MN - Modèles Numériques

Peta scale mULti-gridS ocean-ATmosphere coupled simulatIONs – PULSATION

Impact of small-scale phenomena on systematic biases in climate models

Reduction of systematic biases digital climate models explicitly reproducing the small-scale phenomena in certain areas of small extent, but nevertheless crucial for the large-scale climate, using a new type of ocean-atmosphere simulations say «multiscale «on petaflop computer.

Quantify processes determinants small scale for large-scale climate

Understanding the climate is one of the major scientific challenges of this century . Despite the progress achieved in the understanding of the physics of climate, in the power of super computers , the major source of most through climate models is the lack of adequate representation of some small-scale phenomena ( = 10km) . The advent of petaflop machines as Curie in France, changes the game and allows to drastically increase the density of the computational mesh used in climate models and usually limited to ~ 100km . <br />The aim of our project is twofold : to quantify the improvement of simulations based on increasing the spatial resolution of the mesh used and whether it is possible to improve outcomes, at lower cost, by increasing the resolution only some areas are held phenomena known to have a strong impact on the climate scale small scale. We will focus in particular on some ski characterized by deep, cold water to the surface coastal areas , although representing yet less than 1% of the earth's surface , are critical to the climate of the entire belt tropical . <br />The realization of this project will lead to the development and implementation of petaflop computer , a new type of ocean-atmosphere numerical model that will make it possible to reduce the systematic biases, cost , locally increasing horizontal resolution to better reproduce certain phenomena of small -scale determinants for the large-scale climate .

Our approach is to build the first multi- scale ocean-atmosphere numerical model , introducing regional oceanic and atmospheric high-resolution zooms in several key areas of a global climate model. By following this strategy, we will be able to represent the main oceanic and atmospheric dynamic processes of fine -scale critical climatic regions , and allow these regional climates retroactive climate scale .
Computing power Curie allows us to perform a series of simulations in which we will gradually increase the spatial resolution of the model to achieve a spatial resolution of 9km on certain key areas (such as the South East Pacific ) or the entire field of study, which represents a maximum computation overhead compared with 1000 simulations we usually realize. Each component ( ocean, atmosphere and coupler ) of our model is the culmination of nearly two decades of research. They have a massively parallel software architecture and a possibility of mesh refinement to be adapted for this study.
The validation of the model results with observations of temperature , wind, currents, precipitation , etc. ... we will quantify the improvements simulations with increasing horizontal resolution. The comparison of results from simulations benefiting from very high resolution over the entire area or only in coastal areas , indicate , in addition, to what extent the model bias can be reduced by concentrating the computing power only on those key regions.

The first 6 months of the project were devoted to the preparation and participation in the « Grand Challenge Curie « in which we implemented , tested and turned the ocean and atmosphere models in forced mode and coupled to a spatial resolution of 9km ( 27km instead of the initially planned in the first part of the project) . Participation in the « Grand Challenge Curie « is a pilot project which is technical and scientific challenge. Must be met within the same team , advanced skills on the physics of climate modeling and high performance computing to deploy a new machine size Curie simulations sizing far exceeding anything has been done so far . This work required a heavy investment in terms of developments either in code ( new version of the module and server IO in the ocean ) of pre-and post -processing ( generation of initial states and conditions the limits of the ocean and the atmosphere) or the environment ( management , automated recovery and sustainability of simulations). Our efforts were rewarded by the realization of simulations 3 years with very high resolution (9km) who turned over 8000 cores Curie in parallel.
Finesse and detail phenomena reproduced in these first simulations far exceeds the results obtained so far on these simulations brooding over half the globe. The first step marks the beginning of a long process of analysis to understand and interpret these new findings that open , already , many research opportunities .

The development of an ocean-atmosphere numerical model multi-scale offers the unique significantly improve the next generation of climate model while keeping a reasonable computational cost opportunity.
The methodology developed is directly applicable to many other questions which agitate the Climate Research: cyclones, air-sea interaction on a small scale particularly in frontal regions such as the level of the Gulf Stream, location and training deep water that feeds the slow circulation of the ocean across the globe ...

