Data-Driven Method to Improve High Resolution Flood Risk Assessments

The global increase in the frequency, intensity, and adverse impacts of natural hazards necessitates comprehensive risk assessments at regional to national scales. Floods are among the most destructive natural disasters, causing significant losses to both social and environmental systems. This dissertation addresses key gaps in current vulnerability and flood risk assessment practices by leveraging machine learning (ML) methodologies. Specifically, three major studies were conducted to advance our understanding and modeling capabilities. In the first Chapter, we have developed a block-level Socio-Economic-Infrastructure Vulnerability (SEIV) Index that helps characterize the spatial variation of vulnerability across the Conterminous United States. The SEIV index provides vulnerability information at the block level, takes building count and the distance to emergency facilities into consideration in addition to common socioeconomic vulnerability measures, and uses a machine-learning algorithm to calculate the relative weight of contributors to improve upon existing vulnerability indices in spatial resolution, comprehensiveness, and subjectivity reduction. The second chapter presents a block-level coastal flood risk assessment framework for the Gulf Coast of the United States, a region highly vulnerable to hydrometeorological extremes. This framework integrates hydroclimatic variables, geomorphological factors, and socio-economic infrastructure indicators (from chapter 1) into a comprehensive flood risk assessment. The framework uses supervised machine learning to assign flood risk categories to each block based on observed flood damages over the past two decades. The third chapter addresses the complex phenomenon of compound flooding, where multiple flood drivers such as storm tide, river discharges, and heavy rainfall interact. Focusing on Galveston Bay, Texas, a deep learning algorithm, convolutional neural network (CNN), specifically a U-Net algorithm, was trained to quantify the relative contributions of each flood driver and delineate flood transition zones, areas where their interactions dominate the flooding dynamics. The final chapter of this dissertation summarizes the key findings and emphasizes the primary contributions of this research to the scientific community. It reflects on the progress made in utilizing data-driven approaches to enhance flood risk assessments. Additionally, the chapter outlines potential future directions that could build upon or further improve the work presented in this study. Overall, this dissertation demonstrates the efficacy of ML and DL techniques in enhancing spatial resolution, reducing subjectivity, and improving the comprehensiveness of flood vulnerability and risk assessments. The frameworks developed are scalable and transferable, offering valuable tools for researchers, planners, and policymakers engaged in disaster risk reduction. Through advancing fine-scale, data-driven flood risk modeling, this work contributes to more equitable, effective, and resilient strategies for managing risk from natural hazards.

Grantor dc:publisher

University of Alabama Libraries

Year dc:date.issued

2025

Author dc:creator

• Yarveysi, Farnaz

Advisor dc:contributor.advisor

• Moradkhani, Hamid

Contributors dc:contributor

• Kumar, Muckesh
• Mekonnen, Mesfin
• Elliott, Mark
• Shao, Wanyun

• Flood Risk
• Hydrologic Modeling
• Machine Learning
• Resilience Planning
• Vulnerability Assessment

Statement dc:rights

• All rights reserved by the author unless otherwise indicated.

Language dc:language.iso

en_US, English

Dc Identifier Other

1175938

OAI identifier oai:identifier

oai:ir.ua.edu:123456789/17054