School of Civil and Environmental Engineering

Ph.D. Thesis Defense Announcement

POST-DISASTER DAMAGE ASSESSMENT USING DEEP LEARNING WITH

SURROGATE DATA

By HUILI HUANG

Advisor:

Dr. Max Mahdi Roozbahani (COC) & Dr. J. David Frost (CEE)

Committee Members:  Dr. B. Aditya Prakash (COC),

Dr. Duen Horng Chau (COC),

Dr. Nimisha Roy (COC)

Date and Time:  May, 5th, 2026. 11:00 AM EST

Location: SEB122/ https://teams.microsoft.com/meet/27868450097026?p=7JAII0s7X3YKKAgSyG   

ABSTRACT
Natural hazards such as earthquakes, hurricanes, floods, wildfires, and landslides continue to cause substantial human and economic losses worldwide. This dissertation addresses the cold-start problem in post-disaster damage assessment, where timely and reliable damage information is urgently needed in the early aftermath of a hazard event, yet event-specific labeled data and expert-verified assessments are often unavailable. To mitigate this challenge, this dissertation investigates the use of surrogate data, defined here as timely, opportunistic, or non-traditional observations that provide proxy evidence of damage, disruption, or recovery when authoritative assessments are delayed. Specifically, this work examines nighttime lights as a proxy for regional disruption and recovery, and ground-view imagery from social media and expert reconnaissance as a source of fine-grained, asset-level damage evidence. Building on these data streams, it develops benchmark datasets, scalable annotation protocols, and computational methods for damage assessment under limited supervision, heterogeneous observations, and domain shift, including ScaleNet for multi-scale representation learning, DASeg for weakly supervised damage localization using vision foundation models, and SeismoMind for zero-shot damage assessment using multimodal large language models. Overall, the findings show that surrogate data, combined with robust and generalizable deep learning frameworks, can substantially mitigate cold-start constraints and enable more timely, scalable, and actionable post-disaster damage assessment.