Global Temperature: Adjusting Data (Cowtan & Way)

I noticed there’s a post at Curry’s in which Frank Bosse uses my method (and my program) to adjust global temperature data for things we know cause it to fluctuate. By removing fluctuations of known origin (or at least, our best estimate of them), we hope to sharpen our view of the changes that are happening for other reasons.

As a starting point, Bosse takes the global temperature estimate of Cowtan & Way. The known factors allowed for are ENSO (the El Niño Southern Oscillation), atmospheric aerosols from volcanic eruptions, and variation in the output of the sun.

El Niño is an oscillation of the ocean/atmosphere system. Every few years or so (it’s quite unpredictable really) it exposes more warm ocean water to the atmosphere, which transfers heat from sea to air. Since the air is where we live, and where our weather happens, we see an increase in global temperature. But it’s only temporary, and it doesn’t increase Earth’s total energy, it just moves some of it to where we notice its effect on the weather. Its opposite phase, la Niña, tends to cool off the globe temporarily.

Volcanic eruptions, especially large ones, can throw massive amounts of junk into the air. If the eruption is explosive to boot, that stuff can get thrown high up there (way up in the stratosphere even) where it can take years to finally settle out of the air. And some of that junk is sulfur compounds.

Atmospheric chemical processes make those sulfur compounds into sulfate aerosols, tiny particles, which tend to be reflective, giving the atmosphere a bright haze. It’s subtle, but measurable, and the bright hazy atmosphere scatters incoming sunlight in all directions, some of it right back to space. Less gets through to reach the ground and warm our world. This does change Earth’s total energy, reducing the power we get from the sun. That tends to cool the planet off.

The sun is, ultimately, the source of the energy that powers our weather and our lives. And it literally keeps the planet warm. If the sun gets hotter the Earth does too, and vice versa. The energy output of the sun is very reliable, but it does fluctuate a little, fluctuations which are sure to affect global temperature.

Running my program to estimate the influence of those factors and remove them, to generate adjusted data which (we hope) show us what else is happening to global temperature, I got this (including data through the incomplete year 2018):

The most obvious effect that the strong peaks from the 1998 and 2016 el Niño events aren’t in the adjusted data. Other events, like the drop in 1993 from the Mt. Pinatubo volcano, are also gone.

Let’s look at just the adjusted data:

Some things are obvious. Talk of a “pause” or “hiatus” starting in 1998 or any other time, is just ridiculous. These data suggest two different episodes, with the trend rate changing in 1978 (from changepoint analysis). We can estimate the trend thus:

Another thing to note is that whether using raw or adjusted data, the hottest four years are the last four years (presuming 2018 will come out in the top 4, pretty much a sure thing). Interesting that using the adjusted data, the hottest year of all wasn’t 2016, it was 2017.

Frank Bosse got a different result from using my program:

Notice that the 1998 el Niño peak is still there strongly, as is the 2016 peak. A clue to why his is so different, might be when he says:

For the C&W dataset I downloaded annual averages for 1950…2016. Thereafter the data was passed the “Tamino- filter” (shown in fig.1 and 2 of the cited post) for eliminating the influences of ENSO, volcano and variations in the solar forcing.

The program can’t run on annual averages. It absolutely requires monthly averages. What I’ve done above is adjust the monthly data, and only then compute annual averages for plotting purposes. Even if he used monthly data, something still went horribly wrong because he got the wrong result. Any conclusions based on it are also wrong.

Frank Bosse has a lot to say. I’m not impressed with any of it, in fact, getting the wrong adjusted values for the Cowtan & Way data seems to me just the start of his problems.

This blog is made possible by readers like you; join others by donating at My Wee Dragon.


16 responses to “Global Temperature: Adjusting Data (Cowtan & Way)

  1. This is my version, based on your code but not exactly the same, & probably using slightly different data for exogenous factors (e.g. OSIRIS SAOD after 2012). It looks pretty similar except for 2016 > 2017—the upshot being that Bosse should’ve gotten something much closer to yours.

  2. El Nino “doesn’t increase Earth’s total energy, it just moves some of it to where we notice its effect on the weather”.
    Is it correct that El Nino’s actually very slightly reduce the Earth’s total energy, since the resulting warmer surface will radiate slightly more energy to space?
    Assuming an average surface temperature of 288K and an increase of 0.2K due to El Nino, that would increase the heat loss by a ratio of 288.2^4/288^4 = 1.0028 (via Stefan–Boltzmann Law).

    • Timothy (likes zebras)

      Yes, I would expect you could see that if you looked in the ocean heat content or top of atmosphere radiation data.

      Note, however, that there are consequent effects on clouds that might prove to be a complication, and add considerable noise.

    • But, you have less upwelling in the eastern Pacific, so the Niño 3.4 ocean surface gets warmer because it’s getting more sunlight and long wave. Much of that warming is happening during the El Niño event.

    • I’d read that yes, El Ninos actually do increase OLR, and thus cool the Earth a bit, even as the surface temps increase a bit. Sorry, don’t have a source!

  3. As well as not correcting for the el Nino peaks, I notice the Bosse version of the “analysis” also didn’t do much about correcting for Mt. Pinatubo.

  4. Bosse got the results he was seeking, no reason to let sound methodology get it in the way of that.

    • I would concur with this view of Bosse’s intentions. Bosse tells us he takes as his start-point Folland et al (2018) ‘Causes of irregularities in trends of global mean surface temperature since the late 19th century’ saying:-

      “Inspired by a recent paper ( Folland 2018), I try to replicate the annual development of the observed global mean surface temperature (GMST) for 1950-2016 with a minimum of fitting and smoothing. Unlike it was done in Folland I don’t want to break the time series into sub-periods nor use monthly data when annual data seem to be sufficient.”

      The idea that annual data “seems” sufficient is a bit odd, especially as the period under analysis is only back to 1950 (as opposed to 1891 in Folland et al. when monthly data would have far bigger Confidence Intervals to negotiate.
      But while Folland did examine sub-periods, this was in an extra analysis. The findings for the full period are set out in their Fig 1. Note the residuals 1950-1913 in Fig 1g – as flat as a pancake. Yet without a blink of incredulity, Bosse finds a big fat wobble, the black trace in his Fig 2. And, as smallbluemike asserts, this is exactly what Bosse was seeking.

      “Some internal variability of the climate system which is missed in modeled approaches as this very recent paper (Kravtsov 2018) shows.” (Kravtsov was attempting to find evidence to support the WUWT – Wyatt’s Unified Wave Theory – so loved by blog-mom Judy Curry.)

      And having found his big fat natural wobble, Bosse even has the bare-faced audacity to speak the big fat lie:-

      “All the characteristics described in Folland (2018) are replicated.”

      So to run with his big fat lie, we can conclude that Bosse is effectively saying what Folland et al said – “No important additional factors are needed to explain the two main warming and three main slowdown periods during this epoch.” And to be clear, the “no important additional factors” includes there being no sign of Judy’s big fat natural wobble.

  5. Thanks for looking at this – I agree with your conclusions.

    I think this is important work, and I would still like to work with you at some point on a sensitivity analysis on the regression coefficients using model data, but at the moment what time I have is more than filled with working on observational and methodological biases.

    [Response: Thanks.]

  6. “Frank Bosse has a lot to say. I’m not impressed with any of it, in fact, getting the wrong adjusted values for the Cowtan & Way data seems to me just the start of his problems.”

    Yah, i getting arround this for years with him ( in the german online forum “Wetterzentrale”), absolutely pointless to talk about.

  7. “Denialist screws up data analysis,” truly a mosquito-bites-man story.

  8. I think you’re misunderstanding what the figure is showing. It’s intended to be a reconstruction of observed GMST using the inverse of your adjustment factor added to an anthropogenic trend + another factor.

    • Y, that correct, he done inverse to replicate the observation, but the same boring stuff all the time just with a new face…

      take this from his post

      “The impact of any forcing on the SST will be much more muted in these regions than in other regions of the globe because the surface warming power is mixed down into and has to warm a much bigger layer of water ”

      On first, seems correct, on second, its incorrect, because especially the NA(as he decribed) is also a result of AMOC and the most important, on mean state, the net energy flow is from ocean to atmosphere, because of AMOC the water is warmer then radiativ equilibrium

      If we also assume that weaking of AMOC is a result of global warming, this regionis one of the most impact due antro. climate change.

