SWD Emissions Model (Legacy)

Last update: November 28, 2023 to adhere to IAB Europe’s recommendations for Harmonizing Language Use for Sustainability Approaches in Digital Advertising.

Conducting accurate greenhouse gas emissions evaluations from digital products and services isn’t easy. If you consider a product’s entire life cycle, things quickly get complicated:

• How do you include the embodied energy and materials required to produce a product or service?
• What about the energy required to host a product across servers, cloud containers, and content delivery networks?
• How do you assess the energy requirements of end-users interacting with your product or service across devices over time?

These aren’t easy questions to answer. Digital products and services have many components across multiple (often closed) systems, each of which have their own energy and resource requirements. Finding a blanket one-size-fits-all solution is elusive. 

Over the past several years, Wholegrain Digital and Mightybytes, the creators of this website, collaborated with Medina Works, EcoPing, and the Green Web Foundation to define new open standards for evaluating and assessing carbon emissions from digital products and services.

Our goal is to help anyone interested in designing digital carbon estimation tools—like Website Carbon, Ecograder, or Ecoping.Earth, for instance—a methodology that provides consistent results.

Why Estimating Digital Emissions is so Difficult

Network system boundaries make it challenging to define accurate digital emissions assessments.

• If the boundaries are too tight, you won’t get a realistic representation of energy use.
• Conversely, broad system boundaries add increasingly complex variables to the equation, especially when closed network elements, like Local Area Networks (LANs) and end-user devices are thrown into the mix.

The potential margin of error widens the broader your system boundaries get.

If you’re just getting started with this topic, we suggest reading Tom Greenwood’s excellent post, Why do estimates for internet energy consumption vary so drastically?

For our needs, we defined the widest system boundaries available to represent a comprehensive carbon footprint, but segmented the impact for each sub-system to provide greater insight. We also cross-referenced a number of studies (see ‘References’ below) to get a balanced view of the current data.

System Segments

The system segments and percentages we use below are based on Anders Andrae’s study, New perspectives on internet electricity use in 2030. They can be found in the raw data sheet that accompanies the study:

• Consumer device use: end users interacting with a product or service. This accounts for an estimated 52% of the system. Returning visitors are assumed to be 25%, loading 2% of data.
• Network use: data transferred across the network. This accounts for an estimated 14% of the system.
• Data center use: energy required to house and serve data. This accounts for an estimated 15% of the system.
• Hardware production: embodied energy used in the creation of embedded chips, use of data centers, use of networks, and the use of consumer communication devices. This accounts for an estimated 19% of the system. 

Balancing the Numbers

To devise the formulas below, we read many research papers. The most useful studies are listed at the bottom of this post. Some had narrow system boundaries with higher energy estimates while others had broader system boundaries yet lower energy estimates. 

This posed a significant challenge to our efforts:

• If you overestimate, organizations can get caught in the trap of thinking they’re doing more than they actually are. This can become especially challenging when PR teams get involved.
• On the other hand, underestimating energy use could lead an organization to the conclusion that the effort is not worth an organization’s time because it is not impactful enough. 

While we chose to use numbers from the Andrae study, we cross-referenced those with the Swiss study on inconsistencies in energy estimates, which showed similar (though slightly higher) results. As noted above, we reviewed several other studies as well. As of April 2022, we also updated carbon intensity figures to match those from Ember’s Data Explorer, which provides more recent figures.

We still have some uncertainty on this, but decided to publish in the name of continued progress and better information moving forward.

Estimating Digital Emissions: The Formulas

This methodology is a standardized approach and does not account for all variables of any digital product or service. As an open methodology, this may be adapted to incorporate factors relevant to a specific product or service.

The Key Metric

We chose kWh/GB as the key metric on which to estimate the carbon footprint, as this metric is feasible to measure for most web services and is the unit of measurement used by the majority of studies on this topic.

Energy Consumption

Data used for calculating energy consumption is derived from the raw data for the “Expected 2020 scenario” from the Andrae study.

Carbon Intensity

The default figure used for carbon intensity is the global average carbon intensity of electricity (442g/kWh), which is pulled from the CO2 intensity dataset for “World” of Ember’s Data Explorer. 

• This can be replaced by numbers for the specific country or state where this is known:
  • Table A1.3 in this EIB guide
  • You can also use this global electricity map
  • Table A.III.2 in this IPCC report, which is specific to data
  • The Green Web Foundation has added country codes to CO2.js (also pulled from Ember data)
• If a percentage of renewable energy is known, that can be estimated at approximately 50g of emissions per kWh, which is based on this study from NREL.

Data Center Energy

The methodology assumes that data traffic “within data centers” and “between data centers” are sub-processes of the work that needs to be done to operate web services for end users. Data transfer to end users is the basis of the calculations.

Specific Data Points

We used these data points to define the formulas below:

• Annual Internet Energy: 1988 TWh
• Annual End User Traffic: 2444 EB

• Annual Internet Energy / Annual End User Traffic = 0.81 tWh/EB or 0.81 kWh/GB

• Carbon factor (global grid): 442 g/kWh
• Carbon factor (renewable energy source): 50 g/kWh

Emissions Calculation Formulas

Using the above data, we came up with these formulas: 

Energy per visit in kWh (E):

E = [Data Transfer per Visit (new visitors) in GB x 0.81 kWh/GB x 0.75] + [Data Transfer per Visit (returning visitors) in GB x 0.81 kWh/GB x 0.25 x 0.02]

Emissions per visit in grams CO2e (C):

C = E x 442 g/kWh (or alternative/region-specific carbon factor)

Annual energy in kWh (AE):

AE = E x Monthly Visitors x 12

Annual emissions in grams CO2e (AC):

AC = C x Monthly Visitors x 12

Annual Segment Energy:

Consumer device energy = AE x 0.52

Network energy = AE x 0.14

Data center energy = AE x 0.15

Production energy = AE x 0.19

Annual Segment Emissions:

Consumer device emissions = AC x 0.52

Network emissions = AC x 0.14

Data center emission = AC x 0.15

Production emission = AC x 0.19

Guidance for Using this Methodology

This GHG model is attributional, in that it models the emissions from a system attributable to an organization, for their own reporting purposes. It is designed to work with well known standards like the GHG Protocol Corporate Standard, to allow an organization to make a GHG assessment from digital services, typically as part of their Scope 3, or supply chain emissions.

The model is also designed to support substituting some average figures if you have access to higher precision data for each of the four main sections—for the consumer device network and data center usage, and for the embodied carbon associated with producing the hardware.

For more information, and to better understand which upstream and downstream categories we refer to in the model, as well as the precedents we refer to when doing so, please consult the documentation for the co2.js software library, or open an issue on the project.

What if Our Estimates are Wrong?

The scientific community has yet to reach broad consensus on how, specifically, to measure emissions from digital products and services. Some are critical of the model described above while others are working to hone and further refine it.

To be blunt, we can’t wait while they duke it out. The climate crisis is happening now. Whatever we can do to reduce emissions and remove greenhouse gases from the atmosphere must happen as quickly as possible. 

This freely available GHG model—even if it is imperfect—can be used by anyone or added to any digital product or service to run a GHG evaluation. 

• We’ve made every attempt to ground this work in the latest scientific data available to us. 
• Understanding that these are estimates, we rounded up numbers to account for a margin of error.
• New studies will inevitably get us closer to specific and accurate data; we’ll adjust the GHG model as we learn more.

Please contact us with questions or suggestions on how to improve the formulas above.

Sincerely,

Chris Adams, The Green Web Foundation

Rym Baouendi, Medina Works

Tim Frick, Mightybytes

Tom Greenwood, Wholegrain Digital

Dryden Williams, EcoPing

References

• New Perspectives on Internet Electricity Use in 2030, Anders S.G. Andrae, June 2020.
• Investigating the Inconsistencies Among the Energy and Energy Intensity Estimates of the Internet, Swiss Federal Office of Energy SFOE, June 2021.
• The Overlooked Environmental Footprint of Increasing Internet Use, Renee Obringer, Benjamin Rachunok, Debora Maia-Silva, Maryam Arbabzadeh, Roshanak Nateghi, Kaveh Madani, April 2021.