Geospatial-Based Modelling of Woody Vegetation Patterns and Aboveground Biomass in The Saloum Delta, Senegal: A Pathway to Optimal Land Restoration

Abstract

The interplay of climate and anthropogenic pressures has led to significant degradation of vegetation in many regions, leading to land degradation and affecting ecosystem services. While evidence of land degradation is apparent, some areas also show promising signs of regreening, offering both challenges and opportunities for restoration. The Saloum Delta exemplifies these dynamics, yet there is a lack of detailed understanding of its vegetation patterns, particularly woody tree vegetation, which have implications in the management of the landscape. A spatially detailed assessment is critical to reconcile these contrasting trends and inform sustainable management strategies for optimal land restoration. The first objective assessed woody cover dynamics from 2002 to 2022. Random Forest algorithm (RF) was used for image classification in Google Earth Engine. Post-classification analysis such as change detection, fragmentation, and connectivity analysis was done using R software. The second objective assessed the environmental drivers of the spatial distribution of the woody cover and related habitat suitability. Species Distribution Models (SDM) were applied using GPS coordinates of the woody tree covers as occurrence data and ten environmental variables. Ensemble model with Maxent, General Linear Model (GLM) and RF were used. The third objective estimated the aboveground biomass (AGB) of the different woody tree covers using allometric equations and machine learning. AGB estimation integrated ground inventory data from 138 plots with Sentinel-2 imagery from dry and wet season. The machine learning models, included Random Forest (RF), K-Nearest Neighbor (K-NN), Super Vector Machine (SVM) and XGB (Gradient Boosting Model), to predict the AGB. First, the spatiotemporal analysis revealed that Mangroves dominate both Protected Forests (PF) and Outside Protected Forests (OPF), with significant gains from “Water” and “No Woody Cover.” Plantations in OPF showed progressive expansion, highlighting land-use shifts outside protected areas. Pattern analysis indicated increased connectivity and reduced fragmentation for Mangroves and Close Woodlands in PF. In contrast, Open Woodlands in OPF showed dynamic fragmentation patterns with an increase of small patches in Plantations. The assessment of woody tree spatial drivers reveals key environmental drivers, such as salinity and bulk density for Mangroves, rainfall and salinity for Close Woodlands, burn area index for Open Woodlands, and rainfall and proximity to villages for Plantations with spatial pattern highlithing their suitability for optimising their coverage. Mangroves accounted smallest gab between the actual coverage and suitable coverage with a gap of 3.47%. Strong gab still existed for the other woody tree with 5.49, 6.03 and 6.41% for Close Woodlands, Open Woodland and Plantations respectively. Lastly AGB estimation of the woody cover revealed Close Woodlands had the highest biomass density (295.08 Mg/ha), followed by Open Woodlands, Plantations, and Mangroves. Seasonal variability influenced predictions, with wet-season Sentinel-2 imagery yielding more accurate results. Random Forest models provided the highest accuracy (R² = 0.83, RMSE = 47.20). The findings filled the identified gap by employing advanced geospatial analyses and helped understand woody vegetation patterns, further suitable areas, and biomass potential. supporting future land management strategies for optimising land restoration and policy regreening.