Predicting establishment of non-native fishes in Greece: identifying key features <br />

Abstract

Read online

Non-native fishes are known to cause economic damage to human society and are considered a major threat to biodiversity loss in freshwater ecosystems. The growing concern about these impacts has driven to an investigation of the biological traits that facilitate the establishment of non-native fish. However, invalid assessment in choosing the appropriate statistical model can lead researchers to ambiguous conclusions. Here, we present a comprehensive comparison of traditional and alternative statistical methods for predicting fish invasions using logistic regression, classification trees, multicorrespondence analysis and random forest analysis to determine characteristics of successful and failed non-native fishes in Hellenic Peninsula through establishment. We defined fifteen categorical predictor variables with biological relevance and measures of human interest. Our study showed that accuracy differed according to the model and the number of factors considered. Among all the models tested, random forest and logistic regression performed best, although all approaches predicted non-native fish establishment with moderate to excellent results. Detailed evaluation among the models corresponded with differences in variables importance, with three biological variables (parental care, distance from nearest native source and maximum size) and two variables of human interest (prior invasion success and propagule pressure) being important in predicting establishment. The analyzed statistical methods presented have a high predictive power and can be used as a risk assessment tool to prevent future freshwater fish invasions in this region with an imperiled fish fauna.

Keywords

• Logistic regression
• random forest
• biological invasions
• Invasion pathway
• Non-native fishes
• CART analysis