Methods based on k-nearest neighbor regression in the prediction of basal area diameter distribution

In the Finnish compartmentwise inventory systems, growing stock is describe d with means and sums of tree characteristics, such as mean height and basa l area, by tree species. In the calculations, growing stock is described in a treewise manner using a diameter distribution predicted from stand varia bles. The treewise description is needed for several reasons, e.g., for pre dicting log volumes or stand growth and for analyzing the forest structure. In this study, methods for predicting the basal area diameter distribution based on the k-nearest neighbor (k-nn) regression are compared with method s based on parametric distributions. In the k-nn method, the predicted valu es for interesting variables are obtained as weighted averages of the value s of neighboring observations. Using k-nn based methods, the basal area dia meter distribution of a stand is predicted with a weighted average of the d istributions of k-nearest neighbors. The methods tested in this study inclu de weighted averages of (i) Weibull distributions of k-nearest neighbors, ( ii) distributions of k-nearest neighbors smoothed with the kernel method, a nd (iii) empirical distributions of the k-nearest neighbors. These methods are compared for the accuracy of stand volume estimation, stand structure d escription, and stand growth prediction. Methods based on the k-nn regressi on proved to give a more accurate description of the stand than the paramet ric methods.