Tropical floodplains, such as Pantanal in Central South America, are important
features for land-atmosphere interactions. Schemes to account for floodplains should
therefore be included in Earth System Models, but this requires observations of
flooded area for validation. Satellite data is a possible solution to estimate the
flooded area but it is important to evaluate the different flood detection algorithms
available in order to use the most efficient for the region. This work explores different
methods to estimate the flooded area from the MODIS MOD09A1 satellite surface
reflectance product using spectral indexes (mNDWI, NDMI, NDMI-NDVI) to detect
the presence of water. We include the traditional threshold-based methods but also
some unsupervised classification methods such as the k-means and the Principal
Component Analysis applied on the water-related spectral indexes. The calibration
and validation of these methods are based on the hydrological knowledge of the
region, coming from land surface models, river discharge observation and from
previous satellite estimations of the flooded area. The NDMI index seems too
sensible to the vegetation which leads to error in the estimation of the flooded
area. The other methods were spatially and temporally consistent with previous
studies over the Pantanal.