Classifying tree and nontree echoes from airborne laser scanning in the forest-tundra ecotone

Temperature-sensitive ecosystems such as the forest-tundra ecotone are expected to be particularly affected by the changing climate. A large proportion of the total land area in Norway is represented by the forest-tundra ecotone, and effective monitoring techniques for these areas are required. Airborne laser scanning (ALS) has been proposed for detection of small pioneer trees and its height and intensity data may hold potential for monitoring tasks. The main objective of the present study was to assess the capability of high-density ALS data (e. g., >2 m(-2)) to classify tree and nontree echoes directly from the laser point cloud. For this purpose, the laser height and intensity, a geospatial variable represented by the area of Voronoi polygons, and the terrain variables of aspect and slope were used to distinguish between tree and nontree laser echoes along a 1000 km transect stretching from northern Norway (66 degrees 19' N) to the southern part of the country (58 degrees 3' N). Generalised linear models (GLM) and support vector machines (SVM) were employed for the classification using different combinations of the aforementioned variables. Total accuracy and the Cohen's kappa coefficient were used for performance assessment for the different models. A total accuracy of at least 93% was found irrespective of classification method or model, and Cohen's kappa coefficients indicated moderate fits for all models using both classification methods. Comparisons of Cohen's kappa coefficients revealed equivalent performances for the GLM and SVM classification methods for models consisting of different combinations of the laser height, intensity, geospatial variable, and aspect. However, SVM was superior when laser height and intensity were used together with slope. In summary, the capability of high-density ALS data for the classification of tree and nontree echoes directly from the laser point cloud could be verified irrespective of the classification method.