What’s Really Wrong With the Global Surface Temperature Record

Recently there has been much discussion as to whether the homogeneity adjustments applied to raw surface air temperature records by GISS, NCDC, CRU and BEST might not have manufactured a lot of the global warming allegedly caused by man-made greenhouse gases. Here I look briefly into this question, but more deeply into into the question of whether the published “surface temperature” time series that are presently used to evaluate global warming, such as HadCRUT4, GISS LOTI and NCDC land & ocean, are fit for purpose. And without further ado here are the conclusions I have reached based on an analysis of HadCRUT4, the most commonly-used of the published “surface temperature” series:

      1. The homogeneity adjustments applied to the raw surface air temperature records are suspect but have little impact on HadCRUT4.
      2. The bias adjustments applied to the raw sea surface temperature records are equally suspect and have a larger impact. They don’t add warming (they actually apply a net cooling adjustment between 1880 and the present) but they significantly change the shape of HadCRUT4.
      3. HadCRUT4 combines surface air temperatures and sea surface temperatures that show quite different trends into an apples-and-oranges average that does not provide meaningful results. Consequently HadCRUT4, along with its sister GISS, NCDC and BEST combined land and ocean “surface temperature” series, must be deemed unfit for purpose, particularly when the purpose is to quantify global warming and to guide the world’s multi-trillion-dollar efforts to combat it.
      4. When surface air and sea surface temperatures are considered separately – which is the way they should be evaluated – we find that climate models do a generally good job of hindcasting surface air temperatures but a conspicuously poor job of hindcasting sea surface temperatures.

How HadCRUT4 is constructed

I start with a brief discussion of this subject because it’s important to an understanding of what comes later.

HadCRUT4 is a hybrid series constructed by area-weighting the CRUTEM4 “land” surface air temperature series and the HadSST3 “ocean” sea surface temperature series. (It exists because it’s believed we need a global series to evaluate global warming and the only way we can construct one is by combining land and ocean temperatures.) But because the oceans cover 70% of the Earth and the land only 30% HadSST3 contributes over twice as much to HadCRUT4 as CRUTEM4. Figure 1 plots the three series together. Note how HadCRUT4 follows HadSST3 more closely than it follows CRUTEM4:

Figure 1: The published versions of CRUTEM4, HadSST3 and HadCRUT4

The question of why HadCRUT4 does not yield meaningful results is discussed later; for the moment we will confine ourselves to quantifying the impacts that adjustments to the raw records have had on HadCRUT4. I did this by recalculating HadCRUT4 using unadjusted surface air temperature and SST series in place of CRUTEM4 and HadSST3.

Impacts of the homogeneity adjustments applied to CRUTEM4.

CRUTEM4 is one of the less aggressively adjusted land series, and in “Homogenizing the World” I estimated that the adjustments applied to it added only 0.11C of warming between 1895 and 2000. I therefore simulated the raw surface air temperatures CRUTEM4 was based on by subtracting 0.11C of straight-line warming from CRUTEM4 between 1895 and 2000 (0.125C when projected to 2014) and recalculated HadCRUT4 using these simulated raw data. As would be expected the impacts are minimal:

Figure 2: Impact of recalculating HadCRUT4 using raw surface air temperature data instead of CRUTEM4

So while homogeneity adjustments remain an issue in so far as they affect the credibility of the scientific institutions that apply them, they don’t significantly change current estimates of global surface warming.

Impacts of the bias adjustments applied to HadSST3

The closest we can get to raw SST data is the ICOADS SST series, which is already manipulated (constructing a baseline-adjusted global series from hundreds of millions of point SST readings is not a straightforward exercise) but there’s not much we can do about that. And when we recalculate HadCRUT4 using ICOADS instead of HadSST3 we see some significant changes. First the three series plotted together:

Figure 3: HadCRUT4 recalculated using CRUTEM4 and the raw ICOADS SST series instead of HadSST3

Second, the published version of HadCRUT4 versus HadCRUT4 recalculated using the raw ICOADS data instead of HadSST3.

Figure 4: Impact of recalculating HadCRUT4 using raw ICOADS SST series instead of HadSST3

The bias adjustments applied to ICOADS result in a slight overall cooling adjustment but the main impact is on the shape of HadCRUT4, which changes appreciably.

We will now take a closer look at these bias adjustments, which are applied ostensibly to remove biases thought to have caused by changes in SST measurement methods (from insulated buckets to uninsulated buckets, from buckets to intakes etc) over time. I don’t have the space to discuss how these adjustments were arrived at in this post, but anyone requiring more detailed information will find a full account in this lengthy tome I wrote a few years ago.

