THE FUSION OF INDIVIDUAL TREE DETECTION AND VISUAL INTERPRETATION IN ASSESSMENT OF FOREST VARIABLES FROM LASER POINT CLOUDS

Abstract. In this study we searched the obtainable accuracy of forest inventory based on the individual tree detection (ITD) by using fusion of automatic ITD (ITD_auto) and visual interpretation of laser point clouds. Current ITD algorithms, mostly based on segmentation of canopy height models (CHMs), are not able to utilize the whole information included in three-dimensional point clouds. We hypothesized that visual interpretation of the point cloud could provide so-called "best case" tree detection that could be achievable automatically. We refer to this method consisting of ITD_auto and visual interpretation as ITD_visual. We assessed the plot level accuracies of the ITD_auto and ITD_visual in boreal managed forest conditions using 322 plots. Based on the results the accuracy of ITD can be improved with visual interpretation. Omission trees are mainly missing from both ITD-methods. ITD_visual produced more accurate estimates for all forest variables compared to ITD_auto, e.g. RMSE% in volume decreased from 33.3% to 27.8% and bias% in volume from 4.1% to 2.3%. Area-based approach (ABA) is becoming more general for operational forest inventories with sparser laser data. ITDvisual would be justified if it could replace expensive field work in plot-wise measurements needed for ABA. Further research is needed in the use of ITD results as a reference for ABA.