Using machine learning to synthesize spatiotemporal data for modelling DBH-height and DBH-height-age relationships in boreal forests

Abstract

Sustainable forest management requires the ability to accurately model forest dynamics under a changing environment, which is difficult using conventional statistical methods as many factors that interactively affect forest growth must be considered. As well, statistical model development is often limited by the lack of broad-scale repeated forest measurements needed to capture changes in 1 or more variables and the corresponding changes in forest dynamics (e.g., growth in diameter and height), while assuming other variables do not change, or their changes do not significantly affect the forest dynamics of interest. In many forested countries, comprehensive monitoring programs have amassed large amounts of diverse forest measurement data. Here we propose a new approach for using artificial neural network-based machine learning to synthesize spatiotemporal tree measurement data collected over a vast area of boreal forest in central Canada to model diameter at breast height (DBH)-height and DBH-height-age relationships for 6 dominant tree species. More than 30 potentially important stand structure, site, and climate variables were considered. We used an individual-based modelling approach by considering each individual tree measurement as an instance of the complex relationships modelled; together, broad-scale long-term monitoring data contain many such instances, representing considerable spatial and temporal scale variation in forest growth and growing conditions. Using this approach, we significantly improved DBH-height and DBH-height-age models. And the models developed allowed us to analyze the effects of environmental conditions or changes in these conditions on forest growth. This may be the first attempt at applying this type of approach, which can be used to more accurately model, for example, forest growth, mortality, and how they are affected by changing climate in a variety of forest types.