Replication #2: Selection of Gridcells

The Corrigendum SI stated that:

MBH98 made use of all nearly continuous monthly gridpoint surface temperature records (no single gap greater than 24 months, and no more than 10 years of total missing data.

I checked these criteria against the temperature dataset archived at the Corrigendum SI and found that 242 out of 1082 gridcells selected for the 1902-1993 period failed one or both criteria, while 229 gridcells in the 1854-1993 ( rather than the 219 shown in MBH98) met both criteria. Some curious differences between newer and older HadCRU versions are noticed in passing. More


  1. John A.
    Posted Feb 20, 2005 at 3:26 AM | Permalink


    Could you put in a link back to the weblog from the webpage?

  2. Louis Hissink
    Posted Feb 20, 2005 at 4:21 PM | Permalink


    I think I can see what is happening with the grid cells – it looks like they assign aribtrarily grid cells defined by lat/longs, and average the mean temperatures in a cell. This is a classic error in estimation since they are effectively averaging intensive variables within a pre-defined cell.

    What should have been done is to class the earth surface initially as land -ocean, then refine the land into urban, rural, desert, rainforest etc by assigning polygons – easily done with landsat imagery. Further subdivisions for ice-caps so that the final result is the earth’s surface described into various accurately defined polygons of area according to surface type.

    The next step is to see where the meteorological stations are and the assign each station an area of influence defined as a polygon whose boundary is the 1/2 way point between adjacent stations, (old fashioned way of doing polygonised ore reserves) so that we end up each station having an area assigned to it.

    Each mean temperature of each station is then multiplied by its associated area of influence to yield a countable quantity, (temp*area), which are then all added, and finally divided by the total surface area to yield the global mean temperature for a specific instant in time – say 21 june for a particular year.

    The process is repeated annually but areas might well change – one could see ice caps either growing or shrinking, or whatever.

    This is how the global mean temperature should be computed.

    The other technique using cells merely estimates the mean temperature of the thermometers.

    Steve’s comments: Louis, I know how ore reserves calculations work and, more unfortunately, I know what happens when they don’t work out. On this board, the following point will probably have meaning to you and me. I lived through a project in a narrow-vein mine which had some low-grade intersections and high-grade intersections. The actual area of influence of the low-grade intersections was much bigger than in the polygons used to calculate ore reserves and the actual area of influence of the high-grade intersections was much lower. It struck me that a much more sensible way of calculating ore reserves would be to relate the area of influence to the grade of the ore. I never wrote up the concept, but it wouldn’t be that hard to do a computer program to do this.

    On your point about temperature averaging, I think that you’re missing a step in the procedures. Compilations like CRU average “anomalies” rather than temperatures. “Anomalies” in their jargo means the difference from the average over a reference period (say 1961-1990). So there is no simple averaging of “intensive” properties along the lines that you suggest and so I don’t see that this as being a problem along the lines you suggested. I’ve looked at Essex and McKitrick on this issue and, as of right now, don’t follow their argument. I have no plans to get into the topic for quite a while. Regards, Steve

  3. Peter Hartley
    Posted Feb 20, 2005 at 7:12 PM | Permalink

    In terms of auditing grid cell temperatures, if you are not aware the the work of Warwick Hughes you might find his web site interesting. He has found some particularly surprising things in some of the Russian numbers. On a related note, I also have yet to see a thorought statistical analysis of the effects of the varying numbers of stations included in many of the ground temperature averages. Surely, changes in the cities included in these averages over time has affected the average so calculated. It would be interesting to see how.

    Steve: I’m familiar with Warwick’s analyses. I like that sort of detailed case study.

  4. Louis Hissink
    Posted Feb 20, 2005 at 10:32 PM | Permalink


    Fair comment – but as a first pass estimation methodology, that is where one starts. If experience shows what you detailed in the vein situation, then calculation of variograms etc to see what the area of influence should be, would be the next improvement in accuracy, but just adding temperatures without linking them to a physical object or area, is tantamount to averaging the temperature of the measuring thermometers. However, it is a bit technical for here, so I’ll spend a month or so to create and example.

    Steve’s comments: Louis, my comment about the ore reserve example had nothing to do with the temperature situation. It was just something that was in a back recess of my mind. By the time that we realized the problem, there wasn’t much that we could do about it. Please don’t get that example conflated with temperature averaging; I was really just musing about an old misadventure.

  5. pete
    Posted Sep 9, 2009 at 7:21 PM | Permalink

    The climate2003 site seems to have been cybersquatted, so the “more” link is broken.

    Steve: You should be able to find the relevant page at changing the direction mutatis mutandi.

    • pete
      Posted Sep 9, 2009 at 9:03 PM | Permalink

      Re: pete (#5), checked there already, couldn’t find a page with that name.

One Trackback

  1. […] post Replication #2 pointed out a situation where a simple step in MBH98 like the selection of 1082 temperature […]

%d bloggers like this: