This website will not empower you.

Measuring the web

The author, Daniel Hartley
Daniel Hartley
Reading time: 7 to 8 minutes

In order to build more sustainable websites and apps, we first need to measure the energy they use, and the carbon emissions for which they are responsible.

The CO2 emissions associated with digital products and services depends on many factors and varies depending on which are taken into account.

I am going to concentrate on websites, including web apps, and briefly cover streaming video. The aim is to get a feel for which numbers are important, the units used, and calculate some ballpark values.

Units

The data that makes up a web page or a video are measured in bytes. For video this value is typically expressed in gigabytes (a billion bytes). For example, Netflix equates 1 hour's viewing at 'Standard Definition' to 1 GB.

Web pages are measured in thousands (kilobytes or KBs), or millions of bytes (megabytes, or MBs). The http archive puts the current mean 'page weight' or 'page size' at 2,198KBs (desktop), or approximately 2.2 megabytes (2.2 million bytes of information). For mobile the value is 1,942KBs.

A kilowatt-hour (kWh) is the energy consumed by a 1,000-watt or 1-kilowatt electrical appliance operating for 1 hour. It is commonly used as the billing unit for business and domestic users.

Kilowatt-hours are useful for aggregating electricity use from multiple sources. For example, a toaster rated at 1,000 watts on for 15 minutes a day will use .25kWh. A small fridge might use .5kWhs per day (manufacturers provide annual values because fridges don't use electricity all the time).

The energy use of domestic appliances (the toaster, the fridge, etc.), or the laptop, router, transmission tower, data centre, etc. which contribute to streaming a video, or downloading a web page, can be combined in a single value measured in kWhs.

The cost (measured in energy intensity or carbon dioxide emissions) of downloading a web page, or streaming a video must account for multiple systems and devices. Coming up with a single figure is tricky, and controversial.

The energy intensity of the Internet, expressed as energy consumed to transmit a given volume of data, is one of the most controversial issues. Existing studies of the Internet energy intensity give results ranging from 136 kWh/GB down to 0.0064 kWh/GB, a factor of more than 20,000.
The Energy Intensity of the Internet: Home and Access Networks | Coroama et al.

Greenhouse gas emission intensity (gCO2e/kWh) is measured in grammes of CO2e emitted per kilowatt-hour. Greenhouse gases, including carbon dioxide and methane, are released when fossil fuels are burnt. In 2019, the average amount of carbon dioxide released for each kWh of energy consumed across the EU was 255 grammes. If pledges to decarbonise electricity are honoured (replacing fossil fuels with renewables), this value will fall to 0 in 2050. The figure for France in 2019 was just 56 grammes, a consequence of their reliance on nuclear energy.

Measuring electricity & emissions

There are many online tools and APIs for measuring the carbon emissions associated with Internet data. I maintain a list on the sustainability page.

The table below sets out values used by two popular calculators, and a range of values for energy intensity and CO2e emissions provided by the International Energy Agency (IEA).

Changing input values updates output values elsewhere on the page. The default data amount is 1GB, a good starting point for streaming.

If you are more interested in web pages:-  

Energy intensity & CO2 emissions
ProviderkWh/GBgCO2e/kWh
0.81
475
0.23
276
0.81
33.4
0.077
475
0.077
26

Output values

Are these figures accurate?

Calculating electricity use and emissions currently relies on assumptions and averages. Averages are useful for smoothing out values but they can also disguise distortions - this is why the http archive uses median rather than average values when reporting page size. The average can be affected by very small and very large page sizes, whereas the median expresses typical page size - 50% of values fall either side of the median.

If you are calculating values for a specific website or service, consider evaluating the main components separately - servers (data centres), networks, and devices +. Streaming has a different profile than website downloads; smaller devices like mobiles consume less than laptops, both of which have a considerably smaller energy and carbon footprint than large screen, high definition televisions.

For example, a 50-inch LED television consumes much more electricity than a smartphone (100 times) or laptop (5 times). Because phones are extremely energy efficient, data transmission accounts for more than 80% of the electricity consumption when streaming. Streaming an hour-long SD video through a phone on WiFi (Scenario C) uses just 0.037 kWh – 170 times less than the estimate from the Shift Project.
George Kamiya | IEA

Chris Adams' introduction to marginal costs, energy grids, and computational demands is well worth reading if you want to examine your development stack in detail.

Tom Greenwood describes the problem of where to set emissions' boundaries, and explains how the website carbon calculator measures a site's emissions. Tom uses a value for emissions intensity of 1.8 kWh/GB in his article but I understand the website carbon calculator is now using a value of 0.81 kWh/GB in line with Sustainable Web Design.

Are there data we can compare our figures with?

In order to see if our results make sense, it is useful to look for comparable data. For example, we can use values derived for regional figures - such as those for the UK - from which per capita values are estimated.

Let's compare our values for Internet use with annual per capita averages for the UK.

In the United Kingdom, the average Internet user will generate 140 Gigabytes of Internet traffic per month in 2021, up 159% from 54.0 Gigabytes per month in 2016…
Cisco (PDF)

Using Cisco's figure value for gigabytes of data gives an annual value of (140*12) 1,680 GBs.

