We propose to investigate the ability of climate models to represent the higher-order statistics of daily atmospheric and oceanic variations, with emphasis on the statistics of extreme and high-impact weather events. We will evaluate multi-decadal changes in, and anthropogenic impacts on, these statistics in the Inergovernmental Panel on Climate Change (IPCC) Fifth Assessment Report (AR5) climate model simulations of the 20th century and projections for the 21st century. Specifically, we are interested in going beyond means and variances toward the skewness S (the third moment), excess Kurtosis K (the fourth moment), the risk of exceeding two or more standard deviations, and extreme values in any specified period. This interest is motivated by a need to understand, simulate, and predict the deviations of the probability distributions from Normal distributions, especially because such deviations are intimately associated with the statistics of extreme weather events, including but not limited to hurricanes, tropical cyclones, extra-tropical winter storms, summer heat waves, and droughts and floods.

We will use available Coupled Model Intercomparison Project Phase 5 (CMIP5) coupled model simulations of the period 1871-2008 from 20 international modeling centers, generated as part of the IPCC's 20th century climate simulations with prescribed time-varying radiative forcings. We will also use all IPCC model projections of the 21st century climate available at daily resolution. Using Extreme Value theory and also the new framework of Sardeshmukh and Sura (2009), we will quantitatively assess the association between long-term changes in the dominant decadal/multi-decadal modes of climate variability (such as the Pacific Decadal, Atlantic Multi-decadal, and North Atlantic oscillations) and corresponding changes in the statistical moments, probabilities, and return times of exceedances above specified thresholds of daily quantities. Finally, and importantly, we will repeat the entire analysis for 1871-2008 with the newly available global NOAA-CIRES 20th Century reanalysis (20CRv2) project dataset, and with all other available observational datasets for various subsets of this period.

The applicability of the stochastically generated skewed (SGS) probability density functions discovered by the PIs to Extreme Value analysis extends well beyond the weather and climate contexts considered in this proposal. We expect it to bolster traditional approaches based on Generalized Extreme Value (GEV) and Generalized Pareto (GP) distributions in many other contexts as well. Our project will yield extreme value information on all variables in the 1871-2008 20th Century Reanalysis archive, which will be bundled with the archive to facilitate widespread dissemination through our already well-established institutional website.

The proposed work also includes an important educational and training component that will contribute toward the goal of training the next generation of scientists in the areas of stochastic dynamics and extreme value analysis in general and of extreme weather and climate risk analysis in particular.