As is now traditional, every year around this time we update the model-observation comparison page with an additional annual observational point, and upgrade any observational products to their latest versions.
A couple of notable issues this year. HadCRUT has now been updated to version 5 which includes polar infilling, making the Cowtan and Way dataset (which was designed to address that issue in HadCRUT4) a little superfluous. Going forward it is unlikely to be maintained so, in a couple of figures, I have replaced it with the new HadCRUT5. The GISTEMP version is now v4.
For the comparison with the Hansen et al. (1988), we only had the projected output up to 2019 (taken from fig 3a in the original paper). However, it turns out that fuller results were archived at NCAR, and now they have been added to our data file (and yes, I realise this is ironic). This extends Scenario B to 2030 and Scenario A to 2060.
Nothing substantive has changed with respect to the satellite data products, so the only change is the addition of 2020 in the figures and trends.
So what do we see? The early Hansen models have done very well considering the uncertainty in total forcings (as we’ve discussed (Hausfather et al., 2019)). The CMIP3 models estimates of SAT forecast from ~2000 continue to be astoundingly on point. This must be due (in part) to luck since the spread in forcings and sensitivity in the GCMs is somewhat ad hoc (given that the CMIP simulations are ensembles of opportunity), but is nonetheless impressive.
The forcings spread in CMIP5 was more constrained, but had some small systematic biases as we’ve discussed Schmidt et al., 2014. The systematic issue associated with the forcings and more general issue of the target diagnostic (whether we use SAT or a blended SST/SAT product from the models), give rise to small effects (roughly 0.1ºC and 0.05ºC respectively) but are independent and additive.
The discrepancies between the CMIP5 ensemble and the lower atmospheric MSU/AMSU products are still noticeable, but remember that we still do not have a ‘forcings-adjusted’ estimate of the CMIP5 simulations for TMT, though work with the CMIP6 models and forcings to address this is ongoing. Nonetheless, the observed TMT trends are very much on the low side of what the models projected, even while stratospheric and surface trends are much closer to the ensemble mean. There is still more to be done here. Stay tuned!
The results from CMIP6 (which are still being rolled out) are too recent to be usefully added to this assessment of forecasts right now, though some compilations have now appeared:
The issues in CMIP6 related to the excessive spread in climate sensitivity will need to be looked at in more detail moving forward. In my opinion ‘official’ projections will need to weight the models to screen out those ECS values outside of the constrained range. We’ll see if other’s agree when the IPCC report is released later this year.
Please let us know in the comments if you have suggestions for improvements to these figures/analyses, or suggestions for additions.
Z. Hausfather, H.F. Drake, T. Abbott, and G.A. Schmidt, “Evaluating the Performance of Past Climate Model Projections”, Geophysical Research Letters, vol. 47, 2020. http://dx.doi.org/10.1029/2019GL085378
G.A. Schmidt, D.T. Shindell, and K. Tsigaridis, “Reconciling warming trends”, Nature Geoscience, vol. 7, pp. 158-160, 2014. http://dx.doi.org/10.1038/ngeo2105