There’s a really interesting paper by Mark Richardson, Kevin Cowtan, Ed Hawkins, and Martin Stolpe called Reconciled climate response estimates from climate models and the energy budget of Earth. Ed already has post that explains it all very clearly, so I won’t need to go into too much detail.
The basic issue is that energy balance estimates for climate sensitivity use observed changes in temperature, and estimated changes in forcing, to determine the climate sensitivity. The results tend to suggest values lower than suggested by climate models. However, the temperature datasets used have some coverage bias (not all areas have coverage) and the temperature measurements vary from being air temperatures over land to sea surface temperaures over the oceans. For climate models, on the other hand, there is no coverage bias and the temperatures are typically air temperatures everywhere. Therefore, it’s not a like-for-like comparison.Richardson et al. (2016) illustrate that this can make quite a difference. The figure on the left shows the HadCRUT4 temperatures (grey line), together with the models temperatures with full coverage and using air temperatures everywhere (red line), model temperatures determined using air and sea surface temperatures (red dotted), and model temperatures using air and sea surface temperatures and also masked to account for coverage bias in the HadCRUT4 data (blue triangles). The effect is not trivial, with as much as 0.2oC warming missing due to coverage bias and the use of sea surface, rather than air, temperatures. The figure on the right shows the impact on climate sensitivity etimates (Transient Climate Response (TCR) only). The bottom blue bar shows the original Otto et al. result. The next is the Otto et al. method, but updated using Lewis & Curry forcings. The one above that, however, is the CMIP5 result, but using blended temperatures (air over land, and sea surface over ocean) and masked to compensate for coverage bias in the observational dataset. Essentially, they all produce best estimates of around 1.4oC.
The top two bars in the figure on the right, however, show the CMIP5 estimate using air temperatures only (top) and an observationally-based estimate that uses inferred air temperatures. They produce best estimates of 1.8oC (CMIP5 tas) and 1.7o (observationally-based, inferred tas). The key point is that if you correct the models by using blended temperatures and account for coverage bias, you get a result that is consistent with the observationally-based estimates. Similarly, if you correct the observationally-based estimates to account for blended temperatures and coverage bias, you get a result consistent with the model result. Essentially, this is an argument that there isn’t necessarily a discrepancy between climate model estimates and energy balance estimates; you just need to do a like-for-like comparison.
So, this suggests that there might not really be a discrepency between observationally-based and model-based estimates for climate sensitivity, but also suggests we might have to be careful as to what we mean when we discuss global temperatures. As Ed’s post concludes
Finally, if the reported air-ocean warming and masking differences are robust, then which global mean temperature is relevant for informing policy? As observed? Or what those observations imply for ‘true’ global near-surface air temperature change? If it is decided that climate targets refer to the latter, then the warming is actually 24% (9-40%) larger than reported by HadCRUT4.