EEA Industrial Emissions Portal - landscape for Facility Foundry v1 onboarding

Phase 2 / Workstream B.2 landscape, written 2026-05-10. Reference for the EEA fetcher built against the v1 schema + ADR 008/009 facility-side synthesis + ADR 010/011 lineage substrate. Cross-references: docs/adr/008-cross-source-facility-identity-resolution.md, docs/adr/010-lineage-records-share-canonical-pipeline.md, docs/LINEAGE.md, docs/DATASOURCES.md, docs/phases/02-multi-jurisdictional-and-lineage.md.

Live-data observations (verified 2026-05-10 against the EEA Datashare WebDAV share at https://sdi.eea.europa.eu/datashare/public.php/webdav/). Pre-research assumptions in docs/DATASOURCES.md about a "~30,000+ pollutant-release sites" universe are corrected here: the IED installation register is a strict superset of the historical E-PRTR universe and is materially larger (70K installations / 64K parent facilities). Two pre-research assumptions also failed against the live data and are noted inline below: there is no usable cross-Member-State operator field in the user-friendly distribution, and the canonical distribution is now a versioned snapshot (v16, vintage 2026-02-20), not a daily-rolling product.

1. The portal supersedes the legacy E-PRTR series. Under the 2024 Industrial Emissions Directive update (IED 2.0) the legacy E-PRTR pollutant-release register and the new IED installation register are merged into a single dataset under EEA dataset code eea_industrial-reporting-e-prtr_s (catalogue UUID 9405f714-8015-4b5b-a63c-280b82861b3d). The current published version is v16, snapshot 20 February 2026, covering E-PRTR facility releases 2007-2024, IED installations 2017-2024, and Large Combustion Plants (LCPs) 2016-2024. v15 (Feb 2025) and earlier remain co-published in the same share for reproducibility.

2. The bulk download is a Nextcloud datashare folder, not a direct file URL. Downloads route via:

   https://sdi.eea.europa.eu/datashare/s/sptXqwkQr5g7Bp5
   

   Programmatic enumeration uses WebDAV PROPFIND against https://sdi.eea.europa.eu/datashare/public.php/webdav/ with HTTP basic auth <share-token>:<empty> (the token is the public share key, not a credential - Nextcloud expects the URL share token in the Authorization header for WebDAV requests against public shares). The full share is 1.4 TB across all historical versions + all reporting flows; FF only pulls the current-version v16 tabular bundle (147 MB compressed → 695 MB uncompressed).

3. Concrete download URLs for v16 (current vintage):

   Artifact	URL	Size
   Tabular CSV bundle	…/webdav/eea_t_ied-eprtr_p_2007-2024_v16_r00/User%20friendly%20.csv%20files.zip	147 MB
   Spatial GeoPackage	…/webdav/eea_v_4326_10_m_ied-eprtr_p_2007-2024_v16_r00/ied_SiteMap.gpkg	584 MB
   Schema metadata PDF	…/webdav/eea_t_ied-eprtr_p_2007-2024_v16_r00/EEA_Industrial_Reporting_Metadata_v16.pdf	1.4 MB
   Full Access DB	…/webdav/eea_t_ied-eprtr_p_2007-2024_v16_r00/Industrial_dataset_v_16_2026_02_16.accdb	2.03 GB

   The tabular CSV bundle is the primary FF artifact - it carries the 16 user-friendly tables that mirror what the Access DB exposes via its tabular surface. The Access DB has additional tables (operator details, permit administrative metadata) that didn't get exported into the CSV bundle but are accessible via mdbtools if a future slice needs them. Spatial GeoPackage is optional and only needed if FF wants polygon footprints instead of the lon/lat carried in F6_1 directly.

