A global, 30-m resolution land-surface water body dataset for 2000

Monday, 15 December 2014
Min Feng, Joseph O Sexton, Chengquan Huang, Dan-Xia Song, Xiao-Peng Song, Saurabh Channan and John R Townshend, Global Land Cover Facility, Univeristy of Maryland, College Park, MD, United States
Inland surface water is essential to terrestrial ecosystems and human civilization. The distribution of surface water in space and its change over time are related to many agricultural, environmental and ecological issues, and are important factors that must be considered in human socioeconomic development. Accurate mapping of surface water is essential for both scientific research and policy-driven applications. Satellite-based remote sensing provides snapshots of Earth’s surface and can be used as the main input for water mapping, especially in large areas. Global water areas have been mapped with coarse resolution remotely sensed data (e.g., the Moderate Resolution Imaging Spectroradiometer (MODIS)). However, most inland rivers and water bodies, as well as their changes, are too small to map at such coarse resolutions. Landsat TM (Thematic Mapper) and ETM+ (Enhanced Thematic Mapper Plus) imagery has a 30m spatial resolution and provides decades of records (~40 years). Since 2008, the opening of the Landsat archive, coupled with relatively lower costs associated with computing and data storage, has made comprehensive study of the dynamic changes of surface water over large even global areas more feasible. Although Landsat images have been used for regional and even global water mapping, the method can hardly be automated due to the difficulties on distinguishing inland surface water with variant degrees of impurities and mixing of soil background with only Landsat data. The spectral similarities to other land cover types, e.g., shadow and glacier remnants, also cause misidentification. We have developed a probabilistic based automatic approach for mapping inland surface water bodies. Landsat surface reflectance in multiple bands, derived water indices, and data from other sources are integrated to maximize the ability of identifying water without human interference. The approach has been implemented with open-source libraries to facilitate processing large amounts of Landsat images on high-performance computing machines. It has been applied to the ~9,000 Landsat scenes of the Global Land Survey (GLS) 2000 data collection to produce a global, 30m resolution inland surface water body data set, which will be made available on the Global Land Cover Facility (GLCF) website (http://www.landcover.org).