Climate Science Glossary

Term Lookup

Enter a term in the search box to find its definition.


Use the controls in the far right panel to increase or decrease the number of terms automatically displayed (or to completely turn that feature off).

Term Lookup


All IPCC definitions taken from Climate Change 2007: The Physical Science Basis. Working Group I Contribution to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change, Annex I, Glossary, pp. 941-954. Cambridge University Press.

Home Arguments Software Resources Comments The Consensus Project Translations About Support

Bluesky Facebook LinkedIn Mastodon MeWe

Twitter YouTube RSS Posts RSS Comments Email Subscribe

Climate's changed before
It's the sun
It's not bad
There is no consensus
It's cooling
Models are unreliable
Temp record is unreliable
Animals and plants can adapt
It hasn't warmed since 1998
Antarctica is gaining ice
View All Arguments...

New? Register here
Forgot your password?

Latest Posts


GISTEMP: Cool or Uncool?

Posted on 19 April 2012 by Kevin C

There are three main versions of the instrumental temperature record, HadCRUT3 from the UK meteorological office, GISTEMP from NASA, and the NCDC dataset from NOAA. Of the three, HadCRUT3 shows the least warming over the last 15 years, and GISTEMP shows the most. The difference is quite striking:

Dataset1997-2012 trend
HadCRUT3v 0.013 ±0.142 °C/decade
NCDC 0.049 ±0.132 °C/decade
GISTEMP 0.103 ±0.143 °C/decade

Given the short-term cooling influences which have been operating over the last decade, the GISTEMP trend is much as expected. But the HadCRUT3 and NCDC trends are much lower.

In the previous article we examined the problem of sampling a stratified data set, and how this impacts the HadCRUT3 temperature record. We saw that the land temperature anomalies are both higher and have been increasing faster than ocean temperature anomalies. However land temperatures are under-represented in the HadCRUT3 data over the last decade, and the proportion of land temperatures has been declining. These effects both contribute to an increasing cool bias in the HadCRUT3 data.

Is there another source of bias which might explain the divergence of the three datasets? Studies from the ECMWF and GISS have indentified one such source in the HadCRUT3 data: Poor coverage at high latitudes. Can we find any evidence which might confirm this?

Global coverage of the surface temperature records

Coverage maps for the three datasets, along with values for the proportion of the land and ocean area of the globe, are shown for the month of December 2010 in the following figure:

Coverage of the 3 temperature records for Dec 2010

HadCRUT3 suffers from poor land coverage of the continental interiors apart from Europe and the USA - this effect was identified in the previous article. But both HadCRUT3 and the NCDC data omit the Arctic and Antarctic regions. Could this explain the lower trends of these datasets?

Temperature trends by latitude zone

As with the land/ocean bias, a problem arises when our sample contains unrepresentative proportions of different strata from the population, where the strata behave differently from one another. We saw that the temperature data was stratified into land and ocean strata. Is the global temperature data also stratified by latitude? The University of Columbia group provide this figure based on the GISTEMP data: 

GISTEMP zonal temperature trends 

Note that Arctic temperature anomalies are both higher and increasing more rapidly over recent decades than those for the rest of planet. The Northern mid-latitudes are also warmer than the global mean, but show little change since 2000. The Antarctic also shows a recent warming trend.

This is exactly the same problem as we saw in the previous article: The fastest warming regions of the globe being underrepresented in both the HadCRUT3 and NCDC datasets. The problem has only arisen during the last decade because that is the timescale over which Arctic temperature anomalies have been diverging from the rest of the planet.

Estimating the effect of coverage using GISTEMP

The GISTEMP dataset provides gridded global temperature estimates covering almost the entire planet over recent decades: This data allows us to estimate the effect of poor coverage in the other datasets. For each month, the temperature anomaly map can be modified to eliminate all the map cells which are missing from the corresponding month of the less complete data set. Then a global average temperature is calculated using only the remaining cells. The GISTEMP team did precisely this calculation for the HadCRUT3 data in Hansen et al (2010) figure 13 (full version):

GISTEMP, HadCRUT3v, and GISTEMP with reduced coverage

The result is that if you reduce the coverage of GISTEMP to match HadCRUT3, the bulk of the difference between the two datasets disappears. The difference between the results can be explained by the poorer coverage of HadCRUT3.

