Surface versus satellite; the temperature data set controversy

This post follows on from Euan Mearns’ recent posts on record heat and the Ratpac data set. Its goals are:

  • To clarify some points regarding what the satellite and “surface” temperature records are really telling us.
  • To see if we can define which temperature sets are reliable and which aren’t.
  • To draw appropriate conclusions.

Figure 1 compares the HadCRUT4 global “surface temperature” series with the University of Alabama Huntsville (UAH) global “lower troposphere” series. This comparison is the one we’re used to seeing and indeed about the only one we ever see. It’s the basis of the dispute over which year was really the warmest on record, whether the Earth has warmed by 0.59°C since 1979, as indicated by the HadCRUT4 trend line, or by 0.41°C as indicated by the UAH trend line, and over the length of the post-1998 warming “pause” or “hiatus”, although I’m not going to get into that here:

Figure 1: HadCRUT4 “global surface” versus UAH “lower troposphere” series since the beginning of the satellite record in 1979.

This comparison, however, ignores a number of complications.

A brief note on data sets, sources and treatment before proceeding.  Instead of comparing surface and satellite data globally this post compares them with the global data segregated into “land” and “ocean” components. The four temperature time series considered are:

CRUTEM4 (surface land)

HadSST3 (surface 0cean)

UAH lower troposphere v6.0 land (lower troposphere land)

UAH lower troposphere v6.0 ocean (lower troposphere ocean)

The data for these four series are from KNMI Climate Explorer, as are the “land” and “sea” masks used to segregate the UAH global series into “land” and “ocean” components. The data for the Remote Sensing Systems (RSS) v 3.2 and the “raw” ICOADS series are also from KNMI. The HadAT and Ratpac radiosonde data sets discussed later in the text are referenced individually. All data are annual (Jan-Dec) means with the average of the 1979-1983 means set to zero so that all plots start at about the same level.

We begin with HadCRUT4. HadCRUT4 supposedly measures the Earth’s “surface temperature”, but the two surfaces at which it measures temperatures are quite different. About 30% of HadCRUT4 comes from surface air temperatures measured a nominal 5ft above the land surface (CRUTEM4) and the remaining 70% from sea water temperatures measured anywhere from a foot to maybe 50 ft below the sea surface (HadSST3), and as one might expect temperature trends in these two contrasting environments are quite different (Figure 2). Trend line gradients in fact show the global land warming by over twice as much as the global oceans (0.97°C vs 0.44°C since 1979):

Figure 2: CRUTEM4 “land” vs. HaSST3 “ocean”

Yet still we average CRUTEM4 and HadSST3 together to estimate “global surface temperatures” and assume that the result tells us how much the Earth’s “surface” has warmed. But the surface we define by averaging them doesn’t exist as a physically-identifiable entity. It consists of 70% sea water and 30% air – presumably in the form of bubbles – a medium which is found in limited quantities only on surf beaches and during storms at sea. (This problem wouldn’t exist if we had a data set providing surface air temperatures five feet above the sea surface that we could use instead of HadSST3, but we don’t).

The UAH satellite data, on the other hand, are measured at a continuous, coherent and physically-definable surface and we might therefore expect that they would show less variation. But they don’t. The lower troposphere also shows over twice as much warming over land as it does over the ocean (0.63°C vs. 0.27°C since 1979. According to Dr. Roy Spencer 17% of the UAH signal over land is a result of thermal emission from the land surface, but this causes only a “slight enhancement” of the land vs. ocean trend difference):

Figure 3: UAH “land” vs. UAH “ocean”. The series are segregated by applying the KNMI “land” and “sea” masks

These results show that the big difference in post-1979 warming isn’t between the surface and the lower troposphere at all; it’s between the land and the ocean. As always there are some wrinkles, but the bottom line is that a blanket comparison of global UAH with HadCRUT4 is not a good approach, as I noted to begin with.

And in passing I will briefly touch on one intriguing wrinkle. How long has the land been warming faster than the ocean? Obviously it can’t have been doing so indefinitely. Culled from the depths of my files is Figure 4, which compares the ICOADS ocean SST series with the GISS “meteorological station only” land surface air temperature series since 1880 and plots the difference between the two. The difference plot can be fitted closely with a sine curve with a period of 110 years and an amplitude of 0.63°C. Speculation is invited:

Figure 4: Surface air temperature (SAT) versus sea surface temperature (SST) since 1880. The SAT data are from GISS and the SST data from ICOADS.

