Home/BlogAs a follow-on to my recent post regarding global surface air temperature trends (1979-2025) and how they compare to climate models, this is an update on a similar comparison for tropical tropospheric temperature trends, courtesy of tabulations made by John Christy. It also represents an update to my popular “epic fail” blog post from 2013.
As most of you know, climate models suggest that the strongest warming response the climate system has to increasing anthropogenic greenhouse gas (GHG) emissions (mainly CO2 from fossil fuel burning) is in the tropical upper troposphere. This produces the model-anticipated “tropical hotspot”.
While the deep oceans represent the largest reservoir of heat energy storage in the climate system during warming, that signal is exceedingly small (hundredths of a degree C per decade) and so its uncertainty is rather large from an observational standpoint. In contrast, the tropical upper troposphere has the largest temperature response in climate models (up to 0.5 deg. C per decade).
This shown in the following plot of the decadal temperature trends from 39 climate models (red bars) compared to observations gathered from radiosondes (weather balloons); satellites; and global data reanalyses (which use all kinds of available meteorological data).
The sonde trend bar in the above plot (green) is the average of 3 datasets (radiosonde coverage of the tropics is very sparse); the reanalysis trend (black) is from 2 datasets, and the satellite trend (blue) is the average of 3 datasets. Out of all types of observational data, only the satellites provide complete coverage of the tropics.
Amazingly, all 39 climate models exhibit larger warming trends than all three classes of observational data.
Time Series, 1979-2025
If we compare the average model warming to the observations in individual years, we get the following time series (note that complete reanalysis data for 2025 are not yet available); color coding remains the same as in the previous plot:
The unusually warm year of 2024 really stands out (likely due to a decrease in cloud cover letting in more sunlight), but in 2025 the satellites and radiosondes show a “return to trend”. Of course, what happens in the future is anyone’s guess.
“So What? No One Lives In the Tropical Troposphere”
What is going on that might explain these discrepancies, not only between the models and the observations, but even between the various models themselves? And why should we care, since no one lives up in the tropical troposphere, anyway?
Well, the same argument can be made about the deep oceans (no one lives there), yet they are pointed to by many climate researchers as the most important “barometer” of the positive global energy imbalance of the climate system caused by increasing GHGs (and maybe by natural processes… who knows?).
The excessive warming of the tropical troposphere is no doubt related to inadequacies in how the models handle convective overturning in the tropics, that is, organized thunderstorm activity that transports heat from the surface upward. That “deep moist convection” redistributes not only heat energy, but clouds and water vapor, both of which have profound impacts on tropical tropospheric temperature. While moistening of the lowest layer of the troposphere in response to warming no doubt contributes to positive water vapor feedback, precipitation microphysics governs how much water vapor resides in the rest of the troposphere, and as we demonstrated almost 30 years ago, that leads to large uncertainties in total water vapor feedback.
My personal opinion has always been that the lack of tropical warming is because positive water vapor feedback, the primary positive feedback that amplifies warming in climate models, is too strong. Climate models actually support this interpretation because it has long been known that those models with the strongest “hotspot” in the upper troposphere tend to have the largest positive water vapor feedback.
Will Climate Models Ever Be “Fixed”?
I find it ironic that climate models are claimed to be based upon fundamental “physical principles”. If that were true, then all models would have the same climate sensitivity to increasing GHGs.
But they don’t.
Climate models range over a factor of three in climate sensitivity, a disparity that has remained for over 30 years of the climate modeling enterprise. And the main reason for that disparity is inter-model differences in the moist convective processes (clouds and water vapor) which cause positive feedbacks in the models.
Maybe if the modelers figured out why their handling of moist convection is flawed, models would then produce warming more in line with observations, and more in line with each other.
Much of global warming alarmism arises from scientific publications biased toward (1) the models that produce the most warming, and (2) the excessive GHG increases (“SSP scenarios“) they assume for the most dire climate change projections. Those scenarios are now known to be excessive compared to observed rates of global GHG emissions (and to the reviewer of our DOE report who said this conclusion was in error because I didn’t account for land use changes, no, I removed land use changes from the SSP scenarios… it was an apples-to-apples comparison).
Finally, I don’t want to make it sound like I’m against climate modeling. I am definitely not. I just think the models, as a tool for energy policy guidance, have been misused.
Thank you, Roy. Your research and posts are always interesting and make important points. Now, I’ll sit back and watch the food fight begin (heads up in the cafeteria).
Dr Roy
Thanks for your work and expertise in this field. I have one question concerning the first post, which GHG scenario was used in the tabulation of model results?
Will
I believe John Christy used SSP245, although it doesn’t much matter because through 2025 there is little difference between scenarios.
