Getting a solar quote used to start with a ladder, a tape measure, and a site visit. Solar-Roof-AI starts with a satellite image instead. It is an open source planning workspace that analyzes rooftops, estimates usable area, previews panel layouts, and exports blueprint-style reports, all from a map in your browser (GitHub, retrieved 2026-10-24).
Key Takeaways
What it is: Solar-Roof-AI (amarnath3003/solar-roof-ai, CC BY 4.0) is an interactive solar planning workspace for rooftop mapping, AI-assisted roof detection, panel layout simulation, and financial estimation.
The workflow: Search a property, draw or auto-detect roofs and obstacles, check usable area and a solar exposure heatmap, place panels manually or with auto-pack, then export GeoJSON geometry or a blueprint-style PDF report.
The scale: About 132 stars and 18 forks on GitHub, 138 commits, created 2026-03-31. It is a small, focused project, not a commercial solar platform.
The honest limit: Results are planning-grade, not permit-ready engineering. Detection quality depends on imagery quality, zoom, and roof contrast, and manual edits remain essential for complex roofs.
Why it matters: It shows how satellite imagery plus open tooling can compress the slowest part of solar pre-sales, the roof survey, into a reviewable desktop workflow.
The problem it solves
Solar feasibility work is usually fragmented. One tool holds the map, another holds the spreadsheet, and the panel layout lives in a CAD file somewhere else. Estimators juggle GIS platforms, CAD software, and messy spreadsheets before they can answer a simple question: how many panels fit on this roof, and what is it roughly worth?
Solar-Roof-AI tries to put that whole chain in one workspace. You search for a property, the map centers on it, and you work from satellite imagery outward: roof polygons, obstacle layers, usable area, panel placement, and a financial estimate, ending in an exportable report (GitHub, retrieved 2026-10-24).
This is not a new idea at industry scale. Google's Solar API, launched from Project Sunroof, made detailed rooftop solar potential data available for more than 320 million buildings across 40 countries, fusing overhead imagery, rooftop data, historical weather, and financial values to reduce the need for on-site visits (Renewable Energy World; Google Maps Platform). What Solar-Roof-AI adds is the open, inspectable version: a codebase you can run, fork, and adapt, instead of an API you can only call.
How it works: eight steps, one workspace
The project describes an end-to-end flow that is refreshingly concrete:
Search for a property by address or place name, using OpenStreetMap Nominatim.
Enter the map workspace and switch to satellite imagery mode.
Draw roof polygons and obstacles manually, or run AI-assisted auto-detection.
Review detection results and accept only what looks correct, tuning confidence and area filters.
Calculate usable roof area and inspect solar heatmap hints.
Simulate panel placement manually or with auto-pack.
Review estimated system capacity and financial outcomes.
Export geometry as GeoJSON or a blueprint-style PDF report.
That review step is the design decision that matters most. The AI suggests, the human decides. Detections are previewed before they are applied to the workspace, which keeps a bad detection from silently becoming a bad quote (GitHub, retrieved 2026-10-24).
Six features, mapped to the workflow
Feature | What it does | Why it matters |
|---|---|---|
Property search and mapping | Address search, map-first workspace, high-resolution satellite imagery | No site visit needed to start an estimate |
Roof and obstacle mapping | Manual drawing, separate editable layers, irregular shapes and exclusion zones | Chimneys, vents, and skylights are obstacles, not just roof |
AI-assisted roof detection | Snapshots sent to a Roboflow hosted workflow, plane and obstacle detection, confidence filters | Automates the tedious tracing step, with human review |
Geospatial and solar analysis | Gross area, blocked area, and net usable area, orientation candidates, solar exposure heatmap | Turns a polygon into a number a customer can understand |
Panel layout simulation | Manual or auto-pack placement, capacity-aware panel selection, overlap validation | Shows what actually fits, not just total roof area |
Financial planning and export | Financial dashboard, sizing inputs, estimation charts, GeoJSON and PDF export | Turns a layout into a handoff document |
The stack is deliberately ordinary, which is a compliment here. The frontend is React 18 with Vite and TypeScript, mapping with Leaflet and Turf.js, address search with OpenStreetMap Nominatim, charts with Recharts, and detection through a Roboflow serverless workflow. You need Node.js 20 or newer to run it locally (GitHub, retrieved 2026-10-24).
