1. We know there is considerable short term weather and even climate noise in GMST and this can be removed by at least minimal filtering.
2. But we also know a far better metric is ocean heat content (OHC), and comparing models and obs, at least after 1958, would be more revealing. Values are not viable before IGY (1958) and error bars are certainly greater in earlier years, but also not muich was going on then.
Thanks Kevin! I actually have a post coming up in the next few weeks on the question of filtering, but the results are still somewhat dependent on assumptions near the end-points of the dataset (and its hard to rule out 0.4C per decade over the last 15 years).
OHC is definitely a key metric, and I'd like to do a comparison with climate models there. Its just a bit more challenging to get OHC fields for every model.
I've always wondered about tracking GHG effects by surface measurements, when we know that there is a large signal from heat moving back and forth (ENSO et al) from the ocean to perturb it.
Curious what you think for the cloud/aerosol question. If the choices are internal variability, aerosols, or warmer air causing reductions in cloud albedo (as posited in the write-up—maybe I don’t fully understand things), it’s kind of cold comfort to just hope that its internal variability?
Shouldn’t there be an option for the climatological impacts due to increased influence of geomagnetic storms and the associated solar radiation reaching deeper into our atmosphere?
For the record and contrary to several threads and comments, there’s no evidence that changes in Earth’s magnetic field have any measurable effect on climate or ECS.
1. Magnetic field strength has varied enormously over Earth’s history — including full reversals — without producing climate shifts.
2. The magnetic field does not regulate solar energy input, and ECS depends on radiative forcing, which magnetism cannot influence.
3. Cosmic-ray changes from magnetic variations are tiny compared to greenhouse forcing and too small to affect climate.
4. The observed pattern of warming matches greenhouse physics — not magnetic trends.
5. If the magnetic field had a large impact on ECS, past reversals would show huge climate anomalies, but they don’t.
You summarized the model interpretations very eloquently, sensibly, openly, and impartially. Am a fan of yours now. I just pray we don't find any wrenches in the future that cause quicker heating where we have to change the models to show either a higher ECS or unforseen variables discoverable only at time Earth had heated to a certain extent. Models seem to be doing great but as we heat up we might find new menaces.
Whether directly or indirectly all is connected on Earth and even minute changes are likely in the ECS if the magnetic field changes. Radiation fluxes and other geo changes will factor in even if the ghgs stay constant. My tuppence.
💯 and we have a growing body of evidence that the earths magnetic field is weakening and the Earth is becoming much more sensitive to electromagnetic radiation. Unfortunately, making this connection is still considered blasphemous to many.
Phil, I agree 100%. One problem is many of the current plethora of climate scientists (not all ) are so narrow minded I think they are just alchemists with a degree and spew out hot air just to get their names on papers and get goose bumps when addressed as Dr so and so. Eg, marine cloud brightening vs sai. A recent paper showed ship clouds only caused 5% of the negative radiative forcing.. the 95% was the raw sulfate tropospherical distribution from the ship emissions. Now, they say we need marine clouds where the salt has to be 2 um or whatever to be effective cloud nucleators. We don't need marine clouds. NaCl is up to .97 albedo. Just get the ships to spray ocean water ingo the sky.We might get a few clouds too. Now. Like renewables are weknewables many scientists love wearing Bozo the clown outfits in private.As Earth burns.
And it might be interesting to align 1900-2030 CMIP3 modeling with analyses that better address the 1940s-era heterogeneity in sea surface temperature. So basically a modification of what Dr. John Kennedy did, with the following analyses: DCENT-I, DCENT-MLE, HadOST, COBE-STEMP3, ERA5, and PAGES2K.
The whole cherry tree approach really highlights something interesting about the CMIP6 hot models debate. They might have been overshooting long-term trends, but when you look at just the past 15 years, observations are sitting right in the upper half of that range. Makes you wonder if dismissing those models was premature, or if we're just in a hot phase that'll average out.
