Retrieval of fine resolution land covers in high latitude region of Alaska from integrated phenological characteristics

Arctic vegetation serves as a critical climate change indicator and plays an essential role in land-atmosphere interactions. However, accurate identification of complex Arctic vegetation types at regional scales remains challenging due to the Arctic's remoteness and spatiotemporal irregularities in satellite observations caused by persistent cloud cover. Existing land cover (LC) products often oversimplify Arctic landscapes or struggle with accurate classification in transition zones where vegetation types exhibit subtle spectral differences but distinct phenological patterns. This study introduces a novel Time Series Matching method using All Cloud-free Observations (AO-TSM) that addresses the challenge of utilizing irregularly sampled observation sequences in Arctic remote sensing. The method constructs pixel-scale adaptive matching templates by extracting phenological-spectral patterns from prior knowledge, transforming these into multidimensional vectors that preserve both spectral and temporal characteristics. A dual-metric similarity assessment combining Euclidean and cosine distances enables robust classification under any observation distribution conditions. Applied to Alaska using Sentinel-2 data, AO-TSM generated a 10-m resolution land cover map (AKALC-2020) with 87.32% overall accuracy (OA), demonstrating significant improvements in distinguishing complex tundra vegetation types. Compared to existing products, AKALC-2020 achieved superior performance for erect dwarf-Shrub, grassland, and bare ground classes, with producer's accuracy (PA) and user's accuracy (UA) exceeding 85% while maintaining realistic spatial patterns across heterogeneous landscapes. This adaptive approach offers substantial advantages over traditional statistical composite methods, providing a robust foundation for Arctic ecological monitoring and climate research.