Predictors of hurricane evacuation decisions: A meta-analysis

We systematically review and meta-analyze quantitative prediction models for hurricane evacuation decisions. Drawing on data from 33 prediction models and 29,873 households, we estimate distributions of effects on evacuation decisions for 25 predictors. Mobile home occupancy, evacuation orders, and having an evacuation plan showed the largest positive effects on evacuation, whereas increased age and Black race showed the largest negative effects. These results highlight the importance of both social-economic-structural factors and government action, such as evacuation orders, for enabling evacuation behaviors. Moderator analyses showed that models built using real-hurricane decisions showed larger effects than models of hypothetical decisions, especially for the strongest predictors. Additionally, models in Florida had more consistent results than for other U.S. states, and models with a larger number of covariates showed smaller effect sizes than models with fewer covariates. Importantly, our study improves methodologically and inferentially over previous reviews of this literature.

Routing algorithms as tools for integrating social distancing with emergency evacuation

One of the lessons from the COVID-19 pandemic is the importance of social distancing, even in challenging circumstances such as pre-hurricane evacuation. To explore the implications of integrating social distancing with evacuation operations, we describe this evacuation process as a Capacitated Vehicle Routing Problem (CVRP) and solve it using a DNN (Deep Neural Network)-based solution (Deep Reinforcement Learning) and a non-DNN solution (Sweep Algorithm). A central question is whether Deep Reinforcement Learning provides sufficient extra routing efficiency to accommodate increased social distancing in a time-constrained evacuation operation. We found that, in comparison to the Sweep Algorithm, Deep Reinforcement Learning can provide decision-makers with more efficient routing. However, the evacuation time saved by Deep Reinforcement Learning does not come close to compensating for the extra time required for social distancing, and its advantage disappears as the emergency vehicle capacity approaches the number of people per household.

Network Modeling of Hurricane Evacuation Using Data-Driven Demand and Incident-Induced Capacity Loss Models

The development of a hurricane evacuation simulation model is a crucial task in emergency management and planning. Two major issues affect the reliability of an evacuation model: one is estimations of evacuation traffic based on socioeconomic characteristics, and the other is capacity change and its influence on evacuation outcome due to traffic incidents in the context of hurricanes. Both issues can impact the effectiveness of emergency planning in terms of evacuation order issuance, and evacuation route planning. The proposed research aims to investigate the demand and supply modeling in the context of hurricane evacuations. This methodology created three scenarios for the New York City (NYC) metropolitan area, including one base and two evacuation scenarios with different levels of traffic demand and capacity uncertainty. Observed volume data prior to Hurricane Sandy is collected to model the response curve of the model, and the empirical incident data under actual evacuation conditions are analyzed and modeled. Then, the modeled incidents are incorporated into the planning model modified for evacuation. Simulation results are sampled and compared with observed sensor-based travel times as well as O-D-based trip times of NYC taxi data. The results show that the introduction of incident frequency and duration models can significantly improve the performance of the evacuation model. The results of this approach imply the importance of traffic incident consideration for hurricane evacuation simulation.

Informational Determinants of Large-area Hurricane Evacuations

This study reports on two experiments to investigate the informational determinants of hurricane evacuation decisions (temporal and spatial). Whereas most observational and experimental studies in this domain address the public’s response to forecast information, this study addresses emergency management decisions. Using a subject sample of emergency managers and other public safety leaders, contrasted with a more typical university subject pool, this study presents an experimental design that overcomes the counterfactual problem present in all prior published experiments, by relying on an actual storm (Hurricane Rita) with a known outcome. Several methodological advancements are presented, including the use of an established numeracy instrument, integration of advanced hydrodynamic forecasts, and use of a loss aversion frame to improve generalizability. Results indicate that the availability of additional forecast information (e.g., wind speed, forecast tracks) significantly increases the probability and improves the timing of early voluntary evacuation. However, we observe that more numerate subjects are less likely to avoid relying upon forecast information that is characterized by probability (e.g., the uncertainty in the forecast track, sometimes referred to as the “cone of uncertainty”). Consequently, more numerate emergency managers are almost twice as likely as less numerate ones to provide additional evacuation time to their coastal communities, and they do so by longer than a typical workday (8.8 hours). Results also indicate that subjects knowingly over-evacuate large populations when making spatial mandatory evacuation orders. However, results indicate that numeracy mitigates this effect by more than half in terms of the population subject to mandatory evacuation.