A 30 m annual cropland dataset of China from 1986 to 2021 was developed using a framework integrating time-series Landsat satellite imagery, automated training sample generation, and machine learning and change detection techniques. The dataset, named CACD, provides high-resolution cropland maps at 30 m spatial resolution, enabling detailed analysis of cropland dynamics across China. The pixel-wise F1 scores for annual maps and change maps were 0.79 ± 0.02 and 0.81, respectively, indicating high accuracy. Cross-product comparisons with other datasets highlighted the precision and robustness of CACD. From 1986 to 2021, China's total cropland area expanded by 30,300 km² (1.79%), with significant increases before 2002 and a general decline between 2002 and 2015, followed by a slight recovery. Cropland expansion was concentrated in the northwest, while the eastern, central, and southern regions experienced substantial cropland loss. Approximately 419,342 km² (17.57%) of croplands were abandoned at least once during the study period. The CACD dataset supports sustainable agricultural use and food production through its high-resolution, consistent data. The dataset is freely available at https://doi.org/10.5281/zenodo.7936885. The study introduces an integrated framework for monitoring annual cropland dynamics at 30 m spatial resolution, leveraging baseline land cover maps and the TWDTW discrimination algorithm for automated training sample generation, random forest classifiers for estimating cropland probabilities, and LandTrendr for change detection. The framework enables accurate and consistent cropland mapping, with results showing high accuracy and robustness. The dataset provides valuable insights into cropland dynamics, including spatial and temporal changes, and highlights the potential of change detection algorithms in identifying dynamic land cover changes. The study also addresses challenges in cropland mapping, such as limited training samples and the need for robust algorithms, and demonstrates the effectiveness of the proposed framework in overcoming these challenges. The results contribute to understanding cropland dynamics in China and support sustainable agricultural practices.A 30 m annual cropland dataset of China from 1986 to 2021 was developed using a framework integrating time-series Landsat satellite imagery, automated training sample generation, and machine learning and change detection techniques. The dataset, named CACD, provides high-resolution cropland maps at 30 m spatial resolution, enabling detailed analysis of cropland dynamics across China. The pixel-wise F1 scores for annual maps and change maps were 0.79 ± 0.02 and 0.81, respectively, indicating high accuracy. Cross-product comparisons with other datasets highlighted the precision and robustness of CACD. From 1986 to 2021, China's total cropland area expanded by 30,300 km² (1.79%), with significant increases before 2002 and a general decline between 2002 and 2015, followed by a slight recovery. Cropland expansion was concentrated in the northwest, while the eastern, central, and southern regions experienced substantial cropland loss. Approximately 419,342 km² (17.57%) of croplands were abandoned at least once during the study period. The CACD dataset supports sustainable agricultural use and food production through its high-resolution, consistent data. The dataset is freely available at https://doi.org/10.5281/zenodo.7936885. The study introduces an integrated framework for monitoring annual cropland dynamics at 30 m spatial resolution, leveraging baseline land cover maps and the TWDTW discrimination algorithm for automated training sample generation, random forest classifiers for estimating cropland probabilities, and LandTrendr for change detection. The framework enables accurate and consistent cropland mapping, with results showing high accuracy and robustness. The dataset provides valuable insights into cropland dynamics, including spatial and temporal changes, and highlights the potential of change detection algorithms in identifying dynamic land cover changes. The study also addresses challenges in cropland mapping, such as limited training samples and the need for robust algorithms, and demonstrates the effectiveness of the proposed framework in overcoming these challenges. The results contribute to understanding cropland dynamics in China and support sustainable agricultural practices.