Individual Tree Segmentation and Tree Height Estimation Using Leaf-Off and Leaf-On UAV-LiDAR Data in Dense Deciduous Forests

Accurate individual tree segmentation (ITS) is fundamental to forest management and the studies of ecosystem. Unmanned Aerial Vehicle Light Detection Ranging (UAV-LiDAR) shows advantages for ITS height estimation at stand landscape scale. However, dense deciduous forests with tightly interlocked crowns challenge performance ITS. Available LiDAR points through crown appropriate algorithm are expected attack problem. In this study, a new UAV-LiDAR dataset that fused leaf-off leaf-on point cloud (FULD) was introduced assess synergetic benefits by comparing different types algorithms (i.e., watershed segmentation, layer stacking segmentation) in Northeast China. Field validation conducted four typical stands, including mixed broadleaved (MBF), Mongolian oak (MOF), broadleaf-conifer (MBCF) larch plantation (LPF). The results showed combination FULD (LSS) produced highest accuracies across all (F-score: 0.70 0.85). also better on estimation, root mean square error (RMSE) 1.54 m level. Compared using solely, RMSE reduced 0.22 0.27 m, 12.3% more trees were correctly segmented FULD, which mainly contributed improved detection rate nearly DBH levels accuracy low levels. improvements attributed abundant from bole treetop as well each being included LSS algorithm. These findings provide useful insights guide application when multi-temporal data available future.