We propose to apply a new regression tree algorithm (as demonstrated in Sun and Yu, 2010; Sun et al., 2011) for the retrieval of water fraction under all-sky conditions to better assess and document the severity of floods. Floods are the most frequent natural disasters, resulting in more loss of life and property in the United States than other types of severe weather events. Flood pixels, as seen from satellite, are comprised of water mixed with land, such as vegetation, trees, saturated soils, or impervious areas (e.g., roads and buildings). Measures of water fraction are thus more suitable for representing flooding conditions in varying stages, particularly compared to traditional water classification techniques. Our multistage retrieval strategy for deriving water fraction and flood extent will leverage recently launched Visible/Infrared Imager/Radiometer Suite (VIIRS), as well as MODIS data products for historic flooding events. Floods are usually associated with cloudy conditions. Moreover, a lot of floods occur during nighttime, especially in northern latitudes. Under these scenarios, we will use microwave imagery-sounding data, such as those available from the Advanced Technology Microwave Sensor (ATMS) and historic Advanced Microwave Sounding Unit (AMSU) observations, to estimate water fraction at coarse resolutions. We will then leverage our library of ancillary land surface data at various scales (e.g., DEM, soil type, and land cover/land use fields) to provide estimates of flood distribution at the VIIRS native spatial resolution (< 2.0 km). With this multi-sensor approach, the extent of floodplains and flood-prone areas will be approximated at small to intermediate map scales; thus enabling us to examine the relationships between flood severity and structural damage. Our study will be focused on flooding due to the landfall of tropical cyclones. This proposal is closely and directly relevant to the Disaster programs research focus area Flood prediction, mapping, analysis, and mitigation.