A robust soil quality index for mountain ecosystems driven by XGBoost and network analysis

Soil quality is essential for sustainable agriculture, environmental conservation, and food security. Mountain soils are vulnerable to erosion, water and soil loss, and other disturbances. Sampling in mountainous areas remains challenging. There is a gap in accurate, efficient, and convenient soil quality evaluation. Uncertainty persists in selecting indicators that best reflect sustainable outcomes and the ecological functions of mountain soils. This is critical for preventing degradation. This study selects the mountainous region of the Yunnan-Guizhou Plateau as the research area, combining soil physical, chemical, and biological properties, and using Total Data Set (TDS) and Minimal Data Set (MDS) methods, incorporating soil quality indices with and without weights (SQI_w and SQI_nw), to assess mountain soil quality. We used Principal Component Analysis (PCA), Network Analysis (NA), and machine learning algorithms (XGBoost) to select different soil quality indicators and weights. Notably, these methods consistently identified key indicators such as β-1,4-glucosidase (BG), available potassium (AK), soil organic matter (SOM), microbial biomass carbon (MBC), microbial biomass nitrogen (MBN), nickel (Ni), and arsenic (As), underscoring the critical roles of biological activity, nutrient availability (chemical), and heavy metal concentrations in mountain soil quality. We developed 12 SQI indices, namely: MDS-L-W-SQI_PCA, MDS-NL-W-SQI_PCA, MDS-L-NW-SQI_PCA, MDS-NL-NW-SQI_PCA, MDS-L-W-SQI_NA, MDS-NL-W-SQI_NA, MDS-L-NW-SQI_NA, MDS-NL-NW-SQI_NA, MDS-L-W-SQI_XGBoost, MDS-NL-W-SQI_XGBoost, MDS-L-NW-SQI_XGBoost, and MDS-NL-NW-SQI_XGBoost, and then compared different methods to select the best-performing model for mountain soil evaluation. The results show that the MDS-SQI constructed based on XGBoost has superior performance. The R² values for the TDS-SQI fit were 0.83, 0.85, 0.76, and 0.87. Fewer indicators improve efficiency while maintaining accuracy. The NA-based model performs better than the PCA-based one. Nonlinear scoring methods show higher sensitivity and better adapt to mountain soils. Both NA and XGBoost-based MDS-SQI exhibit higher SI and ER values. This capability is crucial for the Yunnan Jinsha River Basin, where hydropower development is ongoing. The predictive insights support soil management and address the current lack of evaluation frameworks in mountainous areas. This method reduces ambiguity in indicator selection and eclipsing errors. It provides a more scientific and reproducible framework for mountain soil quality evaluation.