Finding what you don’t know: Testing SDM methods for poorly known species

Aim A limitation of species distribution models (SDMs) is that with low sample sizes are difficult to model. Yet, it often important know the habitat associations poorly known guide conservation efforts. Techniques have been proposed for modelling species’ distributions from a few records, but their performance relative one another has not compared. Because these built and evaluated small data sets, sampling error could cause severely biased in environmental space. As result, SDMs likely underpredict geographic given sizes. We perform first comparison methods explicitly promoted or developed predicting ranges very Location North Carolina, USA. Taxon South Mountains Grey-cheeked Salamander (Plethodon meridianus). Methods Using sparse, existing georeferenced records P. meridianus, we using range previous researchers argued should work then tested each SDM’s ability accurately predict independent survey were prior our study. compared omission AUC. Results Roughly half successfully predicted centre, all had high rates exterior. In interior exterior, ‘ensemble models’ technique produced rates. Spatial filtering negligible impact on model performance. Most, all, outperformed predictions distance populations. best-performing methods, an improved map meridianus. Main Conclusions Geographically peripheral populations SDMs, though some clearly inferior set. recommend when low, use Maxent species-specific settings.