LINKING KNOWLEDGE ABOUT GROWTH AND WOOD PROPERTIES IN RADIATA PINE - PAST, PRESENT AND FUTURE

Given the importance of radiata pine ( Pinus radiata D.Don) to the New Zealand forestry sector, there has been a considerable focus on understanding growth and wood properties in this species. However, for many years research focused on understanding and enhancing tree growth was not always linked with research into wood quality. Much of the initial work investigating radiata pine wood quality focused on quantifying the extent of variation in key properties of interest, principally density, within and among trees (Cown and McConchie, 1983, Harris, 1965, Palmer et al., 2013) and how these were affected by forest management (Carson et al., 2014, Cown, 1973, Cown, 1974, Cown and McConchie, 1981, Cown and McConchie, 1982). At the same time research into growth looked at how to improve both productivity and profitability. This has resulted in a reduction in rotation length and also silvicultural regimes that were often characterized by heavy and early thinning (James, 1990). The focus of much of the early tree improvement research was on enhancing growth and improving stem form, rather than on improving internal wood properties (Jayawickrama and Carson, 2000). Not unsurprisingly, there has been concerns that many of these practices have had a negative impact on wood quality (Burdon, 2010, Cown, 1992, Moore and Cown, 2017). In order to better understand the implications of current and future research aimed at enhancing productivity on wood quality, better knowledge about the connection between tree growth and wood properties is required. In this presentation, we describe different modelling and experimental approaches that have aided our understanding of this connection. Empirical models were developed to predict the radial and longitudinal variation in wood density, microfibril angle and spiral grain angle (Kimberley et al., 2015, Moore et al., 2014, Moore et al., 2015) within a tree. These non-linear mixed effects models are based on ring-level data and account for the hierarchical structure in these data, which arises from the fact that they were obtained from radial samples taken at different heights in selected trees from a number of stands. The models enable the effects of ring number, ring width and height in the tree to be accounted for which enables them to be directly linked to growth models. In the case of wood density, genetic improvement can also be accounted for (Kimberley et al., 2016). This modelling approach is suitable for wood properties that are continuous and approximately normally distributed. However, traits such as resin pockets and intra-ring checks, which are major sources of value loss in appearance products (Cown et al., 2011, Cown, 2013), are discrete and are better represented as counts. Previous efforts to try to describe the variation in these traits and the factors associated with their occurrence have generally not yielded many useful results. To identify factors associated with these traits, we have applied machine learning approaches, specifically Random Forests and Gradient Boosting Trees, to historical data on resin pockets and intra-ring checks. Finally, we explore the potential to further link tree growth, form and wood properties through the analysis of high-density airborne LiDAR data acquired from a UAV, terrestrial LiDAR data and wood properties data that were obtained from a large (~10 ha) field experiment that contained a range of different seedlots and stand densities. Ring-level models have enabled the general intra-stem patterns in wood density, microfibril angle and spiral grain angle to be quantified. A large proportion of the radial variation in density and microfibril angle can be accounted for by ring number from the pith and ring width. This enables these models to be linked to growth models and used to predict the effects of factors such as age and stand density on these two wood properties. At the time of writing, we have only just commenced the analysis of data on resin pocket and intra-ring checking occurrence using machine learning approaches, so do not have any results yet which identify any site, stand or genetic factors associated with the incidence of these two traits. Likewise, we have only commenced analyzing the LiDAR dataset that has been collected. We are currently calculating a range of stem and crown metrics for individual trees within this trial and will determine whether any of these are associated with wood density and stiffness measurements made on individual trees. The simple empirical models that have been developed for radiata pine have enabled a better understanding of the effects of some growth-related factors on selected wood properties. However, such models do have a number of limitations. Firstly, they focus on associations rather than explicitly attempting to model the underlying growth and wood formation processes. Secondly, they still predict idealized patterns of intra-stem variation in wood properties and ignore the departures from this pattern due to factors such as the presence of reaction wood. More realistic representations of these patterns is needed in order to better understand the impacts on end product performance. It is hoped that being better able to characterize the growing environment of a tree and its crown structure will lead to an improved understanding of the connection between tree growth and wood properties.