Now that 2015 has blown away previous record-hot years, the global warming deniers are scrambling to blame it on anything but global warming. Their favorite candidate is something that does in fact make Earth’s surface get hotter, something that really did contribute to 2015’s record heat: el Niño.
But how much? A post at Carbon Brief addresses just that question. Their conclusion is that el Niño contributed only about 10% of the record.
Much of their conclusion is based on soliciting opinions from experts in the field. One of those, Gavin Schmidt at NASA’s Goddard Institute for Space Studies, estimates el Niño caused 0.07 deg.C extra warmth in the NASA data. He based his estimate on compensating the data for its el Niño influence, yielding “el Niño-corrected” or “adjusted” data:
It’s an effort I’ve been involved in for a few years now. I’ve compensated for multiple factors, including el Niño, atmospheric aerosols (from volcanic eruptions), and fluctuations in solar output.
Of course the adjustments don’t compensate for these factors perfectly, so lately I’ve pondered some improvements. For one thing, it’s possible that el Niño has an impact which lasts longer than the el Niño itself. We know there’s a lag between el Niño and its effect, but perhaps there’s more than one lag at work. Therefore I’ve allowed for el Niño to have an impact at two different lags (I’ve tried three but that didn’t seem to improve things beyond two).
For another thing, the impact of el Niño may not be linear. I’ve suspected that perhaps a strong el Niño has more effect than a weak one, stronger even than the strength of the el Niño itself. This intuition was based on the fact that models which include an el Niño effect don’t seem to match all of its influence — they don’t seem strong enough when the el Niño is strong.
Finally, there may be a seasonal pattern to the influence of el Niño. Back in 2013 I discussed the research of Kosaka and Xie, and emphasized that one important result was their discovery that the influence of el Niño depends on season. I even closed that post by saying “the regression approach of Foster & Rahmstorf … might be improved substantively simply by allowing for a seasonal pattern in the influence of el Niño..”
So, I’ve added these elements to the mix of factors by which el Niño can influence global temperature. The best model I’ve found so far (there’s a lot more to test) involves a linear el Niño effect which lags only 2 months behind the el Niño itself, a nonlinear el Niño effect which lags 10 months, and a seasonal effectiveness of the el Niño impact. In agreement with the research of Kosaka and Xie, the el Niño impact is strongest in northern-hemisphere winter and weakest in northern-hemisphere summer.
With all those in play, with volcanic aerosols and solar fluctuations to boot, the model compares to the observations thus:
It’s particularly interesting to look at annual averages, of the observations, the model, and the adjusted data:
The model doesn’t just match the observations well, it accounts for over 70% of the variance of the data from a steady linear increase since 1970.
We can also plot the impact of el Niño on each year’s temperature:
My result indicates that el Niño led to 0.08 deg.C warmer temperature in 2015. That’s hardly enough to explain the record heat, which was mainly due to global warming. Note, however, that el Niño caused fully 0.2 deg.C warming in 1998, so the record heat of that year — which the deniers love to point to as the “end” of global warming — really was due to el Niño.
We can even look at the el Niño influence on a month-by-month basis:
This illustrates that el Niño is the main reason that the last few months of 2015 were so much hotter than the preceding months of 2015.
For 2015 as a whole, el Niño contributed in a small way to its extreme heat. But the main factor was the continuing trend. That’s due to man-made greenhouse gases, and it’s called global warming.
If you like what you see, feel free to donate at Peaseblossom’s Closet.
So you can predict at least 2 months in advance what the GISS monthly anomaly will be? More if you have an accurate prediction of ENSO values.
How about adding PDO to the regression?
[Response: Yes, one could predict two months in advance based on the model. As for PDO, its correlation with the residuals is practically nonexistent, which argues against its being a palpable influence on global temperature.]
The best model I’ve found so far (there’s a lot more to test)
Out of interest, how do you test your models?
[Response: Mainly I examine the model coefficients and their standard errors (corrected for autocorrelation). I’m working up to using AIC, but it’s complicated because the likelihood function also depends on autocorrelation.]
I guess cross-validation would also be complicated by autocorrelation? (and by the nature of time series in general…since you can’t randomly split the data into build & test folds…or can you?)
While the stats work out nice, have any of your physicist buddies suggested to you why the model might be physically reasonable as well?
The longer lag on the second component might be involved with the NH winter when much of the moisture raised by El Nino condenses out of the atmosphere. This also would explain why people (including me) were talking of 4-6 month lags for ENSO influence in recent years. This model might make more physical sense than the previous.
I am going to get the terminology wrong. The extra degrees of freedom added by adding 10 month lag, non linearity and seasonal effects, my wet finger thinks they are probably fine for best guess estimates. (AKA if I was forced to make planning decisions, Id use them as the best available estimate.)
Do they pass statistical tests. Hind/fore/casting or residual based?
[Response: Out-of-sample testing is what I’m working toward. But I’m in the early stages, everything so far is just a preliminary.]
Just how high will the latest spike go?
Could you make a prediction for the coming months based on what you have just now, Tamino? Qualifying that it’s based on an unfinished thesis, of course, and it could be updated. It would be interesting to watch that play out, and for you to comment on what you learn from it.
[Response: Right now I’m still trying different models, and I’d like to do more of that before moving on to other things. But I’ll get to it … patience, patience.]
Maybe the AMO?
One minor point: 2015 corresponds to 1997 regards the evolution of the el Nino and 2016 will likely correspond to 1998 so perhaps one should compare 2016 with 1998 in order to assess the relative importance of the el Nino. Just a thought.
I may have a useful piece of information to add.
A while back I was looking at the El Nino signal in the SST data, divided by longitude into Ocean basins: E Pacific, W Pacific, Indian and Atlantic. If I remember correctly, the El Nino temperature response appears first in the E Pacific, than weakly in the Indian, then after about 10 months in the Atlantic. It’s a long time ago, so I may have the details wrong. I haven’t done a literature search to check if this is well known.
[Response: The only research I’m aware of on the topic is Kosaka and Xie, who specifically mention a seasonal pattern to the el Nino effect but don’t (if I recall correctly) consider multiple lags. And, I’m really not up on the literature on the subject.
It sounds as though you’ve identified some possible physics behind the multi-lag nature. Seems to me that this is an avenue very worth exploring.]
I’m curious if you guys are talking atmosphere, SST, or both. I do not know where it falls in the order, but eventually I think there is usually a cooling of the Western Pacific SST.
Trenberth often says the 97-98 El Nino saw a reduction in OHC. It seems to me that has not really happened with El Nino events since then.
“As for PDO, its correlation with the residuals is practically nonexistent, which argues against its being a palpable influence on global temperature.”
How many confident (denialist) predictions of cooling based on PDO have I read over the years? Well, enough that I was momentarily startled to read the comment quoted. But the purported correlation doesn’t look very good on the old WTF ‘sniff test’:
I suspect that comparing residuals makes much more sense, but still.
Because it losing steam when they made their predictions, and gained steam while they were waiting for their predictions to come true, and they never will.
JCH, looks like it’s pretty difficult to predict. While there appears to be some physical basis for the wiggles, it’s barely through two cycles in the instrumental record. I’d guess that’s not nearly enough data to make a prediction or gauge periodicity.
Here’s the same data sets from 1979, picked for the satellite era.
Barry – I’ve posted versions of that at Climate Etc. dozens of times. It was part of that basis for why I guessed the PDO was about to go positive.
>”the El Nino temperature response appears first in the E Pacific, than weakly in the Indian, then after about 10 months in the Atlantic.”
The effect certainly moves around to different locations with different lag times. When I tried this a long time ago now, I found rather that instead of temperature lagging MEI by a certain period like 3 or 4 months it seemed better to adjust temperature based on MEI summed over something like 1 to 7 or perhaps even more months previously. I wasn’t sure that made logical sense thinking different regional variations would tend to cancel out and there would be one optimal time lag. However, now you mention your non linear effect, I am wondering if a strong El Nino can tend to cause impacts in more regions at once. If there was such an effect might a longer period of adding ENSO index data make some logical sense? Of course it is more parameters and might be prone to overfitting, (it wouldn’t surprise me if the better results I got were just overfitting). Wondering if you have or might try such smeared out lag periods?
>” a linear el Niño effect which lags only 2 months behind the el Niño itself, a nonlinear el Niño effect which lags 10 months, and a seasonal effectiveness of the el Niño impact.”
Do the satellite temperatures have similar lags and what do those adjusted data look like? I guess it would be too much to hope they might show downward steps at times of satellite changes?
Hi tamino, this is a long-time-reader again and his one-question-per-year.
I remember an other method you explained (Kosaka and Xie 2013) to account for the influence of ENSO (obtained with historic climate forcings and constrained pacific tropical SST), witch showed a pronounced recent ENSO-induced cooling until last 2000s.
It seems there’s a real divergence between these two methods, since this cooling appears much less pronounced here, but I can’t figure out why.
So maybe someone could help and give me a clue : did I miss the point or is an intermediate level stuff finally needed for me ?
So I guess we have 0.07 K for El Niño and less than 0.03 K from global warming since 2014. That leave 0.06 K from other internal variability to get to the difference of 0.16 K in the temperature from 2014 to 2015.
I think the important totals are 0.13 K from internal variation and 0.03 (or less) from AGW. Of course, the AGW part accumulates while the internal variation averages zero.
JMA yesterday finalised on the annual global anomaly for 2015.
0.42C over baseline – a shade higher than their preliminary result (0.40C), 0.15C warmer than last year’s record anomaly and 0.2C warmer than 1998 in that data set.