5
$\begingroup$

Context: We perform a lot of CFD (Computational Fluid Dynamics) modelling for our clients (geothermal, construction, general underground engineering, etc) where we have to model temperature flows and changes in a soil medium. In this type of modelling small changes in the thermal conductivity value of the soil medium have dramatic effects on the results of the model.

We've identified inaccuracies in our model's soil thermal conductivity value (when compared to the actual thermal conductivity of the real-world soil we are trying to model) as one of the primary sources of error in our models.

Question: What is the most accurate method for estimating the thermal conductivity of a target real-world soil, besides costly direct physical measurement? We can generally get estimates of many soil properties for our target soil from geo-tech studies, such as texture, carbon, rock content, etc. so these can potentially be used as inputs.

Improve this question
$\endgroup$
1
  • $\begingroup$ Moisture levels and mineral identification and content is bound to be important $\endgroup$
    – Stian
    Commented Sep 27, 2021 at 4:15

2 Answers 2

Reset to default
9
$\begingroup$

I don't think this is something that you can model accurately in an ab initio manner, as there is no good method for predicting the thermal conductivity of amorphous materials yet, let alone materials that are both amorphous and have macroscopic inhomogeneities. IMHO the way to go is to collect experimental thermal conductivity data on soil samples with different properties, and train a Bayesian model against the data (for example a simple kriging may suffice).

Improve this answer
$\endgroup$
1
  • $\begingroup$ Thank you, that is exactly what we've done (with a machine learning Neural Network model, instead of a Bayesian), as I describe in my answer. I'm wondering if there are others who have made models that might be better than ours, and I'm just wondering what techniques other modelers use. $\endgroup$
    – Sheldonfrith
    Commented Sep 26, 2021 at 18:15
6
$\begingroup$

Our Method

We have created a machine learning model for estimating the thermal conductivity of any soil, which we have made publicly available at soilconductivity.com.

Our model is significantly more accurate than any other existing method that we are aware of (MAE of around 0.08 W/mK), besides direct physical measurement of the soil, of course. (here is a link to a detailed comparison of our model's accuracy to the best available models described in Evaluation of Soil Thermal Conductivity Models.

Our model DOES require information about the soil. Specifically you need to know the soil texture, carbon content, coarse-particle/gravel content, rough temperature (is it frozen or thawed), and moisture. However these properties are much easier and cheaper to find than a direct measurement of Soil Thermal Conductivity. Additionally, for an even cheaper and quicker option, our website actually includes prediction tools to estimate all of these soil properties from a given geographic location and depth.

Improve this answer
$\endgroup$
2
  • $\begingroup$ I highly recommend not to checkmark your own answer for something like this. Just be happy that it got 6 upvotes and let the community decide which answer they like bette. Otherwise it looks like spam, and it does more harm than benefit to your soilconductivity website. $\endgroup$
    – Nike Dattani - No Free Time
    Commented Sep 30, 2021 at 21:09
  • $\begingroup$ Sorry, have removed the checkmark. $\endgroup$
    – Sheldonfrith
    Commented Oct 2, 2021 at 23:11

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .