Mapping long-term terrace change in the Chinese Loess Plateau: A new approach by decision tree model with digital elevation model (DEM) and land use

Human-created terraces are distributed extensively in the Chinese Loess Plateau, which play key roles in soil conservation, agricultural production and sustainable development. However, large-scale and long-term terrace mapping remains a big challenge due to the complexity of topography, land cover and the deficiency of high-quality historical spatial data. Facing this task, our study aims to develop a new approach for capturing 30 years (from 1990 to 2020) of terrace patterns at macroscales (the whole Loess Plateau, with an area of 6.4 x 105 km2). The decision tree model (DTM) was integrated with digital elevation model (DEM) and land use data to detect terrace change, and terraced samples were extracted from existing findings for spatial validation. Our study confirmed that this new approach can work successfully on identifying cultivated and grassy terraces, as evidenced by receiver operating characteristic (ROC) curves and area under curve (AUC) values. More notably, a decreasing trend was detected in cultivated terraces with continued uneven distribution from 1990 to 2020, while the areas of grassy terraces increased markedly with more-concentrated larger patches. This finding indicated that huge areas of terrace abandonment may have already occurred in this region. More attention thus should be paid to the rising risks of cropland utilization and food security. Since it is the first time to get long-term reliable terrace maps on the Loess Plateau, our efforts can help to better take stock of terrace resources for wiser land use managements and agricultural policy adjustments, finally benefiting socio-ecological sustainability. This study presents a detection method suitable for macro-scale analysis to accurately identify the spatial pattern of terraces and comprehensively quantify and visualize the benefits derived from terracing services.image