These early studies have been two general items (without peer-reviewed) having the project and its integration into the French landscape of high performance computing. The first article was published in the information APRO Bulletin number 72. The second article will be published in a special issue of «The Search« dedicated to the great challenges that have turned on Curie.
The project and our initial results were also presented to the «international supercomputing conference« 17-21/6/2012 in Hamburg (Germany) and the «13th annual workshop WRF users' 25-29/6/2012 in Boulder (USA).

Climate modelling has become one of the major technical and scientific challenges of the century as a fierce controversy has arisen on the issues of climate change. One of the major caveats of climate simulations, which consist of coupling global ocean and atmospheric models, is the limitation in spatial resolution (~100 km) imposed by the high computing cost. This constrain greatly limits the realism of the physical processes parameterized in the model. The imminent arrival of petascale machines in France offers an inestimable opportunity to the climate modelling community to reduce recurrent biases and limit uncertainties in climate simulations and long-term climate change projection.
In this project, we propose to take up this scientific challenge and explore new pathways toward a better representation of the multi-scale physics that drive climate variability. We are targeting to identify and quantify the key mechanisms (named “upscaling”) by which small-scale localized errors have a knock-on effect onto global climate. Our efforts will focus on key upscaling processes taking place in coastal upwelling areas, which hold the models strongest biases in the Tropics. Two major coastal upwelling regions (the Arabian Sea and the southeastern Pacific) that have great societal impacts, but differ in their characteristics and impacts on climate (El Niño and Monsoon) will focus our interest.
Our approach aims at building a modeling platform for multi-scale ocean-atmosphere coupled simulations, by introducing embedded high–resolution oceanic and atmospheric zooms in key regions of a global climate model. By following this strategy, based on a 2-way nesting procedure, we will be able to represent major fine-scale oceanic and atmospheric dynamical processes in crucial areas, and allow these regional processes to feedback on the climate at global scale. To attain this goal, we will combine state-of-the-art and popular models: NEMO for the ocean, WRF for the atmosphere and OASIS for the coupler. WRF and NEMO are among the very few models able to run at global scale and to incorporate the 2-way embedded zooms functionality. Representing the only two models of the climate community in the PRACE benchmark set, these 2 models are also of particularly interest for the HPC community.
Our methodology will consist in several steps. We will first introduce the regional zooms in only one component (ocean or atmosphere) of the system. The comparison of experiments with and without zooms in the Arabian Sea or in the southeast Pacific will allow explore oceanic or the atmospheric upscaling processes and quantify their impact on the large-scale tropical climate and recurrent models bias. In a second step we will introduce a fully coupled regional model into the global climate model, and investigate the upscaling impact of the regional coupled processes. In a third step, we will increase the spatial resolution of the models global and regional grids to reach peta-scale simulations performed on the Bullx machine “Curie”. We will then investigate the sensitivity of our results to the resolution and quantify the advantages of the multi-scale modelling approach.
To achieve this work, we will need to succeed in several technical challenges: development of a coupling interface specific to the use of embedded zooms, timer for the fully parallelized version of the coupler, implementation and optimisation of a multi-scale coupled model on a peta-scale computer.
The completion of this project will end up in the creation of the first multi-scale ocean-atmosphere coupled modeling platform, which constitutes a bridge between the global and regional approaches. This new tool, designed to explore the impact of the highest spatial resolution not yet approachable by the current climate models and investigate the role of selected upscaling processes, offers an unique opportunity to significantly improve the next generation of climate simulations.

Project coordinator


The author of this summary is the project coordinator, who is responsible for the content of this summary. The ANR declines any responsibility as for its contents.



Help of the ANR 582,666 euros
Beginning and duration of the scientific project: January 2012 - 39 Months

Useful links

Explorez notre base de projets financés



ANR makes available its datasets on funded projects, click here to find more.

Sign up for the latest news:
Subscribe to our newsletter