New fishery forecasting model ditches equations and assumptions, improves accuracy

By on March 26, 2015
Mature sockeye salmon approaching their spawning grounds in the Fraser watershed, British Columbia, Canada. (Credit: Shane Kalyn / 4 Element Photos)

What do financial markets, natural ecosystems and one of Canada’s most vital salmon fisheries have in common? A new study from the Scripps Institution of Oceanography found that methods once used to model ecosystems and economies are particularly effective at forecasting fishery performance in British Columbia’s Fraser River.

The method known as empirical dynamic modeling showed greater accuracy in predicting 2014 sockeye salmon recruitment rates at Lake Shuswap than the fishery’s official, traditional forecasting technique. The paper describing the new method was published in the Proceedings of the National Academy of Sciences.

“Classical fishery models — and indeed all equation-based models — are based on assumptions that essentially say we understand how the system works,” said George Sugihara, developer of the method and professor at the Scripps Institution at the University of California San Diego. “However, these models are really just hypotheses written down using the language of mathematics. Writing something down as an equation does not make it true, or even reasonable.”

Unlike traditional fishery modeling, empirical dynamic modeling is based on time series data established for each spawning stock rather than static equations. Classical fishery models, Sugihara says, assume an unchanging environment, and that fish stocks exist independently of other species in an ecosystem.

Population assessments of sockeye salmon conducted near their Fraser River spawning grounds in British Columbia, Canada. (Credit: Shane Kalyn / 4 Element Photos)

Population assessments of sockeye salmon conducted near their Fraser River spawning grounds in British Columbia, Canada. (Credit: Shane Kalyn / 4 Element Photos)

“The fact that these equation-based models do not predict very well should not be surprising,” he said. Some other recent models attempt to limit assumptions, but are still based on hypotheses. “The bottom-line is that they are not successful at real-time prediction.”

Sugihara and graduate student Hao Ye tested the method on nine major spawning stocks in the Fraser River system, including Lake Shuswap, one of the most important sockeye recruitment locations in the system. Collaborator Sue Grant of Fisheries and Oceans Canada provided data for the models, representing stock counts and environmental factors such as temperature and river flow.

The study’s biggest challenge was explaining empirical dynamic modeling to colleagues in fishery forecasting, Sugihara said. While the ideas behind the method aren’t hard to understand, Sugihara said that he and his co-authors met some resistance due to to the considerable differences between the new method and its forebearers.

In spite of that resistance, the study’s authors seem confident that their method will be incorporated into official fishery forecasts. Their commentary in the paper even suggests that moving away from equation-based — and therefore assumption-based — modeling could change the way scientists think about and communicate their work.

“EDM does not make such assumptions,” Sugihara said. “It lets the data speak for itself.”

Top image: Mature sockeye salmon approaching their spawning grounds in the Fraser watershed, British Columbia, Canada. (Credit: Shane Kalyn / 4 Element Photos)

About Alex Card

Leave a Reply

Your email address will not be published. Required fields are marked *