Abstract: For more than 400 years，check dams have been constructed for soil erosion control and agricultural production improvement. The most effective way to conserve soil and water in the Loess Plateau is to construct check dam systems in gullies. Obtaining the check dam information such as location，number，water surface area，dam- controlled area etc.，is very important for soil and water conservation in the watershed. Traditionally，the check dams’ distribution was evaluated by using measurement of topographic maps based on in-situ hydrographic survey with electronic planimeter. However，the traditional approach is usually laborious，costly and time consuming. Advances in remote sensing，geographic information systems and computer technology have made it more convenient and less expensive for obtaining the parameters of check dams. The paper extracted check dam information by making use of multi-sources data in a Loess Plateau catchment. Huangfuchuan Watershed was taken as the study area because it was seriously exposed to severe soil and water loss and had been built larger number of check dams to intercept sed- iment that discharged into the Yellow River. Multi-sources data of the Huangfuchuan Watershed including topogra- phic maps （1976），landsat5-TM data （1990，2007） and Google Earth （2010） were collected for check dam interp- retation. The multi-sources data were preprocessed by geometric correction and eliminating cloud effects under the support of GIS spatial analysis techniques. The cloud cover of every image was less than 5% because the identificat- ion of check dams became somewhat complicated due to the disturbances from shadows. The topographic maps and Google Earth images can be directly used to extract check dams by analyzing the boundary outline of check dams on the GIS software platform. Based on the characteristics of visible light with high reflectivity from the surface，the Lan- dsat5-TM data was used to extract the check dams. The combination of bands 7，4 and 3 were used to separate water bodies in check dams from surrounding surface features by Erdas application. The results were verified based on fie- ld measurement through GPS. The results indicated that 392 check dams in 1976 were extracted from topographic maps，while 445 in 1990，504 in 2007，and 567 in 2010 were extracted from the images. The check dams in the west of the Huangfuchuan Watershed were much less than those in the east. The water surface area of check dams extrac- ted from the Landsat images was highly consistent with those of Google Earth images. 16 randomly selected check dams were chosen to verify the interpretation results through field survey. Whereas the dam-controlled areas between the landsat5-TM data and the field measured data showed good agreement with R2 higher than 0.96. It suggested that the topographic maps and Landsat images can be used for identifying the check dams. The NDAI （Normalized Difference Area Index） and DAI （Difference Area Index） falls within the range between -1 and 1. When the values is close to 0 giving the best linear fit between the surface water areas extract from Google Earth image and Landsat5- TM data respectively. Two indices showed that there was a very good linear fit between the different source data with extremely significance. Historically，dynamics in the check dams number，water surface area of check dams and dam-controlled area indicated their substantial trapping effects on stream flow and sediment load in the watershed. This study proved that the proposed approach can be applied for check dam extraction and may make it possible to analyze dynamic change of check dams. The results can be used to estimate of sediment trapping efficiency resulting from check dams and may provide decision supports for soil and water conservation in the Loess Plateau.

Key words: multi-sources data, Huangfuchuan, check dam, dynamic change