No automated registry built from remote sensing and heuristics is ever perfectly accurate, and DeepPVMapper is no exception. This page is a plain account of where the current version falls short, why, and what to do if you spot one of these problems — or something else entirely — on the interactive map.
Two ways to report a problem: for a false positive, a missing installation, or a polygon that doesn't match the rooftop, use the pencil tools directly on the map — they apply to your view immediately and are reviewed for merge. For anything else (see below), use the Report an issue button at the bottom of the map, or open an issue directly on the GitHub tracker.
The detection pipeline works from aerial/satellite imagery and a learned model, not a ground survey — so a small share of installations are missed (shadowed roofs, small or unusual arrays, imagery gaps), and a small share of detections aren't real installations at all (reflective surfaces, skylights, some roof patterns that resemble panels from above).
Fix: use Add missing installation or False positive directly on the map.
Polygon outlines are extracted automatically and can be offset from the true roof edge, especially on complex, multi-plane, or tightly packed rooftops — the array is real, but its exact footprint on the map may not be pixel-perfect.
Fix: use Redraw on the installation's popup to trace the corrected outline.
Surface, tilt, azimuth, and installation year are estimated from imagery and metadata heuristics, not measured on site or self-reported by an owner — treat them as reasonable estimates rather than exact figures, particularly for older installations where the underlying imagery is coarser.
Fix: use Report an issue → "Missing or incorrect attributes".
The detection pipeline is tuned for rooftop arrays. Occasionally a ground-mounted, utility-scale solar farm gets picked up and classified as a rooftop system instead — there isn't yet a fully robust automatic filter for this distinction, so a handful of these slip through into the registry.
Fix: use Report an issue → "Looks like a utility-scale plant".
Pulled live from gabrielkasmi/openpvmapper-issues — open a data-quality issue there directly if you're comfortable with GitHub.
Loading…
Something else — a question, or a problem that doesn't fit any of the above? If it's general (someone else might wonder the same thing), start a discussion on GitHub. Otherwise, contact Gabriel directly.