Non-destructive estimation of Pinus taeda L tracheid morphological characteristics for samples from a wide range of sites in Georgia

Tracheid coarseness, specific surface, wall thickness, perimeter, and radial and tangential diameter from 119 radial strips of Pinus taeda L. (loblolly pine) trees grown on 14 sites in three physiographic regions of Georgia (USA) were measured by SilviScan. NIR spectra were also collected in 10 mm increments from the radial longitudinal surface of each strip and split into calibration (9 sites, 729 spectra) and prediction sets (6 sites, 225 spectra). NIR spectra (untreated and mathematically treated first and second derivative and multiplicative scatter correction) were correlated with tracheid properties to develop calibrations for the estimation of these properties. Strong correlations were obtained for properties related to density, the strongest R ² being 0.80 (coarseness), 0.78 (specific surface) and 0.84 (wall thickness). When applied to the test set, good relationships were obtained for the density-related properties (R _p ² ranged from 0.68 to 0.86), but the accuracy of predictions varied depending on math treatment. The addition of a small number of cores from the prediction set (one core per new site) to the calibration set improved the accuracy of predictions and, importantly, minimized the differences obtained with the various math treatments. These results suggest that density related properties can be estimated by NIR with sufficient accuracy to be used in operational settings.