Although there are varying definitions for the mediterranean climate, we chose one definition that can be readily mapped with available climate data and has minimal over prediction into areas that are not part of the mediterranean biome (see supporting information Text S1 for further discussion and sensitivity analysis). According to this definition, published by Aschmann [17], an area is within the MCE if it meets five conditions; 1) The winter must be wetter than the summers (>65% of the precipitation falls in the winter half of the year), 2) the annual precipitation must not be too low (>275 millimeters (mm)), 3) nor too high (<900 mm), 4) the winter must be cool (<15°Celsius (C) mean temperature for the coldest month of the year), but 5) it can not have too much frost (<3% of the annual hours are below freezing). We used the WorldClim [18] high resolution (2.5 arc-minutes or ~5 kilometer (km) horizontal resolution at the equator) grids of global climate data summaries from 1960–1990 to map the current MCE where all five Aschmann conditions are met (see supporting information Text S1 for more information about the WorldClim dataset).

Data for projections of future climate conditions were derived from the results of the AOGCMs run to support IPCC's Fourth Assessment Report. The data [World Climate Research Programme's (WCRP) Coupled Model Intercomparison Project (CMIP) phase 3 multi-model dataset] include seven future emissions scenarios. Three of these scenarios are used most often by modeling groups and are considered representative of low (B1 or stabilization at 550 ppm atmospheric CO₂), moderate (A1b or stabilization at 720 ppm atmospheric CO₂) and high (A2 or no stabilization) emission trajectories [19]. We compiled the AOGCM output data for monthly surface air temperature and precipitation flux for the 20^th century and the 21^st century for three future emissions scenarios. While some modeling groups have generated multiple simulations for a given scenario and others have done no simulations for a given scenario, we analyzed all available AOGCM simulations in the CMIP multi-model dataset, including 48 low emission simulations, 52 moderate emission simulations, and 36 high emission simulations, for a total of 136 simulations of future climate (see supporting information Text S1 and Table S1). By doing so, we treat each AOGCM simulation for a given emissions scenario as a unique and probable experimental outcome and average the results, thereby elucidating a more robust set of potential climate outcomes.

To reduce the variability associated with annual climate projections, we averaged the monthly data in the AOGCM simulation results to two 30-year periods; one “current” and one “future”. The WorldClim data is primarily derived from 1960–1990 weather records, so we averaged the monthly data from January 1960 to December 1989 for each of the modeled 20^th century simulations to generate the current time period. The majority of the model simulations end in 2100, so we averaged the monthly data from January 2070 to December 2099 for each of the modeled 21^st century simulations to generate the future time period. We then subtracted the modeled current data from the modeled future data to reduce modeling biases and generate projected climate anomalies. For example, an AOGCM simulation may have modeled the average July temperature for a specific area to be 24°C for the current time period (1960–1989) and 27°C for the future time period (2070–2099), so the projected climate anomaly for that area would be 3°C. We used the change factor approach to downscale the projected climate anomalies from the coarse resolution of the AOGCMs to the finer resolution of the WorldClim data. This method involves interpolating the projected climate anomalies and adding the interpolated data to the current climatology (see supporting information Text S1).

We applied Aschmann's [17] conditions to generate binary maps of the projected future MCE for each AOGCM simulation. The size of the projected future MCE in each region was compared to the size of the current MCE for each AOGCM simulation, and the average and 5, 10, 25, 50, 75, 90, and 95 percentiles of the projected percent change for all simulations in a given emissions scenario were calculated. This provides both a measure of the range and the central tendency of the ensemble of projected changes [20]. We considered using a weighted-average, but since we do not have a testing dataset for the future locations of the MCE, and other studies have found little increases in predictive power with a weighted average compared to a non-weighted average [21], we used only the average.

We spatially combined the current MCE and all of the future MCE projections to calculate the percent of AOGCM simulations predicting an expansion or contraction of the MCE for each grid cell on the globe. We defined seven categories based on whether areas were in the current MCE or not, and the number of AOGCM simulations that project the area will be in the future MCE (Table 1). These categories were mapped using the suite of AOGCM simulations for each emissions scenario.

