Crown width prediction for Larix olgensis plantations in Northeast China based on nonlinear mixed-effects model and quantile regression

Key message New allometric crown width models were developed for Larix olgensis based on a large dataset from plantations in northeastern China using the nonlinear mixed-effects model and quantile regression, and multiple variables were included in the developed models and calibration was performed to enhance their applicability. Crown width (CW) is an essential indicator of the general health, vigor, and stability of living trees. It is used as a predictor in various tree models, such as growth, biomass, mortality, stem taper, and volume models. In this study, models of tree crown width were developed and evaluated using data from 343 permanent sample plots (PSPs) of Larix olgensis plantations in Heilongjiang Province, Northeast China. A logistic function with several predictor variables, including diameter at breast height (DBH), total tree height (H), height to live crown base (HCB), and height-diameter ratio (HD), was selected as the basic crown width model to provide acceptable model generality. Four modeling approaches were evaluated: (1) a mixed-effects model, (2) a three-quantile regression method, (3) a five-quantile regression method, and (4) a nine-quantile regression method. The mixed-effects and quantile regression models were calibrated using simple random sampling (SRS) and different sampling numbers (1 to 12 trees per plot). The evaluation results of the jackknifing technique indicated that both the mixed-effects and quantile regression approaches outperformed the generalized model. The prediction performance of the models improved as the sampling number increased, but the gains in performance gradually decreased. In general, the use of six sample trees per plot was considered a good compromise between the investigation cost and predictive accuracy for calibrating the mixed-effects model and quantile regression methods.