4. The CSV bundle is 16 tables grouped by reporting flow:

   Code	Filename	Rows	Description
   F1.4	F1_4_Air_Releases_Facilities.csv	372,206	Air emissions per facility × pollutant × year
   F2.4	F2_4_Water_Releases_Facilities.csv	-	Water releases per facility
   F3.2	F3_2_Transfers_Facilities.csv	-	Pollutant transfers per facility
   F4.2	F4_2_WasteTransfers_Facilities.csv	853,413	Waste transfers per facility
   F5.2	F5_2_LCP_Energy_Emissions.csv	-	Large Combustion Plant energy + emissions
   F6.1	F6_1_IED_Installations.csv	477,644	IED installation register - facility-axis foundation
   F7.1	F7_1_IED_WI_coWI.csv	-	Waste / co-incineration plants

   Plus aggregate tables (national / sectoral / by-Annex-I-activity) that FF doesn't consume directly. F6.1 is the load-bearing file for the Facility universe; F1 - F5 are useful for emissions-side FacilityAttribute enrichment but are not required to land Workstream B.2's exit criterion (multi-region facility coverage).

5. F6.1 schema is INSPIRE-anchored, not Member-State-local. The 31-column schema (verified by reading the live header):

   PublicationDate, CountryName, reportingYear,
   parent_facilityInspireId, EPRTRAnnexIMainActivity,
   InstallationInspireId, installationName,
   installationNameConfidentialityReason, City_of_Facility,
   Longitude, Latitude, addressConfidentialityReason,
   installationStatus, IEDAnnexIMainActivity, IEDMainActivityName,
   baselineReportIndicator, siteVisitNumber, siteVisitURL,
   dateOfStartOfOperation, publicEmissionMonitoring,
   publicEmissionMonitoringURL, permitGranted, dateOfGranting,
   permitReconsidered, permitUpdated, dateOfLastUpdate, permitURL,
   RelevantChapter, BATConclusion, BATAEL, BATAELStricterCondition
   

   Both parent_facilityInspireId and InstallationInspireId are full URI-form INSPIRE IDs, e.g. https://registry.gdi-de.org/id/de.rp.inspire.pf.bube-eureg/5000671_4731_0. The host varies by Member State (registry.gdi-de.org for Germany, etc.), but the URI is opaque and stable across years

• that's the load-bearing identifier for ADR 008 cross-source matching against e.g. UK EA permit refs (B.3) and Wikidata P9758 (EPRTR Facility ID, when present).

1. Facility-axis cardinality, by-country, status:

   • 477,644 rows in F6.1 (one row per installation × reporting year, 2017-2024 = 8 years).
   • 70,847 unique InstallationInspireIds - the installation universe.
   • 64,041 unique parent_facilityInspireIds - the facility-axis universe (FF's entity for ADR 008 matching). The 6.8K-installation surplus reflects multi-installation facilities (e.g., a chemical complex with separate permits per process unit).
   • Top countries: Germany (105K rows), France (56K), Spain (54K), Italy (53K), Netherlands (33K), then a long tail Finland/Hungary/Sweden/Romania/Portugal at 6-11K each. Coverage spans EU-27 + EEA non-EU contributors (NO, IS, LI, CH, UK historical) - multi-region as advertised.
   • installationStatus distribution: functional 95.1 % / disused 2.9 % / decommissioned 1.2 % / notRegulated 0.9 %. Filter to non-notRegulated for the active facility set; carry disused / decommissioned as LifecycleEvent(kind="decommissioned").

2. Operator-axis is unavailable in the user-friendly bundle. Pre-research assumed an operator_name field would be present per the legacy E-PRTR Article 7 reporting requirement; this failed against the live v16 CSVs. Neither F6.1 nor the facility-axis emission tables (F1.4 / F2.4 / F4.2) carry an operator string. The IED 2018 Implementing Decision moved operator-details reporting into a separate dataflow that the EEA chose not to expose in the user-friendly bundle. Two consequences:

   • FF emits no Entity / OperatorReference / Relationship records from the EEA fetcher in B.2. EEA is a facility-only source - different from MSHA / MRDS / Wikidata which all carry operator names.
   • Operator-axis enrichment for EU facilities goes via the A.6 same_entity_as matcher operating on facilityName × CountryName against entities surfaced by Wikidata's operator chain (P137/P127/P749) on the same facility, after ADR 008 spatial-+-name match resolves Wikidata's facility QID to the EEA InspireId. Two-step: ADR 008 facility match → Wikidata operator chain → Workstream A lineage. EEA never contributes a hard anchor to the operator graph.

   Future B.2.x extension could parse the .accdb's operator tables via mdbtools, but that's a follow-up - not in scope for closing Phase 2.