The majority of AOGCM and emissions scenario projections of the MCE at the end of the 21^st century (or future MCE) are larger than the MCE at the end of the 20^th century (or the current MCE). The median future MCE increases to 106, 107, or 111% of its current size, for the low, medium, and high emissions scenarios, respectively (Figure 1). However, this pattern is not consistent within each region. Instead, there is a disparity between the regions with some projected to experience an increase in the MCE in the future and some projected to experience a decrease in the MCE. Almost all of the AOGCM simulations project an increase in the MCE in the Mediterranean Basin with the median future MCE increasing to 115, 126, or 132% of its current size for the low, medium, or high emissions scenarios, respectively. The median projected increases are greater in Chile/Argentina, ranging from 129% for the low to 153% for the high emissions scenario. In the United States/Mexico region, the projected change in the MCE is less dramatic, with the median future MCE decreasing to 96, 95, and 94% of its current size, for the low, medium, and high emissions scenarios, respectively. In South Africa, greater than 90% of the AOGCM simulations project a decrease in the future MCE, with the median estimates ranging from 83% of the current MCE for the low to 60% for the high emissions scenario. In Australia, the projected area reduction is more extreme, with median estimates ranging from 77% of the current MCE for the low to 49% for the high emissions scenario.

By overlaying all of the future MCE projections, we were able to map all of the grid cells on the globe where the MCE is likely to expand, contract, and remain stable with varying levels of confidence (Figure 2). We show that there are areas of contraction even within regions that are projected to have a net increase in the MCE. For example, the median projection of the future MCE in the Mediterranean Basin is larger than the current MCE, but most AOGCM simulations project contractions in Morocco and in the Middle East. The geographic separation between the areas of contraction and expansion within each region highlighted in Figure 2 will have important implications for adaptation of native plants and animals with limited dispersal or migration capabilities.

In addition to showing where the MCE is projected to contract, we also wanted to determine if there was AOGCM agreement on why it will contract (e.g., a warmer winter temperature, or less annual precipitation). These changes will have important implications for the persistence of native plants and animals in the mediterranean biome. To do this, we determined the level of agreement among the AOGCM simulations on which Aschmann's conditions were no longer met for the areas with a projected contraction. We report the results for areas where at least 90% of the AOGCM simulations agree that the MCE will contract and agree on the changing climatic condition causing the contraction under the high emissions scenario. Across all five regions, we can project with the most confidence that 7.2% of the current MCE will contract. Over half of this projected contraction, or 4.0% of the current MCE, results from warming in the winter. Almost a quarter of the projected contraction, or 1.7% of the current MCE, results from a drop in total annual precipitation below the 275 mm threshold. For one fifth of the projected contraction, or 1.5% of the current MCE, the AOGCM simulations agree that an area will contract, but they do not agree on which condition will change (supporting information Table S2). By country, most of the loss in Australia, the United States, Iran, Israel, and Libya is attributable to a warming winter, while the majority of the loss in Argentina, South Africa, Morocco, and Syria is due to a drop in annual precipitation.

Current land conversion and protection status and configuration relative to these climatic changes will play an important role in determining the extrinsic adaptation potential for the species of the mediterranean biome. Approximately one third of the area in the current MCE has already been converted to agricultural and urban land uses. If most of the converted land is in areas where the MCE is projected to contract, extrinsic adaptation potential will not be significantly reduced because these areas are poor habitat for native species. However, if the areas projected to have a stable or expanded MCE are disproportionately converted, this will exacerbate the negative impacts of climate change on biodiversity in the mediterranean biome. When looking across all five regions, we found the land conversion patterns are similar in areas where the MCE is projected to contract with confidence (23%) and in areas projected to remain stable with confidence (29%), but the regional patterns were more variable (Figure 3). In California and Mexico, extrinsic adaptation potential is conserved in the future because most of the conversion lies in areas that are projected to contract, and there is little conversion in the areas of stability. Similarly, Chile/Argentina and South Africa also have low levels of conversion in the confident stable areas. The opposite is true in Australia, where 64% of the likely stable area and 49% of the confident stable area is already converted, greatly diminishing the extrinsic adaptation potential of native biota.

At 4%, the level of protection for biodiversity in the current MCE is below that of the more expansive mediterranean biome (5%) and well below the global average (12%) for all terrestrial biome types [5]. We wanted to determine if the level of protection is higher or lower in areas with high likelihood of retaining a mediterranean climate. For the entire biome, just over half of the existing protected areas are projected to retain the mediterranean climate with high confidence, even under the high emissions scenario (Figure 4). The projected status of protected lands in some regions is much better, as over 70% of the protected areas in California/Mexico, South Africa, and the Mediterranean Basin are in the confident stable areas. In stark contrast, the MCE is projected to contract or is uncertain in over 75% of the protected areas with mediterranean climate in Australia, and there is almost no projected expansion into other protected areas to offset this loss.

