Missions

##########
# General contexte and objectives

This project is part of MCMET (https://nastar.laplace.enseeiht.fr/pages/anr_mcmet.html), a multi-disciplinary, scientific computing project funded by the Agence National de la Recherche. MCMET is conducted by the EDStar consortium (http://www.edstar.cnrs.fr/prod/fr/), which consists of about fifty researchers working on the physics of energy and climate in various French laboratories.

One challenge raised by the energy transition is the optimal design of facilities with high energy efficiency (solar power plants, buildings...) constrained by meteorological conditions evolving with climate change. This requires fast and versatile scientific computing strategies to solve the direct models as a prerequisite for the optimization.

The scientific goal of MCMET is to develop a novel strategy for simulating complex energy systems by building upon recent advances in Monte Carlo path-sampling methods. The fast ray-tracing techniques used for physically-based rendering (solving the radiative transfer equation to generate an image) have revolutionized the field of movie animation in the last decades. In a multi-disciplinary article recently published in Science Advances, it was demonstrated that this approach can be generalized to the simulation of multi-scale multi-physics models in complex geometries, as long as the models are linear.

MCMET was designed to investigate pathways to extend this capacity to nonlinear models. It is organized in three work packages designed to answer three objectives: The first objective (WP1) is to formulate nonlinear physical models in the path-space paradigm (path integral and associated sampling algorithm) using the null-collision approach. The second objective (WP2) is to develop scientific computation libraries for sampling and analysing these new multi-scale multi-physics path spaces, within the star-engine development environment. The third objective (WP3) is to apply this framework to address specific issues in three domains of application: solar energy, buildings, and climate physics.

The present 18-month research project is at the intersection of climate physics and computer sciences.

The main objectives are:
- upgrade the community code for 3D atmospheric radiative transfer developed at CNRM, htrdr-atmosphere, to state-of-the-art Monte Carlo algorithms that have been concept-proofed in the recent years in EDStar,
- implement a coupled radiation / statistical cloud geometry model and assess its accuracy with respect to other parameterizations used in climate models, using the upgraded htrdr-atmosphere version as a reference.

##########
# Detailed description

One important part of global climate models is the radiation model (or parameterization) that is used to solve radiation propagation throughout the atmosphere. However, it remains extremely difficulty to model radiative transfer in the presence of heterogeneous clouds. Indeed, in typical climate simulations, cloud geometrical properties are only known statistically because temporal and spatial scales of variability of clouds are smaller than a mesh grid. The overall goal of this research project is to improve the models that describe the way in which subgrid clouds impact radiative transfer.

To this end, reference simulation tools have been developed by the TROPICS team and EDStar consortium in previous projects. Detailed, high-fidelity simulations of cloud dynamics are performed with the Meso-NH Large-Eddy Model (http://mesonh.aero.obs-mip.fr). Accurate three-dimensional (3D) radiation can then be simulated in the virtual clouds with htrdr, based on Monte Carlo methods (https://www.meso-star.com/projects/htrdr/htrdr.html). These simulations are currently too expensive to be performed online in climate models but they provide useful reference data to support parameterization development, evaluation and calibration. Yet, as htrdr was originally conceived to study cloud radiative effects, the treatment of gas absorption remains largely approximate and overly memory consuming. In addition, choices that were made early in the conception of htrdr for data partitioning have since then shown their limitations and must be reconsidered. Ideas have been imagined and concept-proofed within EDStar to improve these two aspects but they have not been implemented in htrdr yet. This will be the first objective of this research project.

In another hand, parameterizations that are currently used to model cloud-radiation interactions in climate models suffer from various limitations. One is that radiative transfer is approximated in a two-stream model, which is knowingly inaccurate in clouds because of their large optical depths and highly scattering properties. A second one is that assumptions made on cloud geometry are deeply intricated with the two-stream equations, which makes it difficult to test new ideas on how to represent clouds for radiation. A third one is their inability to represent horizontal transfer of light through cloud sides, thereby neglecting the so-called 3D radiative effects of clouds. In the last two decades, only a few suggestions have been published to deal with these issues. Recently, a proposition has emerged in the EDStar consortium to develop a statistical formulation of radiative transfer that is coupled to an explicit, statistical model of cloud geometry represented as a random field, and to solve this coupled model using Monte Carlo methods. Designing, implementing and evaluating this new model (later refered to as "the coupled model") will be the second aspect of this post-doctoral research project.

This research project is hence twofold, with the following objectives:
- upgrade the community code for 3D atmospheric radiative transfer developed at CNRM, htrdr-atmosphere, to state-of-the-art Monte Carlo algorithms that have been concept-proofed in the recent years in EDStar,
- implement the coupled model and assess its accuracy with respect to other parameterizations used in climate models, using the upgraded htrdr-atmosphere version as a reference.

