 |
|
 |
||||||
|
||||||
 |
Home
| Mission
|
about SciDAC
|
Contact Us |
|||||
 |
 |
 |
||||
|
Alumni ProjectMulti-Resolution Climate ModelingFerdinand Baer, Aimé Fournier, Univ. of MDJoseph J. Tribbia, NCAR Michael Fox-Rabinovitz, Univ. of MD, Mark Taylor, LANL; Affiliates SummaryThe purpose of this project is to create a global climate model with features that will provide the best available climate predictions, predictions that can be used effectively to assist in making useful decisions on issues related to climate on regional as well as global scales. It is noteworthy that regional climate events may be the predominant manifestations of global climate change, and that prediction on the regional scale is essential to the understanding of overall global change. Concurrent with the development of a model that has the capacity to predict climate "seamlessly" across various scales over the Earth, one must recognize the inherent variability of a model. This variability is only overcome by determining model climate predictions from statistics of a number of realizations of the model climate. The model computations must therefore be as rapid as possible, to provide those statistics in a practically useful time frame, given the time constraints large models put on limited computing facilities. The proposed model will be uniquely structured to optimize the time required on state-of-the-art supercomputers. A large community of global climate modelers and model contributors are at work today to create an optimum model to generate reliable climate predictions. Major needs for improvement include a more transparent, efficient and accurate method of producing regional climate predictions involving mesh refinement (in this regard see the collateral study by M. Fox-Rabinovitz), and application of a computing methodology which uses the latest in computing hardware (MPP) most effectively and economically, to produce the best prediction results with the minimal expenditure of resources. Our model construction methodology is ideally suited to satisfy the above requirements. We tile the spherical domain with elements (called spectral elements) that can be sized arbitrarily to meet local scaling requirements. The model can produce predictions on a range of scales over the entire global domain, without any user interference in the computational process. The method also takes optimum advantage of parallel processing computers (the present state and future evolution of supercomputing) by minimizing communications amongst the elements and thus also amongst the computer processors since each element (contiguous group) is assigned to a processor. This procedure has yielded dramatic speedup of the dynamics computations, will similarly speed up the full general-circulation model (GCM), and will make the production of multiple realizations more feasible. We anticipate that the method will also provide a more accurate prediction. Additionally the model will serve as a research and training tool. Since many of the forces that drive the climate system are not yet completely understood, the model can and will be used as an experimental environment to test various hypotheses on system forcing, especially in the area of spatially localized scale interactions. Knowledge gained in this process will help to refine the model, allowing it to produce improved predictions. We have thoroughly tested the dynamical core of SEAM, the principal component of the model, with very simple global forcing (denoted H-S forcing) and have found it to run stably and as expected. We have added topography to the model, and have tested this using Local Mesh Refinement (LMR) to test the ability of SEAM to accept the imbedding of a local fine scale grid. Experiments incorporating the Andes were successful. To test this version of the model against other similar experiments, we ran with high resolution (LMR) over the continental U.S.A. This experiment was first done and published by Fox-Rabinovitz using a stretched grid with the NASA climate model, a project currently funded by the SciDAC/CCPP. To carefully test the model, we ran integrations for three years starting from a resting atmosphere and analyzed the climate statistics of the third year. We ran three mesh size cases (fine, medium and course resolution) on a uniform grid with and without topography and repeated some of these runs with LMR over the USA to capture the fine resolution of the global grid. These experiments demonstrated the advantage of LMR, especially over the mountains, and were in reasonable agreement with the results presented by Fox-Rabinovitz. An interesting sidelight to this study was that the same integrations run using both Fortran 77 and Fortran90 on different machines gave somewhat different results, highlighting the significance of machine operating systems in creating codes. For consistency with the community effort in climate modeling, we have selected to build into SEAM the full physics package available from the Community Atmosphere Model (CAM) developed at NCAR with both University and Federal Agency (DOE, NSF, NOAA, etc.) collaboration. This package comes with a ‘coupler’, a program that is structured to accept a dynamical core (DC) model and link it to the physics package. Although there are now three DCs linked and available for experiments at NCAR, SEAM proved to be more difficult to link because of the unstructured nature of its grid. We are however at the last stages of this coupling and are testing the full model at this time. The following tests are either scheduled or are in progress. Both CAM/EUL (NCAR Eurlerian spectral DC) and CAM/SEAM will be run for five days without physics (dynamical core with H-S forcing), with topography and real initial conditions and for five days. As soon as the physics is coupled to SEAM, the same experiment will be run with full physics. Since CAM/EUL has been used extensively, this should be a good test for SEACM. For the upcoming years, tests with various forms of LMR will be explored with SEACM and comparison studies will be performed with Fox-Rabinovitz using his stretched grid model. Additionally, SECAM with its modified version of the coupler will be made available to the CAM program both for experimentation, and improvement. The computations in SEACM are tuned to MPP architecture and make highly efficient use of it. Since climate integrations require the most powerful available machines, it is essential that we have access to the facilities at NERSC and ORNL, both of which we currently use. We welcome the collaboration of other SciDAC researchers in making our calculations even more efficient. For further information on this subject contact:Prof. Ferdinand Baer
|
||||
Home | ASCR | Contact Us | DOE disclaimer |
|
|
