Democratizing Land Cover Mapping

Drive. Record. Map.
Your dashcam turns every journey into open land cover data for Indonesia.

Join as contributor How it works
—
georeferenced images
—
with GPS coordinates
—
contributors
—
objects detected

Live Land Cover Map

Explore all approved images on an interactive map — filter by date, land cover class, and location.

Open live map

The LULC data challenge

Land cover classification from satellite imagery needs three ingredients: imagery (increasingly accessible), algorithms (sophisticated), and ground truth data — the major bottleneck.

  • Cost barrier: traditional field surveys cost $50–200 per sample, requiring GPS devices, trained personnel, and dedicated campaigns.
  • Temporal gap: historical training data is virtually non-existent, preventing validation of past classifications.

This excludes most researchers and organizations in developing regions from comprehensive LULC research.

The dashcam solution

Consumer-grade dashcams with GPS generate thousands of georeferenced images per day during routine travel. After a one-time $100–300 investment, the marginal cost approaches zero.

  • Opportunistic collection along transportation corridors
  • Web-based collaborative annotation and correction
  • Rich temporal resolution for multi-temporal analysis

How it works

1. Capture & process

Dashcam video is automatically converted into georeferenced still images.

2. OCR extraction

An OCR engine reads GPS coordinates, dates, times, and speed from the image overlay.

3. Quality control

Suspicious coordinates are flagged for human review — automated processing with human validation.

4. Label & export

Collaborators annotate land cover and export datasets as CSV/GeoJSON for model training.

Platform features

Camera position fix

Correct GPS coordinates and OCR misreads directly on an interactive map.

Land cover labeling

Annotate roads, vegetation, buildings, water and more for vision AI analysis.

Map visualization

Browse processed images on an interactive map of vehicle tracks.

Flexible export

Map-based area selection with date and class filters; CSV and GeoJSON output.

Vision AI analysis

AI-powered land cover scene analysis using Anthropic Claude vision models.

Collaborative roles

Viewers browse datasets, contributors process and correct data, admins manage quality.

Applications

  • Training data for Sentinel-2 / Landsat classification
  • Multi-temporal analysis validation
  • Phenological modeling support
  • Urban planning and infrastructure monitoring

Open science

We invite researchers, government agencies, and citizen scientists to build upon this foundation, contribute data, adapt the platform to local contexts, and share improvements with the community.

Together, we can democratize LULC data collection globally.

Turn your daily drive into science

Register as a contributor to process dashcam imagery, correct OCR readings, and label land cover — or browse and download the growing open dataset.

Create an account Log in