Assessing land use/land cover change : detection algorithms using Sentinel-2 satellite time series from 2018 to 2022

Details

Supervisors

Defourny, Pierre ; Lamarche, Céline ; Bos, Audric

Faculty

Faculté des bioingénieurs

Degree label

Master [120] en sciences agronomiques et industries du vivant, à finalité spécialisée: Geo-Information Science and Earth Observation for Environmental Modelling and Management

Abstract

Land Use and Land Cover (LULC) change detection algorithms are essential for monitoring environmental changes. This study evaluates the performance of two prominent algorithms, Continuous Change Detection and Classification (CCDC) and Breaks For Additive Season and Trend Monitor (BFASTm), in the Wallonia region of Belgium. All available Sentinel-2 images acquired between 2017 and 2022 were used. Changes were detected and analyzed across seven different LULC classes. A random stratified sample design was used for assessing the change detection accuracy with 1228 sample pixels. The performance of both algorithms was assessed using Lifewatch data from 2018 to 2022 as the reference dataset. CCDC proved to be effective for detecting changes in almost all land cover classes. It achieved a higher overall accuracy of 60%, accurately identifying true changes in 31% of cases. Conversely, BFASTm identified true changes in 26% of cases, with an overall accuracy of 55%. Both algorithms demonstrated strong performance in detecting changes in grassland areas while exhibiting the poorest performance in wetland regions. The two methods experienced a decline in performance due to high cloud cover, resulting in producing higher commission errors for change pixels, and temporal constraints related to the reference dataset. This study underscores the challenges of achieving high accuracy in cloud-prone regions like Wallonia and emphasizes the need to refine these algorithms to better handle the complexities of LULC change detection.