Automatic extraction technique for the erosion gully in the Loess Plateau based on bidirectional relief-shading method

Gully erosion has been one of the most important reasons for the land degradation in the Loess Plateau. There is a serious threat to the local infrastructure, grain security, and ecological environment. Therefore, it is very necessary to dynamic monitor the erosion gully. However, the traditional methods are very unfavorable for the Loess Plateau with a large area, such as manual measurement and image visual interpretation in this case. In this study, automatic extraction of erosion gully was proposed for the large-region and high-resolution Digital Elevation Model (DEM). Bidirectional Relief-shading served as an excellent erosion gully extraction for the Loess platform/tableland landform using shaded relief, indicating the balanced accuracy and efficiency. The characteristics of outstanding relief shade were used to extract the erosion gully in the area under the simulated light. But there were more false and missing areas in other landforms that caused by special terrains, such as steep slopes and wide flat gully bottom, which was greatly damaged the correctness and continuity of the extraction. The special terrain was first summarized for the false and missing areas in the other landforms. Taking the 3.2 m GF-7 DEM of Wubu County, Shaanxi Province, China, as the research object, the gully networks buffer filling, dilate-erode, and area threshold were applied to eliminate the false and missing areas, and then the erosion gully extraction algorithms were designed for the large-region and high-resolution grid DEM using object-oriented, data partition, image processing, and GIS technology. The 10 small watersheds with an area of 3-6 km2 were evenly selected to verify the extraction accuracy in the study area, and then the validation data was obtained by the visual interpretation with a 0.65 m GF-7 image. The results show that: 1) The false and missing areas were automatically eliminated to ensure the correctness and continuity of the extraction, where the erosion gully map of Wubu County was obtained; 2) In 10 verified watersheds, the extraction was consistent with the reference in general, but the references were better in detail, and the accuracy of the extraction was 81.1%-86.3%, with an average of 83.8%. Therefore, the extraction was supported by Python language, Geospatial Data Abstraction Library (GDAL), OpenCV library, PostGIS database, and Windows 10 operating system. The erosion gully extraction and software were developed for the large-region and high-resolution DEM data with DEM partition, initial extraction, error area, and missing area elimination and splicing as the core in the dynamic monitoring of erosion gully.