Modeling longleaf pine (Pinus palustris Mill) wood properties using near infrared spectroscopy

This research demonstrated model development for important wood properties using near infrared spectroscopy (NIR); it considered the effect of outside sources of error, and the ability of NIR to measure fiber morphology. Strength, stiffness, and density were successfully modeled from wood samples taken throughout 10 longleaf (Pinus palustris Mill) trees. Principal components and multiple linear regression were compared for performance in prediction of density, strength, and stiffness. I found both modeling techniques to yield similar prediction accuracies. However, I found that density could be estimated through Beer-Lambert’s law since the absorbance at all wavelengths increased with density. Also, 5 of 6 wavelengths needed to predict strength were also needed to predict stiffness lending support that similar chemical morphology controls the covariance between strength and stiffness. Klason lignin, extractives, and microfiber angle (MFA) were also measured throughout the tree. I found extractives, lignin, and MFA to decrease from the pith outward regardless of height. A theoretical model was built attempting to explain how lignin content and MFA co-vary. Theoretical and empirical spectroscopic models both predicted MFA with nearly similar root mean square error and supported that lignin was a probable factor responsible for the covariance in spectra with MFA. Tracheid length was another secondary trait investigated. I demonstrated that tracheid length could be predicted with an R2 of 0.71 due to NIR spectra response with age. Accurate tracheid length prediction was possible due to systematic variation of chemistry with age except for at ring 1 and 4 where some other unknown factor was responsible. Finally, blue stain and machine variability were investigated as two sources of extraneous error. It was of interest to know if the common extraneous error would bias a prediction equation. With proper modeling, I found I could avoid the blue stain variation present in the spectra when modeling modulus of elasticity (MOE), modulus of rupture (MOR), density, lignin, and extractives. However, when a calibration was built from one machine and then applied to a population of scans made from a separate machine, blue stain became problematic and prediction of MOE, MOR, and extractives were biased.

Name thesis:degree_name

Doctor of Philosophy (PhD)

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Environmental Sciences

Grantor

Renewable Natural Resources

Year dc:date.available

2004

Author dc:creator

• Via, Brian Kipling

• MOE
• chemistry
• forestry
• wood
• MOR
• spectroscopy
• NIR

Statement dc:rights

• unrestricted
• Release the entire work immediately for access worldwide.

Identifier

etd-11082004-155505
https://repository.lsu.edu/gradschool_dissertations/524

OAI identifier oai:identifier

oai:repository.lsu.edu:gradschool_dissertations-1523