Assessment of soil quality in a heavily fragmented micro-landscape induced by gully erosion

Soil quality degradation induced by erosion significantly inhibits sustainable development worldwide. For assessment of soil quality variations in an area with a heavily fragmented micro-landscape induced by gully erosion, 16 soil quality indicators were tested in laboratory settings and selected by principal component analysis (PCA). Meanwhile, soil quality prediction was conducted by the random forest (RF) model with its quality indicators derived from a 3-dimensional structure of the landscape (resolution, 0.01 m) obtained with an unmanned aerial vehicle (UAV). During RF modelling, 80 % of the Soil Quality Indices (SQIs) estimated by PCA were randomly selected as training data, and the remaining was used to validate the prediction result. The optimal SQIs were shown to include Mn-d, bulk density, silt content, and cation exchange capacity (CEC). Additionally, the PCA-calculated SQI ranging from 0.33 to 0.85 decreased with decreasing elevation in the gully erosional area. Moreover, the spatial soil quality predicted by RF with a satisfied accuracy (R-2 = 0.83 similar to 0.86; RMSE = 0.03 similar to 0.04) was comparable to PCA-calculated SQI. Overall, the spatial variation of soil quality in the gully was attributed to elevation (13.4 similar to 24.1 %), slope gradient (8.0 similar to 13.4 %), relief amplitude (9.8 similar to 12.9 %), and terrain roughness index (10.3 similar to 11.9 %). This study confirmed the excellent performance of RF for SQI prediction, and also indicated that ultra-high-resolution (0.01 m) terrain obtained by unmanned aerial vehicle (UAV) was a competent tool for soil quality assessment in areas with complicated microtopography and limited availability for soil sampling.