Tuning the ICON-A 2.6.4 climate model with machine-learning-based emulators and history matching
<p>In climate model development, “tuning” refers to the important process of adjusting uncertain free parameters of subgrid-scale parameterizations to best match a set of Earth observations, such as the global radiation balance or global cloud cover. This is traditionally a computationally exp...
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Copernicus Publications
2025-06-01
|
| Series: | Geoscientific Model Development |
| Online Access: | https://gmd.copernicus.org/articles/18/3681/2025/gmd-18-3681-2025.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1850208068228349952 |
|---|---|
| author | P. Bonnet L. Pastori M. Schwabe M. Giorgetta F. Iglesias-Suarez V. Eyring V. Eyring |
| author_facet | P. Bonnet L. Pastori M. Schwabe M. Giorgetta F. Iglesias-Suarez V. Eyring V. Eyring |
| author_sort | P. Bonnet |
| collection | DOAJ |
| description | <p>In climate model development, “tuning” refers to the important process of adjusting uncertain free parameters of subgrid-scale parameterizations to best match a set of Earth observations, such as the global radiation balance or global cloud cover. This is traditionally a computationally expensive step as it requires a large number of climate model simulations. This step also becomes more challenging with increasing spatial resolution and complexity of climate models. In addition, the manual tuning relies strongly on expert knowledge and is thus not independently reproducible. To reduce subjectivity and computational demands, tuning methods based on machine learning (ML) have become an active research subject. Here, we build on these developments and apply ML-based tuning to the atmospheric component of the Icosahedral Nonhydrostatic Weather and Climate Model (ICON) at 80 km resolution. Our approach follows a workflow similar to other proposed ML-based tuning methods: (1) creating a perturbed parameter ensemble (PPE) of limited size with randomly selected parameters, (2) fitting an ML-based emulator to the PPE to generate a large emulated ensemble with the emulator, and (3) shrinking the parameter space to regions compatible with observations using a method inspired by history matching. However, in contrast to previous works, we apply a sequential approach: the selected set of tuning parameters is updated in successive phases depending on the results of a sensitivity analysis with Sobol indices. We tune for global radiative properties, cloud properties, zonal wind velocities, and wind stresses on the ocean surface. With one iteration of this method, we achieve a model configuration yielding a global top-of-atmosphere net radiation budget in the range of [0, 1] W m<span class="inline-formula"><sup>−2</sup></span>, and global radiation metrics and water vapour path consistent with the reference observations. Furthermore, the resulting ML-based emulator allows us to identify the parameters that most impact the outputs that we target with tuning. The parameters that we identified to be mostly influential for the physics output metrics are the critical relative humidity in the upper troposphere and the conversion coefficient from cloud water to rain, influencing the radiation metrics and global cloud cover, together with the coefficient of sedimentation velocity of cloud ice, having a strong non-linear influence on all the physics metrics. The existence of non-linear effects further motivates the use of ML-based approaches for parameter tuning in climate models.</p> |
| format | Article |
| id | doaj-art-db3f0061bc5f4affa2130f85d3d87b9b |
| institution | OA Journals |
| issn | 1991-959X 1991-9603 |
| language | English |
| publishDate | 2025-06-01 |
| publisher | Copernicus Publications |
| record_format | Article |
| series | Geoscientific Model Development |
| spelling | doaj-art-db3f0061bc5f4affa2130f85d3d87b9b2025-08-20T02:10:19ZengCopernicus PublicationsGeoscientific Model Development1991-959X1991-96032025-06-01183681370610.5194/gmd-18-3681-2025Tuning the ICON-A 2.6.4 climate model with machine-learning-based emulators and history matchingP. Bonnet0L. Pastori1M. Schwabe2M. Giorgetta3F. Iglesias-Suarez4V. Eyring5V. Eyring6Deutsches Zentrum für Luft- und Raumfahrt (DLR), Institut für Physik der Atmosphäre, Oberpfaffenhofen, GermanyDeutsches Zentrum für Luft- und Raumfahrt (DLR), Institut für Physik der Atmosphäre, Oberpfaffenhofen, GermanyDeutsches Zentrum für Luft- und Raumfahrt (DLR), Institut für Physik der Atmosphäre, Oberpfaffenhofen, GermanyMax Planck Institute for Meteorology Hamburg, Hamburg, GermanyDeutsches