Quantitative effects of soil organic matter on thermal conductivity modeling

• A large dataset consists of 1569 measurements from 208 soil samples was compiled. • Sensitivity of 15 models and 5 soil organic matter (SOM) levels were evaluated. • The normalized model is optimal for predicting thermal conductivity (λ). • λ is largely dependent on bulk density and water content. • The effect of SOM on λ modeling is minimal with SOM content ≤ 2.5 %. Soil thermal conductivity ( λ ) is a crucial parameter for energy transfer between the land and the atmosphere. An increasing number of studies have noted that changes in soil organic matter content ( ϕ som ) considerably influence λ . However, only a few studies have focused on quantifying the effects of soil organic matter (SOM) on λ modeling. In this study, the performances of 15 λ models that consider the effects of SOM were evaluated based on a compiled dataset consisting of 1569 measurements from 208 soils (0 %≤ ϕ som ≤40 %, mass content). Two random forest-based explainable artificial intelligence models, Shapley additive explanations (SHAP) and permutation importance (PI), were also applied to assess the SOM effects. The results showed that the model developed from normalizing Kersten function had the best prediction accuracy because of its realistic function types and parameter settings. The models of Johansen (1975), Balland and Arp (2005), Su et al. (2016), and Yan et al. (2019) performed the best, with NSE ≥ 0.71 and RMSE ≤ 0.3 W m -1 K -1 . Moreover, PI and SHAP demonstrate that the effect of SOM on λ is nonlinear, and the influence of SOM should be considered when ϕ som >2.5 %. There is a strong interaction between SOM and bulk density when ϕ som >2.5 %, significantly influencing λ .