Data Comparisons

Do the Leprechaun’s updates to GISSTemp “matter”?

13 Comments

There were quite a few changes in historic temperature anomaly values in the most recent GISSTemp update; (see GISSTemp Jan….) In comments people wanted to know if this “matters”. As usual, the answer is: “Whether or not the changes matter depends on what question you ask.”

Today, I’ll post some graphs and numerical value for trends we obtain using temperature anomaly data up to Dec. 2008 taken from the current GISSTemp records published in mid February 2009 and value published in January 2009.

The graphs shown will address this question:

  • How much do the changes affect the least squares trend for three periods: Jan 1900-Dec 2008, Jan 1980-Dec. 2009 and Jan 2001-Dec. 2008?

Most of the narrative will be contained on the figures themselves.

Trends

First, let’s compare the surface trend from 1900 through year end 2008:

Figure 1: Surface Trend Since 1900.(Click for larger.)
Figure 1: Surface Trend Since 1900.

If we use the current GISSTemp values, the surface trend is smaller than if we use the previous value. The change is small– about 1% of the mean trend observed.

Now, let’s examine a shorter period, beginning in 1980:

Figure 2: GISSTemp since 1980(Click for larger.)
Figure 2: GISSTemp since 1980

If we compute the trend since 1980, the GISS trend remains strongly positive, but the trend is slightly less than the previous value. Does this matter?

It won’t matter if we do a statistical test comparing the trend to the null hypothesis of “no warming” and assume all residuals from linear are AR1 “noise”.

I don’t know if this will “matter” when I repeat statistical comparisons to model trends– but it actually might. I will be doing this test– because I had planned to wait for the Feb 2009 update of GISS before I finalized some analyses using Dec. 2008 as the final point. (The first GISSTemp update often changes the most recent month’s value quite a bit. I hadn’t expected so many changes in other historic values. I’m waiting for HadCrut next.)

The reason the changes in historic data might matter is that while the effect on the trend is small compared to the observed trend, it is not quite so small compared to the difference between predictions and models. This can matter when some “rejections” of model-mean trends are just on the border of significance using any particular statistical test.

Now, let’s examine a shorter period suitable for testing the forecasting ability of the full modeling process (beginning with formulation of scenarios). Here is a graph beginning in 2001:

Figure 3: GISSTemp trends since 2001(Click for larger.)
Figure 3: GISSTemp trends since 2001

If we compute the Jan 2001-Dec 2008 trend using the current GISSTemp data, the trend remains negative, but it’s less negative than if we use the data in cache. The difference is very small relative to the trend itself. However, once again when doing statistical tests to reject or accept a hypothesis, and the trend is just on the border of a “reject/accept” threshold, small changes in the data can tip the balances. So…. I’ll see what happens. (I’m confident it won’t change the much when we apply the methods described in Santer17 to see if trends match IPCC projections.)

Summary

If we are simply examining the change in the trend based on the historic revisions between Jan 2009 and Feb 2009, the magnitude of the changes is small. For most practical purposes, these temperature changes affect computed least squares trends by less than 1%.

Of course, the effect on these speciric computed trends is not the only thing that “matters”. I’m sure readers are aware that other things matter. The changes always raise questions in people minds. Some will wonder what we would learn if I used the cached version of temperatures published in 2000? Or 1990? Or whenever. The answer is: I don’t know. I would have to download them and check.

The fact that this answer is not easy to obtain matters both in terms of science and in terms of politics.

Note on title

According to the scientific, totally unsecure system web based poll, 74% of votes support the notion that this months changes in GISSTemp are the work of Leprechauns. The poll is still open.

13 thoughts on “Do the Leprechaun’s updates to GISSTemp “matter”?”

  1. The big question in my mind is what causes GISS to revise temperatures that far back. I think I read somwhere that GISS does some automatic trend adjustment which takes account of recent actuals. Is this true? However the adjustments from Jan to feb seem to have no obvious pattern. Any ideas Lucia?

  2. I’ve said this before, but once again, I would like to register my complete and utter amazement and horror that these changes to historic records are allowed to stand.
    Will the Gettysburg Address be changed to reflect new understandings?
    Mike Bryant

  3. Mike Bryant,

    the understanding of the Constitution and the Bill of Rights is almost continuously being revised by Judges, Politicians, teachers, reporters… Why should the temperature record be exempt??

  6. Mike, if the temperature data has a spurious warming bias in it as a result of accumulating UHI effect, would you want to see that bias removed in any analysis?

  8. Mike,

    Well, I presume that the raw data is all still ‘there’ (wherever!). For example, there’s recently been close interest in the READER data for the Antarctic – these figures don’t change when GISS applies adjustments to their analyses. It seems to me that GISS, Hadley, UAH, RSS, etc. are all presenting analyses based on raw data which remains in existence as such – unalysed raw data. I wouldn’t know how to make anything of the satellite raw data, that’s for sure, but then I wouldn’t know what to make of meterological station data records where the siting has changed or where a city has grown up around it. It seems to me that you are declaring there should be no analysis of data!

  9. They can analyze it all they want as long as they leave the old data intact. Don’t call the new analysis the data.
    Mike

  10. Hi Lucia,

    This is a summary of the changes introduced in the GISSTEMP monthly temperature data between 1880 and 2008, comparing the February 2009 version (today’s data) with the January 2009 version (as it existed on Jan 14th):

    1 month has cooled 0.04C (Nov 1999)
    1 month has cooled 0.03C (Jan 1900)
    12 months have cooled 0.02C
    225 months have cooled 0.01C
    123 months have warmed 0.01C
    11 months have warmed 0.02C
    1 month has warmed 0.03C (Nov 1896)

    That means there have been changes in 374 of 1536 months (about 24% of the data).

    The linear temperature trends (in ºC/century) have changed as follows, as a result of these changes in the monthly data. I put the global trend and then the different trends in 30 year periods, as it seems to be the time frame where the direction (sign) of the trend seems to change.

    GLOBAL 1880-2008: (OLD +0.56522) cools to (NEW +0.56333)
    1881-1910: (OLD -0.1287) cools to (NEW -0.1333)
    1911-1940: (OLD +1.2305) warms to (NEW +1.23697)
    1941-1970: (OLD -0.2343) warms to (NEW -0.2311)
    1971-2000: (OLD +1.54494) cools to (NEW +1.52833)
    2001-2008: (OLD -0.1458) warms to (NEW -0.1438)

    Best regards.

  11. Nylo,
    Those figures are very interesting. I wonder what the cumulative changes are since Hansen took the reins at GISS. Would those cumulative changes matter?
    Mike Bryant

  12. Mike,

    I already studied that. There has been a major total change since the first GISS version I have access to, which is the one that was available on December 2005. However most of that change happened sometime between the versions of February 2006 and June 2006. Whichever the reason was for the very big change that took place then, consisting on a strong cooling of the past and warming of the present, I am quite confident that it was something different than the leprechaunts we are dealing with now. However I don’t know the story of that change.

    The leprechaunts we deal with now have different effects one month and the next one. I have no idea why some temperature records change and others don’t, but the net effect seems to be a bit random, sometimes providing more warming, other times cooling.

    There is however something that is pretty common to every monthly change. The monthly temperatures in the time period between 1950 and 1980 change very little compared to the monthly temperatures in the beginning or at the end of the whole data set, and normaly you will see aproximately as many warming months as cooling months in that period. Let’s remember that 1950-1980 is the time period whose average temperature GISS uses for zero anomaly, so it seems like the leprechaunts like to keep the same reference level, come what may. This for me has a very funny effect: the temperatures GISS seems to be more confident about in the whole dataset are those between 1950 and 1980. We should go back in time and start using the same technology we had then lol !

    To illustrate this: I already told that 24% of the historical monthly temperature data in GISS has changed this month. Well, if we focus on the 1950-1980 period, only 15 of the 372 monthly temperatures have changed (4%), with nine 0.01C coolings and six 0.01C warmings. So the changes concentrate OUTSIDE the 1950-1980 interval.

    I don’t ignore that, if there was a generalised warming or cooling over the 1950-1980 period, we would not notice it inside, as its average temperature would move and the anomalies would be related to a new level. However if all the reference period warmed or all of it cooled, we should see a common cooling or warming outside that reference period, respectively, and we don’t see that. In fact, normally we observe opposite changes before and after the reference period: either the past cools and the present warms, or the past warms and the present cools. This is not consistent with a mere change in the reference level. So there is something else going on there that keeps the temperatures of the reference period mostly unchanged. The changes in the 1950-1980 period alse tend to be simetric: either there is cooling before 1965 and warming after it or viceversa. There is kinda pivotal point in 1965.

    For the record, comparing today’s version with Dec 2005 version, the changes are:

    0 months cool by -0.12ºC
    3 months cool by -0.11ºC
    6 months cool by -0.1ºC
    15 months cool by -0.09ºC
    28 months cool by -0.08ºC
    45 months cool by -0.07ºC
    62 months cool by -0.06ºC
    101 months cool by -0.05ºC
    124 months cool by -0.04ºC
    172 months cool by -0.03ºC
    168 months cool by -0.02ºC
    179 months cool by -0.01ºC
    135 months warm by 0.01ºC
    102 months warm by 0.02ºC
    87 months warm by 0.03ºC
    53 months warm by 0.04ºC
    29 months warm by 0.05ºC
    18 months warm by 0.06ºC
    21 months warm by 0.07ºC
    4 months warm by 0.08ºC
    5 months warm by 0.09ºC
    3 months warm by 0.1ºC
    2 months warm by 0.11ºC
    1 month warms by 0.12ºC

    The average temperature change for months in 1880-1950 is -0.024ºC
    The average temperature change for months in 1981-2000 is +0.007ºC

    The Global Trend for 1900-2000 has warmed from 0.519 ºC/century in Dec 2005 to 0.565 ºC/century now, as an artifact of the changes in the historical records (only NEW data post-2005 should have been added and it shouldn’t have affected the 1900-2000 trend at all).

    Best regards.

Comments are closed.