I co-run the website ShouldIBake.com (and its associated Twitter @baking4cast). It uses forecasts from National Grid to show the proportion of solar, wind, and hydro power on Great Britain’s electricity network. I bake bread most weeks, and my basic motivation in developing the project was to help me decide when I could do it with the most renewable electricity.
But there is a potential difficulty here. This is that when thinking about when to use electricity, if your main concern is the associated carbon emissions, it could make more sense to focus on marginal rather than average carbon intensity of. In this blog I quickly explain what this means and, why despite the validity of the marginal emissions approach, I think average emissions are still appropriate to use in this context.
In the margins
The clearest explanation I found of the difference between marginal and average emissions is given in this electricityMap blog by Olivier Corradi.
The basic point is this. When I turn on my oven to bake, I add an (albeit tiny) electricity demand to the overall existing pool of demand. Because supply and demand must always be in balance, a commensurate increase in supply is necessary. This increase in demand is likely to come from the most expensive power source (the cheapest ones are already being fully used).
Because renewables are the cheapest power source to operate, in Britain this marginal increase will almost always come from gas. The change in greenhouse gas emissions associated with this change in generation is known as the marginal emission factor (MEF).
According to this approach, because the impact of turning on my oven would be to increase gas use, the carbon impact of this baking session should be seen as that of burning gas. Let’s say the average carbon intensity of the grid is currently 100g CO2/kWh, but intensity from the marginal gas generation is around 500g. The consequence of my baking now is an increase of 500g CO2/kWh.
Corradi’s blog draws a useful distinction between the consequentialist paradigm, as described above, and the accounting paradigm. In the latter, you look back after the fact and divide grid carbon emissions equally between all users, meaning my baking could be seen as accounting for 100g CO2/kWh. Please check out that blog for a fuller description of this.
What are the implications of this for ShouldIBake.com? It could be reasonable to say that the question “should I bake?” leads us to the consequentialist paradigm, and marginal emissions. The question “when should I have baked?”, on the other hand, leads us to the accounting paradigm. If I used ShouldIBake.com and chose to bake on Monday at 50% renewables rather than Tuesday at 25% renewables, and then later account for my baking-related carbon emissions, this would have been a reasonable action.
If you want to be relatively sure your decision on when to bake to will directly cause the smallest carbon emissions, MEF would arguably be the better indicator to look to, rather than average generation which ShouldIBake.com currently provides. We’ll look into whether we can provide this as an additional information stream, although our capacity to make significant developments like this is somewhat limited. And it would depend on ability to access the information, with the electrityMap API being a possible source.
As a temporary measure, we’ve tweaked the site to advise against baking between the hours of 4–7pm. I haven’t been able to put my hand on recent average MEFs per half hour period for GB, but would suspect it coincides with periods of highest demand. This is when the most expensive plants are likely to be brought online, implying they are less efficient and therefore emitting more carbon per unit of generation. But open to being corrected on this!
However, I also see a number of reasons why keeping a focus on average emissions (or in the case of ShouldIBake.com, proportion of renewable generation) is justifiable.
Emitting emissions, omitting omissions
Marginal emissions factors are a pragmatic and often useful way of thinking about the carbon impacts of electricity-using actions. But I think they are subject to a form of omission bias. This is where people place more weight on the potential impacts of action vs inaction.
In all the examples of MEFs at work, the scenario is set up as follows:
- There’s a whole load of existing demand
- Someone acts to add or remove a demand
- The emissions allocated to that extra or removed demand are accounted for at the MEF.
For me, there is a blind spot here between 1 and 2:
1a. Other people DO NOT act to change their existing demand.
What the marginal emissions approach does is effectively to take prior demand as read, and consider only the new change in demand. This is reasonable and pragmatic insofar as the prior demand is not something under the control of the creator of the new demand. But I think it is important when it comes to matters of attribution, and of individual vs collective responsibility.
To illustrate this, let’s say we have two electricity users (Alice and Bob), and two power sources. One of the power sources is low-carbon hydro power, constantly generating 1 kW. The other is a diesel generator that turns on when demand is higher than 1 kW. User Alice has decided to run a 1 kW heater constantly to heat her greenhouse.
Any other demand that either Alice or the other user, Bob, adds at any point would therefore be accounted at the MEF of the diesel generator. Bob never even has an option to use electricity at anything other than the MEF of diesel. If Alice uses only her heater, then each time Bob decided to turn on his oven, that decision would cause the diesel generator to kick in, making Bob responsible for the associated carbon emissions. Right?
Let’s reframe the situation making explicit the omission bias. Imagine a point in time, T, at which Bob decides to put his oven on. To complete the description of this situation, we also need to capture that fact that at time T, Alice DOES NOT decide to turn off her heater.
