THE K-NEAREST-NEIGHBOR METHOD FOR ESTIMATING BASAL-AREA DIAMETER DISTRIBUTION

The purpose of this study was to examine the use of the non-parametric k-nearest-neighbour method in estimating the basal-area diameter dist ribution. Distance-weighted nearest-neighbour estimation was based on the similarities of known stand characteristics, e.g. basal area and b asal area median diameter, between the target stand and reference stan ds. The weights for the reference stands were determined using similar ity distances. These similarity distance functions were determined by tree species using standardized stand variables. The material of this study consisted of 553 forest stands located in eastern Finland. The a ccuracy of the k-nearest-neighbour method was compared with the Weibul l distribution by tree species, using current stand volume characteris tics and those that should exist after IO yrs' simulation time. The We ibull-based method tn;as, in most cases, more accurate than the neares t-neighbour method, but the results were also quite good for the k-nea rest-neighbour method. Both methods were negligibly biased across the whole material. However, at the extremes of the volume distribution cl ear underestimates were obtained when using the XI-nearest-neighbour m ethod.