Relating Forest Attributes with Area- and Tree-Based Light Detection and Ranging Metrics for Western Oregon

Abstract

Three sets of linear models were developed to predict several forest attributes, using stand-level and single-tree remote sensing (STRS) light detection and ranging (LiDAR) metrics as predictor variables. The first used only area-level metrics (ALM) associated with first-return height distribution, percentage of cover, and canopy transparency. The second alternative included metrics of first-return LiDAR intensity. The third alternative used area-level variables derived from STRS LiDAR metrics. The ALM model for Lorey’s height did not change with inclusion of intensity and yielded the best results in terms of both model fit (adjusted R 2 0.93) and cross-validated relative root mean squared error (RRMSE 8.1%). The ALM model for density (stems per hectare) had the poorest precision initially (RRMSE 39.3%), but it improved dramatically (RRMSE 27.2%) when intensity metrics were included. The resulting RRMSE values of the ALM models excluding intensity for basal area, quadratic mean diameter, cubic stem volume, and average crown width were 20.7, 19.9, 30.7, and 17.1%, respectively. The STRS model for Lorey’s height showed a 3% improvement in RRMSE over the ALM models. The STRS basal area and density models significantly underperformed compared with the ALM models, with RRMSE values of 31.6 and 47.2%, respectively. The performance of STRS models for crown width, volume, and quadratic mean diameter was comparable to that of the ALM models.