AUTHOR=Manyothwane Tshepho , Mengistu Tsidu Gizaw 

TITLE=Challenging global generalizations: superior land cover mapping in Botswana with a locally trained transformer model

JOURNAL=Frontiers in Remote Sensing

VOLUME=Volume 6 - 2025

YEAR=2025

URL=https://www.frontiersin.org/journals/remote-sensing/articles/10.3389/frsen.2025.1654692

DOI=10.3389/frsen.2025.1654692

ISSN=2673-6187

ABSTRACT=IntroductionAccurate and high‐resolution land use and land cover (LULC) classification remains a critical challenge in ecologically diverse and spatially heterogeneous dryland environments, particularly in data-scarce regions. Botswana, with its complex environmental gradients and dynamic land cover transitions, exemplifies this challenge. While global products such as ESA WorldCover, Dynamic World (DW), and ESRI Land Cover provide valuable baselines, their accuracies remain limited (with an overall accuracy of 65–75%) and often fail to capture fine-scale spatial and thematic details.MethodologyThis study presents one of the first applications of Transformer-based deep learning models for national‐scale LULC mapping in Botswana. The model was trained on Landsat 8 OLI imagery, integrating field observations, Dynamic World-derived labels, and Google Earth validation to construct reliable training datasets in data-limited regions. Qualitative assessments were conducted using true and false color composites, vegetation and water indices, and expert validation to evaluate the model’s ability to delineate complex land cover features.Results and discussionsThe Transformer-based model achieved an overall accuracy of 95.31% on the testing dataset, with a Total Disagreement (TD) of 4.69%, primarily driven by Allocation Disagreement (AD = 3.44%) rather than Quantity Disagreement (QD = 1.25%). This indicates accurate estimation of class proportions with some misplacement of classes. F1-scores of 0.80 or higher for most land cover categories reflect strong thematic performance. Compared to the global DW product, the model demonstrated superior spatial detail, class-wise accuracy, and robustness, particularly in urban areas and ecologically sensitive zones such as the Makgadikgadi Pans and Okavango Delta. Temporal LULC trajectories reconstructed for 2014, 2019, and 2024 effectively captured major land change processes, including cropland expansion, grassland regeneration, and seasonal flooding, providing a valuable tool for environmental monitoring and sustainable land management in semi-arid regions.