Estimation of forest parameters using the non-parametric techniques and satellite images at compartment level

Abstract

This paper discusses the use of medium resolution Landsat TM satellite images to support conventional approaches of Lithuanian forest inventory practices. Estimation accuracies achieved using just field measured sample plots, Landsat TM satellite images and two non-parametric k-nearest neighbor and most similar neighbor estimators were studied at a level of compartments. 19 mature forest areas, prepared for final felling with GPS measured borders and all trees callipered, were used for validation. Notably higher estimation accuracies were achieved using field sample plots distributed through the whole forest area studied than just ones located on mature forest stands. The root mean square deviations in estimating compartment-wise volume of growing stock per 1 ha was around 27-28% if the best variant of estimation approach was used. Possible influence of the accuracy in locating the borders of validation areas on the estimations is discussed in the paper, too