To maximize utilization of our forest resources, detailed knowledge of wood property variation and the impacts this has on end-product performance is required at multiple scales (within and among trees, regionally). As many wood properties are difficult and time-consuming to measure our knowledge regarding their variation is often inadequate as is our understanding of their responses to genetic and silvicultural manipulation. The emergence of many non-destructive evaluation (NDE) methodologies offers the potential to greatly enhance our understanding of the forest resource; however, it is critical to recognize that any technique has its limitations and it is important to select the appropriate technique for a given application. In this review, we will discuss the following technologies for assessing wood properties both in the field: acoustics, Pilodyn, Resistograph and Rigidimeter and the lab: computer tomography (CT) scanning, DiscBot, near infrared (NIR) spectroscopy, radial sample acoustics and SilviScan. We will discuss these techniques, explore their utilization, and list applications that best suit each methodology. As an end goal, NDE technologies will help researchers worldwide characterize wood properties, develop accurate models for prediction, and utilize field equipment that can validate the predictions. The continued advancement of NDE technologies will also allow researchers to better understand the impact on wood properties on product performance.

With conifer plantations having an increasingly important role in meeting the fiber needs of society, an understanding of the effect of silvicultural practices on wood quality is critical. The perception of wood quality varies, making it hard to define in a single statement; however, possibly the most succinct definition is “a measure of the aptness of wood for a given use”. In general, properties that have a positive influence on a specific product assist in defining changes in wood quality. Since wood properties exhibit large variability within annual rings, within trees, and among trees in a stand, and have both genetic and environmental components (i.e., vary with different physiographical regions), it is imperative to have an understanding of wood properties at multiple levels. In this paper, we review the typical variation patterns in wood properties of conifers, with specific emphasis on loblolly pine (Pinus taeda L.), and radiata pine (Pinus radiata D.Don), two of the most common conifer plantation species globally. We also describe the impact of conventional silvicultural treatments on wood quality. Modeling efforts to predict variation in wood properties within trees, and in response to silvicultural treatments are also summarized.

Loblolly pine (Pinus taeda L.) is increasingly grown on intensively managed plantations that yield high growth rates. Wood properties, including specific gravity (SG), change with cambial age, and thus intensively managed trees contain a high proportion of low density corewood when harvested because of reduced rotation lengths. This study was undertaken to develop models of ring-level properties (SG and width) in intensively managed loblolly pine plantations. Ninety-three trees from five stands aged from 24 to 33 years were harvested, and 490 disks were obtained from in between the 5.2-m logs that were cut, and at the merchantable top. The disks were cut into pith-to-bark radial strips that were scanned on an X-ray densitometer, and the resultant data analyzed using non-linear mixed-effects models. The fixed effects of the models, which included cambial age and for some models disk height and ring width, were able to explain 56, 46, 54, 16, and 46 percent of the within-tree variation for ring SG, ring width, latewood SG, earlywood SG, and latewood percent, respectively. To assess implications for wood utilization, a modeled tree was built by using height, diameter, and taper equations and these models were linked with the developed ring SG model to produce a tree properties map. The linked information was also used to generate tree and log SG and proportion of corewood values for different rotation ages. The results from this study are a step towards integrating wood quality models into growth-and-yield modeling systems that are important for loblolly pine plantation management.

Near-infrared (NIR) spectroscopy is widely used as a nondestructive evaluation (NDE) tool for predicting wood properties. When deploying NIR models, one faces challenges in ensuring representative training data, which large datasets can mitigate but often at a significant cost. Machine learning and deep learning NIR models are at an even greater disadvantage because they typically require higher sample sizes for training. In this study, NIR spectra were collected to predict the modulus of elasticity (MOE) of southern pine lumber (training set = 573 samples, testing set = 145 samples). To account for the limited size of the training data, this study employed a generative adversarial network (GAN) to generate synthetic NIR spectra. The training dataset was fed into a GAN to generate 313, 573, and 1000 synthetic spectra. The original and enhanced datasets were used to train artificial neural networks (ANNs), convolutional neural networks (CNNs), and light gradient boosting machines (LGBMs) for MOE prediction. Overall, results showed that data augmentation using GAN improved the coefficient of determination (R²) by up to 7.02% and reduced the error of predictions by up to 4.29%. ANNs and CNNs benefited more from synthetic spectra than LGBMs, which only yielded slight improvement. All models showed optimal performance when 313 synthetic spectra were added to the original training data; further additions did not improve model performance because the quality of the datapoints generated by GAN beyond a certain threshold is poor, and one of the main reasons for this can be the size of the initial training data fed into the GAN. LGBMs showed superior performances than ANNs and CNNs on both the original and enhanced training datasets, which highlights the significance of selecting an appropriate machine learning or deep learning model for NIR spectral-data analysis. The results highlighted the positive impact of GAN on the predictive performance of models utilizing NIR spectroscopy as an NDE technique and monitoring tool for wood mechanical-property evaluation. Further studies should investigate the impact of the initial size of training data, the optimal number of generated synthetic spectra, and machine learning or deep learning models that could benefit more from data augmentation using GANs.

Loblolly pine is a major fibre source for the pulp and paper industry. Here we developed the first nonlinear models to predict the within-tree and regional variation of tracheid length and width for planted loblolly pine. Data were obtained from macerated tracheids and near-infrared spectroscopy calibration models from trees sampled in 109 stands across the southeastern United States. The fixed effects for the final tracheid length model, which included cambial age, height of disk within tree, and physiographic region, explained 71 percent of the variation with root mean square error (RMSE) of 0.28 mm, while the fixed effects for the final tracheid width model explained 57 percent of the variation with RMSE of 1.4 μm. There was significant variation in tracheid properties across the growing regions. Tree maps showing within-tree variability in tracheid properties were produced. Five simulated scenarios were compared using the models developed, with mean tracheid dimensions calculated on a whole-tree basis at a first and second thinnings, and at final harvest. Also from the final harvest, the tops of trees, and outerwood chips produced during lumber manufacturing were also simulated. For the whole tree scenarios, both mean tracheid length and width increased with age, increasing from 2.24 mm and 40.5 μm (age 12), to 2.51 mm and 41.3 μm (age 18), and to 2.73 and 41.8 μm at age 25, respectively. The tops of the trees at age 25 had a mean tracheid length of 2.46 mm and a mean width of 41.0 μm, while the chips had a mean tracheid length of 3.13 mm and a mean width of 42.5 μm. Due to the models representing samples collected from across the southeastern United States, and their relatively high precision, they are suitable for incorporation into growth and yield systems allowing for prediction of tracheid properties.