Extreme value statistics of large climate modeled and observed datasets

High-resolution climate modeling offers a promise of enhanced realism in the simulation of extreme storms over current generation CMIP5 climate models. Analysis of sub-daily model output, necessary to quantify the statistical behavior of these storms, together with finely resolved spatial scales results in multi-terabyte datasets to be processed. Recent advances in extreme value statistical approaches allow the incorporation of physical information in the statistical description of rare events. We use time as a covariate to formulate non-stationary distributions and climate variability indices to more accurately calculate long period daily precipitation return values. We also utilize storm categorization information to further condition these estimates. Performance of selected R routines as incorporated in the parallel analysis tool, ViSit, will be discussed.