WOOD MOISTURE CONTENT DETERMINATION BY HANDHELD NEAR-INFRARED REFLECTANCE SPECTROMETER

Abstract

Rapid, accurate determination of wood moisture content is of paramount importance for the wood industry, infrastructure maintenance, studies of plant physiology, and forest management. Near-infrared reflectance spectroscopy (NIRS) is a widely used non-destructive technique for analyzing properties of materials, including moisture content. Small, portable, handheld NIR spectrometers represent an emerging technology with strong potential for rapidly, affordably estimating materials properties. Here, we used a SCiO^TM miniature handheld NIR spectrometer and a partial least squares (PLS) regression model to predict wood moisture content.  The model was developed using spectra (740-1070 nm) collected from increment borer wood samples from 41 representative softwood and hardwood trees, calibrated against gravimetric wood moisture content determined by oven-drying. The calibration and prediction datasets contained 2/3^rd and 1/3^rd of all data, respectively. We explored effects of different spectral preprocessing algorithms (i.e., first and second-order derivatives and standard normal variate transformations) on model performance. First-order derivative spectra with five latent variables yielded the most robust model (R²: 0.72, RMSE_P: 0.32, ratio of performance to deviation: 2.2). Broadly, we demonstrated that relatively low-cost miniature handheld NIR spectrometers such as the SCiO^TM can rapidly estimate percent moisture content in wood of various species.