3. Sector classification: dual-axis EPRTR + IED. Each row carries both EPRTRAnnexIMainActivity (legacy E-PRTR Annex I sectors, e.g. "4(a)(ii)" = chemical industry / organic chemicals) and IEDAnnexIMainActivity + IEDMainActivityName (the more granular IED Annex I taxonomy, e.g. "4.1(b)" with the descriptive sub-activity name). FF emits two Classification records per installation:

   • taxonomy="eprtr_annex_i", code = the EPRTR designator (resolves to a sector-level commodity hint per the EPRTR sector code table)
   • taxonomy="ied_annex_i", code = the IED activity code

   The commodity-axis mapping (EPRTR/IED → USGS commodity code, matching the Wikidata B.1.5 pattern) is a B.2.x slice; B.2.0 just records the schema location.

4. Lifecycle events are richly populated. Per F6.1 row:

   • dateOfStartOfOperation - first commissioning date (sentinel 1000-01-01 and 1900-01-01 appear when reporters didn't have the actual date - filter at parse time).
   • permitGranted / dateOfGranting - original permit issuance.
   • permitReconsidered / permitUpdated / dateOfLastUpdate - permit-amendment timeline; useful for tracking BAT-conclusion compliance updates.
   • Each maps cleanly to LifecycleEvent(kind=...) records: commissioned from dateOfStartOfOperation, permit_granted from dateOfGranting, etc. Reporting-year itself is not a lifecycle event - it's metadata about when the row was reported.

5. License: free / EU open-data. Per the EEA's standard re-use terms (Commission Decision 2011/833/EU), bulk redistribution of derived data is permitted with attribution. No registration / API key / rate limit on the public Nextcloud share. FF's Provenance.notes should include the canonical citation: "EEA - Industrial Emissions Directive 2010/75/EU and E-PRTR Regulation (EC) No 166/2006, dataset version 16, 2026-02-20".

Coverage gap analysis vs Phase 2 exit criteria

EEA satisfies the multi-region facility coverage half of the Phase 2 exit criteria. Numbers compared to existing fetchers:

Total post-EEA facility universe: ~600K source-tier facilities. EU coverage moves from "Wikidata-only and sparse" to dense at the facility-permit level; expect significant ADR 008 matches between Wikidata's EU mining/power facilities and EEA's IED installations.

Cross-source matching expectations:

• EEA × EPA FRS: zero overlap (geographic disjoint EU vs US).
• EEA × MSHA: zero overlap (EU has no MSHA-equivalent register).
• EEA × USGS MRDS: low overlap - MRDS's non-US coverage is mostly Latin America + Asia-Pacific; EU MRDS coverage is spotty and mostly historical mines.
• EEA × Wikidata: this is the load-bearing match. Wikidata's P9758 (EPRTR Facility ID property) was populated on only ~9 entities in the B.1.0 universe survey, so direct-join is not the path; ADR 008 spatial-+-name matching is. Expect HIGH matches for the well-known EU facilities (BASF Ludwigshafen, TotalEnergies refineries, ArcelorMittal steel plants) and MEDIUM matches for the long tail.
• EEA × GLEIF: no direct match - EEA carries facility-axis identifiers, GLEIF carries entity-axis (LEI). Indirect bridge via Wikidata operator chains landing on Wikidata operators carrying P1278 (LEI) - same path as the broader operator-axis story. EEA's zero-operator-axis means it can't bridge to GLEIF on its own.

Fetcher design implications

Cache strategy: BULK_FILE. One zip per vintage; vintage = the metadata-published snapshot date (e.g. 2026-02-20 for v16). Honor the same idempotent download_stream + _extract_* skip- if-cached contract used by GLEIF / CorpWatch. Cache directory layout: data/raw/eea/<vintage>/User-friendly-csvs.zip → extracted to siblings.

Vintage anchor: the v16 metadata XML (Industrial_Emissions_Directive_..._657ac3cb-affa-4295-a4a9-27b4f539adab.xml) carries the canonical publication date. New versions land roughly annually (v15 = Feb 2025; v16 = Feb 2026). FF pins to a specific version-string (v16) and a snapshot date the same way GLEIF pins to a content_date. Operator runs --force to roll the pin forward.

