Assessment of the accuracy and spatial consistency of global land use and land cover maps in semi-arid regions of Zimbabwe.

Semi-arid regions of Zimbabwe, southeastern Africa, comprise fragile ecosystems exposed to anthropogenic pressure and highly sensitive to climate variability. Land Use and Land Cover (LULC) maps are essential for monitoring these dynamic environments, supporting the detection of land cover disturbances and the identification of key drivers of landscape transformation. Global LULC products have gained prominence due to their broad coverage, free access, and fine spatial resolution. Among them, the 10 m products from ESRI (Environmental Systems Research Institute) and Dynamic World (DW), developed by Google and the World Resources Institute, stand out for their annual and near real-time updates, respectively. Selecting the most suitable product remains challenging in heterogeneous semi-arid landscapes. This study (1) analyzes the spatial consistency of 2024 ESRI and DW LULC products in Binga, Gwanda, and Mwenezi, Zimbabwe, and (2) evaluates their accuracy using field-based reference data. Spatial consistency was assessed via map overlay, and accuracy via an error matrix using Overall Accuracy (OA) and Kappa. Agreement between products reached 68.34% in Binga, 71.16% in Gwanda, and 63.89% in Mwenezi, with the highest concordance in the rangeland class (78.84%-68.44%). Divergences were most pronounced in cropland, trees, and built-up areas. Accuracy ranged from 93.30% to 87.01% (95% confidence interval), with Kappa values between 0.91 and 0.83. ESRI performed better in the drier districts of Gwanda and Mwenezi, while DW showed higher consistency in Binga, where slightly greater rainfall promotes environmental heterogeneity. These findings align with studies showing that subtle rainfall and vegetation variations influence global LULC product performance, underscoring the need for regional validation of global datasets in semi-arid environments. This analysis relied on 2024 data, which was the only reference year available due to logistical and financial constraints in field campaigns across these remote, hard-to-access districts. Extending this validation to a multi-year series is recommended for future research to capture interannual variability in product performance.