Monitoring flooded area fraction in floodplains of Paraná basin using passive and active microwave systems

SALVIA, MARÍA MERCEDES; GRINGS, FRANCISCO MATIAS; KARSZENBAUM HAYDEE; PERNA, PABLO; BARBER, MATÍAS; FERRAZZOLI, PAOLO

Foz Do Iguassú, Brasil

Congreso; The Meeting of the Americas 2010; 2010

American Geophysical Union

Over the past two decades, orbital passive microwave systems have proven to be sensitive to flood condition in large floodplains. This sensitivity is rooted in the well differentiated emission properties of calm water with respect to non-flooded land of any kind. For bare soil (or calm water), the emissivity at vertical (V) polarization is higher than the one at horizontal (H) polarization. Basically, an increase of soil moisture produces a decrease of emissivity at both polarizations, and at all microwave frequencies. However, the effect is more important at H polarization and at the lower frequencies. On its turn, the influence of vegetation is dependent on frequency, overall biomass and geometrical properties of vegetation elements. Several studies were aimed at singling out vegetation effects from soil effects. Moreover, in order to eliminate the dependence on surface temperature, a normalized polarization index (PI) was used. It was found that the polarization difference at high frequencies was mostly related to vegetation emission, and that lower frequencies, were best for soil condition monitoring. Passive signatures collected by SMMR and SSM/I instruments were exploited for flood monitoring applications. In these sensors, 37 GHz and 19 GHz channels were used, since lower frequencies were available only at very poor resolutions or not available at all. Although high frequencies suffer atmospheric effects and high canopy attenuation, valid results in some applications were obtained. A simple algorithm, based on the polarization difference at 37 GHz measured by SMMR, was adopted to estimate the fraction of flooded area during several events occurred in the Amazon River and in other large South American rivers floodplains. This algorithm is based on (1) a model of allowed pixel end-members and (2) an additive hypothesis about the emissivity of the footprint. Using these hypothesis and several resourceful methods to estimate the emissivity of footprint end-members and their fractions, several estimations of wetland flooded area were published. These approaches worked well in areas where the pixel model chosen (open water, flooded land, non-flooded land), correctly represented the real situation at pixel scale. Furthermore, one of the strongest hypotheses of this approach is that the fraction of open water is a constant value and can be estimated independently. Recently, we have analyzed AMSR-E signatures of the Paraná River basin during a large flood event. Since we have found a strong correlation between PI and water level at several hydrometric stations, we have applied the algorithm developed in the Amazon case for a region in the Paraná River Delta. For this area there is a general correspondence between the variations of flooded area and variations of measured water level. In this paper, a refined approach to estimate flooded area of the Paraná River Delta based on a more complex pixel model and passive-active synergy is presented. Based on a high resolution landcover map of the area, the pixel model now includes the five mayor pixel end-members present in the area (open water, flooded/non-flooded forest and flooded/non-flooded marsh) and a method to estimate their PIs and fractional area. Moreover, in this new approach, Envisat ASAR and Radarsat 2 ScanSAR mode data are used to estimate flooded area using a simple change detection scheme. This high resolution information about flood condition is then used to recalibrate the passive estimate.