Global PV equipment and project database#

Five evidence-backed tables cover modules, inverters, trackers, combiner boxes and projects. Each row has a stable ID, standardized metadata, source references, creation/update/verification times and an extension field. This is a growing research catalogue: the launch includes 21,605 modules, 2,344 inverters, 5 trackers, 3 combiners and 7,320 projects across 11 countries. Most projects come from the United States EIA-860 inventory; global project coverage is still highly uneven.

Catalogue guide and update contract, data dictionary, expanded research prompt, and launch research evidence.

Interactive viewer and runnable statistics#

Open the interactive database viewer or the Database gallery. Search and filter every published row, select columns, sort, inspect evidence, compare groups and export filtered CSVs. The viewer is regenerated from the exact published Parquet files on each documentation build. Browser JSON projections are derived assets; Parquet and canonical JSONL remain the data formats of record.

Runnable offline examples cover table counts and sizes, all-column completeness, country coverage, equipment queries and separate AC/DC capacity summaries. The reusable energy_analysis.pv_database functions work on local Parquet snapshots. Unknown values remain null; reported partial sums are not total installed capacity.

Download and storage choice#

The successful main documentation build publishes Parquet datasets: modules, inverters, trackers, combiner boxes, projects, manifest, source registry, and SAM notice.

Parquet + Zstandard provides typed nullable columns, compression and efficient column reads. sources and source_urls remain native lists, and timestamps are UTC Arrow timestamps. Nested variable metadata is canonical JSON in designated columns, keeping the wide row stable as new source fields appear. Git stores reviewable JSONL; every binary snapshot is reproducible from it. CSV and Excel are poor canonical choices for heterogeneous nested sources and reliable nullable types.

uv run python -m energy_data_retrieval.pv_catalogue validate
uv run python -m energy_data_retrieval.pv_catalogue build --output artifacts/pv-v1
import pandas as pd

modules = pd.read_parquet(
    "artifacts/pv-v1/modules.parquet",
    columns=["record_id", "manufacturer", "model", "pmax_stc_w", "bifacial", "source_urls"],
)
print(modules.loc[modules["bifacial"].eq(True) & modules["pmax_stc_w"].ge(600)].head())

projects = pd.read_parquet("artifacts/pv-v1/projects.parquet")
print(
    projects.loc[
        projects["country_code"].eq("CHL"), ["name", "capacity_dc_mw", "capacity_ac_mw", "sources"]
    ]
)

For optional SQL analysis, DuckDB can query those Parquet files directly without a server. It is not a required dependency. Always distinguish country coverage and OEM coverage from current market availability, and keep capacity bases separate.

Launch research and known limits#

The SAM/CEC module library was pinned by Git revision and SHA-256, not read from an unversioned moving URL. Its June 2026 module data contains 36 repeated model names with conflicting parameters: all 72 raw rows are quarantined. Case-sensitive OEM suffixes remain distinct and are candidates for further review, not auto-merging.

EIA report-year 2025 provides 7,309 operating PV plant IDs. The adapter joins plant metadata to OP/PV generator records, keeps generator attributes in phases_json, and sums nameplate AC and net DC separately. Missing unit capacities produce a null total. Plant dates and angles are not guessed from the first generator. Non-PV equipment at a mixed facility does not enter PV totals. Topaz is explicitly reconciled with its OEM/owner-sourced project row, leaving 7,320 unique projects including 11 additional projects outside the USA.

The Topaz owner describes 550 MW without an AC/DC basis, while the EIA generators sum to 585.9 MWac. Both are retained with their source and a review flag; they are not silently equated. Historical source dates are retained. last_verified_at means source access, not confirmation of live operation or current certification.

OEM tracker/combiner specifications are sparse where configuration-specific technical sheets are gated. No contact forms were submitted. Coordinates outside the EIA sample are left null where no sufficiently supported location was found. GEM, WRI, CEN, ANEEL and UK REPD were reviewed as expansion candidates; they are not claimed as imported datasets here. Their filtering, licensing, chronology and project/phase distinctions need dedicated handling.

