Case study · Earth observation & restoration

Reviving Victoria.

An Earth-observation-guided restoration plan for the Lake Victoria South Basin, Kenya — integrating two decades of satellite, climate and geodetic data into a single, spatially-prioritised recovery strategy.

Wide aerial panorama of the southern basin of Lake Victoria — wetlands, river deltas and sediment meeting open water
The South Basin — known locally as Nam Lolwe — spanning Kisumu, Migori, Siaya and Homa Bay. 00°22′S, 033°12′E (WGS 84).
Region
Lake Victoria South Basin, Kenya
Framework
Multimodal Earth observation
Study period
1992 – 2025
Classifier accuracy
96.7% OA · κ 0.957

Three decades of fragmented evidence

Lake Victoria is the largest freshwater lake in Africa and a lifeline for millions across Kenya, Uganda and Tanzania. Yet its South Basin has suffered three decades of rapid degradation — deforestation, agricultural and settlement encroachment, artisanal mining, and climate-driven shifts in rainfall.

Earlier studies each captured a single facet — land cover, water quality, or lake level — across different periods and scales. None had synthesised these signals into one basin-wide picture that decision-makers could act on. That gap is what this work set out to close.

No single dataset explains the decline. The recovery plan had to read land, water and climate as one connected system.

A multimodal Earth-observation framework

GWO integrated four independent satellite and reanalysis records into a single analytical framework, processed in Google Earth Engine and QGIS — each chosen to constrain a different driver of change, with uncertainty stated throughout.

Sentinel-2 SR 2017–2025
Surface-water quality indices — chlorophyll-a (NDCI) and turbidity.
Landsat 7/8/9 2005–2025
Random-Forest land-use / land-cover across five epochs.
MODIS Terra LST 2000–2024
Land-surface thermal response.
ERA5-Land 2000–2023
Near-surface (2 m) air-temperature trend.
Copernicus altimetry 1992–2024
Basin-mean lake water-surface height.

What the basin told us

A Random Forest classifier (200 trees, 70/30 split) reached 96.7% overall accuracy (κ = 0.957) and was applied uniformly across all five epochs. The 2005–2025 land-cover transitions are stark:

−26.2%
Vegetation loss — 1,120 km² of forest and woodland cleared.
+75.0%
Urban expansion — 9,574 km² of new built-up land.
+15.0%
Bare soil — 6,364 km² from farming, mining & clearing.
−17.4%
Cropland — 14,275 km² converted to bare soil & built-up.
98.7%
Water persistence — the most stable class in the basin.
+1.21m
Lake-level jump — anomalous 2019–20 rise from IOD rainfall.
Land-use / land-cover classification map of the Lake Victoria South Basin, 2025 — water, vegetation, cropland, urban and bare soil
Fig. 1 — Random-Forest LULC classification, Lake Victoria South Basin, 2025 (Landsat). Urban expansion and vegetation loss are the dominant transitions.

Water quality & a shifting climate

Satellite water-quality indices document persistent eutrophication in the Winam Gulf: the chlorophyll-a proxy (NDCI) rose steadily and peaked in 2023, while falling turbidity signalled a shift from sediment-dominated to algae-dominated waters — a hallmark of advancing eutrophication.

ERA5-Land shows atmospheric warming of 0.17 °C decade⁻¹, while the post-2019 lake-level anomaly has persisted alongside elevated flood frequency — evidence of compounding hydro-climatic stress rather than a single gradual trend.

Satellite view of the South Basin shoreline with sediment plumes entering the lake from river deltas
Fig. 2 — Sediment plumes from the Kuja–Migori catchment entering the Winam Gulf — the near-shore zone where eutrophication is most acute.

Low-cost monitoring with GNSS reflectometry

To make ongoing monitoring affordable, GWO ran an in-house multi-GNSS orbit simulation across all four constellations — GPS, Galileo, GLONASS and BeiDou. A modest ground-based GNSS-R network could deliver 150–250 water-surface measurements per day, centimetre-accurate, at 25–50 km spacing — sub-daily resolution at a fraction of the cost of conventional tide-gauge or radar-altimetry infrastructure.

The recommendation: prioritise, don’t generalise

The signals converge on one conclusion — restoration should be spatially targeted, not uniformly applied. GWO’s central recommendation is a Restoration Priority Index: a single composite layer combining vegetation-loss magnitude, water-quality deterioration and flood exposure.

  • 01

    Targeted afforestation

    Direct replanting to the sub-catchments with the steepest measured forest loss.

  • 02

    Near-shore pollution mitigation

    Concentrate nutrient and sediment controls on the Winam Gulf hotspots flagged by NDCI.

  • 03

    Habitat-linked fisheries management

    Tie fisheries policy to measured habitat availability and water-quality recovery.

  • 04

    Deploy & validate GNSS-R

    Stand up a low-cost receiver network for sub-daily lake monitoring, with Kenyan agencies.

The framework is designed to extend across the trans-boundary basin in partnership with Uganda and Tanzania — turning a one-off assessment into a living monitoring system.

Sources & data

This case study is a condensed summary of independent research. For full data sources, datasets and references, please get in touch.

Work with GWO

Facing a similar land or water question?

If you need a baseline you can defend and a strategy decision-makers can adopt, tell us about the site and the question in front of you.

Lake shoreline at dawn — reeds, wet sediment and still water under soft light