Leveraging Geospatial Techniques to Monitor Restoration Efforts and Assess Associated Forest Ecosystem Services: Case Study of Maasai Mau Forest in Kenya

The Maasai Mau Forest is part of Kenya’s Mau Forest Complex, which is one of East Africa’s most vital tropical water towers. Over the recent decades, it has experienced significant disturbances from encroachment, unauthorised livestock grazing, and illegal abstraction of forest resources. These pressures have significantly altered the structural composition of forests, reduced aboveground biomass, and weakened critical ecosystem services. Currently, countries have engaged in national initiatives aimed at forest ecosystem restoration, including AFR100, REDD+, and SDG15, and Kenya has an ambitious objective to plant 15 billion trees by 2032. These offer a significant opportunity to assess the large-scale recovery of tropical forests. However, monitoring efforts remain limited by weak indicators, a lack of baseline data, and inconsistent reporting systems.

This study examines disturbances and recovery in restored areas of the Maasai Mau Forest using an integrated remote sensing and machine-learning approach. The analysis focuses on assessing vegetation growth, area restored in hectares and carbon sequestered during the process using Sentinel-1 radar, Sentinel-2 optical imagery, GEDI lidar, and ground measurements. They are used to quantify spatial, temporal, and structural (3D) changes in vegetation following disturbance and during recovery.

Data from 2019 to 2025 were processed to develop fused satellite products for the region of interest. Sentinel-1 (VV, VH) was corrected and speckle-filtered using the refined Lee method, while Sentinel-2 imagery was cloud-masked and reduced to relevant spectral bands. From these datasets, radar-based indices, VV/VH ratios, and optical vegetation indices (NDVI, EVI, SAVI, PSSRa) were derived. The indices and selected bands were fused, and principal component analysis (PCA) was performed to generate harmonized inputs for classification.

K-means clustering was applied to the PCA outputs and subsequently labelled as forest and non-forest classes. NDVI was also used to derive annual indices and assess time-series trends. Comparing the classified outputs over time enabled a change detection of forest loss and gain. NDVI-based thresholds and temporal metrics were combined with classified outputs to identify restored areas and map vegetation recovery trajectories.

The results show a clear pattern of forest regeneration. NDVI analysis and satellite-based classification indicate a stable increase in forest cover and a decline in non-forest areas between 2019 and 2025. Dense vegetation increased after 2023, while moderate vegetation declined and sparse vegetation remained relatively stable, with a trend of y = 2336.2x + 30453 (R² = 0.510). PCA-based classification shows forest cover increasing from 32,424 ha to 36,791 ha, while non-forest areas decreased from 13,751 ha to 9,385 ha. Linear trend analysis supports this positive trajectory (forest: y = 1202.7x + 31066, R² = 0.643; non-forest: y = –1202.7x + 15110, R² = 0.643), suggesting a progressive transition from non-forest to forested conditions.

This research shows how tropical forests regenerate after disturbance and enhances understanding of vegetation response to structured efforts. The findings offer valuable evidence for policymakers, conservation planners, and climate practitioners aiming to strengthen restoration outcomes across tropical landscapes.