Comparison of sampling methods for estimation of nearest-neighbor index values

Abstract

Neighborhood-based indices such as mingling index and diameter differentiation are a set of diversity measures that are based on the relationship between a reference tree and a certain number of nearest neighbors (i.e., trees to which it has the lowest horizontal distance). Using stem-mapped data from eight headwater sites, we compared the relative bias and relative root mean square error (relative to the true mean of each site) of several different methods of choosing reference trees for calculation of diameter differentiation ([Formula: see text]) and species mingling ([Formula: see text]) index. Indices were defined using two, three, and four neighbors and methods for selection of the reference tree were random selection of a tree in a fixed-radius plot (FI), random selection of a tree in a variable-radius plot (VA), azimuth selection method (AZ), and nearest tree selection (NT). In general, the relative bias was lower than ±2.5% for [Formula: see text] and lower than ±10% for [Formula: see text] regardless of the method. The FI method consistently had the lowest relative bias and relative root mean squared error. The NT and AZ methods were second in terms of relative root mean squared error for [Formula: see text] and [Formula: see text], respectively. Simplicity of these two methods might outweigh their slightly worse performance.