A conservation-focused NGO working in the Kenyan highlands needed a defensible, spatially explicit basis for deciding where to concentrate a five-year soil-and-water conservation program across a large, degraded catchment. Without a current land-use baseline or a formal watershed delineation, the NGO's targeting had previously relied on field staff intuition rather than catchment-wide erosion risk data.
GeoVeris delineated the catchment into sub-basins from a conditioned digital elevation model, then built a 16-year land-use/land-cover time series from Landsat imagery to quantify how forest cover, cropland, and degraded land had shifted across the catchment. Combining that land-use history with slope, soil erodibility, and rainfall erosivity layers in a RUSLE-based erosion model let us rank sub-basins by estimated sediment yield rather than by visual impression alone.
The resulting maps gave the NGO a catchment-wide erosion risk ranking it had never had before, along with hard evidence of a 26% loss of natural forest cover over the study period — a figure that reframed the scale of the intervention needed. The conservation targeting maps were adopted directly into the NGO's five-year watershed rehabilitation program.
Five sub-basins, together responsible for an estimated 60% of catchment-wide sediment yield, are now the program's first-phase intervention priority.
Methodology
Delineated watershed and sub-basin boundaries from a hydrologically-conditioned 12.5 m ALOS PALSAR DEM using standard hydrology GIS tools, cross-checked against a finer 5 m contour-derived DEM for the upper 30% of the catchment where PALSAR artifacts were most pronounced. Produced a multi-temporal land-use/land-cover classification (2005, 2013, 2021) from Landsat 5/8 imagery using a random-forest classifier trained on 850 field-verified ground-truth points, achieving an overall classification accuracy of 89% (kappa 0.85). Combined the LULC time series with slope, soil erodibility, and rainfall erosivity layers in a RUSLE-based erosion risk model to rank sub-basins for conservation intervention.
Results & Outcomes
- Delineated the 2,400 km² catchment into 14 sub-basins prioritized by erosion risk for targeted intervention planning
- Quantified a 26% loss of natural forest cover across the catchment between 2005 and 2021
- Identified 5 sub-basins responsible for an estimated 60% of catchment-wide sediment yield
- Conservation targeting maps adopted directly into the NGO's 5-year watershed rehabilitation program
Challenges
PALSAR DEM artifacts in the steep upper catchment produced spurious pit-and-peak noise that distorted automated flow-direction results, initially fragmenting several sub-basins into unrealistic slivers. We resolved this by fusing in a higher-resolution contour-derived DEM for the affected 700 km² area and applying an iterative fill-and-breach conditioning workflow, rather than a blanket sink-fill, which preserved real channel network detail a standard fill operation would otherwise have flattened.