Researcher on Strategies for climate model calibration F/M

Climate models are essential tools for understanding the Earth’s climate system and anticipate its response to various external forcings, particularly those of anthropogenic origin. Their development is a long-term and collective endeavour, that brings together dozens of engineers and researchers over decades. This process encompasses a wide range of disciplines, from computational geophysical fluid dynamics to the representation of physical and biogeochemical processes, as well as model evaluation and validation to name only a few aspects of climate model development.

One of the most fundamental characteristics of the climate system is its continuum of spatial and temporal scales. Yet traditional numerical models used to solve the geophysical fluid dynamics equations must introduce a scale truncation: large scales are explicitly resolved using a discretized version of those equations, while smaller-scale processes are represented through physical parameterizations, which describe the statistical effects of smaller-scale processes on the resolved momentum, energy and water budgets. In addition, some processes – such as radiation or phase changes – are inherently outside the scope of fluid dynamics equation and must be parameterized regardless of model resolution.

Physical parameterizations are conceptual models that distil our current understanding of complex, multiscale processes in the climate system. These parameterizations are typically based on theoretical, empirical, or data-driven formulations, which include a number of parameters. Their values are often poorly constrained by observations. With climate models often involving several dozens of such parameters – and given the high computational cost of each model simulation – their calibration is a major bottleneck in climate model development. This calibration process directly impacts key emergent properties of climate models, including their climate sensitivity.

To address this challenge, the French climate modelling community has partnered with mathematicians to develop a novel, semi-automatic calibration framework (Couvreux et al., 2021). This approach integrates uncertainty quantification, to avoid overfitting, and machine learning, to reduce computational costs. It is based on the history matching technique (Williamson et al., 2013) which differs from traditional optimization methods by identifying the range of acceptable (“tuned”) model configurations rather than a single best-fit solution. This method supports the exploration of alternative plausible climates and enables the quantification of remaining uncertainties once performance targets are met within defined tolerances. Furthermore, the framework can accommodate diverse types of observational and theoretical constraints – from global metrics, to local measurements or diagnostics derived from large-eddy simulations or process models – applied across the entire hierarchy of climate model configurations.

As part of the long term DEPHY initiative, and in particular during the recent High-Tune project, the IPSL and CNRM teams have jointly developed ht-explo, a tool implementing the history matching framework for atmospheric model calibration, starting at the process level through comparisons between single-column model simulations and consistent large-eddy simulations. One of the general objective of the PRACCS-PC6-QUINTET project is to adapt/extend its use to the full coupled climate models. Related to this broad objective, the history matching framework now opens the way to better formulate and investigate scientific questions that have been around for decades:

• How can we define the structural errors of a model?

• Can we understand the relationship between physical parameterizations and emerging properties of the model?

• How does model errors propagates across the hierarchy of model configurations (e.g., 1D, 3D, coupled) or across spatial model resolutions?

• Can properties of coupled ocean-atmosphere models be inferred from their standalone components (atmosphere, land, ocean, sea ice)?

• Can we make a better use of novel statistical or machine-learning methods for model emulation or calibration?

We are seeking a motivated and talented researcher to join the CNRM to investigate one or several questions of this non-exhaustive list. The successful candidate is therefore expected to develop her/his own research project, while contributing to the other following objectives:

1. Consolidate and advance the ht-explo tool to transform it into a core component of model development workflows within the French climate modelling community. This work will be conducted within the QUINTET framework in close collaboration with IPSL partners and mathematicians from D. Williamson’s team. It includes:

• Defining an efficient and transparent calibration strategy using a range of model configurations and a well-defined set of evaluation metrics sufficiently well constrained by observations or other relevant reference data.

• Enhancing the tool documentation, modularity, and portability.

• Adding new features in the tool (e.g., emulation of vector metrics, efficient exploration of small parameter subspaces).

• Developing appropriate diagnostics to better leverage ht-explo results.

2. Contribute to the calibration of the CNRM coupled climate model, within the collaborative framework of the CNRM climate model developers, and in particular to support the preparation of the new generation CNRM climate model that will participate in the upcoming seventh phase of the Climate Model Intercomparison Project (CMIP7). This includes identifying and documenting structural errors of the CNRM climate model.

The recruited researcher will publish results in peer-reviewed journals, present work at national and international conferences or workshops, especially those associated with the TRACCS programme and the QUINTET project. She/he will also have opportunities to supervise Master’s students.

The successful candidate will be recruited for a 3-year contract, renewable one (funding for 6 years in total already secured). Applicants are expected to provide a detailed CV and a statement of purpose (max. 2 pages) explaining their vision for the project, the research questions they would like to explore, and their interest in this position. This document will be central during the selection process.

Couvreux, F., Hourdin, F., Williamson, D., Roehrig, R., Volodina, V., Villefranque, N., Rio, C., Audouin, O., Salter, J., Bazile, E., Brient, F., Favot, F., Honnert, R., Lefebvre, M.-P., Madeleine, J.-B., Rodier, Q., and Xu, W.: Process-Based Climate Model Development Harnessing Machine Learning: I. A Calibration Tool for Parameterization Improvement, J. Adv. Model. Earth Syst., 13, e2020MS002217, https://doi.org/10.1029/2020MS002217, 2021.

Williamson, D., Goldstein, M., Allison, L., Blaker, A., Challenor, P., Jackson, L., and Yamazaki, K.: History matching for exploring and reducing climate model parameter space using observations and a large perturbed physics ensemble, Clim. Dyn., 41, 1703–1729, https://doi.org/10.1007/s00382-013-1896-4, 2013.

General scientific knowledge :

The research conducted within this project sits at the intersection of climate science and statistical science, particularly uncertainty quantification. Applicant must hold a PhD thesis in one of these domains and demonstrate a strong interest – or ideally, experience – in the other. Prior work in climate or statistical modelling, model calibration or evaluation, or uncertainty quantification will be appreciated.

Technical skills :

- Proficiency in Unix/Linux environments

- Solid experience with scientific programming and analysis tools such as Python or R

- Familiarity with NetCDF data handling tools, including NCO and CDO

- Basic knowledge of version control systems, particularly Git

Professional skills :

- Ability to work independently and manage time and tasks effectively

- Strong rigour in scientific development, testing, and analysis

- A demonstrated sense of initiative, motivation, and scientific curiosity

- Fluency in English (spoken and written) at a minimum B2 level

- French proficiency is not required, but any level will be a plus for day-to-day integration

Interpersonal skills :

- Strong interpersonal and communication skills

- Proven ability to work collaboratively: you will be part of the TRACCS community and contribute to its development and dynamism

- Responsiveness and availability

Further information :

• Duration of the contract : 3 years, renewable once (funding for 6 years in total already secured)

• Deadline application : 15/06/2025

• Teleworking friendly : Up to 2 days a week

• Required level of education/ Diploma : PhD

• Required level of experience : Beginner

• Required level in french : Not required

• Email contacts for any further information : aurore.voldoire@meteo.fr and romain.roehrig@meteo.fr

Why join us?

Embark on a stimulating adventure in the service of all, alongside men and women committed to the daily challenges posed to our society by the weather and climate. And enjoy the following benefits: flexible working hours, RTT, telecommuting, administrative restaurant or luncheon voucher, 75% contribution to public transport, contribution to health insurance, sports and cultural associations depending on the site (climbing, gym, pottery, theater, etc.).

Other benefits await you, come and discover them!