It’s all about priorities
The MEF approach implicitly prioritises precedence. Essentially, if your demand precedes another (i.e. is on first), you get priority when it comes to access to lower carbon sources of generation. Instead of placing “already-on” and “about-to-come-on” demands on an equal footing, about-to-come-on demands are seen as boosting demand, stimulating higher carbon generation and therefore increasing emissions.
The generalised outcome of this is that that the less likely a particular kind of demand is to be present on the grid, the more likely it is to face a high carbon MEF when it does appear. This is because more constant demands are more likely to be already online.
The flip-side of this situation is that because adding new demands are assigned a high carbon MEF, turning demands off should therefore also be assigned a high value of carbon saving. This is why organisations like the Rocky Mountain Institute advocate an MEF approach when it comes to correctly recognising the carbon impact of turn-down demand response.
This situation does, however, risk leaving electricity users in a quandary. Let’s go back to the baking example. Under an MEF approach, the options available to home bakers are:
- Bake whenever you like. Since in Britain natural gas is the dominant source of marginal generation, the decision to turn on your oven will almost always have a relatively high carbon impact (i.e. at the MEF of gas generation, which varies but not a great deal [I think]— although note above comments on 4–7pm period).
- Cut out baking altogether. By explicitly not-baking at the times you would otherwise have baked, you are credited a high MEF carbon saving because of the marginal gas you have avoided burning.
The answers to the question of when to bake here are, respectively, “whenever”, and “never”. The timing is irrelevant. Not very engaging or motivating!
Let’s do a thought experiment. What if, instead of viewing some demands as pre-existing and others as new (and therefore marginal), the “slate” of loads was wiped clean at every instant in time. At every instant, the demand present on the grid was the result of an explicit or implicit decision taken at that instant — either to turn on a new device, or to keep a device that was previously running turned on. (Put more realistically, we could view the decision to keep a device running as the default re-application of some previous decision to turn it on.)
Moving to this framing has no impact on the amount of demand on the grid at any instant — it is identical to under the precedence framing. However, it does force us to question how we assign a certain unit of demand to certain unit of generation (and therefore emissions).
Under the precedence approach it is easy — a “continued demand” decision precedes a “change demand” decision and therefore ends up lower in the stack of emissions. But let’s go a step further in our thought experiment and imagine that we had no knowledge of the state of a demand prior to the instant in question.
To revisit our previous example, at time T we see Alice has a heater on, and Bob an oven. But we wouldn’t know if Alice had just turned on her heater that instant, or if it had been on in the previous one. (Nor would we know for Bob’s oven.) How, in such a case, could we choose to allocate emissions? I will write more on this separately at some point, but for now, it doesn’t seem unreasonable to take the average emissions approach. We would be agnostic as to which of the demands mean the diesel generator has to run, and allocate the average emissions of the hydro and diesel generators to both demands.
Wider considerations
The idea of marginal generation, costs, and emissions, is well established. It is well-suited to thinking about how electricity works in centralised, fossil-fuelled systems. In these situations, constancy of demand is valued. In that world, where supply follows demand, demands that come and go pose challenges to be met by generation. Perhaps this explains why the act of “turning on/off” gets more focus than the act of “staying on/off”.
But electricity systems are changing. In decarbonised, decentralised grids based largely around renewable generation, demand increasingly must follow supply. Here, the previous paradigm is turned on its head. Now, demands that turn off and on, and can be shifted in time and space, are more valued. Constant demands become more challenging. In that world, it is going to be increasingly hard to overlook “staying on” decisions.
While there are various drivers for demand side flexibility, matching demand to low-carbon generation — including variable generation from renewables — is an important one. That is why organisations like National Grid make the data available that we use to run ShouldIBake.com.
But clearly these services are little-known. Most demand is not operated flexibly. From the point of view of someone who has decided to act flexibly — and if you assume that no-one else takes responsibility for the timing of their electricity use and therefore don’t change their “staying on” decisions — it still makes sense to pay attention to MEFs. But as described above, I question whether allocating responsibility for emissions at the MEF is either fair, or what we need for the future. This is why it is also important to think longer term.
Final thoughts
The idea that the carbon intensity of the grid varies over time needs to be brought to wider attention. Services which recognise this and other valuable aspects of flexibility need to be offered in the retail market. The sooner these things happen the better — because societies that begin making this shift sooner will be better adapted to function in a supply-led, renewables dominated world. While using average generation/emissions data to inform this has its challenges, it is also much easier to interpret that MEFs, and is likely to become increasingly important as more and more of our demand is met by renewables.
