A message in a bottle dropped in the ocean may be found years later somewhere completely unexpected; the routes taken by such bottles have been an enduring mystery of the sea. This month, researchers will attempt to tackle that mystery in the Forecasting Floats in Turbulence competition sponsored by the US Defense Advanced Research Projects Agency (DARPA), with a $50,000 prize at stake.
Oceanographers have an accurate picture of so-called mesoscale features, patterns of ocean currents that cover tens of kilometres or more. But seen in more detail, these coherent structures are surrounded by many smaller currents moving in different directions 鈥� so predicting the motion of a drifting sensor is far from straightforward.
鈥淭his is a chaotic system, so small errors in drifter location or velocity can quickly lead to very different results,鈥� says at the University of Victoria in Canada. 鈥淚f your model puts a drifter slightly on the wrong side of one of these coherent structures, it鈥檚 similar to the difference between a car getting off the highway at an exit or missing the exit.鈥�
Advertisement
Depending on exactly where it is, a drifting object may loop around in an eddy, get becalmed or change direction.
DARPA programme manager , who is running the competition, says success will rely on better computer models with high enough resolution for smaller scale currents. Imagine low-resolution 8-bit graphics compared with the 4K graphics of today, he says. 鈥淭he math to move between these regimes is filled with uncertainty and chaos,鈥� says Waterston.
Techniques for mapping currents have improved since the days when researchers relied on drift bottles instructing the finder to write back with their location, supplemented by convenient accidents. One such accident helped oceanographers map ocean currents for years 鈥� after 29,000 plastic bathtub toys were lost from a cargo ship in the Pacific in 1992, oceanographers logged the locations of the toys as they washed up on the coast of the US and Canada. Modern drifting sensors often have satellite tracking, but prediction is still difficult.
The challenge began on 3 November with DARPA releasing the exact location of 90 drifting solar-powered sensors from US firm Sofar Ocean in the Atlantic each day for 20 days. At the end of that period, competitors must provide predictions of the expected drifter locations for 10 days into the future. The scoring system favours long-term accuracy; a prediction to within 32 kilometres on day 10 scores more than an accuracy of 2 km on day one.
Aksamit notes that competitors may benefit from recent improvements in modelling fluid flows. His own research has also looked at how machine learning can help by filling in the gaps when detailed information isn鈥檛 available.
鈥淒eep-learning methods can tap into the history of the drifter motion, and make future predictions based on how the drifter behaved in the past,鈥� says Aksamit.
The immediate aim of the project is to assist DARPA鈥檚 Ocean of Things, a plan for a vast array of drifting sensors to carry out extended monitoring missions. Better understanding of drifting will help place them to ensure maximum coverage. Improved prediction tools for surface currents will also be useful for tasks like anticipating the spread of oil spills and other pollution, and helping searchers locate survivors after accidents and shipwrecks.
