Estimating Stem Diameter Distributions in a Management Context for a Tolerant Hardwood Forest Using ALS Height and Intensity Data

Abstract

Two types of nonparametric modeling techniques and various metrics derived from airborne laser scanning (ALS) data were examined in terms of their utility for modeling stem diameter distributions in an uneven-aged tolerant hardwood forest in Ontario, Canada. Using an area-based approach (ABA), the frequency distribution of trees across 6 size classes was predicted using k-nearest neighbor (k-NN) imputation and Random Forest (RF) regression. Predictor variables derived from ALS height and intensity data were divided into 3 groups: height only, intensity only, and all metrics. Prediction results demonstrated that the first 2 groups of predictor variables exhibited similar predictive accuracy, whereas the synergy of both resulted in enhanced performance. The utility of intensity-based metrics was corroborated by an importance measure obtained from RF. The size class-specific stem density estimation approach based on RF was more accurate and flexible than the simultaneous estimation approach based on k-NN models. After the predicted diameter distributions were grouped into 9 structural groups, heterogeneous accuracy scores revealed the challenges for predicting select diameter distributions. Although successes were observed for certain size classes, there remains additional research (e.g., development of additional metrics or data types) to be done to accurately predict a complete range of size classes.