steml

Solar data for the age of AI.

steml turns imagery of the planet into structured, decision-ready data. We start with solar: what is installed on every roof, what capacity remains and where demand comes next, for lenders, energy providers and installers in any market.

stem: grow your data

lon 3.7°W · lat 40.4°N · loading coastlines…

Method

  1. 01
    Imagery

    High-resolution aerial and satellite imagery from national mapping agencies and commercial providers. Our pilot runs on 25 cm official orthophotos.

  2. 02
    Detection

    A computer-vision model segments solar arrays on every roof. Pools, greenhouses and shade structures are filtered out by colour checks, and every rejection is logged with its reason.

  3. 03
    Verification

    Before a roof is recorded as having no panels, it is re-examined at 2× zoom, centred on the building. Low-confidence detections are flagged, never counted.

  4. 04
    Attribution

    Each detection is matched to a building footprint (Overture Maps: OpenStreetMap and national cadastres), so every number traces back to a specific roof and image.

  5. 05
    Benchmarking

    Adoption is measured per neighbourhood with 95% confidence intervals and compared with the city, the region and the most similar areas by housing type.

From the pilot: three real roofs

Aerial view of a roof: panels found
Panels found140 m² house roof. Array detected with confidence 0.70 and matched to the building outline.→ counted as installed
Aerial view of a roof: found on the second look
Found on the second lookA small array missed in the town-wide pass. Re-checked at 2× zoom, centred on the roof, then detected at 0.90.→ counted as installed
Aerial view of a roof: uncertain, flagged
Uncertain, flaggedSomething panel-like at 0.37 confidence. We don't guess: it is excluded from counts and from lead lists.→ needs review

Cyan: building outline. White: detected array with model confidence. Yellow dotted: below threshold, sent to review. Orthophoto © Institut Cartogràfic i Geològic de Catalunya (ICGC), CC BY 4.0.

What we provide

Who it is for

Coverage & delivery

Coverage
Global by design: any market with high-resolution imagery and building outlines. Live pilot in Cambrils, Spain. New markets on request.
Refresh
Quarterly releases, folding in new imagery as it is captured and official installation registries where they exist.
Delivery
API, CSV and GIS layers (GeoJSON, GeoPackage).
Validation
Model output reproduces the reference implementation on test imagery. Counts are cross-checked against official installation registries.

Request early access

Tell us where you operate and what you need. We'll be in touch when data for your area is ready.

Only used to tell you when data is ready. No spam.

Solar is our first dataset. If you need something else measured from above (buildings, roofs, land use, infrastructure), tell us in the company field and we'll scope it.