What is the carbon footprint of electricity?
Our article in the Journal of Industrial Ecology compares the direct and life-cycle carbon intensity of electricity across multiple heterogeneous data sources and shows how much the answer depends on method. Explore it live, powered by nxbase.
Ask for the carbon intensity of electricity and you’ll get a number that sounds settled. But the answer is much more uncertain than expected.
Plenty of authoritative sources publish emission factors for electricity, and they disagree — sometimes by a factor of two or more. Not because any of them is wrong, but because they draw different system boundaries, make different assumptions, use data from different years, and model the grid in different ways. The result is a serious question answered with numbers that are all defensible yet tell quantitatively different stories.
Our article in the Journal of Industrial Ecology puts these estimates side by side through a single, reproducible pipeline — and treats the spread between them as a finding in its own right.
Why these numbers matter more every year
This is no longer an academic quibble. As corporate reporting extends beyond Scope 2 into Scope 3, the electricity embedded in a company’s supply chains reaches — at least indirectly — every Scope 3 category. The factor you choose doesn’t stay put: it propagates through the entire inventory.
Scope 2 itself is under revision, chasing ever-finer temporal granularity. That, we’d argue, is a second-order problem. What’s really missing is consistency across sources.
Two questions hiding inside one number
Most of the disagreement comes down to two methodological choices.
- Direct vs life-cycle. Direct emissions are what leaves the power plant’s stack when it burns fuel. Life-cycle emissions also count the supply chain — building the plant, mining and transporting the fuel, manufacturing the panels and turbines. Nuclear and wind have almost no direct emissions but a non-negligible life-cycle footprint; a gas plant, by contrast, is dominated by its direct combustion.
- Monetary vs physical. Monetary models (in our case, input–output models) trace emissions through the economy in units of money, then convert to energy — comprehensive, but coarse. Physical approaches — life-cycle inventories or measured grid data — work directly in physical units: more precise for electricity, but they draw their system boundaries differently.
Cross those two axes and you get the four quadrants that organise the comparison that follows. To put the monetary databases — EXIOBASE, EMERGING, EORA 1, GLORIA and GTAP, all processed with the new version of MARIO, our open-source input–output engine — on the same footing as the physical ones, we bring their money-based factors onto a common grams-of-CO₂-per-kWh basis using country-level electricity prices for 2017 and 2023.
Explore it yourself
The chart below puts every source side by side, drawn from nxbase, our open data infrastructure. Each marker is one source’s estimate for one country; colour marks the method (monetary vs physical), shape marks the source. Toggle between direct and life-cycle emissions and between years; click a source in the legend to hide it, double-click to isolate it, and hover any marker for the exact value.
Carbon footprint and intensity of electricity, by country and data source
Each marker is one data source's estimate for a country; colour = natively monetary (converted to a physical basis via electricity prices) vs natively physical, shape = source. Filter by scope, year and methodology. The spread across markers is the methodological uncertainty.
* Natively monetary factors are converted to a physical basis (gCO₂eq/kWh) using country-level electricity prices. Served through eNextGen nxbase (open tier, anonymous query). For Eora, GLORIA and Electricity Maps, only the datapoints shown in the article are held in nxbase. Source: Citterio et al. (2026), doi:10.1007/s44498-026-00148-3.
A few patterns stand out. Low-carbon grids like Sweden and France sit near the bottom whatever you ask — yet the relative disagreement between sources is widest exactly there, where small absolute gaps loom large. Coal-heavy grids like Poland and Estonia are high everywhere, but the sources still span hundreds of grams per kilowatt-hour. And even within a single family the estimates diverge: two monetary input–output databases can disagree as much as a monetary and a physical one do.
Even for direct emissions from fossil fuels — where you’d expect broad agreement — the estimates vary substantially from source to source. Bring in full life-cycle emissions, and the discrepancies grow wider still.
Zoom in: by generation technology
Behind every national number is a mix of technologies. The next chart drops beneath those averages to the technologies themselves — coal, gas, nuclear, wind, solar, hydro — pooling the estimates across sources. The input–output sources come from nxbase: nxsut 3.0 and the EXIOBASE hybrid (EX3h) each contribute one point per country × technology, for both the direct and life-cycle views, computed from their activity-level emissions; the process-LCA and reference values (Ecoinvent, Electricity Maps, IPCC, NREL) come from the harmonised literature.
Carbon footprint of electricity, by generation technology
Physical-unit emission factors (gCO₂eq/kWh) for EU27 + US, pooled across data sources. Fossil vs low-carbon. Each box spans the interquartile range (median = solid line, mean = dashed); whiskers reach 1.5×IQR; dots are outliers. Switch between direct and life-cycle emissions, and zoom on the clean technologies.
Physical-unit emission factors by technology, EU27 + US, gCO₂eq/kWh. The input–output sources nxsut 3.0 and the EXIOBASE Hybrid contribute one point per country × technology. The literature (Electricity Maps, IPCC, NREL, Ecoinvent) is the article's supplementary distribution: Citterio et al. (2026), doi:10.1007/s44498-026-00148-3. Life-cycle axis capped at 2500 gCO₂eq/kWh (outliers beyond drawn at the edge).
The pattern is stark: fossil technologies sit an order of magnitude above the low-carbon ones. Switch to direct emissions and renewables and nuclear collapse to almost zero — their footprint is nearly all in the supply chain, not at the smokestack.
