Silvilaser 2019 - Poster Presentations »
Tree buffer exclusion improves tree detection on airborne laser scanning surveys in even-aged stands
Individual tree detection (ITD) is one of the approaches to estimate stand attributes using LiDAR data. The most critical part of this approach is the actual detection of the tree or delineation of its crown. Because of that, many studies have been carried out to develop and test algorithms that correctly identify trees over different forest structures. Usually, for even-aged stands, a simple local maxima filter would work, as the canopy structure is homogeneous. In spite of that, even under these conditions the number of trees is under or overestimated in many cases. This study tests an add-in for three tree detection methods for even-aged stand application. We tested two (software FUSION and rLiDAR R package) local maxima filter algorithms (LMF) with three window sizes each, and the watershed approach. Afterward, we performed the “Tree Buffer Exclusion” using radius from 0.5 to 2 meters. This exclusion consists of iterations over all the treetops creating a buffer, keeping the highest value and excluding the others that fall within the buffer. The results were compared to the number of trees found in the forest inventory. The area of study had 470ha of eucalyptus ranging from three to six years divided into 19 stands with two different spacing(3x3 and 4x3m). The LiDAR data was obtained from the Cessna 206 with a point density of 5 points.m-2 and maximum scan angle of 30°. The results indicate that the methods, without buffer exclusion, usually overestimate the tree density in a stand. For the local maxima filters as the window size increases the number of trees is underestimated with no mean near where the error equals to zero. With the tree buffer exclusion was possible to obtain errors ranging from 0.9 to 25% with a relative RMSE(%) of 13.38%. All methods were improved using the tree buffer exclusion. The best overall method was using the FUSION’s local maxima filter with a two meters window size and a 1.5-meter buffer for tree exclusion. However, the best method was different if we analyze stand-by-stand. That might be explained for the differences in age, slope, spacing and scan angle of the stands which might be explored in further studies.