Chemistry and Dynamics of Ocean and Air
in One Earth System Model

PROJECT LEADERS
C. Roberto Mechoso, Professor, Department of Atmospheric Sciences, UCLA Senior Fellow, SDSC
James Demmel Professor, Department of Computer Science and Mathematics, UC Berkeley

alifornians and other denizens of the American Southwest have just been through one of their wettest winters, thanks to El Niño, and are now wondering if they face the prospect of a dry winter, thanks to La Niña. These extreme climate phenomena are known to affect areas of the world far from their origins in the tropical Pacific oceans, and their costly consequences in the shape of storms, floods, mud slides, drought, and forest fires have earned general respect. That respect includes an intensified effort to understand and predict El Niños, La Niñas, and other global climate phenomena. This effort is now pushing the limits of high-performance computing, since it's difficult to make significant predictions by piecing together a dozen or so simulations that are only several months long.

UCLA is the home of a pioneering group of scientists who have been working in the Department of Atmospheric Sciences on computer models of the Earth's weather and climate since the 1950s. (The department itself was founded in 1940.) "Our modern models still incorporate the amazingly robust formulation of atmospheric circulation processes that was developed by Akio Arakawa starting in the 1960s," said C. Roberto Mechoso, leader of the group over the last 10 years. Arakawa himself is still active and has led many revisions and upgrades to keep the models state-of-the-art.

Thus it is not surprising that one of the largest and most ambitious experiments to emerge from the worldwide community of modelers of climatic change is this group's effort to organize and put together a working Earth System Model (ESM). The ESM effort unites a global climate modeling experiment led by Mechoso, a global atmospheric chemistry experiment led by Richard Turco, also of UCLA, and NPACI efforts in both the Metasystems and Programming Tools and Environments thrust areas.

"Our ESM efforts have been funded for the last five years as part of the NASA Grand Challenge Science Teams I and II groups of geophysical and astrophysical advanced modeling efforts," Mechoso said. "The collaborations within NPACI, while newer, are now key elements of the program as a whole."

COMPONENTS OF THE SYSTEM

PROGRESS AND PLANS

MULTISCALE, MULTIRESOLUTION MODELING

COMPONENTS OF THE SYSTEM

"We're trying to combine four large-scale, parallel computer models with the objective of running them all, in simulations as long as a century, in the wall-clock time it now takes to run just the atmospheric model, while feeding information from each model to the others," Mechoso said.

One of the models is the UCLA Atmospheric General Circulation Model (AGCM). The Ocean General Circulation Model (OGCM) is based on POP (the Parallel Ocean Program), originally developed for a Connection Machine at the Los Alamos National Laboratory on the basis of a model known as the Bryan-Cox-Semtner model. The version of POP in the ESM has been optimized by Yi Chao and a group of researchers at the NASA Jet Propulsion Laboratory. Finally, there are two chemical models: an Atmospheric Chemical Tracer Model (ACTM) and an Oceanic Chemical Tracer Model (OCTM), both developed in the Turco group.

"The chemistry of the atmosphere and the oceans can have great influence on climate outcomes," Turco said. "But the incorporation of these effects into atmospheric or ocean dynamical models has rarely been tried." He noted two examples of the interconnections. First, scientists have long known that ocean temperatures (and the all-important sea surface temperature at the boundary with the atmosphere) depend on the 3-D distribution of salinity in any region of the ocean. Second, atmospheric temperatures and radiation properties (absorbing, transmitting, and reflecting heat) are linked to, for example, sulfate aerosol burdens--sulfates contribute to cooling by absorbing incoming solar radiation. All the chemical tracers that are found in and may be exchanged between atmosphere and oceans are included in Turco's models.

"The entire climate system, particularly over decades and longer, is a mass of nonlinear physical, chemical, and biogeochemical interactions and feedbacks," Mechoso said. "A realistic model of the system must include all effects that can have an influence over the time period we want to model."

PROGRESS AND PLANS

Preliminary versions of the couplings between and among the models have been tried. The model codes used in the project are among the most highly optimized of such models in the world. In one experiment done at NASA Goddard Space Flight Center, 10.6 gigaflops was achieved for the AGCM and OGCM running side-by-side on 512 processors of a CRAY T3D; a five-day simulation with high resolution took less than 20 minutes of wall-clock time.

In a more recent experiment, the coupled models on the CRAY T3E have run more than four times faster on the same number of processors. The code run on the T3E scales nearly linearly. A 50-year simulation of the coupled models in lower-resolution versions yielded very realistic pictures of El Niño/Southern Oscillation events during the period (Figures 1 and 2). The ocean model at high resolution has successfully simulated Gulf Stream eddies, essentially "storms" within the Gulf Stream currents whose locations have a strong impact on climate (such eddies may have been a major influence on the "Little Ice Age" of the late 17th and early 18th centuries). In addition to Mechoso and Turco, members of the UCLA modeling group include Mechoso group members Tony Drummond, John Farrara and Joseph Spahr, and Turco group member Mohan Gupta.

"The addition of either chemical model to the fully coupled AGCM/OGCM makes the whole complex run an at least one order of magnitude slower," Mechoso said. "This gives a very good idea of the kind of computational capability that will be needed to run a working ESM."

In its first working version, during the next year, the ESM model output will be produced by the four model components executing in distributed computing environments and validated against observational data sets contained within an object-relational database management system constructed for the ESM. For this purpose, Jim Demmel of the Computer Science Department at UC Berkeley and group members Keith Sklower and Howard Robinson have been building a Distributed Data Broker for the ESM, a program that can request model fields at a specified sample rate and channel them to other models or applications. This is one particularly active facet of the NPACI collaboration in the ESM, part of the Programming Tools and Environments thrust area. "The broker sees the models as producers and consumers of relevant data subfields," Mechoso said. "It registers and exchanges the subfields as requested during coupled runs."

MULTISCALE, MULTIRESOLUTION MODELING

In a separate NPACI project, the Mechoso group will work with the Legion project of the Metasystems thrust area to integrate existing numerical models across global, regional, and local scales, spanning five orders of magnitude in space and minutes to decades in time. The models to be used include the UCLA AGCM, OGCM POP, a Mesoscale Atmospheric System (MAS) used by Jinwon Kim at the Lawrence Berkeley Laboratory, and the San Diego Bay watershed and hydrodynamic model from SDSC.

"We expect to have our data broker running on top of Legion to perform the same tasks it does in the NASA-funded work," Mechoso said. The AGCM has already been coupled to MAS in experimental runs, and Legion has been integrated with the AGCM/MAS system, under the direction of Catherine Holcomb and Andrew Grimshaw of the Legion project at the University of Virginia. "We will be using the IBM SP and the CRAY T3E at SDSC for this work," Mechoso said. After the AGCM/OGCM has been coupled to MAS, integration of the San Diego Bay model will complete the cascade in model scales.

"The question of whether the yield of vital crops such as maize and rice can be maintained at levels that avoid famines has a climate system component that we will then have a hope of answering," Mechoso said. "Our completed Earth System Model should be able to translate global-scale predictions into specific prognostications for regional and local consumption. But we will need the multiteraflops machines, the petabyte databases, and the technological expertise of the future to accomplish this goal." --MM