What to do about it
The paper closes with a practical decision framework: given your purpose — a corporate inventory, a policy comparison, a product LCA — which source is the right one, and what do you trade off by choosing it? Transparency about assumptions and methodological choices is the floor.
| Use case | Recommended data type | Rationale and key considerations |
|---|---|---|
| Territorial electricity benchmarking / producer-side analysis | Direct or life-cycle emission factors in physical units, from primary or official sources | Use when the question concerns electricity generated within a territory. Suitable for producer-side benchmarking and comparability with generation statistics. Not ideal for end-use footprints where imports/exports materially alter the delivered mix. |
| Scope 2 location-based accounting | Consumption- or delivery-based direct factors in physical units; production-based factors only as a proxy when better consumption-based data are unavailable | Prefer in highly interconnected or import-dependent grids. Always disclose whether the factor is production- or consumption-based, together with its data year and treatment of imports/exports. |
| Scope 2 market-based accounting | Supplier-specific or residual-mix direct factors in physical units | Use only for contractual claims that meet Scope 2 quality criteria. Sits outside the main quantitative comparison in the paper. |
| Scope 3 category 3 (fuel- and energy-related activities, incl. upstream electricity and T&D losses) | Life-cycle emission factors in physical units (grams CO₂eq per kWh) — average, supplier-specific, or hybrid | Align with the chosen Scope 2 basis. Explicitly state whether transmission and distribution losses are included. Monetary factors are not appropriate once kWh are known. |
| Corporate sustainability disclosure, hot-spotting | Monetary input–output databases (e.g. EXIOBASE) | Appropriate when expenditure data dominate and many categories or suppliers must be covered. Prefer multiregional models when supply chains are international. Use to identify hotspots, then refine material categories with hybrid or supplier-specific/physical data. |
| Comparability with product LCAs or EPDs | Process-based life-cycle databases (e.g. Ecoinvent) | Ensures methodological consistency with the dominant approach in product-level studies. Disclose cut-off rules, multifunctionality assumptions, capital-goods treatment, geography, and data vintage. |
| Rapidly evolving grids / time-sensitive applications | The freshest, year-matched physical factors; where relevant, sub-annual, hourly, or near-real-time data | If the data lag exceeds ~2–3 years, or the grid mix is changing quickly, run a sensitivity analysis or triangulate with a more recent source. Temporal and spatial representativeness should weigh as much as source authority. |
The ceiling — and the bigger prize — is greater international harmonisation and standardisation of emission factors for the commodities that drive greenhouse-gas inventories, energy carriers first among them. Strengthening the consistency of these foundational datasets would do more for the credibility and comparability of carbon accounting than another decimal place of temporal granularity in Scope 2 reporting.
Open by construction
Everything behind these charts is queryable, anonymously, from nxbase. Neither figure is a static image exported once — each is rebuilt from live queries against the open tier of our database, the same queries anyone can run. The monetary input–output factors are converted to g CO₂eq/kWh on the fly, using governed electricity prices and ECB exchange rates — the same conversion any nxbase user would run.
That’s the point of nxbase: not another dataset, but a place where heterogeneous sources — input–output tables, life-cycle inventories, measured grid data — are mapped to a common structure, kept versioned, and served through one API. A comparison like this stops being a one-off spreadsheet and becomes something you can re-run, extend, and build on.
An update since the paper — and what’s next
Since the manuscript was finalised, we’ve kept moving. The hybrid supply-use table used in the paper has been rebuilt as nxsut 3.0 — an open, reproducible electricity supply-use table that replaces the proprietary inputs entirely. It’s the first fully open tier of nxbase, and it’s already in the chart above (the blue triangle, life-cycle 2023). A dedicated post is coming soon.
There’s a deeper thread here, too. Turning a monetary footprint (per euro) into a physical one (per kilowatt-hour) requires a price — and price is context-dependent, never universal. That conversion is one instance of a broader capability we’re building into nxbase: converting not just units, but quantities, always with the context made explicit. More on that soon as well.
Read more
- Article: Citterio, C., Golinucci, N., Rinaldi, L., Rocco, M.V. (2026). Carbon intensity of electricity: a systematic methodological and quantitative review. Journal of Industrial Ecology. doi:10.1007/s44498-026-00148-3
- Code & data: GitHub
- The data infrastructure: nxbase
Data & licensing
Every series in these charts is served through nxbase. Here’s where each one comes from, and under what licence it is redistributed:
| Source | Licence (as republished) |
|---|---|
| nxsut 3.0 | CC BY-SA 4.0 — eNextGen supply–use model |
| EMBER | CC BY 4.0 |
| eGRID | US EPA, public |
| GTAP | Proprietary - Datapoints embedded in the article under CC BY 4.0 |
| EXIOBASE 3.10.2 | Proprietary - Datapoints embedded in the article under CC BY 4.0 |
| EXIOBASE 3.3.18 (Hybrid) | CC BY-SA 4.0 |
| EMERGING | CC BY 4.0 |
| NREL | CC BY 4.0 |
| IPCC | CC BY 4.0 |
| JRC | CC BY 4.0 |
| EORA 1 | Academic licence — Datapoints embedded in the article under CC BY 4.0 |
| GLORIA | CC BY-NC 3.0 |
| Electricity Maps | Proprietary — Datapoints embedded in the article under CC BY 4.0 |
For the sources with restrictive licence — EORA 1, GLORIA and Electricity Maps — only the datapoints used in the original JIE article are shown.
The JIE article and this blogpost are published under CC BY 4.0. Wherever a source’s licence restricts redistribution, only the individual datapoints that already appear in the article are used.