the Air Vent

Because the world needs another opinion

Comparison of CRU and GISS Temp Data

Posted by Jeff Id on December 10, 2009

A short analysis by Kenneth Fritsch on differences between land surface temperature datasets. I’m extremely busy tonight and have not been able to verify anything here, but am familiar with the quality of Ken’s work. I’m sure he’ll stop by to answer questions from time to time.

Also, reader Ian has an interesting reply with an email from Ed Cook. I want to discuss more in the future but am way too busy today. It shows the fear of doing what’s right in climatology– rather shockingly. It’s comment #52 here.


Guest post Kenneth Fritch,

For those of you who are looking for a complete refutation of global warming, I will disappoint. My point in this whole exercise is more subtle. I wanted to determine whether we could compare temperature series for regions of the globe and over certain time periods where the temperature difference would show statistically significant differences. As far as we know these temperature series are constructed using much of the same raw and perhaps adjusted data. These series, while used separately, like to advertise the fact that they closely follow one on another. I would think that to show statistically significant differences between sets would place uncertainty on all sets as we do not have an independent and absolute standard for comparison. Even smallish differences would place doubts.

To that end, I compared the data sets described below and measured the normalized differences over several global regions and mainly two time periods. I used CRUTEM3+HadSST2 and GISS 1200 km and GISS 250 km for temperature data sets of land and sea for comparisons from the link:

That link is very useful for extracting zonal temperatures from the globe. I used the globe, and the zones (around the globe longitudinally): 0-20N, 0-20S, 20N-40N, 20S-40S, 40N-60N, 40S-60S, 60N-80N and 60S-80S. Unfortunately when comparing polar regions, the CRU data set did not have sufficient data points to do a reasonable comparison. Why the GISS data sets had more data points (filled in?) I do not know at this time.

The results are listed below and a graph is linked here:

For the time period 1900-2008 for the globe we get the following statistical results:
The trend is from a linear regression of data set differences versus time and is in degrees C per century. The adjusted R^2 has its usual meaning. The probability that the trend could happen by chance is given by p carried only to 3 decimal places. The lower and upper trend values at the 95% CIs are listed.

Trend = 0.07; Adj R^2 = 0.26; p = 0.000; Low95% = 0.05; Up95% = 0.09

Trend = 0.02; Adj R^2 = 0.00; p = 0.265; Low95% = NA; Up95% = NA

There is nothing very different about these data sets except that the CRU-GISS1200 is significantly different – but not by much. However, when the difference time series are viewed as, by example in the linked graph above, an apparent step is noted around the mid 1940s. A trend from 1945-2008 yields some very significant differences for the globe as noted below:

Trend = 0.14; Adj R^2 = 0.34; p = 0.000; Low95% = 0.09; Up95% = 0.19

Trend = 0.17; Adj R^2 = 0.34; p = 0.000; Low95% = 0.11 ; Up95% = 0.23

Using these same methods the results below show the differences for the time period 1900-2008 (except where noted) and 1945-2008 for global zonal regions. Results are shown only for CRU-GISS1200. The differences between CRU-GISS250 while different were of the same magnitude.

0-20N for 1900-2008:
Trend = 0.05; Adj R^2 = 0.11; p = 0.000; Low95% = 0.02; Up95% = 0.08

0-20S for 1900-2008:
Trend = 0.15; Adj R^2 = 0.47; p = 0.000; Low95% = 0.12; Up95% = 0.18

20N-40N for 1900-2008:
Trend = 0.02; Adj R^2 = 0.01; p = 0.192; Low95% = NA; Up95% = NA

20S-40S for 1900-2008:
Trend = 0.17; Adj R^2 = 0.44; p = 0.000; Low95% = 0.13; Up95% = 0.21

40N-60N for 1900-2008:
Trend = 0.01; Adj R^2 = 0.00; p = 0.405; Low95% = NA; Up95% = NA

40S-60S for 1956-2008:
Trend = 0.44; Adj R^2 = 0.31; p = 0.000; Low95% = 0.26; Up95% = 0.62

0-20N for 1945-2008:
Trend = 0.16; Adj R^2 = 0.49; p = 0.000; Low95% = 0.12; Up95% = 0.20