Having got that out of the way we can now move on to the main part of the post. How reliable are the HadSST3, CRUTEM4 and UAH series? The best way of finding out is to take the raw data and reconstruct the series completely from scratch, but this is usually beyond the capacity of common mortals, although there is one case where I have in fact done it. The fallback position is to compare the three series with other independently-constructed versions to see how well they match. If they match there is at least a presumption that the series is recording real temperature variations.

We begin with HadSST3:

HadSST3 is arguably the most important of the data sets because it contributes 70% of the value of HadCRUT4. What do we have to compare it with? There are a number of other SST series that show similar results, such as ERSSTv4 and Reynolds OI v2, but the most diagnostic comparison is with the “raw” ICOADS SST data from which these series are constructed (although ICOADS is not exactly “raw” because reducing many millions of point-source SST readings to a common baseline involves a fair amount of statistical manipulation). And when we compare HadSST3 with ICOADS over the period of satellite record since 1979 we see no significant difference:

Figure 5: Raw ICOADS SST data vs. HadSST3

We can conclude from this that HadSST3 has not been adjusted to any significant extent over the period after 1979, although the same can’t be said for the period before 1979. In summary, HadSST3 is probably OK.

Next comes CRUTEM4:

It’s frequently claimed that CRUTEM4 is correct because its “sister” series, such as NCDC land, GISS and BEST, show similar amounts of warming. But all these series apply “homogenization” algorithms to the raw data, and numerous previous analyses, including several published here on Energy Matters, have shown that these algorithms sometimes have a regrettable tendency to manufacture warming where no warming exists. So the fact that CRUTEM4 compares with NCDC, BEST and GISS does not confirm that CRUTEM4 reflects real temperature trends.

What we need is an independently-constructed surface temperature series that uses the same raw data set to compare CRUTEM4 against, and some years ago I constructed one. It uses unadjusted GHCN v2 data from 801 very-carefully-selected surface temperature stations, all of which are guaranteed free of significant UHI gradients. I haven’t updated it since 2010 and probably never will because of the amount of work involved, but 2010 gets us most of the way. Figure 6 compares CRUTEM4 with two versions of my series:

Figure 6: CRUTEM4 vs. two surface air temperature series independently constructed from unadjusted GHCN v2 data by the writer

Why two versions? Because I’m unable to replicate the CRUTEM4 “land mask” exactly and these are limiting cases. The area-weighted series projects surface temperatures out over the ocean – i.e. into HadSST3 territory – and will therefore tend to underestimate land warming, while the station-weighted series will tend to overestimate it because of the concentration of stations in areas that have warmed more than the global average, particularly in mid-high northern latitudes. Both series, however, show less warming than CRUTEM3 – the area-weighted series about 0.1°C and the station-weighted series about 0.3°C projected out to 2015 – suggesting that CRUTEM4 may have overestimated land surface warming since 1979 by about 0.2°C.

But has it? Impossible to say. And what difference does it make if it has? Not that much. Subtracting 0.2°C reduces post-1979 CRUTEM4 warming from 0.94°C to around 0.75°C, still comfortably in excess of lower troposphere warming over land, and it lowers post-1979 HadCRUT3 “global” warming from 0.59°C to about 0.53°C, still considerably more than global lower troposphere warming and certainly nowhere near enough to shut down the IPCC.

My conclusion on CRUTEM4 is that it leaves a lot to be desired, but it isn’t demonstrably wrong to the point where we can reject it out of hand.

Last comes UAH:

Here I present two lines of evidence which demonstrate that the UAH lower troposphere series is robust, with the first being that Remote Sensing Systems (RSS) independently analyzes the same raw data and comes up with an almost identical result:

Figure 7: UAH vs. RSS lower troposphere series, global

The second is RSS’s comparison of raw and corrected RSS and UAH troposphere temperatures with the UKMO HadAT radiosonde data (available in text format here) shown in RSS’s Figure 4 and reproduced below as Figure 8. The HadAT radiosonde data show a warming trend of 0.189°C/decade compared to 0.181°C/decade for RSS and 0.175°C/decade for UAH. These numbers are essentially the same within limits of measurement error:

Figure 8: UKMO HadAT radiosonde data vs. raw and “sampled” versions of RSS and UAH lower troposphere temperatures. The HadAT plot is a weighted average of the HadAT data at different millibar levels to make it directly comparable with the coverage of the RSS/UAH microwave sounding units .

