Improving the accuracy of multisource forest inventory estimates to reducing plot location error — a multicriteria approach

Abstract

A procedure is introduced to reassign satellite image information to field plot data of forest inventory by a multicriteria approach. The method can be utilised in satellite-image-aided forest inventories, e.g., as in the Finnish Multi-Source National Forest Inventory (FMS-NFI) since 1990. This inventory method presumes that the field sample plots are geographically accurately located and each individual field plot can be identified with a specific picture element of the applied satellite image. Small area estimation errors are highly sensitive to the location errors of field plots with respect to the satellite image. In spite of current GPS systems, an accurate location is extremely difficult to achieve due to map errors and errors in rectifying a satellite image on a map. Satellite image information is reassigned to the field plots within an n× n image window around the assumed location. A weighted function of the correlation coefficients of the selected image and field variables is used as a scaling function in this multicriteria optimisation approach. The root mean square errors (RMSEs) of estimates, produced with a nonparametric k-nearest-neighbour ( k-nn) estimation method, are applied as a criterion when judging the final goodness of the relocation. The relocation reduces the pixel-level RMSE of total volume per hectare by 36%.