Jan. 5, 2008 Animal populations and the stock market are hard to forecast. Both are generated by complicated, interdependent systems.
Unlike financial stocks, where trades are meticulously recorded, scientists began estimating animal populations only a few decades ago. But a new technique makes it possible to use the same tools some banks use to forecast the stock market and apply them to ecology.
The newly developed "Dewdrop Regression" can forecast fish populations with 3% the data previously required through other methods, according to Hsieh, Anderson, and Sugihara in an article appearing in The American Naturalist.
The migration of the forecast tools from finance to ecology parallels Dr. Sugihara's own journey. After proposing simplex projection in 1990 with Lord Robert May, later Chief Science Advisor of the UK and President of the Royal Society, Sugihara became a managing director of a major bank for several years.
Returning to academia and ecology, "I realized that even great ecologists were working with time series only a few tens of points long," Sugihara said. To apply data-hungry techniques to short time series, Hsieh et al. take data from several species collected simultaneously over a few years and stitch them together. A few test manipulations need to be applied; but when done properly, the technique is able to forecast with 15-20 points instead of 1,000. "You're doing significantly better than chance within four years," said Anderson.
But does it work for real world ecological problems? Using 40-year time series from 23 California fish species, Hsieh et al. showed that though they were <10% predictable alone, they become >60% predictable with the new procedure, combined with others from the same habitat. "When you consider that we're predicting the change, not just raw abundance, this accuracy becomes very exciting," Anderson said.
This research was published in the January issue of the American Naturalist.
Other social bookmarking and sharing tools:
Note: If no author is given, the source is cited instead.