Sunday, December 6, 2015

Big UAH adjustment.

I noticed in Roy Spencer's latest post the following observation:
Of course, everyone has their opinions regarding how good the thermometer temperature trends are, with periodic adjustments that almost always make the present warmer or the past colder.
It's true that adjustments at his UAH are less frequent. But when they happen, they are large. I decided to plot adjustments to UAH in this year, compared to the adjuatments in GISS (thermometer land/ocean) made over four years. The GISS version of Dec 2011 was the earliest I could find on the wayback machine. UAH brought out v6 in beta during 2015, replacing v5.6, which is however still maintained.

Update. I have found on wayback more GISS data going back to 2005 (the directory name had changed). I won't add it to the original graph; it is too close to the other GISS to show. I've added below the fold a graph of differences between each dataset, new minus old, to show adjustments on a better scale. The accumulation of 10 years of "periodic adjustments" to GISS is still dwarfed by the adjustment made to UAH in 2015.

I've set GISS to the UAH anomaly base, 1981-2010, and smoothed the monthly data with a running 12-month mean. I've used reddish for UAH, and blue for GISS.
Update: I have appended a plot including GISS 2015 an RSS, with better scaling, below.

AS you see, GISS adjustments are much smaller. I should mention that if you use the GISS base of 1951-1980 the adjustments look larger. The reason is that GISS is a much longer record, and adjustments are cumulative, and the earlier base period brings in all the adjustments since 1951.

Eli has a forceful critique of UAH here. Measurement by satellite interpretation of a very indirect signal in a place that is hard to locate exactly is always going to be chancy. As Dr Mears, the man behind the RSS satellite measure, said, in discussing measurement errors:
A similar, but stronger case can be made using surface temperature datasets, which I consider to be more reliable than satellite datasets (they certainly agree with each other better than the various satellite datasets do!).
His comment on agreement was made before UAH v6, which improved the agreement, but not confidence in their stability. I suspect that UAH (and RSS) should adjust more often, but that it is not done because of the inherent uncertainty.
Difference plot below

Here is the difference plot. I can't put the anomaly of the GISS data from September 2005 on the 1981-2010 base, so I have made its average match the full GISS average from 1981-2005.

Now it is clear that even 10 years of adjustments to GISS are small in comparison to the adjustment to UAH this year.

Update - A zip file with the dta and code that I used is here. It just reads the data, resets the anomaly base, and plots.

Update: Here h/t ehak are Dr Spencer's similar plots. His amplitude is actually greater, since it isn't smoothed. But the pattern is the same.

Update: following the suggestion of reader Everett F Sargent below, I have redone the first plot including RSS (green). I have also added the GISS 2005 data and rescaled the y-axis.


  1. Good thing Lamar Smith is going to root out the corruption at UAH.

  2. Looks like things started drifting low in 2005 which would correspond with NOAA-18 coming on line. Other complications may include losing the aqua AMSU in 2011 because of increased noise.

  3. When UAH went from v5.6 to v6beta, some of the monthly changes were huge, especially for subglobal regions.

    Even globally, one month's anomaly changed by +0.30 C. One North Pole land monthly anomaly decreased by over 1.4 C. A USA48 month changed by -0.7 C.

    Imagine the outrage if Karl et al had changed any temperatures by this much.

    1. These changes could be a strong clue as to how the UAH analysis changed.

  4. Carl Mears knows what he is talking about. RSS has performed an uncertainty analysis, and assessed by that, the trend for 1979 through 2012 could be anywhere between 0.06 and 0.2 C/decade with 95% confidence. Here is the 90% confidence interval. The precision in Hadcrut kriging is five times better, 1.78+/-0.13 for the same period, based on their uncertainty ensembles.

    As I can see the trend of UAH v6 (and RSS) starts to deviate from other tropospheric indices about year 2000. The transition period from MSU to AMSU was June 1998 til July 2001 for UAH v6

    RSS and UAH v6 are not the only indices for troposphere temperatures. And the global surface indices fits nicely between the green and orange trendline of these upper air indices

    1. -a minor correction, the trend for Hadcrut kriging above is in C/century, should have been 0.178 to compare with RSS..

    2. Has Mears commented anyplace the divergence between RSS and these other indices?

    3. Well, RSS has this validation tool
      The comparisons end in 2011, but looking at the differences, it is quite clear that RSS starts to diverge from about year 2000
      Ratpac is the only continuosly updated and freely available weatherballon index (I expect an update for the full autumn 2015 by tomorrow, or so. It could be the warmest seasonal anomaly ever in the troposphere)

    4. Olof, re uncertainties in satellite based data.

      Interesting. This is the first I've heard of an uncertainty budget for either the satellite based temperature series. Do you have a link for the Mears paper as opposed to the data file you link to?

    5. Bill H, it should be this paper:
      The paper was submitted in 2010, but the uncertainty ensembles go through 2012, so they have likely done additional work later on..
      There is also some info on their website

    6. Olof: Thanks for those links. I think there might be an interesting post to be written on the difference between observation uncertainites in trends and the standard statistical uncertainty in the trend.

