Using feature importance as an exploratory data analysis  tool on Earth system models

Ries, Daniel; Goode, Katherine; McClernon, Kellie; Hillman, Benjamin

doi:https://doi.org/10.5194/gmd-18-1041-2025

Articles | Volume 18, issue 4

https://doi.org/10.5194/gmd-18-1041-2025

© Author(s) 2025. This work is distributed under
the Creative Commons Attribution 4.0 License.

Special issue:

Theoretical and computational aspects of ensemble design,...

https://doi.org/10.5194/gmd-18-1041-2025

© Author(s) 2025. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 18, issue 4

Development and technical paper

|

24 Feb 2025

Development and technical paper |

| 24 Feb 2025

Using feature importance as an exploratory data analysis tool on Earth system models

Daniel Ries, Katherine Goode, Kellie McClernon, and Benjamin Hillman

Data sets

Code and Data Supplement for Using feature importance as exploratory data analysis tool on earth system models Daniel Ries, Katherine Goode, Kellie McClernon, Benjamin Hillman https://doi.org/10.5281/zenodo.12169923

Model code and software

Code and Data Supplement for Using feature importance as exploratory data analysis tool on earth system models Daniel Ries, Katherine Goode, Kellie McClernon, Benjamin Hillman https://doi.org/10.5281/zenodo.12169923

Short summary

Machine learning has advanced research in the climate science domain, but its models are difficult to understand. In order to understand the impacts and consequences of climate interventions such as stratospheric aerosol injection, complex models are often necessary. We use a case study to illustrate how we can understand the inner workings of a complex model. We present this technique as an exploratory tool that can be used to quickly discover and assess relationships in complex climate data.