Enhanced supervised classification of seasonal pastures on the Qinghai-Tibet Plateau (1990–2020) using Landsat optimal time window

Understanding the distribution and changes of seasonal pastures is critical for guiding livestock production and promoting the sustainable management of grassland resources. However, long-term, high-resolution datasets on seasonal pasture distribution remain scarce, which has significantly constrained related research and management practices. In this study, we developed an automated seasonal pasture classification framework based on the Google Earth Engine (GEE) platform. The framework leverages Landsat imagery within the optimal time window (Days of Year, DOYs 190–280), incorporates algorithms for automated sample generation and refinement, and employs the Random Forest (RF) classifier to enable fully automated seasonal pasture mapping. Using this approach, we produced 30 m resolution seasonal pasture maps for the Qinghai-Tibetan Plateau from 1990 to 2020. Validation with independent samples demonstrates that the overall classification accuracy for all periods exceeded 93.10%, and the spatial details of the results outperformed previous studies. Further spatiotemporal analysis revealed that between 1990 and 2020, the area of warm-season pastures declined by 14.11%, while cold-season pastures expanded by 102.97%. This research fills a critical knowledge gap regarding the spatiotemporal dynamics of seasonal pastures on the Qinghai-Tibetan Plateau and reveals the trends and patterns of their changes. The high-resolution and long-term dataset generated provides essential information to support the scientific management and decision-making for grassland resources in the Qinghai-Tibet Plateau.