Evaluating kNN and regression methods for predicting tree and stand variables from LiDAR data in plantation forests

LiDAR scanning offers the possibility of reducing fieldwork in forest inventories. Growing stock variables can be derived from LiDAR data using k-nearest neighbor (kNN) data imputation or regression modeling. The inventory data needed for forest planning consist of either tree- or stand-level variables, depending on the type of model used for growth prediction. This study compared LiDAR-based regressions with kNN data imputation to obtain forest inventory data from the plantation forests of northeastern China, assuming that either stand- or tree-level models are used in management planning. In the former case, a set of four stand-level variables (dominant height, median tree diameter, basal area, and number of trees per hectare) was imputed with the kNN method or predicted with regression models. In the latter case, the entire diameter distribution of the trees was estimated. The kNN method imputed the distributions directly from the field plots, whereas the regression method predicted the parameters of the Weibull distribution using either a parameter prediction method (PPM) or a percentile-based parameter recovery method (PRMP). Regression modeling and kNN performed similarly in the simultaneous estimation of four stand-level variables, the relative RMSEs being 17.32% (regression) and 15.53% (kNN) for stand basal area and 5.34% (regression) and 7.50% (kNN) for dominant height. When the diameter distributions were estimated, the kNN and PPM methods performed better than the PRMP method. PPM was better than kNN for unimodal diameter distributions, whereas kNN was better for bimodal and multimodal distributions. As the plantation forests of northeastern China frequently have natural regeneration of shade-tolerant species, resulting in bi- and multimodal diameter distributions, kNN imputation may be recommended for future studies.

Lidar · Linear regression · Regression · Regression analysis · Tree (set theory) · Forest Ecology and Biodiversity Studies · Forest ecology and management · Remote Sensing and LiDAR Applications

• The Kolmogorov-Smirnov Test for Goodness of Fit

  Frank J Massey•Journal of the American…•1951

  Add to list