Dynamic trip distribution models for hurricane evacuation

Abstract

The dissertation presents several methods to estimate time-dependent origin-destination (O-D) trip tables for hurricane evacuation using survey data from hurricane Floyd in South Carolina in 1999. A static disaggregate destination choice model in multinomial logit form was developed considering household characteristics, hurricane threats, and destination socioeconomic and demographic characteristics. Models were developed for persons evacuating to friends or relatives, and hotels or motels separately. None of the household characteristics of evacuees were found to be significant in distinguishing destination choice but destination characteristics were. The model was tested by comparing the observed destination choices with predicted values via trip length distribution analysis and no significant difference was found. Based on a sequential logit model developed by Fu (2004), time-dependent travel demand was estimated. Coupled with trip distribution from the static logit destination choice model and an expansion factor, time-dependent O-D trip tables were computed. Dynamic traffic assignment (DTA) was then performed to provide time-dependent O-D travel costs. Other time-dependent variables included the predicted hurricane striking probabilities on destination zones in each time interval, and the estimated accommodation remaining in each destination zone in each time period. The estimated trip length distribution was similar to the observed values. The destination choice model was estimated for peak and off-peak periods separately and it was found that the parameter estimates differed significantly. The model was further validated with split sample tests and a traffic link count comparison. Degree of transferability of the model to other locations was analyzed. As an alternative, a static gravity model was developed which employed an impedance function that describe evacuee’s travel behavior. The model was estimated through a chi-squared minimization process. The model was found to produce a trip length distribution that was statistically similar to observed values. The static gravity model was extended into a time-dependent version where time-dependent travel cost and distance from the hurricane center to the destination is explained by a negative exponential and Rayleigh function respectively. The time-dependent gravity model was found to perform similarly to the time-dependent logit model in terms of matching observed trip length distributions.