Assessing the potential for forest residue classification and distribution over clear felled areas using UAVs and Machine Learning: a preliminary case study in South Africa

Abstract

The use of UAV based images in forestry allows for the coverage of large areas with a high level of detail. The combination of this information with machine learning (ML) techniques provides significant data for management and forest operations. This study focuses on evaluating the potential of UAVs based images and the use of ML algorithms to assess the distribution and classification of forest residues over clear felled areas. A random forest model was built using RGB bands, textural variables, and information from the surface model to classify elements in a clear felled site. The classification resulted in an overall accuracy of 91% with high values for coarse woody debris (CWD) and ground detection. We concluded that the method shows a significant and solid improvement for the classification of forest residues in clear felled sites.