ecoinvent LCI Database — The Definitive Reference

1. What ecoinvent Is — and Is Not

ecoinvent is a life-cycle inventory (LCI) database: a structured collection of tens of thousands of unit-process datasets, each describing the inputs and outputs of producing one unit of a product or service — the energy consumed, materials used, and emissions released. It is the raw material of life-cycle assessment. When an LCA practitioner needs to know the environmental burden of a kilowatt-hour of German grid electricity, a kilogram of primary aluminium, or a tonne-kilometre of road freight, ecoinvent is where that number most commonly comes from.

What ecoinvent is not is equally important. It is not a standard — it does not set rules the way ISO 14025 or the GHG Protocol do. It is not a calculation method — it supplies inventory data, and the practitioner chooses the impact-assessment method applied on top. And it is not a finished footprint — it is the background system that surrounds a study’s foreground data. Confusing the database with the standard, the method, or the result is the root of most misunderstandings about it.

1.1 Where ecoinvent Sits — the Layer Map

The cleanest way to understand ecoinvent is to place it in the stack of things that together produce a published environmental claim. The database is the bottom layer — everything above it consumes its data.

ecoinvent occupies the data layer. The method layer (how to run the LCA) and the declaration layer (how to publish it) both depend on the inventory data underneath. A standard tells you how to build the house; ecoinvent supplies the bricks.

1.2 The ecoinvent Association

ecoinvent began in the late 1990s as a Swiss project to build a consistent, transparent LCI database, and released its first version in 2003. Over the following two decades it grew into the ecoinvent Association, a non-profit organisation based in Zurich that publishes the database and coordinates a large network of industrial and scientific data providers. The non-profit, association-led governance matters: it positions ecoinvent as a community resource rather than a single vendor’s commercial product, and underwrites the peer-review and transparency that give the data its credibility.

1.3 Scale and Coverage

The current release contains more than 26,600 datasets spanning the global economy. The breadth is the point — an LCA can only be as complete as the background data available to it, and ecoinvent’s coverage is what makes economy-wide studies feasible.

Sectors updated in the current version include fuels, electricity, chemicals and plastics, waste, metals, textiles, forestry and wood, agriculture, transport, pulp and paper, and batteries. Geographic coverage runs from country- and region-specific datasets to two important catch-all geographies — GLO (a global average) and RoW (Rest-of-World, everything not covered by a specific region) — which together ensure a dataset is always available even where local data is not.

2. Anatomy of the Database

Understanding how ecoinvent is structured is the difference between using it correctly and mis-citing it. A few core concepts govern everything.

2.1 Unit Process vs LCIA Result — the Two Views

ecoinvent data can be consumed in two fundamentally different forms, and confusing them is a common error:

An EPD generator that hands you “the emission factor for 1 kg of steel” is giving you the LCIA-result view; an LCA practitioner modelling in openLCA or Brightway typically works with the unit-process view so the software can re-solve the network. Both come from the same database; they are different representations of it.

2.2 Activities, Products, and Exchanges

The atomic unit of ecoinvent is the activity — a process that produces one or more reference products. Each activity is described by its exchanges: the flows in and out. Exchanges come in two kinds — technosphere exchanges (links to other activities, like “consumes 0.5 kWh of electricity”) and elementary exchanges (flows to and from the environment, like “emits 2.1 kg CO₂”). Every product with mass carries at least six standard properties, including dry mass, wet mass, water content, and crucially fossil and non-fossil carbon content — and every product carries a price, which is what enables economic allocation in the system models discussed next.

2.3 Market vs Production Activities

ecoinvent distinguishes production activities (a single facility or technology making a product) from market activities (a consumption-mix that blends all the production routes supplying a region, plus the transport to get the product there). When you want “the steel a buyer in Europe actually receives,” you use the market activity; when you want a specific production route, you use the production activity. Selecting a production dataset where a market dataset was appropriate — or vice versa — is a subtle but consequential modelling error.

3. The Three System Models

This is the most important — and most misunderstood — feature of ecoinvent, and the one that distinguishes it from almost every other LCI database. ecoinvent is the only major LCI database published in three parallel system models: Cut-off, APOS, and Consequential. The same physical product has a different environmental result in each, because each model answers a different question about how to share burdens across co-products and recycling. Choosing the wrong one — or worse, mixing them — silently corrupts a footprint.

