Statistical methods in the detection and attribution of climate change

Abstract: Over the last two decades, detection and attribution (D&A) of climate changes played a central role in assessing the human influence on climate. A wide variety of statistical models and methods have been used to this purpose over this period. The main goal of this talk is to present a short overview of these, paying attention to discussing the assumptions underlying each approach. While simple comparisons of observations with simulations by climate models have sometimes been used, the most commonly used approach is based on linear regression models (OLS), sometimes assuming error in the predictor (TLS or EIV). I will then shortly discuss some of the methodological issues related to the inference within these models. Part of this discussion will deal with the estimation of large covariance matrices, which is a key issue in D&A.