Système d'alerte dynamique par télédétection pour l'observation des déforestations en Malaisie

Authors: Perbet, Pauline
Advisor: Béland, Martin
Abstract: Malaysia is under severe deforestation pressure. There is an urgent need to plan oil palm cultivation by turning to sustainable strategies. An alert system targeting illegal deforestation would be a valuable tool for managers. Google Earth Engine is used to set up a warning system of deforestation from medium resolution satellite images. Sentinel-2 and Landsat 8 optical images are combined with Sentinel-1 radar images for high repeatability and limit cloud cover impact. The change vector analysis detects deforestation on each new acquisition based on forest reference image. Results show sufficient accuracy with optical sensors (less than 11% commission and omission error). However, the radar sensor gives lower results (14% commission error and 28% commission error), related to many artefacts caused by the speckle. Therefore, we use a dynamic score related to the observation of 3 sensors over 3 months, which indicates a level of confidence. This method provides 7% of omission error and 0% of commission error for events with a high level of confidence, with a score greater than 20 points.
Document Type: Mémoire de maîtrise
Issue Date: 2019
Open Access Date: 30 November 2020
Grantor: Université Laval
Collection:Thèses et mémoires

Files in this item:
Description SizeFormat 
35483.pdf1.98 MBAdobe PDFThumbnail
All documents in CorpusUL are protected by Copyright Act of Canada.