Record kinds emitted:

• Facility - one per parent_facilityInspireId (deduped across reporting years); coordinate from the latest reporting-year row's Longitude/Latitude.
• FacilityIdentifier(scheme="eea_inspire_id", value=URI) - the load-bearing cross-source key. Add scheme constant to models/_registry_ids.py (SCHEME_EEA_INSPIRE_ID). Also a per-installation FacilityIdentifier(scheme="eea_inspire_installation_id", …) for the installation-axis (B.2.x; not blocking).
• Classification(taxonomy="eprtr_annex_i", code=...) and Classification(taxonomy="ied_annex_i", code=...) per installation; both rendered as MEDIUM-confidence sector hints.
• LifecycleEvent(kind="commissioned", date=dateOfStartOfOperation) with sentinel-date filter (< 1950-01-01 → drop). LifecycleEvent(kind="permit_granted", date=dateOfGranting) conditional on permitGranted=Yes.
• FacilityAttribute(attribute_name="installation_status", …) carrying the status enum (functional/disused/decommissioned).
• No Entity / OperatorReference / Relationship - see the operator-axis gap above.

Memory footprint: F6.1 is 319 MB uncompressed CSV, ~480K rows × 31 cols. Polars scan_csv + lazy collect handles this in well under 1 GB RSS. Streaming row-by-row not required - this is a normal-sized CSV.

Confidence tiering:

• HIGH - installationStatus=functional, INSPIRE ID present, coordinates non-null, dateOfStartOfOperation non-sentinel.
• MEDIUM - anything missing one of the above (typical for the long tail of disused / partially-redacted records).
• LOW - addressConfidentialityReason populated (operator invoked confidentiality on the address; coords likely also jittered - keep but de-rank).

What B.2 sub-steps look like

Following the A.7 / B.1 pattern:

• B.2.0 - landscape doc (this file).
• B.2.1 - fetcher skeleton + BULK_FILE cache strategy + download/extract path. Tests: skeleton + bulk-source flag + metadata XML pin.
• B.2.2 - F6.1 parser → Facility + FacilityIdentifier + Classification + LifecycleEvent emit. Synthetic-CSV unit tests.
• B.2.3 - Live ingest end-to-end + ADR 008 spatial-name match against Wikidata operators on the EU facility set. Spot-check BASF Ludwigshafen, TotalEnergies Antwerp, ArcelorMittal Bremen for HIGH matches via the synthesis facility-side matcher. (No A.6 lineage spot-checks - EEA doesn't surface operators.)

Optional follow-ups (Phase 3-eligible):

• B.2.x.commodity - EPRTR / IED Annex-I → USGS commodity code mapping, mirroring Wikidata B.1.5.
• B.2.x.emissions - Stream F1 - F5 emissions tables into per-pollutant FacilityAttribute records (release amounts, by year). Strong fit for downstream concentration analysis.
• B.2.x.operator - Parse the .accdb's operator tables via mdbtools to recover the operator-axis. Restores EEA→Wikidata lineage bridge directness.

Pre-research assumptions corrected

1. "~30,000+ pollutant-release sites" is wrong. That was the legacy E-PRTR pollutant-reporter count from the pre-IED-2.0 merge. The unified IED+E-PRTR register is ~64K facilities / 71K installations.

2. "E-PRTR series and the IED installation register are separate" is wrong post-merge. v15+ are unified; the legacy E-PRTR-only series is no longer published as a standalone product. Historical E-PRTR-formatted CSVs (v9 etc.) remain accessible at the older data-and-maps/data/... URLs but are frozen in time.

3. "REST API + SPARQL endpoint" is mostly wrong. The industry.eea.europa.eu portal exposes interactive query tooling backed by what looks like an internal SQL surface, but no public REST/SPARQL endpoint is documented or stable enough to anchor a fetcher on. Bulk-file via WebDAV is the only durable access path.

4. "Operator field present per Member State, varying in formality" turned out to be wrong for the user-friendly distribution. Operator info exists in the .accdb, not in the CSV bundle. The operator-axis gap is a real coverage limitation of the EEA path.