Applying Dynamical Adjustment Techniques to Quantify Contributions of Variability to North Atlantic Sea Surface Temperature

Quantifying the relative contributions of external forcing and internal variability to North Atlantic Sea Surface Temperatures (NASST) has important implications for attributing and predicting climate changes around the North Atlantic basin. Virtually all previous methods have approached this problem by estimating the externally forced signal directly, either from climate models or using statistical filtering methods. These methods make strong assumptions about the forced variability which are not universally accepted. In this work, we approach this problem in a fundamentally different way that avoids controversial assumptions about the externally forced variability. Specifically, we estimate internal variability directly and then compute the forced variability as the residual. This method uses a machine learning model to estimate the internal variability of a pattern, namely the NASST basin mean, based on other spatial patterns orthogonal to it. The machine learning model is trained on a multi-model set of pre-industrial CMIP simulations. This technique is a form of dynamical adjustment, although our implementation differs from previous versions. Dynamical adjustment mades no assumptions about forced variability, but it does make strong assumptions about internal variability. This method is applied to a suite of CMIP historical simulations and the resulting estimates are compared to other methods that directly estimate the forced signal.