New research on managing aquatic invasive species in Canada combines the power of machine learning with expertise in biology and statistics to build a simple, easy-to-use tool for environmental managers.
The tool—developed by Mark Lewis, Russ Greiner, and their UAlberta research collaborators—helps
environmental managers decide which approach to take when dealing with invasive species in their waterways by predicting the outcomes of various invasive species management strategies.
The research combines the University of Alberta’s expertise in mathematical biology and machine learning, with a decades-long track record of research excellence in the area.
“The economic cost of invasive species is in the tens of billions of dollars—and because the cost is so high, there is a great deal of interest in preventing and controlling invasions,” explained Lewis professor of mathematics and biological sciences and Canada Research Chair in Mathematical Biology.
There are many kinds of aquatic invasive species in Canada in both marine and freshwater environments, ranging from invasive seaweed to fish to water fleas. Many are spread by shipping routes and through pet trade, like when you release your goldfish into a lake.
Applying the power of AI
“Mathematical biology has many wonderful tools for understanding the eradication of invading species, in general. Machine learning provides many complementary tools, for predicting whether a specific invasion is likely to be eradicated, using some specified approach,” explained Greiener, professor of computing science and principal investigator with the Alberta Machine Intelligence Institute.
Lewis and Greiner, with their former postdoctoral fellow Yanyu Xiao, ran machine learning techniques that produced a simple, easy-to-use decision trees (like the one pictured). From here, environmental managers can use the trees to determine the best strategy to take to deal any invasion, ranging from eradication to containment and mitigation of spread. The research shows that, over the wide of species considered, there are three main elements to consider: habitat type (e.g. sub-tidal or lake), amount of area invaded, and whether containment attempts have been made.
“As the end user, having a computer isn’t necessary,” says Lewis. “All you need is a copy of the decision tree, which is much more useful for environmental managers working in the field.”
The paper, “Evaluation of machine learning methods for predicting eradication of aquatic invasive species,” is published in Biological Invasions.