How GIS is Being Used to Fight Deforestation in the Mau Forest

The Mau Forest Complex is not simply Kenya's largest indigenous forest — it is the country's most critical hydrological infrastructure. From its 400,000 hectares of highland montane forest flow twelve rivers that fill Lakes Nakuru, Baringo, Bogoria, and Turkana; feed the Mara River that sustains the wildebeest migration and the Masai Mara National Reserve; supply water to over five million people in Nakuru, Kericho, Bomet, Narok, and Trans Nzoia counties; and generate a significant fraction of Kenya's inland freshwater available for agriculture and urban supply. When the Mau loses trees, Kenya loses water. And for decades, the forest has been losing trees — through encroachment, excision, charcoal production, and agricultural conversion — at a rate that outpaced both governance capacity and monitoring ability. GIS and remote sensing have fundamentally changed the monitoring equation. This article explains how.

The Scale of the Problem: What Decades of Loss Looks Like in Data

Before examining how GIS is being used, it is necessary to understand what the spatial data actually shows when the entire historical record is compiled. The Mau Forest Complex at its estimated extent in the early twentieth century covered approximately 400,000 hectares — a continuous highland forest belt straddling six counties from Nakuru in the east to Bomet and Trans Nzoia in the west, functioning as a single ecological unit despite administrative boundaries.

By 2008, when the Government of Kenya established the Prime Minister's Task Force on the Conservation of the Mau Forests Complex — the highest-level political recognition that the forest was in existential crisis — an estimated 107,000 hectares had been lost or severely degraded since independence. The Taskforce's own GIS analysis, drawing on digitised aerial photography from the 1970s and 1980s combined with Landsat satellite imagery up to 2008, mapped the progressive loss in spatial detail that had never previously been compiled. The maps were politically explosive: they showed not just how much had been lost, but precisely which excisions had occurred when, under whose authority, and in which constituencies.

The most severely affected section was the Eastern Mau — the block most accessible from the Rift Valley floor — where forest cover had been reduced from an estimated 55,000 hectares in 1973 to approximately 29,000 hectares by 2005. The Transmara section, straddling the Narok-Trans Nzoia boundary, had similarly seen extensive conversion to tea and maize smallholdings, driven by land allocation by successive county councils in the 1980s and 1990s. The South West Mau, home to the Ogiek community whose ancestral tenure within the forest was recognised by the African Court on Human and Peoples' Rights in 2017, showed a more complex loss pattern — legally authorised and illegal clearing intertwined in ways that only fine-resolution spatial analysis could untangle.

The GIS and Remote Sensing Toolkit

The change detection analysis described above, and all the operational monitoring that has followed, depends on a specific set of geospatial tools and data sources. Understanding the toolkit — what each component does and why it is used — is essential for anyone working in forest conservation, environmental monitoring, or land governance in Kenya who wants to understand both the capabilities and the limitations of the spatial data they are receiving.

How Change Detection Actually Works: From Pixels to Prosecution

The phrase "change detection" encompasses a range of analytical techniques, but all of them rest on the same core principle: comparing the spectral characteristics of a landscape at two different points in time to identify where significant changes have occurred. For forest monitoring in the Mau, the process moves through five stages from satellite acquisition to actionable field intelligence.

Stage 1: Image Acquisition and Pre-Processing

Raw satellite imagery is acquired from the satellite provider (ESA, USGS, or Planet) and pre-processed before analysis. Pre-processing involves converting raw digital numbers to surface reflectance values (correcting for atmospheric scattering and absorption), and co-registering images from different dates to ensure they are aligned to the same spatial reference — so that pixel A in the 2015 image corresponds to exactly the same ground location as pixel A in the 2024 image. In cloud-prone environments like the Mau Forest — which receives 1,200–2,000 mm of rainfall annually — cloud masking is a critical pre-processing step. Pixels covered by cloud or cloud shadow in an image are excluded from the analysis, and composite images are built from the clearest observations within a defined time window (typically a three-month dry season period) to maximise usable coverage.

Stage 2: Vegetation Index Computation

NDVI is computed for each cloud-masked image: a pixel-by-pixel calculation that produces a value between -1 and +1 for every cell in the image. Dense forest returns NDVI values of 0.6–0.85 in a healthy state. Grassland and degraded scrub returns 0.3–0.55. Bare soil, recently cleared land, and active agriculture typically return 0.05–0.2. A second index — the Normalised Difference Moisture Index (NDMI), computed from near-infrared and shortwave infrared bands — is added to distinguish moisture stress (which temporarily reduces NDVI) from true vegetation loss.

Stage 3: Temporal Change Analysis

The change detection analysis compares NDVI values from a baseline period (e.g. 2000–2005 mean) to values in the analysis year. Pixels where NDVI has declined by more than a threshold value — typically 0.2 in sustained decline — are flagged as candidate loss areas. Statistical methods such as Continuous Change Detection and Classification (CCDC), developed by Boston University and implemented in Google Earth Engine, fit time-series models to every pixel's historical NDVI trajectory and detect anomalous departures from the expected seasonal pattern — distinguishing the abrupt, step-change signature of clearing (which appears as a sudden, permanent NDVI drop that does not recover with rainfall) from the gradual decline associated with drought or fire damage.

