Modeling Postearthquake Fire Ignitions Using Generalized Linear (Mixed) Models

Abstract

This paper presents a rigorous approach to statistical modeling of postearthquake fire ignitions and to data collection for such modeling and applies it to late 20th century California. Generalized linear and generalized linear mixed models are used for this application for the first time. The approach recognizes that ignition counts are discrete, examines many possible covariates, and uses a small unit of study to ensure homogeneity in variable values for each area unit. Two data sets were developed to explore the effect of missing ignition data, each with a different assumption about the missing data. For one data set, the recommended model includes instrumental intensity; percentage of land area that is commercial, industrial, or transportation; total building area; percentage of building area that is unreinforced masonry; and people per square kilometer. The other includes the same, except area of high-intensity residential development replaces total building area, and median year built over all housing units is also included. The models should be useful in estimating the number and locations of postearthquake ignitions in future earthquakes.