NCAR's HPC climate modelling means of preventing harmful impact of global warming

3 Jun 2010 Hamburg - Dr. Lawrence Buja is working at the National Center for Atmospheric Research (NCAR) in the United States as the Director of the Climate Science and Applications Programme. At ISC'10 he was the third speaker in the session on "Supercomputers for Modelling the Climate and the Roles of Energy Production in Climate Change". He warned the audience that global warming has definitely been observed. The main cause is human activity and people are being affected harmfully by it. This is likely to grow to higher levels which will drive us to extensive adaptation and mitigation.

At the NCAR researchers are heavily occupied with Earth system modellling. There are four NCAR/NSF scientific facilities, including the Earth Observing Laboratory (EOL), the NCAR Earth System Laboratory; the Computation and Information Systems Laboratory (CISL), and the Research Applications Laboratory (RAL).

In former years the weather versus the climate was the topic of study, but currently climate prediction is being focused to looking several years ahead by observing the ocean and the atmosphere, Dr. Buja explained.

This is done by means of the Community Climate System Model (CCSM) which is used to investigate and predict seasonal and interannual variability in the climate, to explore the history of Earth's climate, to estimate the future of the environment for policy formulation, and to contribute to assessments.

In this kind of scientific research, collaborations are critical. Thus, within NCAR there has been founded a

CCSM scientific steering committee and an Advisory Board to set out the guidelines in the research with regard to climate change, the paleoclimate, the climate variability, the societal impacts, and the software engineering needed for the study of these topics.

The software requirements for this type of research are vast There is the need for a large code base, involving 1 million lines of Fortran 90 developed over more than 20 years. The researchers have to be able to work with a userfriendly component parameterization. The model system permits each component to be developed and tested independently on even one processor, if necessary.

A further requirement, according to Dr. Buja, is scientific consistency. A one code base as a "stand-aolone" development code is the same as a fully coupled model. In addition, the design provides extensibility to add new components and new coupling strategies. And the model system has to be performant, efficient and enable easy porting.

The Community Climate System Model provides a scalable and flexible coupling infrastructure, There is parallel I/O throughout model system and a scalable memory for each component. And the capability exists to use both MPI and OpenMPI.

The new CCSM4 architecture has many benefits, Dr. Buja gave a load balancing example that proved the reduced idle time and showed an increase core count for POP.

The Ensemble members run concurrently on non-overlapping processor sets and the communication with coupler takes place serially over the Ensemble members. The cubed-sphere grid overcomes dycore scalability problems inherent with lat/lon grid.

The CCSM is moving to operational petascale now, announced Dr. Buja. The Earth system modelling is at eddy-resolving scales and the climate and regional model "snap shots" have achieved the cloud resolving scale. The system allows for seamless downscaling of integrated weather and climate modelling.

The systems now have 5-30K PEs and researchers are seeing success porting to these platforms. A balanced system is essential and has to include scalable numerics and ananlysis techniques, and robust and fault-tolerant communication frameworks.

Cutting-edge HPC platforms have proven to be very fragile and talented people for HPC code development are the limiting resource. ParallelIO is eliminating all serial code.

The HPC dimensions of climate prediction are guaranteeing new and better science with a set timescale,

data assimilation, spatial resolution and Ensemble size, Dr. Buja explained.

He showed how the North Atlantic and the North American regional climate changes. The goal is to simulate the effects of climate change by improving the predictions.

Researchers are investigating whether anthropogenic climate change is occurring and what the impact of anthropogenic climate change on coupled human and natural systems will be. They are addressing these new, much more complex questions but this requires new approaches and priorities, new science capabilities, collaborators and partners.

There are three broad types of models and analytic frameworks in climate change research. The first one is used for observing the impacts, adaptation and vulnerability. The second type consists of integrated assessment models, and the third type includes climate models, concluded Dr. Buja.
Leslie Verswevyeld