Title: From Simple Linear Regression to Complex Climate Detection

 

 

Abstract: Linear regression is one of the first statistical tools we learn, yet it plays a central role in detecting and attributing climate change. A common strategy models the observed climate pattern as a linear combination of several model-simulated responses plus a term capturing natural variability. Although this setup appears simple, real climate data violate many classical regression assumptions: predictors are noisy ensemble means, signals are highly correlated, and the error structure is high-dimensional and difficult to estimate. These challenges turn a familiar model into a demanding statistical problem that draws on ideas from errors-in-variables regression, shrinkage, regularization, and high-dimensional inference.  This talk introduces the climate model from first principles and illustrates how a basic regression setup leads to deep methodological questions and opportunities for further development.