Optimizing the k-Nearest Neighbors technique for estimating forest aboveground biomass using airborne laser scanning data

Nearest neighbors techniques calculate predictions as linear combinations of observations for a selected number of population units in a sample that are most similar, or nearest, in a space of auxiliary variables to the population unit requiring the prediction. Nearest neighbors techniques have been shown to be particularly effective when used with forest inventory and remotely sensed data. Recent attention has focused on developing an underlying foundation consisting of diagnostic tools, inferential extensions, and techniques for optimization. For a study area in Norway, forest inventory and airborne laser scanning data were used with the k-Nearest Neighbors technique to estimate mean aboveground biomass per unit area. Optimization entailed reduction of the dimension of feature space, deletion of influential outliers, and selection of optimal weights for the weighted Euclidean distance metric. These optimization steps increased the proportion of variability explained in the reference set by as much as 20%, reduced confidence interval widths by as much as 35%, and produced standard errors that were as small as 3% of the estimate of the mean. Published by Elsevier Inc.