Towards Unbiased Estimation and Probabilistic Forecasting of Flash Flood Human and Economic Impacts

Abstract

Rainfall-induced flash floods are characterized by their rapid onset (usually under 6 hours), small spatial scale, and lethality. Little research on the impacts of flash flood has been done at a small spatial scale consistent with their nature. Existing studies have examined patterns in flash flood damages, fatalities and injuries mostly at large spatial scales (e.g., county), but research on flash flood impacts at the neighborhood scale remains scarce. This knowledge gap might be due to lack of integrated data at such small scale. To fill this research gap and to inform effective prepared-ness and mitigation planning, this study provides models for predicting human harm (fatalities or injuries) and economic losses in flash flood events at the census tract scale (as a surrogate for neighborhood). Observations of historical flash flood events indicate that flash flood damages and human harm conform a zero-inflated and highly-skewed distribution, making it difficult to obtain unbiased statistical inferences from conventional statistical models. This study aims to a) assemble flash flood microdata at the census tract scale by acquiring data from multiple online platforms, b) develop probabilistic models with bias-correction methods for predicting flash flood economic damages, fatalities, and injuries, while examining the correlations between these impacts and the built, natural, and social environments in which the flash flood events occurred; c) couple the flash flood impacts prediction models in a Monte Carlo simulation framework to predict the probability of human harm and economic losses in future flash flood events. These objectives are discussed in three individual chapters in this dissertation as follows: a) A rare event logistic regression model for predicting probability of fatality or injury occurrence in flash flood events; b) A two-part mixed effect model for predicting flash flood economic damages at the census tract scale; c) A Monte Carlo simulation for evaluating future flash flood impacts on society.