A cross-disciplinary team of statisticians and climate scientists are to develop and utilize several novel statistical approaches for the analysis of large non-Gaussian data sets primarily in climate science applications. Applications to be explored include a) climate change signal detection and climate model evaluation, and b) investigations of large non-Gaussian fields in climate science that have skewed, heavy tailed or multimodal distributions. New methods that do not rely on Gaussian assumptions include covariance models, Bayesian models for non-Gaussian fields, and sequential inference. Such techniques are needed to overcome the known limitations of using empirical orthogonal functions, which while both popular and simple to use, do not faithfully characterize the entire distributions in many important climate questions ranging from El Nino phenomena, the analysis of precipitation fields and state estimation for non-linear climate systems. Robust and reliable statistical results from the analysis of climate change questions are increasingly vital for societal, economic and political decision making. A key issue in assessing contending climate models is the quantification of model bias, the dependence of which will be investigated.
An important educational goal of the project will enable the cross-disciplinary research and training of Ph.D. students across both lead disciplines, statistics and climatology.
