Geostatistical Modelling of Soil Carbon Spatial Variability and Prediction in the Forest-Savanna Transition Zone of Nigeria

Accurate spatial prediction of soil organic carbon in Nigeria's forest-savanna transition zone spanning the North Central and South West geopolitical zones is critical for national greenhouse gas inventories, soil fertility management, and climate change mitigation planning, yet existing soil carbon maps for this ecologically significant zone are characterised by coarse resolution, outdated source data, and inadequate uncertainty quantification. This study develops and evaluates geostatistical and machine learning methods for high-resolution soil organic carbon spatial prediction in the Nigerian forest-savanna transition zone using a novel multi-depth sampling framework and comprehensive environmental covariate integration. Soil cores were collected to 100 centimetres depth at 420 systematically sampled sites across a 150,000-square-kilometre study area spanning parts of Oyo, Kwara, Kogi, Benue, and Niger States. Organic carbon content was measured at four depth increments and used to compute soil organic carbon stocks in tonnes per hectare. Environmental covariates including climate variables from CHELSA, spectral indices from Sentinel-2, SRTM terrain attributes, and parent material classification from geological maps were integrated in prediction model frameworks. Random forest, gradient boosting, and ordinary kriging methods were compared using nested cross-validation, and uncertainty in predictions was propagated using Monte Carlo approaches. Results demonstrate that gradient boosting achieves the highest predictive accuracy (RMSE = 8.2 Mg C per hectare, R² = 0.71), substantially improving on the FAO/IIASA global soil carbon product (R² = 0.43) when validated against independent held-out sites in the study region. The study produces the highest-resolution soil organic carbon map yet developed for Nigeria's transition zone, with associated uncertainty maps. This work advances digital soil mapping methodology and contributes critical data to Nigeria's REDD-plus national programme. Keywords: soil organic carbon, geostatistics, forest-savanna transition, digital soil mapping, Nigeria.