Cambridge Ethereum Greenhouse Gas Emission Index

 Introduction

Following the release of the Cambridge Blockchain Network Sustainability Index (CBNSI), we want to further investigate environmental externalities attributable to the Ethereum network and provide a daily updated emissions estimate for Ethereum post-Merge.

Driven by a rise in popularity and mainstream adoption, the debate around the sustainability of cryptoassets has intensified. By adding this new index to the CBNSI, we hope to contribute to the debate and offer an approach to quantify Ethereum's ongoing emissions that can subsequently be used as a standalone estimate for mainnet-related emissions, but also as the foundation for any more granular analysis pertaining to emissions allocation at protocol or activity level.

In our prior publication in April, we underscored Ethereum's post-Merge electricity consumption as a pivotal milestone. Although this research signified a crucial preliminary step, it did not capture the geographical distribution of node activity and thus missed a vital component for a more comprehensive environmental impact assessment. Addressing this, our current study builds on those initial findings.

To this end, we are excited to unveil a major extension to our Ethereum research: a novel index tool that leverages information from a wide variety of sources to provide intricate insights into the distribution of Ethereum nodes, laying the groundwork for developing a post-Merge greenhouse gas (GHG) emissions estimate.

Similar to our pre-Merge GHG estimate, we performed a sensitivity analysis to assess how GHG emission estimates vary with changes in key parameters. Accurately calculating exact GHG emissions is challenging, which prompted us to use a hypothetical range encompassing three scenarios. The two extremes constitute the upper and lower limits, encapsulating the most credible estimate.

It is vital to clarify that our index solely evaluates the environmental impact stemming from the electricity consumption of identified nodes and therefore does not constitute a full life-cycle assessment.

Key terms

Table 1: Key terms and abbreviations used

Model parameters and data sources

Table 2: GHG index model parameters

Life-cycle greenhouse gas emissions from electricity generation

Table 3 lists the median total life-cycle emission factors (one-time upstream, ongoing combustion, ongoing non-combustion and one-time downstream) for selected electricity generation technologies. The values are based on a systematic review of published life-cycle assessment (LCA) studies conducted by the National Renewable Energy Laboratory (2021).

 Table 3: Life-cycle emission factors for electricity generation technologies

^a Light-water reactor (including pressurised water and boiling water) only

^b Mean value of photovoltaic (thin film and crystalline silicon) and concentrating solar power (tower and trough)

^c Land-based and offshore

^d Mean value of geothermal, biomass and ocean

Methodology

 This section briefly outlines the core concept of our methodology before delving deeper into how the estimates are calculated.

Structurally, this study closely aligns with the approach of the CBECI by establishing hypothetical lower and upper limits to demarcate the projected range of GHG emissions. This is the basis of the sensitivity analysis, summarised in Figure 1. These boundaries represent extreme scenarios premised on the hypothetical assumption that the entire electricity consumption stems from a single energy source, either exclusively from coal-fired or hydropower. This format, whilst schematic, contextualises the findings and makes the interpretation of the results more intuitive.

It is vital to recognise that while these scenarios offer illustrative insights, they are rooted in stark simplifications, suggesting that Ethereum nodes rely entirely on a single energy source. In practical terms, the electricity consumed originates from a diverse range of energy sources, collectively known as the 'electricity mix'. Identifying the components of this mix is paramount for accurately calculating the emission intensity associated with the Ethereum network. For the purposes of this study, "emission intensity" is defined as the quantity of GHGs emitted per kilowatt-hour (gCO2e/kWh). Table 3 lists the energy sources considered in this study, along with their respective emission intensities.

Figure 1: Sensitivity analysis

Understanding the geographic distribution of node operations is crucial for determining an Ethereum-specific electricity mix. As an example, the emissions resulting from running a node in Sweden would significantly differ from those in Kazakhstan.

