The Diverging Surface Thermometer and Satellite Temperature Records

  • HadCRUT4* and the NASA GISS LOTI time temperature series, that are based on surface thermometers, show a high level of agreement (Figure 1). Offset differences  between the two is roughly 0.2˚C which is largely inconsequential in the bigger picture. One curious discrepancy is that GISS LOTI temperatures have continued to rise since 1998 while HadCRUT4 is more flat.
  • The UAH and RSS satellite based time temperature series also show a high degree of agreement largely within 0.1˚C.
  • A linear regression through the HadCRUT4 data shows a long-term warming trend of 0.48˚C per century.  A linear regression through the UAH satellite data shows warming of 1.11˚C per century. The much higher rate of warming is because the satellite record, that begins in 1980, only samples the recent up leg of what is fluctuating time temperature data (Figure 1).
  • Surface thermometers compared with satellite data also show a high degree of agreement, which is a credit to all the scientists that work on compiling these fundamentally important data sets. The differences are subtle, but the surface thermometer series and satellite data are diverging with time over the relatively short 35 years of satellite data. In this time frame, the surface thermometers are warming at a rate of 0.42˚C per century more rapidly than the satellite data. This is a substantial bias that requires an explanation. This bias is sufficient to send surface thermometers into record breaking territory while the satellite data continues to move sideways and down since 1998.

* all data sources and acronyms are explained at the end of the post

Figure 1 GISS LOTI is biased to higher T anomalies than HadCRUT4. This is largely down to the different base periods chosen for calculating anomalies; 1951-1980 for GISS and 1961-1990 for HadCRUT4. The linear regression through HadCRUT4 is largely matched by parallel rising tops and bottoms (dashed lines). GISS LOTI is rising at 0.69˚C per century (not shown), appreciably higher than HadCRUT4, and caused in part by the recent acceleration in the GISS index.

With the Paris climate and global energy system talks looming there is a lot of interest in whether or not the Earth’s atmosphere (lower troposphere to be precise) is still warming.  For years now, there has been talk of “a pause” in warming since 1998. But now, just in time, there is a flurry of stories about record temperatures being set in England (during one of the coldest summers in recent times), and that 2015 is heading to be the warmest year ever (since records began). The climate science community gloats over rising temperatures; evidently more concerned about the reliability of their models and research grants than they are about the well fare of the Human Race and Earth ecosystems. On the other side of the coin, the sceptics watch the Sun approach a grand solar minimum and hope that the Seine freezes over in December.

I decided it was time to take a closer look at the temperature records upon which all the propaganda is based.  A main message I wish to send is that the differences between the 4 temperature records examined are small and subtle and given the uncertainties, decisions about our energy future are being made in an information framework that is far from certain. The opinion of anyone espousing 97% certainty should in the circumstances be disregarded.

Surface Thermometers

There are at least 5 compilations of surface thermometer records. In this post I examine two of them, the NASA GISS LOTI and HadCRUT4 (Figure 1). These reconstructions are using largely the same data archive but the application of different methodologies means that the various groups arrive at somewhat different answers.

One of the key differences is selection of base period for the calculation of anomalies (Figure 1) . The average temperature of the base period is deducted from the measured temperatures for all months to determine the anomaly and since different base periods are used this results in GISS being linearly displaced to higher temperatures than HadCRUT4 (Figure 1). But that is not all that is going on.

Figure 2 GISS LOTI minus HadCRUT4. The main observations are: 1) the positive bias in GISS most likely comes down to the selection of different base periods; 2) the differences are small, mainly confined to +015 and -0.05˚C; and 3) the cyclical variance, especially between 1920 and 1970 requires an explanation, but that is beyond the scope of this post.

Deducting HadCRUT4 from GISS LOTI quantifies the residuals as shown in Figure 2. It is a rather curious pattern. Most of the dT is confined to +0.15 and -0.05˚C (±0.1) which in the larger picture is a rather small difference. Since the difference is small I will not try to explain it. But it is worth noting that since 1970 GISS LOTI first diverged, then converged and is now diverging again from HadCRUT4. The recent divergence during the last 15 years is evident in Figure 1 where NASA GISS can be seen to be warming at a higher rate than HadCRUT4.

