Rooftop solar is growing fast, and it plays a real role in decarbonizing the power system — it also makes communities more resilient and brings people closer to their own electricity use. But rooftop PV is a genuinely different class of asset, and a hard one to keep track of: in France, rooftop installations account for more than 99% of all grid-connected PV systems. DeepPVMapper sets out to close that gap, combining geospatial data, remote sensing, and crowdsourcing into a registry that aims to be as complete and accurate as possible. The Data and Software pages cover what the project produces and how to use it; this page is about why it exists and where it came from.
Rooftop PV is a blind spot. Distributed PV, a category dominated by rooftop systems, makes up a large and fast-growing share of the solar capacity installed worldwide, yet nobody knows precisely how much of it there is. Small and self-consumed systems are hard to track, even in mature markets, and official figures can be far off. Rooftop PV deserves a focus of its own, not one line in a broader solar inventory.
The map should be free. The value of such a registry comes from people using it and combining it with their own workflows. Putting it behind a paywall only adds barriers, so everything — from the training data to the final registry — is released openly, kept interoperable with OpenStreetMap and reproducible end to end. The aim is not to impose yet another standard, but to provide an interoperable basis that users can adapt to their own formats. The sole goal is to detect PV systems visible from openly available imagery and describe their technical characteristics — never to know who owns them.
The model proposes, people verify. Machine learning does the heavy lifting, but it makes mistakes, and what holds back adoption is less the accuracy of a detector than the lack of a way to measure it. Rather than claiming to represent a definitive ground truth, DeepPVMapper combines remote sensing with crowdsourced verification (PV Check, MapRoulette) to provide the best available information, and to improve it continuously. Verified detections are documented as such in each release and feed back into the training data of future models.
An ecosystem, beyond France. DeepPVMapper is one initiative among others, such as EarthPV and TraceTheSun. It focuses on France and tries to “industrialize” the map; the long-term hope is that the experience gained benefits other countries, with their own detections hosted alongside the French ones. Access is open through direct download, an API and an MCP server, on one condition: no harmful use or scams (see the acceptable use policy).
Read the full manifesto: Rooftop Solar Needs an Open Registry →
DeepPVMapper grew out of a CIFRE PhD thesis — a French industrial research contract pairing an academic institution with a company — conducted between Mines Paris (PSL University) and RTE France, the French electricity transmission system operator. The thesis, “Enhancing the Reliability of Deep Learning Models to Improve the Observability of French Rooftop Photovoltaic Installations,” was defended in 2024 at Université Paris sciences et lettres (theses.fr record, summary).
The starting point was operational. In 2020, early findings — later published in RTE and the IEA’s 2021 joint report on the technical feasibility of a power system with a high share of renewables in France towards 2050 (RTE & IEA, 2021) — showed that the risk of contingencies induced by rooftop PV could increase sharply by 2050, driven by the lack of observability of these small, distributed installations. The first step was to assemble training data and adapt an existing detection approach — building on DeepSolar (Yu et al., 2018) and 3D-PV-Locator (Mayer et al., 2022) — to the French context, which led to the detection algorithm DeepPVMapper and the training dataset BDAPPV.
The thesis also examined how much the algorithm’s performance degrades outside its training conditions, leading to a new feature-attribution method, WAM (ICML 2025), and to follow-up studies on remote-sensing-based PV detection (Environmental Data Science, 2025). It also showed that better forecasts and reduced contingency risk come from resolving the uncertainty around installed capacity itself: in France, aggregate capacity is tracked fairly well nationally, but significant imbalances remain locally. Getting there required a new Bayesian methodology — see the Registry Audit.
By the end of the thesis, 38 French départements had been mapped (roughly 175,000 km²). As the rest of France was mapped, other groups introduced their own detections, notably FRPV (Thebault et al., 2025). They were consolidated with DeepPVMapper’s into OpenPVMapper, which aims to be a reference registry of rooftop PV installations in France. It is updated as new imagery becomes available and keeps improving through crowdsourcing — have a suggestion, or want to help? Get in touch.
@phdthesis{kasmi2024enhancing,
title={Enhancing the Reliability of Deep Learning Models to Improve the Observability of French Rooftop Photovoltaic Installations},
author={Kasmi, Gabriel},
year={2024},
school={Universit{\'e} Paris sciences et lettres}
}