This blog post is part of a series of 7 articles uncovering all you need to know about marginal signals, discussing their challenges and why average emissions factors may be preferable. Find a simpler overview of the complexities to consider when using marginal signals in practice here.

After introducing the concept of marginal signals, the methodology used by Electricity Maps, we discussed why marginal emissions factors are unsuitable for scope 2 carbon accounting and the challenges of using them for real-time decision-making. In this article, we will continue to dive into the limitations of marginal emissions factors, specifically optimizing electricity consumption using marginal signals often does not lead to a reduced carbon footprint.

‍

The topic of carbon-aware electricity consumption has become more widespread with a growing number of applications in carbon-aware computing. An important consideration is choosing the right signals for emissions reporting and decision-making. So, average or marginal emissions factors?

For reporting, there is a clear consensus about using average grid emissions. Marginal emissions factors are unsuitable for scope 2 accounting and have been ruled out by major regulations and leading standards. These standards instead require the use of average emissions factors, of which flow-traced emissions factors are the most accurate version by capturing imports and exports at an hourly level.

However, as highlighted by a study recently published by researchers from the University of Massachusetts and MIT, “there is a lack of consensus on the right signal for carbon-aware optimizations”. Today, both average and marginal carbon intensity signals are used for decision-making. 

This study highlights the implications of choosing one signal over the other, by comparing Electricity Maps’ flow-traced (average) emissions factors with WattTime’s marginal emissions factors across 65 grids worldwide. It shows that both signals are statistically different with very low correlation between them, which in practice means “optimizing for one signal could lead to more carbon emissions from the other signal’s standpoint”. 

To summarize:

• There is a lack of consensus on what signal to choose for optimization but optimizing for one signal could lead to more carbon emissions from the other signal’s standpoint
• There is consensus on using average emissions factors for calculating a carbon footprint and reporting on emissions

As a consequence, optimizing based on marginal emissions factors could increase carbon emissions calculated on average emissions factors and thus the carbon footprint. In other terms, optimizing consumption with a marginal signal may increase a consumer’s carbon footprint after optimization. 

This blog post aims to illustrate the results of this study with examples taken from US and European grids.

How optimizing based on marginal emissions factors impact the carbon footprint

This section focuses on the impact on the carbon footprint of choosing between average and marginal emissions factors for load-shifting strategies. 

As mentioned in the introduction, standards and regulations require using average emissions factors to calculate the carbon footprint of electricity consumption and report on (corporate) carbon emissions. 

Shifting flexible electricity loads (such as computing jobs) in time is one important tool to reduce individuals' and companies’ carbon footprint. We quantify below the change in the carbon footprint from a simple time load-shifting strategy realized over 2024 in 12 grids worldwide: Australia New South Wales. Brazil, Germany, Denmark, Spain, France, Great Britain, Netherlands, Poland, California, Texas, and Mid-Atlantic US. The load-shifting strategy is either run based on WattTime's marginal carbon intensity or Electricity Maps flow-traced carbon intensity.

As illustrated in Figure 1, optimizing based on hourly flow-traced carbon intensity will reduce the carbon footprint in all grids. This reduction fluctuates from 7% in Mid-Atlantic US to 50% in California. The load-shifting based on the flow-traced carbon intensity yields more than 20% of carbon footprint reduction in most of the grids considered.

Optimizing based on the marginal carbon intensity does not always reduce the carbon footprint, and when it does, the reduction remains way below the reduction obtained with the optimization on hourly flow-traced carbon intensity. In Denmark and Poland, shifting loads based on WattTime marginal signal increases the carbon footprint.

Differences in shifting a load based on average or marginal emissions factors: an example from PJM grid

The largest grid in the US (PJM grid) is taken as an example here. The marginal carbon intensity is sourced from WattTime grid emissions data platform. The flow-traced carbon intensity is sourced from Electricity Maps’ historical data, which is available for free on our data portal.

PJM grid operator also publishes the marginal carbon intensity on their openly accessible data portal but warn about their use: “Because of the various constraints involved, PJM cannot make any guarantees as to the accuracy of the information”. Directly measuring marginal emissions factors is in practice impossible. As a consequence, these factors can't be validated and large differences and discrepancies exist between the signals provided by WattTime and other data sources including PJM grid operator. This is discussed in greater details in another blog of this series that focuses on the challenges of validating marginal signals.

Figure 2 shows that marginal and average emissions factors in PJM exhibit very different fluctuations in the first days of January 2024.

The marginal carbon intensity provided by WattTime reached its lowest levels at the time when renewable energy on the grid was also at its lowest levels. Shifting a flexible electricity load over these four days following the marginal carbon intensity from WattTime would have likely redirected consumption to times highlighted in green. However, most of these times coincide with times of high flow-traced carbon intensity which are precisely the emissions factors used to calculate the carbon footprint. This leads to an increase in consumers’ carbon footprint increase after optimization. 

This example illustrates how the lack of correlation between the two signals means that shifting loads with marginal emissions will not reduce (and may even increase) the carbon footprint. Researchers from the University of Massachusetts and MIT found that “among 65 regions, 36 regions (55.4%) exhibit a negative correlation between their average and marginal carbon intensity signal and only 1.5% have a strong positive correlation.” 

For most of the grids worldwide, shifting loads in time based on marginal emissions factors would increase the carbon footprint.

Carbon-aware computing is not only about shifting loads in time though. In many cases, there is also flexibility in moving loads to different locations. Temporal and spatial load shifting is what Google uses Electricity Maps forecasts for. The researchers also considered the implications of choosing one signal or the other for spatial load shifting, and their conclusion remains the same: the two signals give opposite indications of where to shift loads.

Conclusion

With more applications of carbon-aware electricity consumption, choosing the right signals for emissions reporting and decision-making becomes crucial. 

Average grid emissions make consensus for carbon footprint calculations and emissions reporting is on average grid emissions, with flow-traced emission factors being the most accurate variant. Regulations require their use and prohibit the use of marginal emissions factors. 

Both average and marginal emissions factors are used today to optimize electricity consumption. However, optimizing based on one signal may increase emissions from the other signal’s standpoint. For this reason, optimizing consumption with a marginal signal may increase a consumer’s carbon footprint while optimizing based on flow-traced grid emissions will always reduce it.

‍

Learn more about marginal emissions factors in the others parts of this series on our blog. Read the previous part about the challenges of using marginal emissions factors for real-time decision-making, or dive into the next one to learn about the challenges associated with validating marginal emissions factors.