RealClimate logo

Predicting annual temperatures a year ahead

Filed under: — gavin @ 16 September 2016

I have a post at Nate Silver’s 538 site on how we can predict annual surface temperature anomalies based on El Niño and persistence – including a (by now unsurprising) prediction for a new record in 2016 and a slightly cooler, but still very warm, 2017.

The key results are summarized in the figures that show how residual variations in the global temperatures (after detrending) related to the ENSO phase at the beginning of the year (defined using the MEI), and the predictions for 2016 (two methods) and 2017.

Note that this analysis was done with only data until July. With the August data included as well, there is a small reduction in the uncertainty (purple bar), but the mean prediction is basically unchanged. I think this might be the first prediction for the 2017 annual mean specifically – we’ll have to wait ~18 months to see how it pans out.

I’m happy to get deeper into the analysis here if there is interest. Additionally, I’d be interested in ideas and/or suggestions for climate issues that might fit the 538 style for followup posts.

16 Responses to “Predicting annual temperatures a year ahead”

  1. 1
    Bruce Tabor says:

    It’s a relief to see you using 1880-99 as a baseline, which is effectively pre-industrial. 1951-1980 is common, but I’ve seen others – it get’s very confusing. The change in temperature since pre-industrial times is the issue of greatest public interest.

    Is it possible to STANDARDISE THE BASELINE? 1880-99 would be ideal. I realise temperatures are less accurately known in that period, but the uncertainty could be removed by “fixing” it relative to a more accurate baseline like 1951-1980. Or having a best estimate of 1951-1980 relative to preindustrial (e.g. +0.4 Celsius).

  2. 2
    Randall R. Besch says:

    1880- 1899 is hardly “preindustrial”. The industrial age for the modern world started in the 1500’s. Use of wood and coal maybe earlier. Ever since the use of machines over animals and people when coal was dug up and burned one could link it to that. 1951 is when electronics were coming into greater use with the creation of circuits in 1947 to replace vacuum tubes. It may have taken a few hundred years to build up the GHG’s from growth of population and fossil fuel energy. We do know that the 1940’s – 1970’s created a temporary cooling was due to World War I & II added millions of tons of debris into the atmosphere that cooled things till the end of the 1970’s.

    Perhaps I am just not reading it correctly. Pardon me if I am way off here.

  3. 3
    Eric says:

    For those of you who do not look at least occasionally at the online comic xkcd, be sure to look at the temperature trend line he published at Ok, a bit off topic but it will be useful for those who think this is all a joke.


  4. 4
    Cameron says:

    I think an objective analysis of model predictions and observations over different time periods might fit the 538 website quite well

  5. 5
    Silk says:

    #3 – Someone should show that to Malcolm Roberts.

  6. 6
    wili says:

    Many great points made here. The title of the 538 piece is a bit too optimistic, isn’t it? (And I realize that writers don’t always get to pen the titles to their own pieces).

    “Why We Don’t Know If It Will Be Sunny Next Month But We Know It’ll Be Hot All Year”

    But we don’t ‘know’ that for absolutely certain, right?

    Major volcano activity could throw off that otherwise reasonable prediction, as I understand it. I notice that you do mention throwing out years with major volcanic activity when establishing the underlying trend in the article, but perhaps there should be a clearer caveat stated wrt how certain we can be about claims of annual predictions in the title?

  7. 7
    Jim Eager says:

    We do know that the 1940’s – 1970’s created a temporary cooling was due to World War I & II added millions of tons of debris into the atmosphere that cooled things till the end of the 1970’s.

    No Randall, we don’t know that at all. Particulate and droplet aerosols only stay in the troposphere for a few months to a year before being precipitated out unless replenished, and in the lower stratosphere only 1-3 years, as the 1991 eruption of Mt Pinatubo demonstrated. So how could WW1 and even WW2, neither of which injected aerosols into the stratosphere, have had any effect into the 1970s?

    Far more likely to have caused the flat temperature period of the mid 1940s to 1970s was the truly enormous post-WW2 industrial boom in the US, Europe, Japan, and other parts of the developed and developing world that continuously pumped out steadily growing amounts of sulfur dioxide, black carbon soot, ash and dust from every chimney and tail pipe 24/7/365, year after year without any kind of pollution controls. It was like Mt Pinatubo popping off every other year or so for 35 years, totally suppressing the warming caused by the also steadily rising CO2 from those same chimneys and tail pipes.

    And then the Clean Air Act in the US and similar legislation in Western Europe was passed and went into effect and rapidly reduced those sunlight reflecting and blocking aerosol emissions, unmasking the growing warming produced by steadily rising CO2 that was there all along.

  8. 8
    Tim Havard says:

    Would it not be possible to calculate a factor that describes atmospheric pollution per capita thus integrating population growth with industrial expansion(pollution) rather than an arbitrary date. This maybe would stabilise the so called base line.

  9. 9
    Mal Adapted says:


    a slightly cooler, but still very warm, 2017

    I can hear it now: “Global warming stopped in 2016” 8^(.

  10. 10
    gallopingcamel says:

    I would be interested to hear what Gavin Schmidt has to say about this:

    It looks like a “Peer Reviewed” paper showing that temperature has not increased in 20 years in spite of a relentless rise in the Keeling curve.

  11. 11
    Chris O'Neill says:


    showing that temperature has not increased in 20 years

    Temperature has increased at the rate of 1.9 degrees C/century in 20 years.

    I think that paper might have been talking about something other than temperature increase in the past 20 years.

  12. 12

    An obscure paper by Glenn Brier shows that ENSO has deterministic properties back to the year 1525

    Interesting that no one pursued this discovered periodicity and instead skeptics such as Tsonis have seemed to push the idea that ENSO is completely unpredictable.

    The paper by Astudillo et al has also finally been published. This has some clever signal processing substantiating that ENSO is determinsitic

    Astudillo, HF, R Abarca-del-Río, and FA Borotto. 2016. “Long-Term Potential Nonlinear Predictability of El Niño–La Niña Events.” Climate Dynamics, 1–11.

  13. 13
    Silk says:

    #10 – It’s trivial to show that HadCRUT does not show that global mean temperature declined between 2003 and 2014 as the paper claims. will demonstrate that for you.

    (It’s also trivial to show that trying to extract a trend in satellite data over such a short time scale is a pointless exercise)

    Pleasingly (from a scientific, rather than human race, point of view) all three satellite datasets show a strong warming signal if you look at the datasets starting in 1980 through to the present day. And I see no sign of any hiatus.

  14. 14
    Interested reader says:

    That’s great. Can’t wait to be in 2018 and check this out!

    Is such prediction, if accurate, very useful besides demonstrating the scientific capability?

  15. 15
    barry says:


    HadCRUt does show a cooling trend if the reference period is 2003 to 2014 (Dec 2013), as hockeyschtick posted. Try it out.

    hockeyschtick doesn’t realize that the authors of the paper mean Jan 2003 – Dec 2014. The HadCRUt4 mean trend for that period is slightly warming (with considerable uncertainty).

  16. 16
    TPP85 says:

    That is convincing and a very interesting tool. Of course, the climate part is the long-term trend which has the greatest weight on the absolute value. The famous 1998 lost much of his fame now. And we are even not sure to see again a value like the 2010 record. Simply the basics of the expectation from the 90s. The last decade was useful for.. what?