I'm only seeing tangential discussion of positive feedback effects.
The repeated "discovery" of more of these feedbacks over the past 30 years, leads me to predict that the models will continue to lag the accelerating warming. It is a chaotic system, capable of terrifyingly rapid mode changes.
These models ignore much (all?) of that and their use seens to be more about feelings of safety and control than full modelling, or preparing governments for reality.
Has anyone published data from runs of a model that includes day 5 positive feedback mechanisms? There are many more but let's start at 5.
One of the great things climate scientists can take credit for is predicting warming before it occured. Was there a prediction of warming acceleration?
In any case what is the theory as to why an acceleration? One line I’ve seen is that sinks aren’t doing as well as they have in the past.
I am not a scientist, just a very humble analyst and I haven’t seen acceleration theories And there may be no acceleration but if there is it’s probably important to focus on it.
I'm interested in overshoot and carbon capture to pull the surface temperature back down. I asked GPT5 to look at the various carbon capture ideas, and estimate how much CO2 we could reasonably expect to pull from the air. It replied that if everyone takes it seriously, and invest several trillion dollars, we could pull 4-5 brillion tons of CO2 from the air per year.
So I asked, if global temperature increases by half a degree C in the next two decades, how much will that heat catalyst increase the emissions of natural feedback loops (microbes decomposing organic material in the soil, in wetlands, landfills, and permafrost). It replied, 4-5 billion tons per year.
It is currently impossible to determine the heat energy content of Earth’s climate system with sufficient precision and accuracy to allow for meaningful prediction. This is due to the lack of an adequate number of thermometric instruments for the entire volume of the World Ocean, and the lack of precise and accurate thermometrics for the atmosphere prior to the advent of satellite measurement.
Additionally, Earth’s climate system is a chaotic system. The Butterfly Effect got its name, ironically, from that fact.
These facts are heretical in the context of the $$$$$$$ “climate change” industry… but they are the truth.
Oak Ridge National Lab hired me some years ago to do modeling of atmospheric/CO2 variables and their relationships. My views are well-informed.
I agree. I have seen other climatologists demonstrate that there is an energy imbalance with many of the current models that are effecting the predictive qualities of these models.
It seems that it would be an oversight to assume that the quantity of solar radiation reaching and interacting with our atmosphere is static and not something that needs to be modeled, even separately from the climate models.
Yes, solar radiation does vary, but looking at the 11 year cycle doesn’t capture it and the bigger impact from our perspective is due to variation in Earth’s magnetic field.
If you're claiming variation in our magnetic field is contributing to the current rapid climate change, then yes. That's not a consensus claim, so the burden is on you to support it.
Zeke, a question about how this comparison relates to uncertainty in future projections:
To what extent does the observed agreement (or disagreement) between models and historical temperatures affect our confidence in the uncertainty ranges for projections out to 2100? In other words, how should we think about the way historical model error propagates into the “error bars” around future warming under a given emissions scenario?
Good post. I'd be interested in your views as to what the distribution of model results and various summary statistics actually means, given that this is an ensemble of opportunity, and not a random selection of balls from an urn.
See:
"Multi-model datasets are often described as ‘ensembles of opportunity’. This refers to how they are assembled, namely by asking for model results from anyone who is willing to contribute. Various non-scientific aspects determine the size and composition of such ensembles, e.g. whether modelling groups are interested at all and whether they have funding and computational resources to do the requested simulations.
The implication of how these multi-model ensembles are created is that the sampling is neither systematic nor random. The distribution of the models is thus completely arbitrary, and might be different in a subsequent ensemble, therefore changing the result even if the knowledge about the climate system has not changed. Even if we were to solve the problem of weighting the individual members, the posterior would still depend on the prior distribution which is determined by human decisions and cannot be interpreted in a scientific sense. . .
Therefore, it is probable that the range covered by the multi-model ensemble covers a minimum rather than the full range of uncertainty."
