Docs / What was changed in each source

Changes to AGRIBALYSE 3.2

Part of What was changed in each source. What is still open about AGRIBALYSE 3.2 is on the issue tracker under the label agribalyse.

AGRIBALYSE is ADEME's French agricultural database. It arrives as a SimaPro process export, AGB32_final.CSV — 20,440 processes, 6,535,117 exchanges with the environment — so its flows are the rows those exchanges name: 5,528 of them, of which 5,485 reach matching. The other 43 — the Final waste flows compartment — are extracted and then refused on the record (#855): a SimaPro final-waste row is an accounting device at the product-system boundary, not an exchange with the environment, and the consensus vocabulary has no waste compartment. Each refusal is published on this list's correspondence under brightway:excludedSourceConcept, so a consumer counting the vendor's rows against the associations sees a decision, not an oversight.

The file ships no flow identifier of any kind (Export platform IDs: No). Every row is therefore named the way the export writes it — name, compartment and sub-compartment, unit — and the uuids this project gives AGRIBALYSE rows are derived from those same fields. The unit is the one the trailing substance blocks inventory the flow in, not the unit of any single exchange: the 324 radionuclides are inventoried in kBq whatever scale a process happened to state.

Because AGRIBALYSE's names carry SimaPro's habits — Water, river, FR, Gas, natural/m3 — the list is registered with simapro_origin, so it receives every correction in simapro-lineage-manual-fixes.json, the unit-suffix splitter and the SimaPro name rules, exactly as BAFU 2026 v1 and Stepwise 2006 do. The geography splitter is different: this page used to say the list received it too, and no build ran it for any list but BAFU, whose extractor calls it itself (#887). Since the list's manifest says simapro.geography_split: "load", the split runs as the flows are loaded — Water, river, FR goes forward as Water, river in FR, the uuid and the vendor file untouched, and the shipped spelling kept as a name the merge and the published list still answer to. Its compartments go out as the vendor wrote them; where they belong is stated in context-manual-mapping.json, whose rules for this list agree with BAFU's on the eighteen compartments the two spell identically.

Not yet merged

AGRIBALYSE 3.2 is registered and extracted, and merges last (priority 400) when a build names it. No documented shared build includes it yet, so this page records no merge outcomes: no row has been sent anywhere by hand, no registry number corrected, and no name split across two substances. Measured with the read-only lookup against the 2026-08-30 build of f113244, 4,077 of its 5,528 distinct flows already resolve to a consensus flow (73.8%; 75.4% weighted by exchange traffic). The remainder are catalogued in the issues under the label above — land classes, water materials, river ties, and a small set of spelling variants.

Particle rows read into a size window

AGRIBALYSE ships 27 airborne-particle rows, and every one is read into one of the seven size windows the list publishes (Particles (PM10), Particles (PM2.5), and so on). Twenty-five are BAFU's spellings and take BAFU's readings. Two are the list's own, and were read on 2 September 2026 (#921):

Both minted a substance of their own, with no factor, on the first five-list builds. A particle row that no reading covers now stops the merge instead of doing that quietly.

Carbon dioxide that did not say where it came from

Glutamate {CN} U releases 1.9175 tonnes of Carbon dioxide to air — the whole carbon dioxide of a Chinese monosodium glutamate plant. The row names no origin: not fossil, not biogenic, not land transformation, registry number 124-38-9 and nothing else. Carbon dioxide of unstated origin is a substance EF 3.1 ships, but no method characterises it, so those tonnes scored nothing for climate change (#914).

AGRIBALYSE ships the plain row twice, and the two are decided separately, because their datasets have nothing in common:

A third row did state an origin, and this project was not reading it. Carbon dioxide, peat oxidation reached the same unstated-origin substance, because peat oxidation was not one of the phrases the list recognises as saying where carbon came from. It is now, and it is read as a land use change: draining a peatland and letting the peat oxidise is a change of land use, and AGRIBALYSE books it as one — its own Peat degradation emissions on grassland, per kg CO2 {GLO} ships the release as Carbon dioxide, land transformation, and its soil carbon datasets say their land-transformation row is "excluding peat degradation", the same accounting split out rather than a different kind of carbon. Under EF the total climate-change score is the same either way; what the reading decides is whether the mass lands in Climate change-Fossil or in Climate change-Land use and land use change.

