Diameter class-based model for predicting forest growth dynamics and responses to thinning regimes

초록

Forest dynamic models have traditionally used stand age as an indicator of stand development. However, stand structure is critical in determining growth trajectories that age-based frameworks may fail to capture, particularly under active management practices, such as thinning. Therefore, we developed a diameter class-based growth model that integrates machine learning to predict stand development using volume transitions among tree size classes. Using Korea's National Forest Inventory data (2006-2020), we trained models to estimate diameter at breast height (DBH) class volumes and conifer-broadleaf ratios at 5-year intervals, demonstrating strong predictive accuracy (R2 = 0.83-0.96). We applied this framework to 1752 stands aged 41-50 years with similar initial stand volumes and identified four structural clusters using the relative volume proportions of each DBH class. Despite comparable stand ages and volumes, the 100-year projections revealed divergent developmental trajectories among the structural clusters. Although the dominant size class shifted progressively toward larger classes in all clusters, the rate and timing of this transition influenced long-term productivity. Stands with prolonged contributions from small- and medium-sized classes achieved approximately 1.5 times higher volumes than those already dominated by large trees, which exhibited early growth saturation. Thinning simulations indicated that productivity responses depended on the alignment between thinning regimes and dominant growth classes. However, in stands where growth was predominantly focused on the largest classes, thinning failed to increase productivity across all thinning regimes. These findings demonstrate that structure-based predictions of forest dynamics are feasible and that the optimization of silvicultural interventions based on pretreatment stand structural characteristics is essential.

키워드