The topographic maps were geocoded to universal transverse mercator (UTM), zone 49N, using the WGS84 ellipsoid, and were used to delineate the check dams in ArcMap manually. Landsat images were subjected to atmospheric correction, radiometric calibration, and radiometric rectification procedures to facilitate comparability between dates.22 The Landsat images were geo-referenced to 1:50,000 scale topographic maps using second-order polynomial transformation with the root mean square error of below 0.5 pixels (), and registration was verified visually by overlaying and swiping registered images. Due to the strong absorption and reflectance region in the infrared, visible red, and near-infrared bands (bands 7, 4, and 3) provided in the Landsat images, the combination of these bands was used to distinguish water bodies and land. The check dams or reservoirs were extracted by using the maximum likelihood supervised classification method in Erdas Imagine software.14,15 During the interpretation process, each 30-m pixel is classified as check dam (open water surface) or nondam area. Ponds and drainage network were distinguished from check dams through field survey. A database of the check dams (including volumes, built time, location, dyke height, remained/unfilled storage) were collected from Hydrology Bureau of the Yellow River Water Resources Commission. The volumes, built time, and dyke height were obtained through local census, and the remained storage for each check dams was calculated by combining bathymetric survey and storage curve analysis.