Catapult 2026

Site Scouter is a map-first decision-support tool for analyzing a user-selected polygon and recommending subregions for solar panels, wind turbines, and data centers.

What The App Does

The system accepts a polygon, validates it, subdivides it into smaller cells, and scores those cells for multiple infrastructure uses.

• Solar scoring considers rooftop/open-land area, shading, slope, and irradiance proxy.
• Wind scoring considers open land, obstruction, slope, and wind proxy.
• Data center scoring considers contiguous flat land and road access.
• The frontend renders returned candidate polygons and lets users inspect the highest-ranked cells by use type and score.
• The frontend also supports single-asset analysis for solar, wind, and data centers with preset or custom specs, plus a past-year daily generation trend where weather data applies.

Current Pipeline

The backend pipeline is modular and now lives in backend/infrastructure/.

1. Polygon normalization and self-intersection checks.
2. Cell subdivision inside the polygon.
3. Imagery retrieval.
4. OSM building and road ingestion.
5. Terrain sampling for slope.
6. Segmentation or land-cover extraction.
7. Infrastructure scoring and ranked candidate output.

The main orchestrator is backend/infrastructure/pipeline.py.

Data Sources

• Free default imagery: USGS USGSImageryOnly, which is primarily NAIP for CONUS requests.
• Optional imagery: Mapbox Static Images and Sentinel Hub Process API.
• Vectors: OpenStreetMap via Overpass.
• Terrain: OpenTopoData public API.

Undocumented Google tile scraping is intentionally not used. The backend accepts "google" only as a compatibility input and converts it to a supported/fallback path with notes, because the direct tile URL approach is not a supported Google Maps Platform integration.

ML Segmentation Support

The backend supports these segmentation modes:

• rule_based
• hybrid
• auto
• unet
• mask_rcnn

rule_based and hybrid work out of the box. unet and mask_rcnn are wired as remote-service integrations because this repo does not ship model weights or a local inference runtime.

Remote inference env vars:

• INFRA_UNET_ENDPOINT
• INFRA_MASK_RCNN_ENDPOINT

Expected response shape:

{
  "source": "unet-service",
  "cells": [
    {
      "id": "cell-1",
      "vegetation_ratio": 0.12,
      "water_ratio": 0.01,
      "impervious_ratio": 0.42,
      "shadow_ratio": 0.08,
      "building_ratio": 0.25
    }
  ]
}

Repository Layout

• backend: FastAPI service and scoring pipeline.
• frontend: React + Leaflet UI.
• model: Existing training/data-prep experiments.
• data: Local datasets used by model and analysis work.

Important Backend Files

• backend/main.py: API routes.
• backend/schemas.py: request/response models.
• backend/asset_analysis.py: single-asset analysis and weather-driven trend output.
• backend/infrastructure_pipeline.py: compatibility wrapper.
• backend/infrastructure/providers/imagery.py: imagery retrieval.
• backend/infrastructure/providers/vector_data.py: OSM ingestion.
• backend/infrastructure/providers/terrain.py: slope sampling.
• backend/infrastructure/segmentation.py: rule-based and remote ML segmentation integration.
• backend/infrastructure/scoring.py: feature fusion and candidate scoring.

Important Frontend Files

• frontend/src/App.jsx: top-level app state and workflow.
• frontend/src/components/ControlPanel.jsx: collapsible planning panel, asset specs, and result summaries.
• frontend/src/components/MapScene.jsx: map rendering and candidate polygons.
• frontend/src/components/TopBar.jsx: compact settings popover.
• frontend/src/components/TrendChart.jsx: past-year daily generation chart.
• frontend/src/lib/assetAnalysisApi.js: single-asset API client.
• frontend/src/lib/assetResult.js: asset-analysis result mapping.
• frontend/src/lib/infrastructureAnalysisApi.js: infrastructure API client.
• frontend/src/lib/infrastructureResult.js: backend-to-UI mapping helpers.

Local Development

Backend:

cd backend
python -m unittest discover -s tests -p 'test_*.py'
uvicorn main:app --reload

Frontend:

cd frontend
npm ci
npm run test
npm run lint
npm run build
npm run dev

Environment Variables

Frontend env vars:

• VITE_BACKEND_URL
  • Optional.
  • Defaults to http://127.0.0.1:8000.
  • Set this when the frontend talks to a non-local backend.
• VITE_BASE_PATH
  • Optional.
  • Only needed when serving the built frontend from a subpath such as GitHub Pages.

Backend env vars:

• No env vars are strictly required for local startup.

• To run the backend with the fullest live-data path, set the provider/model env vars below.

• MAPBOX_ACCESS_TOKEN

  • Optional. Needed only for imagery_provider="mapbox".

• MAPBOX_STYLE_OWNER

  • Optional. Defaults to mapbox.

• MAPBOX_STYLE_ID

  • Optional. Defaults to satellite-streets-v12.

• SENTINEL_HUB_CLIENT_ID

  • Optional. Needed only for imagery_provider="sentinel".

• SENTINEL_HUB_CLIENT_SECRET

  • Optional. Needed only for imagery_provider="sentinel".

• SENTINEL_HUB_COLLECTION

  • Optional. Defaults to sentinel-2-l2a.

• SENTINEL_LOOKBACK_DAYS

  • Optional. Controls Sentinel date selection.

• SENTINEL_MAX_CLOUD_COVER

  • Optional. Controls Sentinel filtering.

• INFRA_UNET_ENDPOINT

  • Optional. Remote U-Net inference endpoint for segmentation_backend="unet" or auto/hybrid.

• INFRA_MASK_RCNN_ENDPOINT

  • Optional. Remote Mask R-CNN inference endpoint for segmentation_backend="mask_rcnn" or auto/hybrid.

• OSM_OVERPASS_URL

  • Optional. Defaults to the public Overpass interpreter.

• INFRASTRUCTURE_IMAGERY_SIZE

  • Optional. Controls imagery sample resolution for the infrastructure pipeline.

Notes:

• Past-year solar and wind trend data comes from Open-Meteo historical APIs and does not require an API key.
• The free default imagery path is usgs, so you can run the infrastructure endpoint without paid imagery credentials.