A novel framework for identifying artificial grassland planting areas using time-series multispectral imagery with enhanced accuracy

Artificial grassland is a key intervention for restoring degraded grasslands, and rapid large-scale mapping via remote sensing is essential. Current approaches remain highly dependent on field surveys and have notable limitations. To address these issues, we developed a framework that exploits multi-temporal Landsat series imagery together with the Continuous Change Detection and Classification algorithm to identify candidate locations and establishment dates of artificial grasslands. By integrating inferred planting times with annual phenological peak metrics to construct discriminative features, and applying a Random Forest classifier, we delineated planted artificial grassland areas. Ground validation yielded an overall accuracy of 89.87%, and comparisons with previous studies showed higher accuracies (>82.93%). Applying the framework to map 2021–2023 plantings in Qinghai Province produced a correlation coefficient r = 0.8843 ( p < 0.05) with statistical records. These results indicate strong generalizability and accuracy across regions and years, demonstrating the framework's reliability for advancing artificial grassland mapping.