Research in computational science and math for geophysics: TDD and IMAGe
 |
|||
|
|
 |
|
|
|
 |
|
|
|
 |
|
|
|
This is a detailed simulation of a rising bubble of hot air and is similar to the small-scale motion of the atmosphere expressed by nonhydrostatic physical equations. This computation is significant because it is found using highly accurate numerical methods and can be implemented on large, massively parallel supercomputers. As models for the Earth's climate increase in spatial resolution, it will be necessary to simulate the complex motions of the atmosphere using nonhydrostatic fluid equations to accurately represent processes such as thunderstorms or the effect of mountainous terrain. |
|
|
|||
|
|
 |
|
|
|
This figure reports the results from testing a Bayesian statistical method to reconstruct past Northern Hemisphere temperatures based on tree rings and other proxy information. For this test case, the true temperature series has been obtained from a numerical experiment with the NCAR paleoclimate model. Figure (a) depicts some synthetic annual data (green) that is similar in quality from a network of tree ring chronologies and derived from the model temperatures, the radiative input to the atomosphere at a global scale (red) and the model temperature over the period of instrumental temperature records. Figures (b) and (c) are reconstructions (red) of the past tempertures using a Bayesian hierarchical statistical model, and can be compared to the true model temperatures. The reconstruction in (b) also uses the radiation series in a simple statistical form and provides a more accurate estimate of temperature than when the tree ring information is used alone (Figure c). Reconstructing past climate using single types of climate proxies may not reproduce behavior at all time scales, such as the results in (c). This example is important for demonstrating the value of including additional covariate information, such as radiation input, when reconstucting past climate. The statistical framework used to solve this problem faciliates incorporating important physical information and can aid in capturing the patterns and cycles of climate over longer periods. |
|
|
|||
The research activities within CISL's Technology Development Division (TDD) and Institute for Mathematics Applied to Geosciences (IMAGe) support scientific computation, numerical modeling, and the statistical analysis of geophysical data and model experiments. This research is important to maintain an innovative computational and modeling facility at NCAR, and more broadly, to lead the geophysics community in adopting new computational and mathematical approches that enhance scientific research. This mission is aligned with NCAR's strategic priority of "Conducting computer science, computational science, applied mathematics, statistics, and numerical methods R&D."
Given this broad priority, the research in CISL must span several disciplines and address computational science at many levels. These include improvements to network flow and scalability of existing codes to different numerical methods for simulating the flow of geophysical fluids. Integrated with the computational science are areas of applied mathematics that include data analysis, models for multiscale processes, and techniques for assimilating data into numerical models. Because these different elements are coordinated through a single lab, there is an easy transfer of technology and ideas from prototypes and theoretical results in IMAGe, to issues of implementation and workflow in TDD, and finally into incorporation as tools and models for the communities served by CISL. There is also a valuable reverse transfer whereby emerging computational capability and data storage spur particular research that takes advantage of these features.
Institute for Mathematics Applied to Geosciences (IMAGe)
For FY2007, the IMAGe program on statistics and numerical models provided a focus for statisticians to build collaborations with geophysical modeling groups, and for modelers to be aquainted with statistical methods for more efficient design and analysis of numerical model experiments. Other important accomplishments are advances in data assimilation and high-order numerical methods. During this period several large numerical experiments of flows with and without magnetic fields were conducted to test aspects of turbulence theory. In addition, a new research direction was taken in reconstructing paleoclimate.
IMAGe plans for FY2008 are continuations of these activities. Some important milestones are incorporating higher-order numerics and conservation into a dynamical core for the NCAR global atmosphere model, combining several different climate proxies into a climate reconstruction, and numerical experiments of turbulent flows for spherical geometry. In the coming year, we will focus on the applications of RBFs to unsteady fluid flows based on incorporating both local node refinement and filtering. The IMAGe Theme-of-the-Year program for FY2008 will be on Geophysical Turbulence Phenomena. Its goal is to synthesize theory, observations, and computation to advance the understanding of turbulence.
Technology development
TDD researchers developed scalable versions of POP and CICE as well as a sequential CCSM coupler that currently runs on the IBM Blue Gene/L (BG/L) system at production resolutions. This is a significant simplification in design and opens the door to ultra-high-resolution coupled climate simulations using very large numbers of processors.
FY2008 plans are continuations of existing work. Some important milestones include deploying a new computer to support our experimental systems research and TeraGrid activities, and developing a service oriented architecture for the geosciences. Further, development will proceed on an ultra-high-resolution configuration of CCSM that is suitable for execution on 10,000 to 30,000 processors.
Earth system modeling infrastructure
One of the major accomplishments of FY2007 was the successful completion of a beta test of a coupled, ESMF-based COAMPS-NCOM application. The atmospheric portion of COAMPS (Coupled Ocean Atmosphere Mesoscale Prediction System) is in operational use at the Navy Fleet Numerical Meteorology and Oceanography Center (FNMOC). Coupling it to NCOM (NRL Coastal Ocean Model) improves the predictive capability of the system at high resolution. The coupling was accomplished within the ESMF component architecture using the ESMF sparse matrix multiply for regridding. The coupled system is expected to be transitioned to FNMOC operations beginning in FY2008.
In FY2008, the ESMF team will publicly release a version of the framework (v3.1.0) that contains wholly redesigned data structures for the representation of grids and fields. These structures allow for the representation and manipulation of rectilinear and curvilinear grids and to a limited extent, multi-patch grids. This release will also enable users to represent a wide variety of grids in index space, and to perform highly scalable redistribution and sparse matrix multiply operations using them.
Visualization and enabling technologies
Scaffolding upon knowledge gained in the VSTO project, we began to investigate emerging ontology-based semantic systems as a core foundation for several activities. Our experiments spanned the Earth System Grid, the Community Data Portal, the Earth System Curator, and CADIS, and the results were very promising, particularly as they relate to the potential for improved human-computer interfaces.
We will continue this R&D work into FY2008, and plan to deliver an initial production release of powerful new knowledge-based tools during the year. In FY2008 we also plan to investigate scalability and performance strategies in the context of our community analysis and visualization tools. This work will span the intertwined areas of data models, aggregation schemes, complex grid topologies, and display models.
Computational science
The Computational Science Section developed a High-Throughput Computing (HTC) service for BG/L. Designed as a high-performance computing platform for large parallel jobs, BG/L previously allocated resources in 64-CPU partitions. HTC allows each of the 2,048 processors in BG/L to run independent programs, making BG/L a candidate platform for more diverse Grid-enabled workflows in use at NCAR. In FY2007, we developed a Grid-enabled interface supporting HTC on BG/L. This new interface accepts single-processor tasks using the Globus Resource Allocation Manager (GRAM), aggregates HTC tasks into BG/L partitions, and requests partition execution using the underlying system scheduler. By separating HTC task aggregation from scheduling, we provide the ability for workflows constructed using standard Grid middleware to run both parallel and serial jobs on BG/L.
Support
Support by funding agencies other than the NSF is indicated in the individual reports in this section of the CISL annual report.