Machine learning-based mapping of land use and land cover in the Nigerian red soil region

Accurate and location-specific data on land use and land cover (LULC) are essential for quantifying anthropogenic and climate-driven impacts in complex tropical environments. However, many existing LULC studies are constrained by coarse spatial resolution, limited spectral information, or inadequate integration of ancillary data such as topography factors that reduce classification accuracy in heterogeneous landscapes. This study evaluated the performance of two optical remote sensing datasets from Sentinel-2 and Landsat-8 for LULC mapping. Using imagery acquired between October to December of 2023, we tested three machine learning algorithms: Random Forest (RF), Classification and Regression Trees (CART), and Support Vector Machine (SVM) under three scenarios: (i) using only spectral bands (ii) spectral bands with additional bands, and (iii) spectral bands and topographic variables. Five major LULC types namely: water body, forest, low vegetation, built-up area, and bare land were identified and mapped. Scenario (iii) and RF classifier achieved the highest overall accuracy (89%) and a Cohen's kappa coefficient (0.86), while SVM performed least effectively (74% accuracy, kappa = 0.64). Although Landsat-8 imagery yielded slightly higher numerical accuracy, Sentinel-2 imagery resulted in finer spatial detail and more precise class delineation. These findings highlight the complementarity of both datasets and demonstrate that incorporating topographic predictors significantly enhances LULC classification performance. Overall, the study underscores the potential of freely available satellite data and machine learning integration for reliable, high-resolution LULC mapping in humid tropical regions by providing a foundation for improved environmental monitoring, resource planning, and climate adaptation strategies.