The State of European Renewable-Siting Data - 2026

Rasmus Pedersen, Northing · 2 September 2026

31 countries. More than 2,000 probe requests against live services. Machine-accessible grades probed live and independently re-verified; desk-only grades capped and marked as such. 2026 edition, v1. I want your corrections (see the end).

Every solar park, battery site, and Power-to-X plant in Europe starts with the same six questions: who owns the land, what may be built on it, what's protected, where is the grid, is there capacity left, and who got permits nearby? Six datasets, graded here side by side across all 31 countries.

I spent the summer looking for them. It took longer than expected. I didn't just read catalogue descriptions. I connected to the services, counted the records, and graded what a developer's screening workflow can actually use. This index is the result: every EU country plus Norway, the UK, Switzerland and Iceland, graded on the six core siting datasets.

The five findings

1. Two countries publish all six core datasets nationally and machine-readable: France and Slovenia. In July this list had five names. Estonia, Malta and the Netherlands left it when I re-derived both grid axes blind for this revision. Estonia's connection-capacity map carries bands with no megawatt field behind them. Malta's numeric feed covers the low-voltage tier only. The Netherlands' line geometry could not be confirmed national on the operators I could verify. The corrections log has each case. Slovenia is a small state with one national system per dataset. Small does not mean thin: Slovenia's national building-permit register held 296,108 cases when I probed it in August 2026. It means simple: a handful of national authorities instead of a federation of regional ones. France is the one that does not fit. Despite its permitting reputation, it runs six nationally accessible machine-readable channels: national cadastre, national zoning, national protected-area registers, grid geometry down to distribution level, a live per-substation hosting-capacity feed spanning the TSO plus distributors covering nearly the whole network, and a national register of 138,000 classified installations (observed August 2026) that doubles as its environmental-permit trail. That register covers wind; ground-mount solar is largely absent from it, because solar parks are not classified installations. Their permits sit in the other half: a national permit-level dataset of building permits, published as open data, which is the channel France's permit grade rests on. Complete does not mean gapless: the per-country deep-dives record the residual coverage tails inside these six A-grade rows too.

2. The megawatts concentrate where the data is hardest. Three of Europe's five biggest solar markets sit in the bottom two bands of this index. Germany, Europe's largest solar market by installed capacity, has two clean national datasets. One is protected areas. The other, since this revision, is permits, through the federal register that links every registered generation unit to the permit that authorised it. Everything else fragments across 16 Länder or four TSO control areas, or lacks a machine-readable channel entirely. Direct checks of all four TSOs found one banded substation table, one colour map and two walls, but no numeric capacity or queue endpoint; the distributors that own most of the 110 kV tier publish no figures at all. Italy's zoning is a twenty-region patchwork: one region grades A, twelve grade B, and seven have no usable channel. Spain's federated axes fragment seventeen ways, and its grid geometry grades B: a national substation-point layer, a state topographic line layer last issued in 2015 with no voltage attribute, and operator line geometry only in regional pockets. The national TSO publishes map sheets, not data. These are the markets where the most screening questions require local expertise instead of a laptop.

3. The parcel layer is near-universally open; the grid layer is the weakest axis on the continent. Parcel geometry flows through an open national machine-readable channel in 26 of 31 countries. Open parcels with a separately documented ownership channel is the prevailing European pattern. At the other extreme, the two grid axes hold most of the index's B and C grades; they are what keeps most of the Broad band short of Complete:

Regional only = no clean national channel exists for that axis, so every region is graded separately (Germany's 16 Länder, Spain's 17 comunidades, Italy's 20 regioni; all regions, never a sample). A country whose regions all earn the same grade still counts here: sixteen identical regional channels are not one national channel.

With grid capacity scarce across Europe, the most decision-relevant layer is also the thinnest one on the map: confirmed clean network geometry in 4 of 31 countries, with 9 more unresolved because the blind re-derivation could not finish the distribution-tier search, and clean connection-capacity data in 20. Where it is published, capacity constraints can be refreshed programmatically instead of rechecked manually. In the Netherlands, one of the three major DSOs alone maps 588,655 medium-voltage cable features (observed August 2026), alongside a national capacity map. France publishes live per-substation hosting capacity for its TSO and the distributors covering nearly its whole network. Lithuania's TSO publishes multi-year per-zone hosting-capacity forecasts. The EU has noticed: ENTSO-E and the DSO Entity's Capacitypedia (May 2026) is an official pan-European overview of what grid operators publish. Descriptions and links, not data. An official map of the fragmentation, which itself is unchanged.

4. The Nordics share a band, not a data stack. Denmark has A+ zoning and A connection-capacity data but B grid geometry. Norway is close to the mirror image: A grid geometry, capacity and permits, B zoning. Finland's connection capacity is confirmed live at the transmission tier; its line geometry stays unresolved because the distribution-tier search did not finish; Sweden's permit corpus grades B because its national case register covers wind only. A shared band does not mean screening is complete. It means the remaining manual work differs by market, even between neighbours. Most of it is not missing data. Denmark's national building register turned out to hold every building-permit case with its dates, behind the same self-service key its parcel data uses, which is why its permit grade rose to A in this revision; its DSO networks are mapped, just not published. Norway serves its zoning as an open view map and keeps the national vectors for Norge Digitalt partners. Those gaps are gaps in the terms of access: the register exists, and somewhere there is a form, an agreement, or a login between you and it. The exception is real, and it is the same one twice: neither Norway nor Sweden publishes an open national case-level building-permit register. Norway's national case index returns building-permit decisions municipality by municipality, and Sweden's statistics office holds every granted permit and publishes totals; a developer still cannot screen either as one corpus.