Reading the phrase rather than curating the row answers Methane, peat oxidation in the same line — six datasets with the same defect, which #914 does not mention. Those three names are every name carrying the phrase across EF 3.1, all five ecoinvent releases, BAFU 2026-v1, Stepwise 2006 and AGRIBALYSE 3.2.

The third, Dinitrogen monoxide, peat oxidation, is the one the rule cannot finish, and it is worth reading before writing a rule of the same shape. The reading is right about it — the nitrous oxide comes off the same drained peat — but EF ships no nitrous oxide of land-use-change origin to move it to, and qualifying the name is by itself enough to take the row off the flow it was already on. Its registry number, 10024-97-2, reaches two substances: nitrous oxide, and the correction flow for nitrous oxide emitted after a hundred years. What separated them was the row's own name sitting on nitrous oxide as a synonym, put there by the row itself; a name that carries an origin is filed elsewhere, the synonym is not carried, and a row that was correctly placed is published nowhere. So this one is sent by hand, in agribalyse-3.2-match-overrides.json, to the flow it already had.

Registry numbers corrected

AGRIBALYSE's flow list is SimaPro's copy of EF 3.1's, and on a few names it carries a number EF's own row had wrong or that names the element where the name states an ion. agribalyse-3.2-manual-fixes.json corrects each, keyed on the name and the number the vendor wrote, so a row already carrying the right number is left alone.

The file holds the earlier corrections too — Vanadium (V) and Arsenic (V) under the wrong ions' numbers, and the bare graded-ore rows given their elements' numbers (#876).

Curated targets

agribalyse-3.2-match-overrides.json sends a row where the merge would not put it on its own. Beside the water and energy rows described above, three rows named Carbon, organic, in soil or biomass stock go to EF 3.1's Carbon, Organic, In Soil Or Biomass Stock (#915). The rows carry the element's number, 7440-44-0, which #588 ruled groups nothing, and without the target they fell through to the tail of their name and minted a Carbon resource in the ground under elemental carbon: 2,278 datasets read as digging the element out of the ground, Mine infrastructure, phosphate rock {GLO} alone 31,640 kg of it. ecoinvent's and BAFU's rows of the same name are on the stock flow, ecoinvent's Increase row by the same kind of target.

The three differ only in the compartment the exporter wrote. The in ground one is ecoinvent's, carried through 2,307 datasets. The bare Resources one is AGRIBALYSE's own soil carbon accounting, and it puts both directions on a single row — positive when carbon leaves the soil, negative when it is stored, as the vendor writes on the row itself — so it is folded into one flow the way ecoinvent's Increase and Decrease rows are (#588). The third occurs once, in Peat degradation emissions on grassland, per kg CO2 {GLO} - Adapted from WFLDB, and that dataset shows why none of the three is an emission: it already emits the carbon. Beside the 0.2727 kg on the stock row it ships 1 kg of Carbon dioxide, land transformation, and 0.2727 kg is 12/44 of that — the same carbon, booked a second time for the resource accounting, which the dataset's own comment says it is doing. Every dataset carrying a stock row does this: 0.017 kg of stock carbon against 0.0623 kg of CO2 in the annual-crops land use change dataset, −0.0110 kg against 0.0403 kg of Carbon dioxide, to soil or biomass stock in the permanent-meadows one, always 44/12 apart (AGB32_final.CSV). Sending a stock row to a CO2 flow would count that carbon twice. EF 3.1 characterises the stock flow at nothing, which is what a ledger entry should score.

Finding the original row

The pages' convention for a list without vendor identifiers applies: rows are quoted by name, compartment and unit as AGB32_final.CSV writes them, and the project's uuids are stable across builds but are not ADEME's.