Monitoring flood is an important task for disaster management. It requires to distinguish between changes related to water from the other changes. We address such an issue by relying on both spatial and intensity information. To do so, we exploit min-tree that emphasize intensity extrema in a multiscale, efficient framework. We thus suggest a two-step approach operating on satellite image time series. We first perform a temporal analysis to identify images containing possible floods. Then a spatial analysis is achieved to detect flood areas on the selected images. Both steps relies on the analysis of component attributes extracted from the min-tree representation. We conduct some experiments on a flooded scene observed through Sentinel-1 SAR imagery. The results show that flood areas can be efficiently and accurately characterized with spatial component attributes extracted from hierarchical representations from SAR time series.