Supercomputers usher in a 'golden age' of high-res climate modeling

High resolution simulation of the Community Atmospheric model, by Wehner, et al. Credit: Michael Wehner, YouTube

Thursday, November 13, 2014, 6:28 PM - Using computer models to forecast future climate conditions is a challenging task, but with the advent of the latest generation of supercomputers, it seems that the real limitations lie with us, rather than with the computers.

Supercomputers are really blowing away all expectations these days, with speeds ranging up to 10s of petaFLOPS (1 petaFLOPS = 1 quadrillion FLoating-point Operations Per Second), and new records being set all the time. While playing video games on computers like that would be a mind-boggling experience (as that's 10,000 or more times the speed of a Sony Playstation 4), one area of scientific research benefiting from these new supercomputers is climate forecasting.

Running a sophisticated computer model to predict future climate conditions can take a long time. Lower-resolution models, where the data is run in larger blocks, each covering quite a bit of the Earth's area, can do a good job when forecasting general trends of one or two factors and be run in a fairly reasonable amount of time. This works well when you're modeling something like surface temperatures or ocean temperatures or moisture content in the atmosphere, since the larger-scale effects can still give you high confidence in the results. However, the low resolution causes problems with forecasting other important factors.

"In the low-resolution models, hurricanes were far too infrequent," Michael Wehner, a climate scientist at the Lawrence Berkeley National Laboratory, in Berkeley, Calif., said in a Berkeley Lab press release.

"High resolution gives us the ability to look at intense weather, like hurricanes," study co-author Kevin Reed, a postdoctoral fellow in Climate and Global Dynamics at the National Center for Atmospheric Research (NCAR), in Boulder, CO, added in the release. "It also gives us the ability to look at things locally at a lot higher fidelity. Simulations are much more realistic at any given place, especially if that place has a lot of topography."

However, moving up to fine-resolution required slows things down significantly, though. Some models can take years to finish running.

These new petaFLOPS-range supercomputers are coming to the rescue, though, with much more practical results, that reduce those years down to just months.

Wehner, who led the latest IPCC climate assessment, has been working with an international team of colleagues to test a high-resolution climate model developed by the US Department of Energy and the National Science Foundation, called the Community Atmospheric Model, version 5.1. In a recent paper, just published in the Journal of Advances in Modeling Earth Systems, they describe running this climate model down to just a 25-km resolution, with the supercomputer generating amazing results.

CLICK BELOW TO WATCH: A side-by-side comparison of old vs new model runs, both attempting to produce weather effects from July - Nov, 1979, shows the staggering difference in the results.

"I've literally waited my entire career to be able to do these simulations," Wehner told Berkeley Lab News.

"I've been calling this a golden age for high-resolution climate modeling because these supercomputers are enabling us to do gee-whiz science in a way we haven't been able to do before," he added. "These kinds of calculations have gone from basically intractable to heroic to now doable."

Using hindcasting techniques to test the model on these supercomputers, the researchers found that the high-resolution models did much better, coming very close to producing the same number of storms that were seen in reality.

DID YOU KNOW?: Hindcasting, or 'backtesting' is a method of testing how well a climate model can forecast the future, by starting the model run sometime in the past, sometimes decades back, and running it forward from there.

From here, the next steps in the research will be to switch from hindcasting past changes to forecasting future climate.

One other point made by this study was that the limitations that currently exist in climate models - specifically in how much data we can supply them and how well we are able to translate or 'parameterize' real-world weather effects into the model - won't be improved by the high-resolution models. To continue to get good results, we still need to provide the models with enough data, and provide accurate translations of real-world effects into computer parameterization.

The latter - accurately translating real-world effects - is one area where we'll need a new "paradigm shift" for climate modelling, according to what Wehner told Berkeley Lab News, specifically with how we understand and handle clouds.

CLICK BELOW TO WATCH: "I find that these projected future changes to extreme weather and climate in a warmer world to be sobering. The cost and impact of even small increases in the severity of these phenomena can be large, both to our nation as well as others around the world." - Dr. Wehner, 2012.