Harnessing compound flood risk assessment Utilizing a physics-informed GeoAI surrogate framework considering social...

Area of Science:

• Environmental science
• Geospatial artificial intelligence (GeoAI)
• Climate change adaptation

Background:

• Compound flooding (CF) from storm surge, rainfall, and river discharge poses increasing risks to megadeltas.
• Existing assessments often analyze flood drivers separately and neglect social vulnerability.
• Urbanizing megadeltas face escalating compound flood hazards, necessitating integrated risk assessments.

Purpose of the Study:

• To develop and demonstrate a physics-informed GeoAI (PI-GeoAI) surrogate framework for spatiotemporal compound flooding risk assessment.
• To couple hydrodynamic modeling with supervised machine learning for enhanced flood hazard and social vulnerability analysis.
• To evaluate the impact of mid-century climate forcing scenarios on compound flooding risk in the Greater Bay Area.

Main Methods:

• Developed a PI-GeoAI surrogate framework integrating a hydrodynamic model with Random Forest machine learning.
• Utilized Typhoon Mangkhut data in the Greater Bay Area, China, for framework demonstration.
• Conducted factorial perturbation experiments with 120 mid-century scenarios combining storm surge, precipitation, and river discharge intensifications.

Main Results:

• The Random Forest surrogate achieved 72% within-one-class accuracy in classifying flood damage.
• Compound flooding risk is primarily amplified by storm surge under high-end forcing scenarios.
• Flood depth increased by 1.11× and risk by 1.01×, with significant shifts in damage and risk classes.

Conclusions:

• The PI-GeoAI framework offers a scalable and spatially explicit approach for compound flooding risk evaluation.
• Storm surge is the dominant driver of compound flooding risk amplification in the studied megadelta.
• The study identifies socially differentiated hotspots for targeted adaptation strategies in urbanizing megadeltas.