First, the correlated-k distribution model that is currently used for gaseous absorption will be replaced by the line-by-line model of Nyffenegger-Pere et al, and the octrees currently used to provide access to the volumetric data describing 3D cloud fields will be replaced with a Bounding Volume Hierarchy (BVH). Combining these advances opens research questions that are at the intersection of physics and computer sciences: How to combine spatial and spectral partitioning of gas spectral properties (millions of molecular energy transition lines and varying thermophysical conditions)? What physical criteria should be used to partition the spatially heterogeneous cloud fields in a BVH? How to design an efficient ray-tracing algorithm to propagate rays in volumetric fields described by a BVH?

Then, a first version of the coupled model, which was designed in R. Lebrun's PhD thesis and then in P. de Truchis de Varennes' internship and is based on Markovian Processes, will be implemented in htrdr and evaluated against reference simulations. Depending on the remaining time and on the candidate's wishes, a 3D version of the coupled model might also be developed and implemented. Again, research questions at the intersection of physics and computer sciences arise: How to locally sample a random 3D cloud field according to its spatial-structure model? Can this sampling procedure be formulated in the theoretical framework of stochastic processes? How to define and partition an upper-bound concentration field when the raw concentration field is itself a random field which is never fully sampled?

This research will be conducted in close collaboration with Meso-Star, the company who initially developed htrdr, and with other members of EDStar who are also working on radiative transfer in spectrally and spatially complex media: S. Vinatier's team in planetology at Observatoire de Paris (Univ. PSL), G. Parents' team in fire sciences at Laboratoire Énergies & Mécanique Théorique et Appliquée (Univ. Lorraine), R. Lebrun and N. Mourtaday at Laboratoire d'Optique Atmosphérique (Univ. de Lille).

The recruited researcher will publish results in peer-reviewed journals, present work at national and international conferences or workshops in climate physics and computer graphics. She/he will also participate in the annual seminar of EDStar.

The successful candidate will be recruited for a 18-month contract. Applicants are expected to provide a CV and a cover letter (1 to 2 pages) explaining how their previous work relates to the present project and stating their interest in this position.

N. Villefranque et al. In : Science Advances 8.27 (2022). url : https://www.doi.org/10.1126/sciadv.abp8934.
M. Galtier et al. In : JQSRT 125 (2013). url : https://doi.org/10.1016/j.jqsrt.2013.04.001.
Y. Nyffenegger-Péré et al. Dans : PNAS 121 (2024). url : https://doi.org/10.1073/pnas.2315492121.
N. Villefranque et al. In : JAMES 11.8 (2019). url : https://doi.org/10.1029/2018MS001602.

Activities

- Bibliography work on previous research conducted in EDStar (PhD thesis of Y. Nyffenegger-Pere and R. Lebrun)
- Co-animation of collective working sessions on line-by-line sampling algorithms and BVH ray-tracing algorithms
- Software development (C, star-engine, free software), with the Meso-Star company
- Collective redaction of scientific articles

Explore one or several of the following questions
- What formulation for 3D coupling of radiation and statistical cloud geometry models in 3D?
- What stochastic process / random function? Markovian / Gaussian... ?
- Efficient partitioning of an upper-bound field when the raw field is unknown?
- Partitioning and access to cloud sample data?
- New data structures for volumetric data: an opportunity to couple htrdr to MesoNH?

Skills

General scientific knowledge:
The research conducted within this project sits at the intersection of climate physics and computing sciences, particularly computer graphics. 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 atmospheric physics, radiation, statistical modelling, Monte Carlo methods or computer graphics will be appreciated.

Technical skills:
- Proficiency in Unix/Linux environments
- Ease with scientific programming, ideally with the C programming language
- Basic knowledge of version control systems, particularly Git

Professional skills:
- 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 would be a plus

Interpersonal skills:
- Strong interpersonal and communication skills
- Proven ability to work collaboratively
- Responsiveness and availability

Work Context

The Direction de l'Enseignement Supérieur et de la Recherche (DESR) brings together the research entities of Météo-France (mainly CNRM, SAFIRE and LACy), the National School of Meteorology (ENM) and their shared administrative and IT support services (PGA).

The CNRM is a Joint Research Unit (UMR 3589, www.umr-cnrm.fr) under the joint supervision of Météo-France and CNRS. The CNRM conducts research in the field of meteorology and climate, from the observation, understanding and modelling of processes to the development of weather forecasting and climate projection systems that can be transferred to Météo-France's operational services.

The recruited candidate will join the TROPICS team (TROpical Processes, Intraseasonal variability, Convection and Cloud Studies) of the CNRM's meso-scale meteorology research group (GMME). The team's research is on atmospheric physics with a focus of the processes that dominate tropical climate: convection, clouds, precipitation, radiation. The objectives of this research is (i) to better understand how these processes work, how they interact together and with large scale dynamics, as well as their role in climate variability and extreme events and (ii) to better represent these interactions in atmospheric models at all scales.

Constraints and risks

no