Automated Classification of Trees outside Forest for Supporting Operational Management in Rural Landscapes

Abstract

Trees have important and diverse roles that make them essential outside of the forest. The use of remote sensing can substantially support traditional field inventories to evaluate and characterize this resource. Existing studies have already realized the automated detection of trees outside the forest (TOF) and classified the subsequently mapped TOF into three geometrical classes: single objects, linear objects, and ample objects. This study goes further by presenting a fully automated classification method that can support the operational management of TOF as it separates TOF into seven classes matching the definitions used in field inventories: Isolated tree, Aligned trees, Agglomerated trees, Hedge, Grove, Shrub, and Other. Using publicly available software tools, an orthophoto, and a LIDAR canopy height model (CHM), a TOF map was produced and a two-step method was developed for the classification of TOF: (1) the geometrical classification of each TOF polygon; and (2) the spatial neighboring analysis of elements and their classification into seven classes. The overall classification accuracy was 78%. Our results highlight that an automated TOF classification is possible with classes matching the definitions used in field inventories. This suggests that remote sensing has a huge potential to support the operational management of TOF as well as other research areas regarding TOF.