Due to the decentralised nature of the Ethereum network, directly obtaining data about the geographical distribution of mining activity is challenging. To address this challenge, we utilised a network crawler. This tool provides the analytical data necessary to formulate dynamic electricity mix estimates, allowing us to determine the corresponding Ethereum-specific emission intensity.

 The following sections delve deeper into the methodologies employed across the various scenarios.

Lower- and upper-bound estimates

 For the formulation of an appropriate electricity mix profile corresponding to our best-guess estimate, a comprehensive understanding of the geographical distribution of node activity is paramount. Nevertheless, such detailed information is not a prerequisite for computing the upper and lower bounds. These bounds serve as hypothetical boundaries, delineating best and worst-case scenarios that encompass our best-guess estimate.

In the upper-bound scenario, it is postulated that Ethereum nodes exclusively utilise coal-fired power – an energy source characterised by the highest emissions per kilowatt-hour of generated electricity. In contrast, the lower-bound scenario stipulates that electricity is entirely sourced from hydropower.

By applying this assumption to both scenarios, we simplify the conversion of electricity consumption into GHG emissions. Even as the geographical landscape of node activity evolves, the use of a single energy source ensures that the emission intensity, I_s,d, remains unchanged. Thus, potential shifts in the distribution of node activity, which would typically alter the network's electricity mix profile, have no bearing on either scenario.

The necessary data to compute GHG emissions, namely emission intensity and electricity consumption, are readily available. The GHG emission estimate for any given day d is determined by multiplying the emission intensity of energy source s, where s ∈ S, with the respective electricity consumption for that day (Equation 1).

Best-guess estimate 

For our best-guess estimate, understanding the geographical distribution of node activity is paramount, as it underpins the creation of an Ethereum-specific electricity mix. To this end, we draw insights from the Armiarma crawler, a network monitoring tool. This tool harnesses the node discovery protocol, discv5, to detect peers for message exchange. Information derived from these exchanges is stored and subsequently utilised for various purposes, one of which includes the creation of a map illustrating the global distribution of node activity.

 However, there are instances where data may be absent, necessitating strategies to address these gaps. Such situations arise primarily under two circumstances: when making estimates before the inaugural data point on our node distribution map, and when the latest data point does not align with the current date. To navigate these challenges, we have delineated three distinct time intervals, each adopting its unique methodology, to ensure that electricity can be converted into GHG emissions, irrespective of the availability of corresponding geographical data.

 The three intervals and their underlying mechanisms will be discussed in the following segments.

Creating the three intervals

 This section elucidates the three time intervals foundational to the emission intensity calculations of our best-guess estimate.

The geographical distribution of node activity is instrumental in determining the electricity mix. However, consistent data availability remains a challenge. For context, our inaugural electricity consumption estimate dates to 13 December 2021, whereas data regarding the geolocational distribution of node activity was only available from February 2022. Consequently, methods were needed to account for periods before geolocational data became available and for any subsequent times when the most recent data point is not contemporaneous with the present date.

In response, GHG emission estimates are delineated into three distinct time intervals: historical, assessed, and predicted, as depicted in Figure 2.

Figure 2: Time intervals

The issue of unavailable data during certain periods only applies to the best-guess estimate, since the upper- and lower-bound estimates are based on a single energy source (coal-only or hydro-only) and, hence, changes in geographical hashrate distribution do not impact the emission intensity. By assuming a single energy source, the emission intensity stays constant over time and always corresponds to the value in Table 3, regardless of where the hashrate originates.

For the best-guess estimate, different methodologies were devised for each time interval to compute GHG emissions. The estimates for the historical and predicted intervals are based on notable simplifications, while computing estimates for the assessed interval is more complex:

• The historical interval, covering December 2021 to January 2022, postulates a global distribution of node operations proportional to each nation's share of total global electricity production.

• The assessed interval offers enhanced granularity, with data on the global distribution of node activity being available throughout this timeframe.

• The predicted interval, covering the period from the most recent data point to the present date, assumes that the latest geolocational data available serves as a reliable indicator of the current node distribution.

