RealClimate: Comparison Update 2024

One more dot on the graphs for our annual model-observations comparisons updates. Given how extraordinary the last two years have been, there are a few highlights to note.

First, we have updated the versions of a few of the observational datasets: UAH TLT/TMT are now on version 6.1, and the NOAA NCEI surface temperature data are now version 6. We use the same collations of Hansen81/Hansen88/CMIP3/CMIP5/CMIP6 model output as previously. The comparisons cover surface air temperatures, sea surface temperatures, tropospheric atmospheric temperatures (TLT, TMT), stratospheric temperatures, and a few variations on these themes that have been of interest in the past. (It would be nice to have some non-temperature variables in the mix – feel free to suggest some if you can point to (or post) an archive of the models historical+projected results).

With respect to the GMSAT, it’s striking how close the real world is to the Hansen et al. (1988) ‘Scenario B’ (this scenario had ‘business as usual’ concentration rises in CO2, but too much growth in CFCs and CH4. However, the prize for most skillful projection still goes to the CMIP3 ensemble; even after 20 years, it’s still pretty much spot on.

The detailed issues that lead to some angst around the CMIP5 models – mis-specifications of the forcings, the importance of the SST/SAT blend vs. SAT trends have somewhat faded in importance. These are/were real issues, but they are small compared to the ongoing trends. With respect to CMIP6, the observations (across a swath of temperature related diagnostics) are still best matched by the sub-sample of screened models (i.e. discarding those that ‘ran hot’).

The updates with respect to the atmospheric temperature profiles (MSU/AMSU derived diagnostics), have become slightly more favorable to the models, though the structural variation between the RSS data and the UAH/NOAA STAR retrievals is still clear. For the sea surface temperature, the real world seems to warming at the upper end of expectations, but still (just!) within the screened spread.

One of the main reasons to maintain these comparisons is to see where discrepancies arise. To that end, multiple versions of the observational data are obviously useful since they can give an estimate of the structural uncertainty (this has been very important in the MSU/AMSU comparisons for instance). In other instances, we have less concerns about the observational data, but we are concerned that the models are not being given the right inputs. For example, since the internal variability in stratospheric temperatures is much less than in the lower atmosphere, incorrect forced signals can emerge faster. I think we may be seeing some of that in the SSU comparisons…

The match to the models is very good over the historical period (to 2014), but post 2015, there is some mismatch between the model variance and the obs. There are two potential issues – the timing of the solar cycle 25 (a solar max warms the stratosphere) – which happened earlier and bigger than expected by CMIP6, and the presence of the Hunga Tonga volcano (from 2021) which is having complex impacts on the stratospheric temperatures. Nonetheless, the long term trends are still well-modeled.

As always, if someone knows of expanded model diagnostics and relevant observational data sets to compare with, let me know and we can add it to the page.

Thanks as always to the data centers who provide the observational data, the CMIP committees who organised this storage of the outputs, the modeling centers that did the runs, and the authors who produced the derived data sets we are using directly here (full refs on the above listed page).

See you next year!