Disentangling greenhouse warming and aerosol cooling to reveal Earth’s climate sensitivity (T. Storelvmo, T. Leirvik, U. Lohmann, P. C. B. Phillips & M. Wild; Nature Geoscience 9, 286–289 (2016) doi:10.1038/ngeo2670) doesn't seem to have garnered much attention. I glanced at it, I think, thought "that told me what I thought I knew already", and thought no more. Life is so much simpler when you need think no more. But then a correspondent who wishes to remain anonymous (and before you start guessing, no, it isn't JA, he is quite forthright) offered me some thoughts, and I thought I'd share them with you.

I notice that subsequent to me typing this in but prior getting round to press submit, there's another Storelvmo paper out, though she isn't the lead of that one. JA wasn't impressed.

Essentially, what the paper is trying to do is estimate TCS by trying to take account of the cooling affect of aerosols. Which is does by taking the GEBA dataset and throwing at it a pile of "Statistical methods commonly applied to economic time series"; or as they say *Observations from the ∼1,300 surface stations considered were used to estimate the free parameters of a set of equations predicting temperature at individual stations as a function of CO2,eq and DSRS, using dynamic panel data methods that allow for potential long-run cointegrating links among component global time series* yadda yadda. Or as my correspondent put it, and I paraphrase, the novelty of the study is use of statistical techniques from econometrics which are unfamiliar to climate researchers... the climate research community has not in general understood the paper... the techniques are totally inappropriate in this context.

So, what do we have? "The analysis further yields a best estimate of the TCS of 3.1 K over land, with a 95% confidence interval of 1.7–4.4 K"; this is fair enough on its own terms; then "Given that land has warmed at exactly double the rate of the ocean over the past century" they deduce that "TCS for the entire globe is estimated to be ∼2.0 K (95% confidence interval 1.1–2.9 K)", by setting the bit-due-to-ocean-fraction to half that due to land. That seems a bit of a stretch, although not obviously wrong. However, back in the abstract, the TCS is "(2.0 ± 0.8 K)". This is also numerically different to their text. I've never been too hot on such things, but I gather that those who care about such things do care about the difference between CI's and +/-'s as well.

Also, I found their eqn 1 puzzling. In it, they define TCS as F_2x * delta_T / delta_F. F_2x is the forcing due to doubling CO2. delta_F is the "net RF" and delta_T is "change in global mean temperature". But they've already defined TCS with the conventional defn of "the temperature change that occurs at the time of CO2 doubling" so they're not allowed to re-define it. Presumably they mean that eqn 1 may be used to estimate TCS using present-day observations.

If you look at their fig 1, you'll see that the correlations between SO2 emissions, and DSRS (Downward Solar Radiation at the Surface), is quite high: R(1964−1985) = −0.83, R(1986−2010) = −0.78. It isn't clear whether that is important or not. Certainly, they say they are relating DSRS directly, not via SO2. However, on the off chance that it does matter: notice that they don't give the correlation for the full period. I'm told that if you work it out for yourself for the full period, its an unimpressive and certainly unsignificant -0.2.

### Refs

- Log in to post comments

An econometrist tried to make me look into cointegration. It sounded like a neat trick, but also surprising when it would work. Thus I am curious about why "the techniques are totally inappropriate in this context."

“Given that land has warmed at exactly double the rate of the ocean over the past century”The models give a factor 1.6. For most latitudes this factor fits in the observations. The land in the Arctic warmed a lot more than expected (which could also be natural variability) and the Southern ocean did not warm that much (which is where I trust SST the least).

I would not dare to say at this moment whether 1.6 or 2 is the right factor.

Did your Deep Throat elaborate on why the techniques were inappropriate?

I don't think nitpicking the specific application of econometric methods would be very fruitful here. The reality is that observed DSRS changes are large. If you take these to be indicative of aerosol forcing you're going to find a large negative aerosol forcing (see Cherian et al. 2014 for example), and that would suggest a relatively high climate sensitivity in an energy balance model. If the study failed to find a high sensitivity from these data that would suggest a problem with the method. It could be getting roughly the "right" answer by accident I guess.