computer-vision survey · greater metro manila · last calibrated 2026-05-10
A satellite-imagery survey of rooftop solar across Greater Metro Manila.
We trained a custom computer-vision model on labeled examples of rooftop solar from public mapping data, then validated it against held-out examples it had never seen during training: it identifies solar panels at 96% precision. The model surveyed every neighborhood across NCR and the surrounding cities in Bulacan, Cavite, Rizal, and Laguna, examining tens of thousands of high-resolution satellite tiles. It identified 515 rooftops with detected solar (280 high-confidence above a 0.85 model score, 235 below-threshold candidates included for review). Of the 277 high-confidence detections that fell inside a city polygon, 242 (87%) were not already on a prior public map of solar at the time of the scan. Click any city for the count. Zoom in and click any rooftop to see the panels in satellite imagery.
How to read this map
Each dot marks a rooftop where the model identified solar panels in satellite imagery. Orange dots aren't yet on any public map of solar; green dots are already documented. City color shows where solar is densest. Zoom in to see the actual panel area, segmented from the imagery, traced on each rooftop.
Color encoding: orange = newly identified (not on any prior public map of solar); green = corroborated by an existing public map; gray = lower-confidence detection (below the high-confidence threshold). Full methodology: how this map was built.
largest detections by capacity
Top six rooftop installations identified.
Ranked by estimated installed capacity. Each detection corresponds to a building footprint from public mapping data; capacity is derived from the segmented panel area (panel area divided by 6 m² per kWp, capped at the building footprint).
cities ranked by detection count
cities ranked by detection density
Density (high-confidence detections per km² of city area) normalizes across cities of different sizes - smaller LGUs disproportionately represented in the dataset surface here. Raw data and reproducibility scripts: github.com/xmpuspus/solar-map-ph.
Model outputs, not official records. Calibrated to 96% precision on a held-out validation set; expect roughly 1 in 25 high-confidence detections to be incorrect. Treat as a research artifact for further verification, not a permit list.
Statistical indicators derived from public data (Esri World Imagery, OpenStreetMap, ESA, Microsoft, NOAA, NASA). Patterns may have legitimate explanations. "Not on any prior public map" is computed as OSM features tagged generator:source=solar within 200 m of the detected tile centroid, the same proximity convention used by DeepSolar (Stanford, 2018) and SPECTRUM (ICSC, 2025). Residential rooftops are aggregated into counts only; see /privacy.
Cite as: SolarMap.PH (2026Q2), https://github.com/xmpuspus/solar-map-ph. CC-BY-4.0 on data. Imagery © Esri, Maxar, Earthstar Geographics; building data © OpenStreetMap contributors.