Robust detection of forced warming in the presence of potentially large climate variability

Warming of the climate system is unequivocal and substantially exceeds unforced internal climate variability. Detection and attribution (D&A) employs spatio-temporal fingerprints of the externally forced climate response to assess the magnitude of a climate signal, such as the multi-decadal global temperature trend, while internal variability is often estimated from unforced (“control”) segments of climate model simulations (e.g. Santer et al. 2019). Estimates of the exact magnitude of decadal-scale internal variability, however, remain uncertain and are limited by relatively short observed records, their entanglement with the forced response, and considerable spread of simulated variability across climate models. Hence, a limitation of D&A is that robustness and confidence levels depend on the ability of climate models to correctly simulate internal variability (Bindoff et al., 2013).

For example, the large spread in simulated internal variability across climate models implies that the observed 40-year global mean temperature trend of about 0.76°C (1980-2019) would exceed the standard deviation of internally generated variability of a set of `low variability' models by far (> 5σ), corresponding to vanishingly small probabilities if taken at face value. But the observed trend would exceed the standard deviation of a few `high-variability' climate models `only' by a factor of about two, thus unlikely to be internally generated but not practically impossible given unavoidable climate system and observational uncertainties. This illustrates the key role of model uncertainty in the simulation of internal variability for D&A confidence estimates.

Here we use a novel statistical learning method to extract a fingerprint of climate change that is robust towards model differences and internal variability, even of large amplitude. We demonstrate that externally forced warming is distinct from internal variability and detectable with high confidence on any state-of-the-art climate model, even those that simulate the largest magnitude of unforced multi-decadal variability. Based on the median of all models, it is extremely likely that more than 85% of the observed warming trend over the last 40 years is externally driven. Detection remains robust even if their main modes of decadal variability would be scaled by a factor of two. It is extremely likely that at least 55% of the observed warming trend over the last 40 years cannot be explained by internal variability irrespective of which climate model’s natural variability estimates are used.

Our analysis helps to address this limitation in attributing warming to external forcing and provides a novel perspective for quantifying the magnitude of forced climate change even under uncertain but potentially large multi-decadal internal climate variability. This opens new opportunities to make D&A fingerprints robust in the presence of poorly quantified yet important features inextricably linked to model structural uncertainty, and the methodology may contribute to more robust detection and attribution of climate change to its various drivers.

 

Bindoff, N.L., et al., 2013. Detection and attribution of climate change: from global to regional. IPCC AR5, WG1, Chapter 10.

Santer, B.D., et al., 2019. Celebrating the anniversary of three key events in climate change science. Nat Clim Change 9(3), pp. 180-182.