Examples of Internet data usage
TitleData (GBs)
1
0.002
263.76
1,680
432

Output values

kWhs of energy
g ( tonnes) of CO2

How does this compare to UK average electricity use per person of 4,500 kWhs?

There is either no data, or you have disabled JavaScript which is necessary to view charts on this site.

How does this compare to average UK CO2 emissions per person of 5.48 tonnes (2019)?

There is either no data, or you have disabled JavaScript which is necessary to view charts on this site.

The energy and emissions associated with websites and streaming is divided between the user (this value is included in their bill), the content provider and myriad intermediaries, so direct comparison means little. But comparing data does provide a sense of scale, and an idea of what is typical.

Sources of truth

Calculations for energy intensity and carbon emissions are complex and must combine contemporary and historical data sets from multiple sectors. There are also differences between authors. Most use kWhs but some prefer machine hours (especially for streaming). There is wide variance in how inclusive models are (the scope).

In his update on French sustainability accounting, Gauthier Roussilhe explains how French modelling includes a full Life Cycle Assessment (LCA) of a service, not only its immediate energy and emissions requirements.

To sum up, French designers and developers tend to develop sustainable digital services knowing that they must aim to reduce these four factors: GHG emissions, water consumption, consumption of resources and consumption of primary energy.
Digital Sustainability: a French update

There is also a paucity of sources. Not only do many website carbon emissions calculators refer to a handful of papers, the majority of the most popular calculators are built on common APIs:

Mightybytes and Wholegrain Digital collaborate at Sustainable Web Design, and provide the community with two free calculators, EcoGrader, and the Website Carbon Calculator. EcoPing and Zifera use the WebsiteCarbon carbon API (with modifications). Beacon uses data from The Green Web Foundation and Google PageSpeed Insights (whose data underpins many performance tools). The Carbonalyser browser extensions, EcoInfo, ecometer, and EcoIndex.fr rely on data and models from The Shift Project.

Collaboration is healthy - these companies have put a great deal of effort into understanding the problem, sharing resources, and building useful, and extensible tools. I recommend trying them all (each one has unique features), and reading their authors' discussions on methodology and practice.

But if you use a website carbon calculator, question its findings, and verify its sources. Understand that the field is changing quickly - more data, new patterns of consumption, more devices, more users, different users, changes in hardware efficiency, and shifting programming paradigms.

Develop a feel for the numbers and units involved.

The spat between The Shift Project and Carbon Brief and the IEA is a salutary example as to why we should question assumptions. The trouble arose over the carbon emissions attributable to watching Netflix. The parties involved scrutinised the data, and resolved their differences, providing a useful insight into methodology, rigour, and the scientific method. Unfortunately, misleading data got out and spread quickly. Authoritative websites including Phys.org and The Guardian maintain stories with inaccurate information.

When the BBC reported on emissions resulting from their programmes, they highlighted discrepancies between their estimates and those of Carbon Trust. Difference and comparison promote scrutiny and trust.

The results and comparisons here reveal just how challenging it can be to model complex systems. This is clear from the differences between the Carbon Trust and iPlayer estimates that resulted from alternative assumptions – which are necessary ingredients to any model. However, despite these differences, our results show good accordance with the Carbon Trust study. Research in this area highlights the value of using robust science to enhance awareness of the carbon impact of TV services. This is essential if we are to reduce our emissions as an industry and would not be possible without the continued collaboration of media organisations and academics. (Ed. emphasis mine)
The carbon impact of streaming: an update on BBC TV's energy footprint

What next?

Despite exponential growth in data usage, electricity demand has remained nearly flat because devices, networks, and data centres operate more efficiently year on year.

According to Andrae and Edler this looks set to change. In a best case scenario, ICT (Information and Communication Technology) will consume 8% of global electricity use by 2030; in a worst case scenario this rises to 21%, with the majority of the increase expected to come from data centres and networks.

A counter view from, among others, Koomey and Masanet, is that newfound efficiencies will continue to account for increases, and that a proliferation of smaller devices may lead to a fall in demand.

The relatively low climate impact of streaming video today is thanks to rapid improvements in the energy efficiency of data centres, networks and devices. But slowing efficiency gains, rebound effects and new demands from emerging technologies, including artificial intelligence (AI), IoT, and blockchain, raise increasing concerns about the overall environmental impacts of the sector over the coming decades.

Appendix: ICT energy consumption

Main components of ICT energy consumption (2021)GreenIT - iNUM : impacts environnementaux du numérique en France
There is either no data, or you have disabled JavaScript which is necessary to view charts on this site.

Climate care includes the energy used to manufacture the hardware.

Main components of ICT energy consumption (2017)climate care - The carbon footprint of the Internet
There is either no data, or you have disabled JavaScript which is necessary to view charts on this site.

Based on average viewing habits, my updated analysis shows that viewing devices account for the majority of energy use (72%), followed by data transmission (23%) and data centres (5%). In contrast, the Shift Project values show that devices account for less than 2% of total energy use, as a result of underestimating the energy use of devices (4x) while substantially overestimating the energy use of data centres (35x) and data transmission (50x).
George Kamiya | IEA

References

Related content