Integrating morphometric and spectral evidence for species discrimination in Amazonian trees: a case study of Carapa (Meliaceae)

Semírian Campos Amoêdo , Caroline da Cruz Vasconcelos , Isolde Dorothea Kossmann Ferraz

Reliable species-level identification of Amazonian trees from herbarium and field material is essential for conservation, management, and economic use of non-timber forest products, yet overlapping morphology and inconsistent identifications often hinder it. Using three Amazonian species of Carapa (C. guianensis, C. surinamensis, and C. vasquezii) as a case study, we tested whether leaf morphometry, near-infrared (NIR) reflectance spectroscopy, and their combination can discriminate species from vegetative material and whether sample imbalance explains the poor recognition of the rarest species. We measured 12 vegetative morphological characters and obtained NIR spectra (350–2500 nm) from herbarium specimens and field-collected leaves. Species discrimination was evaluated using Linear Discriminant Analysis (LDA), Partial Least Squares Discriminant Analysis (PLS-DA), and Random Forest (RF), considering both the original imbalanced dataset and a class-balanced dataset. Morphometric data provided the highest classification performance (balanced accuracy ≈ 0.68), outperforming models based on NIR spectra alone or on the combined morphometric and spectral datasets. Overall, LDA and PLS-DA performed better than Random Forest. Class imbalance suppressed recall of C. surinamensis (as low as ~0.01–0.22), which improved markedly (0.41–0.55) after correction, at the cost of C. guianensis recall, showing that imbalance explains much, but not all, of the difficulties in recognizing this species. Residual overlap, mainly between C. surinamensis and C. vasquezii, is discussed as morphological and spectral similarity rather than evidence for or against taxonomic boundaries.