Performance tradeoffs in target-group bias correction for species distribution models

question to the fi tting of statistical distribution models to P-O datasets (i.e. model calibration; Chefaoui and Lobo 2008, Vanderwal et al. 2009, Barbet-Massin et al. 2012). A fundamental assumption in SDM calibration is the unbiased sampling of available conditions in the environmental space (Araú jo and Guisan 2006, Pearce and Boyce 2006). However, sampling bias is a widespread issue since the vast majority of SDMs are calibrated with opportunistic data, which are not collected following strict random or systematic sampling methods (Yackulic et al. 2012). Th is issue is even more problematic in global datasets (Loiselle et al. 2003, 2008, Newbold 2010, Boitani et al. 2011). If the existing geographic bias correlates with gradients in environmental conditions, P-O SDM predictions and performance are likely aff ected (Kadmon et al. 2004, Phillips et al. 2009). Th is might happen when observers have Ecography 39: 001–012, 2016 doi: 10.1111/ecog.02414 © 2016 Th e Authors. Ecography © 2016 Nordic Society Oikos Subject Editor: Christine Meynard. Editor-in-Chief: Miguel Ara ú jo. Accepted 10 August 2016