Digital Soil Mapping of Soil Taxonomic Classes: Prospects and Challenges for Machine Learning Integration in Nigeria

Reliable spatially explicit soil information is important for land-use planning, agricultural management, and food-security planning in Nigeria. Digital soil mapping (DSM), which combines soil observations with environmental covariates through spatial prediction functions, provides an opportunity to complement and progressively update existing soil information. This review examines the prospects and challenges of integrating machine learning (ML) into DSM of soil taxonomic classes in Nigeria. Peer-reviewed literature was sythesised to assess soil classification systems, environmental covariates, modelling approaches, validation practices, and the relative development of continuous soil-property mapping and discrete taxonomic-class prediction. The evidence retrieved at the time of literature search indicates that ML-based DSM of continuous properties, such as soil organic carbon, particle-size fractions, and other nutrient indicators is more developed in Nigeria than direct prediction of formal soil taxonomic classes. This highlights an important methodological and information gap, constraining evidence on direct ML-based taxonomic-class prediction and limiting the availability of spatially explicit taxonomic information for land-use planning, soil management, suitability assessment, and broader soil-resource decision-making. Key challenges identified include sparse and spatially clustered legacy data, limited high-resolution covariate availability, inconsistent harmonisation between classification systems, constrained computing and technical capacity, and poor model transferability across agro-ecological zones. Prospects for advancing the field include renewed legacy-data mobilisation, cloud-based access to high-resolution remote-sensing covariates, hybrid and deep-learning approaches, and deliberate investment in taxonomic-class-specific training datasets. Future Nigerian DSM research may prioritise direct classification of soil taxonomic units using ML, alongside refinement of continuous-property mapping, to build a coherent, nationally consistent soil information system.