ORIGINAL Nonlinear multivariate modeling of strand density from near-infrared spectra Brian Via • Timothy McDonald • John Fulton Received: 14 December 2010 / Published online: 20 January 2012 Ó Springer-Verlag 2012 Abstract The variation of juvenile wood concentration within a southern pine feedstock can impact strand density variation during composite processing. Higher strand density variation can equate to increased variance in product performance and higher manufacturing costs. In this study, near-infrared spectroscopy coupled with linear and nonlinear methods of calibration was used to predict strand density. The best performing model was developed with a 1st derivative pretreatment and 6 factors including a quadratic term and exhibited a root mean square error of pre- diction (RMSEP) = 0.0566, R 2 = 0.84, and a ratio of performance to deviation (RPD) = 2.30. When only the radial surface was presented and a linear model was utilized, the RMSEP was lowered to 0.033 and the RPD increased to 3.93 and confirmed that a random surface orientation will decrease model precision. The Box-Behnken design was found useful in providing a competitive nonlinear cali- bration model but with a smaller sample size. Introduction Product density is the fundamental property to influence wood mechanical and physical properties. It is well known that an increase in wood density can result in an increase in product stiffness and strength. Alternatively, an increase in juvenile wood content will decrease density and can result in lower composite properties (Kennedy 1995). Any variation in feedstock density will translate into increased variation in composite density resulting in higher costs to manufacture. B. Via (&) School of Forestry and Wildlife Sciences, Auburn University, 520 Devall Drive, Auburn, AL 36849, USA e-mail: bkv0003@auburn.edu T. McDonald  J. Fulton Department of Biosystems Engineering, Auburn University, 200 Tom E. Corley Building, Auburn, AL 36849, USA 123 Wood Sci Technol (2012) 46:1073–1084 DOI 10.1007/s00226-012-0467-x