Insights into the prediction uncertainty of machine-learning-based digital soil mapping through a local attribution approach

Rohmer, Jeremy; Belbeze, Stephane; Guyonnet, Dominique

doi:https://doi.org/10.5194/soil-10-679-2024

Articles | Volume 10, issue 2

https://doi.org/10.5194/soil-10-679-2024

© Author(s) 2024. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/soil-10-679-2024

© Author(s) 2024. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 10, issue 2

Original research article

|

30 Sep 2024

Original research article |

| 30 Sep 2024

Insights into the prediction uncertainty of machine-learning-based digital soil mapping through a local attribution approach

Jeremy Rohmer, Stephane Belbeze, and Dominique Guyonnet

Supplement

https://doi.org/10.5194/soil-10-679-2024-supplement

Data sets

Data to run the synthetic test case Hannah Meyer https://github.com/HannaMeyer/CAST/tree/master/inst/extdata

Model code and software

R script for computing group SHAPLEY dedicated to prediction uncertainty Jeremy Rohmer https://doi.org/10.5281/zenodo.13838496

Short summary

Machine learning (ML) models have become key ingredients for digital soil mapping. To explain why the ML model is reliable, we apply a popular method from explainable artificial intelligence to the uncertainty prediction, with an application to the mapping of hydrocarbon pollutants on urban soil. We show the benefit of a joint analysis of the influence on the best estimate and the uncertainty to improve communication with end users and support decisions regarding covariates’ characterisation.