Figure 5 plots the adjusted HadSST3 series, the raw ICOADS series and the difference between them, which defines the bias adjustments applied to ICOADS:

Figure 5: Bias adjustments applied to raw ICOADS series to construct HadSST3

An argument can be made that if bias adjustments this large are necessary to make the ICOADS SST record “correct” then it was too heavily distorted to have been used in the first place. But if such adjustments are to be applied to a temperature series as critically important as HadSST3 an ironclad justification is needed for doing it. So what is the justification?

The accepted way to correct for measurement biases is to segregate the measurements into categories based on how they were taken – buckets, intakes, sensors, whatever – then to quantify the bias levels between the categories and finally to adjust them to match each other. Historic records, however, do not provide  enough information on how SST measurements were taken to allow biases to be quantified this way. So the adjustments are based on the assumption that once all the biases are removed SSTs and air temperatures over the oceans will show the same trends, or if you like that one is a proxy for the other.

And once this assumption is made the adjustment process is greatly simplified. All that’s necessary is to adjust the SSTs to match air temperatures – usually night marine air temperatures measured on board ships – and biases in the SST record are automatically removed.

The question, however, is whether the assumption is valid. In my literature searches I was unable to find a single study which concluded “air temperature trends over the ocean and SST trends are the same and here’s the proof”. I did, however, find a lot of evidence to  suggest they aren’t the same. A few examples are shown in the following Figures:

Figure 6 (Figure 17 of the lengthy tome) compares the ICOADS SST series with an ocean air series I constructed using 174 GHCN v2 records from stations on islands in the oceans plus a few on exposed coastlines (note that the ICOADS plot omits the distorted SST data during World War 2). According to this comparison ocean air temperatures and SSTs do not move together. Instead they oscillate around each other, with SSTs cooling by about 0.5C relative to air temperatures between 1890 and 1910, warming by about a degree C relative to air temperatures between 1910 and 1970 and cooling by about 0.5C relative to air temperatures after 1970. The oscillation can be fitted with a sine curve with a period of 120 years and an amplitude of 1.1C:

Figure 6: Author’s ocean air temperature series versus raw ICOADS SST series

Figure 7 (old Figure 18) replaces my ocean air series with the GISS “meteorological station only” global surface air temperature series. The result is the same:

Figure 7: GISS “meteorological station only” air temperature series versus raw ICOADS SST series

CRUTEM3 and HadSST3 show land air temperatures warming by about a degree C relative to ocean SSTs since 1880, but the oscillatory pattern is still there despite the adjustments applied to these series:

Figure 8: CRUTEM4 versus HadSST3

Even climate models agree that air temperature and SST trends are different. Figure 9 plots the IPCC’s “worst case” RCP8.5 temperatures as projected by the CMIP5 climate models the IPCC used in the AR5. The models show air temperatures over the oceans (I applied a global sea mask) warming by a degree C relative to SSTs between 1880 and 2100.

Figure 9: CMIP5 climate models, air temperatures over oceans versus SSTs, RCP8.5 scenario

The conclusion must be that adjusting the raw SST data to match air temperatures does not remove SST measurement method biases. It distorts the SST record instead.

Figure 9 conveniently leads in to the next and key question, which is:

Is HadCRUT4 meaningful?

Let’s make another unlikely assumption – that the CMIP5 models shown in Figure 9 correctly predict what temperatures will do between now and 2100. In this case the red and blue lines in Figure 10 show what CRUTEM4 and HadSST3 will look like and the black line shows what HadCRUT4 will look like in 2100 assuming 30% land and 70% ocean. CRUTEM4 shows an air temperature increase of 6.4C in land areas since 1880, HadSST3 shows SSTs increasing by only half as much and HadCRUT4 shows an increase of 0.3 times 6.4C plus 0.7 times 3.2C = 4.2C in “global surface temperature”.

Figure 10: What CRUTEM4, HadSST3 and HadCRUT4 will look like in 2100 if they accurately predict 21st century temperatures

But exactly where is this “global surface” at which HadCRUT4 measures temperatures?

It isn’t anywhere. It doesn’t exist as a physically-definable entity. HadCRUT4 mixes air temperatures measured a nominal five feet above the ground with SSTs measured anywhere between a foot and fifty feet below the surface of the ocean, meaning that the average HadCRUT4 measurement will have been taken maybe a foot below the surface in a medium consisting of 30% air and 70% sea water. Climate models can’t simulate temperatures in this non-existent environment, so HadCRUT4-compatible model output has to be generated by averaging air temperature and SST model means in the same way as observational means are averaged to generate HadCRUT4.

And where on the Earth’s surface in the year 2100 will we have experienced the 4.2C of “surface warming” shown by HadCRUT4? Not on the land, nor in the oceans. We won’t have experienced it anywhere.

