Saturday, September 6, 2014

Fragility of the "pause"

This post relates to a technical meaning of the pause, often dwelt on on skeptic sites. Some number of years for which the global temperature trend to present has been negative or zero. Of course the pause itself should be more broadly defined as a period where the trend is substantially less than some expected value. But the zero trend seems to attract people, so I thought some prognosis of it might be interesting.

Actual zero trend tends to get mixed up with trend not significantly different from zero. I'll stay away from the latter as I think it is a misuse of statistical significance (SS). If you have SS, you can infer something. If you don't, you can only infer that a test has failed. Maybe too much noise; maybe an inadequate test.

Werner Brozek runs monthly articles at WUWT. He looks at a variety of indices, and notes the number of years of zero trend. He also looks at SS tests, which I think are misplaced. Anyway, something is happening there. The pause given by most indicators is shrinking.

Lord Monckton runs monthly posts with titles like Global Temperature Update – No global warming for 17 years 11 months. He is always referring to the MSU-RSS index, which is not surface, but lower troposphere. Dr Spencer, who manages the other LT index from UAH, wrote about how UAH and RSS are diverging, and advised:
"But, until the discrepancy is resolved to everyone’s satisfaction, those of you who REALLY REALLY need the global temperature record to show as little warming as possible might want to consider jumping ship, and switch from the UAH to RSS dataset."
Lord M is following that advice, as we shall see. Indeed, from 18 years ago to present, RSS has zero trends. But as you'll see below, all other indices have trends from 0.5°/Century to 1°/Century. No 18 year pause there.

Anyway, I took a number of indices (most sources, graphs and some tables here), and plotted for each index the trend from time x in the past to now (July 2014). I've plotted the last 18 years, to match Lord M, skipping post-2012 since short trends are large and variable and mess up scales. Here is a resulting plot:

The skeptic convention is that the pause goes back to the earliest crossing of the x axis. So for example HADCRUT 4 would be "paused" since about 2001, and you can see Lord M's 18 years for RSS. You can also see why he likes RSS. It really is an outlier. Interestingly, UAH is almost an outlier in the other direction, with a pause of about six years. The surface measures are fairly consistent.

The ups and downs of the curves follow peaks and valleys in the temperature curve itself, and I've shown faintly the UAH time series near the bottom (12-month centered running mean). A high temp lowers the trends in later times. To get a broader view, I'd recommend the Moyhu trend viewer. Here for example is HADCRUT 4 - in the original you can click on both the triangular plot and the time series to find out various numerical information. The top corner looks like this:

The right end is where trends ending at present are found, and as you go down, the starting point gets earlier (difference in years shown on axis). As you go left, the end point gets earlier. The diagonal shows trends of 1 year duration. So the plot above represents colors along the right edge. Brown represents zero trend, and you can see a bluish (negative) area bottom right, where the brown boundary tangles with the edge. The brown points on the edge correspond to the crossing points (red HAD4 crossing the x xis) in the graph.

You'll see something similar in other plots that you can get by pressing buttons. The key thing is that we are reaching the edge of that region. It will soon be left behind, and won't leave any brown on the axis. The pause will contract dramatically.

Here is a movie plot that shows that. I have plotted how the trend plot would look if you started in March 2014, April etc. And I have padded the future with reflected temperatures - August supposed to be the same as July, September =June etc. This enables us to see how the arithmetic pans out if the present warmth continues. There are in fact data for UAH and MSU-RSS for August, which I have used (both went down). Click the buttons at the top to flick through.


You'll see from March onward, all the plots are moving up, so the pause has tended to shorten. Not so much for RSS, though, as Lord M has been repetitively noting. Now the interesting thing is that projecting through to November, all the indices except RSS have cleared the axis, and UAH still has the 2010 dip. Even HADCRUT clears, though only just. No more pause at all for surface indices!


