Context
A national environmental agency needed a defensible, repeatable land cover change dataset to support a stalled policy decision on protected-area boundary revisions.
Approach
We processed a 10-year Landsat 8 and Sentinel-2 time series through Google Earth Engine, applying a random-forest classification trained on 1,200+ ground-truth points, followed by post-classification change detection between 3 epochs (2014, 2019, 2024).
Outcome
The resulting change dataset — with quantified deforestation, urban expansion, and wetland loss statistics per administrative region — gave the agency the evidence base to move the boundary revision forward.
Methodology
10-year Landsat 8 and Sentinel-2 time series ingestion via Google Earth Engine; random-forest supervised classification trained on 1,200+ field-verified ground-truth points; post-classification change detection across 3 epochs (2014/2019/2024); accuracy assessment against an independent validation set (overall accuracy 89%).
Results & Outcomes
- Classified land cover across the full national extent at 10m/30m resolution
- Quantified a 14% net forest cover decline over the 10-year study period
- Delivered per-region change statistics used directly in policy submission
- Achieved 89% overall classification accuracy against independent validation
Challenges
Cloud cover in the country's humid southern region required a multi-sensor, multi-date compositing approach to assemble cloud-free annual mosaics rather than relying on single-date imagery.
“The remote sensing change-detection analysis gave us defensible evidence for a policy decision that had been stalled for two years.”
Fatima Al-Rashid, Environmental Scientist
A National Environmental Agency