Fertilizers are commonly applied to improve the productivity and quality of bamboo. However, the nutrient responses of bamboo components after regular fertilization are not fully understood. This study was carried out to determine the effects of regular fertilization on the nutrient distribution of biomass components (i.e., culms, branches, leaves, roots, rhizomes) in a Moso bamboo stand in southern Korea. The study site was fertilized regularly for approximately 30 years to produce edible bamboo shoots. A total of 20 bamboo plants (10 fertilized and 10 unfertilized) were cut to measure the nutrient (C, N, P, K, Ca, Mg) concentrations of each bamboo component. Belowground roots and rhizomes were sampled at a 30-cm soil depth. The N, P, and K concentrations and stocks of aboveground biomass components were increased by regular fertilization, whereas the C, Ca, and Mg stocks were attributed to culm densities. The nutrient stocks of belowground roots were significantly lower in the fertilized plots than those in the unfertilized plots, except for the P stocks. The results indicate that regular fertilization could be a key factor to maintaining bamboo shoot productivity because of the increased responses of the nutrient concentration and stocks of bamboo components.

Accurate estimation of tree height from diameter at breast height (DBH) is essential for forest inventory, biomass estimation, and management applications, yet conventional height–diameter (H–D) models often show substantial bias when applied across structurally heterogeneous stands. This study evaluates whether incorporating forest structural information can improve H–D model performance and reduce unexplained variation beyond size-based approaches. Using large-scale forest inventory data from South Korea (197,685 trees across 14,357 plots and 216 species), together with independent datasets from the United States, we applied a stepwise modelling framework combining generalized additive models and machine-learning-based residual analysis. DBH provided a strong baseline predictor of tree height but did not fully capture observed variation. Including forest structural variables, primarily clear bole height, substantially improved model performance (RMSE reduced from ~2.94 m to ~1.98 m; R² increased from ~0.486 to ~0.770) and reduced systematic deviations across stand conditions. Variance partitioning showed that structural variables explained a larger independent fraction of variation than DBH, while environmental variables (including altitude, slope, distance to road, temperature, and precipitation) contributed smaller but spatially structured effects. Machine-learning models improved predictions for size-only models; however, these gains declined markedly when structural predictors were included, indicating that much of the improvement arises from representing underlying stand conditions rather than model complexity. These results demonstrate that incorporating forest structural information provides a practical and effective approach for improving H–D models, reducing prediction bias, and enhancing model transferability across diverse forest conditions. The proposed framework can be readily applied in forest inventory and modelling systems to improve the accuracy and reliability of tree height estimation.