Are different facets of plant diversity well protected against climate and land cover changes ? A test study in the French Alps

W. Thuiller (wilfried.thuiller@ujf-grenoble.fr), M. Guéguen, D. Georges, L. Chalmandrier, J. Renaud, C. Roquet and S. Lavergne, Laboratoire d’Ecologie Alpine, UMR CNRS 5553, Univ. Joseph Fourier – Grenoble 1, BP 53, FR-38041 Grenoble Cedex 9, France. – L. Garraud and J. Van Es, Domaine de Charance, Conservatoire Botanique National Alpin, Gap, FR-05000, France. – R. Bonet, Parc National des Ecrins, Gap, FR-05000, France. – N. E. Zimmermann, Landscape Dynamics, Swiss Federal Research Inst. WSL, CH-8903 Birmensdorf, Switzerland.

bring the same level information. For instance, if a 2.5x2.5km pixel has been sampled twice using single occurrence method, they will have a maximum of 2 species. Instead, if the same sites were sampled using a phytosociological method, the actual number of species could range from 1 species to more than one hundred.
We thus built two maps to represent the sampling effort, one for each sampling method. For each sampling method, the weight value of each pixel corresponds to the number of sampling units. Each map is then re-scaled by the maximum of sampling units in the study area for a given sampling method. To give more weight to the phytosociological method that is more complete in terms of sampling, we multiplied the final map by 0.7 and the final map for single occurrence method by 0.3. The two maps were then summed to give a single weighing map of each pixel in function of the type and number of sampling units.
We used this protocol to build a single weighing map for each of the incremental grid: 250m, 1km, 2.5km and 5 km. For each resolution, we then compared the observed and projected species richness weighted by the sampling effort map (Fig. S1). We also tested the sensitivity of our differential weighing protocol for the two sampling methods.
The 2.5km resolution was finally retained given it was the best trade-off between high resolution and robustness. The differential weighing protocol did not influence the results (Fig. S1).
Description of the three selected regional climate models.
We selected three different Regional Climate Models (RCMs) fed by three different Global Circulation Models (GCMs) in turn, to reflect variation in the degree of projected warming by 2100. RCMs downscale the output from GCMs for a given region by taking the GCM output as boundary input, while the processes within the study area are downscaled based on physical, meteorological processes. Such downscaling is usually performed to a spatial resolution of ca. 20x20km (10'x10') or similar. We selected the following pairs of RCMxGCM for our analyses: HadRM3xHadCM3, CLMxECHAM5, and RCA3xCCSM3.
These three model combinations are provided to the user community by the EU project ENSEMBLES (http://www.ensembles-eu.org), in which a larger number of RCMs runs were produced to reflect the best current knowledge on the future of the European climate.
We selected these three model combinations, because they represent well the variability of the climate future presented in ENSEMBLES for the A1B scenario. The runs by HadRM3xHadCM3 represent a high degree of warming (~4.9°C and ~5.0°C warming of annual or summer temperature) as is illustrated in the figure A2 below. The runs by CLMxECHAM5 represent an average degree of warming compared to all ENSEMBLE model runs (ca. +3.8°C for both annual and summer temperature). Finally, the RCA3xCCSM3 runs project a low degree of warming relative to all ENSEMBLE runs (ca. +2.5°C and +2.3°C for annual and summer temperature). All three RCM x GCM model combinations project a relative decrease in summer precipitation (-8 --12%) while there is no clear trend in annual precipitation change.        A1b.rca and A2.rca: CCSM3xRCA3 climate model driven by the A1b and A2 scenarios and the GRASS and BAMBU storylines, respectively. The protected area network corresponds here to protected areas with sustainable use of natural resources (Ia, II, III, IV, V and Natura2000).