mrchrisadams
commented
2 years ago I've been chatting with other people on the GSF SCI data project, and based on that discussion I'll see if it's realistic to have an annual, marginal figure as a simplified fallback.
You can see more about that project below:
https://github.com/Green-Software-Foundation/sci-data/
If that doesn't work, I've found out that the dataset compiled by Ember is Creative commons Sharealike Attribution:
https://ember-climate.org/global-electricity-review-2021/data-explorer/
This would allow us to include by-country figures, but the difference is that I think these figures would not be marginal figures in the same sense as the dataset described here.
I've arranged a call to speak to them to ask a few more questions, because for many cases, having the non-marginal figures would still be useful for attributional use-cases, like annual reporting, or plugging into tools like cloud carbon footprint.
I'll add further notes about the Ember dataset, but I imagine the implementation would be almost the same as this one here - simply doing a key value lookup by country and perhaps year #11
I've been looking around, and it looks like there's a helpful set of annual emissions factors for countries that have been commissioned and created by the United Nations International Financial Institutions project.
TODO
this is the TLDR for this issue
In particular, there's section about why they were created in the first place on the project, and I think it's worth sharing here:
These provide:
mapor json file to read).I understand the intention of these to be widely available, providing some kind of sensible baseline for discussions about carbon emissions from activity. I can't see how you'd do this without the information being open, because if the information isn't open, then only the people who can afford the data get to make any data-informed arguments.
That seems to go against the stakeholder expectations, although this is a thing we'd need to confirm though, obvs.
A sample of the data
Here's the link to the data in the spreadsheet - Harmonized IFI Default Grid Factors 2021 v3.1
The first four numerical numbers are combined margin emission factors - as I understand it, this looks like it factors in future deployment, which is in most cases expected to be lower carbon than the existing infra.
What the columns mean
So, in other words:
The first four columns with numbers are combined margin grid emission factors. These are different to the OM ones, as you can see.
As I mentioned before, I think the OM figures are higher because they're not trying to incorporate future planned infrastructure - you might care about this if you were making an investment with a 10-20 year time horizon (i.e. do I build this solar farm?), but it's less relevant for understanding the carbon footprint of electricity today.
Guidance on their use
There is some guidance on where you would use these country level figures from Methodological Approach for the Common Default Grid Emission Factor Dataset - AHG 001:
Tier 1 here basically means default energy figures at country level, when you have no other info available. It's one step better than a global estimate i.e. 440g CO2/KWh.
Tier 2 and tier 3 refer to higher resolution, like grid or even grid region level, where you do have information where the number might be different where the generation is in a country, and when it's being run. This might be the case if you had access to one of the other APIs we have support for.
For the purposes of understanding your emissions from compute you use now, where you aren't doing anything clever like moving compute loads through time or space in response to the grid itself changing, the OM figures look like they would work as a baseline.
Here's some quoted guidance their usage as outlined in Methodology/approach to account project emissions associated with grid electricity consumption - AHG-002
Emphasis is mine:
This makes me think they would work as a useful default set of numbers for the countries, to use for calculating emissions from an infra set up at present - a baseline of sorts.
Comparing these to higher time resolution marginal emissions numbers from APIs
I don't think this is the same as the data from marginal computing APIs, but I'll readily admit that even now, I'm not sure.
I understood marginal intensity from API s on a short time horizon was for people figuring out right now for example, whether turning something on or off would have much of an affect on the carbon intensity on the grid.
If you wanted to work out the carbon savings from clever scheduling of compute you might compare these numbers with a marginal API, like that provided by Wattime, Electricity Map's Marginal API, the Carbon intensity in the UK provided by the National Grid, and so on.
These APIs give you some idea of what each marginal unit of usage might be - so rather than looking at the whole fleet of power plants, and allocating a share to you that's the same as the energy your server is using, they're looking at the ones that would need to be switched on based on their current view of the conditions on the entire grid, as a consequence of you running that compute, and causing the extra demand for electricity to power that server.
These aren't the same, and they tend to be higher, as you'd typically switch on faster responding fossil fuel generation to meet this need.
These seem conceptually similar to the marginal figures in the dataset linked above, but after reading the guidance, I get the impression that the annual, by-country figures in the dataset are more about deciding whether to finance a whole new energy project, rather than deciding if existing, often dirtier capacity should be spun up to meet an uptick in demand.
It may be the case that these balance out anyway over the long term, but I'm leaving a note here, for discussion, and to ask others for pointers,