One of the major benefits of certain artificial intelligence models is that they can speed up menial or time-consuming tasks —- and not just to whip up terrible “art” based on a brief text input. University of Leeds researchers have unveiled a neural network that they claim can map an outline of a large iceberg in just 0.01 seconds.
Scientists are able to track the locations of large icebergs manually. After all, one that was included in this study was the size of Singapore when it broke off from Antarctica a decade ago. But it’s not feasible to manually track changes in icebergs’ area and thickness — or how much water and nutrients they’re releasing into seas.
“Giant icebergs are important components of the Antarctic environment,” Anne Braakmann-Folgmann, lead author of a paper on the neural network, told the European Space Agency. “They impact ocean physics, chemistry, biology and, of course, maritime operations. Therefore, it is crucial to locate icebergs and monitor their extent, to quantify how much meltwater they release into the ocean.”
Until now, manual mapping has proven to be more accurate than automated approaches, but it can take a human analyst several minutes to outline a single iceberg. That can rapidly become a time- and labor-intensive process when multiple icebergs are concerned.
The researchers trained an algorithm called U-net using imagery captured by the ESA’s Copernicus Sentinel-1 Earth-monitoring satellites. The algorithm was tested on seven icebergs. The smallest had an area roughly the same as Bern, Switzerland and the largest had approximately the same area as Hong Kong.
With 99 percent accuracy, the new model is said to surpass previous attempts at automation, which often struggled to tell the difference between icebergs and sea ice and other features. It’s also 10,000 times faster than humans at mapping icebergs.
“Being able to map iceberg extent automatically with enhanced speed and accuracy will enable us to observe changes in iceberg area for several giant icebergs more easily and paves the way for an operational application,” Dr. Braakmann-Folgmann said.
