Of all climate-related disasters, floods account for the largest average annual losses. Only a limited climatic perspective on floods in the United States exists. This includes the identification of the seasonality and typical mechanisms (e.g., frontal or connective precipitation) important for floods by subregion. Climate change analyses have led to either no clear assessment of changes in flood potential, or to projections of dramatically increased frequency of extreme floods. The anticipated intensification of the atmospheric hydrological cycle and the increased atmospheric moisture holding capacity under warming, render increasing flood risk plausible. However, it is unclear whether the climatic processes associated with extreme floods are well modeled in global and regional climate models, and whether such models provide predictability for assessing the frequency and intensity of rainfall responsible for extreme floods in the United States with spatial specificity relevant for hydrological analysis of floods.
Our work shows that extreme floods (annual exceedance probability less than ~ 0.1) in most river basins in the United States are associated with a distinct atmospheric moisture transport pattern, where the moisture source is typically in the oceans rather than associated with local convection. Over much of the Western United States, we have been able to demonstrate statistical predictability of the annual maximum flood conditional on pre-season Pacific SSTs. For a region in Brazil we are able to demonstrate that the annual maximum flood at each of the stations can be modeled using concurrent large scale, seasonal climate predictors, and a spatial scaling model for the flood process indexed to the drainage area of the site. Consequently, our hypothesis is that river basins aggregate the spatio-temporal climate signal in terms of synoptic and seasonal atmospheric moisture transport in a way that allows empirical connections to be drawn between slowly varying climate fields and the severity, incidence and location of extreme floods over N. America. If these connections can be quantitatively assessed, modeled and understood, then a basis for assessing changes in flood risk using GCMs or empirical methods could be developed for seasonal prediction and for climate change projections.
The research proposed here seeks to develop an exploratory statistical-dynamical approach for “downscaling” flood risk from climate models through an analysis of the causal structure of the entire ocean-atmosphere-land chain of the flood process. This entails (a) use of historical, reanalysis and GCM data for the diagnostic analyses of the causal structure from the spatiotemporal hydroclimatic data associated with the extreme floods in each of the regions of the United States; (b) Bayesian model development for assessing the conditional probability distributions across the causal chain, leading to a conditional flood risk estimate given either GCM state variables or observed/re-analysis data fields, and (c) assessments of projections of flood risk at selected locations for the upcoming season or for a climate change scenario.