I could add a third line of evidence by comparing UAH/RSS with another radiosonde data set – NOAA Ratpac (data in text format here) which Euan Mearns discussed in detail in his recent eponymous post. Unfortunately Ratpac can’t be compared directly with UAH or RSS TLT because Ratpac data are given at specific millibar levels while UAH and RSS TLT are weight-averages over a range of millibar levels (as noted in the caption the HadAT plot shown in Figure 8 is a weighted average of different tropospheric levels). We can, however, compare it with HadAT at the 700mb level, which is in the most heavily-weighted portion of the TLT window. The two give effectively the same result:

Figure 9: HadAT vs. Ratpac radiosonde data, 700mb level.

In summary , the global UAH series is replicated by the RSS series, gives substantially the same results as the independently-derived HadAT radiosonde data set and HadAT also compares closely with the Ratpac radiosonde data set. It would be difficult to do much better. We can reasonably accept that the series is correct to within normally-accepted limits of error and go from there.

And where exactly do we go? Well, if the world insists on measuring the progress, or lack thereof, of “global” warming it needs a robust, coherent and global temperature data set to do it, and the only one it has is UAH (or RSS). UAH doesn’t measure temperatures at the surface, where global warming should ideally be measured, and it goes back only to 1979, but offsetting this defect is the fact that the surface data sets, in particular HadSST3, become progressively more corrupted by “adjustments” before 1979 to the point where their reliability before 1950/60 is questionable. And as shown in Figure 10 the Ratpac and HadAT2 radiosonde data might be good enough to allow us to project the series back to 1958 anyway (The 700mb data are plotted illustration purposes. “lower troposphere” temperatures would probably show less overall warming):

Figure 10: HadAT vs. Ratpac radiosonde data, 700mb level, all data since 1958.

But this of course isn’t going to happen. The world will continue to use the HadCRUT4 “surface temperature’ series as its global warming yardstick basically because it shows more global warming than the lower troposphere. But is this because HadCRUT4 overestimates surface warming or because the surface really has warmed more than the lower troposphere? I’ll leave that question open for discussion.

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

20 Responses to Surface versus satellite; the temperature data set controversy

  1. It might be instructive to compare the UAH global lower troposphere series with the series of global surface air temperature T2m (2m above the ground) of the ERA-Interim Reanalysis, which also can be obtained from KNMI (via the menu item Monthly reananalysis fields). Reanalysis is a scientific attempt to develope a comprehensive record of how weather and climate are changing over time, which makes use not only of synoptic observations at climate stations, but also of radiosonde, satellite, buoy, aircraft and ship reports. Observations over land are blended with values from the background forecast model over sea. A plot of UAH Global Land&Ocean TLT 90S to 90N versus ERAI T2m anomalies (with respect to 1981-2010 base period) trailing 12 month average shows a good concordance of both series until 1998, in 1998 UAH displays much more warming than ERAI, less significant also in 2010, in 2006 and 2007 ERAI displays more warming than UAH, in 2012 UAH displays a considerable cooling, but ERAI does not. Since 2013 both series seem to run parallel, but ERAI moves at a higher level. I would have liked to upload the plot I made.

  2. Flocard says:

    I would like to have the view point of experts.
    I have been told also that satellites data is also affected from a systematic error as time progresses.
    They measure integrated data over the entire width of atmosphere and use an algorithm to deduce the temperature in the first 5000 m. The temperature which is being discussed here.
    On the other hand as time goes, the altitude of the satellite decreases so that the algorithm has to be rescaled.
    In a sense there is also a global man-made intervention in a set of otherwise objective data.
    Is anybody knowledgeable on that point.

    • Euan Mearns says:

      Hubert, its true that the satellite data are subject to much processing and corrections. See here:

      And so there is no perfect solution or answer. UAH and RSS both give same result now processing the same data. Had, GISS, NOAA and BEST give same result for thermometers processing more or less the same data. The agreement between thermometers and satellites is actually remarkably good, apart from they seem to be drifting apart in recent years. HadCRUT3 actually matched the satellites pretty well until they “upgraded” to HadCRUT4.

      Watch the Vid where Roy Spencer and Gavin Schmidt are interviewed and decide for yourself which one is the more credible scientist.

