Predicting the spread of aquatic invaders: What can we learn from 200 years of continuous invasion by zebra mussels?
Dispersal plays a fundamental role in the context of the effects of global habitat fragmentation, climate change, and biological invasions. Understanding factors controlling dispersal and spread of species is crucial to improve the management of natural populations. Dreissena polymorpha is considered the most aggressive freshwater invader in the northern hemisphere, and is a convenient model for invasion biology offering one of the best aquatic examples for quantifying the invasion process. Using data on 553 of the 1,040 glacial lakes in the Republic of Belarus (Europe) that were examined for the presence of zebra mussels in 1971-1996 we developed a predictive model of Dreissena spread. The major goal of the current study was to test the model by re-examining lakes that were free of zebra mussels during the initial survey. In this case, we had the rare opportunity to use a predictive model, test its performance through time, and then modify that model and improve our understanding of the key parameters associated with the future spread of an important invader. We used the Random Forests classification algorithm with 16 environmental variables to determine the most important factors that differed between invaded lakes and those lakes suitable for invasion that have not yet been invaded.
In spite of 200 years of continuous invasion, by 1996 zebra mussels were found in only 16.8% of all lakes studied. Of those lakes without zebra mussels in 1996, 66% were predicted to be susceptible to invasion by zebra mussels in the future, and 33% were predicted to be immune to successful invasion due to their water chemistry. Eighty lakes free of zebra mussels in 1996 were reexamined from 1997-2008. Of these, zebra mussels successfully invaded an additional 31 lakes, all of which were classified initially as suitable for zebra mussels; none of the lakes previously classified as unsuitable were invaded. Distance to the nearest infested lakes was found to be the most important variable, followed by the lake area, color, average depth, concentration of chloride, magnesium and bicarbonate. This study provides a useful approach for predicting the spread of an invader across a landscape with variable habitat suitability that can be applied to a variety of species and systems.