The 30m seamless annual leaf-on Landsat composites from 1985 to 2024 were generated using a comprehensive framework designed to ensure high-quality, consistent data across decades. Starting with preprocessed Level-2 surface reflectance images from multiple Landsat sensors, the dataset is restricted to the Leaf-On season, with rigorous cloud and shadow masking applied based on quality assessment bands. To maintain consistency across sensors, spectral harmonization is conducted, followed by annual composite generation using the medoid method to capture peak vegetation conditions. The resulting composites are structured into a spatially consistent data cube, facilitating efficient analysis and monitoring of vegetation dynamics over time. The band naming convention follows Landsat TM standards, with bands designated as Blue (B1), Green (B2), Red (B3), NIR (B4), SWIR1 (B5), and SWIR2 (B7). Both qualitative and quantitative evaluations were conducted to validate the data quality. Here, we provide 2023 image data covering southwestern forest regions of China as a sample for testing. For access to the full dataset, please visit Google Earth Engine at this link, and Earth Engine App (Landsat Yearly Composite Viewer) at this link. The dataset has now been updated to include data up to 2025. Data citation: Cai, Y., Li, X., Zhu, P., Nie, S., Wang, C., Liu, X., & Chen, Y. (2025). China Earth Observation Data Cube: The 30m Seamless Annual Leaf-On Landsat Composites from 1985 to 2023. Journal of Remote Sensing. DOI: 10.34133/remotesensing.0698 For data-related inquiries, please contact Dr. Yaotong Cai at caiyt33@mail2.sysu.edu.cn.