The average warming rates are:

GISS LOTI 0.69˚C per century
HadCRUT4 0.48˚C per century

The difference is in fact small and likely within the uncertainty. I will return to these surface thermometer records following a description of the satellite data.


The procedure for measuring the temperature of the lower troposphere from space employing microwave data is complex. The satellites that do this are in decaying orbits, burn up and need to be replaced on a regular basis. Roy Spencer at UAH provides a good description.

There have been several satellites and the data are merged into a single data set. This data is analysed by two separate groups. One led by Roy Spencer at the University of Alabama, Huntsville (UAH) and the other by Remote Sensing Systems (RSS). The temperature records are compared in Figure 3.

Figure 3 Comparison of the two satellite records. They are very closely alined, with dT lower than 0.15˚C (Figure 4). 

Figure 3 shows how the two analyses are closely alined but are offset by about 0.15˚C, presumably because of calibration and or normalisation differences (Figure 4). The gradients are identical (1.1 v 1.2˚C per century). The satellite data is often used to argue that there has been no warming since 1997, and indeed a regression through the post-1997 data has a negative gradient. The last year plotted is 2014, but it is plain to see that the satellite data is nowhere close to recording record high temperatures. The warmest year remains 1998 followed by 2010.

Figure 4 dT between RSS and UAH determined by subtracting UAH from RSS (Figure 3).

Comparison of Thermometers and Satellites

Having set the scene, I now get to the main point of the post and that is to compare the thermometer and satellite records. To do this I have removed the small biases described above by re-normalising the data so that the mean of the 5 year period 1980 to 1984 = 0.00˚C. To do this the following adjustments have been applied:

GISS LOTI -0.242˚C
HadCRUT4 -0.085˚C
UAH +0.145˚C
RSS + 0.059˚C

This has the effect of making all of the indexes begin at the same datum point of 0.0˚C at the beginning of the time series. It does not affect structure or gradient of the data. The result shown in Figure 5 shows a number of things: 1) the two satellite records and the two thermometer records are very closely alined, 2) in a general sense there is good agreement between thermometers and satellites and 3) it is evident that the thermometers and satellites are diverging (Figures 6 and 7).

Figure 5 The two satellite and two thermometer records compared. One should not lose sight of the fact that these two very different approaches to measuring the mean temperature of the lower troposphere are providing quite closely alined results. 

Figure 6 A simplification of Figure 5 showing the means of thermometers and satellites. While the differences are small and subtle, the outcome is not. Thermometers mark 2014 as the warmest year, satellites show 2014 as nothing out of the ordinary.

Simplifying one stage further, Figure 6 shows the simple arithmetic means of the thermometer and satellite data and shows that since 1980, the thermometer record has been warming +0.4˚C per century more than the satellite record. This is a small difference but it is significant because it pushes the surface thermometers into the territory of “record warmth” while the satellites are nowhere close to recording new temperature highs. Note that the warming trends are for the steep up leg since 1980 (Figure 1) and should not be extrapolated into the future since we know that natural oceanic cycles (e.g the Atlantic Multi-decadal Oscillation or AMO) will drive temperatures sideways or down at some point in the future.

Figure 7 Subtracting satellite temperatures from thermometers provides this picture. Up until 1998 there is no clear bias but after that the surface thermometers go into over drive. But note the absolute differences are tiny, about 0.15˚C.

Finally, Figure 7 shows the difference between the thermometer and satellite trends (Figure 6) and shows that the thermometers have been steadily diverging from the satellites since 1980, especially since 2004, with two notable exceptions during the big El Nino years of 1998 and 2010 where the satellites run warmer than the thermometers. Removing these two years from the regression, the R^2 improves to 0.51.

