Discrimination of caatinga species based on bark using near infrared spectroscopy

Background: The Caatinga biome has high deforestation rate, so the correct identification of species is important to conserve resources. The objective of this study was to evaluate the potential of NIR spectroscopy to distinguish tree barks from eight species from the Caatinga biome based on the development of multivariate models. Three trees of each species were felled, and the trunk was cut at six positions to obtain bark sample discs: 0%, diameter at breast height (DBH) (1.30 m from ground), 25%, 50%, 75% and 100% of commercial height. Spectra were collected with resolution of 4 cm-1 and wavenumber ranging from 10 000 to 4 000 cm-1 using a probe with 2 mm aperture. All discs obtained from the six positions were approximately 5 mm from the probe, and 24 spectra were collected from each disc, for a total of 144 per tree and 432 per species. Classification methods were based on all spectra and only the DBH position, by applying linear discriminant analysis, support vector machine and k-nearest neighbors (K-NN). Results: Better results were obtained with K-NN and first derivative spectra, with accuracy of 0.91 (all tree positions) and 0.85 (only DBH). NIR spectroscopy with multivariate analysis has potential to discriminate Caatinga species based on spectra of bark samples. Conclusion: The use of near infrared in forest can confirm the correct species before cut on forest management, contributing to conservation of Caatinga resources and an adequate use of species with high aggregated value.