Bi-level optimization for risk-based regional hurricane evacuation planning

Almost all engineering evacuation models define the objective as minimizing the time required to clear the region or total travel time, thus making an implicit assumption that who will or should evacuate is known. Conservatively evacuating everyone who may be affected may be the best strategy for a given storm, but there is a growing recognition that in some places that strategy is no longer viable and in any case, may not be the best alternative by itself. Here, we introduce a new bi-level optimization that reframes the decision more broadly. The upper level develops an evacuation plan that describes, as a hurricane approaches, who should stay and who should leave and when, so as to minimize both risk and travel time. The lower level is a dynamic user equilibrium (DUE) traffic assignment model. The model includes four novel features: (1) it refocuses the decision on the objectives of minimizing both risk and travel time; (2) it allows direct comparison of more alternatives, including for the first time, sheltering-in-place; (3) it uses a hurricane-scenario-based analysis that explicitly represents the critically important uncertainty in hurricane track, intensity, and speed; and (4) it includes a new DUE algorithm that is efficient enough for full-scale hurricane evacuation applications. The model can be used both to provide an evacuation plan and to evaluate a plan’s performance in terms of risk and travel time, assuming the plan is implemented and a specified hurricane scenario then actually occurs. We demonstrate the model with a full-scale case study for Eastern North Carolina.     

Long-term regional hurricane hazard analysis for wind and storm surge

This paper introduces a new method to estimate the long-term regional hurricane wind and storm surge hazard. The output is a relatively small set of hurricane scenarios that together represent the regional hazard. For each scenario, the method produces a hazard-consistent annual occurrence probability, and wind speeds and surge levels throughout the study area. These scenarios can be used for subsequent evacuation or loss estimation modeling. This optimization-based probabilistic scenario (OPS) method involves first simulating tens of thousands of candidate hurricane scenarios with wind speeds and approximate surge depths. A mixed-integer linear optimization is then used to select a subset of scenarios and assign hazard-consistent annual occurrence probabilities to each. Finally, a surge model is used to estimate accurate surge depths for the reduced set of events. The method considers the correlation between winds and surge depths and the spatial correlations of each; it is computationally efficient; and it makes explicit the tradeoff between the number of scenarios selected and the errors introduced by using a reduced set of events. A case study for Eastern North Carolina is presented in which a final set of 97 hurricanes provides unbiased results with errors small enough for many practical uses.