COTQ 10-m Product Matches ESA WorldCover on Structural Metrics Across Eight Quebec Tiles

The evaluation harmonized legends across COTQ, ESA WorldCover, ESRI LandCover, and Google DynamicWorld then applied object-size distributions, shape complexity, Adjusted Rand Index, Intersection over Union, and Sentinel-2 reflectance separability. Photo-interpretation targeted disagreement zones. Tiles covered temperate forest to tundra gradients. COTQ displayed object statistics and spectral signatures nearest to ESA WorldCover while diverging on urban, wetland, and cryptogamic classes due to differing thematic priorities.

ESA WorldCover documentation (2021) and Dynamic World v1 release notes confirm class boundary differences that explain the observed offsets. COTQ's annual update cadence for soil artificialisation tracking provides operational continuity absent in the global products. Structural similarity to ESA reduces re-training costs for downstream models already tuned to WorldCover outputs.

For Quebec land-management agencies the results indicate COTQ can replace global layers inside provincial reporting pipelines without major retraining. Remaining class-definition mismatches require targeted post-processing masks for wetlands and rocky surfaces. Full-province rollout now depends on extending the eight-tile sample to all 48 Sentinel-2 tiles covering Quebec.

No autonomous-vehicle mapping implications are stated in the paper; land-cover consistency metrics remain relevant only for environmental monitoring workflows.