0-20S for 1945-2008:
Trend = 0.20; Adj R^2 = 0.58; p = 0.000; Low95% = 0.16; Up95% = 0.24

20N-40N for 1945-2008:
Trend = 0.08; Adj R^2 = 0.23; p = 0.000; Low95% = 0.05; Up95% = 0.12

20S-40S for 1945-2008:
Trend = 0.43; Adj R^2 = 0.82; p = 0.000; Low95% = 0.37; Up95% = 0.48

40N-60N for 1945-2008:
Trend = -0.09; Adj R^2 = 0.13; p = 0.002; Low95% = -0.15; Up95% = -0.03

60N-80N for 1945-2008:
Trend = -0.24; Adj R^2 = 0.08; p = 0.015; Low95% = -0.43; Up95% = -0.05

15 Responses to “Comparison of CRU and GISS Temp Data”

  1. Kenneth Fritsch said

    A short analysis by Kenneth Fritsch on differences between land temperature datasets.

    Jeff ID, those data sets are land and sea.

  2. Jeff Id said

    Sorry for the typo. I updated.

  3. Jeff Id said


    Do you know anything about the Giss urbanization correction? The last time I read up on it was almost a year ago.

  4. Kenneth Fritsch said

    Jeff ID, GISS uses satellite observed “night lights” to locate and classify rural stations which are then used to adjust “reasonably” nearby more urban ones for UHI effects. GISS previously used a population rating to differentiate rural and more urban stations and I suspect that is what they continue to do for the rest of the world. This US/ROW difference has been a major bone of contention for Steve M and he has noted it many times at CA.

    GHCN for its latest USHCN temperature series version uses breakpoints in the temperature time series to determine non-homogeneities for required adjustment, including those for UHI. The difference between GISS US and USHCN mean temperatures for 1920-2008 is statistically significant and is due primarily to the greater adjustment GISS calculates for UHI.

    The GISS series is primarily derived from the GHCN unadjusted data that it adjusts using its own unique algorithms.

    I want to post the differences series for the various land temperature data sets with an explanation of the CRU adjustments, and then finally compare the lower troposphere UAH and RSS to the other land surface temperature series from 1979 forward.

  5. Kenneth Fritsch said

    I should have been clearer above that the GISS satellite observations are used for adjusting the US temperatures only.

  6. Jeff Id said

    #5 You and another blogger have inspired me to finally start digging into temp records. This should be interesting.

  7. Kenneth Fritsch said

    Using the same KNMI source, linked below and that I used for comparing CRU and GISS land and sea temperatures in my introductory post for this thread, in this post I am doing difference series for land temperature data sets for CRUTEM3, NCDC (GHCN) and GISS 1200km. The comparison/differences were for the zonal regions, again, 0-20N, 0-20S, 20N-40N, 20S-40S, 40N-60N, 40S-60S, 60N-80N and 60S-80S and for the globe.

    I included breakpoints that are calculated from the R program that I copied from Steve M at CA and for which the code is listed below. The results are presented in the links listed below.

    I thought the breakpoint (or change point) was the proper tool for looking at the differences in temperature data sets because that is the tool that GHCN is using to find non-homogeneities in their data. The results show many breakpoints in most difference series. Also the straight line trends that the breakpoints segment the series into are statistically significant. Also in a number of cases the trends in the difference series are large compared to the overall trends calculated from the individual series.

    What really astounds me is that I have seen no literature where these comparisons/analyses are made by the authors of these series. It is almost a strong indication that their interest lies in avoiding looking at differences and instead touting the agreements. These series, as far as we know, use primarily the same raw data (from GHCN) and adjust it using their own algorithms. That being the case we are looking at adjustment differences which when disaggregated into regions and time periods can be large.

    bp = breakpoints(x ~ year,h=5)
    #if bp = 0 do not continue, but use this script:
    plot(x, type=”l”, xlab= “Years”, ylab=”Difference Temperature Anomaly Degrees C”, main= “No Breakpoints for GISS-GHCN for Globe”)
    make.bp=function(x,nbreaks=6) {
    bp = breakpoints(x ~ year,breaks=nbreaks, h=5);
    fac0 = breakfactor(bp)
    fm0 = lm(x ~ year);
    fm1 = update(fm0,x ~ year+fac0*year)
    plot(year,A$x, type=”l”, xlab= “Years”, ylab=” Difference Temperature Anomaly Degrees C “, main= “Breakpoints GISS-GHCN for Globe”)

  8. Kenneth Fritsch said

    The links to my results from the post above are listed below. Sorry about the inconvenience.>>>>

  9. Kenneth Fritsch said

    Another attempt:

    The links to my results from the post above are listed below. Sorry about the inconvenience.

