Improving species distribution models for invasive non‐native species with biologically informed pseudo‐absence selection. (7th April 2019)

• Record Type: Journal Article
• Authors: Chapman, Daniel; Pescott, Oliver L.; Roy, Helen E.; Tanner, Rob
• Is Part Of: Journal of biogeography. Volume 46:Number 5(2019:May)
• Page Start: 1029
• Page End: 1040
• Publication Date: 2019-04-07
• Subjects: benchmarking -- Biomod -- climate envelope -- ecological niche model -- invasive alien species -- Maxent -- pest risk assessment -- presence–absence -- presence‐background -- presence‐only; Biogeography -- Periodicals; 578.09
• DOI: 10.1111/jbi.13555
• Languages: English
• ISSNs: 0305-0270

Abstract

Abstract: Aim: We present a novel strategy for species distribution models (SDMs) aimed at predicting the potential distributions of range‐expanding invasive non‐native species (INNS). The strategy combines two established perspectives on defining the background region for sampling "pseudo‐absences" that have hitherto only been applied separately. These are the accessible area, which accounts for dispersal constraints, and the area outside the environmental range of the species and therefore assumed to be unsuitable for the species. We tested an approach to combine these by fitting SDMs using background samples (pseudo‐absences) from both types of background. Location: Global. Taxon: Invasive non‐native plants: Humulus scandens, Lygodium japonicum, Lespedeza cuneata, Triadica sebifera, Cinnamomum camphora . Methods: Presence‐background (or presence‐only) SDMs were developed for the potential global distributions of five plant species native to Asia, invasive elsewhere and prioritised for risk assessment as emerging INNS in Europe. We compared models where the pseudo‐absences were selected from the accessible background, the unsuitable background (defined using biological knowledge of the species' key limiting factors) or from both types of background. Results: Combining the unsuitable and accessible backgrounds expanded the range of environments available for model fitting and caused biological knowledge about ecological unsuitability to influence the fitted species‐environment relationships. This improved the realism and accuracy of distribution projections globally and, generally, within the species' ranges. Main conclusions: Correlative SDMs remain valuable for INNS risk mapping and management, but are often criticised for a lack of biological underpinning. Our approach partly addresses this concern by using prior knowledge of species' requirements or tolerances to define the unsuitable background for modelling, while also accommodating dispersal constraints through considerations of accessibility. It can be implemented with current SDM software and results in more accurate and realistic distribution projections. As such, wider adoption has potential to improve SDMs that support INNS risk assessment.