Add to this the fact that HadCRUT4 combines two series with different temperature trends and the conclusion is inescapable. HadCRUT4 is not meaningful. We should not be using it. If we want to find out what’s really going on with the Earth’s climate we must consider surface air temperatures and SSTs separately.

Which brings up the final question. How might this change our perspective on global warming?

Well, it won’t make global warming go away, but it will make a difference, and potentially a large one, to climate model/observed temperature comparisons. Climate model performance is presently checked by comparing hybrid series like HadCRUT4 with HadCRUT4-compatible temperatures constructed by combining air and sea temperature model output, and these comparisons show a reasonably good global match. Figure 11, which reproduces Figure TS-7a of the IPCC AR5, is an example:

Figure 11: IPCC AR5 model-versus-observed temperature comparisons. The observational series are HadCRUT4, GISS LOTI, NCDC land and ocean and Japan Meteorological Agency

Yet rarely do we see a global comparison of modeled land surface air temperatures with the CRUTEM4 land series, which is surprising because except for the recent warming “pause” the fit is quite good:

Figure 12: CMIP5 climate model land surface air temperature means versus CRUTEM4

And I don’t think I’ve ever seen a comparison of modeled SSTs against HadSST3. Maybe this is why:

Figure 13: CMIP5 climate model SST versus HadSST3

And Figure 14 shows what may be the only comparison you will ever see of modeled SSTs versus the raw ICOADS SST series:

Figure 14: CMIP5 climate model SST means versus ICOADS SST series


One key question remains. It’s not really within the scope of this post but I will touch on it anyway. Is there any chance the raw ICOADS series shown in Figure 14 is representative of actual SST temperature trends? If so global warming becomes a different ball game. One line of evidence which suggests it might be is that the only excursion in the series that can be conclusively identified as a bias effect is the large “spike” in SSTs during World War 2, and as discussed in the lengthy tome linked to earlier this spike wasn’t caused by the SST measurement method changes that are alleged to have caused it. It was caused by World War 2.

Another is the global relative sea level rise series I put together from 382 tide gauge records a few years ago. After looking at it for a while it struck me that I’d seen a series very like it somewhere else. Then it occurred to me which series it was:

Figure 15: Author’s relative sea level rise series versus ICOADS SST series. Note that case the WWII “spike” in ICOADS has been replaced with a straight line.

A little too close to be dismissed as coincidence, I think.

Data sources:

The published series used in the post are available from the following sources.

CRUTEM4, HadSST3 and HadCRUT4 from: http://www.cru.uea.ac.uk/cru/data/temperature/#datter

ICOADS from: http://research.jisao.washington.edu/data_sets/global_sstanomts/

The GISS meteorological station only series from: http://data.giss.nasa.gov/gistemp/

The CMIP5 climate model data, along with most of the series listed above, from KNMI Climate Explorer: http://climexp.knmi.nl/selectfield_obs2.cgi?someone@somewhere

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

21 Responses to What’s Really Wrong With the Global Surface Temperature Record

  1. rogue says:

    Roger, thank you for this article. On your final point about ICOADS, are you in touch with Nic Lewis about how this might affect energy balance models?

  2. Euan Mearns says:

    Roger, thanks for this informative post. I agree with your comments on Figure 5. If you have to make adjustments this large then you can’t use the data. It smacks of fiddling. And Figure 15 looks too good to be true. Being a cynical skeptic you will have to convince me. Worth a post on its own detailing how your sea level construction may differ from the official version and hwo you may be justified chopping out the WWII data.

    I know the warmists pride themselves in saying they have warmed the past SST record. They say “we warmed this” and “we cooled that” as if they were Gods. Lost site all together of the fact that they have data and should be developing a theory to fit the data. Instead they have a theory and are determined to mash the data to make it fit.

    My guess is the target is to have a Global temperature history that matches Law Dome and Mauna Loa exactly.

    • And Figure 15 looks too good to be true. Being a cynical skeptic you will have to convince me.

      All I can say is:

      1. There was no prior collusion

      2. Thermal expansion is a major contributor to sea level rise.

      they have a theory and are determined to mash the data to make it fit.

      Not only are they determined to do it, they’ve done it.

      My guess is the target is to have a Global temperature history that matches Law Dome and Mauna Loa exactly.

      If so they’re not doing a very good job of it. The ongoing warming pause in particular could use some more “mashing”.

  3. Jeff says:

    The scale and shape of your sea level graph do not match those here: http://www.cmar.csiro.au/sealevel/

    Can you justify the difference?

    • The CSIRO graph shows absolute (or geocentric, or eustatic) sea level rise. My graph shows sea level rise relative to the coastline. The CSIRO graph contains “corrections” for vertical land movement. Mine contains no corrections.

