To ensure financial soundness of their businesses, the insurance industry is increasingly relying on the use of natural catastrophe models that reflect a much wider spectrum of plausible event extents, hazard intensities and probabilities. While catastrophe earthquake and hurricane models are very well-established in the market, catastrophe flood models are a relatively new development. Leveraging the increased electronic data availability, tremendous increase in computing power and cheaper storage, a few catastrophe flood models have been developed in recent years. Typically, these models are developed at country or continental scales, enabling flood risk assessment at a high spatial resolution, and are based on thousands of years of simulations. For this, comprehensive high-resolution continuous precipitation modeling, including running global circulation and regional numerical weather models in a coupled framework, is done. Conceptual rainfall-runoff models are then employed to estimate riverine and pluvial flood hazard in both space and time, also enabling classification of each flood event. The modeling community has generally found and opined that physically-based simulation models are better suited for catastrophe flood risk models as these adequately preserve the spatio-temporal meteorological and river basin dynamics inherent in the flooding process. However, such modeling effort comes with an excessive time, computational, and economic costs. Statistical simulation approaches have the potential to significantly reduce these costs but generating realistic flood events exhibiting inherent correlation has been challenging. This study presents a more thorough treatment of inherent spatio-temporal correlations and river dynamics in statistically generating flood events employing copula-based methods. Application of dynamic copula framework and nested copulas is illustrated for statistically generating thousands of flood events for a sub-basin in the Ohio river basin. The advantages, challenges, and limitations of this statistical approach over the physically-based modeling are also presented.
Note: Authors prefer this to be included as a poster and not as an oral presentation.
