APPLICATION OF GEOSPATIAL AND STATISTICAL ANALYSIS IN HAZARD ASSESSMENT

Abstract

The extent and severity of the hazard impacts experienced by a region are uncertain and highly dependent on its intensity and the geospatial characteristics of that location (e.g., proximity to the coast). This dissertation considers multiple hazards, including floods, sea-level rise (SLR), and tropical cyclones (TC), and investigates their potential adverse effects on contaminated sites and nuclear power plants (NPPs). Several geospatial and statistical tools are used to assess the hazards' characteristics and their interaction with physical and social features.First, the possibility of inundation at contaminated sites (including brownfield, superfund, and toxics release inventory sites) due to flood and SLR is investigated. A framework is proposed that assigns a score (I-Score) to each site based on the probability of inundation, social vulnerability, and the population of communities near each site. The analysis shows that many contaminated sites with a relatively high likelihood of inundation are located in areas with a high Social Vulnerability Index (SVI) and population. A density-based representation of the I-Score for recognizing potential hotspots is presented. The framework proposed in this study could support policymakers in prioritizing contaminated sites for further assessment and action. Second, geospatial and statistical tools are used to characterize the uncertainty associated with TC predictions and to compare forecasted and observed hurricane track information. Specifically, hurricane track and wind radii (wind radius is the distance from the center of TC that is under the effect of the specific level of wind speed that TC generates) are considered, and uncertainty associated with the timing and location of landfall (in this dissertation is defined as the onset of various storm conditions which is different from the prevalent definition as an intersection of the center of TC with coastline) and the duration of direct storm impacts (e.g., high wind speeds) is characterized. A particular focus is placed on presenting the uncertainty associated with the temporal and spatial characteristics of the TC so that the hazard may be characterized for input into the probabilistic risk assessment (PRA) for NPPs. Third, models are developed to support the external hazard PRAs for NPPs. NPPs may implement a range of actions to prepare for and protect against natural hazards such as hurricanes. These preparatory actions are intended to mitigate the potential damage caused by these events. However, successful implementation of these preparatory actions requires adequate notifications and sufficient warning time. Thus, comprehensive risk assessments require an understanding of TC forecast errors and uncertainty. The variation of forecast errors in specific parameters (e.g., warning time and strong wind duration) relative to TC geospatial location and maximum winds is explored. Statistical models are developed and adopted to predict the adverse effects of TC and their timing. Then, a Bayesian network (BN) is developed to probabilistically model the relationship of multiple variables, including the relationship between TC-generated wind, flood, and warning time. The BN developed in this dissertation can be connected to other BNs related to human reliability analysis and system failure analysis to provide information regarding external hazards (flood, wind, and their timing).