5. Half the data I first graded "absent" wasn't absent - it was invisible in English. Before publishing, I re-hunted every negative finding: 91 claims that a dataset is not published, re-searched using only each country's own language. Polish for Poland, Greek for Greece, all of them. 47 of the 91 fell. Thank god for AI, it's much better at languages than I am. The registers existed: in a regional geoportal, a cantonal gazette API, a municipal permit portal. Live, often machine-readable, just never in English and often not published by the operator you would think to ask. The grades and bands below were re-checked against every verified find and survived the hunt (confirmation passes before and after it moved another 25 grades, 21 of them upward; grades move when the evidence moves, in both directions). In European siting data, the search language is part of the infrastructure. Screening a market without its language is not screening it. That is why the corrections invitation at the end of this index is not a formality.

The index

Bands, not rankings: countries within a band are listed alphabetically. Band placement counts clean-national core datasets out of six (an unresolved cell does not count): Complete = 6, Broad = 4-5, Partial = 3, Fragmented = 0-2. The full per-axis grade matrix follows the lists.

Complete - all six, nationally, machine-readable (2): France, Slovenia.

Broad - 4-5 of six (17): Bulgaria, Cyprus, Denmark, Estonia, Finland, Greece, Iceland, Ireland, Latvia, Lithuania, Luxembourg, Malta, Netherlands, Norway, Portugal, Slovakia, Sweden. Broad, not finished: every country here still lacks one or two of the six, most often grid geometry.

Partial - 3 of six, workable with effort (8): Croatia, Czechia, Hungary, Italy, Poland, Romania, Spain, United Kingdom.

Fragmented - 0-2 of six (4): Austria, Belgium, Germany, Switzerland.

(I'm Danish. Denmark did not make the top two. The grades go where the probes point.)

The full matrix

The grades behind the bands: all 31 countries, all six datasets. Grades, not sources: this matrix shows what each country publishes and how usable it is, not where it lives, in which portal, under which name, or in which language. That, with field-level detail, record counts, and update cadence, is what the per-country deep-dives carry.

A+ authoritative national register of record with verified national coverage: open, machine-readable · A open national machine-readable channel · B exists with material friction: fragmented, partial, paywalled, or view-only · C no workable channel · ^d grade rests wholly or partly on desk verification: some or all of the channel could not be live-probed, and the grade is capped per Method · ^r national grade rolled up from per-region live probes rather than one national-scope probe · ^e a case-level register exists behind a national identity login and is not otherwise reachable; the grade reflects what a cross-border reader can open · ^u the grade rests on a channel that is not a supported interface: no published documentation or export control, or a vendor-hosted service whose own metadata does not name the publisher · ^s the freshest open edition is stale beyond one and a half times the cadence the publisher declares or has demonstrated · ^v the grade rests on a verified union of several operators' channels, or on an association-run aggregator carrying the operators' own values · unresolved: the blind re-derivation of the grid axes (v1.2) could not settle a deciding clause; the cell shows the floor the evidence supports and the ceiling it allows, and counts as not clean · cells like B×14 C×2: no clean national channel exists; the per-region grade distribution is shown instead (every region graded, never a sample).

Put the index to work

Screening sites in Denmark now, or opening another European market? Send me the market, the technology, and one to three blinded sites. I will tell you which of the six questions can be kept source-current there, which still require local work, and whether a scoped deep-dive or pilot makes sense. Denmark runs live in Northing today. The free screening is real, current data.

If you are scoping technical diligence on a portfolio, the matrix shows how much of each market's screening can be refreshed from source.

If you work these markets and I have something wrong, or something missing, the corrections channel below is the front door. Verified finds ship credited in the next revision.

Method

31 countries, six core datasets each; federated countries graded region by region: all regions, never a sample (Germany's 16 Länder, Spain's 17 comunidades, Italy's 20 regioni). Grades reflect probes run June-August 2026, and for the two grid axes a blind re-derivation from live probes run 8-15 September 2026 (v1.2, described in the corrections log); live services drift, which is one more reason the corrections channel below exists. I recorded more than 2,000 probe requests against live services during the research; where no machine access exists to probe, grades are capped and marked desk-only. Machine-accessible load-bearing claims were independently re-checked against the live source before publication; two confirmation passes moved 25 grades, 21 of them upward. Every negative finding was then re-tested under native-language-only search: 91 absence claims across all 31 countries; 47 yielded real sources, each independently re-probed live before acceptance (finding 5). Grades measure data accessibility for siting decisions, not permitting speed, not market attractiveness. The boundary is standard access: a grade reflects what a screening workflow can reach anonymously or through a documented self-service channel, and where a channel sits behind an account, a key, a contract, or a fee, the per-country deep-dives say so. The six axes are the foundation layer every technology shares; wind and hydro add technology-specific layers (resource data, setbacks, water rights) that this index does not grade. A country can have beautiful data and a brutal permitting process; the index tells you which questions can be refreshed from public data and which still require local or legal diligence. Nor is a grade a completeness guarantee: presence and usefulness are graded separately where they diverge. Network geometry and connection capacity are different axes precisely because substations on a map are not the same as knowing what you can connect there. A dataset that exists but is thin, stale, or attribute-poor is graded for what it actually delivers, with the specifics (fields, record counts, update cadence) recorded in my underlying data.