This would make a good keynote talk at the upcoming ICCC (International Conference on Climate Change) in DC this April. I’ll be there.
(Preferably without a food fight.)
Santer et al. 2018 contradicts Spencer’s claims of wholesale model failure. Santer concludes that, with appropriate statistical testing and updated datasets, discrepancies between modeled and observed tropical tropospheric trends are not robustly significant in some layers and time periods.
https://link.springer.com/chapter/10.1007/978-3-319-65058-6_5
Ben twists himself up trying to explain away discrepancies, but he’s still wrong.
It’s impossible to evaluate the differences you claim without confidence intervals and explicit uncertainty quantification.
Your analysis commits a common error in model-observation comparisons, treating the multi-member ensemble mean as the expected outcome for a single realization of the climate system. Climate model ensembles are built to sample internal variability as well as the forced response, whereas you present observations as one single realization which can therefore legitimately fall toward one end of the ensemble distribution over a finite time period without implying systematic model bias.
This issue is particularly evident for CanESM5, which stands out in your opinion article as having one of the largest mean discrepancies. This particular ensemble comprises around 50 simulations, implying substantial spread due to internal variability.
The appropriate test of consistency is whether observed tropical tropospheric warming rates fall within the ensemble spread, not whether they match the ensemble mean. But I’m sure you already know that.
IMHO.
Well put.
Arkady,
Are you suggesting that a confidence interval should be calculated from an ensemble model?
A funny thing is here that among the authors of the article
Consistency of Modeled and Observed Temperature Trends in the Tropical Troposphere
you find these three:
– Leopold Haimberger (U Vienna, Austria)
– Carl Mears, Frank Wentz (RSS, Santa Rosa)
*
The first one is, if I well remember, a radiosonde specialist who worked in the early 2000s with Spencer/Christy on homogenisation of radiosonde data using UAH’s satellite data, and developed with collabotators the RICH and RAOBCORE software packages (designed to improve homogeneity in radiosonde data).
*
The latter two are the head of REMSS, a group showing satellite data differing a lot from UAH’s (RSS v4.0).
Roy,
Isn’t the tropical troposphere extremely sensitive to the ENSO state?
Thus, our particular ENSO history matters a great deal.
It would be worthwhile to compare only to models with a similar ENSO history to the one we had.
Hard to compare the two updates.
Looks voluntary.
Care to elaborate on what you mean?
You said the observations moved up one place since 2024.
What’s the move since 2013?
Christy’s comparisons were misleading because they used unrepresentative data (tropics only) cherry picked to make the models look bad.
The average model from the mid 1970s, when programmed with actual CO2 growth since then, predicted 0.2 degrees C warming per decade.
Surface warming in those years was 0.2° C per decade. Total warming predicted was accurate, even though CO2 was over emphasized & increased absorbed solar radiation was under emphasized.
UAH GAT increased at a slower rate than surface measurements.
But nobody knows if UAH is more accurate than surface measurements.
There is no perfect measurement of GAT that we can compare Surface measurements or UAH to. Knowing the accuracy of these measurements is impossible.
Its even more staggering to realize that planetary motions have an impact here as well. They account for some of the variations seen the instrument and proxy records, along with variations in solar brightness.
Berkeley Earth just released their annual report on global temperatures for 2025.
It sure did:
https://berkeleyearth.org/global-temperature-report-for-2025/
NOAA as well.
https://www.ncei.noaa.gov/news/global-climate-202513
“Annual Highlights:
NOAA ranks 2025 as the third-warmest year in its global temperature record, which dates back to 1850.
Upper ocean heat content was record high in 2025.
Annual sea ice extent for both the Arctic and Antarctic regions ranked among the three lowest years on record.
The Northern Hemisphere snow cover extent was the third lowest on record.
There were 101 named tropical storms across the globe in 2025, which was above average.”
Meanwhile, so called skeptics will dismiss this by insisting that the long term trend has been global cooling for 3000 years.
They really said nothing at all,
“They really said nothing at all,”
Could you clarify what you are referring to and who this is directed toward?
Don’t mind Ian. He’s a nice chap, but also an ankle biter.
Let’s frame it this way:
Globally, there is not one single day in 2025 that was cooler than its 1991-2020 average.
(Via Andrew B. Watkins.)
How can NOAA claim it has records going back to 1850.,when it did not exist before 1970? They may have sourced data, but it is theirs.
How can you be post here while being so ignorant and unwilling to learn?
All institutions worldwide collect as much hstorical data as possible, and so doesn NOAA.
They started around 1990 GHCN-M (also named GHCN V1), the first project acquiring and maintaining as much monthly station data as possible.
