Identifying Forest Dynamics for Change Detection Using Multitemporal SPOT Imagery

Change detection based on satellite remote sensing relies on the comparison of multispectral reflectance acquired at different dates. A major problem in forest change detection is to separate the signal change related to forest conversion from other sources of noise such as the ever-changing state of a forest stand from the regrowth to the mature age. This research aims at characterizing spectral reflectance over the whole forest succession in order to quantify the reflectance dynamics and to identify the most appropriate spectral signal thanks to a robust object-based change detection approach. From a large sample of spruce stands, spectral trajectories of forest cycle were derived for green, red and NIR reflectances and the NDVI derived from SPOT-HRVIR images. These trajectories were found typical for spruce stands and consistent between satellite images. Based on the object-based change detection method using image differencing, the combination of all spectral bands was proved more efficient to detect changes and more discriminant to distinguish them than any single spectral band or NDVI. Two forest changes, namely the clearing and the regrowth, have been distinguished based on multivariate analysis for different temporal resolutions. Finally, tills study emphasized the need of a better characterization of changes of interest for defining appropriate forest monitoring systems.