Local-scale accuracy assessment of vegetation cover change maps derived from Global Forest Change data, ClasLite, and supervised classifications: case study at part of Riau Province, Indonesia

摘要

Massive deforestation in Indonesia drives the need for proper monitoring using appropriate technology and method. The continuing mission of Landsat sensor extends the observation to almost 30?years back, initiating the ability to monitor the dynamics of vegetation intensively. By taking the advantage of the Landsat archive, advanced semi-automatic classification method, namely ClasLite developed by Asner et al. (J Appl Remote Sens 3:33543–33543, 2009) and a new end-product of 30?m Global Forest Cover Change cover (GFC) datasets developed by (Hansen et al. in Science 342:850–853, 2013a), offered the ability to easily monitor deforestation and forest degradation with little or few knowledge of mapping. This study aims to assess the performance of these newly available products of GFC and the ClasLite method against the traditional pixel-based supervised classification of minimum distance to mean (MD), maximum likelihood (ML), spectral angle mapper (SAM), and random forest (RF). Visual image interpretation of pan-sharpened Landsat was carried out to measure the accuracy of each final map. Result demonstrated that GFC and CLaslite performance has 3 to 18% higher overall accuracy for mapping vegetation cover change compared with the conventional supervised analysis using MD, ML, SAM, and RF with ClasLite as the most accurate method with 78.14?±?2%. Further adjustment of the cover change map of GFC by using forest extent from ClasLite was able to increase the accuracy of the original GFC data by 10%. Therefore, GFC and ClasLite ensure the ability to monitor vegetation cover change accurately in a simple manner.