Imagine a world with no storms.
School children would never know the joy of a snow day, the S.S. Minnow would never have been lost, and Dorothy would have never journeyed to Oz.
Such is the atmospherically vanilla world portrayed by conventional climate models. In conventional climate model simulations, precipitation only falls in moderate showers, leaving climate scientists to complain that the models rain too weakly, too often. And according to climate scientists, a virtual world with no stormy weather isn’t just boring – it’s unrealistic. In fact, the Intergovernmental Panel on Climate Change, the synthesis of climate change research that is considered the final word on the state of the science, says that their data should not be used to project global precipitation pattern changes.
A more recent innovation in climate modeling, however, seems more capable of simulating the statistically rare and intense storms seen in nature. The added realism could greatly improve scientists’ ability to sort out the dynamics of future global precipitation patterns, but the new technology has its limitations as well. Super-parameterized or multi-model framework (MMF) models miss a particularly important source of precipitation – large-scale, organized storms formed in the unique atmospheric conditions surrounding inland mountain chains.
Mike Pritchard, a graduate student researcher at Scripps Institution of Oceanography, at UC San Diego, set out to find out why MMF models were not capturing this particular brand of storm in their simulations. He chose a storm system that forms in the lee, or downwind side, of the Rocky Mountains each summer to focus his investigative efforts.
“The storms are born in a wonderfully whooshy part of the earth’s atmosphere,” said Pritchard. Variations in surface temperatures across the rugged Rocky Mountain terrain tilt the air pressure systems above them. The seesawing bands of pressure, in turn, create a bellow effect that draws moist air from the Gulf of Mexico and pulls it across the landscape east of the Rockies. The moist air then feeds these massive, organized storms that roll across the plains during the growing season. The storms last for days, dumping nightly rains over the central United States as they travel east, according to Pritchard.
The summer storm systems that sweep across the central U.S. are similar to other seasonal convective storms that we see forming in the lee of mountain chains around the world. It's an important sort of storm for climate models to get right if they are to reliably project how rainfall patterns in the continental interiors will change in the future.
These massive storms and the crop-soaking rains they deliver, however, were missing in MMF simulations that Pritchard ran during his research. He suspected that the model was incapable of representing the atmospheric physics of these particularly massive systems, and he wanted to understand why. So, he launched an investigation.
“I went looking for trouble,” he said.
But that's not what he found.
Pushing the Limits of Computational Horsepower
Conventional climate models, according to Pritchard, were built with statistical equations that assume a limited range of behaviors for clouds.
“These models suppose that all clouds are like the popcorn clouds we often see strewn across an otherwise blue sky in the tropics,” said Pritchard. Popcorn clouds bloom, live, and die all in one place. They don’t organize, build, and move like these enormous storm systems we see in the central U.S., he said.
Conventional climate models could be configured at high enough resolutions to capture these organized systems and shake up the bland-weather scenario that currently characterize their simulations. But conventional models running at that level of atmospheric detail are computationally expensive, according to Pritchard; too expensive to be used to make long-range projections for precipitation. More computing power is needed for every level of detail added, and the details matter when it comes to the kind of storm he was investigating.
The eventual goal in the community, according to Pritchard, is to have a full global cloud-resolving model that pixelates global cloud activity down to 1-kilometer squares.
“Running a conventional climate model that could represent cloud behavior down to that scale with the technology we have now would take an entire day to generate one simulated day of global activity,” he said.
That doesn’t help climate scientists who want to make projections about what will happen in 50 to 100 years.
Spreading the Computational Load
“My group is working on an interim technology,” said Pritchard. He’s a doctoral candidate at Scripps collaborating with specialists in multi-scale modeling frameworks. MMFs embed a network of miniature cloud resolving models within a larger framework. The mini-models supply the details of cloud and rain activity to the overall framework without overtaxing the computers that run them.
“Some researchers liken the mini-cloud models in MMFs to opinion polls,” said Pritchard. “It takes a lot of work to collect the opinions of everyone in the state; so to save time and effort, you collect the opinions of a select few and extrapolate their views to the larger group.”
The MMF approach relies on a technique of splitting the embedded cloud models into thousands of discrete pieces so that the computational work they generate can be distributed across many nodes inside a supercomputer. This divide-and-conquer method makes MMFs more computationally affordable, but splitting the embedded cloud resolving models into tiny separate pieces could explain why the models have trouble simulating storms systems that travel – like the summer storms that well up in the lee of the Rocky Mountains.
“The problem may have something to do with the fact that the small, fast-moving physics of storm propagation are artificially restricted to these comparatively small, discrete squares inside the mini-models,” said Pritchard. The isolated chunks of simulated atmosphere don’t talk to each other directly. And Pritchard assumed the lack of communication prevented the simulated storm from propagating east and bringing the rain.
Spotting the Storm
Pritchard ran new simulations in 2010 with Super-Parameterized Community Atmosphere Model 3.5 (SP-CAM 3.5), an MMF model used by his research group. He changed the simulation parameters in the model, and began animating portions of data to see if he could find where the model was breaking down. The change in perspective revealed something that he had not seen before — evidence of convection and organized precipitation brewing in the highest echelons of the system.
“The microphysics of the mini-model clouds were wrong,” he said. “Something was causing the rain to evaporate before it hit the ground.”
Pritchard and Scripps colleague Gabe Kooperman are now collaborating with outside groups to improve the mini-models’ handling of precipitation.
“What really impressed me,” said Pritchard, “was how the virtual storm was propagating in the model.” It moved as a large, organized system despite the fact that it was travelling across a patchwork of tiny, independent, cloud-resolving models. “It made me re-think the possibilities with MMFs,” he said.
With a little help, the MMF model was able to represent a weather phenomenon that scientists had assumed it could not handle. Pritchard said that the discovery makes him optimistic that other shortcomings in the MMF approach can eventually be reconciled.
Pritchard points at his laptop screen to an enormous red swirl looming ominously over much of the western Pacific Ocean. “We call it The Great Red Spot,” he said. “It’s a super-monsoon that doesn’t actually exist in nature.” It’s an aberration we only see in MMF simulations.
Some researchers suspect that the same communication breakdown that makes it difficult for MMF to represent large-scale storm propagation also creates occasional anomalies like The Great Red Spot. “Phantom weather events like this in the model will have to be resolved before MMF can be used for future climate projections,” he said.
But if Pritchard was able to coax a missing storm into an MMF simulation, perhaps others will have similar success getting the mystery monsoon out.
“Barring a major improvement in super-computing speeds,” said Pritchard, “MMFs are our most immediate hope for getting climate models to tell us something useful about future precipitation patterns.”
Pritchard has recently reported his findings in The Journal of Atmospheric Sciences along with co-authors Mitchell Moncrieff of the National Center for Atmospheric Research in Boulder, Colorado; and Richard Somerville, a climatologist and emeritus professor at Scripps Institution of Oceanography.
Their research was supported by the Center for Multiscale Modeling of Atmospheric Processes (CMAP), a National Science Foundation (NSF) Science and Technology Center managed by Colorado State University, and by the U.S. Department of Energy’s Atmospheric Science Program Atmospheric System Research.
— Donna Hesterman