Wetlands are among the most ecologically important ecosystems on Earth and their sustainability depends critically on the water resources. In a scenario of climate change and increased anthropogenic pressure, detailed monitoring of the water resources provides a fundamental tool to assess the ecosystem health and identify potential threats.
Doñana wetlands, in Southwest Spain, dry out every summer and progressively flood in fall and winter to a maximum extent of 30,000 ha. The wetland filling up process was monitored in detail during the 2006–2007 hydrologic cycle by means of twenty-one Envisat/ASAR scenes, acquired at different incidence angles in order to maximize the observation frequency. Flood mapping from the two uncorrelated ASAR channel data alone was proved unfeasible due to the complex casuistic of Doñana cover backscattering. This study addresses the synergistic utilization of the ASAR data together with Doñana's digital elevation model and vegetation map in order to achieve flood mapping.
Filtering and clustering algorithms were developed for the automated generation of Doñana flood maps from the ASAR images. The use of irregular filtering neighborhoods adapted to the elevation contours drastically improved the ASAR image filtering. Edge preservation was excellent, since natural edges closely follow terrain contours. Isotropic neighborhoods were assumed of a single class and their intensities were averaged. As a result, intensity fluctuations due to speckle and texture over areas of the same cover type were smoothed remarkably.
The clustering and classification algorithm operate on individual sub-basins, as the pixel elevation is more accurately related to the cover classes within them. Vegetation and elevation maps plus knowledge of Doñana backscattering characteristics from preceding studies were initially used to select seed pixels with high confidence on their class membership. Next, a region growing algorithm extends the seed regions with new pixels based on their planimmetric adjacency and backscattering Mahalanobis distance to the seeds.
During the seed region growth, new pixels' possible classes are not constrained to their cover type according to the vegetation map, so the algorithm is able to capture temporal changes in the vegetation spatial distribution. Comparison of the resultant classification and concurrent ground truth yielded 92% of flood mapping accuracy. The flood mapping method is applicable to the available ASAR images of Doñana from six other hydrologic cycles.