Page 22: of Marine Technology Magazine (March 2018)

The problem with model resolution Réunion Island during the model month of July 2015.”

Modeling the global ocean is inherently dif? cult. Through- And their results were remarkable. According to Durgadoo, out the history of ocean model development, from the earli- “while it is impossible to pinpoint an exact location, we found est models using very basic computing products by modern that the origin of the ? aperon is likely to be to the west rather standards, through to the modern state-of-the-art ‘million- than southwest of Australia. More importantly, based on our lines-of-code’ behemoths, researchers have struggled to deal analysis, the chance that the ? aperon started its journey from with issues of resolution i.e. the geographical scale at which the priority search area is less than 1.3 percent.” a model runs - where the smaller your grid resolution is, the The team had used their model to conclude that search ef- better your representation of the ocean. forts along the priority zone were highly unlikely to achieve

According to Professor Sergey Danilov, who works on ocean success in ? nding the aircraft. Indeed, with the plane still model development at the Climate Dynamics Division of the missing today, the fate of ? ight MH370 remains a mystery.

Alfred Wegener Institute, Bremerhaven, Germany, the main [Editor’s note: Since the author’s writing, challenge has always been to make the models reproduce the the search for MH370 has resumed] water mass characteristics and circulation that we observe in the real ocean. Technology drives model advances “Motions at small spatial and temporal scales cannot be For ocean models to have reached this level of sophistica- modeled and are therefore parameterized,” he says. “This cre- tion today, the technology driving their development has had ates errors, which can accumulate over time. So, modelers try to have been wide ranging; from the observational units de- to reduce these by increasing resolution, improving the ? del- ployed at sea for acquiring accurate data, to the state-of-the- ity of parameterizations, or improving numerical algorithms.” art supercomputers used to make future predictions.

This sentiment is echoed in former MIT oceanographer Carl “Developments on the computer hardware side allows one

Wunsch’s book Modern Observational Physical Oceanog- to use more resources,” says Professor Danilov, “meaning we raphy, where the author explains that no model has perfect can explicitly resolve processes that were previously param- resolution. This means that some processes are always omit- eterized. There is hope that new computational technologies ted – an obstacle that nature does not face. “The user must involving GPUs – Graphics Processing Units – will lead to an determine whether the omission of those processes is impor- increase of model throughput.” tant,” writes Wunsch. “Even were it possible to perfectly nu- “On the physical side,” he adds, “new data are becoming merically represent the assumed equations, errors always exist available through modern technology, helping to better tune in computer codes.” or constrain parameterizations used in the models. Satellite altimetry and Argo ? oats are of particular importance.”

The search for MH370 But Danilov points out that progress in computational power

Nevertheless, scientists who specialize in ocean model de- is the main driver at present. Running global models at a high velopment have made massive strides in their quest for per- resolution – around one kilometer grid size – is already possi- fection. When a ? aperon (part of a plane wing) from missing ble, meaning processes down to that level are being resolved.

Malaysian airlines ? ight MH370 turned up on La Réunion Is- “Models that resolve mesoscale motions will become a real- land in the Indian Ocean in July 2015, Dr. Durgadoo and his ity in the foreseeable future,” he says. “But such model runs colleagues had a brilliant idea. By using their state-of-the-art are still too computationally expensive, meaning they take a ocean model, they reasoned that it should be possible to help lot of time to run and generate a lot of data. So, the distinction ? nd out where the plane had crashed. should be made between what is possible in principle, and “The mere fact that debris belonging to MH370 had been what can be used as a research tool.” found on beaches of the Indian Ocean suggested that they In fact he believes that the future of ocean modeling may ? oated for months on the ocean surface,” he says. “In theory, follow a similar pathway to that of weather prediction, where given the right information, trajectories could be simulated in ensembles of model runs are performed to get a feeling for the hope to locate the ? aperon’s possible start position, and multiple potential future states of the ocean – not just one. hence shed some light on the location of the demised aircraft.” “The problem is,” he says, “that even with perfect initial

And that’s exactly what they did. By using their model and data, there is a horizon of predictability, because after a cer- back tracking the debris using a method called Lagrangian tain time, prediction becomes more dif? cult. The ocean has analysis, the researchers were able to estimate the location of intricate internal dynamics – which are chaotic – and so a the plane. Durgadoo described the process in a 2016 article. numerically simulated ocean will diverge from observations “The idea was that we could use an ocean model to track the over time.” ? aperon back in time to establish the ? ight’s crash location. “Better numerics and parameterization will improve the

But the ocean is a chaotic place; it makes no sense to simulate ocean’s predicted mean state and variability,” he says. “But the path of a single ‘virtual ? aperon’ backward in time. There- the overall computational effort is rather big.” fore a ‘strength in numbers’ strategy is what we used when we “So our ability to simulate the ocean will improve, but grad- placed close to ? ve million virtual model ? aperons around La ually.”

March 2018 22 MTR

MTR #2 (18-33).indd 22 MTR #2 (18-33).indd 22 2/20/2018 3:34:01 PM2/20/2018 3:34:01 PM