Assessment of Global Land Use and Land Cover datasets for national-scale applications in Pakistan

Global Land Use and Land Cover (LULC) datasets are widely used for environmental monitoring, sustainable land management, and climate resilience planning. However, their reported global accuracies do not necessarily reflect their reliability at national and regional scales, particularly in countries with heterogeneous landscapes. This study evaluated the accuracy of seven widely used global LULC datasets across Pakistan, a geographically diverse country characterized by complex topographic, climatic, and ecological conditions. Four datasets at 10-m (ESA WorldCover, Dynamic World, FROM-GLC10, and ESRI Land Cover) and three datasets at 30-m resolutions (GLAD, GLC-FCS30D, and FROM-GLC30) were used. All datasets were evaluated using a standardized workflow implemented in Google Earth Engine, including harmonization of native classes into a six-category IPCC-based classification scheme. In addition, a class-wise multi-dataset compositing approach was explored as an exploratory strategy to evaluate whether integrating class-specific strengths across datasets could reduce class ambiguities and improve thematic consistency. Accuracy was evaluated using stratified random validation samples interpreted independently with ground truth information and high-resolution Google Earth Pro imagery. The classification performance was assessed using overall accuracy, user’s accuracy, producer’s accuracy, F1-score, and kappa coefficient. Among the 10-m products, ESA WorldCover achieved the highest overall accuracy (72.88%) and showed particularly strong performance in Settlement, Wetland, and Grass classes. Among the 30-m datasets, GLAD performed best overall (66.63%) and demonstrated relatively stronger classification for Crop and Wetland classes. Taken together, this study provides a national-scale evaluation of global LULC datasets in Pakistan and offers practical guidance for selecting suitable products for environmental assessment and land management applications in data-scarce regions.