Uncertainty Quantification
arXiv:2603.15681v1 Announce Type: new Abstract: Flash floods are the most destructive natural hazard in Himachal Pradesh (HP), India, causing over 400 fatalities and $1.2 billion in losses in the 2023 monsoon season alone. Existing risk maps treat every pixel independently, ignoring the basic fact that flooding upstream raises risk downstream. We address this with a Graph Neural Network (GraphSAGE) trained on a watershed connectivity graph (460 sub-watersheds, 1,700 directed edges), built from a six-year Sentinel-1 SAR flood inventory (2018-2023, 3,000 events) and 12 environmental variables at 30 m resolution. Four pixel-based ML models (RF, XGBoost, LightGBM, stacking ensemble) serve as baselines. All models are evaluated with leave-one-basin-out spatial cross-validation to avoid the 5-15% AUC inflation of random splits. Conformal prediction produces the first HP susceptibility maps with statistically guaranteed 90% coverage intervals. The GNN achieved AUC = 0.978 +/- 0.017, outper
arXiv:2603.15681v1 Announce Type: new Abstract: Flash floods are the most destructive natural hazard in Himachal Pradesh (HP), India, causing over 400 fatalities and $1.2 billion in losses in the 2023 monsoon season alone. Existing risk maps treat every pixel independently, ignoring the basic fact that flooding upstream raises risk downstream. We address this with a Graph Neural Network (GraphSAGE) trained on a watershed connectivity graph (460 sub-watersheds, 1,700 directed edges), built from a six-year Sentinel-1 SAR flood inventory (2018-2023, 3,000 events) and 12 environmental variables at 30 m resolution. Four pixel-based ML models (RF, XGBoost, LightGBM, stacking ensemble) serve as baselines. All models are evaluated with leave-one-basin-out spatial cross-validation to avoid the 5-15% AUC inflation of random splits. Conformal prediction produces the first HP susceptibility maps with statistically guaranteed 90% coverage intervals. The GNN achieved AUC = 0.978 +/- 0.017, outperforming the best baseline (AUC = 0.881) and the published HP benchmark (AUC = 0.88). The +0.097 gain confirms that river connectivity carries predictive signal that pixel-based models miss. High-susceptibility zones overlap 1,457 km of highways (including 217 km of the Manali-Leh corridor), 2,759 bridges, and 4 major hydroelectric installations. Conformal intervals achieved 82.9% empirical coverage on the held-out 2023 test set; lower coverage in high-risk zones (45-59%) points to SAR label noise as a target for future work.
Executive Summary
This article presents a novel approach to flash flood susceptibility mapping in Himachal Pradesh, India, utilizing Graph Neural Networks (GNNs) and conformal uncertainty quantification. The GNN model outperforms traditional pixel-based machine learning models, achieving an AUC of 0.978. The results highlight the importance of considering river connectivity in flood risk assessment and provide statistically guaranteed 90% coverage intervals for susceptibility maps.
Key Points
- ▸ Graph Neural Networks (GNNs) are used for flash flood susceptibility mapping
- ▸ The GNN model outperforms traditional pixel-based machine learning models
- ▸ Conformal uncertainty quantification provides statistically guaranteed 90% coverage intervals
Merits
Incorporation of River Connectivity
The use of GNNs allows for the consideration of river connectivity, which is a crucial factor in flood risk assessment.
High Accuracy
The GNN model achieves a high AUC of 0.978, indicating excellent predictive performance.
Demerits
SAR Label Noise
The lower empirical coverage in high-risk zones suggests that SAR label noise may be a limitation of the study.
Limited Generalizability
The study is focused on Himachal Pradesh, and the results may not be directly applicable to other regions.
Expert Commentary
The article presents a significant advancement in flood risk assessment, demonstrating the value of incorporating river connectivity into predictive models. The use of GNNs and conformal uncertainty quantification provides a robust and reliable framework for susceptibility mapping. However, further research is needed to address the limitations of the study, including the impact of SAR label noise and the generalizability of the results to other regions. The study's findings have important implications for climate change research, infrastructure planning, and policy development.
Recommendations
- ✓ Future studies should investigate the application of GNNs and conformal uncertainty quantification to other regions and flood risk assessment contexts.
- ✓ Research should focus on addressing the limitations of the study, including the development of methods to mitigate SAR label noise and improve the generalizability of the results.