The origin of the trends in Figures 6 and 7 require an explanation. Either the satellite or thermometer records are biased in some way. Note that the differences between thermometers and satellites is minute – only +0.15˚C in 2014 and when you consider that this is supposed to represent the average temperature of the lower troposphere the difference is trivial. What is not trivial is the trend through the data which as already mentioned is sufficient to have thermometers recording record temperatures while the satellite record since 1998 is actually trending down. This minute difference will be used by the climate science community to argue for vast penalties to be imposed upon humanity in the Paris talks. The stakes could hardly be higher.

My own prejudice, it is the thermometer record that is subtly biased for a number of reasons. First, the climate science community seems over eager to find warming wherever they look and to exaggerate dubious observations. A good example of this would be UK MET office claims of new record temperatures at Heathrow Airport in July of this year which seems likely to have been caused by hot air from a jet. Credible scientists would not seek to use or publicise data from such a dubious source as Heathrow Airport. The surface thermometer record is also fraught with problems one being that many records are unreliable. Rather than discarding unreliable data the climate science community choses instead to try and correct them employing dubious homogenisation algorithms. This opens the door to data manipulation. And for example, in the case of Iceland, well curated records have been changed creating the aura of record warmth on that island that quite simply does not exist. Finally, I believe that land surface changes such as deforestation and irrigation may impact surface thermometers in a way that goes beyond the well documented urban heat island effect. This in essence may create near surface anthropogenic global warming that is not caused by greenhouse gasses and may be viewed as benign.


Acronyms, Abbreviations and Data Sources

HadCRUT4 stands for Hadley Research Centre at the UK Met office and Climate Research Unit of the University of East Anglia and Temperature and version 4. Data downloaded from this link:

NASA is the US National Aeronautics and Space Administration. GISS is the Goddard Institute for Space Studies. LOTI is the Land Ocean Temperature Index. Data downloaded from this link:

UAH is the University of Alabama, Huntsville where Roy Spencer and John Christie are the curators of the satellite microwave data. Data for V6 were downloaded from this link:
Selecting the file tltglhmam_6.0beta3

RSS is Remote Sensing Systems, the other curator of the satellite microwave data. Data were downloaded from this link:

Clicking on FTP Air Temp Time Series and then selecting the first file called


This entry was posted in Climate change and tagged , , , , , , , . Bookmark the permalink.

34 Responses to The Diverging Surface Thermometer and Satellite Temperature Records

  1. mbe11 says:

    Interesting presentation. Have you tried doing this same thing with the temperature data from two generations back to see if the satellite data still tracks? If the divergence is more in the present that would suggest, at least to me, something is seriously wrong with the homogenization scheme.

  2. oldfossil says:

    Our good buddy Steve Goddard, who does seem to have a bit of a one-track mind, frequently posts charts of raw T data vs the homogenised version. They show that temperature adjustments are always upwards in the modern era and downwards in the early 20th century. This puzzles me somewhat. Commonsense says that the adjustments across a large number of stations should average out to nil. This leads one to suspect a systematic bias in the homogenisation.

    On the other hand Dr. Judith Curry, whose integrity I do not question, was part of BEST before she turned lukewarmist/sceptic. She stands by the validity of the BEST series, which shows the same apparent systematic bias.


    • Euan Mearns says:

      It has been established that homogenisation adds about 0.2˚C to the global land record since 1880, but given that the land accounts for only 29% of the total, the net effect is expected to be rather small.

    • A C Osborn says:

      I think that your question about BEST is valid and perhaps I can provide a quote from Steve Mosher (who is involved with BEST) from another post about the discrepancies found throughout BEST “Final Output”.
      In response to this statement “However, after adjustments done by BEST Amundsen shows a rising trend of 0.1C/decade.
      Amundsen is a smoking gun as far as I’m concerned. Follow the satellite data and eschew the non-satellite instrument record before 1979.”
      He wrote
      “BEST does no ADJUSTMENT to the data.