Comparing the trends

All three temperature records provide gridded datasets, so it is fairly simple to reproduce this calculation; the only minor complication being that GISTEMP uses a different grid. The GISTEMP data was regridded to match the others, and then masked to match the poorer coverage of first the HadCRUT3 data, and then the NCDC data. Temperature series were calculated from the resulting maps. The resulting trends over recent periods in °C/decade are given in the following table:

masked to

Note that when the GISTEMP data is masked to reduce the coverage to match one of the other datasets, the resulting temperature trend is a good match for the trend in the incomplete dataset. The datasets agree over the regions where they share coverage. The differences in the global trends between the three datasets must arise from the regions where the HadCRUT3 and NCDC datasets have no coverage.

But is it right?

The GISTEMP result differs because GISTEMP covers the polar regions missing from the other datasets. But could this additional data be wrong?

The polar coverage of GISTEMP arises mainly from the fact that GISTEMP allows each weather station to contribute to an area of radius 1200km around the station - this distance was determined by examining how temperature changes with distance in regions with good coverage (see Of Averages and Anomalies - Part 1B). In addition, land stations are allowed to provide temperature estimates for ocean cells where no sea surface temperature is available - in practice, this means that coastal stations around the Arctic provide temperature readings for the pole.

Is this approach reasonable? We can compare the temperature maps obtained from the different datasets. The first three panels of the following figure show maps of the changes in the pentadal mean temperature between the years 1996-2000 and 2006-2010. (Cells without 30 months of data in both periods are omitted).

Anomaly maps for the major records and NCEP reanalysis

Again we can see that the three datasets agree well in the regions where all have coverage. The smoothing effect of the 1200km radius in the GISTEMP data is also apparent. In both the HadCRUT3 data and the NCDC data the top edge of the observed region is mostly red, and the few isolated Arctic cells in the HadCRUT3 data show large positive anomalies. That provides tentative support for a larger temperature increase in the Arctic.

There is another source of data which can help with this question. The NCEP/NCAR reanalysis dataset provides an alternative method of determining global temperatures. Instead of just performing a statistical analysis of the weather station and sea surface observations, the reanalysis attempts to construct a complete model of the state of the Earth's atmosphere at any point in time. This is achieved by using a modern weather model to produce a 'nowcast' for any date in the past using not just the weather stations, but also data from weather balloons, satellites and radar. This data is less useful for determining long term trends because the results are influenced by the introduction of new data sources (in particular the satellite data in 1979; e.g. Chen et al, 2008), but is useful for determining a physically consistent geographical picture.

The fourth panel in the previous figure shows the results (which can be recalculated here). The resolution is rather higher than the instrumental records, but shows the same hot Arctic and Antarctic features.


  • The difference in recent trends between GISTEMP and the other datasets arises primarily from the incomplete sampling of the Arctic and Antarctic regions in the HadCRUT3 and NCDC datasets.
  • Most of the difference between the temperature series can be accounted for by this difference in coverage. The datasets agree over the region where they all have coverage. Thus the lower trends of the incomplete HadCRUT3 and NCDC datasets provide no evidence against higher trend of the more complete GISTEMP data.
  • There is some internal evidence to support the 'hot Arctic' of the GISTEMP dataset, particularly from the isolated Arctic stations of the HadCRUT3 dataset.
  • Further evidence is provided by the NCEP/NCAR reanalysis data, which uses a much more diverse range of observations coupled to a modern weather model.

Additional reading: Glenn Tamblyn's series 'Of averages and anomalies' contains more information on the methods used in determining the instrumental temperature records: Parts 1A, 1B, 2A, 2B.  Dana also has a first Look at HadCRUT4.

The data for the original three temperature records along with the two masked GISTEMP records may be downloaded as a spreadsheet file for further study. An obvious experiment would be to calculate correlation coefficients between all the datasets over the past 30 years.

