Leveraging Predictive Science, Artificial Intelligence and Citizen Science for Risk-Informed Disaster Damage Assessment

Abstract

A reliable assessment of damage and loss is an important function in the disaster management cycle. A dependable loss prediction model is essential for effective pre-disaster preparedness, that has direct consequence on the severity of the disaster impact. Similarly, robustness and efficiency of post-disaster damage assessment dictates the success and failure of post-disaster response and recovery operations. A thorough damage assessment facilitates an effective and informed response to the disaster impact and results in enhanced performance of post-disaster functions. An inadequate damage assessment may potentially be a result of (a) Inaccurate or insufficient reporting of damage data, (b) Delays in the resource allocation or damage inspection, (c) Unreliable inferences from the collected damage data, and (d) Ignoring the uncertainties associated with the quantified damage metrics.

In this dissertation, some novel techniques to overcome these limitations are proposed, and thereby improving the quality and reliability of the engineering predictions, specifically in the domain of disaster risk-assessment and damage predictions. The idea is to move towards accurate predictions as well as more defensible, transparent, and credible assessment of the underlying uncertainties. The overarching goal is to have an accurate and well-informed estimate of ex-ante risk as well as ex-post disaster evaluation at a higher resolution. Both these aspects of disaster management are essential for effective disaster mitigation and timely recovery of the affected community. In particular, the use of advanced predictive models to quantify the losses, and the use of Artificial Intelligence (AI) and crowdsourcing to assess post-disaster damages is discussed in this dissertation.