John Christy's Misleading Graphs from 2016

It's been over a decade, but occasionally I still see people sharing the graphs John Christy's presented as testimony in 2016 as "evidence" that models can't accurately predict observations. The graphs have been heavily criticized at RealClimate (see here and here) for very good reasons. 

Sometimes these graphs get dressed up with a different aesthetic while keeping the comparison exactly the same. The version below is from a Heritage Foundation "report" published in April of 2026, but the authors didn't even bother to update observations to 2025. They just ignore the last 10 years of observations, possibly because none of the authors had the competency to update the observational data to today.

Since Benestad and Schmidt have done such a thorough job debunking these types of graphs, I don't see any need to duplicate their efforts, but here's a brief summary of the what's misleading about these graphs:

1. Christy chose comparisons of atmospheric temperatures up to 50K ft, even though the primary point of comparison between models and observations should be at the surface where people live. Christy didn't show surface comparisons because then he couldn't exaggerate the mismatch by as much.
2. Christy chose to plot the model mean and observational datasets so that the linear trends for each would converge in 1979. This has the visual impact of making it appear that they deviate from each other more quickly than is warranted. Christy should have chosen to plot both models and observations to a common baseline, such as 1979 to 1983.
3. Christy shows either a spaghetti plot of 102 model runs or a single line for the model mean (or single trendline). It would have been better to plot a 95% confidence envelope with the model mean.
4. There is far more uncertainty in satellite observations than in thermometer datasets. RSS, UAH and NOAA-STAR show significantly differing trends. Christy collapses these into a single mean, but he should have included the margin of error for observations.
   
5. Satellites do not measure mid-tropospheric temperatures. They measure the radiances at wavelength bands emitted by oxygen molecules at various layers of the atmosphere, from which temperatures are inferred using a model. To compare models to observational data, Christy should have chosen surface thermometers, rather than satellites that interpret observational data using a model. 

At the time of Christy's testimony, Schmidt showed the observational data at the time would fall mostly within the 95% confidence envelope of models, even though the trends for each of the observational datasets were on the low end of CMIP5 models. And it's true that there was a model mismatch between TMT observations and models at the time (that Christy exaggerated), and this had as much to do with uncertainties in satellite datasets as anything else (this is discussed here and here).

However, it's been over 10 years since Christy's testimony, and so those continuing to share these graphs are adding two more problems to the above:

1. The CMIP5 models are no longer current. AR6 has been published with the CMIP6 ensemble.
2. The observational data they are sharing are now over 10 years out of date, and both RSS and UAH have updated what is the current version of each dataset.

Recently, Zeke Hausfather has developed a Climate Dashboard (it looks like there's a similar project in development by Berkeley Earth) that allows anyone to see near real time comparisons between observations and models (including CMIP3, CMIP5, and CMIP6). The comparisons on this site are current, properly plotted, and show the most meaningful comparisons - how GMST changes in observations compare to the three most current CMIP ensembles, with a model mean and 95% confidence envelope. I'd encourage you to check out that site, but here is how models are performing at the time that I'm writing this post.

Observations are Running through the Middle of the CMIP6 Envelope

Observations are Running through the Middle of the CMIP3 Envelope

And here's how model trends compare to observations:

CMIP6 Model Trends vs Observations

It's clear that properly plotted comparisons of models and surface observations demonstrate that models are doing just fine at predicting changes in GMST. Of course, that does not mean that models are doing equally well at predicting local and regional effects of AGW. Anyone determined to find model mismatches will be able to find them - cherry picking are still possible. But models are doing a good job of predicting the main point of comparison between models and observational data, which I guess explains why contrarians never show that primary point of comparison. The they must either cherry pick or use outdated, misplotted/misleading comparisons.