Tebaldi, C., & Knutti, R. (2007). The use of the multi-model ensemble in probabilistic climate projections. Philosophical transactions of the royal society A: mathematical, physical and engineering sciences, 365(1857), 2053-2075.
The fact that CMIP-style models are an ensemble of opportunity is not ideal; one of our recommendations for the CMIP7 cycle is to focus more on perturbed physics ensembles, where variations of a single model can often more thoroughly explore the solution space. For example, a single perturbed physics experiment with HadCM3 gives a range of carbon cycle feedback responses twice as large as the C4MIP ensemble even when constrained to reproduce historical CO2 concentrations: https://www.carbonbrief.org/analysis-how-carbon-cycle-feedbacks-could-make-global-warming-worse/
Emulators are another way to potentially more thoroughly explore the solution space for aggregate variables like GMST, as they can be calibrated based on a full PDF of plausible values (e.g. for climate sensitivity or carbon cycle feedbacks) rather than relying on the distribution of ESM values for those variables: https://gmd.copernicus.org/articles/17/8569/2024/
Nayak, M. S., Armour, K. C., & Battisti, D. S. (2025). Evaluating physical climate model emulators for global warming projections. Geophysical Research Letters, 52(23), e2025GL118591.
It would be much more rigorous to compare projections made at time X with the future evolution of projected variables at time X+Y, and then deconvolving reasons for accuracies or inaccuracies, skill or lack of skill, etc..
Showing a multi-model mean and big envelope of model results based on prescribed forcings does not tell us much about the fidelity of any particular model or allow researchers to identify strengths or weaknesses of individual models.
Schmidt, G. A., et al. "Practice and philosophy of climate model tuning across six US modeling centers." Geoscientific Model Development 10.9 (2017): 3207-3223.
More generally, GAST is not relevant to mitigation or adaptation in the real world, so it is not clear what a retrospective analysis of the model ensemble with respect to GAST actually signifies, other than the fact that modelers can create an envelope of results that contains observations. One good way to do this of course is to have a big envelope.
Obviously, models do pretty poorly at the regional level for variables like precipitation.
Vautard, Robert, et al. "Evaluation of the large EURO‐CORDEX regional climate model ensemble." Journal of Geophysical Research: Atmospheres 126.17 (2021): e2019JD032344.
At the end of the day, plausibility for climate sensitivity is not determined using these models but rather by a range of expert opinion based on various lines of evidence. The question is then what are these models for? There doesn't seem much use in repeatedly showing model envelopes on blogs as if it means something.
Thanks Zeke. Two comments.
1. We know there is considerable short term weather and even climate noise in GMST and this can be removed by at least minimal filtering.
2. But we also know a far better metric is ocean heat content (OHC), and comparing models and obs, at least after 1958, would be more revealing. Values are not viable before IGY (1958) and error bars are certainly greater in earlier years, but also not muich was going on then.
Thanks Kevin! I actually have a post coming up in the next few weeks on the question of filtering, but the results are still somewhat dependent on assumptions near the end-points of the dataset (and its hard to rule out 0.4C per decade over the last 15 years).
OHC is definitely a key metric, and I'd like to do a comparison with climate models there. Its just a bit more challenging to get OHC fields for every model.
Thank you!
I've always wondered about tracking GHG effects by surface measurements, when we know that there is a large signal from heat moving back and forth (ENSO et al) from the ocean to perturb it.
Agreed
Great stuff as usual. Will be interesting (and scary) to watch the next few years.
I am particularly curious to see if we can get to the bottom of the reduced cloud albedo aspect of things? See this recent write-up: https://www.earth.com/news/rapid-surge-in-global-warming-may-have-an-unexpected-source-low-cloud-albedo/
Curious what you think for the cloud/aerosol question. If the choices are internal variability, aerosols, or warmer air causing reductions in cloud albedo (as posited in the write-up—maybe I don’t fully understand things), it’s kind of cold comfort to just hope that its internal variability?
Shouldn’t there be an option for the climatological impacts due to increased influence of geomagnetic storms and the associated solar radiation reaching deeper into our atmosphere?
For the record and contrary to several threads and comments, there’s no evidence that changes in Earth’s magnetic field have any measurable effect on climate or ECS.
1. Magnetic field strength has varied enormously over Earth’s history — including full reversals — without producing climate shifts.
2. The magnetic field does not regulate solar energy input, and ECS depends on radiative forcing, which magnetism cannot influence.
3. Cosmic-ray changes from magnetic variations are tiny compared to greenhouse forcing and too small to affect climate.
4. The observed pattern of warming matches greenhouse physics — not magnetic trends.
5. If the magnetic field had a large impact on ECS, past reversals would show huge climate anomalies, but they don’t.
Dr. Hausfather,
You summarized the model interpretations very eloquently, sensibly, openly, and impartially. Am a fan of yours now. I just pray we don't find any wrenches in the future that cause quicker heating where we have to change the models to show either a higher ECS or unforseen variables discoverable only at time Earth had heated to a certain extent. Models seem to be doing great but as we heat up we might find new menaces.
Do you think that changes to earths magnetic field would impact ECS?
Whether directly or indirectly all is connected on Earth and even minute changes are likely in the ECS if the magnetic field changes. Radiation fluxes and other geo changes will factor in even if the ghgs stay constant. My tuppence.
💯 and we have a growing body of evidence that the earths magnetic field is weakening and the Earth is becoming much more sensitive to electromagnetic radiation. Unfortunately, making this connection is still considered blasphemous to many.
Phil, I agree 100%. One problem is many of the current plethora of climate scientists (not all ) are so narrow minded I think they are just alchemists with a degree and spew out hot air just to get their names on papers and get goose bumps when addressed as Dr so and so. Eg, marine cloud brightening vs sai. A recent paper showed ship clouds only caused 5% of the negative radiative forcing.. the 95% was the raw sulfate tropospherical distribution from the ship emissions. Now, they say we need marine clouds where the salt has to be 2 um or whatever to be effective cloud nucleators. We don't need marine clouds. NaCl is up to .97 albedo. Just get the ships to spray ocean water ingo the sky.We might get a few clouds too. Now. Like renewables are weknewables many scientists love wearing Bozo the clown outfits in private.As Earth burns.
I wouldn't say it's blasphemous, only that it isn't gospel yet. Can you cite some of that "growing" body of evidence?
How much of the work that ESA is doing on the subject are you aware of?
I thought observed forcing exceeds RCP8.5's post-2005 projected forcing:
1) RCP8.5 projection: https://pik-potsdam.de/~mmalte/rcps/
2) observed: https://github.com/ClimateIndicator/forcing-timeseries/blob/main/output/ERF_best_aggregates_1750-2024.csv [from: https://essd.copernicus.org/articles/17/2641/2025/ ]
Assuming RCP4.5's post-2005 forcing is less than RCP8.5, shouldn't RCP8.5 be used instead of RCP4.5 when comparing CMIP5 models to observed warming?
Similarly, should a scenario other than SSP2-4.5 be used when comparing CMIP6 models to observed warming?
And it might be interesting to align 1900-2030 CMIP3 modeling with analyses that better address the 1940s-era heterogeneity in sea surface temperature. So basically a modification of what Dr. John Kennedy did, with the following analyses: DCENT-I, DCENT-MLE, HadOST, COBE-STEMP3, ERA5, and PAGES2K.
