Automatic mapping of tree crowns in scattered-tree woodlands using low-density LiDAR data and infrared imagery

Abstract

Accurate estimation of canopy cover (CC) and its delineation in tree-grass ecosystems (TGE) such as savannas and silvopastoral systems are crucial to analyze and model the functioning of these systems and the role of trees at different scales. At large-scale, remote sensing is a key tool, and assessments of land use and landscape elements often rely on satellite and aircraft sensor imagery and light detection and ranging (LiDAR) data. This study addresses automatic mapping of CC in TGE using high-resolution infrared orthophotographs and low-density LiDAR data from the Spanish Aerial Orthophoto National Plan (PNOA). Canopy cover mapping was performed in two areas with different structural complexity (with or without shrub layer) by an object based image analysis (OBIA) approach applied on canopy height model (CHM) generated by the LiDAR point cloud, infrared imagery, and both sources combined (LiDAR-imagery fusion). Overall accuracy (OA) was more than 91% with the two separated methods and more than 95% combining them. The results show that low-density LiDAR data is not a reliable source for the automatic mapping of canopy of scattered tree in TGE, OBIA on high-resolution infrared orthophotographs allows a more accurate automatic delineation of tree canopy, and the combined approach was the only way to obtain acceptable mapping in shrub-encroached stands, where errors were still greater than 15% with single-source based methods.