Crowd-Sourced Data, Data-Driven Techniques, and Model Predictive Control to Improve Understanding, Prediction, and Mitigation of Urban Coastal Flooding

Author: ORCID icon
Sadler, Jeffrey, Civil Engineering - School of Engineering and Applied Science, University of Virginia
Goodall, Jonathan, Engineering Systems and Environment, University of Virginia

Warmer average global temperatures have already caused and will likely cause further increased flooding in coastal cities. This increased flooding is due to an increase in average storm intensity and a rise in sea level, results of warmer global temperatures. In support of a coastal city’s ability to adapt in place to increased flood risk, this dissertation focuses on understanding, predicting, and mitigating coastal flooding through four studies. Each of the four studies will focus on one of these three overarching objectives: to understand, to predict, and to mitigate coastal flooding. In the four studies, emerging technologies, approaches, and datasets are investigated as alternatives to more traditional approaches whose limits are pushed by complex urban coastal environments. Geographically, this dissertation focuses on two coastal cities in Southeastern Virginia USA: Norfolk and Virginia Beach, an area particularly vulnerable to coastal flooding due to land subsidence. The first study focuses on the importance of rainfall measurements close to flood-prone intersections. The objective of this study was to quantify the impact of nearby rain gauges in estimating areal rainfall for watersheds which drain to flood-prone street intersections in Virginia Beach. The second study focuses on data-driven prediction of coastal street flooding. This study demonstrates data-driven models using crowd-sourced data as a step toward street-level flood prediction. The third and fourth studies focus on the ability of active stormwater controls, particularly model predictive control (MPC) for mitigating flooding. The third study describes a software tool developed to simulate MPC of a stormwater system using the Environmental Protection Agency Stormwater Management Model (EPA-SWMM5). The fourth and final study provides insight as to how the utility of active stormwater control will change over time given different sea level rise projections and control configurations. This research found that (i) the data from a rain gauge within 0.5 km of a watershed centroid in Virginia Beach can reduce estimation variance by 50-100\%, (ii) using crowd-sourced flood reports as training data, data-driven models can predict within one flood report on average for rainfall event in Norfolk, and (iii) that in addition to a tide gate, MPC can further reduce overall flooding with an average effective percent reduction of 32\% in Norfolk. Norfolk and Virginia Beach are experiencing the effects of sea level rise earlier than many coastal cities. The methods developed and insight gained in this dissertation could be applied to and benefit other coastal cities who, in the coming decades, will likely experience sea level rise effects similar to those currently being experienced in Norfolk and Virginia Beach.

PHD (Doctor of Philosophy)
Coastal Flooding, Model Predictive Control, Stormwater systems, Machine learning, Crowd-sourced Data, Python, Parallel computing, Real-time control
Sponsoring Agency:
National Science FoundationMid-Atlantic Transportation Sustainability Center University Transportation Center
Issued Date: