Excellent work and communication...great science in action.
I was in the Australian Bureau of Meteorology in the 2000's when the sceptics were using the same tactic about the Australian high quality station data record ACORNSAT.
Measuring things accurately is not easy and often raw data can be misleading.
But again, same thing - comparing the all Australian raw temperature to the quality controlled data it was actually showed more warming.
I think my two favorite arguments are (1) "the climate has always changed" (at which point I'll teleport* them to a period of the planets history with the highest CO2 levels), and (2) "CO2 is plant food" (at which point I'll lock them in a room set at varying temps above 35°C, with no water and tons of dehydrated food).
_______________
*Or maybe just ask them how the hell they know that the climate has changed in the past. ("Were you there?")
They'd have to admit that they got the information about previous climates from ·scientists·, at which point they need to describe how they choose which science to believe.
I assume the origin of the figure is Christy's "Decline" paper in Theoretical & Applied Climate. I've been looking into that. Obviously the big problem is the unadjusted dataset, but there are several others. One is that Tmin thresholding omits a validity check, so that `if( jval(iyr,idy) .le. ntval(idy)` returns true for missing days (-999) as well as very cold days, resulting in a cold wave overcount. There's also an edge-miscount at the start and end of warm and cold seasons. Seasons miscount leap years, which may create another spurious cold day (but at least that's temporally uniform). A result of all this is that the paper's stated Tmax overcount potential (0.9%) is understated by a factor of 3-4, and the error grows with time due to time of observation shifts.
The results in the paper really aren't reproducible, because the .f code linked in the supplement is an "example", not the actual code used, as evidenced by differences in filenames in the header from those provided. A couple of metadata files required to run it aren't provided. 90% of the result is due to data changes, not methods, but the data files provided don't document which points are original, and which were edited or extended (though you can generally work this out after the fact by comparing with the raw original). Some of the "augmented" data series are just bonkers, and disagree with surrounding stations and even UAH TLT.
I have to say though that the USHCN chain is not entirely replicable either, at least so far as I can determine. There's no public source for the 52j adjustment code corresponding with the 2015 tech note. There's no daily USHCN, and the corresponding daily GHCNd is raw-only. The time-of-observation metadata isn't distributed with the codes - you have to dig it up from an old station history file - and the time stamp in GHCN is wrong for stations that have a morning precip/afternoon temp schedule. In the end it's not wrong, but this is the kind of fuzz the denial industry feeds on. I would be happy to be proven wrong on this.
"...the heat of the 1930s was a regional event..."
----
Ask most Americans what percentage surface of the globe the entire fifty states represents, and they respond on the high side. The total area of all fifty states represents →2%← of the globe's surface.
There is an important bias in counting the number of record high temperature readings as an indicator of climate trends. It is a biased statistic because, in a stable climate, the probability of setting a temperature record in the Nth year of measurement at a particular location is 1/N. If the rate of record high temperature measurements were constant, that would indicate a warming climate.
This is looking at a different problem, however. If you take the 125 year record and plot the years where a record occurred, it should be flat for a stationary climate.
Which is interesting taken together with this June 2026 paper https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2026GL122424 saying the same thing, the unmasking of warming by reduction in aerosols is behind recent changes in atmospheric dynamics that are driving heatwave persistence and intensity and summer temperature increases in Europe.
If the climate is described by a standard zero-drift Brownian motion, if you don't mind calling it stable climate, you'd still have records here and there. In this case, I don't know what you mean by "flat".
The of setting a temperature record in the Nth year of measurement depends on what stochastic process that you assume. E.g., even if the expectation is the same the left tail can be heavier than the right tail or the other way around. So, generally, it is not 1/N.
In the linked post, you "define" a stable climate as one that doesn’t have any forced climate change in it. I don't know what that means. The planet has a lot history of unforced (by humans) climates that were changing by a lot and sometimes very fast.
A model is a just model. (All models are wrong but some are useful.) I think the model runs that you are referring to only mean that the model lacks realistic inter-annual variability, as we know we all do. So it doesn't prove anything.
I would define climate by some set of weather statistics such as average temperature, daily low and high temperature plus other variables such as humidity, precipitation, wind, etc as a function of the time of year. I would define a stable climate as one for which the expectation values of these variables does not change from year to year. One could model such a climate by choosing the values for each year from some probability density which does not change with time.
Brownian motion is definitely not stable. Expectations for future years are given by the most recent value. Expectations would fluctuate with the present weather and drift off in an unbounded way.
On a planet with constant insolation, constant albedo and constant atmospheric composition, there will still be internal dynamics which produce variability over multi-year time scales. These could perhaps be described by a stochastic process, but that process would need to be mean-reverting to reproduce a stable climate. Brownian motion is not mean-reverting.
On another topic, more related to the original post, I'd like to add my support to the criticism of those who are obsessed with using the "raw data". I worked for many years as an experimental physicist and the idea that we could gain a better understanding of the physics by focusing on the raw data is utter nonsense. Any scientific instrument needs to be calibrated and bad data needs to be filtered out. The idea that this is some sort of conspiracy reveals a complete unfamiliarity with experimental science.
Your definition of stable climate doesn't work (yet) because you need to explain what you mean by expectation. You need a probability measure or something ad-hoc that would allow independent observers to look at the climate record and determine whether it is stable.
If you write dT = \sigma * dW and integrate in time you'll get T(t)=T(0)+\sigma*integral(dW). After taking expectation, the stochastic term vanishes and you have E[T(t)] = E[T(0)]. In words, expected future temps are the same as today. In my opinion, this is a possible and reasonable definition of a stable climate.
I don't see why the process needs to be mean-reverting because BM will eventually return to the initial state with prob 1 anyway. But I don't mind using Ornstein–Uhlenbeck process instead. You will still get occasional records and next year temps will be conditional on previous.
When you discuss records set in 1930 vs now, any record set now is above 1930 temperatures values, only higher values are called record temperatures now, right?
what we're doing here is: for every station, we find the year with the record-setting maximum temperature. then, for each year, we count how many stations had records in that year.
Thanks Andrew ....i think most of us can understand the frustration of having to deal with the, shall we say, misrepresentation of the facts , when if someone was really just sceptical they could do the research themselves to check the facts .
whatever.....moving on...
on a slightly different note , i have been following the recent surge in both EV sales and comment around new battery tech ......especially sodium ion and solid-state battery's....
and leading on from that , there has been increasing comment about the reduction in demand for oil
another article mentioned that a Chinese Province (Hainan) was now finalised a ban on new ICE car sales from 2030 .....it is the first Chinese Province to do so
I've written about the Iran War and the true cost of fossil fuels here: https://www.theclimatebrink.com/p/the-war-in-iran-shows-us-another. The recent request from the Pentagon for $70B-ish shows that it's just getting more expensive . I predict that, when the history of energy is written, the Iran War will be a pivotal event in speeding up decarbonization.
Ethiopia banned ICE imports in January 2024, and has been moving headlong into a "green economy" in terms of electrification and electricity generation. It's one of many countries that are tired of buying imported fuel (or crude to refine). Of course Ethiopia not only has abundant sun, but it is covers part of the African Rift Valley which provides shallow geothermal resources, so it's an easy move for them. (They've also arrange to import Chinese electric semi components for domestic assembly.)
Every adjustment reduces some uncertainty while introducing new assumptions. The challenge is not to eliminate uncertainty, but to make every assumption transparent and continually testable. That, to me, is how science progresses.
Perhaps I am missing something, but doesn’t a continuous level of daily record highs imply that the highs are ratcheting upward? Fine to correct for biases, but even without the correction we are looking at bad news.
Retired hydrologist here. I wanted to second that adjusting (or working with, or analyzing) raw data is very common and not controversial at all. In fact for historical climatic or hydrologic data it is almost always required as part of the data analysis stage of a study to produce robust, defensible results. And, there is no sleight of hand: the analysis applied to the raw data is described in the publication for the study so that readers can understand what was done (and so that peer reviewers can assess the study methodology).
Great post and thanks for sharing. Could you comment on how Tmin and Tmax data are converted to a daily estimate of average air temperature when creating homogenized GHCNd datasets? There is not much (good) literature on this topic as far as I'm aware. Is it simply Tavg = 1/2(Tmin + Tmax)? Maybe this doesn't matter at the annual scale, but it likely matters for both evaluating models (which simulate the full 24h cycle of air temperature) and possibly seasonal trends. Thanks
In this post, we're analyzing Tmax. Tavg is the average of Tmin and Tmax for older measurements, but when hourly measurements became available with electronic thermometers, they switched to averaging those to determine Tavg.
I see. I thought the plot showing the global land area with raw and adjusted surface temperature anomaly might be the average, not tmax.
If what you’re saying is correct that seems like another potentially important adjustment to make in reconstructing unbiased daily air temps. (I’m sure someone has addressed this; I’m just interested in the small details of doing this right).
Excellent work and communication...great science in action.
I was in the Australian Bureau of Meteorology in the 2000's when the sceptics were using the same tactic about the Australian high quality station data record ACORNSAT.
Measuring things accurately is not easy and often raw data can be misleading.
But again, same thing - comparing the all Australian raw temperature to the quality controlled data it was actually showed more warming.
https://www.bom.gov.au/climate/data/acorn-sat/#tabs=Methods
But still ended up with attacks pointing at particular stations and the suggestions of tampering.
Yes, climate deniers have a very small number of very bad arguments.
I think my two favorite arguments are (1) "the climate has always changed" (at which point I'll teleport* them to a period of the planets history with the highest CO2 levels), and (2) "CO2 is plant food" (at which point I'll lock them in a room set at varying temps above 35°C, with no water and tons of dehydrated food).
_______________
*Or maybe just ask them how the hell they know that the climate has changed in the past. ("Were you there?")
Lame
They'd have to admit that they got the information about previous climates from ·scientists·, at which point they need to describe how they choose which science to believe.
This should be used in a philosophy of science class. Thanks.
I assume the origin of the figure is Christy's "Decline" paper in Theoretical & Applied Climate. I've been looking into that. Obviously the big problem is the unadjusted dataset, but there are several others. One is that Tmin thresholding omits a validity check, so that `if( jval(iyr,idy) .le. ntval(idy)` returns true for missing days (-999) as well as very cold days, resulting in a cold wave overcount. There's also an edge-miscount at the start and end of warm and cold seasons. Seasons miscount leap years, which may create another spurious cold day (but at least that's temporally uniform). A result of all this is that the paper's stated Tmax overcount potential (0.9%) is understated by a factor of 3-4, and the error grows with time due to time of observation shifts.
The results in the paper really aren't reproducible, because the .f code linked in the supplement is an "example", not the actual code used, as evidenced by differences in filenames in the header from those provided. A couple of metadata files required to run it aren't provided. 90% of the result is due to data changes, not methods, but the data files provided don't document which points are original, and which were edited or extended (though you can generally work this out after the fact by comparing with the raw original). Some of the "augmented" data series are just bonkers, and disagree with surrounding stations and even UAH TLT.
I have to say though that the USHCN chain is not entirely replicable either, at least so far as I can determine. There's no public source for the 52j adjustment code corresponding with the 2015 tech note. There's no daily USHCN, and the corresponding daily GHCNd is raw-only. The time-of-observation metadata isn't distributed with the codes - you have to dig it up from an old station history file - and the time stamp in GHCN is wrong for stations that have a morning precip/afternoon temp schedule. In the end it's not wrong, but this is the kind of fuzz the denial industry feeds on. I would be happy to be proven wrong on this.
Yes, the data are a rat's nest. We'll have more to say about Christy's paper in an upcoming post.
Happy to share my detailed findings (about 9 pages) if helpful.
👍 analysis, Zeke
As Earth 🔥, heat record breaking will accelerate.
The old records, eg, heated 30s USA, will be toppled.
And, lo and behold, it is happening.
Thanks for the teaching software tip! Very timely for me!
Nicely done. Thanks.
"...the heat of the 1930s was a regional event..."
----
Ask most Americans what percentage surface of the globe the entire fifty states represents, and they respond on the high side. The total area of all fifty states represents →2%← of the globe's surface.
There is an important bias in counting the number of record high temperature readings as an indicator of climate trends. It is a biased statistic because, in a stable climate, the probability of setting a temperature record in the Nth year of measurement at a particular location is 1/N. If the rate of record high temperature measurements were constant, that would indicate a warming climate.
Yes, you're exactly right, as I wrote here: https://www.theclimatebrink.com/p/why-are-there-more-high-temperature?utm_source=publication-search
This is looking at a different problem, however. If you take the 125 year record and plot the years where a record occurred, it should be flat for a stationary climate.
Richard Crim (RIP) was first commenter on this 2023 post with his own about reduction in aerosols being behind accelerated warming. https://richardcrim.substack.com/p/the-crisis-report-39?r=a66g&utm_campaign=post&utm_medium=web&triedRedirect=true
Which is interesting taken together with this June 2026 paper https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2026GL122424 saying the same thing, the unmasking of warming by reduction in aerosols is behind recent changes in atmospheric dynamics that are driving heatwave persistence and intensity and summer temperature increases in Europe.
Crim was right about the aerosol masking. Great article and thank you for sharing it.
You’re very welcome, Jeff 🙏🏼
If the climate is described by a standard zero-drift Brownian motion, if you don't mind calling it stable climate, you'd still have records here and there. In this case, I don't know what you mean by "flat".
The of setting a temperature record in the Nth year of measurement depends on what stochastic process that you assume. E.g., even if the expectation is the same the left tail can be heavier than the right tail or the other way around. So, generally, it is not 1/N.
if you look at a climate model, it follows 1/N almost exactly:
https://www.theclimatebrink.com/p/why-are-there-more-high-temperature?utm_source=publication-search
In the linked post, you "define" a stable climate as one that doesn’t have any forced climate change in it. I don't know what that means. The planet has a lot history of unforced (by humans) climates that were changing by a lot and sometimes very fast.
A model is a just model. (All models are wrong but some are useful.) I think the model runs that you are referring to only mean that the model lacks realistic inter-annual variability, as we know we all do. So it doesn't prove anything.
I would define climate by some set of weather statistics such as average temperature, daily low and high temperature plus other variables such as humidity, precipitation, wind, etc as a function of the time of year. I would define a stable climate as one for which the expectation values of these variables does not change from year to year. One could model such a climate by choosing the values for each year from some probability density which does not change with time.
Brownian motion is definitely not stable. Expectations for future years are given by the most recent value. Expectations would fluctuate with the present weather and drift off in an unbounded way.
On a planet with constant insolation, constant albedo and constant atmospheric composition, there will still be internal dynamics which produce variability over multi-year time scales. These could perhaps be described by a stochastic process, but that process would need to be mean-reverting to reproduce a stable climate. Brownian motion is not mean-reverting.
On another topic, more related to the original post, I'd like to add my support to the criticism of those who are obsessed with using the "raw data". I worked for many years as an experimental physicist and the idea that we could gain a better understanding of the physics by focusing on the raw data is utter nonsense. Any scientific instrument needs to be calibrated and bad data needs to be filtered out. The idea that this is some sort of conspiracy reveals a complete unfamiliarity with experimental science.
Your definition of stable climate doesn't work (yet) because you need to explain what you mean by expectation. You need a probability measure or something ad-hoc that would allow independent observers to look at the climate record and determine whether it is stable.
If you write dT = \sigma * dW and integrate in time you'll get T(t)=T(0)+\sigma*integral(dW). After taking expectation, the stochastic term vanishes and you have E[T(t)] = E[T(0)]. In words, expected future temps are the same as today. In my opinion, this is a possible and reasonable definition of a stable climate.
I don't see why the process needs to be mean-reverting because BM will eventually return to the initial state with prob 1 anyway. But I don't mind using Ornstein–Uhlenbeck process instead. You will still get occasional records and next year temps will be conditional on previous.
https://en.wikipedia.org/wiki/Ornstein–Uhlenbeck_process
When you discuss records set in 1930 vs now, any record set now is above 1930 temperatures values, only higher values are called record temperatures now, right?
what we're doing here is: for every station, we find the year with the record-setting maximum temperature. then, for each year, we count how many stations had records in that year.
Thanks Andrew ....i think most of us can understand the frustration of having to deal with the, shall we say, misrepresentation of the facts , when if someone was really just sceptical they could do the research themselves to check the facts .
whatever.....moving on...
on a slightly different note , i have been following the recent surge in both EV sales and comment around new battery tech ......especially sodium ion and solid-state battery's....
and leading on from that , there has been increasing comment about the reduction in demand for oil
https://www.cnbc.com/2026/07/10/iea-world-oil-demand-declines-iran-war.html
another article mentioned that a Chinese Province (Hainan) was now finalised a ban on new ICE car sales from 2030 .....it is the first Chinese Province to do so
https://www.electrive.com/2026/07/15/first-chinese-province-bans-sales-of-combustion-engine-vehicles/
would you care to comment at some stage ?
I've written about the Iran War and the true cost of fossil fuels here: https://www.theclimatebrink.com/p/the-war-in-iran-shows-us-another. The recent request from the Pentagon for $70B-ish shows that it's just getting more expensive . I predict that, when the history of energy is written, the Iran War will be a pivotal event in speeding up decarbonization.
Meanwhile, the war is also restricting the supply of sulfur, required for processing guano into water-soluble phosphate fertilizer.
Guano is true bs ( bat s). The war is all bs.
Also, hopefully, the fall of Trump. MAGHGATS against MAGHGATS.
Note their sick directive:
Make A Greenhouse Gas Armageddon.
Ethiopia banned ICE imports in January 2024, and has been moving headlong into a "green economy" in terms of electrification and electricity generation. It's one of many countries that are tired of buying imported fuel (or crude to refine). Of course Ethiopia not only has abundant sun, but it is covers part of the African Rift Valley which provides shallow geothermal resources, so it's an easy move for them. (They've also arrange to import Chinese electric semi components for domestic assembly.)
Was there more concrete, asphalt, and urbanization in the 1930s or less than now? Just asking for clarification and common sense purposes.
Every adjustment reduces some uncertainty while introducing new assumptions. The challenge is not to eliminate uncertainty, but to make every assumption transparent and continually testable. That, to me, is how science progresses.
Perhaps I am missing something, but doesn’t a continuous level of daily record highs imply that the highs are ratcheting upward? Fine to correct for biases, but even without the correction we are looking at bad news.
I see that Roger Pielke and Chris Martz responded but I don't see your reactions to their comments.
Retired hydrologist here. I wanted to second that adjusting (or working with, or analyzing) raw data is very common and not controversial at all. In fact for historical climatic or hydrologic data it is almost always required as part of the data analysis stage of a study to produce robust, defensible results. And, there is no sleight of hand: the analysis applied to the raw data is described in the publication for the study so that readers can understand what was done (and so that peer reviewers can assess the study methodology).
Thanks for that. My experience as well.
Great post and thanks for sharing. Could you comment on how Tmin and Tmax data are converted to a daily estimate of average air temperature when creating homogenized GHCNd datasets? There is not much (good) literature on this topic as far as I'm aware. Is it simply Tavg = 1/2(Tmin + Tmax)? Maybe this doesn't matter at the annual scale, but it likely matters for both evaluating models (which simulate the full 24h cycle of air temperature) and possibly seasonal trends. Thanks
In this post, we're analyzing Tmax. Tavg is the average of Tmin and Tmax for older measurements, but when hourly measurements became available with electronic thermometers, they switched to averaging those to determine Tavg.
I see. I thought the plot showing the global land area with raw and adjusted surface temperature anomaly might be the average, not tmax.
If what you’re saying is correct that seems like another potentially important adjustment to make in reconstructing unbiased daily air temps. (I’m sure someone has addressed this; I’m just interested in the small details of doing this right).