Modeling Hurricane Evacuation Traffic: Testing the Gravity and Intervening Opportunity Models as Models of Destination Choice in Hurricane Evacuation [Report]

US Transportation Collection2006PDFTech ReportWilmot, Chester G.Modali, NaveenChen, BinLouisiana State University (Baton Rouge, La.). Department of Civil and Environmental EngineeringLouisiana Transportation Research CenterLouisiana State University. Department of Civil and Environmental EngieeringLouisianaHurricanesEvacuationTraffic modelsEmergency managementTraffic distributionGravity modelsTestingInformation processingEvacuationdestination choicetrip distributionFHWA/LA.06/407LTRC Project No. 03-1SSState Project No. 736-99-1116The test was conducted by estimating the models on a portion of evacuation data from South Carolina following Hurricane Floyd, and then observing how well the models reproduced destination choice at the county level on the remaining data. The tests showed the models predicted destination choice on the remaining data with similar accuracy. The Gravity Model predicted evacuation to friends or relatives in 110 different counties with an average error of 1.55 evacuations over all destinations, while the corresponding error for the IOM was 1.64. For evacuation to hotels or motels in 70 different counties, the Gravity Model gave an average error of 1.48 evacuations and the IOM an average error of 1.50. However, when the IOM was modified to make the sequencing of opportunities sensitive to the direction of evacuation relative to the path of the storm, the modified IOM performed slightly better than the Gravity Model with average errors of 1.55 and 1.43 evacuations to friends and relatives, and motels and hotels, respectively.The transferability of the Gravity Model for evacuations to friends and relatives was also tested in this study by applying the model estimated on the Hurricane Floyd data in South Carolina to data from Hurricane Andrew in Louisiana. Transferability was tested by comparing the trip length frequency distributions from the two data sets, the similarity of friction factors from models estimated on each data set, and the ratio of the Root-MeanSquare-Error (RMSE) of destination predictions of a locally-estimated model to a transferred model on the Andrew data. No significant statistical difference was found between the trip length frequency diagrams or the sets of friction factors at the 95 percent level of significance. The ratio of RMSEs on the Andrew data was 0.67, indicating that the average error of a locally-estimated model was 67 percent that of the transferred model.96