 The assumptions that underpin each interval have significant implications for the estimated emission intensity during that interval. However, as depicted in Figure 3, both the historical and predicted intervals only play a minor role due to the short period of time before geolocational data was available and the continuous updates the network crawler provides.

Figure 3: Ethereum-specific emission intensity

The following sections delve into the rationale behind each interval, detailing the computational steps and data sources used.

Historical interval

The scope of this study would have been considerably limited if only the period for which data was available was considered. Hence, the historical interval was established to account for the absence of geolocational data from December 2021 to January 2022, and an alternative approach was devised to approximate GHG emissions before February 2022.

Given the above assumption, the estimated ann. GHG emissions G_d can be computed using Equations 2 and 3.

and,

Assessed interval

The assessed interval provides annualised estimates of GHG emissions spanning from February 2022 to the last day of the month for which node distribution data is available. This interval represents the segment with the most available underlying data for determining Ethereum's GHG emissions.

While the historical interval's estimate utilises the simplified assumption that the world electricity mix accurately reflects how the electricity consumed by beacon nodes worldwide is generated, the assessed interval accounts for variations in the electricity mix at a country level. This is achieved by leveraging geolocational information on beacon node activity and utilising a multi-step procedure that classifies countries into two categories delineated in Table 4.

Table 4: Country categories

 

This categorisation is essential due to the absence of requisite data for computing country-specific emission intensities across all nations. Category 1 comprises countries (referenced in Appendix 1) for which the essential data is at hand. In contrast, countries classified under Category 2 exhibit node activity, yet lack specific information about their electricity mix. As such, a separate methodology is employed to determine an emission intensity for these nations, which will be discussed later.

After computing the estimated ann. GHG emissions of each country (G_c,d) in the two categories, the resulting country-level GHG emissions are aggregated to form the GHG emissions that can be attributed to the entire Ethereum (G_d), as shown in Equation 4.

As described below, the associated electricity mix profile is used for countries belonging to category 1, while an alternative approach has been adopted for countries falling under category 2.

Equation 5 shows the calculation of country-specific GHG emissions. Each country's GHG emissions (G_c,d) are determined by multiplying the country-specific emission intensity (I_c,d) by the country's share of total electricity consumption (E_c,d). This approach ensures that regional differences in the emission intensity of power generation and beacon node activity are considered.

While Equation 5 is used for countries in both categories 1 and 2, the equation used for calculating country-specific emission intensities (I_c,d) differs.

For each country in category 1, the emission intensity (I_c,d) is calculated using the country-specific electricity mix profile. Hence, the value of each country's emission intensity is based on the energy sources used to generate electricity. The share of each energy source (S_c,s), derived from the electricity mix of every country, is multiplied by the corresponding emission intensity (C_s) of that source. The weighted emission intensities of all sources are then aggregated to form the country's distinct emission intensity (I_c,d), as exemplified in Equation 6.

For all countries in category 2, no distinct emission intensities can be computed. Given the absence of information pertaining to country electricity mix profiles, for all countries in this category, the electricity mix profile is assumed to equal the world electricity mix profile, thus, S_s,c,d = S_s,d. Consequently, the emission intensity (I_c,d) of all these countries equals the world emission intensity, as shown in Equation 7.

Besides examining the differences in country-specific emission intensities, it is also important to distinguish between the scale of node activity in each country. To accomplish this, the electricity consumption associated with node activity in each country (E_c,d) must be determined. As illustrated in Equation 8, this value is computed by multiplying each country's share of global node activity (N_c,d) by Ethereum’s electricity consumption (E_d).

Predicted interval

The predicted interval is a method employed to estimate GHG emissions when up-to-date geographical data on beacon nodes is unavailable. This approach is relevant when there is a disparity between the most recent data available and the current date, as demonstrated in Figure 2. This discrepancy results in a data gap between the latest information on node distribution and current electricity consumption data, rendering the calculation of GHG emissions estimates infeasible while this gap persists.

