Abstract:The objective of the "Tree Extraction" project organized by EuroSDR (European Spatial data Research) and ISPRS (International Society of Photogrammetry and Remote Sensing) was to evaluate the quality, accuracy, and feasibility of automatic tree extraction methods, mainly based on laser scanner data. In the final report of the project, Kaartinen and Hyyppä (2008) reported a high variation in the quality of the published methods under boreal forest conditions and with varying laser point densities. This paper summarizes the findings beyond the final report after analyzing the results obtained in different tree height classes. Omission/Commission statistics as well as neighborhood relations are taken into account. Additionally, four automatic tree detection and extraction techniques were added to the test. Several methods in this experiment were superior to manual processing in the dominant, co-dominant and suppressed tree storeys. In general, as expected, the taller the tree, the better the location accuracy. The accuracy of tree height, after removing gross errors, was better than 0.5 m in all tree height classes with the best methods investigated in this experiment. For forest inventory, minimum curvature-based tree detection accompanied by point cloud-based cluster detection for suppressed trees is a solution that deserves attention in the future.

No abstract

No abstract

No abstract

The reference sample plot (RSP) method is a distance-weighted k nearest neighbour estimation method, which allows simultaneous interpretation of several variables. In the RSP method, the k spectrally nearest field plots are looked at separately for each unknown pixel, and the area weight of the unknown pixel is divided as a function of the spectral distances to the nearest plots. The RSP method was examined in a forest inventory for estimating stem volumes by tree species groups using different satellite materials. Two methods were tested both in searching for and weighting the nearest field plots. Euclidean distance functions worked steadily with all the volume variables studied. The other distance measure tested was based on regression modelling. With more than 15 plots, both covariance weighting and inverse distance weighting gave similar results. Considering the field data of this study, the suitable number of the nearest plots in plotwise estimation appeared to be between 10 and 15 plots. With Landsat TM, SPOT XS and SPOT P, the differences in standard errors were minor. When combined TM and SPOT P were used, the plotwise standard error of total volume was still over 60 per cent.