- CMIP3 modeling: https://archive.is/GI3hf
- Dr. John Kennedy: https://diagrammonkey.wordpress.com/2024/11/26/people-are-afraid-to-merge/
- DCENT-I / DCENT-MLE: https://doi.org/10.31223/X5V16S , https://doi.org/10.26050/WDCC/DCENT_MLE_v1_0
- HadOST: https://doi.org/10.1175/JCLI-D-18-0555.1 , https://x.com/khaustein/status/1250013281676394500
- COBE-STEMP3: https://climate.mri-jma.go.jp/pub/archives/Ishii-et-al_COBE-SST3/gm/
- ERA5: https://psl.noaa.gov/data/atmoswrit/timeseries/
- PAGES2k: https://doi.org/10.5194/essd-2024-500
The whole cherry tree approach really highlights something interesting about the CMIP6 hot models debate. They might have been overshooting long-term trends, but when you look at just the past 15 years, observations are sitting right in the upper half of that range. Makes you wonder if dismissing those models was premature, or if we're just in a hot phase that'll average out.
What's the process for dealing with known anomalous aerosol impacts, like Pinatubo or the removal of sulfur from shipping fuel?
I'm only seeing tangential discussion of positive feedback effects.
The repeated "discovery" of more of these feedbacks over the past 30 years, leads me to predict that the models will continue to lag the accelerating warming. It is a chaotic system, capable of terrifyingly rapid mode changes.
These models ignore much (all?) of that and their use seens to be more about feelings of safety and control than full modelling, or preparing governments for reality.
Has anyone published data from runs of a model that includes day 5 positive feedback mechanisms? There are many more but let's start at 5.
One of the great things climate scientists can take credit for is predicting warming before it occured. Was there a prediction of warming acceleration?
In any case what is the theory as to why an acceleration? One line I’ve seen is that sinks aren’t doing as well as they have in the past.
I am not a scientist, just a very humble analyst and I haven’t seen acceleration theories And there may be no acceleration but if there is it’s probably important to focus on it.
I'm interested in overshoot and carbon capture to pull the surface temperature back down. I asked GPT5 to look at the various carbon capture ideas, and estimate how much CO2 we could reasonably expect to pull from the air. It replied that if everyone takes it seriously, and invest several trillion dollars, we could pull 4-5 brillion tons of CO2 from the air per year.
So I asked, if global temperature increases by half a degree C in the next two decades, how much will that heat catalyst increase the emissions of natural feedback loops (microbes decomposing organic material in the soil, in wetlands, landfills, and permafrost). It replied, 4-5 billion tons per year.
Do those figures match what you are expecting?
still?
lol
It is currently impossible to determine the heat energy content of Earth’s climate system with sufficient precision and accuracy to allow for meaningful prediction. This is due to the lack of an adequate number of thermometric instruments for the entire volume of the World Ocean, and the lack of precise and accurate thermometrics for the atmosphere prior to the advent of satellite measurement.
Additionally, Earth’s climate system is a chaotic system. The Butterfly Effect got its name, ironically, from that fact.
These facts are heretical in the context of the $$$$$$$ “climate change” industry… but they are the truth.
Oak Ridge National Lab hired me some years ago to do modeling of atmospheric/CO2 variables and their relationships. My views are well-informed.
John Anderson jbuand@gmail.com
I agree. I have seen other climatologists demonstrate that there is an energy imbalance with many of the current models that are effecting the predictive qualities of these models.
It seems that it would be an oversight to assume that the quantity of solar radiation reaching and interacting with our atmosphere is static and not something that needs to be modeled, even separately from the climate models.
What, is solar variation not included in the models?
Detailed data sets not long enough?
Yes, solar radiation does vary, but looking at the 11 year cycle doesn’t capture it and the bigger impact from our perspective is due to variation in Earth’s magnetic field.
citation needed, Phil B.
You need a citation that the amount of solar radiation reaching earth is impacted by changes to our magnetic field?
If you're claiming variation in our magnetic field is contributing to the current rapid climate change, then yes. That's not a consensus claim, so the burden is on you to support it.
Zeke, a question about how this comparison relates to uncertainty in future projections:
To what extent does the observed agreement (or disagreement) between models and historical temperatures affect our confidence in the uncertainty ranges for projections out to 2100? In other words, how should we think about the way historical model error propagates into the “error bars” around future warming under a given emissions scenario?
