Constructing a virtual forest: Using hierarchical nearest neighbor imputation to generate simulated tree lists

Abstract

A nearest neighbors method for generating simulated tree lists has been developed. The method employs an implicit two-scale hierarchy to incorporate information from a coarse scale representing the distribution of stand attributes across a region and a fine scale representing the distribution of tree attributes within a stand. The tree list generation method was implemented and tested using data from untreated, naturally regenerated and planted forests in western Oregon, western Washington, and southern British Columbia west of the Cascade Mountains. Simulated tree lists were generated from stand scale attributes for each of the actual tree lists in the data. Distributions of stand scale and tree scale attributes were estimated and used to compare the simulated and actual tree lists. At the stand scale, distributions of quadratic mean diameter and average height for the simulated and actual stands were in very good agreement, having approximately 98% of their probability mass in common for each attribute. At the tree scale, comparisons of the distributions of diameter at breast height, height, and species composition between the simulated and actual stands were more variable, with approximately 84% of the simulated stands identified as statistically similar to their respective actual stands.