New Dataset from RSS: End of the Satellite “Pause”?

Carl Mears and team at RSS have published a new paper describing a revision of their data for atmospheric temperature. The focus is on improving the “diurnal correction,” which is necessary because different regions of Earth are observed at different times of day. The upshot is that the lower atmosphere has warmed faster than was previously believed.

Perhaps the abstract says it best:


Temperature sounding microwave radiometers flown on polar-orbiting weather satellites provide a long-term, global-scale record of upper-atmosphere temperatures, beginning in late 1978 and continuing to the present. The focus of this paper is the middle tropospheric measurements made by the Microwave Sounding Unit (MSU) channel 2, and the Advanced Microwave Sounding Unit (AMSU) channel 5. Previous versions of the RSS dataset have used a diurnal climatology derived from general circulation model output to remove the effects of drifting local measurement time. In this paper, we present evidence that this previous method is not sufficiently accurate, and present several alternative methods to optimize these adjustments using information from the satellite measurements themselves. These are used to construct a number of candidate climate data records using measurements from 15 MSU and AMSU satellites. The new methods result in improved agreement between measurements made by different satellites at the same time. We choose a method based on an optimized second harmonic adjustment to produce a new version of the RSS dataset, Version 4.0. The new dataset shows substantially increased global-scale warming relative to the previous version of the dataset, particularly after 1998. The new dataset shows more warming than most other middle tropospheric data records constructed from the same set of satellites. We also show that the new dataset is consistent with long-term changes in total column water vapor over the tropical oceans, lending support to its long-term accuracy.

The effect on TMT (middle-troposphere temperature) globally looks like this:


The black line shows the old version, the light blue line the new. Note that the overall trend in the new version is 60% bigger than in the old version. The green line at the top shows the effect of their improved diurnal correction.

This should dampen the enthusiasm of deniers like Ted Cruz who have relied on satellite data from RSS to dispute global warming. But I doubt; I suspect instead that Ted Cruz will either find some new reason to deny global warming, or will go on a witch-hunt of Carl Mears and the RSS team, accusing them of fraud because the data don’t say what Ted Cruz and his ilk want it to say.

If you like what you see, feel free to donate at Peaseblossom’s Closet.

74 responses to “New Dataset from RSS: End of the Satellite “Pause”?

  1. The question I have is, will UAH also make similar adjustments given the problems cited by Mears et al? They are, after all, supposed to preparing to release their methodology for UAHv6.

    • It gets complicated for the ‘skeptics’, doesn’t it? We all remember Mears’s statement that he considered the surface datasets to be more reliable, partly because they agree with each other better than the sat datasets do. So the “satellite data is the gold standard” crowd got all chuffed when UAH6 significantly improved the agreement with RSS, which pulled the rug out from under Mears’s statement.

      But the RSS modification is clearly going to affect that, and maybe throw them back out of whack again.

      It’s hard to be a ‘skeptic.’

    • Has UAH EVER released a complete file including source code on its methodology for any version or subversion such that another group could analyze precisely what they are doing?

  2. Tamino, will you please post an update of your comparison of RATPAC to RSS 3, RSS 4, UAH 6? Looks like RSS 4 eliminates the post 2000 divergence!

    [Response: I was planning to.]

    • Tamino, can I use your graph for my own blog.

      [Response: It’s not my graph, it’s from the paper by Mears and Wentz.]

    • Note that the posted graph is TMT, not TLT, and TMT is currently the only data available on their website. Other graphs in the paper indicate that RSS is now “middle of the pack” in terms of trends, both global and tropical, while UAH is now the outlier. So, getting closer one increment at a time.

  3. Craig Tevis

    Even if the RSS data wasn’t corrected Cruz will lose his 18 year 8 month pause. The UAH Feb 2016 global anomaly is out, and it’s the highest in satellite history. No amount of cherrypicking will show a lack of warming.

  4. Chris O'Neill

    The upshot is that the lower atmosphere has warmed faster than was previously believed.

    Of course they say that. The global warming conspiracy requires them to.

  5. So the global warming thugs have coerced RSS into rigging their numbers? The conspiracy widens…
    (This is meant as sarcasm)

    • Oh, it’s not sarcastic at all. I expect this to be the first thing they’ll say on WUWT. The appropriate response is:

      “What’s your recommendation? Leave a known bias in the data? Or correct the data to remove the bias?”

