Salt effects on soil thermal conductivity: Database construction and impact quantification

Soil thermal conductivity (STC) is a key thermal property that affects transport of heat and water. Although the influences of bulk density, water content, and temperature on STC has been widely reported in previous studies, the effect of various salt types at a wide range of concentrations has been understudied, which leads to unsatisfactory accuracy of many STC predictive models. To quantify the impact of salts on STC, the most comprehensive STC database (to the best of our knowledge) considering salt effects was established in this study. Meanwhile, three machine learning models (i.e., Extreme Gradient Boosting-XGBoost, Random Forest-RF, and Backpropagation Neural Network-BPNN), explainable artificial intelligence (e.g., feature importance ranking and partial dependence plots), and Spearman correlation analysis were employed to quantify the effects of salt types and concentrations on STC. The results show that XGBoost performs best within the full STC range (R 2 = 0.86, RMSE= 0.16 W m −1 K −1 ), while BPNN performs better at greater STC range (e.g., >2.0 W m⁻¹ K⁻¹). STC gradually transforms from significantly negative to significantly positive correlation as NaCl concentration increases from 0∼2 mol L −1 to 4∼6 mol L −1 , but the opposite trend was observed for CaCl 2 . In addition, STC increases with increase of positive temperature, bulk density, water content, and decrease of sub-freezing temperatures.