In situ quantification of soil thermal properties by integrating active-distributed temperature sensing and Bayesian inference

Accurate soil thermal property estimation is vital for geotechnical and geothermal applications, yet conventional methods face low resolution and high uncertainty. This study proposes an in situ framework integrating active-distributed temperature sensing with Bayesian inference. This method improves upon conventional techniques by providing distributed, depth-resolved measurements, utilizing the composite-medium line-source (CMLS) model to mitigate errors from oversimplified thermal models, and quantifying result uncertainty via Bayesian inference. A case study in Kunshan, China, demonstrates that the proposed method can successfully delineate the vertical profile of soil thermal properties from distributed temperature data. It clearly illustrates the relationship between different lithological units and the magnitude of their thermal properties. To enhance applicability, the influence of parameter selection and prior was investigated. The results indicate that more accurate estimates are obtained when the thermal conductivity of the backfill material and the temperature rise across the sensing cable are treated as unknown parameters for joint estimation. Furthermore, employing informative priors improves robustness. The analysis highlights the advantages of variational inference, as the posterior distribution reveals the combined effects of measurement noise and low parameter sensitivity. This study offers a practical tool for in situ thermal measurement, enhancing soil thermal behavior simulations and subsurface investigations.