Interactive web tool for visually verifying coconut plantation training labels across all 38 districts of Tamil Nadu. Built for collaborative validation of the Descals et al. (2023) coconut training dataset using Google satellite imagery.

Live tool: athithiyanmr.github.io/coconut_verification_tool

Coconut plantation mapping using satellite imagery requires high-quality training labels. The Descals et al. (2023) dataset provides raster-based coconut labels derived from Sentinel-2 imagery, but visual verification against high-resolution satellite imagery is essential to assess label accuracy.

This tool vectorizes the raster labels into 90,438 individual polygons, overlays them on Google Satellite imagery, and provides a streamlined interface for multiple users to collaboratively verify whether each polygon actually represents a coconut plantation.

• District-based navigation — Select any of 38 districts; polygons load on demand
• District boundary outline — White dashed boundary shows the district extent on the map
• Polygon overlay toggle — Show/hide the training label overlay to inspect the satellite image beneath
• One-click verification — Mark each polygon as "Coconut" or "Not Coconut"
• Auto-advance — Automatically moves to the next polygon after verification
• Progress tracking — Real-time progress bar with Yes / No / Pending counts
• Filter polygons — View All, Pending, Yes, or No polygons in the sidebar

• Shared cloud backend — Verifications are stored in Google Sheets via Apps Script
• Multi-user support — Each user enters their name; verifications are attributed (e.g., "by Athithiyan")
• Real-time sync — Auto-refreshes every 60 seconds; manual Refresh button available
• Conflict-safe — Assign different districts to different people to avoid overlap

• Leaflet.draw integration — Draw new coconut polygons directly on the map for areas not in the training set
• Metadata capture — Each drawn polygon records the user, area (ha), timestamp, and a note
• Shared storage — Drawn polygons are saved to the cloud and visible to all users
• Delete support — Only the user who drew a polygon can delete it

All exports include a source column: training_label for original polygons, user_drawn for manually drawn ones.

GitHub Pages (frontend)          Google Sheets (backend)
┌────────────────────────┐       ┌──────────────────────────┐
│  index.html             │       │  verifications tab        │
│  style.css              │  ───► │  key | status | user | ts │
│  app.js                 │  GET  │                          │
│  data/*.geojson         │  POST │  drawn_polygons tab       │
│  Leaflet + Leaflet.draw │  ◄─── │  id | district | geom... │
└────────────────────────┘       └──────────────────────────┘
         │                                  │
         ▼                                  ▼
   Google Satellite               Apps Script Web App
   Tiles (basemap)               (CORS-enabled JSON API)

• Frontend: Static HTML/CSS/JS hosted on GitHub Pages — zero build step
• Maps: Leaflet.js with Google Satellite + Labels basemap
• Drawing: Leaflet.draw plugin (polygon tool only)
• Backend: Google Apps Script web app → reads/writes Google Sheets
• Data: Per-district GeoJSON files + district boundary GeoJSON

Just open the live tool — no setup needed to browse polygons and view the map.

To enable saving verifications across users:

1. Go to sheets.google.com → create a new blank spreadsheet
2. Name it "Coconut Verifier Data"
3. Rename the first sheet tab to verifications
4. Add headers in Row 1: key | status | user | timestamp
5. Create a second sheet tab named drawn_polygons
6. Add headers in Row 1: id | district | geometry | area_ha | user | timestamp | note

1. In the Google Sheet → Extensions → Apps Script
2. Delete any existing code
3. Paste the contents of google_apps_script.js from this repository
4. Click Deploy → New Deployment
5. Type: Web app | Execute as: Me | Access: Anyone
6. Click Deploy → Authorize → Copy the Web App URL

1. Edit app.js line 8
2. Replace PASTE_YOUR_APPS_SCRIPT_URL_HERE with your Web App URL
3. Commit and push — GitHub Pages will update in ~1 minute

git clone https://github.com/Athithiyanmr/coconut_verification_tool.git
cd coconut_verification_tool
python -m http.server 8000
# or: npx serve .

Open http://localhost:8000 in your browser.

The raw raster was processed into web-ready GeoJSON using Python:

labels_coconut_raw_2020_tamilnadu.tif
        │
        ▼  rasterio.features.shapes() — vectorize coconut pixels (value=1)
   90,438 raw polygons
        │
        ▼  geopandas.sjoin() — assign each polygon to a district
   Polygons tagged with district name
        │
        ▼  Reproject to UTM 44N (EPSG:32644) — calculate area in hectares
   Area calculated per polygon
        │
        ▼  Unique ID per district (1, 2, 3... per district)
   Polygons numbered for systematic verification
        │
        ▼  Simplify geometry + round coordinates to 5 decimal places
   Web-optimized GeoJSON per district
        │
        ▼  district_boundaries.geojson — simplified district outlines
   38 files in data/ folder

coconut_verification_tool/
├── index.html                    # Main page
├── style.css                     # Styles
├── app.js                        # Application logic
├── google_apps_script.js         # Google Apps Script (paste into Apps Script editor)
├── server.js                     # Optional Express backend (for local/self-hosted use)
├── README.md
├── data/
│   ├── districts.json            # District index (name, polygon count, center, bounds, file path)
│   ├── district_boundaries.geojson  # Simplified district outlines for map display
│   ├── ariyalur.geojson          # Per-district polygon data
│   ├── chennai.geojson
│   ├── coimbatore.geojson
│   └── ... (38 district files)

• Descals, A., et al. (2023). High-resolution map of coconut palms using deep learning and openly available satellite images. Zenodo dataset
• Sentinel-2 satellite imagery: Copernicus Programme
• Tamil Nadu district boundaries: Survey of India

MIT

Athithiyan — GitHub | Geospatial Data Analyst & Climate Scientist

Built with geospatial Python (rasterio, geopandas, shapely) and vanilla web technologies.