Responding to the CO2 Coalition's "Fact #21" on Climate Models

CO2 Coalition's "Fact #21" claims that "IPCC models have overstated warming by up to three time too much." According to this claim to fact, John Christy's testimony on "February 2016 to the U.S. House Committee on Science, Space & Technology included remarkable charts that document just how much the models overestimate temperatures. The red line in the chart shows the average of 102 climate model runs completed by Christy and his team at the University of Alabama at Huntsville using the models on which the IPCC itself relies. Also shown on the chart are the actual, observed temperatures. The models exaggerate warming, on average, two and a half times the actual temperature (or three times over in the climate-crucial tropics). Here's the graph they use to support this claim.

The above graph reports to show 32 models and 102 model runs within the CMIP5 model ensemble. They are limited to those runs in the KNMI Climate Explorer. The models are specified but the model runs are not; we don't know, for instance, what RCP scenarios were used or whether Christy has loaded the dice by using runs from high RCP scenarios. And of course, these are not "model runs completed by Christy and his team" at UAH. That's a lie. The plots of what CO2 Coalition calls "actual temperatures" are claimed to be averages of 4 radiosonde datasets, 3 satellite datasets and 3 reanalyses. CO2 Coalition does not tell us what any of these datasets are. The problems with this graph are numerous, and this graph (and those like it) have been severely criticized over the years.

1. Since we're not told the observational datasets, it's hard to reproduce this graph.
2. Christy's "baseline" was a single year (1979) instead of a 5-year or 10-year mean; this was chosen to exaggerate discrepancies between their chosen models and chosen datasets.
3. Christy claims to be using 5-year averages, but in fact the end points use 4-year and 3-year averages.
4. Christy offered a spaghetti plot of model runs, but he did not include the model spread. The 95% confidence envelope for the models is completely missing, but they are essential for an accurate assessment of whether observations agree with models.
5. Christy cherry picked tropical mid-tropospheric temperatures (TMT), when the primary point of comparison between models and observations should be between models and surface temperatures. Changes in GMST anomalies are the most significant and useful comparisons with model predictions.
6. Christy has deliberately chosen an area where observational data is weak (satellite observations in the upper atmosphere can be contaminated by the stratosphere, which is cooling). The fact that models don't predict tropical TMTs well is hardly an argument that they don't predict GMST changes well, but Christy (and the CO2 Coalition) didn't show that comparison.
7. To assess whether warming is consistent with IPCC's expectations using model predictions, you would need to filter the model runs by their calculated TCR or ECS values. That is, if models that produce TCR or ECS values near the IPCC's estimate produce good predictions for GMST changes, then that's a good indication that observations are conforming to the IPCC's expectations.

Correcting for the graphing errors on Christ's part, Gavin Schmidt produced the below graph comparing tropical TMTs to the CMIP5 model mean and spread with a 10-year baseline. While observations appear to be following a trend with less warming than the models, observations are still mostly within the model spread. There is still a model mismatch between models and observations on tropical TMTs, and this has been discussed in the literature. Christy (and the CO2 Coalition) made (and promoted) graphs that exaggerate this mismatch and then use it to say models aren't performing well. But as we'll see below, models estimating ECS near 3 C are actually doing a remarkably good job of predicting GMST changes, even if issues involving models and satellite accuracy mean that there's a continued mismatch with tropical TMTs.

It should also be noted that CO2 Coalition cites a paper by Christy and McNider,[1] which unsurprisingly does not include any version of Christy's graph above, since it's riddled with errors. Instead it shows CMIP5 simulated global lower troposphere temperatures (TLT) from 1975 to 2025. In this graph, global TLTs in those models were expected to increase by ~1.19 C above the 1979-1983 baseline with a 95% range of about 1.04 C to 1.24 C.  The range for 2024 appears to be exactly 1.0 C to 1.2 C. Because the graph below had to be published in a peer-reviewed journal, Christy couldn't get away with the errors and tricks he used in the above graph during his 2017 testimony. This plot didn't cherry pick tropical TMT, and it used a five-year baseline. It also used annual means instead of 5-year means with inconsistent end points.  However, his graph still cherry picks what amounts to an outlier dataset (UAHv5.6) and ignores other satellite data estimating TLT temperatures.

Here's the same graph with RSSv4.0 through Feb 2024 below. With more recent RSS data, global TLTs are closer to the models than the CO2 Coalition would have you to believe. The data shows warming spikes corresponding to El Nino years reaching mean values expected from these CMIP5 model runs with La Nina years typically below the model mean. In other words, RSS shows more temperature variability due to ENSO than CMIP5, but overall warming is not off by as much as Christy shows. I also question the narrow range shown by the 95% CI in Christy's graph. I doubt very seriously that 95% of the model runs fit within that window. 

Since satellite observations have known issues and inaccuracies that have not yet been resolved, there's only so much we can make of this, but it's still an important point of continued research. However, models that calculate TCR near 2 C (ECS near 3 C) have done a remarkably good job of predicting the increase in GMST temperatures. And that's actually the most significant central point here. We'd expect models with high ECS values to predict too much warming. That doesn't mean that observations don't match ECS values near 3 C. CMIP5 models actually calculate a large range for ECS, with some models calculating ECS to be higher than what the IPCC considers most plausible. The central estimate for ECS is ~3 C, and models near that value for ECS are performing quite well, meaning that models generally agree with ECS being near 3 C. Below CarbonBrief plotted the correspondence between GMST temperatures from 1950-2023 and CMIP6 Models filtered for TCR. That is, models with a TCR near the IPCC's central estimate. I don't think you can ask for a better agreement between the two.

CO2 Coalition will not show you the above graph. As I point out elsewhere, satellites simply are not that accurate compared to surface thermometers. In fact uncertainties are ~5x greater with satellites than with surface thermometers. However the model mismatch between tropical TMT observations and models gets resolved, the fact remains that models consistent with central estimates for TCR and ECS are doing a remarkably good job predicting the most accurate datasets that are most relevant to us - global temperatures at the surface where we live.

References:

[1] Christy, J. R., & McNider, R. T. (2017). Satellite bulk tropospheric temperatures as a metric for climate sensitivity. Asia-Pacific Journal of Atmospheric Sciences, 53(4), 511–518. doi:10.1007/s13143-017-0070-z