OpenPVMapper: A Multi-source, Nationwide Database of Rooftop Photovoltaic Systems in France

Summary OpenPVMapper is an open, multi-source database of rooftop photovoltaic installations in mainland France. It aggregates DeepPVMapper detections (AI pipeline on IGN BD ORTHO imagery), OpenStreetMap, FRPV (per-cadastral-parcel probability), and manual corrections, resolved via a fixed hierarchy (correction > OSM > DeepPVMapper > third-party detections). 1,135,850 installations, ~15.0 GWp estimated installed capacity. Coverage : metropolitan France CRS: EPSG:4326 (WGS84) for all files. Format: GeoJSON, packaged as .zip. Licence: CC-BY. Last update: July 2026. Data validation Precision was assessed via manual review of 1,862 installations (two independent stratified samples: by source combination and by power class). Global precision, weighted by true stratum population: ~74-75%. Corroboration across sources matters a lot: precision goes from 71.5% (single source) to 96.9% (2 sources) to 98.2% (3 sources). 430,940 installations (~38% of the database) are corroborated by 2 or more independent sources. File organization raw.zip ├── D001_2020/ │ └── arrays_characteristics_D001_2020.geojson ├── D001_2022/ │ └── arrays_characteristics_D001_2022.geojson ├── ... (one folder per département × available vintage) └── index.json # {"D001": [2020, 2022], "D002": [2021], ...} enriched.zip ├── enriched-national.geojson # 1,135,850 installations └── regions/ ├── ile-de-france.geojson ├── ... (12 mainland regions) latest-dpvm.geojson.zip └── latest-dpvm.geojson # 593,671 installations Attributes raw/ — raw detections by département and vintage One row = one DeepPVMapper detection for a given département and vintage, before any cross-source matching. Field Description array_id Unique detection identifier insee INSEE commune code dpt Département rnb_id Building identifier (Référentiel National des Bâtiments), if matched surface Polygon surface area (m²) tilt Estimated tilt (degrees) azimuth Estimated azimuth (degrees) kWp Estimated installed capacity year (vintage) is not a column in the file itself — it is carried by the folder/file name (D0xx_20xx) and by index.json, not duplicated on every row. raw/index.json: {"D0xx": [year, year, ...], ...} — availability index to navigate without unpacking all of raw.zip. enriched/ — resolved national database, all sources One row = one installation, after cross-source matching (IoU / centroid distance) and resolution to a single reference geometry. Schema shared by enriched-national.geojson and the regional exports. Field Description array_id Unique installation identifier insee INSEE commune code dpt Département rnb_id Building identifier, if matched sources List of contributing sources (DPVM, OSM, FRPV, correction) surface Polygon surface area (m²), rounded to 2 decimals tilt Tilt (degrees), integer azimuth Azimuth (degrees), integer kWp Estimated installed capacity, rounded to 2 decimals frpv_proba FRPV probability (0–1, floored to 2 decimals), if available first_seen / last_seen First / last vintage in which the installation is detected false_positive_source Manual validation status (normal / false / unknown), if reviewed Note: nom (commune name) was dropped — redundant with insee. latest-dpvm.geojson.zip — single-vintage export, DPVM only Simplified export: one row per installation detected by DeepPVMapper on its latest available vintage only (no multi-source merge, no history). This is the export used in the upcoming analysis of grid connection data paper. Field Description array_id Unique identifier insee INSEE code nom Commune name dpt Département rnb_id Building identifier (RNB), if matched surface Polygon surface area (m²) tilt Estimated tilt (degrees) azimuth Estimated azimuth (degrees) kWp Estimated installed capacity year Detection vintage dpt_source Source département (3-digit code, e.g. "001") The yield_kwh / kwp_approx fields from the previous release have been dropped. Record counts File Number of rows enriched-national.geojson 1,135,850 latest-dpvm.geojson 593,671 raw/ (total, all départements × vintages) not counted — see index.json for the per-département breakdown Sources & citation DeepPVMapper: detection pipeline built for this project. Source code: https://github.com/gabrielkasmi/deeppvmapper/. The code used to build this database (multi-source aggregation, geometry resolution) will be released soon. OpenStreetMap: © OpenStreetMap contributors. FRPV: NEROT, Boris; THEBAULT, Martin, 2024, "FRPV - Presence of Rooftop Photovoltaic (RPV) systems on French buildings.", https://doi.org/10.57745/BXXYW4, Recherche Data Gouv, V3. Code availability DeepPVMapper source code (detection, segmentation, characterization): https://github.com/gabrielkasmi/deeppvmapper/. The code used to build this database (source aggregation, resolution, export) will be released separately, soon.