Dynamical Adjustment of Surface Air Temperature Fields from Large Initial-Condition Ensemble Simulations: Ensemble Spread and Multi-Decadal Trend Attribution

Boreal cold season surface air temperature (SAT) variability and trends attributable to fluctuations in large-scale patterns of sea-level pressure (SLP) anomalies are investigated in a large initial-condition (IC) ensemble of general circulation model climate change simulations using partial least squares (PLS) regression, a statistical method that specifies a predictand time series or field using multiple mutually orthogonal factors generated from a field of predictors. This specification represents a dynamical adjustment to the raw SAT field, which reduces ensemble spread by accounting for the influence of spontaneously-occurring atmospheric circulation changes unique to each ensemble member (EM). When applied independently to the simulated Northern Hemisphere (NH) SAT field from individual EMs, the methodology yields a dynamical adjustment that accounts for nearly 50% of the cold season SAT variance in each EM, validating a previous application of PLS regression to observations. Dynamical adjustment is also applied to SAT trends across the large IC ensemble, enabling the attribution of multi-decadal trends to external forcing and internal variability as well as dynamical and radiative/thermodynamic processes. When applied to the ensemble of SAT and SLP trends over a multi-decadal mid-21st Century reference interval, the methodology yields a dynamical adjustment that explains over 75% of the intra-ensemble variance of the NHSAT warming trend and accounts for about 5% of the ensemble-mean NHSAT warming trend. This methodological framework may be applied to other large IC ensembles, other climatological variables, and across a wider variety of time scales in order to investigate the role of internal variability in modulating the magnitude and patterns of historical and projected climate change.