Predicting Flash Flood Economic Damage at the Community Scale: Empirical Zero-Inflated Model with Semicontinuous Data

Abstract

Rainfall-induced flash floods are characterized by their rapid onset and small spatial scale. With little lead time for warning, floodwater can accumulate rapidly and its force can damage roads, swamp houses, destroy bridges, and scour out channels. Having data-driven estimates of potential economic losses from flash floods (before they occur) helps authorities make informed decisions about planning and prioritizing mitigation projects. This article provides a probabilistic predictive model to estimate flash flood economic damage at the census tract scale. To simplify model utilization and avoid strong assumptions about property value and replacement costs, the model predicts the total cost of property and infrastructure damages for individual census tracts (expressed in 2019 prices). The model was developed based on a flash flood data set for a 15-year period (2005–2019) in Texas. The data set was assembled by integrating disparate data from multiple platforms. The occurrence of economic damage was found to be a zero-inflated problem. Therefore, we developed a two-part mixed-effect model. The model first estimates the probability that economic damage will occur (zero-inflated part) and then predicts the dollar amount of the economic damage (continuous part). Utilization of the developed model was demonstrated in an application to Harris County, Texas.