Zeke,
Good post. I'd be interested in your views as to what the distribution of model results and various summary statistics actually means, given that this is an ensemble of opportunity, and not a random selection of balls from an urn.
See:
"Multi-model datasets are often described as ‘ensembles of opportunity’. This refers to how they are assembled, namely by asking for model results from anyone who is willing to contribute. Various non-scientific aspects determine the size and composition of such ensembles, e.g. whether modelling groups are interested at all and whether they have funding and computational resources to do the requested simulations.
The implication of how these multi-model ensembles are created is that the sampling is neither systematic nor random. The distribution of the models is thus completely arbitrary, and might be different in a subsequent ensemble, therefore changing the result even if the knowledge about the climate system has not changed. Even if we were to solve the problem of weighting the individual members, the posterior would still depend on the prior distribution which is determined by human decisions and cannot be interpreted in a scientific sense. . .
Therefore, it is probable that the range covered by the multi-model ensemble covers a minimum rather than the full range of uncertainty."
Tebaldi, C., & Knutti, R. (2007). The use of the multi-model ensemble in probabilistic climate projections. Philosophical transactions of the royal society A: mathematical, physical and engineering sciences, 365(1857), 2053-2075.
The fact that CMIP-style models are an ensemble of opportunity is not ideal; one of our recommendations for the CMIP7 cycle is to focus more on perturbed physics ensembles, where variations of a single model can often more thoroughly explore the solution space. For example, a single perturbed physics experiment with HadCM3 gives a range of carbon cycle feedback responses twice as large as the C4MIP ensemble even when constrained to reproduce historical CO2 concentrations: https://www.carbonbrief.org/analysis-how-carbon-cycle-feedbacks-could-make-global-warming-worse/
Emulators are another way to potentially more thoroughly explore the solution space for aggregate variables like GMST, as they can be calibrated based on a full PDF of plausible values (e.g. for climate sensitivity or carbon cycle feedbacks) rather than relying on the distribution of ESM values for those variables: https://gmd.copernicus.org/articles/17/8569/2024/
Agreed
Emulators are problematic as well, see:
Nayak, M. S., Armour, K. C., & Battisti, D. S. (2025). Evaluating physical climate model emulators for global warming projections. Geophysical Research Letters, 52(23), e2025GL118591.
https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2025GL118591
It would be much more rigorous to compare projections made at time X with the future evolution of projected variables at time X+Y, and then deconvolving reasons for accuracies or inaccuracies, skill or lack of skill, etc..
Showing a multi-model mean and big envelope of model results based on prescribed forcings does not tell us much about the fidelity of any particular model or allow researchers to identify strengths or weaknesses of individual models.
Schmidt, G. A., et al. "Practice and philosophy of climate model tuning across six US modeling centers." Geoscientific Model Development 10.9 (2017): 3207-3223.
More generally, GAST is not relevant to mitigation or adaptation in the real world, so it is not clear what a retrospective analysis of the model ensemble with respect to GAST actually signifies, other than the fact that modelers can create an envelope of results that contains observations. One good way to do this of course is to have a big envelope.
Obviously, models do pretty poorly at the regional level for variables like precipitation.
Vautard, Robert, et al. "Evaluation of the large EURO‐CORDEX regional climate model ensemble." Journal of Geophysical Research: Atmospheres 126.17 (2021): e2019JD032344.
At the end of the day, plausibility for climate sensitivity is not determined using these models but rather by a range of expert opinion based on various lines of evidence. The question is then what are these models for? There doesn't seem much use in repeatedly showing model envelopes on blogs as if it means something.
https://aeon.co/essays/todays-complex-climate-models-arent-equivalent-to-reality
"The immense complexity of the climate makes it impossible to model accurately. Instead we must use uncertainty to our advantage"
There is no correlation between the monthly ENSO data and the monthly Mauna Loa CO2 data.
There is no correlation between the monthly ENSO data and monthly Mauna Loa CO2 values
How is that relevant, Ken?