“For years,” Paul Morin wrote[*], “those of us that made maps of the Poles apologized. We apologized for the blank spaces on maps, we apologized for mountains being in the wrong place and out-of-date information.” Now, after a decade of painstaking work, the time for apologies is over. A major collaboration between universities, the U.S. government and a software company has produced an unprecedentedly accurate map of the poles – and it was made possible by supercomputing.

Morin is the founder and director of the Polar Geospatial Center at the University of Minnesota, where he and dozens of other researchers help the National Science Foundation (NSF) map the Earth’s poles. Morin also liaises between the NSF and the National Geospatial-Intelligence Agency (NGA) and serves on the National Academy of Sciences’ Standing Committee on Antarctic Geographic Information. 

In short: if you’re interested in polar mapping, he’s your guy.

“It’s to serve places like this,” Morin said in a recent NSF-hosted webinar, pointing out a field camp in the dry valleys of Antarctica. “When we’re out there working, we’re sleeping in tents. […] As we were working, we didn’t have access to the kind of resources we have now. And so […] we flew around in helicopters, we had differential GPS, and we were geo-referencing air photography that was collected often in the 80’s, 90’s or the 00’s.”

Morin’s point is well-taken: for those working on or over the poles – not just researchers, but National Guard and Air Force servicemen as well – the accuracy of polar maps is a day-to-day, functional concern. (“I mean, this is the way that we get to work in the morning,” Morin said.)

The scope of the project was staggering. Antarctica is 15 million square miles – 50 percent larger than the contiguous U.S. “We can use all the standards superlatives – the highest, the driest, the coldest – but from my standpoint,” he said, “it’s just big.” But Antarctica, of course, is only one part of the equation. On the other end (quite literally): the Arctic, which is twice the size of the contiguous U.S.

Luckily, Earth-observing satellites tend to be in a polar orbit, constantly taking images of  the poles. The problem, then, became wrangling what Morin calls an “incredible fire hose of imagery” from NASA, the European Space Agency and commercial satellite operators. The imagery that the researchers were able to request allowed for pinpoint accuracy. “If you were to look at the ground in the valleys,” Morin said, “and if you were to put a single oak leaf in a specific location, you could detect the chlorophyll in that oak leaf in a 1.8 meter square pixel.”

But a single, detailed map wasn’t enough.

“You […] just don’t get the repeat that science would need, because the Earth’s surface is always changing,” Morin said of older surveying methods. “All these things – we want to be able to measure and see what the difference is.”

Then, five years ago, the U.S. gained the chairmanship of the Arctic Council and announced plans to create a robust elevation map of the Arctic. Morin and his colleagues realized that this was their opportunity to create an evolving topographic dataset for polar regions. The following year, President Obama announced a project with the NSF and the NGA to create that dataset for Alaska within one year and the Arctic within two.

With NGA’s satellite imagery contracts now at their fingertips, the newly formed team needed tools to process that massive amount of data. They turned to Ohio State University (OSU) and the National Center for Supercomputing Applications (NCSA) at the University of Illinois. OSU provided software that allowed the team to feed stereo imagery into an HPC system and receive a digital elevation model (DEM) with very little human intervention. The NCSA, of course, provided the firepower: Blue Waters, a hybrid Cray supercomputer that delivers roughly 13 petaflops, over 1.5 petabytes of memory and about 26 petabytes of storage. Over time, the team received allocations on Stampede2 and Frontier as well.

They got to work. The team produced a five-meter resolution elevation model of Alaska, then refined it to two meters. Then the Arctic: 12 percent of the Earth mapped at a two-meter resolution. Then Antarctica – another 8 percent. They produced REMA (the Reference Elevation Model of Antarctica) and, later, ArcticDEM, a tool for extracting those two-meter Arctic DEMs from Blue Waters.

Morin walked through the particulars of how granular these maps could be – individual trees being logged, ice melt, lava flows. “We now have better topography for the ice on Earth than we do for the land on Earth,” Morin said. “There really isn’t anywhere else on the planet that we just have this much repeat.”

The project was a success, and NGA and NSF have extended their collaboration and their time on Blue Waters – this time, with the aim to extend the polar mapping project to the entire surface of the Earth.

“When we began this, we just didn’t have HPC experience,” said Morin. “Last time I touched HPC before this project, the computer was a Cray-2. We needed software like Swift and Parsl for sub-scheduling – we’re doing hundreds of thousands of jobs, huge networking and automation. The community just isn’t used to this – you know, the next version of the poles is probably two petabytes! […] These projects are too big for any one agency – we’re talking public, private, multiple agencies, civilian defense… we have to bring everybody to bear on projects this large.”

To Morin, though, this is clearly still just the beginning. Morin cites a project (“Planet”) that is launching hundreds of shoebox-sized satellites for geospatial mapping. “There’s so much data coming through there that we just can’t think of how we’re going to process it even now,” he said. Of course, he does have some ideas: he recalls another project (“Iceberg”) using machine learning algorithms to detect permafrost in the Arctic.

“So,” he says excitedly, “if we can keep throwing imagery at this…”

[*] Paul Morin’s talk, “The use of NSF HPC for the Production of the Earth’s Topography,” was held last week as part of the NSF Office of Advanced Cyberinfrastructure’s Cyberinfrastructure Webinar Series. To read more about the talk, click here.