Assessing the environmental requirements of invaders using ensembles of distribution models

Abstract

Aim Understanding the environmental factors determining the establishment of invasive populations is a crucial issue in the study of biological invasions. By taking into account the uncertainty of predictions, ensembles of niche-based models can provide useful information. Therefore, we explored the use of consensus techniques to generate a quantitative description of the environmental conditions favouring the establishment of four problematic invasive decapods: Cherax destructor, Eriocheir sinensis, Pacifastacus leniusculus and Procambarus clarkii.Methods We collected both native and invasive distribution records from multiple sources. From these data, we modelled the potential distribution of the four decapod species using eight correlative models comprising regression, classification and machine learning methods. The relative influence of the environmental variables in single models was averaged to achieve a consensus contribution of the variables. Ecological requirements were investigated by means of consensus suitability curves, a spatial analysis procedure that shows the variation of consensus suitability along the gradients of environmental variables. ResultsThe predictive accuracy of single models ranged from fair to very good. Still, the variability between predictions was high. Similarly, the influence of each variable in different models was also uneven. Consensus analysis identified the variables related to temperature as highly influential for all invaders. Consensus suitability curves show that C. destructor and Procambarus clarkii have reduced suitability in colder areas whereas the suitability for P. leniusculus is greatly reduced in warmer areas. The distance to the ocean was highly influential in E. sinensis models, with suitability showing an exponential decay as distance increased.Main conclusions We show that the information about the species-environment relationships obtained from niche-based models is highly dependent on the characteristics of the models used. In this context, we demonstrate that ensembles of models and consensus approaches can be used to identify such relationships while also allowing the assessment of the uncertainty of the achieved knowledge.

The effect of sample size on the accuracy of species distribution models: considering both presences and pseudo‐absences or background sites

“…Equal numbers of pseudo-absences or background sites as presences have been used for GLM and GARP models by Stockwell and Peterson (2002), for eight modelling techniques (including artificial neural network -ANN, BRT, classification tree analysis -CTA, GLM, GAM, MARS, mixture discriminant analysis -MDA and RF) by Capinha and Anastácio (2011), and for BRT, GLM, MARS, GARP and Maxent by Mateo et al (2012). Some have adopted specific ratios of the number of pseudo-absences or background sites to the number of presences.…”

Early assessment of the impact of alien species: differential consequences of an invasive crayfish on adult and larval amphibians

“…The American red swamp crayfish, Procambarus clarkii, is native of eastern North America and Mexico, but has been introduced world-wide and is currently invasive in wide areas of Europe (Gherardi & Panov, 2009;Capinha & Anastácio, 2011). Procambarus clarkii is a keystone IAS that can have multiple consequences on both native species and ecosystem functioning.…”

Projection of red spruce (Picea rubens Sargent) habitat suitability and distribution in the Southern Appalachian Mountains, USA

“…SDMs have recently considered multiple factors to predict more realistic ranges of organisms. For example, SDMs developed based on maximum entropy theory using Maxent and ensemble approaches considered many factors, such as vegetation classes, climate factors, topographic factors (Araújo and New, 2007;Capinha and Anastácio, 2011;Elith et al, 2010;Larson and Olden, 2012;Marmion et al, 2009;Phillips et al, 2006;Thuiller et al, 2009). These models projected species distributions based on direct correlations between Fig.…”