One of the first challenges to projecting the impacts of climate change on mediterranean ecosystems worldwide is defining the MCE. The extent of the mediterranean biome is typically mapped based on a combination of climate characteristics and plant assemblages that vary from region to region [2]. Projecting how plant assemblages will shift in response to climate change is subject to significant uncertainty because it requires compounding the uncertainty with projecting climate change with the uncertainty inherent in projecting future distributions of individual species [24], [25]. In this analysis, we minimize the uncertainty and focus on mediterranean climate shifts in the future. As such, we utilize a conservative definition of mediterranean climate [17] that is consistent across all mediterranean regions and minimizes “false-positives” (areas that are not considered part of the mediterranean biome). Some areas that are traditionally considered part of the mediterranean biome are not within our current or future MCE, including the south coast of France, western Italy, northeastern Spain, portions of central Chile, and the southern coast of South Africa. Most of these areas receive less than 65% of their rain in the winter, and thus do not meet the first Aschmann condition (Figure 5). Despite the conservative nature of this definition, we do include some false-positives, including parts of Argentina and the Middle East. These commissions could be the result of a lack of climate station data in these more remote and mountainous areas. We performed a sensitivity analysis with the Köppen definition of mediterranean climate, which is less conservative (Figure S1). This definition identifies more of the traditional areas, but also includes large areas of false-positives. However, when focusing in on the five mediterranean regions, the results of the projected fate of the MCE are very similar using both definitions (see the full analysis in supporting information Text S1)

While there are some significant discrepancies between our map of the current MCE and commonly used maps of the mediterranean biome, such as the one mapped by Olson et al. [2], preliminary analysis indicates significant overlap between the current MCE and the mapped hotspots of plant richness and endemism within the mediterranean biome. Our current MCE corresponds well with the biogegraphical sectors with high incidences of plant endemism in the Mediterranean Basin [7], areas of high modeled endemic Banksia species richness in western Australia [11], and areas of high modeled endemic plant richness in the California Floristic province [14]. In South Africa, an analysis of both modeled and observed plant richness shows the area identified in the current MCE in the Western Cape as a hotspot [26]. This correlation between areas of high plant richness and the MCE does not diminish the need for conservation action in all areas of the mediterranean biome, but it does provide support for the theory that the core mediterranean climate is an important driver of plant endemism and diversity, and that changes in climate could threaten the survival of these endemic plants.

Across the entire mediterranean biome, most of the 23 AOGCMs project a minor increase in the MCE, with significantly large increases in some regions, and significant decreases in others. Not surprisingly, the physical characteristics in each region help to explain some of the disparity between areas of contraction and expansion. High topographic diversity and contiguous land toward the nearest pole provide room for the expansion of the MCE in the United States, Chile, Argentina, Greece, Turkey, Spain and Portugal. In South Africa, there is topographic relief so the MCE can retreat upslope into the Western Cape Fold Mountains, but there is no contiguous land toward the south pole cutting off future expansion of the MCE. Similarly in Morocco, the Atlas Mountains provide topographic diversity, but the Mediterranean Sea blocks expansion toward the north. Southwestern Australia is a flat highly weathered plateau [27] and is contained by the Indian Ocean to the south, resulting in the largest projected contraction of all the regions.

While this is the first global assessment of the impacts of climate change on the MCE, our results are consistent with other regional biome and species level analyses. The IPCC determined that the mediterranean biome as a whole is threatened by desertification from expansion of semi-arid and arid system under relatively minor warming and drying scenarios, and projects significant regional vegetation and species range shifts [28]. In South Africa, despite differences in the current spatial extent, Midgley et al. projected areas of future mediterranean biome contraction and stability in similar areas as our analysis [15]. Similarly, Williams et al. and Hannah et al. found the higher elevation areas of the Western Cape in South Africa support high levels of endemic Proteaceae species richness, and will be important habitat for dispersal by 2050, while the low-lying areas north of Cape Town are high in richness now, but are not protected to support the species in the future [29], [30]. These areas correspond well with our projected areas of higher elevation stability and lower elevation contraction in South Africa. Fitzpatrick et al. studied potential range shifts for native Banksia species in Western Australia and found the areas of greatest percent loss in richness in the arid interior, while the projected loss was less severe in the coastal areas and Fitz-Sterling Ranges [11]. These results are consistent with the results presented in Figure 2, although Fitzpatrick et al. did project species richness increases along the western coast of Australia, while our analysis found almost no projected expansion of the MCE in Australia. Loarie et al. identified areas of future refugia for species with projected range reductions in the mountainous areas along the central coast and foothills of California, USA and Baja California, Mexico [14], which correspond well with our areas of projected MCE stability. Benito-Garzón et al. studied a series of tree species in the Iberian Peninsula and found that some of the mediterranean species had the least projected range reductions and could potentially expand north and into higher elevations [13], which is consistent with our findings in Spain and Portugal. In general, the results from these biome and species level analyses that focused on one region at a time support our findings as well as the urgent need for conservation action to increase the extrinsic adaptation potential of the native biota of the mediterranean biome.

