Factors controlling spatial variation in soil aggregate stability in a semi-humid watershed.

• Soil properties were the primary controls on SAS variation across the watershed. • Topography can affect SAS by influencing SOC dynamics, allocation of LUT or NDVI. • Grassland restoration was recommended for preventing soil erosion in the study area. • SEM gave new insights into the complex network of SAS drivers and the potential paths. Soil aggregate stability (SAS) is a key soil property that affects soil erosion and soil ability to support ecosystem functions. The effects of different environmental factors on SAS are extensively documented. However, the relative importance of the factors that drive variation in SAS at watershed scale is not entirely clear. To investigate the effects of the interactions of environmental variables on spatial variation in SAS, 88 sampling sites were selected across an entire watershed (1.1 km2) on the Chinese Loess Plateau (CLP), from where undisturbed soil samples were collected at the 0–10 and 10–20 cm soil depths. Three indices were used to evaluate the SAS — water-stable aggregates greater than 0.25 mm (WSA >0.25 , %), mean weight diameter (MWD, mm) and mean geometric diameter (MGD, mm). The results showed that variation of SAS across the watershed was moderate, with coefficient of variation (CV) of 23.5–38.9 %. From combined Spearman's correlation analysis (r), redundancy analysis (RDA) and structural equation modelling (SEM), it was found that soil intrinsic properties, mainly soil texture and organic carbon content (SOC), were the primary control on SAS variation. Topographic attributes, primarily wetness index (TWI) and altitude, were also important controls on SAS. These controls were either the direct or indirect effect through SOC dynamics, spatial distribution of land use (LUT) or vegetation cover (NDVI). The effect of LUT on SAS was mainly driven by SOC and TWI at the 0–10 cm depth but by NDVI and TWI at the 10–20 cm depth. SAS was positively correlated with sand content and SOC, but negatively correlated with silt content, altitude, TWI and NDVI. For LUT, SAS in the apple orchard was significantly lower than in shrubland and grassland, however, it was comparable with that in forest. Considering the effects of improving soil structure and the related economic cost, natural restoration of grassland was a good choice for preventing soil erosion in the study area. The results of this study could deepen our understanding of the controls on SAS variation and therefore become useful in soil management and vegetation restoration decisions on CLP and other regions with similar conditions. [ABSTRACT FROM AUTHOR]