What each axis had to deliver

The criteria below were written down before grading and applied to all 31 countries. Where a case sat on a borderline, the ruling was recorded once and then applied to every country that showed the same pattern. This is what a source had to deliver before a grade counted.

Cadastre. A means parcel geometry published nationally, open and machine-readable, downloadable or queryable by machine. Ownership may sit on a separate channel (a register, an eID login, a fee, a contract) without lowering the grade; that channel is recorded, not penalised. A free self-service API key with no approval step still counts as open. B means vector geometry behind a subscription or per-area fee with only a picture view free, or one constituent nation or region with no register at all. C means no workable channel.

Zoning. A+ means one national register of record serving classified plan polygons for every municipality. A means a live national aggregation of classified zoning geometry, open, with minor caveats such as naming variance or coverage still rolling in. B means a national channel that is view-only, a document index without geometry, or materially incomplete. C means an authentication-walled register, or nothing aggregated at any level. Federated countries are graded region by region; the roll-up is a distribution, never a national grade.

Protected areas. A means a live, open, national vector layer of the dominant designation categories: the EU Natura 2000 sites plus the national designations. A+ means that plus exhaustive category coverage, with live counts checked against official totals, internal zonation layers, or two independent open channels. B means partial or view-only. C means nothing machine-readable.

Grid network. A means machine-readable vector line geometry spanning both the transmission backbone and the major distribution tier, published by the operators or through a consolidated national register or state topographic product. B means one tier only, partial coverage, substation points without lines, a raster picture only, or a corpus behind an access wall. C means no machine-readable geometry at any tier after a completed search. Since this revision a cell can also be unresolved. That means a deciding clause, most often whether the distribution-tier search was complete, could not be settled from the evidence, and the cell prints its floor and ceiling instead of a letter.

Connection capacity. A means a live numeric MW or MVA capacity or queue dataset covering one nationally complete tier, transmission or major distribution, per substation, connection point or zone: an API, a feature service, or a regularly updated spreadsheet that parses. A spreadsheet on a cadence qualifies; an API is not required. Coverage is weighted by customers across the tier's operators, with 80 percent as the line. A+ means that plus an explicit firm or N-1 basis stated by the publisher or by the method it cites; an A− would mean a publisher-stated blend or flexible-only figure, and no cell carries one. B means usable with a structural limit, named in the corrections log: document-bound tables, bands or colours instead of figures, coverage short of the line, an express reuse restriction, or figures for a tier that is not a criterion tier. C means nothing usable after a completed search. Unresolved is defined as for grid network. Currency is printed as a mark, not graded: a publisher's own caveats (indicative, confirm with us, no validity date stated) travel with the cell rather than lowering it.

Permits. Two halves, building permits and environmental or energy permits, both enumerated before grading. A means at least one half is a nationally complete, open, machine-readable, case-level corpus relevant to renewables siting, with the state of the other half recorded. B means the stronger half is search-only, aggregate statistics only, or a single-technology subset. C means no retrievable case-level channel from either half.

Two rules apply across all six. A or A+ requires a live probe; anything that could only be verified by desk research is capped at B and marked desk-only, however good it looks on paper. And a grade never moved on a claim; it moved on a re-probe of the live service.

This index is a measurement, not a verdict. Help me sharpen it. I probed in every country, but you live in one of them. To submit a correction: send the country, the axis, the public source, and how it can be reached by machine. I will probe it live; verified changes ship in the next revision with credit to whoever pointed me there. The corrections log will be public.

What has changed?

Three editions so far. v1.2 (21 September 2026) re-derived both grid axes, grid geometry and connection capacity, blind for all 31 countries: 20 capacity cells moved (13 up, 7 down), 6 geometry cells moved (2 up, 4 down), 10 cells print as unresolved, and the Complete band went from five countries to two. v1.1 (7 September 2026) re-probed the permit axis with a stricter method: seven grades rose, none fell, and two marks joined the legend. v1 (2 September 2026) was the first edition. Every entry, with its reason, is in the corrections log. Read the full corrections log →

Northing normalizes Europe's fragmented siting data into one screening workflow. This index is the research it is built from. Get the deep-dive for your market → · See it working on Danish data →

2026 edition, v1.2 (2026-09-21) · corrections welcome; verified ones ship in the next revision, credited · the corrections log has its own page · a per-country deep-dive series follows.