Abstract
AbstractAimSpecies distribution models (SDMs) are widely used to forecast potential range expansion of invasive species. However, invasive species occurrence datasets often have spatial biases that may violate key SDM assumptions. In this study, we examined alternative methods of spatial bias correction and multiple methods for model evaluation for seven invasive plant species.LocationNorth America.TaxonCommon Tansy (Tanacetum vulgare), Wild Parsnip (Pastinaca sativa), Leafy Spurge (Euphorbia virgata), Common Teasel (Dipsacus fullonum), Brown Knapweed (Centaurea jacea), Black Swallowwort (Vincetoxicum nigrum) and Dalmatian Toadflax (Linaria dalmatica).MethodsWe employed bias‐correction measures for both occurrence sampling and background sampling inputs in a factorial design for Maxent resulting in six potential models for each species. We evaluated our models for complexity, model fit and using commonly employed evaluation metrics: AUC, partial AUC, the continuous Boyce index and sensitivity. We then developed a structured process for model selection.ResultsModels developed without occurrence or background bias correction often were overly complex and did not transfer well to expanding range fronts. Conversely, models that employed occurrence and/or background bias‐correction measures were less complex, had better AICc scores and had greater projection into incipient areas. These simpler models were also more likely to be selected when evaluated using a process that integrated multiple evaluation metrics. We found that invasion history (e.g. established versus incipient) was associated with the effectiveness of spatial bias correction techniques.Main ConclusionsWhile challenges exist in building climate‐based correlative species distribution models for invasive species, we found that methods relying on maximizing AUC performed poorly for invasive species. We advocate for the use of multiple and diverse metrics for model evaluation. Users of species distribution models need to incorporate explicit consideration of model discrimination, model fit and model complexity into their decision‐making processes if they are to build biologically realistic models.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.