  1. I'm just a layperson who follows the climate blogs. This is how I look at it. The strongest trend is to 2010.33. It's stout. So I do not see any meaning in pauses that extend far back from 2010.33 as the earth was warming then at a solid pace. What I have read was the 2nd strongest La Nina in the record did a snot knocker on the trend starting in 2010.33 (aqua trend that dives down into the 2012 PDO trough), and despite the fairly aggressive warming since 2012, the trend trend since 1993.58 has drifted lower since. Imo, the pause will be dead, paws up, when trend from 1993.58 starts increasing. ~19-year trends

    1975 to present: 0.0166246 per year (not graphed)
    1993.58 to 2010.33: 0.0198242 per year (red)
    1993.58 to 2012: 0.0163056 per year (green)
    1993.58 to present: 0.0141508 per year (blue)
    1993.58 to 2023.58: ? (won't be surprised if it exceeds the red trend to 2010.33)

    1. Using the RCP8.5 CMIP mean at Climate Explorer, trends for 1993.58 to ? only exceed 1993.58 to 2010.33 in 2029. Under the other, more moderate, scenarios the trend is never exceeded. Of course, that doesn't take into account internal variability.

  2. Hi Nick,
    Seems to me that the measured warming consistently falls at about half the IPCC projected warning. Maybe if everyone could agree to discount half of the IPCC warming, then rational public policies could be enacted.

    Steve Fitzpatrick

    1. Yes, there has been reduced warming. Personally, I think it is a cool stretch - La Nina's etc. At some stage, it may convincingly be significantly less than models show. But not yet.

    2. What would make the rate of warming "convincingly be significantly less than the models show"? (Serious question.)

      I would agree the models' projections of warming are likely correct if there were an increase in rate of warming to well over the model mean (to compensate for the recent shortfall in warming); say an average of 0.3C per decade for a decade and a half. What objective criteria would convince you they are wrong?

      Steve Fitzpatrick

    3. "What would make the rate of warming "convincingly be significantly less than the models show"? (Serious question.)"
      Steve, it's studied in papers like this. Even the sort of analysis that Lucia used to do, though I don't think she included all the variability.

      There's a bias in those studies, though. They are only done when the trends has been low. The question is posed - what is the probability of such a small trend by chance? There's a test for that if you ask once at random. But if you watch the record, and only test when prospective, that's harder to evaluate.

    4. Nick,

      I know that there are lots of studies which examine the now rather obvious divergence between models and reality. The problem (which you seem to hint at) is that those studies offer potential explanations (some would say "excuses") for the divergence only after the inaccuracy of model projections is clear. The proposed explanations are susceptible to bias, and that bias will always be in the direction of preserving the dominant paradigm (as Kuhn might have noted). If you look at the list of explanations in the paper Judith discussed at your link, nowhere is what seems to me the most obvious explanation: the models have large errors in how they treat clouds, so they are too sensitive to forcing. I have little doubt that after another decade with lower than projected warming, we will still be seeing papers with new explanations, and none of them will conclude the models are too sensitive to GHG forcing.

      That is why I asked you for objective criteria for judging model failure.... post hoc studies which offer explanations for model inaccuracy are, IMO, very unlikely to ever be remotely objective, and so are essentially without value in determining the credibility of current model projections. If people want to base public policy on the sensitivity to forcing diagnosed by GCMs (and many people clearly do) then the accuracy of that diagnosed sensitivity must be verified via prediction, not post-hoc explanations for failed past projections. The argument on offer is: "Yes, the earlier projections of warming were too high, but we are sure that was just due to 'sdfkljlkfsdvnad kfbvadigvfbh' and/or 'weopifeqrlgkja soasja ogj', maybe in combination with the influence of 'peefjhvalsdffhqliy', so there is nothing really wrong with the models, and we should enact public policy based on their current projections." This argument would be funny were the stakes involved no so high. Preposterous seems a better adjective.

      So I ask again: Is their any objective criteria of inaccuracy in model projections that would convince you the models are just wrong?

      Steve Fitzpatrick

    5. Steve,
      Fyfe et al is titled "Overestimated global warming over the past 20 years". So they are obviously looking at ways models might be wrong. The brief abstract says:
      "Recent observed global warming is significantly less than that simulated by climate models. This difference might be explained by some combination of errors in external forcing, model response and internal climate variability."

      I think models get a lot right. I often point to movies like this to show what CFD plus forcing and topography etc can do. But yes, the cloud treatment, for example, could be out by quite a bit.

    6. Nick,

      I'll take that as a 'no' answer to my question

      Steve Fitzpatrick.

    7. Steve,

      I'll give you a simple answer for "an objective criteria" (sic):

      If oceanic heat content ceases to rise for increases in CO2 I would regard the models as having failed.

      This is the primary prediction of the climate models: increasing CO2 concentration leads to global warming, predominantly in the form of oceanic warming. Direct evidence from ARGO and indirect evidence from sea level rise since 1979 confirm this prediction. Secondary predictions of the models concern the partitioning of this heat between ocean, atmosphere and ice melt, and here the models have shown shortcomings. Atmospheric warming has been less than expected, while ice melt has been greater than predicted. For instance, the current accelerating trend in melting of Antarctic land ice, far greater than the current increase in Antarctic Sea Ice, was not predicted by the models. How do you think policy should be modified in the light of this?

    8. bill: or instance, the current accelerating trend in melting of Antarctic land ice, far greater than the current increase in Antarctic Sea Ice, was not predicted by the models.

      This idea isn't dissent free. The main evidence for an accelerating trend comes from GRACE, but GRACE has very large systematic corrections that need to be made (if I remember correctly, even the sign of the trend comes from the correction of the data).

      A couple of related papers ICECAP suggest that mass gains exceed loses.

      An argument that the ice loss shown by GRACE are too large.

      So this may be a case where the models are getting it closer to right than (some of) the measurements. My own guess is, even if the rate of ice loss in the interior suggested by GRACE is correct, it will not be a sustained ice loss. As you know, most of the surface of the Antarctica is well above 1000 meters, and expected to stay well below the freezing point of water.

      So to me It seems hard to come up with a model that can explain accelerated mass loss in the interior. It seems more plausible to me that better measurements and analysis will not bear out this result.

    9. There's also this recent analysis suggesting that the earlier trend estimates were done wrong. When they correct for autocorrelation, they find:

      Despite the higher uncertainties, we find significant (2-sigma) accelerations over much of West Antarctica (overall increasing mass loss) and Dronning Maud Land (increasing mass gain) as well as a marginally significant acceleration for the ice sheet as a whole (increasing mass loss).

      But this doesn't address whether the GRACE systematically-corrected numbers are bias free, of course.

      The Antarctica is a pretty big land-mass compared to Greenland of course. So this does (to me anyway) raise the question of whether GRACE has enough fidelity to accurately measure the rate of ice loss in Greenland.

    10. Thanks, Carrick. These are good references, but I'll have to give them considerable study.. The Zwally et al. paper suggests that Icesat and GRACE are actually in pretty good agreement for Greenland despite your suggestion that the GRACE data might not be reliable here.

    11. Bill Hartree,

      Seems you miss my point. Yes, GHGs cause warming. Yes, there is accumulation of heat in the ocean (How could there not be with the measured surface warming since 1900?).

      Neither of these things tell us about the accuracy of model projections. Nor does measured ice loss (especially in light of Carrick's comments on uncertainty in GRACE data). Models have consistently overstated warming AND overstated ocean heat accumulation. That appears most consistent with lower sensitivity to forcing than the models diagnose.

      Steve Fitzpatrick

    12. bill there are actually a range of values given for GRACE for Antarctica by different researchers. I believe Zwally is arguing for the lower end of the range.

      I believe this is a case where you don't really need a full-blown global circulation model to describe the accumulation and loss of ice mass. This is just my prejudice--but I think it should be a relatively robust result.

    13. There's another paper by Matt King here which also pushes for lowered estimates of ice loss from the Antarctica.

      Again the problem isn't GRACE, per se. I think it's an awesome satellite and does a great job doing what it was designed to do.

      The problem is with using GRACE to estimate trends.

      If the issue is that important, and it could be argued that it is, having several satellites in orbit at the same time would greatly reduces systematic uncertainties associated with trend estimation. Fortunately they are doing just that with the GRACE Follow-on missions.

