H3-H4 temperature differences

In this post I look at the effect of changes in the original set of CRUTEM3 stations.

Shown below are the spatial changes in temperatures between HadCRUT3 (2013) and the latest HadCRUT4.6 (2019) from 1998 to 2010. These are calculated using just the original set of CRUTEM3 stations and their counterparts in H4. Each annual 5×5 deg. cell is averaged over  12 months.

Fig 1: Map showing the effects of CRU adjustments in station temperature between 1998 & 2010

There is essentially no change in SST between H3 (HadSST2) and H4.6 (HadSST3) so the increased warming trend since AR5 is simply caused by changes to the underlying CRU station data. Most of the significant changes occur in Asia and N. America. Here are the static temperature differences shown on the same scale as the temperature anomalies for 1998 and 2010.

Fig2. Annual temperature anomalies calculated using CRUTEM3 (2013) and HadSST2.

and now 2010.

Fig 3. HadCRUT4.6 calculated using only modern versions of the original CRUTEM3 stations combined with HadSST3.

Finally  I show  the differences between H4 and H3 temperature anomalies plotted  on the same scale. In an idea world this plot should be pale blue with zero difference.

The difference between Figure 4 and Figure 5

The yellow to pink areas are roughly 0.5 to 1.0C warmer than HadCRUT3 , demonstrating how  after 7-years an apparent increase in ~4000 global land temperatures can explain why the AR5 hiatus essentially evaporated.  The original weather station data cannot have changed, so these effects are probably caused by merging of nearby local stations, homogenisation between regional stations, correcting errors, or something else. Probably only Tim Osborne or Phil Jones could explain.

This entry was posted in AGW, Climate Change, climate science, CRU. Bookmark the permalink.

10 Responses to H3-H4 temperature differences

  1. Olof R says:

    The change from HadSST2 to 3 had an effect on the pause. The “pause” trend 1998-2012 increased by 0.035 C/decade in HadSST 3
    Here is a comparison of old and new HadCRU datasets trends through Wood for Trees:


    The exact calculated trends can be seen if you press “raw data”

    The increase in SST trend is very much what one can expect from the ship/buoy adjustment that was introduced in HadSST3. Buoy data was adjusted up ~0.12 C compared to ship ERI data. The proportion of buoy data increased approximately from 50% in 1998 to 90% in 2012. Thus, the adjustment should increase the temperature by 0.048 over the period. Dividing this by 15 years suggests a trend increase of 0.032 C/decade, which is close enough to the actual 0.035

    The version change in Crutem increased the land trend more, ~ 0.095 C / decade. Despite the fact that land only constitutes 30% of the global surface, the land component must have had a somewhat larger effect on the global trends, compared to that of SST.

  2. A C Osborn says:

    Is not HADCRUT data based on NASA data?
    If so I suggest that you take a look at the analysis carried out by E M Smith on the GCHN data changes between versions.
    They are extremely revealing.

    • Nick Stokes says:

      “Is not HADCRUT data based on NASA data?”

      • A C Osborn says:

        How odd.
        From the HADCRUT website
        “Station sources
        This is a two digit source code, in Jones et al. (2012) and Brohan et al. (2006). References to sources are in these publications. Few stations have a single source of data due to real-time updating (from WMO’s CLIMAT messages or decadal World Weather Record publications or from the publication Monthly Climatic Data for the World). The code given here is therefore where most of the data series has been obtained.”
        It specifically states that 440 stations’s data for Europe, East Asia, Africa, USA, S. America and Australia, 113 stations for Middle East, E. Asia and N. Africa come from http://www.ncdc.noaa.gov/ghcnm/v2.php

        I must be misundesrtanding what “is therefore where most of the data series has been obtained.” means.

  3. Frank says:

    Clive: I’m confused about what is shown in your plots of H4-H3 plots. I’d assume – perhaps wrongly – that you are working with temperature anomalies. In the average anomaly in 2010 is the difference between 2010 and the 1960-1990 or 1950-1980 average anomaly. If so, the data for 2010 is roughly the temperature in 2010 minus the temperature in 1975 and includes the changes caused by homogenization over a half-century. If so, homogenization has almost the same effect in 1998 and in 2010. So why show how this varies from year to year?

