A new model for prediction of soil thermal conductivity

Abstract

This study presents a gene expression programming (GEP) method for determining soil thermal conductivity (k). A large and reliable database containing 444 thermal conductivity measurements was collected and employed to develop the GEP model. Four governing parameters (i.e., porosity, saturation degree, quartz content and temperature) were selected as inputs based on a Pearson correlation analysis. Through the determination of the optimal parameter setting of GEP, a predictive equation was first proposed. Then, the proposed GEP model was evaluated and compared with several previous models. The results indicate the proposed GEP model has higher prediction precision than the other models. In addition, a monotonicity analysis was performed to validate the rationality of the proposed GEP model. According to the monotonicity analysis results, the soil thermal conductivity increases with the increase of saturation degree, quartz content and temperature, and it decreases with the increasing of the porosity. Finally, a parametric analysis was conducted to investigate the significance of the inputs. The results show that the effect rank of these four input variables on soil thermal conductivity is saturation degree > porosity > temperature > quartz content, which can provide a reference for similar projects.