Hurricane evacuation demand models with a focus on use for prediction in future events
Hurricane evacuation demand models with a focus on use for prediction in future events
复制标题
飓风疏散需求模型，重点用于预测未来事件
DOI：
10.1016/j.tra.2016.02.012
复制
发表时间：
2016
影响因子：
6.4
通讯作者：
S. DeYoung
中科院分区：
文献类型：
--
作者：
Kecheng Xu;R. Davidson;L. Nozick;Tricia Wachtendorf;S. DeYoung
虽然大量的文献存在了解飓风疏散行为，很少有研究开发的模型，可用于预测疏散率在未来的事件。在本文中，我们开发了新的有序概率模型疏散使用的调查数据收集在飓风多发的北卡罗来纳州在2011年和2012年。由于模型中的所有协变量都可以从人口普查或基于位置获得，因此新模型可以用于预测未来任何飓风的疏散率。新模型的样本外预测能力使用交叉验证在单个家庭层面进行评估，并使用飓风艾琳（2011年），飓风伊莎贝尔（2003年）和飓风弗洛伊德（1999年）的可用数据进行汇总。模型的结果也与现有的参与率模型，并从文献中获得的逻辑回归模型进行比较。在单个家庭层面的结果表明，大约70%的家庭的疏散行为将被正确预测。错误平均分为假阳性和假阴性，当实际撤离的人数百分比为零或全部时，准确率增加到100%，当人口被划分时，大约一半的家庭实际撤离时，准确率下降到大约50%。总的结果表明，新的模型相比，现有的，平均总疏散率误差为5个百分点。
Although substantial literature exists on understanding hurricane evacuation behavior, few studies have developed models that can be used for predicting evacuation rates in future events. For this paper, we develop new ordered probit models for evacuation using survey data collected in the hurricane-prone state of North Carolina in 2011 and 2012. Since all covariates in the models are available from the census or based on location, the new models can be applied to predict evacuation rates for any future hurricane. The out-of-sample predictive power of the new models are evaluated at the individual household level using cross validation, and the aggregated level using available data from Hurricane Irene (2011), Hurricane Isabel (2003) and Hurricane Floyd (1999). Model results are also compared with an existing participation rate model, and a logistic regression model available from the literature. Results at the individual household level suggests approximately 70% of households’ evacuation behavior will be predicted correctly. Errors are evenly divided between false positives and false negatives, and with accuracy increasing to 100% as the percentage of people who actually evacuate goes to zero or all and decreasing to about 50% when the population is divided and about half of all households actually evacuate. Aggregate results suggest the new models compare favorably to the available ones, with average aggregate evacuation rate errors of five percentage points.