      All the data is used to create an ESTIMATE, a PREDICTION

      “At the end of the analysis process,
      the “adjusted” data is created as an estimate of what the weather at
      this location might have looked like after removing apparent biases.
      This “adjusted” data will generally to be free from quality control
      issues and be regionally homogeneous. Some users may find this
      “adjusted” data that attempts to remove apparent biases more
      suitable for their needs, while other users may prefer to work
      with raw values.”

      With Amundsen if your interest is looking at the exact conditions recorded, USE THE RAW DATA.
      If your interest is creating the best PREDICTION for that site given ALL the data and the given model of climate, then use “adjusted” data.

      See the scare quotes?

      The approach is fundamentally different that adjusting series and then calculating an average of adjusted series.

      in stead we use all raw data. And then we we build a model to predict
      the temperature.

      At the local level this PREDICTION will deviate from the local raw values.
      it has to.

  3. Peter Lang says:

    Interesting post. Very clear. Looks credible, unbiased, honest.

    Question: Why can’t climate ‘scientists’ provide honest analyses and clear, believable explanations like this?

  4. gweberbv says:

    With this data you can do absolutely nothing unless you have a reliable model which you can adjust to the measured data. The data in Fig. 1 you could reproduce with a sinus function or with an exponential function. Or you simply fit a straigt line through it. Or you use an extremely sophisticaed modell with 1 million adjustable parameters that can even fit an elephant fighting with a crocodile. Each approach will give you a very different prediction for the year 2050 or 2100.

  5. Four graphs below:

    The first compares the CRUTEM4 (land) and HadSST3 (ocean) series, the two series that are averaged together to construct HadCRUT4. The difference plot at the bottom shows that warming trends over the land and the oceans since 1979 are quite different, with CRUTEM4 showing about 0.5C more warming over land than HadSST3 shows over the oceans. This is why temperature trends over the land and the ocean must be considered separately and not averaged together as HadCRUT4 (and GISS LOTI) do. More about this in these two posts:

    The second graph compares UAH lower troposphere temperatures over land with UAH lower troposphere temperatures over the ocean. The trend line shows the troposphere over land warming by about 0.2C relative to the troposphere over the ocean but this is largely an artifact of the shape of the plot. The absolute land-ocean temperature difference is about the same now as it was in 1979.

    The third compares the UAH “land” series with the CRUTEM4 land series. CRUTEM4 shows about 0.3C more warming than UAH since 1979.

    The fourth compares the UAH “ocean” series with the HadSST3 ocean series. Both show the same amount of warming since 1979 (the trend line is dead flat).

    Preliminary conclusion? UAH and HadSST3 agree, CRUTEM4 is the odd man out. The implication is that the 0.3C of warming shown by CRUTEM4 relative to UAH land may be a result of “adjustments”, which if it is could explain why the world continues to set new temperature records on a regular basis.

    • Euan Mearns says:

      Roger, great information! I think this is worthy of a 500 word post in its own right. Starting with your 4th chart, this kind of shows that the satellite data are reliable, does it not? And the second chart – UAH land v sea, shows that the troposphere over land is warming more rapidly than over the sea.

      Does this not point to greater radiative forcing over land? And is this not linked to land use changes etc as I suggest. The biases are all so small. But I’m not sure that its necessary to call on adjustments to explain this. Surface thermometers over land will be more prone to excess warming from human activity – like the jet at Heathrow.

      • A post on a subject this complicated would probably be closer to 5,000 than 500 words 🙁

        Does this not point to greater radiative forcing over land?

        Asking questions that would cross Paradise’s eyes ….

        (From “Fiddler on the Roof”) 😉

  6. Halken says:

    I believe there is also two weather balloon temperature datasets, that confirm the satellites. AFAIK.

    • There are indeed radiosonde records that more or less match the satellite records – I think ESRL has them – but whether the match means anything is questionable because they’ve been “adjusted” to show what global warming theory says they should show:

      Previously reported discrepancies between the amount of warming near the surface and higher in the atmosphere have been used to challenge the reliability of climate models and the reality of human-induced global warming. Specifically, surface data showe substantial global-average warming, while early versions of satellite and radiosonde data showed little or no warming above the surface. This significant discrepancy no longer exists because errors in the satellite and radiosonde data have been identified and corrected.

      From the US Climate Change Science Program 2006

  7. A C Osborn says:

    Euan, you said “Offset differences between the two is roughly 0.2˚C which is largely inconsequential in the bigger picture.”

    Do you realy think that 0.2˚C is “inconsequential” when the current so called “Record Temperatures” being banded about are talking about 0.01˚C differences to previous annual values?

  8. From a plant operations chemical engineer, this sort of difference is like meh. The trends are similar.

    Yes making the historic claim is not correct but we get people doing that all the time from one reading ignoring others. It is personal bias as opposed to looking at the data objectively.

    • Nial says:

      As a plant operations chemical engineer, what’s your reaction to the records since the end of the Little Ice Age?

      As an electonic engineer, mine is also ‘meh’.

  9. Sebastian says:

    Nice article. You are right about the thermometer trends being trivial compared to the downward course of the satellite record. From 1998 to 2008 there was a drop of half a degree! in world temperatures but I don’t remember hearing about that from climate scientists. I agree with you that the satellite record must be the reference data set – it has no UHI and has have global coverage – and is nicely are out of the way of manipulation by people with a vested interest in creating a rising trend. If thermometers don’t match the satellite record we can all guess why.

  10. Javier says:

    The average warming rates are:

    GISS LOTI 0.69˚C per century
    HadCRUT4 0.48˚C per century

    The difference is in fact small and likely within the uncertainty.

    It is not small in relative terms. GISS LOTI is showing a rate of warming 44% higher than HadCRUT4. In my opinion that is a huge difference and clearly one of them is a lot closer to the actual average rate of warming.

    One aspect that you haven’t got into is that this is the picture as of today. But while for the last decade and a half present temperatures have changed mostly through adjustments, past temperatures have changed too. So when we look at a recent graph, past temperatures do not look like they looked in an old graph. This constant rewriting of old records is unscientific as every analysis performed on previous data becomes invalidated. And this is not due to obtaining new supposedly better data as in other disciplines but to fiddling with computer codes that modify the only data available. It is a sad situation and if it continues we will have to go to proxy paleorecords to have an idea how warm the temperatures were in the 1930s.

    The publicity coming from thermometer temperature curators every time an irrelevant record is made by amounts within error of measure is a public proclamation of their bias. The adjustments net effect of warming the present, cooling the past, and eliminating 60-year cyclicity from the data, is a confirmation that their adjustments align with their bias. Even if the effect is small the damage made to science is large. Efforts to understand the hiatus and climate cyclicity are derailed if there is no hiatus as publicly claimed and temperatures raise linearly. These are not small potatoes. It is not the first time that science prostitutes but this time everybody is looking.

  11. Hmm, the language and highlighting here seems a little biased. You also failed to mention that the UAH data set is in beta and that, as far as I know, only Spencer knows the algorithms for correcting the data. HADCRUT4 is biased low due to the lack of coverage at the poles. Cowtan and Way used satellite data to correct HADCRUT4 for areas where coverage was weak and came up with a data series that is also counter to the UAH data.

    Of course, anyone wanting to reinforce their own view is likely to believe whichever series appears to fit with that view. I’m sure I’m no different. But taking actual thermometer measurements rather than microwave data, those data series agree. Personally, I think it likely that UAH is applying incorrect adjustments, and hope that someone will take the raw microwave data and re-do the work to convert to temperature measurements, adjusting for orbital decay and timing (and probably a bunch of other things).

    Examples of bias in the article:

    “One curious discrepancy is that GISS LOTI temperatures have continued to rise since 1998 while HadCRUT4 is more flat.” Actually, they have both continued to rise. “More flat” is misleading and probably nonsense, grammatically.

