OSU Harnesses Personal Computers for Climate Research

Comment Off

resources_750_2At the Oregon Climate Change Research Institute (OCCRI) at Oregon State University, the power of thousands of personal computers was harnessed in one of the highest-resolution simulations of the western United States ever performed. OCCRI director Dr. Phillip Mote and his colleagues performed this computation through an international project called Weather@Home, which is administered by www.climateprediction.net. Weather@Home is run through the Berkeley Open Infrastructure for Network Computing (or BOINC, to be cute), which utilizes the computing power of thousands of volunteers’ personal computers from across the world.

BOINC unites volunteers interested in supporting science with researchers. Since users can select which projects to donate time to, this system allows the public to have a say in the direction of scientific research. There are more than a billion personal computers in the world. At any given time, a huge number of these computers are idle. Volunteers all around the world have donated this idle time to scientific computing projects, enabling massive projects to get done without having to compete against projects like nuclear weapon design and espionage for time on standard supercomputers. Currently there are close to 40 projects using the BOINC platform, furthering human knowledge in fields such as cryptography, molecular biology, astrophysics, and pure mathematics.

How can these thousands of computers work together on the same problem? BOINC is an example of parallel computing. Parallel computing is effective in cases where a problem can be split into parts. For example, suppose you had a bucket full of loose change and wanted to count it. It would take a long time for one person to sit and count the coins one by one. But if you recruited 10 friends, you could split the money into 10 piles, have each person count a pile, then collect the totals and quickly produce an answer.

To get accurate statistics, you need to have a large sample size. A major challenge for climate modeling is that because global climate models take such enormous resources to run, it’s only generally possible to run a handful of repeat simulations. The computing resources available through Weather@Home have allowed Dr. Mote and his colleagues to make significant progress in this direction. “When you have 30,000 modern laptop computers at work, you can transcend even what a supercomputer can do,” said Mote in a recent OSU press release.

Even though global climate models are run on some of the most advanced supercomputers in the world, they are limited to grid sizes between 50 and 300 kilometers. Suppose you have a global climate model of a respectable resolution of 100 kilometers (62.5 miles). If you wanted to provide an estimate of total precipitation for Corvallis, you would get one number to summarize the weather in a box going from Salem to Eugene and from Lebanon to the coast. On a global scale, that’s good enough. But if you need to make a recommendation to city planners, that isn’t going to be very useful—the weather on the coast is very different from the weather in the valley.

By limiting the study region to the western US and taking advantage of volunteer computing power, the OCCRI researchers were able to both reduce the grid size to 25 kilometers (15.6 miles) and greatly increase the sample size. “With this analysis we have 140,000 one-year simulations that show all of the impacts that mountains, valleys, coasts, and other aspects of terrain can have on local weather. We can drill into local areas, ask more specific questions about management implications, and understand the physical and biological climate changes in the West in a way never before possible,” said Mote.

The Weather@Home simulation results are in general agreement with observations. However, there are regions where the model needs to be improved—Weather@Home tends to be too cool in a few mountain ranges and too warm in arid plains, including the Snake River Plain and Columbia Plateau, especially in summer, as detailed in a paper published in the Bulletin of the American Meteorological Society.

New participants in the Weather@Home project are always welcome. Instructions are available at www. climateprediction.net; just click “Join!”

By Daniel Watkins

Be Sociable, Share!
In : Feature