      • Professor Ingemar Nordin over at the Swedish blog “Klimatupplysningen” has already denounced this new result, beating Watts:
        “Carl Mears apparently twisted the arm of his own staff to get a measly little increase trend. Carl Mears has always been a true believer and quite pissed that the satellites have not been sufficiently pliable. One wonders how many “corrections” that has been scrapped along the way?”

      • How did you guess?

        “The ‘Karlization’ of global temperature continues – this time RSS makes a massive upwards adjustment”

      • John Mashey

        Ingemar Nordin was one of the “300 scientists (sic)” urging Lamar Smith to waste US taxpayer money. Actually, about 30% of the signers were from outside US, including a whole bunch of the Klimatupplysningen folks. To assess the calibre of discussion there, try the website, which is in Swedish, but Google Chrome works OK for translation.

        “Ingemar Nordin is a professor in the Department of Philosophy, University of Linköping. He has for many years been interested in natural science and technology; their structure and motivations. For many years it has been the policy’s impact on the scientific research that has been the focus. Climate research today is perhaps the prime example of such influence and corruption of science since eugenikens days. Ingemar Nordin has written a couple of books on art, politics and science as well as hundreds of articles on the subject.”

        Ingemar Nordin was a defender of Murry Salby’s CO2 nonsense.

        SO, NO SURPRISE.

      • John Mashey, my wife says you just can’t trust those philosophy majors. (And yes, I am guilty as charged.)

  6. I’m glad this is resolved. It is obvious something was wrong, and its great that smart people figured it out.

  7. Intuitively I would expect tropospheric warming to always be a bit slower than surface warming – is that what the science says, and climate models predict? Or should the satellite series show the same warming trend as surface datasets, if they’re accurate?

    • You are correct, and RSS v4 does indeed show less warming than the surface. So all is well.

    • Why do you expect less warming?

      As I understand it, you expect overall more atmospheric warming mainly because of changes in the moist adiabatic lapse rate. One way of thinking of it is extra evaporation below equals more condensation and latent heating aloft although that’s not complete.

      • Interesting Mark, thanks for the RealClimate link.

      • Pierre-Normand Houle

        It’s true that changes in the adiabatic lapse rate (which grows closer to the moist adiabat, mainly in the tropics) causes the troposphere to warm faster than it would were it not for the negative lapse rate feedback. However, precisely because of the large latent heat exchanges at the ocean surface/atmosphere boundary the tropospheric temperature is more tightly coupled with the ocean surface temperature than it is to the land surface temperature. Since the ocean surface warms overall at about only half the rate of the land surface (due to the larger thermal inertia), it is to be expected that the lower troposphere wouldn’t warm as fast as the global surface average.

      • One effect is that TMT and TLT still include a large effect from upper atmosphere (and this is even based on theoretical models of how they *ought* to react), so they will both see some salami slice of cooler upper tropospheric radiance in their figures.

  8. Is there any information on how this will affect the TLT figures?

  9. Is the new dataset available somewhere? I couldn’t find it.

    [Response: Here.]

  10. The gig is up. Mears used the word “adjustments”, which everyone knows is a synonym for “tampering”. /s

    I cant wait to see what this will do to Monckton’s 18 year fill-in-the-blank months pause. Grabbing popcorn.

    • Monckton’s cherry pause is gone even without the new adjustments: Feb RSS TLT anom is 0.97. That will make the trend from any start date positive.

      • You know, you’re right! And by the narrowest of margins. I just downloaded the full (unadjusted) RSS dataset into Excel and ran a LINEST function on every period ending in Feb. 2016, and they are all positive. The minimum, from December 1997 to present, is 0.000098789, but that is still positive. The pause has ended even without Mears’ latest ‘subterfuge’, and I’m stuck with all this Orville Redenbacher on my hands.

  11. Spencer just came out with his Feb 2016 data, & they show the largest anomaly ever, +0.83° C, breaking the Mar 1998 record. He literally had to add another box to the top of his graph. And that’s after his v6 cooling adjustments that subtracted almost 0.2° C warming.

    We’ll see what the Spencer/Christy v6 methodology methods say. They’ve already revised them several times.

  12. Tamino, we’re all looking forward to your updated comparison of the RATPAC radiosonde data with the latest UAH & RSS data. Thank you!!