    “This is a substantial bias that requires an explanation.” You meant a substantial difference.

    “while the satellite data continues to move sideways and down since 1998” No, since 1998 (i.e. 1999 and later) they continue to move up, but much more slowly.

    “there is a flurry of stories about record temperatures being set in England (during one of the coldest summers in recent times)” The focus on one area of the planet is irrelevant.

    “and indeed a regression through the post-1997 data has a negative gradient” Your emphasis and also a cherry picked starting point. What does the regression post 1998 show?

    • Euan Mearns says:

      Mike, you seem awfully keen on the idea of a world warming under the influence of GHGs. But some of this is fair criticism. But I’d point out that I do declare my prejudices. So the pause that has caused so much soul searching no longer exists? Is that the new narrative?

      I would try to address your comments in more detail, but I hope you appreciate I just don’t have the time. The points I would make are 1) RSS and UAH are in very close alinement, and so if you are implying that Roy Spencer has somehow made an error or has cooked the books then so have RSS, and 2) I’d point to the charts that Roger posted, esp no 4 that shows close agreement between UAH and SSTs over oceans.

      The fact that all the land-based reconstructions agree with each other means about as much as the fact that the satellite reconstructions agree with each other. They are all using variants of the same input data and variants of the same physics / algorithms to get their results.

      My main message stands. The difference between satellites and thermometers is tiny, but there is a bias / trend of divergence. If this continues then one or other of the data sets will have to be proven wrong.

  12. gweberbv says:

    Could someone provide a link to download these data sets?

  13. Owen says:

    Is there not a problem here in that the satellite data doesnt go back far enough to discern an overall trend in temp ?

    • Euan Mearns says:

      Yes, its the biggest limitation of the satellite data. But the trend through 35 years of satellite data is different to the trend through 35 years of surface thermometers.

  14. Luís says:

    Hi Euan, nice post, this divergence is capital aspect in the climate change matter. I would note however, that HADCRUT3 is not divergent at all with the satellite records. In fact, it produces almost the same trend as the UHA dataset:

    The difference to HADCRUT4 is quite visible, and mainly due to the last decade:

    I believe the key to unveil this mystery is to understand why HARDCRUT4 warmed up relative to HARDCRUT3 the way it did.


    • Euan Mearns says:

      Hi Luis, thanks for that. It kind of sniffs, doesn’t it. Seamless transition from worrying about the pause to writing the pause out of history. I’m to do another post on this and will include this astute observation. Interested to hear what Roger has to say – its possible he’s already made this observation.

      • Sometimes, things get written out of history because they weren’t real. I’m amazed by this underlying notion that somehow there is a grand global conspiracy to keep the climate truth from emerging when it’s far more likely that the reverse is the case. The more our scientists examine this stuff, the more they learn that the situation is worse than they thought. The few science contrarians can’t even agree on the different views they take. Meanwhile, nature doesn’t give a toss about what people think and ecosystems continue to adjust as humans perturb them.

        • Euan Mearns says:

          The more our scientists examine this stuff, the more they learn that the situation is worse than they thought.

          Since 1997 the temperature trend is at best downwards, at worst using revised data, rising much more slowly than before. And you want to argue that things are worse than believed before. I’m afraid arguments like that have zero traction here.

    • Elvis says:

      They added many new stations in northerly latitudes. And as these places are warming faster, they push up the global trend.

      Note also that the satellite record is for the lower troposphere, so it is measuring something different and might be expected not to be the same as terrestrial data. And also the satellite measurements get confused by ice, of which there is more in northerly latitudes. So there seems plenty of scope for disagreement between data-sets.

  15. Luís says:

    As an addendum: the declining trend in the RSU dataset is now almost 19 years old:

    A cold 2016 or 2017 can quickly render this trend significant at the climatic scale.

  16. Pingback: AWED Energy & Environmental Newsletter: October 5, 2015 - Master Resource

Comments are closed.