The amount of data encompassed by the global climate in its entirety, from its winds and temperature conditions to shifting formations of cloud, is almost unimaginably complex. Layer in the ever-evolving small- and large-scale interactions involved, and the thought of creating a model of the entire Earth system seems beyond possibility. That is, until recent HPC-powered simulations reached a level of accuracy in many ways “indistinguishable from nature.” This image shows the cloud field from a global simulation of the atmosphere, at 7 km resolution, meaning that the model simulates the winds, the temperature and clouds, and all their interactions, everything that happens in the atmosphere, in little cells 7 km on side. For an image of the size you now see on your screen, that is below the threshold of human vision.
So explains Venkatramani Balaji, Head of the Modeling Systems Group in Princeton's Program in Atmospheric and Oceanic Sciences and the National Oceanic and Atmospheric Administration's Geophysical Fluid Dynamics Laboratory. For more than 20 years, Balaji’s work has centered on building ever-more-sophisticated models of the Earth’s atmospheric conditions. This relies on the processing heft on some of the world’s most powerful high-performance computing systems, including those at Argonne National Laboratory, where simulations are run on up to a million parallel threads. On campus, Balaji and his team also leverage the Tiger cluster made available through Research Computing and PICSciE, the Princeton Institute for Computer Science and Engineering.
The need for such a system goes beyond scientific curiosity. In the face of global warming and its related constellation of localized threats, systemic vision at the global level may prove critical to some of civilization’s most critical environmental projects, with impacts on everything from agriculture to public health. It’s a project that started at Princeton when global warming was first identified as a planetary threat. “Our lab actually pioneered the study of global warming using models, back in the 1950s and 60s. In fact at Princeton we trace our lineage back to the dawn of computing, when computer pioneer John von Neumann at the Institute for Advanced Study suggested that we could use this new tool to build first-principles models to simulate and predict weather and climate.”
The number of factors taken into account in Balaji’s work is awe-inspiring. “Creating a complete-Earth model involves solving the equations for the atmosphere — think of it as a thin layer of fluid, gases and liquids; chemicals of various kinds; dust, clouds and other particles that absorb, reflect and scatter heat. All act in concert on a large wobbling, spinning sphere, which is being heated on one side by the sun, losing heat on the other side,” Balaji explains. This is just the beginning.
“On our planet we have an ocean, which also absorbs heat and CO2 and other elements from the atmosphere, and has its own complicated flow. It can store and release heat over long periods of time — millennia ,” Balaji says. “And ecosystems are of course the crucial element of the Earth system — the thin green layer on land and ocean, that supports all the known life on the planet. Climate science has shown that humans are collectively and measurably affecting the state of the entire system, and we are using our models to work out the implications.”
Beyond being complex, the systems involved are also non-linear and defined by chaos, as first noted by famous MIT meteorologist Ed Lorenz, who drew on the mathematical insight of Poincaré. Without HPC, the complicated dance among these systems would remain largely invisible. “Computing is key to the story: if you think of climate models as a scientific instrument for looking at the Earth system, like a telescope... that telescope is becoming more and more powerful, allowing us to zoom in on more and more features.”
As the field zooms in on the atmosphere at higher and higher levels of accuracy, we reach the stage where our models pass what Balaji’s Oxford-based colleague Tim Palmer calls a “climate Turing test”: where one cannot tell if we are looking at observations or a simulation. The hope is for greater predictive power, especially when it comes to potentially devastating atmospheric events. By running multiple simulations in concert at an ever-greater level of accuracy, Balaji and other researchers aim to secure a better understanding of the nuanced conditions that power catastrophic events like Superstorm Sandy. Moreover, new HPC-driven data is helping researchers understand how such events are influenced — or even driven — by climate change.
Balaji explains that Princeton’s earliest studies of global warming, by Syukuro Manabe and colleagues, involved hundreds of data points and up to five variables. “We are now routinely running simulations using a billion points, hundreds of variables to represent the system state (including many chemical and biological species) and running those models for hundreds and thousands of simulated years.”
The simulations shed light on factors we can at least in part control, while limiting experimentation to the virtual. “You can make an analogy with clinical trials in medicine. We normally test a treatment for safety by running a trial where some patients get the treatment, and others don't,” Balaji explains. “When it comes to climate change, we have a single patient, the planet itself, and cannot afford to put it through trials!”
“We instead rely on our models — we convince ourselves by rigorous sets of tests that the model indeed resembles Planet Earth — and then put it through various scenarios: what would happen if we instituted a carbon tax? What would happen if we did nothing at all? Our models tell us the consequences.”
With an eye on the exascale supercomputing systems expected to be available in the early 2020s, the future of the field promises to be as exciting as the present. The hope: as our understanding of particular phenomena evolves, the model will evolve in kind, and perhaps trace a path towards a healthier planetary future. “When the ‘pale blue dot’, as Carl Sagan called it, was first photographed from the Moon in 1969, it helped us see the planet, and ourselves, as interconnected. I think climate modeling takes that holistic thinking one step further,” Balaji muses.
“Not only can we see the actual Earth system in an image like the one above, we can see the various might-have-been Earths in our models, what the Earth would look like if we had never polluted it, what it might have looked like in the past when dinosaurs roamed and tropical flowers and ferns bloomed where there is only tundra today, what pathways may enable us to ensure a sustainable energy future for all of us. Computing and simulation are central to that vision.”
Learn more about PICSciE’s HPC resources, and visit Dr. Balaji’s page on the Geophysical Fluid Dynamics Laboratory website to learn more about this and his other research projects.