3.1 Why One Product Has Three Numbers

The divergence is not error or uncertainty — it is a deliberate consequence of three different, internally consistent rules for handling the two hardest questions in LCA: how to allocate the burden of a process that makes several products at once, and how to credit recycling. Consider a single illustrative material flowing through all three models:

3.2 Cut-off (Allocation Cut-off by Classification)

The Cut-off model — formally “Allocation, cut-off by classification” — applies the recycled-content (or “100/0”) perspective. A primary producer carries the full burden of primary production. Recyclable material leaves a product’s system burden-free at end of life, and a product made from recycled material enters carrying only the burden of the recycling process, not the original primary production. The producer of waste gets no credit for recyclability; the user of recycled material gets a low-burden input. This is the most widely used model, the default in many tools, and the one underlying most EPDs.

3.3 APOS (Allocation at the Point of Substitution)

The APOS model — “Allocation at the Point of Substitution” — keeps recycling and treatment processes connected to the upstream system that generated the waste, allocating burdens at the point where a recycled material substitutes for a primary one. Where Cut-off makes a clean break, APOS distributes burdens across the whole chain that produced and used the material. The result is a fuller, more interconnected attributional picture in which by-products and waste-treatment carry a share of upstream burden rather than dropping out.

3.4 Consequential

The Consequential model answers a different question entirely: not “what share of existing burdens belongs to this product” but “what happens to total emissions if we make one more unit of it.” It is built from marginal suppliers — the producers who would actually expand to meet additional demand — and uses system expansion with substitution to credit avoided production. Consequential modelling is the right tool for decision and policy analysis (“should we do more of this?”) and the wrong tool for attributional reporting (“what is our footprint?”). It is the most complex model and the least used in corporate reporting.

3.5 Which Model to Use — and the Rule You Cannot Break

4. Data Quality & the Pedigree Matrix

ecoinvent does not just publish numbers — it publishes a structured assessment of how good each number is. The mechanism is the pedigree matrix, a framework that converts qualitative judgements about a dataset’s representativeness into a quantitative uncertainty estimate.

4.1 The Pedigree Matrix

Every data point is scored against five independent quality indicators, each on a scale of 1 (best) to 5 (worst). The five indicators capture the different ways a dataset can fail to represent the situation you are actually modelling.

The five pedigree indicators, each scored 1 (best) to 5 (worst). The scores are not averaged into a grade — they are converted into uncertainty factors. Intermediate scores (2–4) interpolate between the anchors shown.

4.2 From Pedigree Score to Uncertainty

The pedigree scores are not a report card — they feed a calculation. Each score maps to an uncertainty contribution, and the contributions combine (with a basic uncertainty term) into a lognormal distribution around the dataset’s value. That distribution is what makes ecoinvent data usable in Monte Carlo uncertainty analysis: a practitioner can propagate the uncertainty of thousands of background datasets through a model and report not just a central footprint but a confidence range around it. Few other databases ship this uncertainty information natively, and it is one of ecoinvent’s defining strengths for rigorous LCA.

5. Versions & Reproducibility

ecoinvent is a living dataset, updated every year. That continuous improvement is a strength, but it creates a discipline requirement that catches many practitioners out: a footprint is only reproducible if the exact database version and system model are pinned and disclosed.

5.1 Release Cadence and Numbering

Since 2003 the database has been updated yearly, with each version building on all previous versions. Releases are numbered on a 3.x line, and each ships an extensive set of release notes documenting what changed. Recent releases:

5.2 Why Version-Pinning Is Non-Negotiable

Two studies that both say “we used ecoinvent” are not necessarily comparable and not necessarily reproducible. Background values shift between versions as supply chains are re-reviewed, new data arrives, and methods are refined. A footprint that does not record its exact version cannot be reproduced, cannot be fairly compared with another, and cannot be defended in assurance. The minimum reproducibility disclosure for any ecoinvent-backed result is three items together:

5.3 What Changes Between Versions

A version update is not cosmetic. Each release can change background values through new and updated datasets, re-reviewed supply-chain links, expanded geographic coverage, and corrections — ecoinvent publishes known data issues openly and folds fixes into subsequent releases. Because the system models are re-computed on the updated data, a version change can move results even where the specific dataset you use was not itself edited. This is why moving an existing inventory to a new ecoinvent version is a deliberate, documented exercise, not a silent background refresh.

6. LCIA Methods on Top of ecoinvent

A point that trips up newcomers: ecoinvent supplies inventory — the list of emissions and resource flows — not impacts. Turning that inventory into a climate, acidification, or water-scarcity number requires a life-cycle impact-assessment (LCIA) method, and the method is chosen separately from the database. The same ecoinvent dataset produces different impact results under different LCIA methods, entirely independently of the data.

