Forest stand characteristics estimation using a most similar neighbor approach and image spatial structure information

In this study, information on variograms is integrated with image interpretation of standwise volume of tree stock by using a nonparametric method based on a distance-weighted mean of most similar neighbors (MSN). The usability of the various indicator attributes, including image pixel characteristics and features derived from a variogram curve, in the interpretation is studied by analyzing the accuracy of the results at the compartment level. The accuracy of the volume estimation at the stand level was improved when the empirical variogram values were included in the set of indicator attributes in the MSN analysis. With this kind of MSN analysis, the chosen k nearest neighbors have a similar spatial variation structure in the image material as in the target stand. It was found that increasing the number of similar neighbors beyond 3 did not improve the accuracy. When the variogram information was included in the indicator attribute set, the root mean square error (RMSE) of volume estimate was at its lowest at 18% and the bias was then −0.6%. When the variogram information was not used, the RMSE was at its lowest at 24–27% and the bias was 0.2–1.8%, depending on the number of indicator variables used.