Scientists call for a climate research agency

Scientists call for a climate research agency

BY PATRICIA DAUKANTAS | GCN STAFF

A panel of climate scientists has called for a new federal agency to advise policy-makers on climate research and projections.

The Climate Service they envision would be organized differently from other government research programs, said Richard B. Rood, a senior scientist at NASA's Goddard Space Flight Center Data Assimilation Office in Greenbelt, Md. It would employ about 150 scientists and software engineers and have a $50 million annual budget, he said.


NASA scientist Richard Rood envisions a Climate Service with a $50 million budget to advise the government on climate research.
No existing agency has the range of resources needed to solve large problems in climate change, so the government needs a multiagency, product-driven approach to get the right mix of operations and science, Rood said.

The panel's December 2000 report also said U.S. scientists are hampered by a lack of the shared-memory vector supercomputers now made in Japan. Some scientists believe vector systems perform more efficiently than the distributed-memory, parallel computers and clusters favored in U.S. laboratories.

The vector advocates scored a victory in late February when U.S. manufacturer Cray Inc. of Seattle announced it will resell SX-5 vector systems from NEC Corp. of Japan.

Rood's panel consisted of five government climate scientists and one consultant in organizational behavior. The Office of Science and Technology Policy and the interagency U.S. Global Change Research Program commissioned the panel in January 2000.

The term product-driven research describes what usually is called applied research, Rood said.
In discovery-driven research, individual scientists or small groups seek funding from outside agencies and publish peer-reviewed findings. Applied or product-driven research, in contrast, aligns research with timely delivery of technical reports requested by policy-makers.

European cousin

The organization most similar to the one proposed is the European Centre for Medium-Range Weather Forecasts in Reading, England, a consortium backed by 21 countries. It produces operational forecasts and conducts research in new numerical forecast methods. Rood said the center has a more executive-oriented style of management than its U.S. counterparts.

The panel left open the question of whether a Climate Service should be an independent agency like NASA and the National Science Foundation or come under the aegis of a cabinet department.

Simply reorganizing current projects into a new agency without rethinking the underlying management structure would not work, Rood said.

He has long advocated optimizing hardware and software for feeding real-time, remote-sensor observations into Earth science models while they are running [GCN, Oct. 25, 1999, Page 1].
But researchers are still hitting the limits of scalability at 100 to 200 processors. Software optimization and faster processors would improve scalability, he said.

Massively parallel supercomputers rely on commercial microprocessors with clock speeds topping out around 1 GHz. Vector systems have fewer processors that run at up to 10 GHz, Cray spokesman Steve Conway said.

The European center's Fujitsu VPP/5000 vector supercomputer, with only 100 processors, ranks as the 12th-fastest in the world.

American-built supercomputers that rank higher than the Fujitsu system have 1,084 or more CPUs.

The Commerce Department's 1997 anti-dumping case against Japanese supercomputer vendors has made their vector systems essentially unavailable to U.S. researchers, the Rood report said. Cray, at that time a division of SGI, complained after the University Corporation for Atmospheric Research in Boulder, Colo., awarded a supercomputer contract to NEC. Subsequently, Japanese vector system vendors were slapped with tariffs as high as 454 percent.

On Feb. 28, the now-independent Cray said it would drop the dumping complaint in exchange for rights to rebrand and market the NEC SX-5 or its successor in the United States for 10 years. NEC will also invest $25 million in Cray.

Commerce's antidumping decision probably benefited IBM Corp. and Compaq Computer Corp. more than it ever helped Cray, Conway said.

Over the past five years, IBM and Compaq have contracted to build the largest supercomputers in the Energy Department's Accelerated Strategic Computing Initiative [GCN, Sept. 4, 2000, Page 8].

Cray officials expect the SX-5 will hit the U.S. market around June 1. A single-cabinet configuration would cost $2 million to $3 million, Conway said.

Software deficiencies have arisen from the U.S. focus on kiloprocessor commodity clusters, the Rood report said.

'With the restriction of U.S. centers to distributed-memory, commodity-based processors, a large investment is needed in applications software, and there is substantial risk that the performance is intrinsically limited,' it said.

The Rood report is available on the Web at www.usgcrp.gov/usgcrp/Library/models2001/default.htm.

inside gcn

  • machine learning

    Mitigating the risks of military AI

Reader Comments

Please post your comments here. Comments are moderated, so they may not appear immediately after submitting. We will not post comments that we consider abusive or off-topic.

Please type the letters/numbers you see above