Intercomparison of land cover products for Arctic tundra environments in northern Sweden : A geospatial analysis

Sammanfattning

Arctic tundra environments exhibit particularly fragmented landscapes with high spatial and temporal variability, requiring both high spatial resolution and accurate classification to capture fine-scale land cover heterogeneity. Modern land cover products often differ considerably due to variations in source data and methodological approaches. This study used geospatial analysis in QGIS to assess and compare the classification accuracy of three land cover products with different spatial scopes and intended scales – WorldCover 2021 v200, CCI Permafrost v2.0 and NMD 2018 v1.1 – within an Arctic tundra study area near Abisko, northern Sweden. Validation points were extracted from each dataset, aggregated, and manually interpreted using a study-specific, expert-generated classification scheme. Producer’s, user’s, and overall accuracies were subsequently calculated using confusion matrix–based accuracy assessment methods. The results showed that all datasets consistently overrepresented low-growing vegetation classes, while underrepresenting wetland, barren, forest and built-up classes. NMD achieved the highest overall accuracy at 0.64, followed by WorldCover at 0.57 and CCI Permafrost at 0.50. The national adaptation of the classification scheme and inclusion of LiDAR-derived vegetation data in NMD likely contributed to the superior separation between forest and low-growing vegetation classes. Despite being specifically designed for Arctic environments, CCI Permafrost exhibited the greatest confusion between classes, particularly between forest and tundra. This can likely be attributed to structural and spectral similarities between vegetation classes, temporal environmental variability, and broad class definitions between shrub tundra and forest.