The platform for watershed intelligence
The platformto analyze
Empowering communities and conservationists to understand what our planet is telling us — turning silent satellite observations into thoughtful, proactive care for every living watershed.
49.1%Model 1 mean IoU
7land-cover classes
10mSentinel-2 resolution
3trained sites
6-bandR,G,B,NIR,NDVI,NDWI
— Capabilities
Everything you need.
Nothing you don't.
01
Land Cover Segmentation
A U-Net (ResNet18 encoder) reads a 6-channel Sentinel-2 stack and classifies every 10m patch into 7 land-cover types.
02
Temporal Change Detection
Diffs two Model 1 passes into new water, new construction, degradation, and vegetation gain — no separate training needed.
03
Watershed Condition Scoring
A transparent, weighted composite of the land-cover mix — not a black box, every point traces back to a class weight.
04
Field Photo Verification
Geo-tagged photo EXIF GPS extraction, matched against the satellite-predicted class at that exact coordinate.
— How it works
Three steps.
Infinite precision.
workflow.ts
Running...
Four trained sites.
Untouched inference anywhere.
4
Training Sites
7
Land Cover Classes
49.1%
Mean IoU
Coverage ZonesLive --:--:--
Kadwanchi Watershed
Maharashtra
Tamhini Ghat Forest
Maharashtra
Donimalai Iron Mine
Karnataka
Custom Coordinates
Live Geocoder
Performance you
can measure.
Live --:--:--0
Satellite Scenes Processed
0.0%
Model Mean IoU
0-band
Multispectral Input Stack
0
Land Cover Classes
Built with data from
ISROMinistry of Rural DevelopmentSIH 2026BhuvanEarth Engine CommunitySTAC IndexESA WorldCover