| Popis: |
The central challenge to flood risk management is in designing flood mitigation practices and strategies that avoid the human and economic costs of under- and over-investment. Designing to avoid these costs requires accurate recurrence probability estimates for decision-relevant flood impacts such as water depths, financial damages, and water volumes. The challenge of estimating flood impact probabilities is exacerbated in coastal settings with non-independent compound flood drivers such as rainfall and storm surge. While techniques for compound flood impact probability assessments have been proposed, insight into the decision relevance of the choice of methodology has not been explored. Our work begins to address this gap by comparing flood-volume exceedance curves resulting from three approaches in a 1.9km2 coastal urban watershed in Norfolk, Virginia. Watershed runoff and storm sewer flow are represented by 1144 links, 1128 nodes, and 869 subcatchments in a U.S. Environmental Protection Agency Stormwater Management Model (SWMM).The first flood impact probability assessment technique follows a traditional design storm approach in which the joint probability of storm surge and rainfall are mapped directly onto the modeled flood volumes. In contrast, the second and third techniques involve modeling many years of stochastically generated rainfall and storm surge time series and empirically estimating the probability distribution of the resulting flood volumes. These two techniques differ in their approach to stochastic weather generation, one fitting a probability distribution to local rainfall observations to allow for extrapolation outside the record, and the other using only historical rainfall observations but across a wider regional domain. Each approach is grounded in statistical and physical theory but leads to different estimates in flood-volume exceedance curves and their associated uncertainty. Since these estimates would influence flood mitigation design, we show that the choice of technique has design implications. |