The practical consequence is that “method choice” is a third axis of variation alongside version and system model. An EN 15804+A2 EPD, for instance, pairs the EN 15804-aligned Cut-off dataset with EF 3.1 (or TRACI 2.1) to produce exactly the indicator set the declaration requires. Reporting a result without naming the LCIA method leaves the impact numbers unanchored.

7. ecoinvent in the Standards Stack

ecoinvent is not a standard, but it is the data engine that nearly every product-environmental standard runs on. The relationship in each case is the same: the standard sets the rules and the foreground data, and ecoinvent supplies the background system. The boundary between the two — what the manufacturer measures directly versus what comes from the database — is itself a critical modelling decision.

8. ecoinvent vs Other Databases

ecoinvent is the most-used LCI database but not the only one, and choosing a data source is a real decision with real trade-offs. The most important distinction is structural: ecoinvent is a process-based database (bottom-up, physical unit processes), which is a fundamentally different thing from an environmentally-extended input-output (EEIO) database (top-down, built from economic sector data).

8.1 Process-LCA vs EEIO — the Core Divide

The two are complements, not rivals. Process data gives precision where you have specific information; EEIO gives completeness where you only have spend. Sophisticated Scope 3 inventories use both — EEIO to screen the whole value chain, process data to refine the hotspots.

8.2 The Database Landscape

The pattern is clear: free government factor sets like DEFRA and EPA eGRID are excellent for operational Scope 1 and 2 reporting but shallow on upstream supply chains; the free ICE database is a fast carbon-only option for construction materials; EEIO databases give complete-but-coarse spend coverage; and ecoinvent is the deep, multi-impact, process-resolution option you pay for when product-level accuracy matters. They are different tools for different jobs.

9. Access, Licensing & Tools

9.1 Is ecoinvent Free?

No. ecoinvent is a licensed database, not a free public dataset — this is one of the most common points of confusion. The non-profit Association funds its ongoing data collection, peer review, and annual updates through licence fees, with tiered pricing for academia, small organisations, and commercial users. This is the practical trade-off against free government factors: ecoinvent’s depth, multi-impact coverage, uncertainty data, and three system models come at a licensing cost that free carbon-only factor sets do not carry. For an organisation doing operational carbon accounting, free factors may suffice; for one doing product LCA or EPDs, the ecoinvent licence is usually unavoidable.

9.2 The Tools Landscape

ecoinvent is consumed through LCA software rather than used raw. The major environments:

Note that the database licence and the software are separate purchases: openLCA is free but the ecoinvent data within it is licensed; SimaPro and the data are typically bundled but distinct line items. Budgeting for an LCA capability means budgeting for both.

10. Practical Use & Common Pitfalls

10.1 Choosing the Right Dataset

Selecting a dataset is a matching exercise across three axes: the right product/activity, the right geography (specific region > regional > RoW/GLO), and the right technology. Where an exact match does not exist, a proxy dataset — a similar product or a broader geography — may be acceptable, but the proxy and its justification must be documented, and the pedigree scores will (correctly) reflect the weaker match as higher uncertainty.

10.2 Common Errors

11. ecoinvent & Corporate GHG Accounting

Although ecoinvent originates in product LCA, it is increasingly central to corporate Scope 3 accounting, where the same process-versus-spend trade-off from section 8 reappears. A spend-based Scope 3 screen uses EEIO intensities to estimate emissions from procurement value; a more accurate process-based estimate refines material hotspots using ecoinvent unit-process data. Mature inventories combine the two: EEIO for breadth, ecoinvent for the categories that matter most.

This connects directly to data-quality scoring in financial-sector accounting. The PCAF data-quality hierarchy rewards primary and process-based data over spend-based estimates — exactly the gradient from EEIO screening toward ecoinvent-grade process data. And as product-level carbon data moves between companies through frameworks like the WBCSD Pathfinder Framework, the provenance question — primary foreground data versus ecoinvent background, and which version and system model — becomes part of the exchanged record. The same discipline that makes an LCA reproducible is what makes Scope 3 data trustworthy when it crosses an organisational boundary. Throughout, results are expressed in CO₂-equivalent using the LCIA method’s chosen GWP set, which is why naming that method matters as much as naming the database.

12. Frequently Asked Questions

Related GreenCalculus References