Nano clay mitigation of wind induced soil loss across aggregate sizes through wind tunnel experiments and explainable machine learning

Abstract

This study aimed to investigate, under wind-tunnel conditions, the effects of different nano-clay application rates (0, 0.94, 1.88, and 3.75 g m⁻²) on wind-induced soil loss across different aggregate sizes (0.5, 1.0, and 2.0 mm) and wind speeds (8, 10, and 12 m s⁻¹). In addition, the study aimed to model wind-induced soil loss processes using machine learning and explainable artificial intelligence methods. The wind-tunnel experiments revealed that nano-clay applications significantly reduced soil loss across all aggregate sizes and wind-speed conditions. The highest protective effect was obtained at the 3.75 g m⁻² nano-clay rate; soil loss decreased by 80.8% in the 2 mm aggregates, by 93.2% in the 1 mm aggregates, and by 95.9% in the 0.5 mm aggregates. The protective effect of the nano-clay applications was found to be particularly pronounced at high wind speeds and in fine aggregates. Among the evaluated models, XGBoost showed the strongest development-stage performance under repeated nested group-aware cross-validation (R² = 0.835 ± 0.064, RMSE = 11.896 ± 2.229, MAE = 6.686 ± 0.854) and was selected before independent holdout evaluation. On the group-held-out test set, the selected XGBoost model achieved R² = 0.882, RMSE = 13.298, and MAE = 9.443. The lower performance of the linear baseline suggested that the tree-based models captured predictive structure not adequately represented by a simple linear formulation. SHAP analysis indicated that nano-clay dose made the largest mean absolute contribution to the fitted model predictions, followed by wind speed and aggregate size. Overall, nano-clay substantially reduced wind-induced soil loss within the controlled experimental conditions examined here; broader practical applicability requires validation under field conditions.