The stated assumption addresses potential data inconsistencies by positing that the most recent geolocational data serves as an adequate representation of the current node distribution. Consequently, it becomes feasible to compute daily estimates of GHG emissions, even in the absence of geolocational data on node distribution. This implies that if, for instance, distribution data is available until date t, but the current date is t₁, the geographical distribution of date t is used as a proxy during the interim.

As shown in Figure 4, after new geolocation data has been released, t shifts to the right and represents the month for which the latest data is available. The availability of this new data means the GHG emission estimates of the previously predicted interval are retroactively adjusted based on the observed geolocational distribution during this period and become part of the assessed interval. Therefore, the data points in the predicted interval should be viewed with caution, as they are revised each time more recent geolocational data becomes available.

Figure 4: Shift from the predicted interval to the assessed interval after the release of new mining map data

In the predicted interval, ann. GHG emissions (G_d) are estimated by multiplying the ann. total Ethereum electricity consumption (E_d) by the latest emission intensity (I_t) in the assessed interval, as illustrated in Equation 9.

The latest available emission intensity (I_t) is derived by dividing the ann. GHG emission estimate at date t by the ann. Ethereum electricity consumption estimate at date t, as illustrated in Equation 10. To reiterate, date t represents the last day for which geolocational data was available and, thus, the latest available emission intensity estimate is also at date t.

Limitations of the model

Our estimation of Ethereum's GHG emissions draws upon prior research concerning Ethereum’s electricity consumption and node distribution data; thus, the respective assumptions and limitations of these inputs are inherited. Additionally, this study introduces specific assumptions and limitations pertinent to GHG emissions estimation.

This research, as highlighted in the summary, focuses on the environmental implications of Ethereum nodes, specifically the emissions resulting from the electricity needed for their operation. (A more detailed explanation of what this comprises can be found on our website.) Consequently, this study does not constitute a full life-cycle assessment. For simplicity's sake, we operate on the premise that marginal emissions equal average emissions. This implies that the emission intensity (in terms of gCO₂e/kWh) for any given location remains unaffected by the added demand from node activities.

In addition to the assumptions and limitations inherited from previous studies, estimating GHG emissions required introducing a new set of assumptions, which include the following:

• The median GHG emission intensities of electricity generation technologies stated in Table 3 are appropriate for all countries.

• The world electricity mix is a fair approximation for all countries where no specific electricity mix data is available.

• The estimated geographical distribution of Ethereum nodes based on observed node activity by the network monitoring tool is an accurate reflection of the actual node distribution.

• The emission intensity derived from the utilised location-based method adequately reflects the actual emission intensity of node operators.

• Power demand of node operators is assumed to remain constant throughout the day, and the accessed electricity mix does not vary between daytime and night-time.

• Our estimates do not account for any activities that could reasonably be expected to reduce emissions, such as using flare-gas, off-grid power access (i.e., utilising own generation facilities), waste heat recovery or carbon offsetting.

 While the majority of limitations may not substantially affect the model's efficacy, we are cognisant of its imperfections. The CBNSI is an ongoing project, being constantly updated in response to new data becoming available. All modifications are transparently documented in the Change Log.

If you have suggestions on how we could improve the Index, please feel free to send us a message.

How does our estimate compare to other estimates?

Historically, numerous studies have sought to assess Ethereum's environmental impact. Notably, there are only a small number of live indices that continuously track emissions. It is pivotal to note that emission estimates can fluctuate considerably over time; thus, the publication date becomes crucial when comparing different studies. Furthermore, certain studies exclusively consider carbon dioxide (CO₂) in their assessments. Table 5 provides a non-exhaustive list of estimates from other entities.

Table 5: Available studies and articles on Ethereum's environmental footprint

Appendices

Appendix 1

Countries in category 1

 Note: Snapshot that will be expanded should node activity be detected in non-listed countries.