Soil Thermal Conductivity AI

This software is continuously updated by the researchers and programmers at Umny Inc. to further improve its accuracy and speed. If you have questions, feedback, or experimental data you would like to test, please feel free to send us an email.

Accuracy of Autofill Estimates

When inputs are not all manually input by the user, they are generated using predictive algorithms and databases to generate their inputs. The accuracy of the two in use for this version of the software can be found at the links below.
Soil Solids Property
Ground Temperature
This predictor can estimate the following soil properties for any location on Earth:
  • Bulk Density
  • Organic Carbon
  • Coarse Particles/Rocks
  • Sand
  • Silt
  • Clay


Due to a lack of soil property data at deep depths, we have not been able to do a full accuracy study of our Soil Solids Property estimator. In our internal testing we have found cases where the combined inaccuracies of the predictions for all 6 soil properties can produce errors in the final Thermal Conductivity result of 0.5 - 1 W/mK. However, using our Soil Solids Property estimator is still going to give you much more accurate results than a random guess or using a "standard" soil profile.


Our estimates of soil properties are produced by looking at the ISRIC's SoilGrids 2.0 data for the given location, which only covers the top 2 meters of the soil. We then use a piecewise function to extrapolate that surface data down to deeper depths, drawing also on the ISRIC's Depth to Bedrock dataset as a variable in determining the shape of that piecewise function. The primary source of inacurracy in our method is that our extrapolation function does not account for any potential irregular stratigraphic layers within the deep soil profile (this is simply because this stratigraphy data does not exist for 90% of the Earth's surface).
Because our Soil Solids Property predictor relies heavily on the data from the ISRIC's SoilGrids 2.0 dataset. The accuracy of our estimates will be nearly identical to the accuracy of that dataset at shallow depths (with more uncertainty as the depth increases).
Our predictions also rely on the ISRIC's Depth to Bedrock dataset.

Recommended Uses:

For many areas of the world which are lacking good soil data, our soil property predictions are the best available estimates. Our predictor is also the fastest method for getting soil property estimates, and so can be useful if speed and cost are your main priorities.

Alternative Methods:

Besides direct measurement of the soil the best alternative method for estimating the physical properties of your target soil is by referencing borehole/well data near your desired location. If you are not an expert on soil properties you may have to hire one in order interpret the borehole/well data.

Accuracy of Thermal Conductivity Predictions

Comparison to Best Analytical Models

The current best analytical models available for predicting Soil Thermal Conductivity can be found in“Evaluation of soil thermal conductivity models”.This chart compares the accuracy of our SoK predictor to the 3 models evaluated in that study. As you can see our model is significantly more accurate accross all soil types than the best analytical models available (and its much easier to use!).
barry-macaulay, cote and konrad, lu et al., balland and arp, soil thermal conductivity comparison

Average and Worst Prediction Errors by Soil Characteristics

Umny SoK Soil Texture Triangle, Thermal Conductivity

Soil Texture

Soil Texture Thermal Conductivity Accuracy SoK
Sand Silt Rock Clay Thermal Conductivity SoK Umny
More than 50% particles are USDA Sand
More than 50% particles are USDA Silt
More than 40% particles are USDA Clay
None of the above apply

Saturation %

Soil Saturation Thermal Conductivity Accuracy SoK
Saturation % W/mK Thermal Conductivity SoK Umny
No Sat.:
Saturation less than 1%
Low Sat.:
Saturation between 1% and 25%
Med Sat.:
Saturation between 25% and 50%
High Sat.:
Saturation between 50% and 75%
Full Sat.:
Saturation above 75%

Rock Content

Rock Coarse Particle % vs W/mK Thermal Conductivity Accuracy SoK
Saturation % W/mK Thermal Conductivity SoK Umny
No Rocks:
No particles larger than 2mm in soil
Some Rocks:
Rocks/Coarse Particles between 0% and 50% of soil volume
Mostly Rocks:
Rocks/Coars Particles above 50% of soil volume

Organic Carbon Content

Soil Organic Carbon g/cm3 Thermal Conductivity Accuracy SoK
Organic Matter Carbon % W/mK Thermal Conductivity SoK Umny
Low Carbon:
Organic Carbon density less than 0.002 g/cm3
Med. Carbon:
Organic Carbon density between 0.002 and 0.02 g/cm3
High Carbon:
Organic Carbon density above 0.02 g/cm3

Soil Temperature

Ground Temperature Soil degrees Celcius vs thermal conductivity accuracy
Machine Learning Error ground temperature vs thermal conductivity
Soil temperature above 1 °C
Transition Zone:
Soil temperature between 1 and -3 °C
Soil temperature below -3°C

Comparison Charts: Measured vs SoK Predicted Soil Thermal Conductivity Values

The following graphs plot the measured versus GTP predicted ground temperatures across the validation dataset. These datapoints are organized by geographic zone, and are chosen from a wide range of research and government institutes. Numerical values for this validation dataset, as well as mapped locations of their test sites are presented below.
sok accuracy in sandy soils divided by organic mattersoil thermal conductivity in sandy soils by total soil saturationthermal conductivity measured vs predicted for silty soils by carbon levelsthermal conductivity of soil prediction accuracy for silt dry and wetthermal conductivity accuracy by carbon, clay and rocksoil thermal conductivity clay, gravel, dry and saturatedsoil thermal conductivity accuracy in loam soils by carbon contentsoil k accuracy in loam vs water saturation

Download Raw Data from our Validation Tests