  13. RSS 4.0 is very close to the TMT trend calculated by STAR (0.121 K/dec).

  14. Another scientists Justin Data! KINSPEERCY!!!!1



  15. Lars Karlsson

    Is this related to the work of Po-Chedley, Thorsen and Fu?

  16. In case anybody’s curious, here’s what Christy’s favourite graph looks like with RSS v4.0:

    And here’s a fairer comparison, without the 5-year running mean suppressing variability, and aligning both series to the RSS baseline instead of forcing the trends to cross at 1979:

    The red data-point is an estimate of 2015 values, assuming the increase over 2014 is the same as that in RSS v3.3. Not sure why v4.0 doesn’t have it yet.

    Looks to me like the models are doing pretty well.

    [Response: One should also bear in mind that the new data are TMT, not TLT, so they’re influenced by the cooling of the stratosphere.]

  17. How ’bout that? Of course, as just about everybody above mentioned, the weeping and wailing about this latest ‘fudging of the data’ (the rawness of which should have been a Holy Trust) will be cacophonous and protracted.

    My regret, though, is that my plan to observe (and document) the implosion of Lord Monckton’s ‘pause’ graph is going to be seriously compromised.

  18. Everett F Sargent

    [Response: One should also bear in mind that the new data are TMT, not TLT, so they’re influenced by the cooling of the stratosphere.]

    AFAIK, TMT and TLT use the same channels two (2) and five (5) … per the RSS website … Sounding Products

    IMHO, I believe both TMT and TLT are somewhat influenced by a cooling stratosphere (e. g. cooling bias). Pure conjecture.

    I kind of doubt that whenever RSS releases v4.0 TLT data (if they continue to release a TLT product) that there will be a corresponding ~60% increase in TLT trend (1.2C/century*1.6 = 1.92 C/century). More likely is something along the lines of the same delta (0.125 – 0.078 = 0.047C/century).

  19. Well, it is possible to estimate what the new RSS TLT v4 trend would be..
    I took the last 14 months from the old v3.3 TMT product and spliced it to the new TMT (which ends in 2014).
    Then I made a TTT-index (1.1*TMT-0.1* TLS). The resulting trend (1979-now) was 0.179 C/decade, and the trend from year 2000 was 0.146, so the pause has essentially disappeared.
    If UAH v6 TMT is replaced by the new RSS v4 TMT in the UAH v6 TLT- formula, the trend for 1979-2016 becomes 0.21 C/ decade..

    • Interesting. I did something similar, in a somewhat different way. I used the UAH v6 formula too, but with RSS data for TMT, TTS. and TLS. Not quite ideal, since TTS doesn’t start until 1987, but it avoids mixing data from different groups. Using TMT v3 in this way gives me a TLT record with a good match to the real RSS one – monthly errors of around 0.1K, but no significant difference in trend. Using TMT v4, my version of the TLT gives a trend of 0.2 K/decade, from 1987-2015, matching yours nicely. I also compared it to HadCRUT4, and it matches up pretty well, with a difference in trend of 0.04 K/decade.

      So, early days, but it’s looking like RSS v4.0 is more consistent with a) the surface record, b) the models, and c) itself, given that the optimised diurnal corrections do seem to work.

      Incidentally, v4.0 is now available here, with data right through February 2016.

      [Response: That link seems to be for data from v3.3, not the new v4.0. I haven’t yet found a v4.0 TLT.]

      • Yeah, there isn’t a v4.0 TLT yet; Olof and I are talking about a couple of different ways to emulate one. UAH v6 works out TLT as a linear combination of TMT, TLS and TTS. So you can construct a TLT using v4 TMT and a different source for the other two series – v3.3 in my case, UAH v6 for Olof. Mixing different data sources is less than ideal, but the fact that we both got essentially the same results, plus the fairly good match with HadCRUT4, makes me happy enough with it as a rough approximation. Hopefully we’ll have the real thing to compare it to soon.

      • MartinM, RSS has published a v4 TTT now, updated through Feb 2016. The full period trend is 0.172 C/ decade, and the trend from 2000 is 0.136.
        There is still some divergence from RATPAC after year 2000. I guess that the new v4 has not fully resolved the NOAA-14/15 divergence, but it is much better now.
        The TLT v4 might be slightly different (when it arrives). However, the TTT is considered to be the best 850-300 mbar emulation AFAIK

  20. A little off-topic, but look what Canadians coast-to-coast-to-coast were treated to this evening on CBC’s flagship TV news program The National, titled “The future of clean-energy”:

  21. RSS brief on the new data (TMT and TTT only for now), with a handful of links.

  22. Tamino, your predicted witch-hunt has begun over at Watts. The topic is “The ‘Karlization’ of global temperature continues – this time RSS makes a massive upwards adjustment,” and some of the comments are vicious.

