Canopy height dataset for 7 cities in China with 30 m and 1 km resolution

Abstract: Urban vegetation canopy height is a critical parameter, as it helps assess urban ecological structure and greening quality and plays an important role in carbon sink evaluation. Traditional manual measurement methods are inefficient and costly, while existing remote sensing approaches face several challenges, including insufficient accuracy, single-model limitations, and data discontinuity at the urban scale. To address these challenges, this dataset presents a high-resolution fused estimation product developed using multi-source remote sensing data combined with ensemble learning methods. The data product uses spaceborne LiDAR GEDI data as reference labels and integrates multi-source information, including Landsat-8 optical imagery, PALSAR and PALSAR-2 synthetic aperture radar data, SRTM topographic data, and spatial factors. The AutoGluon ensemble learning framework was employed to enable continuous estimation and mapping of vegetation canopy height from the footprint scale to the urban regional scale. The dataset covers three typical Chinese cities, Beijing, Shanghai, Chengdu, Changsha, Wuhan, Shenyang and Shenzhen, providing reliable data support for urban ecological assessment, forest carbon sink monitoring, and urban planning. Methods: The processing workflow consists of three main steps. (1) Multi-source data preprocessing and feature integration. GEDI LiDAR data are quality filtered and converted to extract vegetation canopy height metrics (rh95). Landsat 8 optical imagery, PALSAR/PALSAR 2 radar data, and SRTM topographic data are uniformly resampled, band calculated, and feature extracted to construct a feature dataset encompassing optical, radar, topographic, and spatial factors. (2) Footprint scale height estimation modeling. Multiple linear regression, random forest, support vector machine, and the AutoGluon ensemble learning method are employed to build vegetation canopy height estimation models. Ablation experiments and feature selection are conducted to optimize input variable combinations and improve model interpretability, and the AutoGluon ensemble model is applied to achieve footprint scale height estimation across seven urban areas. (3) Regional scale mapping and residual correction. Based on the optimal model, footprint scale estimation results are extrapolated to the entire urban area to generate 30 m and 1 km resolution vegetation canopy height distribution maps. The model is applied pixel by pixel to produce 30 m and 1 km resolution canopy height products, meeting application needs at different scales.File naming: Follows the convention: “[year][city name][resolution].tif”. For example, “2019_beijing_1km.tif” represents the 1 km resolution canopy height data for Beijing in 2019. All files are in GeoTIFF format and can be opened directly with GIS software.Contact information: Dr. Wenli Huang ([e-mail]). Funding: This work was supported by the National Key Research and Development Program of China (Grant No. 2022YFF1301102). Citation: The recommended citation for the Zenodo record is: Huang, W., Jiang, H., Min, W., Song, Y., Lu, Y., Ming, Z., & Shen, H. (2026). Canopy height dataset for 7cities in China with 30 m and 1 km resolution (2019-2022) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.20711795. The content of the two versions is identical; users may cite either as appropriate.