Tree species diversity mapping using UAS-based digital aerial photogrammetry point clouds and multispectral imageries in a subtropical forest invaded by moso bamboo (Phyllostachys edulis)

Abstract

Moso bamboo (Phyllostachys edulis) tends to invade any surrounding forest areas due to its aggressive characteristics (fast growth and clonal reproduction), where it changes the species composition and canopy structure of the forests, and has negative effects on forest diversity and ecosystem functions. Unmanned aerial system (UAS)-based remote sensing has the capacity to provide high-resolution, continuous spatial data that can be used to detect forest invasion dynamics. In this study, UAS-based RGB and multispectral image data and digital aerial photogrammetric point cloud (PPC) were acquired and used to detect areas of bamboo invasion in a subtropical forest of Southern China. First, a point cloud segmentation (PCS) method was applied for individual tree detection (ITD) using photogrammetric point clouds (PPCs). A random forest (RF) classifier was used to perform tree species classification based on PPC metrics, vegetation indices, and texture metrics. Finally, based on the results of the ITD and tree species classification, alpha-diversity (i.e., the species richness (S), Shannon-Wiener (H’), Simpson (D), and Pielou’s evenness index(J)) and the spatial variation in species composition along the altitude gradient (beta-diversity) in the invaded forests were assessed. Results demonstrated that PCS worked well for tree detection in invaded forests (F1-score = 80.63%), and the overall accuracy of tree species classification was 75.69%, with a kappa accuracy of 73.76%. The forest diversity analysis showed that all alpha-diversity values were generally predicted well (R2 = 0.84–0.91, RMSE = 0.05–0.84). The diversity showed a decreasing tendency with increasing bamboo invasion, and the predominantly broad-leaved invaded forests had higher diversity than the predominantly coniferous invaded forests. The human intervention had a significant impact on bamboo invasion. The ANOVA of the dispersion of the dissimilarities along the elevation gradient showed significant differences in abundance-weighted similarity among the altitude classes (ANOVA of the Bray-Curtis dissimilarity, F4,40 = 6.453, P = 0.0004***; ANOVA of the Jaccard dissimilarity, F4,40 = 5.20, P = 0.0017**). This study indicated the potential benefits of using UAS- based remote sensing data to identify tree species and predict forest diversity in bamboo-invaded forests. Our results suggested that tree species diversity can be directly estimated using individual tree detection results based on PPC data instead of modelling the relationship between field-measured indices and remote sensing data-derived metrics, and revealed the influence of human intervention on bamboo invasion.