Optimizing the selection of hazard-consistent probabilistic scenarios for long-term regional hurricane loss estimation

Long-term regional hurricane loss estimation is typically conducted by simulating the occurrence of many hurricanes in a region using a historical or synthetic set of hurricanes; estimating losses caused by each hurricane; and using the resulting loss database to calculate loss-exceedence curves for each location. This paper introduces a new way to identify a set of hurricanes to be simulated, so as to substantially reduce the number of events required and thus the computation, while maintaining the spatial coherence of individual scenarios. Using a mixed-integer linear programming formulation, a relatively small subset of all possible hurricanes is identified (tens to a couple hundred), and the annual occurrence probability of each is adjusted so that the regional hazard estimated from the reduced set matches the regional hazard estimated from the full synthetic hurricane simulation. The user can explicitly tradeoff number of hurricanes (i.e., computation) and error based on the intended use of the analysis. The method also allows the user to force errors to be smaller at geographic locations or return periods that are particularly important for a given analysis. In a case study application for North Carolina, the method provides unbiased results with errors that are small enough for practical use.