A long-standing scientific question is whether anthropogenic radiative forcing has affected key modes of natural climate variability. We address this question here using large initial condition ensembles (LEs). LEs provide many different realizations of internal variability superimposed on the underlying climate response to historical forcing changes. Subtraction of the ensemble-average forced response from each realization yields an estimate of residual internal variability (RIV). We seek to determine whether RIV differs significantly from the pure internal variability (PIV) obtained from an unforced control run performed with the same climate model. We rely on LEs and control runs generated with five climate models, conducting statistical hypothesis tests on variability spectra for the temperature of the mid- to upper troposphere (TMT). Tests are performed for TMT averaged over different spatial domains and frequency bands.

In all five LEs, the RIV and PIV distributions of the average tropospheric temperature spectra in the ENSO frequency band are always significantly different. In addition to differences in the overall distributions of band power, the mean and the variance of each RIV band power distribution are always significantly larger than in the corresponding PIV distribution. Significant differences occur for tropospheric temperatures averaged over the tropics and over a global domain.

In the Arctic, warming-induced reduction in sea-ice extent contributes to significant decreases in the amplitude of RIV on annual and semiannual timescales. In addition to analyses of the satellite era, we also consider three different 50-year periods spanning 1950 to 2099. For global and tropical TMT variability on 2- to 7-year ENSO timescales, RIV is almost always significantly larger than PIV, irrespective of the choice of model LE and analysis period. The robust nature of the latter result suggests that future temperature extremes are likely to be influenced not only by human-caused global warming, but also by forced increases in the amplitude of ENSO-driven temperature variability.