In 1997, there were about 6000 stations; GHCN V2 and V3 then had 7280 stations worldwide.
Currently the V4 version is active, with over 27000 stations.
From where, do you think, does the data recorded by the numerous stations of GHCN daily?
For example:
ASN00001006 -15.5100 128.1503 3.8 WYNDHAM AERO
ASN00001006 1750 2023 274 WYNDHAM AERO
located in Australia?
What about starting here, instead of believing Robertson’s lies?
https://www.ncei.noaa.gov/products/land-based-station/global-historical-climatology-network-monthly
Ian,
The UK met office has 80% of their stations which are classified as junk that is a few degree centigrade out. Not only that, but some sites don’t physically exist or located in the sea at low tide.
If the UK is supposed to be a mark of quality then it’s anyone guess at how poor NOAA data is.
Anon for a reason
A paper titled ‘State of the UK Climate in 2024’, published in July 2025, reveals that sea surface temperatures near the UK have risen by nearly 1°C compared to the 1961-1990 baseline.
This can’t be urban heat island effect or bad station placement. So what’s the explanation?
https://rmets.onlinelibrary.wiley.com/doi/10.1002/joc.70010
Eldrosion,
I will have a detailed look at the report later. Initial scam did make me laugh.
Climate change cause spring to be a couple of days early. Taken by itself, then yes that is one possible causes. But their next remark that autumn was also a couple of days early suggests something else. Minor orbital changes. But characters like Entropic think you can work out orbital changes on a piece of paper.
But either way thanks for highlighting the paper.
You’re welcome.
I appreciate your commitment to learning more about this topic.
I am new to this and am confused about a fundamental point. When comparing a climate model prediction for 2025 to the observational data in 2025, how recent was the observational data used to construct the climate model? In other words, was the climate model constructed based on data until 2024 or based on data until some earlier year? If the latter, what earlier year? Also if the climate model projecting temperature to 2025 was based on observational temperature data until say 1990, did it then project CO2 levels from 1990 to 2025 or use observed C02 levels from 1990 to 2025 in the projection?
The CMIP6 models used observational data through 2014. From 2015 onward they ran a set of emissions scenarios that postulate emission levels out to the year 2100.
There’s an explanation and overview article at this link:
https://www.carbonbrief.org/cmip6-the-next-generation-of-climate-models-explained/
Mark,
These emission scenarios are based on economists, and unless they have a crystal ball then it’s worthless. Did they predict any pandemic, super scaling of ai data centers, work from home ?
In the UK we have zero trust in economists, doesn’t matter which political party you vote, the majority are very skeptical.
Stu at 2:57 PM
Adding to Mark B’s comment above, the CMIP6 models do not use observed temperatures to project future climate. Historical atmospheric concentrations of CO2, CH4, N2O, and halogenated gases are given as externally specified time series for the period 1850-2014; from instrumental measurements for recent decades, and ice-core reconstructions for earlier periods. Beginning in 2015, scenario-based concentrations are used.
So, when comparing model output to observations in 2025, any agreement or disagreement in 2025 reflects the model’s forced response to the scenario plus internal variability.
You can read more about it here: https://confluence.ecmwf.int/display/CKB/CMIP6%25253A%252BGlobal%252Bclimate%252Bprojections?utm_source
Ark, I think Mark’s point and your response shed light on a very important point, and flaw, in all these models. The models all have to START with temperature and other important physical conditions of the atmosphere, oceans, and cryosphere. In three dimensions. From 1850 to 1900. Really? And then project that out to 2100. Step, by step. Sounds like a Las Vegas magic trick. Entertaining, but not having much real world value. All that computing power would be better spent on AI or bitcoin mining.
After the insults die down, can you or someone else tell me why I am wrong?
Two thoughts:
1) One could argue that incomplete and imperfect observational data to set model initial conditions is a flaw in the data and a challenge for modeling, rather than specifically a flaw in the model. The way it is dealt with, as I understand it, is to run the model for an extended period under the conditions of the CMIP6 “piControl” experiment which holds relevant conditions constant at levels consistent with 1850 with a recommended run time of 500 years. From this one can evaluate consistency of the metrics produced with whatever observations are available.
Another experiment “AMIP” runs from 1850 to near present using historical forcing data to produce metrics whose consistency can be evaluated against observations such as they exist.
2) You’ve identified a real issue, one of many really, which is a fine trait for scientific inquiry. You’ve then used it to discard climate modeling as a useful tool of scientific inquiry without any apparent understanding or evaluation of how the issue is addressed or how plausible it might be that it has been reasonably mitigated.