    14. Steve F. "Models have consistently overstated warming AND overstated ocean heat accumulation."


    15. I should clarify: reference for the second part of this statement would be v. interesting

    16. "What would make the rate of warming "convincingly be significantly less than the models show"? (Serious question.)"

      Judging from the magnitude of natural variability, I would suggest a difference in the trend of the global SAT over the last 30 years of at least 0.15 °C per decade, if the models, the data, and the dates are not cherry-picked and if, very important, the model was run with the same forcing as occurred in the reality.
      To detect smaller differences in the trend would require even longer periods then 30 years.

  3. "Even HADCRUT clears, though only just. No more pause at all for surface indices!"
    But there is also BEST. Although they have no monthly updates, what would the trend compared to the other indices?

    1. BEST and Cowtan/Way are rarely cited as evidence of pause, for obvious reasons. I haven't checked in detail, but I expect they are well clear of the line.

    2. There's not much distance between Cowtan & Way and HadCRUT unless you cherry pick the interval.

      Have you thought about generating one of your "trend-grams" for C&W?

      It'd be interesting to compare it to HadCRUT for example.

    3. Carrick,
      Yes, I will do C&W. I did check BEST (land) there, and I wasn't quite right. The trend is up through the early 2000's etc. But BEST reaches a marked peak in 2007 (as does CRUTEM) so periods going back up to a decade up to a decade that include 2007 tend to have negative trend.

    4. Carrick,
      I've added C&W, BEST LO, and NOAA SST to the trendviewer. I'll blog about it soon. C&Wkrig just touches zero trend if you start in 2005, and it goes quite negative if you start near 2010. But it's decidedly positive if you start before 2005.

  4. Steve F. "Seems you miss my point"

    Don't think so, old sport. You challenged Nick to name an objective criterion that would demonstrate that climate models were wrong. I decided to accept that challenge by naming a lack of OHC increase for continuing increase in CO2 concentration. How does that miss the point?

    1. Bill Hartree,
      Why not instead say that the models would be wrong if the sun did not rise tomorrow? In light of the measured surface warming over the past hundred years, lack of OHC increase would be similar to the sun not rising. OK, so there is no failure in forcast that would convince you the models are too sensitve to forcing. Better if you Would just say that and be done with it.

      Steve Fitzpatrick

    2. Steve,

      I thought your point about "the models running hot" was their failure to predict the "pause" since the late 1990s, so all this stuff about a century of surface warming seems beside the point. My point is that continuing OHC in conjunction with continuing increase in CO2 concentration is the primary prediction of the models and is what is seen. The secondary predictions concern how the heating of the top 100 m or so of the ocean is then partitioned to: (i) the deeper ocean, (ii) ice melt, (iii) the atmosphere ( which would give the "surface" warming whereof you speak). At the moment the models are tending to overestimate (iii) and underestimate (ii). So:
      Primary prediction: verified. Secondary predictions: not verified.
      Conclusion: models have limited predictive success, which is rather different from your statement that they are "just wrong".

  5. Steve F., One other thing, I'm still waiting for a reference for your assertion:
    . "Models have consistently overstated warming AND overstated ocean heat accumulation."

  6. Carrick: having looked into the references that you mention, I note none of them mentions Cryosat. Like Icesat this estimates land ice coverage by altimetry, but has much better coverage and resolution. The cryosat data actually give a rate of ice loss GREATER than GRACE, though the two agree within uncertainty. So, GRACE no longer looks like a high outlier as has been suggested in the past. The "dissenting" arguments based on Icesat now look considearbly weaker.


    1. Bill, thanks for the reference. It'll be interesting to see if this survives scrutiny.

    2. Carrick, you're very welcome. Curiously enough, I discovered the Shepherd et al. paper via a post by, of all people, Anthony Watts. He actually accepted the conclusions of the paper, but argued that the effect of the melt would not give significant sea level rise. The fact that it also undermined the whole "increasing antarctic ice" story was clearly lost on him.