The roof engine ingests storm imagery, isolates individual roofs, and applies preset
analysis recipes to each structure. The result is a clean, per-roof dataset with
consistent severity calls that drop straight into GIS, claims systems, and field
workflows.
Built around real hurricane and hail events across the Gulf Coast, the system is
tuned for messy, post-storm conditions — tarps, debris, partial failures,
and mixed construction types.
Production-grade pipeline
Per-roof, not per-pixelDesigned for insurers, contractors, and coastal agencies
At a high level
From storm imagery to a usable roof dataset.
Everything the engine does is geared toward one outcome: a structured table of roofs,
each with a geometry, severity scores, and practical flags that connect directly to
field work and decision-making.
Roof-centric. We treat each roof as a single object, even when imagery is noisy.
Preset-driven. Damage, hail, wind, and tarp presets can be combined as needed.
Consistent. The same logic runs for every roof in the project, avoiding manual drift.
Core inputs
Post-storm orthomosaics from drones or crewed aircraft.
Optional pre-storm imagery for change detection.
Optional building footprint layers where available.
Processing flow
What happens when you run a project.
Step 1
Imagery ingest
Orthomosaics are validated, normalized, and tiled. Large projects are chunked so
everything stays responsive.
Step 2
Roof detection
Roofs are located as distinct footprints, aligned with any existing building
layers where available.
Step 3
Preset analysis
The preset bundle you choose (Damage, Hail, Wind, Tarp, etc.) is applied to each
roof and converted into scores and flags.
Step 4
Data products
Results are assembled into GIS layers, CSVs, and report-ready summaries for
download or API use.
Output focus
Per-roof records
Each row represents a single structure with geometry, severity scores, and preset flags.
Storm-aware
Hurricanes & hail
Tuned on Gulf Coast events: Ida, Laura, Michael, Ian, and others with real-world roof damage.
Deliverables
GIS + reports
GeoPackage / CSV for GIS, plus human-friendly summaries for management and field teams.
What goes in
The engine is designed to work with the imagery and data sources you already use
after a storm.
Drone or fixed-wing orthomosaics (RGB).
NOAA/NGS Post-event Imagery.
Optional pre-event imagery for change analysis.
Optional building footprints or parcel layers.
What comes out
The main output is a simple roof table plus geometry — easy to join into your own systems.
Roof ID, location, and polygon.
Damage, hail, wind, and tarp scores.
Preset bundle used and model version tags.
Why it stays consistent
Every roof in a project is treated the same way. You’re not relying on a different
person on a ladder with a tape measure and a phone on every street.
Preset configurations are locked per project run.
Severity bins are stable and repeatable.
Outputs can be re-generated later with the same settings.
Behind the scenes the system continues to improve, but the presets you select
always act as a stable contract for that project.