It gives the impression of looking for a rationalization to discount climate modeling without applying a commensurate level of skepticism to the why your conclusion might be poorly reasoned. That is, “climate modeling useless because 1850 observations are sparse”, isn’t self evident.
The “after the insults die down” comment is bad form if you’re really interested in good faith discussion of the science.
1/16: What’s causing Colorado’s crazy warm and dry winter? And what is climate change’s role?
https://www.tiktok.com/@weatherchris/video/7595994271501602103?_r=1&_t=ZP-937yEr2ZPaQ
Spencer’s characterization that “While the deep oceans represent the largest reservoir of heat energy storage in the climate system during warming, that signal is exceedingly small (hundredths of a degree C per decade) and so its uncertainty is rather large from an observational standpoint” is contradicted by the latest peer-reviewed observational analysis of ocean heat content (OHC) presented in this paper: https://link.springer.com/content/pdf/10.1007/s00376-026-5876-0.pdf
The recent study finds that global upper 0-2000m OHC continued to increase in 2025, reaching record levels and showing a statistically robust warming signal well beyond trivial changes.
1/ From 2024 to 2025 the upper 2000m of the ocean gained ~23 ± 8 ZJ (zettajoules) of heat due to changes in Earth’s energy imbalance directly tied to GHG forcing.
2/ Long-term trends in OHC have accelerated from ~0.14 ± 0.03 W/m2 (1960-2025) to ~0.32 ± 0.14 W/m2 (2005-2025).
3/ The continued accumulation of ocean heat is consistent across independent datasets (IAP/CAS, Copernicus Marine, ocean reanalysis products), which substantially reduces the observational uncertainty.
So, ocean heat uptake is neither negligible nor poorly observed, and is consistent with observed Earth energy imbalance. And, because over 90% of excess heat from GHG forcing is a b s o r b e d by the oceans, OHC is a more reliable metric of AGW than surface temperature.
Dyakuyu za vashu uvahu.
1/ There is no valid science to verify “GHG forcing”.
2/ OHC is NOT measured in W/m².
3/ OHC is only evidence that Earth is in a warming trend, as also indicated by UAH Global. Earth is in a natural warming trend.
” OHC is NOT measured in W/m^2. ”
It is absolutely evident that the worldwide renowned lunar motion specialist knows that much better than all people who contributed to the paper Ivanovich linked to:
https://link.springer.com/content/pdf/10.1007/s00376-026-5876-0.pdf
*
As I have explained so often: the ball-on-a-string syndrome is mostly present in people suffering also from the Trump addiction syndrome.
That’s correct Bindi. OHC is NOT measured in “W/m²”. Thanks for quoting me correctly.
That link does get it correct in Fig 1, where the graphs are labeled using Zetta Joules (ZJ). So you’ve proven me right once again.
And you bring up Moon without having a viable model of “orbiting without spin”. Which proves me right yet again, as I repeatedly point out that you’ve got NOTHING.
I never get tired of being right.
Clint R is ignorant to such an extent that he doesn’t even grasp the trivial and fundamental difference between ‘warm’ – expressed in Joule – and ‘warming’ – expressed in Watt/m^2/decade, what is plain correct.
What else could we expect from a 360 degree denier who is not even abkle to accept science backed by Sir Isaac Newton?
*
However, one has to blame here Ivanovich on his wrong pasting of the article’s original, as he forgot to add ‘(10 yr)⁻1 in his post.
Bindi must be working 24/7 trying to find ways to attack me. He’s becoming as obsessed with me as are Willard, Nate, Norman, and Ball4. But, they always end up just proving me right.
In his latest effort, Bindi shows his ignorance of thermodynamics.
* “Warm” is NOT “expressed in Joule”! “Warm” refers to temperature, which is expressed in one of the temperature scales such as Fahrenheit, Celsius, Absolute, etc.
* “Warming” is NOT expressed in “Watt/m^2/decade”! “warming” refers to “heat”, which is the transfer of thermal energy from hot to cold. “Heat” has units of “Watts”, or “Joules/sec”.
* “W/m²” is NOT energy, it is “flux”.
Can hardly wait for Bindi to prove me right again. We know it’s coming….
Hey Puffman, riddle me this –
Alabama is finding new ways to criminally charge undocumented immigrants. Federal courts in the state are using a law last applied during the U.S. internment of people of Japanese descent during World War II to charge immigrants who don’t register themselves.
https://www.al.com/news/2026/01/alabama-uses-japanese-internment-era-law-to-charge-immigrants-who-dont-self-register.html
Does that give you warm feelings expressed in whatever units you please?
“The data doesn’t matter. We’re not basing our recommendations on the data. We’re basing them on the climate models.”
–Prof. Chris Folland, Hadley Centre for Climate Prediction and Research