We’ve been working on marginal emissions since 2018, when we developed our first algorithm using machine learning. Realizing it is a new concept for non-experts, we wrote an explanation of what marginal emissions are in 2019, and wrote about their applicability to real-time decisions in 2022.
Collaboratively working with scientific experts and grid operators for years, we have come to realize significant shortcomings:
Marginal-based methods are non-compliant with current standards and passed regulation.
Scientists and grid operators warn that marginal signals fail to capture the grid's complexity.
Using marginal emissions exposes companies to PR risks.
Values and fluctuations of marginal emissions are unintuitive for end-users.
Marginal emissions have been ruled out by the GHG Protocol Scope 2 Guidance, as well as all other major regulation. Recent legislations from the US government and the European Commission prohibit their use. Further, marginal emissions are unsuitable for Scope 2 Accounting.
Projects using marginal-based methods are non-compliant with current standards and passed regulation.
Scope 2 Guidance: “Companies shall not use marginal emission factors [...] for a location-based scope 2 calculation”
"This guidance does not support an 'avoided emissions' approach for scope 2 accounting"
Corporate near-term criteria: “Avoided emissions fall under a separate accounting system from corporate inventories and do not count toward near-term science-based emission reduction targets.”
Production of renewable liquid and gaseous transport fuels: “The emission intensity of electricity shall be determined following the approach for calculating the average carbon intensity of grid electricity.”
US Department of Energy - Clean Hydrogen "45v" Tax Credit
“The level of the credit is based on the lifecycle greenhouse gas ("GHG") emissions that result from the process of producing clean hydrogen.”
GHG Protocol - Estimating and reporting the comparative emissions impact of products
"To be consistent with the requirements of the GHG Protocol corporate accounting and reporting standards, comparative impacts should not be used to adjust scope 1, 2, and 3 emissions."
European Commission - Corporate Sustainability Reporting Directive
International Sustainability Standards Board - Climate-related disclosures
California Air Resources Board - Climate Corporate Data Accountability Act
UK Department of Transport - RTFO Guidance for renewable fuels of non-biological origin
GHG Protocol
On the surface, marginal emissions are simply the emissions caused by the power plant ramping up (or down) in response to a change in consumption. In reality, the electricity grid is a vast and complex interconnected system, having many interdependent components that all affect each others.
Grid operators acknowledge the marginal concept is an oversimplification of the reality they operate in. They state that the accuracy of these signals can't be assessed and verified in practice and therefore caution against their use.
Scientific experts warn about flaws of marginal emissions that prevent them from accurately estimating the impact of load shifting.
“Determining the correct [marginal] power plant is extremely complex or even impossible. [...] Furthermore, it is never possible to find out retrospectively whether the signal is correct”.
"Because of the various constraints and complexities involved, PJM cannot and does not make any guarantees as to the accuracy of the information nor that it is fit for any purpose."
“Short-run marginal emission factors neglect impactful phenomena and are unsuitable for assessing the power sector emissions impacts of hydrogen electrolysis”.
Q. Xu et al., System-level impacts of voluntary carbon-free electricity procurement strategies
I. Riepin et al., Spatio-temporal load shifting for truly clean computing
W. Ricks et al., Minimizing emissions from grid-based hydrogen production in the United States
50 Hertz Grid Operator
At a time when sustainability claims come under heavy scrutiny, verifiability and auditability are key. Auditing a product feature based on marginal emissions is very difficult.
Some companies have been called out by the press and by experts for their use of marginal emissions.
Big Tech’s bid to rewrite the rules on net zero: [...] will allow companies to report emissions numbers that bear little relation to their real-world pollution.”
The once in a generation chance to fix corporate emissions reporting: "Some of those global corporate giants are proposing an emissions offsetting approach that will weaken climate targets and open loopholes that allow them to claim success without delivering more ambitious – yet still attainable – climate outcomes."
Hidden Power, Broken Rules: How companies are gaming emissions reporting rules and undermining global climate targets: “[...] pushing for new accounting rules that would allow companies to underreport their emissions by up to 90%.”
Financial Times
Marginal emissions cannot be used to calculate end-users' footprint as presented in a historical usage dashboard. Recommendations based on marginal emissions factors may worsen the user’s historical footprint (calculated with hourly flow-traced emissions factors).