      In the second link, the TTT profile looks a lot like the profile for UAH6.0 (the 5.6 profile was rather lower in the atmosphere). Have any of you tried comparing UAH6.0 TLT against RSS TTT?

    7. Kevin, Yes I have compared UAHv6 TLT and RSS TTT. They are very similar in trend and shape. I have also made a UAHv6 TTT which is nearly identical as well. (It is quite easy, it's just 1.1*TMT-0.1*TLS. You dont have to do this with every single measurement, it works with monthly global averages as well)
      I have also constructed a NOAA STAR TTT-index, with the recipe above, which has a larger trend, 0.14 C/decade, compared to 0.11 for the others.

      Kevin, I have a special task for you if you have spare time and find it interesting..
      Make a troposphere hybrid temperature index with your hybrid infill technique (used on Hadcrut 4), in this case take the UAH TLT v6 (or other) global field and "anchor" it in the Ratpac 850-300 hPa station data. An alternative description of such a product could be "a radiosonde corrected satellite dataset"..

    8. Oh, interesting idea. Basically our existing hybrid method but using ratpac for the obs and uah for the spatial component, under the assumption that ratpac is temporally homogenous but spatially incomplete, but uah is spatially homogenous but temporally inhomogeneous.

      It's a good idea, I'll add it to my to-do list. I can't promise to get to it straight away.

    9. I've already done things the other way around, producing a UAH average masked to Ratpac stations:

    10. Yes, I saw that. I think the tests you've done are the right ones.

      Further on looking at the RATPAC data, the documentation is rather discouraging. I suspect the problems with the data are going to dominate, so I think that going much further is going to involve rather more work that just plugging the data into existing methods. At this point I think there is probably more to be gained by a spatial analysis of the differences between UAH 5.6 and 6.0.

  5. Nick,

    I think what's MUCH more interesting is to compare {UAH 56 - RSS 33] and [UAH 60b4 -RSS 33] for the LTL using the 1981-2010 baseline anomaly period (as you have done).

    I then split those into halves (1/1979 thru 5/1997 and 6/1997 thru 10/2015).

    UAH 56 versus RSS 33 (1/1979 thru 5/1997): R^2 = 0.8531
    UAH 60b4 versus RSS 33 (1/1979 thru 5/1997): R^2 = 0.8864

    So some improvement for the 1st halves.

    UAH 56 versus RSS 33 (6/1997 thru 10/2015): R^2 = 0.7792
    UAH 60b4 versus RSS 33 (6/1997 thru 10/2015): R^2 = 0.9238

    So MAJOR improvement for the 2nd halves.

    Will do the RMS for each of these in a little while.

    It appears quite obvious (to me at least) what Woy & Co, are doing. Conceed that RSS is 'right' about the LTL time series, fudge their numbers to get closer to RSS than say "see two 'independent' LTL time series that agree with each other within the 95% confidence interval slopes" to go up against the four surface based SAT's (NOAA/NASA/Hadley/BEST).

    Monkers will then get two LTL time series (RSS and UAH) to pick from (currently using RSS) for ~6/1997 thru to current.

    I simply don't trust either RSS or UAH LTL time series as being representative of SAT at all.

    1. Oops, LTL should be TLT, I don't know what I'm doing there, but I did the exact same thing at RR.

    2. Some additional numbers from the two halves comparison:

      UAH 56 - RSS 33 1/1979 thru 5/1997 (min,max,sigma(rms)) -0.157,0.264,0.0632
      UAH 60b4 - RSS 33 1/1979 thru 5/1997 (min,max,sigma(rms)) -0.171,0.221,0.0594

      UAH 56 - RSS 33 6/1997 thru 10/2015 (min,max,sigma(rms)) -0.187,0.232,0.0824
      UAH 60b4 - RSS 33 6/1997 thru 10/2015 (min,max,sigma(rms)) -0.126,0.118,0.0483

  6. Nice, as usual.

    Do us all a favour and rip "A simple demonstration of chaos and unreliability of computer models" to shreds. You've pretty well done it in the comments, but it would be nice immortalised in a post.

    1. Thanks, William. The WUWT discussion was in a way too scattershot to take seriously, but I did feel a general ramble on stability of ODE solution coming on. I'll see what I can do.

  7. William,

    Don't know what WTFUWT? is doing, but I used a di = 0.357 (which strictly speaking has nothing to do with time but is iteration step size), n=6 to converge, at twice that iteration coefficient it appears to approach infinite iterations and anything above 0.714 oscillates about the correct solution at an increasing rate as di goes up.
    Those numbers are for tinf=243,t0=300,beta=170.

    Oh, and of course T^4 = T*T*T*T (and Excel sucks (although I do use it all the time)).

  8. Can I use some of your graph for my blog? With proper reference of course.

  9. RSS 4.0 on its way:

    In better agreement with other temperature related measurements. Like water vapor.