Wood density prediction using near-infrared hyperspectral imaging for early selection of Eucalyptus grandis trees

Abstract

Wood is a heterogenous material whose properties vary over time, making it difficult to predict the wood properties at a given age of trees in the future. The site and climate are also factors affecting wood heterogeneity. To improve the accuracy of early selection of trees in drier sites, it is thus important to study inter-annual variations in wood density in conditions of contrasting water availability. We tested the use of near-infrared hyperspectral imaging (NIR-HSI) to assess inter-annual wood density and predict wood density at a future age to evaluate the accuracy of early selection of Eucalyptus grandis trees for wood density and to see if a drier site influences early selection. We sampled 38 six-year-old trees growing under two different water regimes: (i) 37% throughfall reduction (–W), to simulate a dry site, and (ii) undisturbed throughfall (+ W). NIR-HSI images were used to build high-resolution wood density maps of the whole cross section. After the annual growth rings were delimited, the average wood density at each age and in growth ring was extracted to evaluate juvenile–mature correlations in the wood. The NIR-HSI images calibrated with a locally weighted partial least square regression (LWPLSR) model, using raw spectra, performed well in predicting the wood density of the whole cross section. Correlations for wood density between ages 1–3 and 5–6 were strong (r = 0.85 to 0.94), while correlations between rings 1–3 and 4–5 were moderate to strong (r = 0.51 to 0.87). In − W plots, juvenile–mature correlations were slightly lower than in + W plots. Our results suggest that early E. grandis selection for wood density is feasible to predict wood density at 6 years of age.