Automated Decision Support to Address Community Resilience Challenges

The 2017 hurricane season revealed that the built environment of US cities is far from resilient to wind and flood damage. To reach a target level of resilience, it is imperative to employ rigorous decision-making approaches to prepare for and mitigate the impact of such events. However, empirically-calibrated and structure-specific vulnerability models, a critical input required to formulate decision-making problems, are not currently available. To provide the innovations necessary to minimize casualties and socio-economic loss due to the failure of our buildings during disasters, a much more proactive approach is needed. Using automatically generated pre-disaster street-view photos and leveraging real-world image recognition, critical visual contents on the images will be automatically identified to determine probabilistic vulnerability models directly linked with the current state of the built environment. Valuable post-disaster datasets collected from recent hurricanes within US will be used to calibrate and validate the methodology.

An overview of our approach.

The year 2017 was the costliest year on US record regarding natural disasters. 16 disaster events surpassed 1B USD in damages, and the total cost of damages in all events exceeded 300B USD (so far). In particular, the three major hurricanes, Harvey, Irma, and Maria, cost more than 250B USD in flood damages. After each event, hundreds of FEMA inspectors must visit affected buildings to verify the loss prior to any approval for financial aid for the owner. With finite resources and a lack of adequate planning procedures, serious backlogs are to be expected. The average wait for an inspection was 45 days in Texas and about a month in Florida. These backlogs have tremendous societal impacts because before the post-disaster asset inspection is complete, no insurance payments can be made, the hurricane victims remain in limbo, and rebuilding (and thus recovery) cannot commence. Critical information necessary for performing post-event inspections, namely asset vulnerability models, is missing. Our three-year research objective is to improve the resilience of the built environment to hurricanes through an automated vision-based system that generates actionable information in the form of probabilistic pre-event prediction and post-event assessment of damages. The central hypothesis is that street-view images contain sufficient information to construct pre-disaster probabilistic vulnerability models for assets in the built environment. Our one-year research objective is to develop preliminary results to demonstrate the feasibility of the central hypothesis in pursuit of external funding. The rationale for this proposal stems from the fact that probabilistic damage prediction/assessment is the most critical input for formulating the decision-making problems under uncertainty targeting the mitigation, preparedness, response, and recovery efforts. An overview of the proposed research is shown in the figure.


Funding Source:

Ali Lenjani
Ph.D. Candidate

Ali received his Bachelor’s and Master’s degree from the Iran University of Science and Technology and the Northern Illinois Univerity in Mechanical Engineering. He is currently a Ph.D. candidate at the Predictive Science Lab at Purdue University.