Hybrid AI Framework Targets Urban Flood Risk and Green Infrastructure Planning

Urban flooding has become a major challenge in rapidly urbanizing regions, driven by growing impervious surfaces, drainage limitations and increasingly extreme rainfall events. A study published in Theoretical and Applied Climatology on 23 October 2026 proposes a new computational framework, called GIStorm-Net, designed to bring together flood-risk prediction and green infrastructure planning in a single decision-support system.

The framework integrates several components that existing approaches typically treat separately: hydrological feature engineering, GIS-based spatial validation, a hybrid flood-risk prediction model combining XGBoost with a Graph Neural Network, and a hybrid Grey Wolf Optimizer–Particle Swarm Optimization (GWO–PSO) algorithm for planning where green infrastructure should be placed. According to the authors, this integration is intended to overcome the fragmentation of hydrological modeling, spatial analysis and green infrastructure planning into independent workflows.

In the reported experiments, GIStorm-Net achieved an accuracy of 0.9798, precision of 0.9504, recall of 0.9466, an F1-score of 0.9485 and a ROC-AUC of 0.9869 for urban flood-risk prediction. These metrics describe the model’s performance on the data used in the study; the authors note that further validation using real-world hydrological observations and field-scale implementation would be needed to strengthen the framework’s practical applicability.

Beyond prediction, the framework includes scenario-based optimization for green infrastructure allocation. Under the optimized allocation strategy, the study estimated potential improvements of 39.6 percent in groundwater recharge and 25.2 percent in runoff reduction. These are model-based scenario estimates rather than measured outcomes from implemented infrastructure.

The authors also describe the framework as a foundation for future integration with satellite-enabled networks and IoT environmental monitoring systems, which could allow remote-sensing observations and real-time environmental sensor data to be incorporated for continuous urban flood-risk monitoring and adaptive planning. As described in the study, these integrations remain prospective extensions rather than features demonstrated in the current work.

Subject of Research: A hybrid machine learning and optimization framework for urban flood-risk prediction and green infrastructure planning

Article Title: GIStorm-Net: a hybrid XGBoost–GNN framework with GWO–PSO optimization for urban flood risk prediction and green infrastructure planning

Article References: Du, W. (2026). GIStorm-Net: a hybrid XGBoost–GNN framework with GWO–PSO optimization for urban flood risk prediction and green infrastructure planning. Theoretical and Applied Climatology, 157(11), Article 746. https://doi.org/10.1007/s00704-026-06695-w

Image Credits: AI Generated

DOI: 10.1007/s00704-026-06695-w

Keywords: urban flooding, flood-risk prediction, machine learning, XGBoost, graph neural network, green infrastructure, optimization, hydrology, GIS, decision support, GIStorm-Net, hybrid