Tapping into the same tools that banks and hedge funds have used to forecast financial markets, scientists at Scripps Institution of Oceanography at UC San Diego have developed a powerful new method for forecasting fish populations. The new technique could have a hand in future ocean management decisions that rely on ecosystem-based data.
Scripps professor George Sugihara applies his skills in applied mathematics and theoretical biology to explore natural systems and more fully understand their nonlinear or “chaotic” patterns, not unlike the way financial analysts attempt to forecast stock market fluctuations.
In 1990, Sugihara and Lord Robert May of Oxford University took a new mathematical concept for understanding such systems and extended it to make forecasts using data derived in nature.
This work led Sugihara to the world of investment banking where he consulted for Merrill Lynch and then became a managing director for Deutsche Bank Securities. After five years in the financial world, Sugihara returned to academia.
In a paper published in the latest issue of American Naturalist, Sugihara and his colleagues describe a newly developed technique called “dewdrop regression,” a method based on equations used for financial forecasting on Wall Street. The technique is gaining attention for its ability to make forecasts based on a small fraction of information required under other methods.
Dewdrop regression, Sugihara believes, could be an important tool in future ocean management plans as governments and municipalities institute ecosystem-based regulations.
“Dew drop regression allows one to make better forecasts of future stock abundances,” said Sugihara. “This knowledge is essential to setting harvest targets. For instance, managers would not want to set a high harvest target if the stocks are on the verge of decline, due to changing weather patterns, for example.
In describing the method in American Naturalist, the researchers took data from the California Cooperative Oceanic Fisheries Investigations for 23 California fish species and constructed population forecasts using only three percent of the information previously required to make similar population forecasts. Individually, each species was less than 10 percent predictable. When stitched together using dewdrop regression, however, population trends of the 23 species became more than 60 percent predictable.
The key behind dewdrop regression is its ability to combine data sets, broadening forecasting skill by melding ecologically similar species.
American Naturalist paper co-authors from Scripps include Chih-hao Hsieh and Christian Anderson.
--Mario C. Aguilera