近地面激光雷达点云密度对森林冠层结构参数提取准确性的影响

Forest canopy structure parameters extracted and derived from light detection and ranging (LiDAR) technology could be regard as a noble view and new dimension to the traditional forest ecology research. The near-surface light detection and ranging on small multiple-rotors drone is more flexible and more efficient to gain local-scale and community-scale forest plots′ high-density point cloud than airborne light detection and ranging on lager fixed wing aircraft. However, unexpected low-density sample was spotted in the relative high-density target region, which affects the accuracy of canopy structure parameters calculating. This project was based on point cloud collected by near-surface light detection and ranging from 4 large long-tern morning forest dynamic plots. Firstly, strip decomposition method was applied to analyze the reason of low-density samples in each plot. Secondly, point cloud thinning simulation algorithm were utilized to fit deviation curve. We compared deviation curve among plots, parameters and sampling scales to demonstrate the effect of low density on extracting forest canopy structure parameter accuracy. Finally, the necessary point cloud density which could guarantee adequate accuracy of parameter extraction was calculated under each condition. The results showed that: 1) the landform or (and) the ill-considered near-surface remote sensing programing and design were main reasons of occurrence of low-density sample in forests dynamic plots. It is relative hard to design and gather high-density point cloud data in rugged and complex dynamic forest plots (around 30 points/ m²), such as Gutianshan and Xishuangbanna plot. Low-density sample was often spotted in the deep valley and high elevation samples in such plots. In the flat forest region such as Changbaishan 1&2 plot, it is easier to gain high-density point cloud data (more than 150 points/ m²). However, the imperfect design result to 1 hm² low-density area in the north part of Changbaishan 1 plot. 2) As point could density dropping, the accuracy of parameters was also accelerated dropping as well, which showed a negative exponent power function. The deviation curve showed clearly difference between parameters and sample scales other than plots. 3) We could gain 95% extracting accuracy at 5—20 sample scales with a point cloud density of 16points/ m² according to the deviation curve. In conclusion, in order to better apply near-surface light detection and ranging technology in forest ecology research, we should pay more attention to the rationality of remote sensing design and programing. Researchers should fully understand the impact of terrain and control the point cloud data quality from the source. If low-density point cloud sample already existed in the target region, ecologist could coarse the sampling scale appropriately to gain a satisfied accuracy of canopy structure parameters and extracted point cloud density-sensitive canopy structure parameter at fine sample scale cautiously.