Impact assessment alignment with ecoinvent 3.13

For most impact assessment methods, LCIA results calculated from ecoinvent 3.13 life cycle inventories in Earthster are closely aligned with results obtained directly from ecoinvent via ecoQuery (see below). The remaining categories differ for a small number of well-understood reasons, such as:

None of these reflect calculation errors — they reflect either deliberate, documented modeling choices or genuine gaps in the underlying LCIA methods themselves.

This page documents these differences in detail to ensure full transparency and reproducibility.

Earthster's calculated impacts are benchmarked against ecoinvent's own ecoQuery-reported results, per impact category, across all 27,400 ecoinvent 3.13 cycles, to isolate implementation differences from inventory differences.

Differences therefore reflect LCIA implementation choices only (e.g. characterization factors, context mapping, proxy selection), and not differences in inventory data. For each impact category, differences are summarized statistically (mean, median, and maximum), allowing both typical and extreme deviations to be identified.

The benchmark results fall into two broad groups: impact categories with perfect alignment, where Earthster and ecoQuery produce identical results for all datasets, and impact categories with documented differences, where systematic and explainable deviations occur.

Table 1. Impact categories where Earthster and ecoQuery produce identical results for all 27,400 ecoinvent 3.13 cycles, grouped by method.

Not all impact categories were included in the benchmark comparison. Exclusions were applied deliberately to avoid misleading or redundant comparisons, and are explained below.

Endpoint categories. Endpoint categories (e.g. ReCiPe 2016 Endpoint H/I/E) were excluded because they are aggregations of midpoint categories. Their benchmark behavior therefore directly reflects that of their underlying midpoints and does not provide additional insight. An exception was made for ImpactWorld+ endpoint categories: because ecoinvent does not provide benchmarks for most ImpactWorld+ midpoint categories, two endpoints were benchmarked as proxies for the non-benchmarked midpoints.

Water consumption–related categories. All water consumption–type categories were excluded from the benchmarks. Ecoinvent applies a different accounting convention for these categories than the one used elsewhere in the comparison, making direct numerical comparison with Earthster methodologically inconsistent and not meaningful (see Reason 6 below for the specifics of this convention gap).

Methods and categories not available in ecoinvent. Impact methods or categories not provided through ecoinvent ecoQuery were also excluded, including:

Since EN15804+A2 and ECI are based on EF 3.1, their numerical behavior and reasons for divergence closely mirror those observed for EF 3.1 categories, and are documented accordingly.

Impact methods ReCiPe and TRACI leave ambiguities in how characterization factors should be assigned to specific emission contexts.

For example:

Because of these gaps, different databases and software systems apply different mapping rules. Earthster follows the context mapping curated by FEDEFL, which aims to be conservative and method-consistent.

Example context mapping differences

In total, 251 substances are affected by these context mapping differences.

ReCiPe has no explicit guidance for low-population-density air compartments (e.g. rural troposphere). Ecoinvent excludes these emissions from its own characterization; Earthster applies a conservative interpretation and characterizes them, consistent with EF3.1's own practice. This produces higher impacts in earthster for the affected datasets, and is the main driver of ReCiPe's Ecosystem/Human damage ozone formation, Stratospheric ozone depletion, and Terrestrial acidification differences.

Many inventory substances do not appear explicitly in LCIA methods. Both Earthster and ecoinvent address this gap using proxy mapping — approximating an uncharacterized substance with a similar one that the method does characterize — but the two systems differ in coverage and specificity.

3.1 Earthster maps to a proxy, ecoinvent leaves the substance unmapped

Both Earthster and ecoinvent extend the published methods with their own proxy mappings where a substance isn't natively covered — this is a normal, expected part of implementing an LCIA method against a real inventory. Earthster's extensions cover every substance ecoinvent itself maps, plus additional substances that ecoinvent leaves unmapped, resulting in more complete coverage overall (there remain some substances neither system maps). Where ecoinvent leaves a relevant substance unmapped, this shows up as a difference simply because Earthster reports a value and ecoinvent reports none — not because either implementation is incorrect.