    • Clive Best says:


      Yes they are always anomalies. The plots just show the annualy averaged difference in anomaly for each lat,lon bin. If the station data for HadCRUT3 and HadCRUT4 were identical then the plots would be all light blue = zero difference. So the net effect of changes to the original H3 stations temperatures explains why 2010 slowly edged warmer than 1998 with each new release of HadCRUT4.

      • Frank says:

        Clive: Thanks for the reply. In Figure 1 (which changes with year), if you focus on the grid cells with the most extreme changes, those changes stick around for a number of years and then may vanish. This suggests that a breakpoint requiring a large or different correction was identified in one version or the other. The extreme corrections that don’t persist presumably must reflect a less radical adjustment at the next breakpoint. In the long run, the only thing that really matters is the total amount of warming and trend over an important (and therefore relatively long period to minimize the impact of unforced variability).

        The grid cells with large corrections that remain relatively constant must be corrections before 1998, including to the period that defines a temperature anomaly of zero. In the US, corrections for breakpoints induced by a change in TOB are presumably present in every grid cell and refinements in breakpoint correction methodology could produce constant changes in those grids.

        While respectful of your work, this way of presenting the data doesn’t help me deal with the problem. The big picture is how much the the global record has been changed by these mostly ad hoc corrections. Representative corrections at individual stations provide the details. If one doesn’t know the cause of a breakpoint, one doesn’t know if it should be corrected or not. Consider a station move from a location that has urbanized to a nearby less biased location that resembled the conditions when measurements were being made a century or half-century earlier. That breakpoint arguably shouldn’t be corrected. Or consider station maintenance that makes the screen more reflective or better ventilated and restores original measurement conditions. Although some breakpoints (TOB) clearly should be corrected, it is obvious to me that all should not be corrected. I don’t know how to tell which ones should be corrected.

        The one thing I am sure of is that the best time to reliably detect a breakpoint is when it is windy and the temperature of the air inside the screen is representative of a large, well mixed mass of air near the surface. If one thermometer starts behaving differently from its neighbors on calm days, it could be reliably reporting on temperature differences in it micro-environment. If one thermometer starts behaving differently from its neighbors on windy days, micro-environment is not likely to be responsible.

        • Clive Best says:

          Yes you’re right. We need a simpler story for what really has changed between V3 and V4. Chiefio (E.M. Smith) did a huge amount of work looking into this but the conclusions remain a bit obscure. It cannot be statistics. GISS quote their errors on the annual temperature anomaly as 0.05C yet V4 seems to have systematically added up to 0.05C of warming onto the post 1990 V3 results. There is something very fishy going on.

  4. Clive, have you looked closely at the temperature data adjustments? My understanding is that for CRUTEM3 adjustments were made by the CRU and by the GHCN (or whatever the organisation was previously called). The major change for CRUTEM4 was that responsibility for data adjustments was pushed out to the national meteorological services (NMSs).

    As I explained in my audit of the HadCRUT4 dataset, if the NMSs followed the recommendations of the WMO, all adjustments would have been by a constant amount regardless of whether the recorded temperatures has been increasingly distorted over time (e.g by increasing UHI). Typically the constant amount by which the data is adjusted is determined at the time of the greatest distortion, such as when a weather station is relocated to a new site.

    In these circumstances the adjustment by a constant has two notable features. First, the earliest data is excessively adjusted. Secondly the false trend created by the increasing distortion will still be present in the data.

    Given that adjustments are made to ALL previous data, multiple data adjustments over time will mean multiple changes to the earliest data.

    If you need a copy of the HadCRUT3 station data to compare with HadCRUT4 station data then let me know and I will send you the .zip file produced by the CRU.

    Just be aware that not only did the number of stations change but in some cases the name of the station changed too.

Leave a Reply