Four decades (1984–2024) of land use and land cover change in Zambia’s Upper Kafue River Basin using ensemble machine learning

Abstract

Land use and land cover (LULC) change is a critical driver of ecosystem services, biodiversity, and water resource sustainability. The Upper Kafue River Basin (UKRB) in Zambia has undergone significant transformations over the past four decades, and this study provides the first multi‑decadal ensemble‑based classification of LULC change in the basin. Multi-temporal Landsat Level-2 surface reflectance imagery (1984, 1994, 2004, 2014, 2024) was classified into five categories agriculture, grassland, forest, built-up land, and water using three base classifiers; Support Vector Machines (SVM), Random Forest (RF), and k-Nearest Neighbours (k-NN). A stacking ensemble with XGBoost as the meta-learner integrated the base model outputs. Accuracy was assessed using independent stratified Google Earth samples and field knowledge. The ensemble consistently outperformed individual classifiers (Overall Accuracy >93%, F-scores ≥0.94). Agriculture expanded by 106.9%, built-up land by 33.5%, and grassland by 41.0%, while forest declined by 19.9%. Water showed a negligible net increase of 0.08%. The ensemble improved classification consistency and reduced misclassification, particularly for agriculture and built-up areas. These findings highlight the growing anthropogenic footprint in the UKRB. The ensemble approach substantially enhances reliability of LULC monitoring in heterogeneous landscapes, providing a robust evidence base for policy interventions. Results underscore the urgency of sustainable land management, balancing agricultural and urban growth with forest conservation and water security.