Multi-scale analysis of soil property variability in Northeast China's black soils using advanced geospatial models

Abstract

Black soils are critical to global agriculture but are increasingly threatened by fertility decline due to intensive land use, particularly in Northeast China. Accurately mapping and understanding the spatial variability of soil properties in these spatially heterogeneous landscapes is vital for sustainable soil management. However, existing models often fail to capture the intricate multi-scale environmental drivers that influence soil dynamics. This study aimed to assess whether geographically weighted artificial neural networks (GWANN) as a localized nonlinear model can more effectively capture the spatial variability of soil organic carbon (SOC), total nitrogen (TN), and total phosphorus (TP) than global models such as artificial neural networks (ANN) and random forest (RF). Additionally, two-dimensional empirical mode decomposition (2D-EMD) and semivariogram analysis were applied to identify scale-dependent variation patterns, alongside variation partitioning analysis to quantify the contributions of climatic, topographic, and biological soil-forming factors. Results showed that GWANN outperformed RF and ANN, achieving reductions in RMSE by 0.063 g/kg and 0.362 g/kg for SOC, 0.013 g/kg and 0.028 g/kg for TN, and 0.005 g/kg and 0.006 g/kg for TP, providing more accurate predictions across all three soil properties. The 2D-EMD analysis revealed that meteorological factors predominantly drive large-scale variability (374–483 km) across all three soil properties. At medium (62–118 km) and small (14–28 km) scales, biological soil factors emerged as the main contributors for SOC and TN, while TP was influenced by meteorological factors at medium scale and by biological soil factors at small scale. Although topographic factors did not dominate at any particular scale, their relative contribution increased at medium and large scales compared to the small scale. This study provides valuable insights for optimizing soil fertility management and promoting sustainable land use practices in black soil regions.