Structure characterization on Mediterranean forest stand using terrestrial laser scanning

Forest stands plays an important role in Western Mediterranean ecosystems, their characterization it is needed for a whole comprehension of natural dynamics and for an efficient forestry management. The definition of the structural parameters of forest vegetation is a useful information for different environmental applications, such as studies on carbon dynamics, sustainable forest management, ecological studies, forest fuel studies and fire risk management. A precise description of the forest is particularly important for fire hazard mitigation planning because allows predictions of the potential fire behavior and its destructive effects. Obtaining detailed information on forest stand and canopy variables requires extensive, difficult, and laborious field campaigns. Remote and proximal sensing techniques for forest monitoring have become popular in recent decades. Specifically, Terrestrial Laser Scanner (TLS), based on Lidar technology, has demonstrated its potential to overcome the limitations of the conventional ground-based forest inventory techniques, but the accuracy and applicability of TLS techniques for estimation of tree attributes and canopy characterization, presupposes a correct separation between points representing shrubs, woody material, leaves and small branches and needs further investigations. In this work we developed and tested an automatic procedure based on the point density algorithms DBSCAN, to correctly separate points representing shrubs, woody material, leaves, and small branches at plot level, in order to identify woody material volumes, tree density and canopy cover on a forest stand. The study was carried out in several areas located in Sardinia, Italy, mainly covered by pine forest, mixed forest, and oak forest with different understory types. Destructive and non-destructive measurements were done inside circular plots of 10 m radius. TLS data sets were collected in field by multiple scanning of the plots. The 3D point clouds were processed for isolating trees, ground and understory and subsequently for separating wood from foliage. Cloud points were partitioned in cubic volumes (voxels) that were used as input to separate stand components (by applying principal component analysis) and to generate wood and no-wood clusters (by applying the point density algorithm DBSCAN). The first results obtained show that the proposed method allows to correctly identify foliage, trunk and main branches especially when the underlying layer is dominated by low herbaceous vegetation. However, further studies are needed to assess the ability of this method in forest stands characterized by high and dense undergrowth and with different species of trees.