关键词:
UAS
Hyperspectral
LiDAR
Biochemical traits
Vegetation indices
Fusion
Three-dimensional distribution
Age growth
摘要:
Biochemical traits in forest vegetation are key indicators of leaf physiological processes, specifically photo-synthetic and other photochemical light pathways, and are critical to the quantification of the terrestrial carbon cycle. Advances in remote sensing sensors and platforms are allowing multi-dimensional and continuous-spatial information to be acquired in a fast and non-destructive way to quantify forest biochemical traits at multiple spatial scales. Here we demonstrate the use of high spectral resolution, hyperspectral data combined with high density three-dimensional information from Light Detection and Ranging (LiDAR) both acquired from an unmanned aerial system (UAS) platform, to quantify and assess the three-dimensional distribution of biochemical pigments on individual tree canopy surfaces. To do so, a DSM based fusion method was developed to integrate the 3D LiDAR point cloud with hyperspectral reflectance data. Regression-based models were then developed to predict a number of biochemical traits (i.e., chlorophyll (Chl) a, b, total Chl and total carotenoids (Cars) content) from a suite of common spectral indices at three vertical canopy levels, and were evaluated using a leave-one-out cross-validation approach. One-way ANOVA and Duncan's multiple comparison post hoc tests were used to investigate the vertical distribution of biochemical pigments on individual tree canopy surfaces, and in response to age and species. Our results demonstrated that a number of vegetation indices, derived from the hyperspectral data, were strongly correlated with a number of biochemical traits (Adj-R-2 = 0.85-0.91;rRMSE = 5.19-6.38%). In general, models fitted using leaf samples from the upper, middle and lower canopies separately (AdjR(2) = 0.85-0.91;rRMSE = 5.19-6.38%) had similar accuracy to the models developed with pooled data (AdjR(2) = 0.87-0.90;rRMSE = 5.21-6.11%). The differences between separate models and global models were not statistically significant (P > 0