0 0

Printable Version  |  Link to this page


Comments 1 to 7:

  1. On the previous post, Skywatcher pointed out quite rightly that the rectangular projection I have used in the maps can create a misleading impression, because the area is exaggerated as you move away from the equator. This inflates the apparent impact of poor coverage at high latitudes. All the conclusions of this post are based on the global temperature averages, not on the illustrations, which are provided as an attempt to explain what is going on. However there is certainly one point which needs to be demonstrated more clearly, I'll try and clarify. In determining the impact of poor coverage at a given latitude, there are three factor at play: The fraction of the Earth's surface covered by that latitude band, the amount by which temperature anomalies in that latitude band differ from the global mean, and how much of that latitude band is missing. In the case of the Arctic, these pull in different directions. The Arctic covers a relatively smaller area than a corresponding latitude range at the equator, but the temperature anomalies are changing much more quickly and the coverage deficit is much more significant. How do these effects play out? That is shown in the following figure: To generate this figure, I took the GISTEMP gridded maps and reduced the coverage to match HadCRUT3 over a 30 degree latitude band, leaving the rest of the map untouched. I then calculated a temperature series from the gridded map series. I repeated this 6 times to produce 6 temperature series. I then took the difference with the original full-coverage data. The results are shown above. Each line represents the bias added to the total temperature series by the loss of coverage in that latitude zone alone. As you can see, the Arctic is the biggest contributor to the recent cool bias of HadCRUT3, followed by the northern mid latitudes. The Antarctic actually provides a warm bias, but one which has been declining since 1998, thus actually adding to the cooling bias since that date. In fact every zone except the southern tropics has been contributing to the cooling trend in HadCRUT3 over the last 13 years. In summary: When calculating area averaged global temperature anomalies, the GISTEMP data suggests that the Arctic, despite it's comparatively small area, is the largest single contributor to the cool bias due to poor coverage in HadCRUT3. The NCDC data suffers from the same problem at high latitudes, although its mid-latitude coverage is better. The implications for HadCRUT4 are interesting.
    0 0
  2. How's the polar coverage for the satellites that give us the RSS and UAH lower troposphere datasets, compared to NCDC and HadCRUT3? Do they also show relatively hot poles where those records lack coverage?
    0 0
  3. WheelsOC: The UAH dataset description gives coverage as from 85S to 85N, excluding some terrain over 1500m - coverage is poor over the "Tibetian Plateau, Antarctica, Greenland and the narrow spine of the Andes". While TLT temperatures are not directly comparable to surface temperatures, it would certainly be very interesting to see how the maps compare. It would be interesting from a social perspective to publish a HadCRUT3 dataset with a coverage bias correction derived from the UAH and/or RSS data. In fact there is scope for a range of indices of this kind and a suite of online analysis tools which provide a uniform view across all the data in any combination. And I guess (and this is not my native field, so my guesses are wrong more often than not) that the agreement will be much better than the current incomplete data. We could also do a mix-and-match on land and ocean series. This is beginning to sound like real research rather than a part time blog effort though. I guess I should look at funding channels for public engagement with scienceĀ·
    0 0
  4. Kevin C - It would also be interesting to look at distance correlation factors per band in the RSS/UAH data, to see whether the regions not covered could be extrapolated as per the GISS data. And to see what that extrapolation might indicate. My guess would be that both would shift to higher trends if polar regions were considered as well, as opposed to left out as they are now.
    0 0
  5. It might be interesting for each of these three sets of data to see what the temperature change has been over the recording period for all the recording stations in 5 degree bands of latitude. Latitude 0 to 5, 5 to 10, 10 to 15 etc. Then we could compare what each of them comes up with for the same band of latitude.
    0 0
  6. William: For GISTEMP, you can generate it for yourself here. Given that the others agree well over the region where they have common coverage, I see no reason to expect that the results will be different, however the grey areas will be bigger. For my own experiments with the HadCRUT3 data I see latitude bands in which only a single station is present right up to the present, and the results from such bands are essentially noise.
    0 0
  7. UAH satellite data: trends for the poles. North Pole 0.47C/decade South Pole -0.05C/decade (see comments above for caveats)
    0 0

You need to be logged in to post a comment. Login via the left margin or if you're new, register here.

The Consensus Project Website


(free to republish)

© Copyright 2024 John Cook
Home | Translations | About Us | Privacy | Contact Us