Monitoring the dynamics of wetland inundation by random sets on multi-temporal images

General models are required for better understanding the dynamics of wetland inundations and their water regimes. Extraction by imagery snapshots and use of a crisp data model do not value the inherent uncertainties in space and time. This study addresses parameterization of a mixed Gaussian random set model in a multi-temporal analysis. The model is applied to monitor annual variation of wetland inundation extents from a series of Landsat TM images in 2004 and HJ images in 2009 on the Poyang Lake national nature reserve (PLNNR) in China. We use related indices to represent spatial uncertainties of inundated areas and to delineate the transition zone between wetland vegetation and open water. The PLNNR water regime is investigated by accumulating a series of random sets during one year and determining the water covering days (WCD) at the pixel level. Random sets provide detailed spatial configurations of the WCD which has a strong negative correlation with the underwater DEM. Comparing 2004 and 2009, the study shows that almost half of the PLNNR area experienced drought. We conclude that the mixed Gaussian random set model with three components presented in this study serves as a general method to parameterize the random set model for large datasets. Moreover, it is well suited to capture detailed information on spatial temporal dynamic of wetland inundation and contributes to our understanding of wetlands water regimes from multi-temporal images.

Research highlights
► Model wetland inundation as random sets and extract transition zones.
► Develop a mixed Gaussian model to parameterize random sets for large datasets.
► Random sets provide detailed representation of inundation by mapping transition zone.
► Model wetland inundation dynamics by integrating a series of random sets.
► More precise spatial configuration of water covering days captured by random sets.