GeoAI land degradation mapping, Garissa County
Spot degrading land years earlier, from space.
Machine learning on Sentinel-2 and Landsat time series in Google Earth Engine to map where vegetation cover and soil condition are declining, flag hotspots and track recovery after interventions such as reseeding and Prosopis clearance.
The problem
Degradation spreads unnoticed until pasture and farmland are lost.
What we built
- 1
Build multi-year vegetation and bare-soil indices from Sentinel-2 and Landsat in Earth Engine.
- 2
Train a classifier on field-verified sample points to separate stable, degrading and recovering land.
- 3
Validate on locations the model has not seen, and map uncertainty alongside results.
- 4
Publish hotspot maps for planners and track change after each intervention.
Highlights
- Free, frequent satellite coverage instead of costly field surveys
- Trend analysis over many seasons, not a single snapshot
- Designed to measure the effect of restoration work
Built with
- Google Earth Engine
- Sentinel-2
- Landsat
- Random forest
- Python
Honest status: Work in progress. Maps will be published once field validation is complete.
TovuTech, in progress.
Want something like this?
We design, build, host and secure systems like this end to end, and train your team to run them.
More success stories