SolarShift.
IMAGE → INTELLIGENCERooftop solar workspace
Connecting to vision engine
SOLAR INTELLIGENCE / LIVE WORKSPACE

Davis California, US

READY TO ANALYZE
38.5465° N / 121.7442° W
A NEW WAY TO SEE YOUR ROOFTOP

The energy above us.

Find roofs. Follow the sun.
See what a photograph can reveal.

Real imagery · pretrained AI · solar physics
RoofPotential canopyShadow
20 m
N
USGS / NAIP · public domain orthophotographyNo detection results until analysis runs
12:00LOCAL TIME · PDT
☼
—Sun elevation
—Azimuth
—GHI · W/m²
00:00Sunrise —12:00Sunset —24:00
21 JUN 202610-minute steps · clear-sky scenarioWaiting for analysis
ⓘ

Preliminary planning, grounded in physics. Heights and roof tilt are estimates. Confirm scale and boundaries; weather and terrain are excluded.

TRANSPARENT BY DESIGN

From pixels to potential.

01 / Computer vision

Real masks. Real imagery.

A pretrained RF-DETR model segments buildings in overlapping image tiles. OpenCV traces and simplifies the masks; overlapping candidates are deduplicated. Confidence comes from the model. Building masks are roof candidates, not a measurement of the roof surface.

02 / Spatial assumptions

A useful approximation.

Polygons are extruded to editable heights. Scale and north direction connect pixels to meters. Vegetation uses an excess-green color heuristic and may include grass; review each region. Rooftop obstacles can be added manually.

03 / Sun & shadow

Let geometry do the work.

pvlib computes NREL SPA solar position from coordinates, timezone, date and time. A caster's shadow length is its height above the receiving roof divided by tan(solar elevation). Swept polygons preserve concave footprints; roof sample points determine direct-light shading.

04 / Clear-sky energy

An estimate you can inspect.

Ineichen clear-sky irradiance is transposed onto the selected panel tilt. Shading reduces direct irradiance; diffuse light remains unshaded. Power = capacity × irradiance / 1000 × performance ratio. Daily kWh integrates 10-minute power samples. Module efficiency 22%, coverage 65%, performance ratio 0.80 and module area 2 m² are defaults.

This is preliminary planning. One image cannot measure heights, hidden obstacles or roof slope. No weather, terrain horizon, electrical string mismatch, module temperature or certified engineering assessment is included. Map shadows on ground and rooftop exposure use different receiving heights.
Model card ↗ pvlib documentation ↗ MANUAL CORRECTION

Add a region.

Click boundary points on the image. Press Enter to finish, Escape to cancel. Obstacles need a height above the receiving roof to cast a shadow onto it.

 SCALE CALIBRATION

Give pixels a distance.

Choose two points on the image spanning this known distance. Scale will update from their separation.