      • Jeff says:

        Is plotting sea level changes relative to a moving reference point meaningful if the movement of the reference point is not quantified too?

  4. clivebest says:


    You have put a lot of work into this . In fact it was your and Euan’s effort that got me looking again into the whole GST processing. I think there is potentially a boasting problem in the whole methodology adopted for measuring temperature Anomalies.

    The orthodox method is to define station temperature anomalies relative to a 30 year period 1961-1990. These are gridded on a 5×5 degree base and averaged weighted by cos(lat) to form a global monthly and annual average. Since many stations overlap in GHCN and CRU (because GHCN imported all CRU stations) they give comparable results. As you show there are some small trend differences introduced by the data correction and homogenisation process.

    However this method causes stations with poor coverage in the reference period to be dropped. Efforts to avoid that use interpolation or values derived from near neighbours in the same grid. BEST uses an interpolation method from near neighbours. This generates fake data which may or may not affect the final result, but I want to ask a separate question..

    Is the orthodox method correct? Is it the only way to eliminate sampling biases? Does the use of a 30 year period itself bias the result in some way? Is there group thinking happening ? It seems to me perfectly reasonable to question this, especially as this has become such a political hot potato.

    I have been trying different methodologies which use all available station data. and indeed get a warming trend 0.2 deg less than CRUTEM4. I am still looking into any possible biases in the definition of temperature anomalies.

    If by trying another approach we get a slightly different result, it does not mean our result is wrong since there is no ‘correct result’. It just means that the methodology is different unless it can be shown to introduce artificial trends. Perhaps the use of a 30 year normal based on a warming climate itself introduces artificial trends!

    We will see.

    • clivebest says:

      ‘Boasting problem’ should read ‘biaising problem’ The iPad I am using kindly auto-corrected it for me!!

    • Euan Mearns says:

      Clive, what you have been doing deserves a lot of exposure. Its just these climate based posts consume sooooo much time, but I’ll get around to it.

      • clivebest says:

        No problem. The whole area of determining surface global temperatures is complex, obtuse and specialised yet the message given to the public is as if it were simple. It isn’t. First define what you mean by a global surface temperature. It doesn’t exist. The only defined temperature is the ‘effective radiative temperature’ of the earth which is ~ 255K

  5. Stuart says:

    It will be interesting to see the next results of NASA’s OCO-2 mission. As covered here http://euanmearns.com/co2-the-view-from-space/

    The homepage for OCO-2 (see here http://oco.jpl.nasa.gov/) seems to suggest that the next data release will show that thawing Arctic Tundra will prove to be an enormous source of atmospheric CO2.

    The industrial epicentres that are Europe, USA and the Far East appear to be utterly dwarfed by the CO intensity over Canada and Siberia.

    The shear scale of the seasonal disparity between the November map and the current map on the homepage suggests that the CO2 cycle is highly seasonal and closely correlated with photosynthesis rather than the robotically continuous output of human industry.

    • Euan Mearns says:

      Thanks for the heads up on the latest from OCO-2. I’d note that the graphic they present goes nowhere near the Arctic though – but it may well be seasonal thawing. Its going to be an interesting story. All that blue over the Southern Ocean sure looks like a sink.

  6. Roger
    Another very well argued post. You have shown that the HadSST3 does not agree to the climate models pre-1940 (Fig 13), and that the “raw” ICOADS data is even more deviant (Fig 14). Further you argue below Fig 5 the reasons for the adjustments. In particular you highlight the assumptions about how the data moves between land and sea.
    An extension of this point is to look at temperature trends over either land or sea. For the homogenized data see http://data.giss.nasa.gov/gistemp/maps/ , particularly at the 250km smoothing radius.
    Even with this smoothed data, there are huge variations.
    Then look at Euan’s plots for number of stations e.g. for Australia
    Adjusting for the measurement biases of station moves and UHI, along with making the data homogenous to allow infilling of the gaps between temperature stations, is based on information from adjacent stations. I propose that the less the number of stations, the greater the amount of smoothing between different areas is likely to be. I believe that you have already shown this with the BEST data, particularly with Manaus, where pre-1920 data was infilled from the Caribbean.
    The implication, which needs to be investigated, is that the early C20th land warming is likely to have been smoothed to a much greater extent than that post 1976. Therefore the close fit of the early 20th land data to the climate models in Fig 12 could be the result of homogenization methods.

  7. Pingback: AWED Energy & Environmental Newsletter: June 1, 2015 - Master Resource

  8. Pingback: Recent Energy And Environmental News – June 1st 2015 | PA Pundits - International

Comments are closed.