Modelling dry soil thermal conductivity

Abstract

Dry soil thermal conductivity (λdry) is a critical parameter required to estimate soil thermal conductivity with models based on the normalized concept. A great variety of models to estimate λdry are available in literature, but there hasn’t been any systematic evaluation of their performance. This may be attributed to the fact that they are only based on a limited number of measurements and were validated with data from a limited number of soil types. The objectives of this study were to (1) conduct an extensive review of the current available λdry models; (2) collate and compile a large dataset containing λdry measurements of various soil types; and (3) evaluate the performance of available models and to establish new models with wider applications. A total of 48 models from the literature were assessed using a large dataset consisting of more than 350 soils (659 measurements) from 34 published articles. All 48 models performed unsatisfactory with root mean square error (RMSE)>0.09 W m−1 oC-1, Nash-Sutcliff Efficiency (NSE) < 0.49, and Akaike’s information criterion (AIC)>∼-500. We developed eight new empirical thermal conductivity models with a non-linear regression method. These new models outperformed the 48 published models, with RMSE < 0.09 W m−1 oC-1, NSE > 0.49 and AIC < ∼-3210. They can be chosen depending on the availability of soil information (e.g., texture, quartz content, bulk density and porosity).