Identifying terraces in the hilly and gully regions of the Loess Plateau in China

Terrace identification is the basis for understanding quantity, quality, and land covers of terraces and their effects on agriculture production and various surface processes, for example, hydrological and ecological processes. However, there are some drawbacks and limitations in the automatic extraction of terraces, such as difficulty of outlining the overall boundary for individual terrace and the limitation of applying methods and parameters. In this study, we used machine learning methods on the basis of Geographic Object-Based Image Analysis using K-nearest neighbours and spectral angle mapper algorithms to extract terraces in the hilly and gully regions on the Loess Plateau, China. This study relied on medium-resolution image Landsat-8 combined with Advanced Spaceborne Thermal Emission and Reflection Radiometer Global Digital Elevation Model to extract the overall boundary of terraces and relies on high-resolution images GaoFen-1 to extract the terraced edges of terraces. It used the methods of principal component analysis and Laplacian convolution filter to enhance and extract terraced edges inside of the boundary of terraces. Our estimates are with overall accuracy of 62.2% in K-nearest neighbours and 74.8% in spectral angle mapper methods, indicating the advantages of the proposed method despite the use of much lower resolution data than previous studies that used both high-resolution terrain and remote sensing imageries data. This study highlights the importance of using appropriate methods plus reasonable spatial resolution of remote sensing data for outlining the overall boundary of terraces in the hilly and gully regions.