Rapid Urban Flood Mapping

Urban flooding is a critical hazard that threatens public safety, disrupts infrastructure, and damages property in cities across the United States. Jackson, Mississippi, has a long history of both pluvial and riverine flooding, leading to repeated inundation of neighborhoods and transportation corridors.

Through this project, Amisha has developed a machine-learning surrogate model trained on high-fidelity hydrodynamic simulations to produce fast, accurate urban flood predictions. By combining physics-based models and advanced data-driven methods, this research aims to support emergency response, planning, and resilience strategies for urban communities.

This research integrates traditional hydrodynamic modeling with machine learning to create a surrogate flood prediction system. The workflow includes:

• Simulating urban flooding with HEC-RAS for multiple historical flood events to generate reference flood maps.

• Extracting geospatial features (terrain elevation, HAND, land cover, distance to streams, rainfall, etc.) that influence flood behavior.

• Training a Random Forest surrogate model on these data to predict flood depths and inundation extents rapidly.

This hybrid modeling approach leverages detailed physics-based simulations to teach the machine-learning model how urban floodwater distributes under varied conditions.