Updating and extending#

The research specification defines the regional rotation, source hierarchy, provenance, null handling and conflict process. New research accumulates on an open [PV database] PR; after that PR closes or merges, the next substantive batch starts a new one. Recurring PRs remain open for review. Scheduling is supplied by the user’s automation, not an unauthenticated GitHub cron.

New metadata should reuse one of the 148 canonical columns. Use a namespaced extension for unclassified metadata, then promote it through a versioned schema change when a stable definition is established. Code, schema and records are validated in CI; Pages exports the validated Parquet snapshot after merge.

Spanish official gazette notices#

Use pv_catalogue.boe.read_boe_pv_summary with the raw bytes, SHA-256 and publication date of a BOE daily JSON summary. The reader returns the issue number, publication date, notice ID and official notice XML URL. Review the operative text and compare permit, applicant, municipality and phase identifiers against the catalogue before accepting any facts. Preserve design and decision dates, and keep grid access, active inverter power and peak DC power separate. BOE source and reuse terms.

Bounded ODRE batches#

The offline ODRE staging ledger supports an inclusive minimum_kw and exclusive maximum_kw. Omitting both retains the original >=10 MW scope. The minimum must be at least 1000 kW; invalid or reversed bounds fail closed. The complete national individual-PV identity export is still mandatory, including names below the selected capacity band. Neither capacity selection nor a reviewed ID overrides aggregate, phase, privacy, existing-identity or shared-commune holds. See the reviewed catalogues/pv/research/20261102T0000_odre_review.json 5–10 MW batch.

Dated enrichment review: 2 November 2026#

The 16:01 review fills missing grid and industrial-service-date fields for 155 existing French projects using unique EIC identifiers in the 31 August 2026 ODRE snapshot. It also promotes explicitly reported historical operators for 236 complete GEM locations. Neither operation adds project identities or claims live operating status. Twelve Suntree component tables support two additional named platforms and four platform enrichments; voltage-labelled component ranges remain qualified within each platform. Rebuild the catalogue for current totals.

Details and exclusions: dated research ledger.

The 17:00 review enriches 257 existing French projects from the Occitanie 1–5 MW PHOTV exact-EIC crosswalk and 176 complete GEM locations with historical operator evidence. It adds ten exact KSTAR inverter models from five tables in the 2026 C&I EN V2 brochure, retaining per-MPPT arrays, options and derating conditions. Eight contradictory/unclear equipment fields remain held. No new projects, countries or manufacturers are claimed. Current counts and field completeness are in the generated manifest; launch totals remain historical.

The 18:00 review adds 333 existing-project enrichments (820 fields: 487 direct values and 333 source-qualified evidence objects), 27 exact KSTAR inverter models and nine BENY combiner platforms. Named configurable platforms remain one row, with changeable fuse/current options qualified. No new project or country is claimed. Conflicting specifications, uncertain identities and restrictive-reuse tracker evidence remain in the dated research review.

ITER Tenerife installation and monthly-series review#

The Tenerife adapter reads the Cabildo de Tenerife Open Data Portal infrastructure CSV and individual monthly JSON series. Pass pinned source bytes and their SHA-256 digests to pv_catalogue.tenerife.read_tenerife_inventory and pv_catalogue.tenerife.read_tenerife_monthly_energy. The monthly reader also takes the reviewed native installation ID. A reader result is a review candidate. Use the existing upsert API after checking the native ID, catalogue names, location and whether the source represents a whole installation.

Tenerife’s nominal kW has unspecified AC/DC basis. The dataset classifies PV explicitly and gives a publisher installation point. The monthly dictionary labels both import and export as Kw/h. No normalized energy observation was published for this edition. The 9 November research ledger records the checked releases, attribution, exclusions and next partitions.