Scaling up tree diversity inventories across Amazonian ecosystems using field spectroscopy

Species identification in Amazonian forest inventories is challenging due to a shortage of taxonomists, high biodiversity, and morphological similarities leading to taxonomic errors. Near infrared spectroscopy (NIRS) is a promising tool for improving species identification efficiency and reliability. This study assessed the effectiveness of NIRS in discriminating against 26 abundant tree species across three Amazonian ecosystems: upland forest, white-sand ecosystems, and floodplain forest, using spectral data from different tree tissues (outer bark, inner bark, and fresh leaves). Each tissue was tested using Linear Discriminant Analysis (LDA) spectral models with two cross-validation methods: leave-one-out and 70/30 holdout. Results showed high discrimination accuracy for all tissues and ecosystems. The general models achieved 86% accuracy for outer bark, 97% for inner bark and 98% for fresh leaves. The most informative spectral bands varied by tissue type: SWIR I (1300 to 1900 nm) for outer bark, and VIS (400 to 700 nm) + SWIR I (1300 to 1900 nm) for inner bark and fresh leaves. A general model integrating species across ecosystems confirmed NIRS as an effective tool for in-field tree identification. These findings highlight the potential of VIS-NIR spectroscopy to Amazonian biodiversity inventories, contributing to more accurate species identification, refining forest management and conservation efforts.