SUPPORTING INFORMATION
Predicting the future effectiveness of protected areas for bird conservation in
Mediterranean ecosystems under climate change and novel fire regime scenarios
APPENDIX S4: Complete description of binary conversion procedure: Sensitivity analysis in selecting thresholds.
To quantify the changes in the effectiveness of N2000 between 2000 and 2050, high quality habitats (hereafter optimal habitats) for the species need to be firstly identified .
We applied cut-off points or thresholds that convert maps with continuous suitability habitat values into maps with more areas hierarchically prioritised according to their relevance for the species. Thus, probability outputs were hierarchically ranked in two levels of increasing suitability. Among the different methods available to transform model probability outputs into maps of suitable-unsuitable areas (Liu et al., 2005), we followed Arcos et al. (2012) and Herrando et al. (2011). Thus, for all species, the threshold below which the absence areas for species are defined corresponds to the average tenth percentile of the data used for developing the models (i.e. 10% of the presence data with the lowest suitability). (Pearson et al., 2007). The average mean of the values within the areas above the first threshold provided a second threshold allowing the identification of optimal habitat areas (i.e. areas with habitat quality above the mean), which can be interpreted as critical habitats for our study species in line with the conservation mandate of the European directives.

This appendix provides a sensitivity analysis assessing the effects of the selection of different cut-off points or thresholds that convert continuous, probability maps into categorical maps for which optimal habitat is identified. In particular, we used the average of 10 th
percentile of the data used for developing the models (i.e. 10% of the presence data with the lowest suitability); the average of 20 th
percentile of the data (i.e.
20% of the presence data with the lowest suitability, and the lowest suitability value of the presence data as first threshold (see Methods section for a detailed description of procedure). The Fig. S4 shows the changes in the effectiveness of N2000 to preserve the optimal habitats for six representative species (i.e. covering the different pattern of response to our future scenarios) of the whole suite of target species.
The results showed that the direction of the effect (positive/negative) and the response patterns under different scenarios for the species analyzed remained unchanged across the different thresholds used. Our conclusions are therefore consistent independently of the use of different thresholds. While differences between thresholds in species habitat availability inside and outside protected areas were minor in terms of changes in protected area effectiveness, it is interesting to note that in larger thresholds (20 th percentile), were more restrictive in selecting optimal habitats for species and reduce the overall amount of species habitat preserved in protected areas.
We finally chose the percentile 10 th
as threshold for the following reasons:
1) The 10th percentile is a widely-used measure since the lower percentile, the lower the error of omission (the model predicts absence in areas where the

species is found), and the greater the sensitivity of the model (Pearson et al.,
2007).
2) Applying low percentiles allow us to adopt a more conservative outlook of the loss of habitat for many species which is very appropriate given the inherent uncertainty associated to modeling approaches (see new section in the discussion).
3) As we work with relative (instead of absolute) units of change and the thresholds are equally applied inside than outside N2000, we found minimal changes in the effectiveness between thresholds so our conclusions on the increasing role of
N2000 in the near future remains the same (see fig. S4 in the new appendix S4).
References
Arcos J.M., Bécares J., Villero D., Brotons L., Rodríguez B., & Ruiz A. (2012)
Assessing the location and stability of foraging hotspots for pelagic seabirds: An approach to identify marine Important Bird Areas (IBAs) in Spain. Biological
Conservation , 156 , 30–42.
Herrando S., Brotons L., Estrada J., Guallar S., & Anton M. (Eds. ). (2011) Catalan
Winter Bird Atlas 2006–2009.
Institut Català d’Ornitologia and Lynx Edicions,
Barcelona.
Liu C., Berry P.M., Dawson T.P., & Pearson R.G. (2005) Selecting thresholds of occurrence in the prediction of species distributions. Ecography , 28 , 385–393.
Pearson R.G., Raxworthy C.J., Nakamura M., & Townsend Peterson A. (2007)
Predicting species distributions from small numbers of occurrence records: a test case using cryptic geckos in Madagascar. Journal of Biogeography , 34 , 102–117.
Figure S4.
Predicted percentage of change in the effectiveness of N2000 for preserving the optimal habitats for the target species between 2000 and 2050 under each scenario of future environmental changes according to the combined models and using: a) the lowest suitability value of the presence data, b) the average of 10 th percentile and c) the average of 20 th percentile as a first threshold. The whiskers indicate the 5% and 95% percentiles, the hinges indicate the first and third quartiles (in blue and grey for
A2 and B2 climate change scenarios, respectively) and the central black line indicates the median value

across the different MEDFIRE runs simulating land cover changes. See Tables 1 and 2 for scenario and species acronyms, respectively.