Why satellite roof detection works at all
Detecting roofs from above is a well-studied computer vision problem, and the numbers behind it explain both the promise and the caution. Stanford researchers analyzed more than 1 billion high-resolution satellite images and identified 1.47 million solar installations across the contiguous 48 US states; their DeepSolar model correctly identified an image as containing solar panels 93 percent of the time (Stanford News).
More recent research keeps pushing accuracy higher. A 2025 study that manually labeled more than 20,000 roof areas over 45.7 square kilometres of satellite imagery reported a maximum roof estimation accuracy of 84.84 percent IoU on test regions, with an average of 77.11 percent (MDPI Sustainability). Those are strong results for planning, and they are also a reminder that no model is perfect: every automated roof outline deserves a human glance before it becomes a proposal.
Google's own expansion shows the same direction of travel. A satellite expansion added 125 million new buildings with Solar API data across 23 countries, extending potential coverage by 1.9 billion additional buildings based on available satellite imagery (Google Research). The industry is moving toward imagery-first solar assessment. Solar-Roof-AI is the small open source version of that shift.
The honest limits
Solar-Roof-AI is upfront about its limits, and any fair write-up should repeat them. Detection quality depends heavily on imagery quality, zoom level, and roof contrast. Pitch, aspect, and height values are estimation-grade, derived from 2D imagery. Results are intended for planning and pre-sales exploration, not permit-ready engineering. Manual edits remain essential for complex edge cases (GitHub, retrieved 2026-10-24).
There are two more practical caveats. First, the README gives no detection accuracy percentage for its own model, so none should be claimed for it; the accuracy figures above belong to other published systems. Second, the detection step calls a hosted Roboflow workflow, so the AI part is not fully self-hosted even though the workspace itself is open source. If you need a fully offline pipeline, that hosted dependency is worth noting before you adopt it.
Should you try it?
Try Solar-Roof-AI if you are a solar estimator, a developer exploring solar tooling, or a student learning geospatial computer vision. It is a compact, readable way to see the full chain, from satellite image to panel layout to PDF report, without signing up for a commercial platform. Run it locally, try it on a roof you know well, and compare its usable-area number against a manual measurement. That comparison is the fastest way to learn where it helps and where it still needs your judgment.
Skip it, or treat it as a learning project only, if you need permit-ready engineering documents, guaranteed accuracy figures, or a fully self-hosted detection model. For those jobs, a commercial survey and engineering review remain the right call.
Want the 30-second version? Watch the repo signal for this project on repomaniac.com, and read this article in Indonesian for the translated edition. If you are new to open source licensing, our What Is Open Source explainer covers the basics behind licenses like the CC BY 4.0 license used here.
Frequently Asked Questions
What is Solar-Roof-AI?
Solar-Roof-AI (amarnath3003/solar-roof-ai) is an open source interactive solar planning workspace for analyzing rooftops, estimating usable area, and previewing panel layouts from satellite imagery. It is licensed CC BY 4.0 and has about 132 stars on GitHub (GitHub, retrieved 2026-10-24).
How does it detect roofs?
It sends map snapshots to a Roboflow hosted workflow for plane and obstacle detection, then lets you preview results, tune confidence and area filters, and apply only the detections you accept. You can also draw roofs and obstacles manually (GitHub, retrieved 2026-10-24).
What can I export from it?
GeoJSON export of roof and obstacle geometry, plus blueprint-style PDF reports for planning handoffs, alongside a financial dashboard with sizing inputs and estimation charts (GitHub, retrieved 2026-10-24).
Is it accurate enough for a real installation quote?
For planning and pre-sales exploration, yes, with human review. For permit-ready engineering, no. The README states results are estimation-grade from 2D imagery and manual edits remain essential for complex roofs (GitHub, retrieved 2026-10-24).
What do I need to run it?
Node.js 20 or newer, plus the frontend dependencies. The stack is React 18, Vite, TypeScript, Leaflet, Turf.js, and Recharts, with OpenStreetMap Nominatim for address search (GitHub, retrieved 2026-10-24).