Assessing wheat yield response to soil compaction using machine learning

— A cross-validated study of soil and nutrient dynamics for precision agriculture —

抄録

This study evaluated the impact of soil compaction, expressed as bulk density (BD), on spring wheat using field trials with vibratory plate-compacted subplots (BD: 1.17–1.30 Mg/m³). Key soil properties, nutrient uptake, and yield were measured under varying BD. PLSR with cross-validation identified key predictors and was compared with MLR/LR for yield modeling. Moderate compaction (BD3: 1.24 Mg/m³) optimized nutrient utilization, producing 173.6 % more yield than highly compacted soil (BD1). The Control subplots (BD: 1.17 Mg/m³) yielded 143.4 % more than BD1. Spatial analysis showed that combined phosphorus–calcium–magnesium dynamics explained 97.2 % of yield variation (R² = 0.972). Nutrient maps supported precision fertilization planning, showing that soil compaction significantly affects wheat productivity through multiple pathways.