Effect of wood moisture variation on the performance of NIR discrimination models for three Dalbergia species from Madagascar

Randriambinintsoa Tiavina, Chaix Gilles, Ramananantoandro Tahiana. 2024. Effect of wood moisture variation on the performance of NIR discrimination models for three Dalbergia species from Madagascar. In : International Conference on Tropical Wood - Advancing the sustainable use of tropical forests. Book of abstracts. Ramananantoandro Tahiana (ed.). University of Antananarivo, IUFRO. Antananarivo : University of Antananarivo, Résumé, 19-20. International Conference on Ttopical Wood (ICTW), Antananarivo, Madagascar, 26 Août 2024/28 Août 2024.

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.

Auteurs et affiliations

• Randriambinintsoa Tiavina, CIRAD-BIOS-UMR AGAP (FRA)
• Chaix Gilles, CIRAD-BIOS-UMR AGAP (FRA) ORCID: 0000-0003-2015-0551
• Ramananantoandro Tahiana, ESSA (MDG)

Source : Cirad-Agritrop (https://agritrop.cirad.fr/613029/)

Métriques

[ Page générée et mise en cache le 2025-09-22 ]