Identifying large-scale terrace landscape pattern in the Loess Plateau of China: A new approach based on DEM and land use data

Abstract

The historic significance and development value of terracing is extremely important, which is an unavoidable requirement to ensure national agricultural production and to restore the soil environment. Effective capture of spatial information from large-scale terrace areas can provide data reference and support for macro-planning, timely management and decision making in terrace construction. Overcoming the limitation that remote sensing cannot effectively expand temporal and spatial scales, in this study, we present a novel method for rapidly identifying large-scale spatial patterns of terraces over multiple time periods. Based on the decision tree (DT) model algorithm, this paper took the existing research results as the sampling data of terraces, combined digital elevation model and land use data, and excavated the rules of terrace identification. The spatial distribution of terraces in the Loess Plateau from 1990 to 2020 was then analyzed visually, based on the terrace recognition rules, and its accuracy was checked against the existing literature. Overall, the results show that terrace recognition rules based on the DT model can effectively identify cultivated and grassy terraces with AUC values greater than 0.7. Furthermore, the kappa coefficients for the cultivated and grass terraces, which were 0.92 and 0.86, were in agreement with the sampling data. Between 1990 and 2020, cultivated terrace area decreased and patch distribution remained patchy, as the area of the grassy terraces increased and the plots continued to aggregate. Once verified, the overall effect of the terrace identification results is substantial, and there is no significant departure from the relevant literature. The aim of this study is to explore an efficient identification method to capture information on the spatial distribution of terraces at a macro scale, providing a new approach to quantitative terrace research from a large-scale perspective.