Grainger College of Engineering, All Events

Multi-Scale Dynamics of Urban Flood Persistence: Forcing Structure, Subsurface Memory,

and Reduced-Order Sewer Hydraulics for Probabilistic Prediction

Advisor: Professor Marcelo Garcia

Abstract:

Urban flooding in large cities is shaped by the interaction of intense rainfall, complex sewernetwork

hydraulics, and slowly evolving subsurface conditions that carry memory from one storm

to the next. While peak rainfall intensity receives the most attention, flood damage — particularly

basement flooding in combined-sewer systems — is often driven more by the duration and

persistence of surcharge than by its peak magnitude. Understanding and predicting this persistence

requires resolving physical processes that span minutes (hydraulic transients in sewer networks),

hours to days (subsurface storage and recession), and seasonal timescales (antecedent wetness

conditioning) — a multi-scale challenge that existing urban flood models do not address in an

integrated manner.

This dissertation investigates the fundamental multi-scale dynamics that control urban flood

persistence, developed and demonstrated on a city-scale Chicago drainage network. Five

interconnected studies form a computational pipeline spanning observation science, atmospheric

characterization, dynamical reduction, and coupled modeling. First, a vulnerability analysis of

multi-year flood impact records reveals that different flood pathways — street flooding versus

basement flooding — exhibit distinct temporal memory structures, establishing persistence as the

scientifically appropriate prediction target. Second, a forcing–response analysis identifies

physically interpretable storm-structure metrics that substantially reduce uncertainty in the

hydraulic response mode, with residual uncertainty governed by antecedent subsurface conditions.

Third, a constrained downscaling framework produces kilometer-scale soil moisture fields from

satellite observations, bridging the resolution gap between remote sensing and urban drainage

applications. Fourth, a physics-structured reduced-order model compresses the high-dimensional

sewer hydraulics into a low-dimensional representation with explicit stability guarantees,

achieving the computational speedup necessary for ensemble simulation. Fifth, a slow–fast

coupling framework connects the satellite-constrained wetness state to the fast hydraulic model

through bounded exchange, enabling explicit representation of how antecedent conditions

modulate flood duration and recovery.

Together, these contributions resolve the physics that underpins urban flood persistence and,

consequently, enable ensemble-based probabilistic prediction with hundreds of members

completing within the warning window required for urban flash floods. The results provide a

scientific foundation for the next generation of impact-oriented, persistence-aware urban flood

forecasting systems.