A study on the multifunctional assessment and spatial differentiation of soil in the Jianghan Plain based on machine learning

The assessment of soil multifunctionality is crucial for regional ecological security and sustainable development. This study aimed to construct a comprehensive soil multifunctionality assessment framework for the Jianghan Plain and elucidate its spatial distribution patterns. Principal component analysis was employed to screen 14 core indicators from 27 soil parameters, establishing a Minimum Data Set (MDS). Based on five key soil functions (primary productivity, water purification and regulation, climate regulation and carbon sequestration, biodiversity maintenance, and nutrient supply and cycling), predictive models were developed using five machine learning algorithms: Random Forest, Support Vector Machine (SVM), XGBoost, Neural Network, and Gradient Boosting Machine. Spatial distribution characteristics of soil multifunctionality were analyzed using the Getis-Ord Gi* method. The results revealed that: (1) The MDS achieved a 48.15% dimensionality reduction rate with a correlation coefficient of 0.816 (R²= 0.667) compared to the full dataset, retaining 91.72% of the original information while significantly improving assessment efficiency. (2) Different machine learning algorithms demonstrated varying advantages in predicting specific soil functions. XGBoost performed optimally for primary productivity (R²= 0.9797) and biodiversity maintenance (R²= 0.8722) predictions, while SVM excelled in water purification and regulation (R²= 0.9392) and nutrient supply and cycling (R²= 0.8025) predictions. Neural Network showed superior performance in climate regulation and carbon sequestration (R² = 0.9599) prediction. (3) Soil function spatial distribution exhibited a distinctive pattern characterized as "low in the north, high in the south, and moderate in the central region." Soil organic carbon density and clay content emerged as the primary driving factors. Multifunctional hotspots accounted for only 1.03% of the study area, while edge-significant regions comprised 25.71%, indicating substantial trade-offs among soil functions. This research establishes a machine learning-based technical framework for soil multifunctionality assessment, thereby providing scientific support for precision soil function management and ecological conservation strategy development in the Jianghan Plain.