Constructing a crown width prediction model for Pinus koraiensi using BLS data

Backpack Laser Scanning (BLS), as a portable new laser radar technology, has been rarely applied in forestry surveys. This study aims to extract the individual tree factors and crown width of Pinus koraiensis using BLS. A crown width prediction model was then established using BLS data. 12 sample plots of Pinus koraiensis plantations were selected in Mengjiagang Forest Farm, Jiamusi City, Heilongjiang Province. Backpack laser scanning was utilized to acquire the point cloud data from these plots. The preprocessing steps included denoising, ground point normalization, and single tree segmentation. The processed point cloud data was then registered with the actual field measurements for the individual tree matching. Parameters of each single tree were extracted to calculate their extraction accuracies. A prediction model was then constructed for the crown width of Korean pine using extracted parameters. Eight commonly used crown width prediction models were evaluated to determine the best-performing as the basic model. Furthermore, a generalized model was obtained to incorporate the stand and single tree factors. Additionally, a nonlinear mixed prediction model was constructed for the crown width to consider the random effects at the plot level, particularly for the artificially cultivated Korean pine. The point cloud data from the backpack laser scanning was matched well with the actual field data, with an average matching rate of 98.5%. The accuracies of extraction were 0.964, 0.871, and 0.928 for the breast height diameter, tree height, and crown width, respectively. The extracted parameters of the point cloud showed significant correlations with the measured ones. The correlation coefficient (R²) between point cloud-extracted and field-measured breast height diameter was above 0.9 (0.904~0.973). The R² between extracted and measured tree height was above 0.650 (0.650~0.740). The R² between extracted and measured crown width was above 0.8 (0.817~0.888). A quadratic function-based model was provided for the best fitting and prediction ( ${}_{}^{}{R}_a^2$= 0.528, RMSE = 0.718, MAE = 0.580 , and MAPE = 0.157). Stand mean diameter was introduced into the quadratic model at breast height, height-diameter ratio, and the total basal area per hectare of trees larger than the subject tree. The ${}_{}^{}{R}_a^2$was improved by 11.15%, whereas, the RMSE was reduced by 6.731%. A better performance was achieved in the mixed-effects model with sample plots as a random effect, compared with the basic model. The accuracies were ranked in the descending order of the nonlinear mixed model ( ${}_{}^{}{R}_a^2$= 0.655, RMSE = 0.620, MAE = 0.484, and MAPE = 0.130), generalized model ( ${}_{}^{}{R}_a^2$ = 0.578, RMSE = 0.669, MAE = 0.501, and MAPE = 0.136), and basic model ( ${}_{}^{}{R}_a^2$ = 0.528, RMSE = 0.717, MAE = 0.580, and MAPE = 0.157). Backpack laser scanning shared better scanning effects on the breast height diameter, crown width, and tree height. The crown width prediction model with point cloud data can be expected to effectively predict the crown width of Pinus koraiensis. Point cloud data can be combined with field measurements to assist in forestry surveys using backpack laser scanning. The BLS can also be applied in dense forest stands and forestry surveys.