Methodology & Research

GreenByte's CO₂ estimates are grounded in peer-reviewed research, international emission databases and established energy models. Here we explain exactly how we calculate, why we trust our numbers, and where they come from.

~85%
Estimated accuracy
90%+
Desktop target accuracy
200+
Countries supported
9+
Scientific & institutional sources

How the calculation works

1

Activity classification

The extension categorizes each network request by content type: video streaming, audio, images, documents, API calls, and general browsing. Each category has a distinct energy profile based on data center processing and network transfer requirements.

2

Data transfer measurement

We measure the volume of data transferred (bytes sent/received) per activity using the browser's Performance API and webRequest API. This provides the foundation for energy estimation: more data transferred = more energy consumed across the network.

3

Energy model application

Each byte is converted to energy (kWh) using a hybrid 3-layer model built on top of the 1-byte methodology. Unlike simple "1 byte = X CO₂" calculators, we apply category-aware coefficients (video, audio, browsing, AI) across three energy layers: data center processing, network transmission, and end-user device consumption. Coefficients are derived from Shift Project, IEA and peer-reviewed studies.

4

Country carbon intensity

The energy consumed (kWh) is multiplied by the carbon intensity factor of the user's country electricity grid (gCO₂/kWh). We use Ember's annual Global Electricity Review data, covering 200+ countries. A country with more renewables produces less CO₂ per kWh.

5

Real-time aggregation

Results are aggregated per session, day, week and month. The dashboard shows trends, comparisons and personalized reduction tips powered by AI analysis of your usage patterns.

Emission factors by activity

Activity Energy factor Source
Video streaming (HD)0.077 kWh/GBIEA, Carbon Trust
General browsing0.06 kWh/GBShift Project
Audio streaming0.04 kWh/GBIEA
Video calls0.1 kWh/GBGreenspector, Purdue University
AI queries (ChatGPT, etc.)~0.001–0.01 kWh/query*IEA, Luccioni et al. (2023)
Cloud storage sync0.06 kWh/GBMasanet et al. (2020)

* Depends on model size, token count, provider infrastructure and inference length.

Core formula

CO₂ = Traffic × Energy Coefficient × Carbon Intensity
Data center + Network + Device
The energy coefficient combines three components from the 1-byte model: data center processing, network transmission, and end-user device consumption. The carbon intensity is based on the electricity grid of the user's country.

Disclaimer — GreenByte provides scientifically informed estimates, not direct physical measurements. Actual emissions depend on hardware, network infrastructure, server utilization and electricity mix at the time of use.

Example calculation

Here's how GreenByte calculates emissions for 1 hour of YouTube (1080p) in Germany:

YouTube · 1080p · 1 hour
3 GB
transferred
×
Energy coefficient (HD video)
0.077
kWh/GB
×
Germany
Carbon intensity (Germany)
385
gCO₂/kWh
=
Result
≈ 89 g
CO₂

Country carbon intensity coefficients

Carbon intensity varies dramatically by country depending on the energy mix. GreenByte uses annually updated data from Ember's Global Electricity Review and the European Environment Agency (EEA). Here are several examples:

Sweden
Sweden
45 gCO₂/kWh
Brazil
Brazil
70 gCO₂/kWh
France
France
85 gCO₂/kWh
United States
United States
180 gCO₂/kWh
Ukraine
Ukraine
260 gCO₂/kWh
Germany
Germany
385 gCO₂/kWh
Poland
Poland
450 gCO₂/kWh
India
India
490 gCO₂/kWh
Australia
Australia
510 gCO₂/kWh
China
China
530 gCO₂/kWh

Research & data sources

Our methodology is built upon peer-reviewed research, institutional reports and open datasets from leading organizations:

Data Centres and Data Transmission Networks

International Energy Agency report on global energy consumption of data centres and networks. Provides baseline energy-per-byte coefficients used in our model.

Read article

Lean ICT: Towards Digital Sobriety

The Shift Project's comprehensive analysis of ICT energy consumption. Source of the 1-byte model methodology and video streaming emission factors.

Read article

Global Electricity Review 2024

Annual dataset covering electricity generation, capacity and emissions for 200+ countries. Primary source for our country-level carbon intensity coefficients.

Read article

Recalibrating global data center energy-use estimates

Masanet et al. (2020) in Science. Recalibrated global data center energy models, showing efficiency gains. Used to refine our per-byte energy coefficients.

Read article

Power Hungry Processing: Watts Driving the Cost of AI Deployment?

Luccioni, Viguier & Ligozat (2023). Comprehensive energy measurements of AI model inference. Source of our per-query AI emission estimates.

Read article

Carbon impact of video streaming

Carbon Trust & DIMPACT study on the carbon footprint of streaming one hour of video. Cross-referenced with our video emission factors.

Read article

CO₂ emission intensity of electricity generation

European Environment Agency dataset on CO₂ intensity per kWh across EU member states. Supplementary source for European country coefficients.

Read article

The overlooked environmental footprint of increasing Internet use

Obringer et al. (2021). Purdue University study quantifying the carbon, water and land footprints of internet activities including streaming, gaming and video calls.

Read article

Green Software Foundation: SCI Standard

Industry consortium defining the Software Carbon Intensity (SCI) specification. Provides standardized methodology for measuring software carbon emissions across infrastructure.

Read article

Known limitations

Like any estimation model, GreenByte has inherent limitations. We believe transparency about these makes the tool more trustworthy:

CDN and edge caching may reduce actual server load below what we estimate
ISP-level caching and compression are not visible to the browser extension
Hardware differences (CPU, GPU, display) affect real energy use but are not measured by the extension
Browser buffering and prefetching can overcount transferred bytes
Server-side energy (data center cooling, storage) is modeled, not directly measured
AI query energy varies significantly depending on model size and inference length

How we ensure accuracy

Our estimation model combines multiple scientific approaches to maximize precision within browser-level constraints:

Cross-referencing coefficients from 9 independent scientific sources
Country-specific carbon intensity from Ember's annual datasets for 200+ countries
Category-aware energy models — video, audio, browsing and AI each use tailored coefficients
Annually updated emission factors aligned with latest IEA and Ember publications
Based on the peer-reviewed 1-byte model methodology (Shift Project, Masanet et al.)
Desktop app (coming soon) will add hardware-level telemetry for 90%+ accuracy

Transparency is our principle

We openly share our methodology, data sources and limitations. As research evolves, so will our models. The desktop application is designed to improve estimated accuracy beyond 90% with system-level power telemetry, GPU sensors and per-process energy attribution.

greenbyte.ukr@gmail.com