A Novel Entropy-Based Method to Quantify Forest Canopy Structural Complexity from Multiplatform Lidar Point Clouds
Xiaoqiang Liu,Qin Ma,Xiaoyong Wu,Tianyu Hu,Zhonghua Liu,Lingli Liu,Qinghua Guo,Yanjun Su
DOI: https://doi.org/10.1016/j.rse.2022.113280
IF: 13.5
2022-01-01
Remote Sensing of Environment
Abstract:Forest canopy structural complexity (CSC) describes the three-dimensional (3D) arrangement of canopy ele-ments, and has become an emergent forest attribute mediating forest ecosystem functioning along with species diversity. Light detection and ranging (lidar), especially the emerging near-surface lidar platforms (e.g., terrestrial laser scanning/TLS, backpack laser scanning/BLS, unmanned aerial vehicle laser scanning/ULS), can depict 3D canopy information with high efficiency and accuracy, providing an ideal data source for forest CSC quantification. However, current existing lidar-based CSC quantification indices may share common limitations of getting saturated in structurally complex forest stands and not fully capturing within-canopy structural var-iations. In this study, we introduced the concept of entropy into forest CSC quantification, and proposed a new forest CSC index, namely canopy entropy (CE). Two major bottlenecks were addressed in the CE calculation procedure, including (1) using a Mann-Kendall (MK) test-based resampling strategy to address the issue of incongruent sampling chances of canopy elements at different locations from different lidar systems, and (2) using a kernel density estimation (KDE)-based method to reduce its dependence on point density. The effec-tiveness and generality of CE were evaluated by simulating TLS and ULS point clouds from nine forest stands and collecting TLS, BLS, and ULS point clouds from 110 field plots distributed in five forest sites, covering a large variety of forest types and forest CSC conditions. The results showed that CE was an effective forest CSC quantification index that successfully captured CSC variations caused by both tree density and the number of vertical canopy layers. It had significant positive correlations with four widely used CSC indices (i.e., canopy cover, foliage height diversity, canopy top rugosity, and fractal dimension; R2: 0.32 to 0.67), but outperformed them by overcoming their common limitations. CE estimates from multiplatform lidar point clouds agreed well with each other (R2 >= 0.70, RMSE <= 0.10), indicating it has generality in cross-platform forest CSC quantification practices. We believe the proposed CE index has great potential to help us unravel the correlations among forest CSC, species diversity, and forest ecosystem functions, and therefore improve our understanding on forest ecosystem processes.