Different approaches to assess the local invasion risk on a threatened species: Opportunities of using high-resolution species distribution models by selecting the optimal model complexity

Abstract Species distribution models (SDMs) are potential tools for the evaluation of the risk of biological invasions to threatened plant species and to prioritize areas for local conservation initiatives. This study seeks to analyze if the use of predictions of less overfit high-resolution SDMs can be highly accurate and can provide a spatial criterion on where the management actions for this local conservation initiatives can be focused, even when being calibrated with sub-optimal datasets of the species distribution ranges. We developed this approach in the Galician massif of Monte Pindo (NW Spain) with a small disjunct isolated population of Quercus lusitanica, which could be threatened by the expansion of Paraserianthes lophantha, a recently detected potentially invasive species from Australia. We compared the predictions of Maxent SDMs calibrated in different geographic extensions of the species distribution range, fitted with different variables, resolutions and levels of complexity. Two different extensions, a small local area restricted to Monte Pindo for both species and a greater regional area extended to the Galician invaded range for P. lophantha, were selected to fit SDMs using bioclimatic, topographic and land cover variables at fine-grain 25 × 25 m spatial resolution. They were compared there with those of SDMs fitted in the whole native ranges of each species with climate (WORLDCLIM datasets) and bioclimatic variables coarse-grain 1 × 1 km spatial resolution. The native models of P. lophantha were projected onto the invasive range environmental variables. Tuning experiments of Maxent regularization multiplier setting were applied to minimize the overfitting. Quantitative and qualitative evaluations were used to select the optimal SDMs of each species, which were combined to build Risk Models (RMs) in the small local area. RMs enabled the assessment of the invasive potential risk of P. lophantha on the population of Q. lusitanica. Results show that the use of high-resolution models, either less overfit or large-scale, can be highly accurate and provide a better spatial criterion of where the conservation management actions can be focused than the models calibrated with all the available data on the geographical native range of the species. This comes in useful to prioritize areas for local conservation initiatives, being especially effectual with early-stage biological invasions. SDMs and RMs generated using the approach of this study could be a good starting point to take action against P. lophantha, which may be a threat to the Monte Pindo disjunct population of Q. lusitanica.