Profiling ecosystem vulnerability to invasion by zebra mussels with support vector machines

Decades since the initial establishment of zebra mussels (Dreissena polymorpha) in North America, understanding and controlling the invasion of aquatic ecosystems continues to be a problem in continent-wide conservation and landscape management. While the high economic and conservation burden of this species makes accurate predictions of future invasions a research priority, forecasting is confounded by limited data, tenuous model assumptions, and the stochasticity of the invasion process. Using a new method for niche identification, we profiled invasion vulnerability for 1,017 lakes in the Great Lakes region of the Unites States. We used a nonparametric geoadditive regression model to test for effects of two water quality variables on the present distribution of zebra mussels. We then used the support vector data description (SVDD), a support vector machine for one-class classification, to estimate the boundary of the ecological niche. By disentangling niche estimation from distributional assumptions, computational niche models could be used to test an array of fundamental concepts in ecology and evolution, while species invasions forecasting is representative of the wide range of potential applications for niche identification in conservation and management.