High-resolution numerical climate modelling
Most of the future climate studies hereby described are based on climate simulation results, performed with coupled (CGCM) or forced global climate models.
Since several years, numerous tools have been developed to ease processing and analysis of numerical simulations. In the near future, our team will prepare the next generation of CGCMs to be used for various studies such as medium-range predictions and climate scenarios.
The permanent collaboration between us and researchers and engineers of Centre National de Recherches Météorologiques (CNRM) and Laboratoire d’Océanographie et du Climat- Expérimentation et Analyse Numérique (LOCEAN, CNRS) should be further fostered. In this way, the behaviour of a state-of-the-art medium resolution coupled model will fit the researcher needs and allow an iterative validation of each technical improvement.
We propose now to set up a scientifically validated high resolution configuration of our ARPEGE-NEMO climate model (less than 50Km horizontal resolution for the atmosphere and less than a 1/4 degree for the ocean) to simultaneously prepare vector to scalar computing migration and improve climate model accuracy and its biases reduction.
We noticed that many laboratories are exploring this scientific thematic, in Europe (Max Planck Institute with IFM Geomar, UK Met Office and IPSL) and all over the world (NCAR, Oakridge, L. Livermore NL, NOAA, COLA, U. Tennesse, U. Tokyo). One reason of the high resolution models popularity lies in their capacity to fully take advantage of the newly available scalar supercomputers, without no need to deeply modify their algorithms (ARPEGE cells side can be reduced until 10 to 1Km). Scientifically speaking, higher resolution coupled model are mandatory to simulate flows explicitly down to smaller scales and to capture potential nonlinear interactions between a wider range of spatial and temporal scales of the different component of the Earth system.
At CERFACS, we have been exploring since several years the possibility to build up a coupling system integrating high resolution components (participation to JAMSTEC Earth Simulator - IPSL joint project, ARPEGE-Climat t359 / ORCA 1/12 degrees experimental configuration on NEC SX9). At the same time, we checked components' performances on new generation of scalar machines (IBM Blue Gene/L, Blue Gene/P, SGI Altix ICE ...).
Those experiments should now be concluded by the setting of scientifically validated configurations (and not only working versions initially built for technical testing). At the same time, we will continue to address the technical high-resolution related issues (efficiency and scalability of coupling software, data compression and reduction, fault tolerance ...)
Those coupled models should offer better speed performances than one hour per simulated month, integrate state-of-the-art model components and exhibit smaller biases to observed mean state and variability or at least, equivalent errors, than those of existing models.
Two aspects of model stability at high parallelism level will be investigated, namely a new fault tolerance MPI (i.e. designed to enable the application to continue operation, rather than failing completely, when some part of the computing system fails) and OASIS coupler scalability.
Since several years, numerous tools have been developed to ease processing and analysis of numerical simulations. In the near future, our team will prepare the next generation of CGCMs to be used for various studies such as medium-range predictions and climate scenarios.
The permanent collaboration between us and researchers and engineers of Centre National de Recherches Météorologiques (CNRM) and Laboratoire d’Océanographie et du Climat- Expérimentation et Analyse Numérique (LOCEAN, CNRS) should be further fostered. In this way, the behaviour of a state-of-the-art medium resolution coupled model will fit the researcher needs and allow an iterative validation of each technical improvement.
We propose now to set up a scientifically validated high resolution configuration of our ARPEGE-NEMO climate model (less than 50Km horizontal resolution for the atmosphere and less than a 1/4 degree for the ocean) to simultaneously prepare vector to scalar computing migration and improve climate model accuracy and its biases reduction.
We noticed that many laboratories are exploring this scientific thematic, in Europe (Max Planck Institute with IFM Geomar, UK Met Office and IPSL) and all over the world (NCAR, Oakridge, L. Livermore NL, NOAA, COLA, U. Tennesse, U. Tokyo). One reason of the high resolution models popularity lies in their capacity to fully take advantage of the newly available scalar supercomputers, without no need to deeply modify their algorithms (ARPEGE cells side can be reduced until 10 to 1Km). Scientifically speaking, higher resolution coupled model are mandatory to simulate flows explicitly down to smaller scales and to capture potential nonlinear interactions between a wider range of spatial and temporal scales of the different component of the Earth system.
At CERFACS, we have been exploring since several years the possibility to build up a coupling system integrating high resolution components (participation to JAMSTEC Earth Simulator - IPSL joint project, ARPEGE-Climat t359 / ORCA 1/12 degrees experimental configuration on NEC SX9). At the same time, we checked components' performances on new generation of scalar machines (IBM Blue Gene/L, Blue Gene/P, SGI Altix ICE ...).
Those experiments should now be concluded by the setting of scientifically validated configurations (and not only working versions initially built for technical testing). At the same time, we will continue to address the technical high-resolution related issues (efficiency and scalability of coupling software, data compression and reduction, fault tolerance ...)
Those coupled models should offer better speed performances than one hour per simulated month, integrate state-of-the-art model components and exhibit smaller biases to observed mean state and variability or at least, equivalent errors, than those of existing models.
Two aspects of model stability at high parallelism level will be investigated, namely a new fault tolerance MPI (i.e. designed to enable the application to continue operation, rather than failing completely, when some part of the computing system fails) and OASIS coupler scalability.
Back
