Machine Learning Algorithms for Forest Stand Delineation Using Yearly Sentinel 2MSI Time Series

Forest stand types maps is fundamental tools for sustainable forest management that needs to be regularly updated. This work aims at the use of satellite images time series and machine learning techniques to automate and improve the efficiency of forest stands maps production. A time series of "sentinel2MSI" satellite images has been classified using supervised classification using various machine learning (ML) algorithms whose effectiveness has been proven in several research studies. The produced stands map, when each type is defined according to conventional criteria, was assessed and gives satisfactory results with a high variability depending on the used classification algorithm. Temporal segmentation of the archive appears to be a feasible and a robust means of increasing information extraction from the Sentinel archive.