      • A C Osborn says:

        Euan, you say “Had, GISS, NOAA and BEST give same result for thermometers processing more or less the same data. ”
        But that was not always true.
        Each group has made adjustments to the data, mortly NCDC & GISS and then the other groups have followed when questions have been asked about the discrepency that the adjustments have produced.
        The Australian BOM and New Zealand Met Office have also followed suit.
        The big Elephant in the room is that those Adjustements mean the new “data” no longer matches the original data that was completely backed up by anecdotal evidence from the time they were measured.
        So on one hand we have the Raw Data and matching Anecdotal Data Vs data completely made up by Climate Scientists based on what they think the Temperatures were supposed to be.

        I know which I believe.

      • sod says:

        The debate is on FOX. I actually think that Gavin is handling the loaded questions rather well.

    • Richard Mallett says:

      John Christy of UAH addressed these objections at – if you call this ‘global man made intervention’ then that applies to all data sets that attempt to derive a global temperature from thousands of measurements from individual stations / buoys / balloon ascents / etc.

      The main disadvantage of the satellite data series is that they start in 1979. If we want to consider the effect of solar cycles, oceanic cycles, etc. on global temperatures (giving us warming / cooling / stasis periods) then we need at least 120 years of data – 37 years is nowhere near long enough.

  3. JerryC says:

    I guess my main questions are A) How accurate are any of these global temperature measurements, really? and B) Even if one accepts that global temperature measurements are accurate to a tenth of a degree, how much temperature stability is it reasonable to expect from a “pristine” earth? All of these measurement sets are being tracked within the confines of a single degree Centigrade, which is arguably pretty darn stable.

    Put another way, to what extent is the scientific establishment spending billions of dollars to measure noise rather than signal?

    • 1C is significant warming. And the warming is continuing.

      Regarding adjustments and what to expect from a pristine earth, I was reminded of this study(PDF) which compares a network of pristine US measurement sites (over the last 10 years or so) with a matched subset of the normal NOAA adjusted data (so this isn’t a measurement of US average surface temperatures, just a subset). This comparison provides validation of the algorithms used to adjusted raw data. So I think we can be confident that the adjustments made to raw data give reasonably accurate results.

      • Roger Andrews says:

        “1C is significant warming”.

        Once again this depends on how one defines “significant”, but 1C of warming represents an increase of only about 0.3% in global mean “surface” temperature relative to absolute zero (from ~288K to ~289K.)

        And if you live in mid-temperature latitudes you can offset 1C of warming by moving about 100 miles north or about 300 feet higher up the hill.

        Not very much at all, really.

        • Relative to absolute zero? Hmm, OK.

          As regards moving up or polewards, yes, that would do it, for temperature, though moving up limits how much more adaptation a species can perform and doesn’t do anything for daylight hours (for instance). Sadly, not all species seem to be adept at moving so quickly. One would have to move further on land, than in sea, but if 1C is of no concern to you, then you’re all set.

          • JerryC says:

            You do realize that every species that is around today has alrewdy survived at least several glacial-interglacial transitions, yes? The world didn’t begin in the mid-19th century when the temperature records started.

  4. sod says:

    In his most recent temperature update (January, whcih was incredibly hot) Roy Spencer from UAH explains how they make changes:

    “We had been concerned that the LT temperature trends over land were too warm compared to the ocean. One hint that something might be wrong was that the trends over very high elevation portions of the Greenland ice sheet and the Himalayas were much colder than the surrounding regions (see Fig. 4 here). Another was discontinuities in the trend patterns between land and ocean, especially in the tropics.

    We determined this is most likely due to a residual mismatch between the MSU channel 2 weighting function altitude on the early satellites versus the AMSU channel 5 weighting function altitude on the later satellites. We already knew AMSU5 peaks lower than MSU2, and had chosen Earth incidence angles in each to get a match based upon theory. But apparently the theory has some error, which we find equates to about 150 meters in altitude. This was enough to cause the issues we see….land too warm at low elevations, too cold for elevated ice surfaces.

    We therefore changed the AMSU5 reference Earth incidence angle (from 35.0 to 38.3 deg.) so that the trends over Greenland and the Himalayas were in much better agreement with the surrounding areas. We also find that the resulting LT trends over the U.S. and Australia are in better agreement with other sources of data.”

    So the agreement with other sources is not confirming how good the UAH data is, it is instead how the UAH data is made.

    I have 3 big problems with the UAH Data:

    1. the time range is too short. Basically a restriction to UAH data means, that we know next to nothing about global temperature.

