Developing an Original Digital Soil Mapping Framework for Precision Soil Classification in Nigeria Using Machine Learning and Multi-Source Geospatial Data

This dissertation develops and validates an original digital soil mapping framework for precision soil classification across Nigeria, integrating machine learning algorithms with multi-source geospatial environmental covariates to produce the first national-scale soil classification map of Nigeria at 90-metre spatial resolution. Existing Nigerian soil maps are based on 1970s and 1980s surveys at small scale with inconsistent classification and sparse sampling, creating a severe knowledge gap for national and regional soil-based planning for food security, climate modelling, and land degradation assessment. Using a systematic sampling design, 1,200 pedon observation points were collected across all six geopolitical zones, with full morphological description and horizon-by-horizon laboratory analysis for classification. Environmental covariates including SRTM DEM derivatives, Landsat spectral indices, MODIS land cover time series, geologic maps, and bioclimatic layers were assembled in a GIS. Three machine learning algorithms (random forest, gradient boosting, and convolutional neural network applied to covariate stacks) were compared for classification accuracy at the Soil Taxonomy subgroup level. An ensemble classifier combining all three approaches was selected as the final mapping model. The model was validated by ten-fold cross-validation and an independent holdout pedon dataset. The dissertation proposes the Nigerian Digital Soil Classification Framework (NDSCF) as an original methodological and applied contribution, achieving 78 percent overall accuracy for great group classification and 71 percent for subgroup, representing a substantial improvement over existing map accuracy. A web-based interactive viewer of the classification map is provided as a public good. Keywords: digital soil mapping, machine learning, Nigeria, soil classification, geospatial data.