Résumé
Since 2013, Madagascar's Dalbergia species have been listed in Appendix II of the CITES, calling for a reliable tool to identify these trees. Near Infrared (NIR) spectroscopy is often used for this, but its field use is hindered by factors like wood moisture. This study aimed to create discrimination models to differentiate three Dalbergia species despite moisture variations : D. chlorocarpa; D. orientalis; D. purpurascens.These species were chosen among the other Dalbergia species because they exhibit sufficient repetitions to calibrate discrimination models (at least 20 samples). Ninety-nine wood samples of this three species of Dalbergia from Madagascar were analyzed, with moisture levels set at 8%, 12%, 16%, and 20%. Each sample's heartwood was scanned six times. These samples were divided into 3 sets : S0 for calibration of discrimination models, S1 for correction or standard for the robustness of the model against the moisture variations, and S2 for the tests of models performances. Partial Least Squares Discriminant Analysis (PLSDA) models, initially calibrated with S0 at 12% moisture, were used to identify S2 samples at different moisture levels. This showed how moisture affects model accuracy. Three correction methods (PDS, Update, EPO) were tested using S1 set. PDS correction showed performance differences in identifying species within S2, highlighting moisture's impact on NIR spectroscopy. Models from Update and EPO corrections showed similar performances, with about 70% accuracy in classifying spectra at different moisture levels. While promising, broader wood sampling, including more species, and more varied moisture content are necessary for these models to be directly applicable in the field.