    2. Those big el nino peaks are completely messing up the data. It is obviously much more sensitive to random events than other datasets are.

    3. we do not live in the lower troposphere.

    • Javier says:

      While we do not live in the lower troposphere, the CO2 hypothesis rests on layer of outgoing IR emission being higher and layer of incoming IR emission being lower with increased CO2. Necessarily that has to produce a higher warming in the lower troposphere specially over the tropical area. That lower troposphere higher warming that the theory predicts has not been observed.

      The El Niño warming and La Niña cooling are direct ocean-atmosphere heat exchanges that only secondarily warm land surface. That satellites show higher low troposphere warming during El Niño than surface (and opposite during La Niña) is proof that they work correctly. This did not affect warming rates during 1979-2003 when RSS had the same warming rate as Gistemp, and should not affect warming rates during 2003-now when their warming rates have diverged.

      And third, being too short only means that we have something to trust from 1979, not the opposite.

      Those reasons should therefore not be of concern.

  5. Roger Andrews says:

    The fact that the satellite record doesn’t start until 1979 isn’t the kiss of death. According to the IPCC man-made global warming didn’t begin until the second half of the 20th century, and according to Figure 10 it began abruptly in 1976. The satellite record therefore covers almost all of the “AGW” period and can therefore be used to make estimates of “post-AGW” warming rates. What happened before 1976 is of course of interest but doesn’t impact these estimates.

  6. Javier says:


    Thank you very much to you and Euan for these posts. I have three comments:

    1. I would like to see the UAH land-CRUTEM4 and UAH Sea-HadSST3 comparisons to see how much difference is coming from each.

    2. Regarding the difference in warming rate between Land and Sea, I have taken the liberty to overlay your graph with a warming rate graph from UK Met Office distributed by BBC some time ago. It shows essentially the same. There appears to be a cycle of about 110-120 years period that is not related to the important 60 years cycle in temperatures:

    We can only speculate about the nature of such cycle. It seems to accumulate heat in the oceans for about 60 years (1910-1970) and then release it to the atmosphere (1970-present). My guess is that it may be linked to El Niño frequency, a phenomenon that does exactly that, it accumulates energy in the oceans during La Niña and releases it to the atmosphere during El Niño. My longest MEI chart goes only to 1950, but it shows that from 1950-1976 the world was dominated by La Niña and since then (40 years) it has been dominated by El Niño, so it is congruent with that hypothesis. However the extended MEI is not very supportive and I don’t know how good it is in reproducing actual ENSO conditions of the past.

    3. I would like to call your attention over another feature of temperature differences between satellites and surface. It seems that both datasets agree very well between 55°N and 55°S, and most of the warming differences are coming from polar areas, specially the Arctic region. This was shown by Bob Tisdale at his post:
    The Impact of GISS Replacing Sea Surface Temperature Data With Land Surface Temperature Data
    And by Okulaer in his post:
    Why “GISTEMP LOTI global mean” is wrong and “HadCRUt3 gl” is right

    So it seems that the temperature battle is going to be fought over the polar temperatures where nobody is measuring in a significant way.

    • Thank you Javier. To respond to your comments:

      1. I would like to see the UAH land-CRUTEM4 and UAH Sea-HadSST3 comparisons to see how much difference is coming from each.

      2. Regarding the difference in warming rate between Land and Sea, I have taken the liberty to overlay your graph with a warming rate graph from UK Met Office distributed by BBC some time ago. It shows essentially the same. There appears to be a cycle of about 110-120 years period that is not related to the important 60 years cycle in temperatures:

      Here’s the graph:

      The 110-year cycle doesn’t seem to relate to any recognized solar or oceanic cycle, but it’s visible on most land/ocean temperature comparisons and I don’t think its an artifact of the data. It’s just another one of those things that tell us how complex the Earth’s climate is and how little we really understand about it. I do, however, tend to agree that the PDO “phase change” in 1976 is somehow related to it and that releases of heat from the deeper levels of the ocean since then could explain at least some of the recent warming. It’s certainly difficult to explain the abrupt shift from cooling to warming in 1976 entirely as a result of anthropogenic forcing.

      Your comment 3: I really haven’t looked into Tisdale’s results in any detail but it wouldn’t surprise me if the differences were coming from the Arctic, which because temperature swings there are much larger than anywhere else contributes disproportionately to the swings seen in the global record. The Okulaer post I haven’t looked at at all.

Comments are closed.