Historical contributions to warming: what, who, where?29 Dec 2020
Reasonably recently, I noticed a tweet (it’s always twitter…) asking for an estimate of the contribution of agriculture to global warming. There were a few responses, but they had quite a wide spread and in general seemed inconclusive. I decided, quite some time after seeing the tweet, to see if I could give my own estimate of this using the probabilistic simple climate model ensemble I used in my previous post. As is the way with these “straightforward” projects, it naturally turned into something a bit more comprehensive than intended. This post is the result - my attempt to estimate what (specific forcing drivers), who (sectors) and where (countries) is causing anthropogenic global warming.
NOTE: I corrected some the figures relating to the sectoral attribution in this article on 12/01/2021 as some of the CEDS emissions weren’t categorised correctly. This had little bearing on the results or conclusions; it just resulted in a larger proportion of aerosol emissions being attributed to the Industrial Processes sector, reducing its historical contribution to warming.
Three important caveats to start
Before I start, I think it’s important to mention a few things to bear in mind when reading this post or drawing conclusions from it. Although I have attempted a “serious” analysis here, it does just represent one estimate that is dependent on the methodologies I’ve used. I’ll try to note any assumptions etc. as I go along, but there are a few key points to emphasize right from the off:
- This is NOT in any way peer-reviewed, or even checked by anyone else (yet). It is analysis that I have carried out myself, and as such, could (and probably does) contain mistakes in any assumptions I make or computations I carry out. If you have any suggestions, corrections or comments, please let me know by commenting below or messaging me so I can make relevant changes!
- This analysis is carried out using a parameter ensemble within a single simple model. The model parameter ensemble is selected to match assessed ranges of radiative forcing from the 5th Assessment Report1, and constrained to be consistent with the current level and rate of global warming (including the key sources of uncertainty); but it may contain structural deficiencies that impact the results. It is likely that another simple climate model would produce different results – though this being said I would expect them to be largely consistent.
- It is possible (likely) that our understanding of some of the drivers of radiative forcing included in the model will be updated in the future, which may render some of the parameter selection and associated results presented here incorrect. This comment is most relevant to aerosol forcing, and some of the more minor forcing drivers which currently have large associated uncertainties such as stratospheric ozone.
The aim of this analysis
Within this post, I’m going to attempt to quantitatively estimate how: a) different forcing agents; b) different sectors; & c) different countries have contributed to the current level of global warming. It’s meant partly to inform discussions of what the key components of anthropogenic global warming are, partly as a bit of documentation for the forcings within FaIR, but also as an example of the type of analyses that are possible with FaIR. I hope that the results are interesting and useful.
Throughout this post, I’ll discuss the methods fully, but the figures only a little bit in the text, largely leaving it up to you to explore them properly - they are fully interactive. If any of them aren’t clear enough or are confusing, please let me know with a comment or message!
Analysis and results
Once again, I’ll be using the probabilistic FaIR ensemble2 I used in a previous post. I’m going to be using a detection-attribution style experiment to estimate the contribution of each component. This means:
- run a baseline simulation with all components included;
- re-run, but leave out a specific component;
- subtract the perturbed run from the baseline, the result is the attributed impact of that particular component.
I do this (rather than just running each component through by itself) as FaIR is weakly non-linear. The only non-linear components of FaIR are the carbon- and methane-cycles due to their temperature & uptake / concentration feedbacks. Although I don’t think it would make a huge difference (you can check this eg. by summing anthropogenic forcing components and checking against the total), I expect it would make some difference. Hence all the results are framed as “what is the impact if a particular component didn’t exist?” rather than “what is the impact of a particular component in isolation?”. The question of additivity in FaIR does need some more exploring, something I’m aiming to do when I revise the paper as it was also noted in a seriously helpful review from Glen Peters!
I make use of five main datasets during this analysis:
|RCMIP protocol emissions3||baseline emission scenario excluding CO\(_2\). SSP2-45 from 2015 onwards.|
|Global carbon project4||baseline emission scenario for CO\(_2\).|
|PRIMAP-histTP v2.15,6||Proportion of CO\(_2\), CH\(_4\), N\(_2\)O, HFCs, PFCs, NF\(_3\), & SF\(_6\) emissions per sector or country.|
|CEDS v_2020_09_117,8||Proportion of aerosol precursor emissions per sector or country.|
|UNEP Ozone9||Proportion of ODS emissions per sector or country.|
First, it’s important to demonstrate that FaIR is able to reasonably reproduce observed global warming to date (otherwise there would be no reason to trust any attributed impacts…) Below is the FaIR baseline run I use throughout this analysis, compared to the recent HadCRUT510 observational global temperature dataset. It appears to do a pretty reasonable job. Perhaps slightly too sensitive to volcanic eruptions, but in general pretty decent. This is unsurprising since this parameter set is more or less tuned to reproduce the current level & rate of warming, as described fully in a previous post, but it’s still good to check! You may notice that I’ve decided to adopt the more usual 1850-1900 reference for this post, which makes very little difference compared to the 1861-1880 reference I’ve tended to use previously. I tend to use the 1861-1880 one simply because it’s less influenced by volcanic eruptions, but in reality it doesn’t affect analysis like this significantly so I’m using the conventional pre-industrial reference period here.