LCI databases: choosing between ecoinvent, EF and sector datasets

In an LCA, foreground data describes your product and the background database describes the rest of the world. The second is often waved through as a default, yet it shifts the result noticeably and is a decision that has to be justified in the report.

ecoinvent and its system models

ecoinvent is the broadest and most widely used LCI database. The critical point is that the same version is published with several system models, and they give different results for the same activity.

• Allocation, cut-off by classification — recycled material burden stays with the first user; the default for EPD work
• Consequential — models market responses; for policy and large-scale change analysis
• APOS (allocation at the point of substitution) — distributes co-product burden
• Using two different system models in one study is a methodological error
• The report must state which version and which system model was used

The version matters too. Results for some activities change noticeably between ecoinvent versions; electricity mixes are updated and process data refreshed. Re-running a study from a year ago on a new version therefore produces different numbers — which is expected behaviour, not an error.

EF reference datasets

The European Commission's Environmental Footprint (EF) framework publishes its own reference datasets and characterisation factors. If you are doing PEF work or need PEFCR compliance, using those datasets is mandatory; they are not freely interchangeable with ecoinvent.

Sector datasets

Many trade associations publish their own average datasets: steel, aluminium, plastics, cement. These are typically weighted averages of member sites and can be more representative than generic database records. Their geographic scope needs careful checking, though — a European average may not represent Turkish production.

The no-mixing rule

The most common and hardest-to-spot error is combining records from different databases in one model. Each database has its own system boundary conventions, its own allocation rules and its own background chain. Mixed together, the resulting number belongs to no coherent methodology.

Where it is unavoidable — a material existing in only one source, say — that line should be flagged separately in the report and its effect shown in the sensitivity analysis.