Users receive multiple other sources of information in their lives such as alerts from their electricity provider, or records of renewable generation in the news. These often contradict the recommendations formulated based on a marginal signal. Marginal emissions factors are commonly perceived as unintuitive and confusing for users, hindering trust and reducing engagement.
Hourly and flow-traced data incorporating electricity exchanges and time fluctuations is crucial to represent the grid's physical reality. Such signals differ substantially from yearly-averaged (“average”) emissions that are called out for inaccuracy. In contrast to marginal signals, they are verifiable, backed by grid operators, and intuitive for all users.
The best signal to reduce scope 2 and scope 3 emissions
The best signal for long-term grid decarbonization
The most intuitive signal for end-users
Different consumers impact the power grid in different ways. An iPhone starting to charge would represent a vanishingly small fraction of the grid’s load while a data center could represent a substantial increase. The former won’t impact the grid while the latter will considerably do (it might cause the dispatch of multiple power plants). There is no single power plant that can be attributed the responsibility of responding to changes across all of these use cases.
PJM, the largest US grid operator answers this misconception in a primer about marginal emissions factors: “[...] if the customer decreased their power usage, the coal generator would burn less coal. In an extremely simple scenario, this is true. The PJM system is vast and dynamic, however, with millions of values changing from one moment to the next.”
The power grid is a complex interconnected system. Multiple electricity markets are coupled, and the actual dispatch can deviate from market orders. Therefore, while the marginal power plant remains a useful economic concept, it does not depict a physical reality.
The calculation of historical footprints should follow an attributional framework and therefore use hourly flow-traced grid emissions factors. Several signals are used today for optimization: short-run marginal emissions, flow-traced grid emissions, renewable energy share, and combined wind and solar generation….
Scientists demonstrated that optimizing consumption based on short-run marginal emissions does not contribute to decreasing grid emissions in the long term.
Optimizing based on marginal emissions raises several challenges discussed in this webpage (lack of validation, unintuitive for end-users,...). Furthermore, marginal emissions factors would overstate emissions reduction from load-shifting and could worsen the carbon footprint calculated with grid average emissions factors. The optimization would not reduce the company’s scope 2 emissions. It would also not incentivize the development of more renewables as it would optimize for the cleaner grid margin instead of the cleaner grid mix.
Read more about the applicability of marginal signals for decision-making in this blog post, and why they are not suitable for scope 2 accounting in this one.
There are claims that science has reached a consensus on using marginal signals for load shifting. Unfortunately, these claims often reference studies that don’t directly assess the use of marginal signals for load-shifting. For example, they compare hourly marginal factors to a yearly average, focus on assessing the impact of policies instead of load shifting, or calculate the consequential impact of renewables investments.
One has to be careful about what types of use cases are investigated. Marginal emissions are useful for consequential accounting. Several scientific articles use them to assess the impact of introducing new climate policies or developing new renewable energy projects.
However, no conclusive study has yet been published that compares the long-term effects of using various signals for load-shifting. There is no scientific consensus about using marginal emissions factors for load shifting.
Our research on more than 100 grids worldwide shows no correlation between times of low marginal emissions and times of abundant renewable energy. Using marginal emissions could redirect demand to times with less renewable energy on the grid as illustrated by this example based on PJM grid operator data in the US.
Working with consumer-facing applications for years, we’ve seen customers who initially powered their features on marginal emissions switch to a renewable energy share signal later. Combined wind and solar generation, or the renewable energy share are much better signals to follow to ensure consumption during windy and sunny times.
Curtailment is more complex than just an excess of renewables. It is the result of complex dynamics including grid congestion and generators’ dispatch. Data published by the Californian grid operator shows that 80% of curtailment happens because of grid congestion and other local grid constraints. It also means that the tale of consuming from excess renewables is only true at an exact specific location on the grid and not elsewhere.
Predicting curtailment requires advanced and precise data that is only made available by a limited number of grid operators. It also requires models with very high spatial granularity to ensure additional consumption will not worsen transmission issues on the grid but enable consumption of wasted renewable energy. No models uphold this level of granularity today.
Combined wind and solar generation, or the renewable energy share are much better signals to follow to align consumption with renewable energy generation and avoid wasting clean energy.
Please reach out to us to join the discussion.