Evaluation of portable NIR instruments for discrimination of 6 similar tropical wood species by PLS-DA

Supervising wood exploitation can be very challenging since there are many similar species and not enough experts able to identify them by wood anatomy analyses to meet the demand. There is evidence that valuable endangered wood species are being smuggled disguised as other species, and because of this identification techniques must be available to verify the identity of wood being cut and commercialized. Near infrared spectroscopy (NIRS) allied to multivariate calibration has been successfully used to discriminate between tropical wood species using high resolution instruments. Portable instruments provide an alternative to identify wood in situ and in real time. In this study the performance of a handheld NIRS spectrometer was evaluated in the discrimination of six similar tropical wood species: Mahogany (Swietenia macrophylla) and Cedar (Cedrela odorata), both high value tropical timber species which were included in Appendixes II and III of the Convention on International Trade of Endangered Species (CITES), respectively; Crabwood (Carapa guianensis); Cedrinho (Erisma uncinatum); Curupixá (Micropholis melinoniana); and Eucalyptus (Eucalyptus grandis), using partial least squares for discriminant analysis (PLS-DA). The measurements for model development and validation take into account spectra taken in the laboratory as well as in the field. Laboratory measurements were taken in triplicate from 103, 214, 72, 109, 87, 63 block samples of Mahogany, Cedar, Crabwood, Cedrinho, Curupixá and Eucalyptus, respectively. Samples had approximately 2 cm3 and were measured in all wood growth directions. Field measurements were taken in sawmills and log yards in the State of Acre from 25 Cedar samples including green and dry wood boards and trunks and 25 Mahogany samples including dry boards and trunks with 25 replicates of each sample. Spectra were acquired using microPhazir Analyzer (Thermo) with a resolution around 8 nm, 5 scans, spectral region of 1595-2396 nm, and in absorbance scale (log(1/R)). Half of all field samples and 2/3 of all laboratory samples were used to build the PLS-DA models and the remaining samples were used for validation. Since wood is such a heterogeneous material, replicates were not averaged. In order to minimize the spectra variation not related to species differences, the region of 1610-2290nm was selected. Data was preprocessed by 1st derivative and mean centered. Only samples with considerably high residues in the dependent variable were considered outliers and excluded from the calibration set, so that the model could cover as much spectral and sample variation as possible. The correct classification rates for the validation dataset were 98.9% for Mahogany, 99.5% for Cedar, 97.7% for Crabwood, 99.6% for Cedrinho, 99.3% for Curupixá, and 99.7% for Eucalyptus. Therefore, the portable NIR spectrometer combined with PLS-DA succeeded in discriminating between the six species with high correct classification rates, even with samples measured in the field and under different conditions. Models can be uploaded to the instrument and used to predict new samples in real time. These results stimulate the application of portable spectrometers in the supervision of wood exploitation, which can increase its efficiency and viability, contributing to species preservation.