Amazonian species evaluation using leaf-based spectroscopy data and dimensionality reduction approaches

Abstract

Sampling trees in natural environment can be used in studies ranging from floristic composition and phytogeography to management and growth modelling, and accurate inventories are based on highly labor-intensive methods. Relying on hyperspectral approach, this study aimed to differentiate spectral libraries of four Amazon tree species. We first prepared the spectroradiometer data on representative bands on foliar biochemistry, followed by reflectance inflection difference and finally, we applied vegetation indices. Next, the discriminant analysis was reasoned on multivariate approach, were successfully discriminated the spectral curves related to each of evaluated tree species. By visual analysis, some regions of the electromagnetic spectrum with higher differentiation in reflectance responses can be seen, in portions of the visible spectrum (0.5–0.65 μm), near-infrared (0.913–1.25 μm) and short-wave infrared 2 (2.1–2.5 μm). There was a higher contribution in distinguishing between species based on specific RID (Reflectance Inflection Difference) heights, such as seen on specific representative bands, where RID approach reached 99.87% of data variability related to principal component 1 (PC1) and 99.72% for leaf structure-based bands in PC1. Principal component analysis applied to the vegetation indices brought satisfactory results, with PC1 highly related to the variability of the vegetation indices results (99.37%). Adopting this approach in hyperspectral data at the leaf level and well-defined classes results in good responses. We emphasize the importance of using combined vegetation indices, with greater contributions by indices developed for quantization or absorption of electromagnetic radiation by chlorophyll, which are based in the visible region. These results can improve further research by using remote sensing techniques, as create brand-new data for Amazonian tree species policymaking, conservation and research.