Uncertainty inside each temperature dataset now outweighs differences between datasets

Why it matters

Studies that estimate climate sensitivity from observed warming have mostly compared datasets and ignored each dataset’s internal ensemble. The authors report that this understates observational uncertainty. Studies of this kind may need to use the full ensembles to give honest ranges.

What they did

The authors used the MIT Earth System Model with a Bayesian fingerprint-style comparison to observed decadal warming in four latitude bands. They ran it on NOAA (1000 members), GISTEMP (200), HadCRUT5 (200) and Berkeley Earth uncertainty bounds. They split NOAA’s ensemble into land and ocean parts and compared infilled with non-infilled HadCRUT5. They also fed 600 ensemble members into a global energy balance calculation from Sherwood et al. (2020).

Key findings

• Medians of climate sensitivity from the newer datasets span 3.26 to 3.46 °C, down from 2.93 to 3.37 °C with older versions, a 54.6% smaller range.
• Within one dataset, NOAA’s ensemble gives median climate sensitivity from 2.82 to 3.97 °C, five to six times the spread between datasets.
• Land temperature uncertainty adds about 11 times more spread to climate sensitivity, and about 7 times more to TCR, than ocean uncertainty.
• Spatial infilling raises climate sensitivity estimates by about 0.2 °C.
• In the global energy balance calculation, the median estimate ranges across ensemble members from 2.47 to 3.59 °C, a spread of 1.11 °C.

Limitations

• Decadal, four-band averaging squeezes the data to 28 points, so it cannot capture the pattern effect of ocean warming or the full spatial detail of ensemble members.
• Each data center builds its ensemble differently, and Berkeley Earth gives only minimum and maximum bounds. Ocean heat content data were the same for all runs.
• The Sherwood et al. calculation already includes some observational uncertainty, so using full ensembles there may double count it. This is a preprint that has not been peer reviewed.

Glossary

• Equilibrium climate sensitivity (ECS): Long-term global surface warming after atmospheric CO2 doubles and the climate settles.
• Transient climate response (TCR): Global warming at the moment CO2 has doubled when it rises 1% per year.
• Spatial infilling: Estimating temperatures in grid cells with few observations from nearby observed areas.
• Pattern effect: How the spatial layout of sea surface warming changes the planet’s radiative response and feedbacks.