Derivation of 30-m-resolution Water Maps from TERRA/MODIS and SRTM

The high temporal resolution and large coverage of coarse- to moderate-resolution satellite imagery, such as NOAA/AVHRR and EOS-TERRA/MODIS, are very advantageous for flood monitoring, but their coarse spatial resolution hinders their wider application. Overcoming this limitation is an interesting scientific challenge with substantial application potential. When both the total water area and topography are known, it is possible to locate regional water distribution in small regions, e.g., less than 1 km2. The water fraction derived from TERRA/MODIS not only provides the water area in 500-m pixels, but it also contains geo-location information in each pixel, providing a potential way to derive high-resolution water maps with the use of SRTM digital elevation model (DEM) data. In this study we introduce an integration method to downscale MODIS 500-m water fraction maps to 30-m water maps with SRTM 30-m DEM data. The method uses an iteration cycle to compare water fraction by increasing water surface level based on the minimal elevation (searched from the 30-m SRTM cells within a 500-m pixel) to water fraction retrieved with dynamic nearest neighboring searching (DNNS) method from TERRA/MODIS multi-channel data, and iterations are terminated when the two fractions are equal to each other. The final water surface level is viewed as the maximal water surface level and smoothed by a uniformity process to reduce the impact from TERRA/MODIS water fraction retrieval errors. Then, cells in SRTM 30-m DEM with elevation between the minimal and maximal water surface levels are assigned as water to generate 30-m water maps. The presented method is applied to the Mississippi river flood of May 2011 and shows promising results. A validation analysis is performed using simultaneous Landsat TM 30-m data, Worldview-2 2-m data and river gauge observations and shows high consistency between integrated MODIS 30-m water maps and high resolution real-time satellite water maps, and the matched rate with TM is above 93%. The calculated maximal water surface levels used for integration are also close to the water levels observed at river gauges along Mississippi River. The validation result indicates the feasibility of this method to derive high-resolution water maps from coarse- or moderate-resolution satellite data, which can significantly improve the capability of these satellite data in flood analysis for decision-makers and downstream users.