Hierarchical eco-zone delineation in data-scarce Afrotropical river landscapes using global datasets

• Machine-learning framework for hierarchical eco-zonation in data-scarce Afrotropical regions. • Integrates open-source global datasets and free-access software for bio stratification. • Delineates 5 ecoregions subdivided at 3 hierarchical bioregion levels for Uganda. • Ecological coherence validated at coarse tiers; finer tiers limited by thematic resolution. • Framework relevant for supporting biodiversity and conservation planning and assessment. Existing environmental management frameworks in the Afrotropical region often rely on drainage divides, taxonomic patterns, or sector-specific classifications; however, they frequently fail to capture the ecological heterogeneity required for broad ecological applications. These shortcomings stem from the scarcity of reliable, localised datasets and appropriate stratification methods. This study integrates multiple machine learning algorithms within a systematic workflow, using open-source software and globally accessible datasets, to develop a hierarchical eco-zonation framework. Using Uganda as a case study, climate, topography, and hydrological variables were used to model eco-zonation maps. At the broadest level, five ecoregions (EcoR) were delineated and subsequently subdivided at three finer hierarchical levels termed Bioregions (BR). Ecological relevance was assessed against potential natural vegetation (PNV) using a spatially dispersed 10-fold cross-validation approach. Predictive accuracy declined from 80.35% at the EcoR level to 52.69% at BR (III). The two coarsest tiers (EcoR and BR (I)) demonstrated strong ecological coherence, whereas BR (II) and BR (III) showed weak coherence. Although finer tiers lack empirical ecological validation, they represent experimental abiotic subdivisions designed to capture theoretical fine-scale heterogeneity. The decline in ecological coherence across scales highlights a fundamental limitation of best-available, globally and regionally curated biota datasets: their thematic resolution is insufficient for robust, finer-scale ecological stratification. This underscores the need for local curation of consistent datasets or development of integrative, high-resolution datasets. Overall, the framework offers a flexible, open-data and open-software approach that can be applied hierarchically across multiple national and subnational scales for ecological stratification and biogeographical analysis supporting biodiversity assessment and conservation planning in data-scarce Afrotropical landscapes.