  10. Kenneth Fritsch said

    And one more result link while at the same time attempting to inflate may embarassingly low number of posts in this thread.

  11. Ryan O said

    First thought . . . GISS seems to hold more faithfully to GHCN. I wonder if this is partly due to the fact that CRU uses a smaller subset of GHCN stations? And while I’m not a huge fan of breakpoint analysis, the year 1920 (or thereabouts) sure comes up a lot. It seems to be something with CRU, though, since the CRU/GHCN and CRU/GISS show it, and GISS/GHCN do not.

  12. I have updated the AIS climate data graphing to include CRU station temperature data. You can plot stations as well as compare the CRU to the GHCN data.

  13. Kenneth Fritsch said

    In completion of my intended comparisons, I calculated difference series for the land and sea temperature series UAH, CRU and GISS 1200km using the KNMI repository of climate data that I used above. As before, I calculated breakpoints for the difference series. The results are shown in graphical form in the links below. There are fewer breakpoints in these differences than those with the non-satellite data and that might well be because of the shorter time duration of the comparison (1979-2008). I used the zonal regions as before and included the global series differences.

    These difference series produce trend differences which on a degrees C per century are larger than when the non-satellite data set comparison were made. To show those larger trends, I have listed the trend slope and 95% CIs for the global differences and the larger zonal differences below.

    CRU-UAH 1979-2008:
    Globe: trend slope = 0.39 +/- 0.30; p = 0.015
    0-20N: trend slope = 0.62 +/- 0.26; p =0.004

    GISS-UAH 1979-2008:
    Globe: trend slope = 0.39 +/- 0.28; p = 0.009
    20N-40N: trend slope = 0.71+/- 0.46; p = 0.003

  14. Kenneth Fritsch said

    Alan Cheetham, thanks for the link. I’ll need to play with your filters, but it looks like it might be useful for future analyses.

    Ryan O, I would estimate that the raw data are similar for all data sets, excluding, of course, the satellite and radio sondes. I think we are looking mainly at differences in adjustments and that is why I judge relatively smaller differences to be important. GISS and GHCN use essentially the same raw data for the US and the adjustments give significantly different trends in difference series. Same goes for USHCN Ver. 1 versus Ver. 2. By the way GHCN uses breakpoints nearly exclusively to adjust temperatures for non-homgenieties.

    The way I have been looking at breakpoint calculations is that they are not subjective and I would much rather draw a straight line through segments of a time series that at least have a semblance of a linear relationship than attempting to draw one through a whole series of peaks and valleys.

    For final completion on these temperature data set comparisons I need to determine in more detail how CRU adjusts temperature data – if that information is available.

  15. milan mitic said

    Mitic CLIMATE ENGINEERING more rain



    Erosion trigger channel + huge tides = huge erosion of land tidal channels = low cost excavation with erosion = land desalination = more clouds = more rain = cooler climate = huge carbon sink

    Ask the farmer that got trouble with erosion because of rain

    what erosion would huge 12m tides do.

    Ask the scientist how big will evaporation be in bone – dry scorching hot desert if tidal system of canal and channels is made by erosion assisted excavation.

    1. evaporation from saline tidal water, canals, channels, tidal lakes, tidal marshes
    2. transpiration from mangroves and other sea water tolerating plants
    3. transpiration from rain forest around, ( tidal evaporation 1 and 2 = more rain = rainforest 3)

    Ask the engineer if it can be done.

    Ak the economist would project be economical
    if less: cyclones,floods, droughts, bushfires,

    more hydro energy

    Greener deserts and more clouds, cooler
    climate, and
    more water in rivers lakes and soil

    for more see:

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s

%d bloggers like this: