CONICET | Buscador de Institutos y Recursos Humanos

We propose a multi-temporal approach to discriminate macrophyte vegetation types in the Lower Paraná River floodplain. During a low intensity flood pulse, seven X-Band COSMO-SkyMed HImage images were acquired, covering a nine-month period. The region was first segmented with a mean-shift segmentation method using the information from the complete temporal series. Next,objects were classified with the expectation maximization algorithm into spectral classes. These spectral classes were assigned to six information classes defined through field sampling: Bulrush marshes, Short broad-leaf marshes, Tall broad-leaf marshes, Short grasslands and grass marshes, Tall grasslands and grass marshes, and Water. Class interpretation was based on the backscatter dynamics in relation to plant coverage and to hydrometric level. We related backscattering coefficient changes to interaction mechanisms: mirror reflectance (water-covered areas), volume dispersion (emergent vegetation with medium backscatter) and double-bounce (emergent flooded vegetation with high backscatter when water is present). The accuracy of the obtained product was assesed by comparing it with 55 field samplings. Global accuracy was 71.2%, whereas Kappa index was 63.4%. This work points out the usefulness of X-Band data for flood monitoring and macrophyte vegetation type discrimination. In a mosaic of herbaceous wetlands, the dynamics associated with flood pulse may change within patches in different geomorphological settings and topographical positions. The knowledge on the relation between vegetation, local settings and floods is essential for interpreting and predicting how backscattering coefficients and other SAR-derivated parameters vary with flooding.

enviar mensaje