Risk of invasion predicted with support vector machines: a case study on northern pike (Esox Lucius, L.) and bleak (Alburnus alburnus, L.)

Abstract

The impacts of invasive species are recognised as a major threat to global freshwater biodiversity. The risk of invasion (probability of presence) of two avowed invasive species, the northern pike (Esox Lucius, L.) and bleak (Alburnus alburnus, L.), was evaluated in the upper part of the Cabriel River (eastern Iberian Peninsula). Habitat suitability models for these invasive species were developed with Support Vector Machines (SVMs), which were trained with data collected downstream the Contreras dam (the last barrier impeding the invasion of the upper river segment). Although SVMs gained visibility in habitat suitability modelling, they cannot be considered widespread in ecology. Thus, with this technique, there is certain controversy about the necessity of performing variable selection procedures. In this study, the parameters tuning and the variable selection for the SVMs was simultaneously performed with a genetic algorithm and, contradicting previous studies in freshwater ecology, the variable selection proved necessary to achieve almost perfect accuracy. Further, the development of partial dependence plots allowed unveiling the relationship between the selected input variables and the probability of presence. Results revealed the preference of northern pike for large and wide mesohabitats with vegetated shores and abundant prey whereas bleak preferred deep and slightly fast flow mesohabitats with fine substrate. Both species proved able to colonize the upper part of the Cabriel River but the habitat suitability for bleak indicated a slightly higher risk of invasion. Altogether may threaten the endemic species that actually inhabit that stretch, especially the Júcar nase (Parachondrostoma arrigonis; Steindachner), which is one of the most critically endangered Iberian freshwater fish species.

More information

Identifiers

Catalogue record

Export record

Altmetrics