For example, Earthster characterizes Kaolinite via a documented approximation to ReCiPe's "Ball clay" flow. ReCiPe's own method defines no native characterization factor for Kaolinite at all, so ecoinvent reports none, while Earthster's proxy value shows up as a difference.

3.2 Earthster maps to a more specific proxy

Where both systems use a proxy for the same substance, Earthster generally selects the one that is most chemically or functionally specific, rather than the broadest available match — in general, Earthster tends to map to more specific substances than ecoinvent does. For example, ecoinvent may map a substance to a generic category such as "herbicides," while Earthster maps the same substance to the specific herbicide that best reflects its chemical behavior. In some cases the method provides a characterization factor for the exact substance; Earthster uses it directly, while ecoinvent assigns the value of a broader proxy instead.

This distinction explains part of the systematic differences observed between Earthster and ecoinvent in several toxicity- and resource-use– related categories.

We found cases where ecoinvent's own CF tables lack a value for a substance, even though the published LCIA method itself provides a characterization factor for it. For example, EF3.1, ReCiPe, and TRACI all define CFs for a number of substances that ecoinvent's implementation simply omits. Earthster assigns the method's own documented value in these cases, rather than under-reporting impacts for the affected substances.

Conversely, a small number of ecoinvent CFs (for example, TRACI Freshwater Ecotoxicity's ammonium, BOD, COD, nitrate, and NO₂ entries) have no traceable source in TRACI's own published documentation, and earthster excludes them on that basis.

Ecoinvent assumes an energy content of 560,000 MJ/kg for uranium; earthster follows the FEDEFL/World Nuclear Association standard of 500,000 MJ/kg. This produces roughly 12% lower impacts in earthster for energy-related categories — CED's nuclear categories, EF3.1's fossil resource use, and CML's abiotic depletion.

Interpretation. The true energy content of uranium is not a fixed physical constant. It depends on several factors, including reactor technology, fuel burn-up rates, enrichment levels, and the efficiency of converting nuclear fuel into usable electricity. Ecoinvent's higher value is derived from efficiency assumptions representative of specific national reactor fleets (notably Germany). While valid for those contexts, this assumption may not be representative of global average conditions. Earthster applies the FEDEFL assumption to avoid embedding country-specific reactor efficiencies into globally scoped LCIA results. Earthster currently retains the FEDEFL assumption, while acknowledging the inherent uncertainty in uranium energy conversion and the legitimacy of alternative values.

Some LCIA methods define different characterization factors for synonymous substances (for example, Mecoprop-P versus (R)-Mecoprop in EF3.1), producing different results depending on which synonym a given exchange is matched against.

Ecoinvent 3.12 introduces minor modifications to a small number of characterization factors in the IPCC 2021 and EF methods. These differences originate from inconsistencies between the main IPCC report and its supplementary materials, which do not always report identical values.

The resulting impact differences are generally very small (typically below 3%, and limited to a small number of exchanges).

Earthster applies the original characterization factors as documented in the main IPCC 2021 report and does not adopt these minor adjustments. Given their limited magnitude and unclear methodological justification, Earthster will retain the original values until future method revisions (e.g. EF 3.1) explicitly resolve these discrepancies.

ReCiPe distinguishes between Methane (34 kg CO₂-eq.) and Fossil methane (36 kg CO₂-eq.), but does not explicitly define a characterization factor for Methane, land use.

In Earthster, Methane, land use is treated as Fossil methane, consistent with the approach used in IPCC, EF, and ImpactWorld+. This reflects the assumption that land-use methane emissions originate from long-term carbon stocks in soils.

In ecoinvent, Methane, land use is instead mapped to Methane (biogenic). This difference in proxy selection leads to a small discrepancy of approximately 5% in calculated climate change impacts for land-use methane emissions, with Earthster reporting slightly higher impacts.

Earthster considers alignment with other major LCIA frameworks to be a more consistent and transparent choice than ecoinvent's proxy mapping in this case.