  23. With all respect to Mears and his colleagues, the continuously shifting nature of the baseline corrections still doesn’t suggest that the microwave-derived atmospheric temperature measurements (or calculations) constitute a stable enough data set to reliably or accurately interpret long-term trends on the magnitude of ~0.1 K/decade.

    This may be inherent in the raw data available from the fifteen or so MSU or AMSU-equipped satellites over the past 37 years, or in the models of emission vs. height used to deconvolve the atmospheric signal, but I think they may be trying to squeeze more juice out of the lemon than it actually contains.

  24. Slightly off topic, but when looking at the satellite temperature record, the big el nino years seem really warm. Could it be not so much that the whole atmosphere is warmer, but more that the warmth moves up to the height at which they measure?
    I’m obviously being naive, but just looking for a way of explaining how years like 1998 really stand out in the satellite record, much more so than temperatures measured at ground level.

    • I will try to find it again and will probably butcher the explanation, but a website/paper showed that during an EL Nino the heat rises far above the equator… like a tall, very narrow belt. During that time the thermometer series measure the narrow belt of warming at the sea and land surface, but the satellites do not see it until atmospheric circulation starts moving it toward the poles. And that is when the satellite’s El Nino spike emerges.

  25. So any guesses on when we may expect the TLT V4.0?

    • Harold Brooks

      David Appell reported he was told (I would assume by RSS people) TLT would be updated in about 6 months.

  26. RSS (v3) plot including Feb 2016.

    While it’s consistent, it’s hilarious that the ‘skeptics’ have been banging on about a statistically non-significant ‘pause,’ only to fret that it’s gone because of a very slight, statistically non-significant warming trend.

  27. Here is a comparison of new and old RSS vs RICH radiosonde data (If you are tired with RATPAC and want an alternative). I have managed to rip the RICH netcdf and produce a zonally weighted global index through 2015.

    The new TTT looks much better than the old TLT, no major divergence after 2000..

  28. Spencer and Christy have blog-commented on RSS 4:

    Some good comments by folks asking about Spencer and Christy’s balloon comparisons in that blog post, for example, Spencer and Christy are using RATPAC-B instead of the officially NOAA-recommended RATPAC-A.

    I’m confused about the bar chart of R^2 balloons versus satellites there. Is it possible that those R^2s reflect the match in month-to-month variation but fail to reflect the match of trends? If so, those R^2s are not the most relevant, given that Eli and Tamino and others had been pointing out the balloon-satellite divergence after the year 2000 is a drift despite continuing good correspondence in peaks and valleys.

  29. Probably a dumb question but: how does one establish statistically that data have no trend? If there is some variability, and the trend line is very close to zero, the uncertainty will include the possibility for either warming or cooling (or not trend).

    Ie, after 100 years of global temp on planet X, we have a trend of 0.02 +/- 0.04.

    I’ll guess that the answer will be that the null hypothesis has not been disproved, but that doesn’t mean there is definitely NO trend. We just don’t have enough information to say anything. Sure looks like a stable series, though.

    If that’s the right answer or close, it makes me think that you cannot use linear regression to determine a flat trend. Is that the case?

    [Response: To quote a wise man, “Proof is for theorems and alcoholic beverages.”

    Suppose an omniscient being told us the data *do* follow a straight line (maybe a flat one) plus normally distributed random noise. Then we know (there’s a theorem, Gauss-Markov) that linear regression gives the best answer. But if the slope comes out 0.000000001 +/- 0.000000004, we can’t be *sure* there’s no trend.

    But we also know its slope is almost certainly very very very small. Probably negligible — meaning, it’s a good idea to neglect it. And we can’t even be certain, just “almost certain.”

    That’s the nature of statistics. Uncertainty is a part of life. We can’t get rid of it. But we can deal with it.]

  30. Commenters Miker and Christian got different R^2 than Christy and Spencer did, for RSS against/with RATPAC and UAH against/witth RATPAC. Olof R explained why RATPAC-A is better than RATPAC-B:

    Christy via Spencer replied: “OK, according to John Christy, it’s because you are using the 850-300mb layer to estimate MT, which is a little apples and oranges. You need to do appropriate weighting of all pressure levels up into the lower stratosphere for MT. If you use only 850-300, you won’t sample the statospheric cooling that MT includes, the warming trend will increase, and so you will get better agreement with RSS because it has a warmer trend.”

