Tree Density Dataset of Mountainous Regions in Northeast China

This dataset represents tree densities in mountainous regions of Northeast China, derived from field surveys of trees with a diameter at breast height (DBH) of ≥10 cm across 1926 plots. Using recursive feature elimination (RFE), six key variables influencing tree density were identified: soil silt content, soil clay content, elevation, NDVI, precipitation in the wettest month, and precipitation in the coldest quarter. A stacking ensemble learning algorithm, combining extreme random trees (ERT), support vector regression (SVR), CatBoost, and a ridge regression metamodel, was used for tree density estimation. The algorithm significantly improved model performance, with an average R² increase of 43.69% and reductions in RMSE and MAE by 11.16% and 10.14%, respectively. The dataset includes a 30 m spatial resolution map of tree densities, estimating approximately 27.497 billion trees in the region. This dataset provides valuable insights for forest carbon sequestration modeling, targeted forest conservation strategies, and carbon management practices.本数据集收录中国东北山区的林木密度数据，该数据源自对1926块样地内胸径（diameter at breast height, DBH）≥10厘米的林木开展的野外调查。本研究通过递归特征消除（recursive feature elimination, RFE）筛选出6个影响林木密度的关键变量，分别为土壤粉粒含量、土壤粘粒含量、海拔、归一化植被指数（Normalized Difference Vegetation Index, NDVI）、最湿月降水量以及最寒冷季度降水量。本研究采用堆叠集成学习算法开展林木密度估算，该算法融合了极端随机树（extreme random trees, ERT）、支持向量回归（support vector regression, SVR）、CatBoost以及岭回归元模型。该算法显著提升了模型性能：决定系数（coefficient of determination, R²）平均提升43.69%，均方根误差（root mean square error, RMSE）与平均绝对误差（mean absolute error, MAE）分别降低11.16%与10.14%。本数据集包含空间分辨率为30米的林木密度空间分布图，经估算该区域内林木总株数约为274.97亿株。本数据集可为森林碳汇建模、精准森林保护策略制定以及碳管理实践提供重要参考依据。