Estimating daily surface water fraction by using regression tree and MODIS derived water indices

Emergency response and management to flooding, the costliest disaster, require timely synoptic observations. Being able to acquire information on Earth surface efficiently, remote sensing has been used to monitor water for decades. However, current sensor design leads to an inevitable compromise between temporal and spatial resolution, which limits the application in sudden or extensive floods. This study aims to address that compromise by estimating surface water fraction from the MODIS images, which are daily available but with a relatively low spatial resolution of 500 m. The Automated Water Extraction Index (AWEI) is designed to better perform in shadowed and built-up areas, and this index is used in the proposed method for water classification and fraction estimation. A relationship between MODIS derived water indices and surface water fraction is first fitted by a pair of a coarse MODIS image and a fine water classification map from a Landsat image (30 m resolution), using a regression tree. The fitted tree is then applied on other MODIS images to produce a series of daily surface water fraction maps. In the experiment, the root-mean-square error is about 6.5%. By mapping water extent daily and with a higher resolution, the proposed method allows near-real-time emergency responses to and long-term analyses of inundation.