    • Well, I have read the dataset descriptions again, and there are more differences between Ratpac A and B, beside the simpler method of global averaging in the B version.
      Ratpac B data is not adjusted for inhomogeneities after 1996 and should be used with caution.
      Ratpac A also uses “raw” data from IGRA after 1996, but the First difference method provide homogenisation when data is merged to regional, zonal and global averages

      There is also a new dataset available, Ratpac v2 beta. Same products and methods, but based on the new more comprehensive IGRAv2 beta database. The Ratpac raw data seems to be more complete/fewer gaps in the v2 version.

  31. RSS monthly anomalies in txt form (land / ocean / land+ocen)..

    Currently v3 and v4 both appear on that page for TMT and TTT.

  32. RSS respond to criticism (including from UAH) of the new data set.

    • Thanks for that link. Over at Climate Etc. Olof did a great job of explaining these issues, which blunted the intent there to spread doubt about the new RSS series.

    • I like the little barb at the bottom of the references on that link:

      “University of Alabama, Huntsville Data: No relevant paper has been published.”

  33. I don’t understand. Isn’t Christy’s approach on matching to RATPAC (i.e., appropriately weighting all pressure levels) the correct one? Do you guys have a rebuttal or should I prefer his R^2s?

    • I’m not following this in detail, but Olof’s comment certainly seems germane:

      If the inhomogeneities in RATPAC-B are serious enough, the weighting function is not the most significant issue. Whether or not that is the case, I don’t know. I suspect there is a case to be made for each choice.

    • Do you know Christy’s method? His weighting?

      Guess not. Christy is a master of hiding what he is doing.

      • You can get RSS weighting functions here; they should be basically identical.

        IIRC, properly weighted, RATPAC-A falls roughly between RSS v4.0 and UAH v6.0b5 in terms of trend, with similar R^2 for both. RATPAC-A has pretty good coverage so, after weighting, makes for a reasonable comparison to the satellite record. Other radiosonde records have relatively poor coverage, and so a direct comparison isn’t ideal; much better is to subsample the satellite record to match the radiosonde coverage, a la Mears and Wentz 2009. Looks to me like Christy and Spencer didn’t do that. Here’s what I get (if I can get the damn images to load this time) for RSS v4.0 and UAH v6.0b5 subsetted to match HadAT:

        And RAOBCORE:

      • This time it was Christy’s weighting of the different radiosonde levels that is hidden. Anyone who has seen them?

      • Martin M. If one focuses on the alleged pause period, e.g. 2000-2015, I got the weighted Ratpac A TMT-trend to 0.18 C/decade, far above the satellites, but RSS v4 TMT or possibly UAH v5.6 TMT must be closest to that.

        The subsampled graph didn’t make it trough. Anyway, Good work. Subsampling is laborious and easily avoided for laymen working with spreadsheets only (like me). Too much work and too little fun.
        Paul S did UAH v6 TLT subsampling here at Tamino’s blog last year.
        I have borrowed and reworked data from his graphs:
        Subsampling seem to ameliorate the trend-break at 2000, but that is mostly an effect of using unhomogenised data from Ratpac B that have a lower trend. Also, I am not sure how Paul S handled drop-out of stations i Ratpac..

        Ratpac B is not a good alternative for subsampling efforts. HadAT (and IUK v2) ends in 2012. However, RICH and RAOBCORE goes through 2015 and can be found here in netcdf:

      • Aaargh. OK, let’s try this once more, without photobucket this time:


        Direct link, in case that didn’t work.

        And RAOBCORE:

        Direct link.

    • Since Cristy has never published his detailed algorithms nor his source code, at least to the best of my knowledge, there is no way to judge this.

  34. I ginned up a couple of graphs based on Tamino’s post some time back running a trend to (Dec) 1997 with extended error bars, and then plotting the post ’98 data. Mine’s very amateurish: the ‘error bars’ are ad hoc and I used monthly anomalies. But I wanted a visual for the latest RSS anomaly based on Tamino’s presentation. Here they are – best viewed in consecutive tabs, clicking from one tab to the other.

    Tab 1 : Tab 2