Inventory of the quantitative characteristics of single oak trees with nonparametric methods of Support Vector Machines and Decision Tree on satellite images of WorldView-2 and UAV

Abstract

To achieve effective forests management it is necessary to obtain reliable statistical data like the number of stands, diameter at breast height (DBH), and ‎crown volume. While traditional methods of the forests measurements are very labor intensive and time consuming, remote sensing can provide up-to-date and low cost ‎data. In comparing ot other sensors, the satellite WV-2 ‎generate very high resolution images that can be used in the forest management practices. In the present study, we aimed to estimate parameters related on the single trees characteristics using decision tree ‎method and Support Vector Machines classification with complex matrix evaluation and Area ‎under operating characteristic curve (AUC) method. We also used UAV Phantom 4 ‎Pro images from two distinct geographic regions. Support Vector Machines ‎classification method generated the highest accuracy in estimating single trees parameters. This study confirms that using WV-2 data it is possible to extract the necessary ‎parameters of the single trees and relied them in the forest management practices.