ORNL
Search Magazine  
   

Climate models for a changing world

Simulations predict possible climate outcomes


Using ORNL's Jaguar supercomputer to capture glimpses of Earth's future, researchers are trying to bridge the gap between science and public policy. Upgrades to Jaguar may make that gap smaller.

Ongoing upgrades to Jaguar, one of the fastest supercomputers in the world, have allowed internationally recognized climate scientist Warren Washington and his colleagues to see Earth's future climate in the long term and with much more clarity.

Recent advances in modeling have improved prediction of the movement of ice sheets. The simulated (right) versus the observed (left) flow velocities for Greenland ice sheets match well. Simulation data: Steve Price, Los Alamos National Laboratory; Observational data: Jonathan Bamber and colleagues in Journal of Glaciology, 46, 2000.
Recent advances in modeling have improved prediction of the movement of ice sheets. The simulated (right) versus the observed (left) flow velocities for Greenland ice sheets match well. Simulation data: Steve Price, Los Alamos National Laboratory; Observational data: Jonathan Bamber and colleagues in Journal of Glaciology, 46, 2000.

Washington and Jerry Meehl, climate researchers at the National Center for Atmospheric Research (NCAR), have paired up with Jaguar to shed light on the confounding and often controversial relationship between greenhouse gas emissions and their effects on climate change.

They head a group of scientists and computer experts that has carried out extensive climate change research for the 20th and 21st centuries and beyond for the Intergovernmental Panel on Climate Change assessment reports, including the 2007 report that concluded the Earth is definitely warming and human activity probably contributed to the warming. A subsequent assessment that will take advantage of model improvements is expected in 2014.

"The US Department of Energy and other parts of the federal government want to know how different energy strategies affect the environment," Washington says. "Tools like the computer climate models allow them to ask conditional, 'what if,' questions."

The Community Earth System Model, a virtual time machine supported by DOE and the National Science Foundation, is a set of coupled computer programs used on Jaguar to simulate Earth's past, present and future climates. Predicting what the future has in store requires a mixture of model development subprojects representing numerous Earth system climate components and lengthy, complex simulations of climate variability and change.

"We have to make some assumptions about how society is going to deal with greenhouse gases," Washington says. "For example, if the world predicts a 'business as usual' scenario, meaning the world continues to emit greenhouse gases at the present rate or even faster, our CESM shows future projections of a much warmer climate by the end of the century."

Jaguar can also produce ensembles of experiments that separate natural climate fluctuations from those caused by man. These experiments show what will happen when there are greenhouse gas mitigation efforts. If emissions of greenhouse gases are cut back significantly, the model shows a much more moderate increase in global warming.

"Our research tells the public and the policy makers what future climate change will look like, given different scenarios," Washington says.

In 2012, DOE granted Washington's team and their project, the Climate End Station, a total of 86 million processor hours through the Innovative and Novel Computational Impact on Theory and Experiment program. The team has 56 million processor hours on Jaguar and 30 million processor hours on Argonne National Laboratory's supercomputer to generate climate simulations. This is equivalent to the power of 28 million dualcore laptops for one hour. However, unlike millions of separate laptops, Jaguar's massive array of parallel processors are interconnected, allowing them to perform millions of calculations simultaneously and making more complex simulations possible.

The CESM can use up to 100,000 of Jaguar's almost 300,000 processors at a time. "We carry out the calculation by breaking the globe up into areas to which we assign a certain number of processors," Washington says. Depending on the resolution of the model, the processors may run for hours or days.

These models give Washington and his colleagues a much more detailed picture of potential climate change, including information regarding the warming of the polar regions, the extent and frequency of droughts and whether heat waves will increase in the future.

The team's history of working with Jaguar has allowed them to examine the relationship between carbon dioxide emissions and the potential for global climate change. In the coming months, upgrades to Jaguar will incorporate high-performance graphics processing units, or accelerator chips. The newly installed, highly specialized GPUs will offer a faster way to process information than relying on central processing units alone. When the upgrades are complete, the system will be renamed Titan.

While Titan will have many advantages over Jaguar, Washington's computer code in its current state is unable to take full advantage of the supercomputer's improved power and speed. However, ORNL's and NCAR's computing experts are already working with the CESM codes to take advantage of the added capabilities.

"GPUs are meant to accelerate really heavy parts of the computation, while CPUs handle the lighter parts," said Matthew Norman, a computational climate scientist within the Scientific Computing Group in ORNL's National Center for Computational Sciences.

The CESM incorporates interacting components: Earth's atmosphere, land surface, oceans, sea ice and land ice. Computations involving the atmosphere take up most of the time and supercomputing resources when the team runs the higher resolution models. This is where Titan's GPUs, which effectively power through dense calculations, will be most useful.

"We will take certain dominant parts of the atmosphere simulation and put those calculations on GPUs where they will run 75 percent more efficiently," Norman says. "Through the use of GPUs, Washington's colleagues will use fewer computing resources to achieve their groundbreaking simulations."

Its use of GPUs will place Titan in a unique position, allowing it to run more sophisticated climate models effectively and efficiently. Researchers will be able to test more phenomena, helping the climate models more faithfully depict the future.

"We are greatly indebted to DOE, and to ORNL in particular, for providing computer time so we can carry out these types of experiments," Washington said. "The use of Jaguar has been vital to pursue this very important research objective for the nation and the world." —Jennifer Brouner