This analysis shows that climate change puts areas with the some of the highest levels of plant diversity and endemism on the Earth at risk. As shown in Figure 2, the Mediterranean Basin, Morocco and Israel contain large areas of projected contraction with no adjacent areas of expansion. The mediterranean portions of Morocco contain almost 13 plant species per 1000 km², while Israel contains 200 [7]. The cape region in South Africa and southwest Australia are the other two regions with large projected losses, and they contain 95.5 and 71 plant species per 1000 km², respectively [7]. For comparison, there is 1 plant species/1000 km² in Europe, 6.5 in Brazil, and 40 in Columbia [7]. The current MCE in these four threatened countries contain 22,400 plants species, 12,925 of which are found no where else [7].

The adaptation potential of the species native to the mediterranean biome will be further limited beyond the direct impacts of climate change analyzed here. Native plant species in all mediterranean regions, except perhaps Chile, are well adapted to natural fire regimes, but a hotter and drier climate has been observed to promote [31] and is projected to promote significant alterations to the fire regime beyond those created by decades of human fire management [28], [32], [33]. While rising atmospheric CO₂ levels could provide benefits to mediterranean plant species [34], the effects are altered when multiple factors of change are considered, including fire, drought, temperature increase, nitrogen deposition, and invasive species [35]–[38]. The rich plant diversity of mediterranean systems is explained in part by the plant adaptations to survive in low nutrient soils, such as California's serpentine soils, Australia's kwongan, and South Africa's fynbos [3], [4], [9]. The patchy nature of soils will act as a barrier and will make species migration in response to a changing climate more difficult. The intrinsic adaptation potential of some mediterranean endemics, particularly in South Africa and Australia, is limited by the relatively short seed dispersal distances and lack of colonization ability of these plants [11], [29], [39], [40]. While these indirect and interacting impacts of climate change are not explicitly considered in this analysis, they are likely to further limit the adaptation potential of mediterranean species.

Despite the significant projected contractions in the MCE, this analysis does offer some reasons for hope and some guidance to direct future conservation action. Approximately 50% of the biome is projected to remain stable with confidence, even under the high emissions scenario. Establishment and management of protected areas in these areas in all five regions represent sound investments given our current understanding of future change, and will help to secure future refugia for endemic species from other threats such as land conversion. However, given the uncertainties associated with the indirect effects of climate change, a conservative conservation approach should also include gene banking and ex-situ conservation for the rich floras of highly threatened regions like Israel, Morocco, South Africa, and Australia.

Since this analysis was conducted using a consistent methodology across all five regions, we can use the results to determine the highest priority regions for action. As shown in Figure 3, South Africa and Australia have the large projected contractions in the MCE. The existing protected area network covers almost 7% of the current MCE in these two regions. In South Africa, the protected areas are concentrated in the higher elevations where the MCE is projected to remain stable, so 77% of the current protected areas are projected to retain mediterranean climate and only 3% are projected to have a different climate in the future with high confidence, even under the high emissions scenario. In contrast, the protected areas in southwest Australia are concentrated in the drier inland portions of the MCE, so only 10% of the current protected areas will remain stable while 17% will likely shift to a new climate. These results suggest that some of the first strategies to enhance the extrinsic adaptation potential and to reduce the threat of climate change to the rich biodiversity of the mediterranean biome might include establishing new protected areas within areas projected to remain stable, improving land management, and restoring native habitat in southwest Australia. In particular, restoration efforts could focus on creating corridors or stepping stones of native habitat to connect isolated remnant vegetation patches in the areas with projected MCE contraction to the native habitat in the areas projected to remain stable. While this strategy has been advocated before [41], [42], and is currently underway in projects like the Gondwana Link (www.gondwanalink.org), this analysis highlights specific areas where this strategy could be implemented to improve the extrinsic adaptation potential of native species in the face of climate change.

Text S1

Table S1

Table S2

Figure S1