Onthe Synergestic Use of ENVISAT/ASAR Imagery and Ancilliary Spatial Data for Wetlands Monitoring
Marti-Cardona, B.; Dolz-Ripolles, J.
Universitat Politècnica de Catalunya, SPAIN
Doñana wetlands are located on the right bank of the Guadalquivir River, in South West Spain. The marshes completely dry out every summer and flood again during fall and winter. The flood extent depends on the cumulated precipitation of the hydrological year, and can reach a maximum area of approximately 30,000 ha in the wetest cycles. The marshes filling up process was monitored between 2006 and 2010 by means of Envisat/ASAR images, acquired at different incidence angles in order to increase the observation frequency.
As usual in marshland areas, Doñana's topography is extremely flat, with a maximum elevation difference of 2.50 m in its entire extension. Despite its subtleness, Doñana's topography has a crucial effect on the hydroperiod or time that each zone remains flooded, what in turn determines the vegetal species growing in it. In 2002 a digital terrain model of Doñana marshes was built based on the elevation data collected in a LIDAR survey flight. It is well-known that the utility of radar images for land cover mapping is seriously undermined by the speckle phenomenon. The literature on speckle filtering is vast, but all filters need to be applied over locally stationary neighborhoods, so that the filtered value does not smear edges or mixes up backscattering values from different cover types. Given the strong correlation between Doñana covers and its topography, neighboring pixels at the same elevation are very likely to belong to the same class. Hence, the DTM provides exquisite information to choose homogeneous filtering neighborhoods, which will be large in flat areas and will become smaller as the terrain gets steeper. Filtering neighborhoods defined in this way are more likely to follow natural edges than square or other fixed-geometry windows used by generalist methods.
This article describes the filtering procedure applied to Doñana ASAR images under the guidance of the DTM. Results are compared to those obtained by filtering the images with common speckle filters. The terrain information is then implemented into a clustering algorithm in order to segment and classify the images, with the final aim to produce flood maps.
The rapid increase in the availability of precise DTMs, together with the fact that physical and ecological parameters in natural environments are often tightly linked to the terrain elevation, point at the interest of implementing DTMs information into the filtering and processing of SAR images.