Choosing a database and emission factors

Once you have an inventory, every line needs a number: kg CO₂-eq per kg of steel, per kWh, per tonne-kilometre. Those numbers come from a life cycle inventory (LCI) database. Picking the database, and then the right entry in it, is the LCA equivalent of choosing which price list to cost from. Choose badly and the model is tidy but wrong.

Datasets and emission factors

A database is a collection of datasets. Each dataset describes one process or product, such as “steel sheet, hot rolled, European average”, with its inputs and its emissions. When you run an impact method over a dataset (for example IPCC AR6 GWP₁₀₀), you get a single number per unit: the emission factor.

Some sources publish only emission factors. Full LCI databases publish the underlying flows too, so you can calculate other impact categories and see what drives the number. For an EPD or a formal LCA you usually need the full datasets. For a first screening, emission factors are often enough.

Free and licensed databases

Database	Who runs it	Coverage	Access
Federal LCA Commons (including USLCI)	A group of US federal agencies	US processes: energy, materials, transport, agriculture	Free
Agribalyse	ADEME (France)	Food and agricultural products consumed in France	Free under the Etalab Open Licence, with attribution; some distributions include ecoinvent data and then need an ecoinvent licence
EF reference packages	European Commission (JRC)	Data and methods for the EU Environmental Footprint	Terms depend on the use; read them first
ecoinvent	ecoinvent Association	Global, thousands of processes across most sectors	Paid licence
Software-vendor databases (for example Sphera’s)	The vendor	Broad, industry-focused	Licensed with the software

Free databases are genuinely useful. They are narrower than the big licensed ones, so expect gaps, especially outside their home region. See also free data and tools.

Tip

If a PCR, a customer or a programme operator names a database and version, use exactly that. It is not a suggestion.

Matching a dataset to an inventory line

For each line, look for the dataset that best matches on three things:

1. Technology. Is it the same material made the same way? Recycled aluminium and primary aluminium have very different footprints. So do blast-furnace steel and electric-arc-furnace steel.
2. Geography. Where was it made? Electricity is the extreme case: the same kWh has a very different footprint in a coal-heavy grid and a hydro-heavy one. Use the grid where the energy is consumed.
3. Time. How recent is the dataset? Energy mixes and industrial processes change. Prefer data that represents the period you are studying.

When there is no good match, use the closest proxy, say that it is a proxy, and flag it for the sensitivity check. That is exactly the kind of assumption a reviewer wants to see written down.

Watch for double counting

In some databases, a “market” dataset already includes average transport to the consumer region, and sometimes losses. If you then add your own transport on top, you count it twice. Read the dataset description before adding transport.

Data quality

LCA 101 lesson 5 lists the data quality aspects ISO 14044 asks you to describe. In practice, for each dataset you pick, record:

• its name, database and version;
• how well it matches on technology, geography and time (good, fair, poor);
• whether it is a proxy, and for what.

This is your evidence. It also tells you where to spend effort next: a hotspot with a poor-quality dataset is the first place to look for better data, often from a supplier (lesson 3).

System models

Databases make methodological choices before you ever see a number, especially on how to share burdens between co-products and how to handle recycling. ecoinvent, for example, publishes its data in several system models, including “allocation, cut-off by classification” (a cut-off approach to recycling like the one in lesson 1), a variant aligned with EN 15804 for construction EPDs, and a consequential model. Two datasets from different system models are not directly comparable. Pick the system model your standard requires, and stay with it.

Mixing databases is possible but needs care: flow names, units, system models and reference years can all differ. A flow that fails to map can silently disappear.

Check the licence before you share or upload

This is the part people skip, and it can cause real trouble.

Licensed databases come with an end-user licence agreement (EULA) that sets what you may do with the data and with results calculated from it. The ecoinvent EULA, for example, allows calculated results to be shared statically, such as in a PDF report, but does not allow them to be shared with non-licence holders in an automated way through a tool or application. Other databases have their own terms. Free data also has conditions: Agribalyse, for example, requires you to credit the source and version.

Before you:

• paste licensed datasets or factors into an AI assistant or any online tool,
• put factors into a calculator or spreadsheet that others will use, or
• publish or send results to a customer,

read the licence and check that the use is allowed. If in doubt, ask the data provider. The learnLCA skills are built to work with free data or with factors you are entitled to use, and to cite every one.

Exercise

Your inventory has these lines. For each, which candidate dataset would you choose, and why?

1. Aluminium extrusion made by your supplier in Norway from 75% post-consumer scrap. Candidates: (a) primary aluminium, global average; (b) aluminium, secondary, from old scrap, Europe; (c) your supplier’s EPD for the extrusion.
2. Grid electricity for your assembly plant in Poland. Candidates: (a) electricity, medium voltage, Europe average; (b) electricity, medium voltage, Poland.
3. A specialist adhesive with no match in any database. Candidates: (a) leave it out, it is only 0.2% of the mass; (b) use a generic “adhesive” or “chemical, organic” dataset as a proxy.

Answer

1. (c) if the EPD is valid, covers the product you buy and uses a compatible standard (lesson 3 shows how to check). Otherwise (b), with a note that it represents recycled content only approximately. Never (a): primary aluminium would badly overstate a mostly recycled product.
2. (b). Electricity should represent the grid where it is used. A European average would hide a large difference.
3. (b), labelled as a proxy, unless your standard’s cut-off rules explicitly allow the exclusion and you have checked it is not environmentally significant. Mass is not the only test: a small amount of a very intensive chemical can matter.

What you should now be able to do

• Explain the difference between a dataset and an emission factor.
• Name free and licensed databases and what each covers.
• Match a dataset to an inventory line on technology, geography and time, and record proxies.
• Explain why system models matter when combining data.
• Check a database licence before sharing results or uploading data to an AI tool.

Sources

• Federal LCA Commons, lcacommons.gov.
• ADEME, AGRIBALYSE documentation: data access and conditions of use.
• ecoinvent, End User Licence Agreement (EULA) and system models.
• European Commission JRC, European Platform on LCA.
• ISO 14044:2006, data quality requirements, iso.org.

Checked: 2026-09-29