Zentrum für Luft- und Raumfahrt (DLR), Institut für Physik der Atmosphäre, Oberpfaffenhofen, GermanyDeutsches Zentrum für Luft- und Raumfahrt (DLR), Institut für Physik der Atmosphäre, Oberpfaffenhofen, GermanyInstitute of Environmental Physics (IUP), University of Bremen, Bremen, Germany<p>In climate model development, “tuning” refers to the important process of adjusting uncertain free parameters of subgrid-scale parameterizations to best match a set of Earth observations, such as the global radiation balance or global cloud cover. This is traditionally a computationally expensive step as it requires a large number of climate model simulations. This step also becomes more challenging with increasing spatial resolution and complexity of climate models. In addition, the manual tuning relies strongly on expert knowledge and is thus not independently reproducible. To reduce subjectivity and computational demands, tuning methods based on machine learning (ML) have become an active research subject. Here, we build on these developments and apply ML-based tuning to the atmospheric component of the Icosahedral Nonhydrostatic Weather and Climate Model (ICON) at 80 km resolution. Our approach follows a workflow similar to other proposed ML-based tuning methods: (1) creating a perturbed parameter ensemble (PPE) of limited size with randomly selected parameters, (2) fitting an ML-based emulator to the PPE to generate a large emulated ensemble with the emulator, and (3) shrinking the parameter space to regions compatible with observations using a method inspired by history matching. However, in contrast to previous works, we apply a sequential approach: the selected set of tuning parameters is updated in successive phases depending on the results of a sensitivity analysis with Sobol indices. We tune for global radiative properties, cloud properties, zonal wind velocities, and wind stresses on the ocean surface. With one iteration of this method, we achieve a model configuration yielding a global top-of-atmosphere net radiation budget in the range of [0, 1] W m<span class="inline-formula"><sup>−2</sup></span>, and global radiation metrics and water vapour path consistent with the reference observations. Furthermore, the resulting ML-based emulator allows us to identify the parameters that most impact the outputs that we target with tuning. The parameters that we identified to be mostly influential for the physics output metrics are the critical relative humidity in the upper troposphere and the conversion coefficient from cloud water to rain, influencing the radiation metrics and global cloud cover, together with the coefficient of sedimentation velocity of cloud ice, having a strong non-linear influence on all the physics metrics. The existence of non-linear effects further motivates the use of ML-based approaches for parameter tuning in climate models.</p>https://gmd.copernicus.org/articles/18/3681/2025/gmd-18-3681-2025.pdf |
| spellingShingle | P. Bonnet L. Pastori M. Schwabe M. Giorgetta F. Iglesias-Suarez V. Eyring V. Eyring Tuning the ICON-A 2.6.4 climate model with machine-learning-based emulators and history matching Geoscientific Model Development |
| title | Tuning the ICON-A 2.6.4 climate model with machine-learning-based emulators and history matching |
| title_full | Tuning the ICON-A 2.6.4 climate model with machine-learning-based emulators and history matching |
| title_fullStr | Tuning the ICON-A 2.6.4 climate model with machine-learning-based emulators and history matching |
| title_full_unstemmed | Tuning the ICON-A 2.6.4 climate model with machine-learning-based emulators and history matching |
| title_short | Tuning the ICON-A 2.6.4 climate model with machine-learning-based emulators and history matching |
| title_sort | tuning the icon a 2 6 4 climate model with machine learning based emulators and history matching |
| url | https://gmd.copernicus.org/articles/18/3681/2025/gmd-18-3681-2025.pdf |
| work_keys_str_mv | AT pbonnet tuningtheicona264climatemodelwithmachinelearningbasedemulatorsandhistorymatching AT lpastori tuningtheicona264climatemodelwithmachinelearningbasedemulatorsandhistorymatching AT mschwabe tuningtheicona264climatemodelwithmachinelearningbasedemulatorsandhistorymatching AT mgiorgetta tuningtheicona264climatemodelwithmachinelearningbasedemulatorsandhistorymatching AT figlesiassuarez tuningtheicona264climatemodelwithmachinelearningbasedemulatorsandhistorymatching AT veyring tuningtheicona264climatemodelwithmachinelearningbasedemulatorsandhistorymatching AT veyring tuningtheicona264climatemodelwithmachinelearningbasedemulatorsandhistorymatching |