Mapping dynamic wetland processes with a one year RADARSAT-2 quad pol time-series

Remotely sensed data are widely used to identify, delineate and characterize wetlands. Optical data provide interesting information on land-use and land cover but are limited to cloud-free periods and to a description of the top layer of the vegetation strata because penetration depth is very small. For these reasons, it is not possible to precisely inventory wetland vegetation and agricultural practices, as well as water cycles and water levels in these areas with passive remote sensing techniques. The objective of this article is to evaluate fully polarimetric RADARSAT-2 time series datasets to determine the water cycle dynamics, in order to delineate precisely potential, effective and efficient wetlands. To that end, the development and validation of a supervised PolSAR segmentation including multi-temporal analysis of wetland evolution, and the investigation of polarimetric decomposition methods for quantitative physical parameter inversion algorithms are presented. The proposed methodology is based on the segmentation of a polarimetric descriptor, the Shannon Entropy, which has been shown to be a very sensitive parameter to the temporal variability of flooded areas. The results have been validated using soil moisture measurements in the field and a LiDAR image. They show that it is possible to produce detailed water feature maps useful for delineating water tables as well as water-saturated areas, and for monitoring water area dynamics. These products provide useful information to identify and delineate wetlands in order to support conservation and management in these ecosystems across large areas.