      Also the handling of forcing, observed warming is not just a result of fast response in climate system also due thermal inertia slow responses, regional forcing and regional responses.

      Thats Circle running for years,….

    • I did myself do a bit of digging as to what this fool Bosse actually did do. And it was pretty-much as paulskio says it is, except to add that it’s a wholly worthless piece of work.

      (1)The start point was not the annual C&W data which was then “passed (through) the “Tamino- filter” for eliminating the influences of ENSO, volcano and variations in the solar forcing.” Rather the monthly C&W data along with the monthly data adjusted for SOL VOL & ENSO were downloaded from the link at the bottom of this Tamino post. The C&W data & the Adjusted(SolVol&ENSO) data was then presumably annualised.
      (2) The annual Adjusted(SolVol&ENSO) data was then correlated against a the sum of Anthro, this said to be sourced from Lewis & Curry (2018) which actually uses the IPCC AR5 Annex II data (which needs extending from 2011: not a great problem as it is a pretty smooth curve). The method of this “regression” was presumably OLS but when such an operation is carried out, the result is signficantly different from that quoted by Bosse. Bosse says it is:-
      . . . T(anthro) = – 0.33 + 0.335F (+/- 0.031 [2sd]).
      Yet I make it a rather similar but different:-
      . . . T(anthro) = – 0.33 + 0.43F (+/- 0.030 [2sd]).
      Bosse completes his second step by calculating a synthetic T(C&W) by adding back in SolVol&ENSO adjustemnt.
      Using my version, the result is remarkably similar (but not entirely) to the CW(reconstructed) shown in Bosse’s Fig 3, and so do my residuals when compared with the data presented by Bosse in his Fig 2, these graphed residuals which Bosse says come from this step 2 regression.
      (3) Of course, Bosse is actually looking for Curry’s big natural wobble and these residuals do present such a one. It is not as big as Bosse says (” the ampitude of about 0.25°C”) possibly because he doesn’t know what ‘amplitude’(sic) means, or possibly because he is compelled to overstate the max-to-min value+noise. Strippling the noise away, its amplitude is 0.04°C (or 0.08°C peak-to-peak).
      Bosse doesn’t explain how he works the AMO into his analysis. The residuals from step 2 could be greatly shrunken by a correlation against AMO using OLS but this would be littl more than curve-fitting. The AMO is out of phase with the residual ‘dip’ by 5 or 6 years and the ‘dip is better explained by the change in the rate of Forcing which at least has a physical basis. Bosse (& Curry) would probably invoke Wyatt’s Unified Wave Theory (WUWT) which can be used to fit the timing of any aberrant wobble into its grand scheme.
      The correlation factor appearing in Bosse Eq 1 for the AMO (which he synthesises to provide a de-trended version) is 0.105. This is not a million miles from the factor I obtain = 0.162.

      The problems I describe I think demonstrate that the analysis from Bosse is worth nothing.

  9. I read this:

    “the IPCC Special Report on Global Warming of 1.5C – calculates that a 1.5C warming trajectory would require the global economy to cut greenhouse gas emissions by around 45 per cent between 2010 and 2030, before then delivering a net zero emissions economy by 2050.

    Even meeting the Paris Agreement’s upper target of limiting temperature increases to 2C above pre-industrial levels would require emissions to fall around 20 per cent by 2030, before achieving net zero emission status by around 2075.”

    and wondered, has anyone charted/modeled what the annual or monthly co2ppm numbers should look like to achieve either of these paths? I realize there are a lot of variables that impact how/when CO2 emissions fall during the two time frames, but it seems like the kind of task that modeling and mathematics can address.

    When folks post up graphs that track emissions/accumulations, it might be helpful to always have the anticipated numbers that lead to the 1.5 and 2.0 degree outcome that are discussed in policy. It might be more informative than just saying over and over that we are failing to make the changes that limit warming to 1.5 or 2.0 degrees.