As an important indicator of ecological restoration, soil roughness changes exhibit a synergistic evolutionary relationship with near-surface characteristics such as soil and vegetation. However, the main controlling factors for changes in soil roughness in different ecosystems are still unclear. This study was conducted to clarify the changes in soil roughness in different ecosystems and to decouple and quantify the impact of near-surface characteristics on soil roughness in the Horqin Sandy Land (China). Soil roughness was measured using the structure-from-motion (SfM) technique, showing a pattern of farmland (14.08 mm) > semi-mobile dune (12.59 mm) > meadow (6.19 mm) > woodland (5.67 mm). Redundancy analysis (RDA) results showed that the near-surface characteristics of the dune-meadow cascade ecosystem that explained soil roughness, in descending order of importance, were as follows: sand, soil moisture content (MC), silt, plant stem diameter (SD), clay, pH, soil organic matter (SOM), soil shear resistance (SC), litter thickness (LT), water-stable aggregates (WSA), litter density (LD) and bulk density (BD). Specifically, in the semi-mobile dune ecosystem, the order of total effects on soil roughness was as follows: sand (1.311) > pH (0.447) > clay (0.387). In the woodland ecosystem, SC (0.226), pH (|-0.216|), and LD (0.179) were important factors controlling soil roughness. In the farmland ecosystem, MC (0.380), LT (|-0.378|), and SOM (|-0.336|) were the main controlling factors of soil roughness. In the meadow ecosystem, silt (0.574), SOM (|-0.413|), and clay (|-0.380|) significantly influenced soil roughness. Compared to other machine learning models and stepwise regression models, the k-nearest neighbors (KNN) model could predict soil roughness changes more accurately. The findings of this research are of great significance for ecological restoration and sustainable development of the ecologically fragile areas in northern China.
Read full abstract