Stage 4: Classification and Area Quantification

Detected change pixels are classified into loss categories — forest to agriculture, forest to bare soil, forest to settlement — using a combination of spectral signatures and contextual rules. Classification accuracy is assessed through independent validation points: locations where the satellite-derived classification is checked against high-resolution imagery or field visits to determine whether the detected change is real or a false positive. For GFW's annual loss product, overall classification accuracy across Africa has been independently validated at 85–92% for confirmed tree cover loss. Change area statistics are then computed: how many hectares were lost in each administrative unit (county, gazetted forest block, specific conservation area) over the analysis period.

Stage 5: Alert Generation and Field Dispatch

GLAD near-real-time alerts — based on comparison of the most recent 8-day Landsat composite against the baseline — are automatically generated and delivered by email or API to subscribed users. In Kenya, the Kenya Forest Service and several NGOs receive these alerts at the gazetted forest boundary level. The alert includes the geographic coordinates of the detected loss event, the area of the alert polygon in hectares, the date of confirmed detection, and a confidence level. A KFS ranger station that receives a 3-hectare high-confidence alert in the Eastern Mau can have a team at the GPS coordinates within hours — before any additional clearing has occurred.

Four Ways GIS is Being Applied in the Mau Today

A Decade of GIS-Informed Policy: Key Milestones

What GIS Cannot Do: Honest Limitations

A balanced account of GIS in Mau Forest conservation must acknowledge what the technology cannot deliver — both to set appropriate expectations for practitioners and to identify where complementary approaches remain essential.

Satellite change detection identifies deforestation events; it does not prosecute them. The gap between a detected alert and a successful prosecution is filled by rangers, prosecutors, courts, and political will — all of which are subject to pressures and failures that no remote sensing system addresses. Kenya's forest crime prosecution rate remains low relative to the number of detected violations, reflecting resource constraints in the prosecution system rather than failures in detection.

Cloud cover limits Sentinel-2 effectiveness in the Mau's wet seasons. The forest's rainfall pattern — bimodal, with long rains from March to May and short rains from October to December — means that some of the most active encroachment periods coincide with the highest cloud cover. Annual NDVI composites mitigate this by drawing on the clearest observations within a season, but rapid clearing events that initiate and complete within a cloudy period may not generate a timely GLAD alert. This is the operational gap that the more costly Planet Labs daily tasking is designed to fill for critical hotspot areas.

30 m Landsat resolution misses small-scale incremental encroachment. A pattern of slow boundary advance — where settlers incrementally expand their plots by a few metres per year — may not generate a detectable alert in any single year but can amount to significant cumulative loss over a decade. This type of encroachment is better detected through annual high-resolution drone surveys of the forest boundary rather than satellite-based change detection.

Finally, and most importantly: the fundamental drivers of Mau deforestation are socioeconomic and political, not technical. Population pressure on forest margins, land tenure insecurity among adjacent communities, the economic value of agricultural land relative to standing forest, and the political economy of forest excisions — these are the forces that GIS can monitor but cannot directly address. The technology is a governance tool, not a governance solution. Its value depends entirely on the institutional capacity and political commitment to act on what the maps reveal.

What Comes Next: Emerging GIS Applications in Mau Conservation

The next generation of GIS applications in the Mau Forest are moving from retrospective monitoring toward predictive and preventive intelligence — shifting the role of spatial data from recording what has been lost to forecasting where loss is most likely to occur next and enabling pre-emptive governance responses.

Deforestation risk modelling uses historical loss data combined with socioeconomic variables — road proximity, population density, land parcel size, agricultural commodity prices, and proximity to existing encroachments — to generate probability maps predicting which forest parcels are at highest risk of clearing in the next three to five years. These models, developed by the Spatial Finance Initiative and KEFRI in partnership, enable KFS to deploy ranger resources proactively to high-risk areas rather than reactively to detected events.

Integrated land parcel overlays are linking Ardhisasa land registry data to the GIS forest boundary layer in real time — flagging every land transaction involving a parcel adjacent to or within the gazetted boundary for automatic review. When a parcel transfer is registered on Ardhisasa and the parcel centroid falls within 200 metres of the gazetted forest boundary, the transaction is flagged and a verification request generated for the county environment department. This administrative integration of GIS and land registry data is one of the most promising governance innovations currently being piloted.

Carbon credit monitoring and verification for Mau restoration projects under the voluntary carbon market is increasingly demanding satellite-based quantification of above-ground biomass change — using methods that combine LiDAR canopy height models with NDVI time series to produce biomass estimates at the plot level. As the market for African forest carbon credits grows, the technical rigour of monitoring, reporting, and verification (MRV) methodology will determine which projects succeed in attracting international climate finance.