New hybrid machine learning predicts lake ecosystem responses to climate change

In the mid-20th century, phosphorus inputs from detergents and fertilizers degraded the water quality of Lake Geneva in Switzerland, prompting authorities to take action to address the pollution in the 1970s.

“The obvious remedy was to reverse the phosphorus load, and that simple idea helped tremendously, but it didn’t return the lake to its previous state, and that’s the problem,” said biological oceanographer George Sugihara. at the Scripps Institution at UC San Diego. of Oceanography.

Sugihara, Ethan Deyle of Boston University and three international colleagues spent five years researching a better way to predict and manage Lake Geneva’s ecological response to the threat of phosphorus pollution, to which must now be added the effects of climate change. The team, which includes Damien Bouffard from the Swiss Federal Institute of Aquatic Science and Technology, publishes its new hybrid approach to empirical dynamic modeling (EDM) on June 20 in the journal Proceedings of the National Academy of Sciences.

“Nature is much more interconnected and interdependent than scientists often would like to think,” said Sugihara, McQuown Professor of Natural Sciences at Scripps. EDM can help in this context as a form of supervised machine learning, a way for computers to learn patterns and teach researchers about the mechanisms behind the data.

“You pull a lever and everything else changes, like a mole. Single-factor experiments, the hallmark of 20th-century science where everything is held constant, can teach you a lot in principle, but that’s not how the world works,” he said.

“If this were not the case, if nature behaved more like single-factor experiments and were less connected and interdependent, we would be able to predict outcomes with simple models where relationships do not change.”

Interdependence and changing relationships are the reality of ecosystems and they are also the reality of financial markets where prediction is so difficult, Sugihara noted. EDM was refined in the crucible of financial forecasting from the mid-1990s to the early 2000s, when Sugihara was managing director of Deutsche Bank.

Sugihara has drawn on his financial background to design market tools to support sustainable marine fishing over the past 20 years at Scripps. He calls EDM “mathematics without equations”.

But EDM is not a black box method, Deyle said, referring to quantitative methods based on mysterious mathematical or computer formulas. It’s a criticism he says is often leveled at machine learning.

“Instead, it uses the data to tell you in the most direct way, with minimal assumptions, what’s going on. What variables matter? How are relationships changing over time? It has a mechanism and transparency that comes directly from the data.

What Sugihara’s team attempted is a departure from traditional modeling methods used over the past few decades. As Deyle notes, parts of well-established patterns are represented by constants.

“The fixed and constant force of gravity, or the shape and depth of a lake for example. Therefore, the physical processes in the lake can be modeled very well with simple equations,” he said.

This is not the case for the evolution of ecology and biochemistry.

“The organisms responsible for change in an ecosystem like that of Lake Geneva have changed over the past two decades. The food web has changed and is constantly changing, as has the biochemistry of the lake,” Bouffard said.

“Standard tools are ill-suited to such problems,” said Deyle, who earned her doctorate in biological oceanography from Scripps Oceanography with adviser Sugihara in 2015.

“Lake Geneva is one of the best-studied systems in the world. It is no coincidence that this was an opportunity to push the boundaries with a machine learning approach to ecological prediction,” Deyle said.

The authors demonstrate that their hybrid approach leads not only to better prediction, but also to a more actionable description of the processes (such as biogeochemical and ecological) that determine water quality.

Notably, the hybrid model suggests that the impact on water quality of an increase in air temperature of 3 degrees Celsius (5.4 degrees Fahrenheit) would be of the same order as the phosphorus pollution of the century. precedent, and that best management practices may no longer involve a single control lever such as reducing phosphorus inputs alone.

“One of the intellectual cornerstones of all of this is minimalism,” Sugihara said. “Extract insights from data with the fewest assumptions.”

A simple model that predicts target data that has not yet been collected is more convincing than a complex model that can agree with current thinking and can be designed to “fit” the story remarkably well. but does not actually “predict” events yet to be seen. This was the major problem in financial applications, where it’s easy to find things that “match”, but nearly impossible to find anything that “predicts”.

“The more complicated something is, the easier it is to get it wrong,” he said. “Our hybrid approach seems to have a balance that works.”

This research was funded by the US Department of Defense’s Strategic Environmental and Development Program, the National Science Foundation, the Department of the Interior, the McQuown Fund, and the McQuown Chair of Natural Sciences, UC San Diego. Study co-authors include Victor Frossard, University of Savoie Mont Blanc; Robert Schwefel and John Melack, University of California at Santa Barbara.

/Public release. This material from the original organization/authors may be ad hoc in nature, edited for clarity, style and length. The views and opinions expressed are those of the author or authors. See in full here.