The six-predictor GLM explained 63.4 per cent of the deviance in current patterns of species richness based on two climatic and four non-climatic parameters (). Compared with the model proposed by Kreft & Jetz (2007)
, this explained about 2 per cent less of the deviance, but yielded very similar estimates of species richness (rs
Table 1. Generalized linear model (GLM) results of a model combining six predictor variables. (Since spatial autocorrelation might affect traditional statistical tests, we additionally performed spatial linear models to scrutinize p-values obtained from the GLM (more ...)
There was a strong interaction effect between temperature and CSR for different classes of water balance (a and ), and a model with an interaction between mean annual temperature and water balance provided stronger relative support than a model including only the main effects (ΔAIC = 106.5). In humid regions with positive water balance, there was a clear positive relationship between CSR and temperature (slope = 16.17 ± 0.65; standard error, p < 2 × 10−16). For regions with negative water balance up to −500 mm yr−1, this relationship was significantly shallower yet positive (slope = 3.16 ± 0.65, p = 0.006), and it was negative for the more arid areas with less than −500 mm yr−1 (slope = −13.82 ± 3.06, p = 1.05 × 10−05). As the water balance was predicted to get more negative in many regions according to the future climate scenarios (b–d), this leads to a predicted decrease in CSR in these regions.
Figure 1. (a) Observed current effects of temperature on plant species richness in 1032 geographical units worldwide. Residuals from the species–area relationship (log–log) are plotted against log10 transformed mean annual temperature (in K) for (more ...)
The projected changes in future CSR relate to the magnitude of the projected temperature rise in a way that global average CSR declines stronger in scenarios with a higher expected temperature rise ( and ). The geographical distribution of future CSR per grid cell for the A1FI scenario differed significantly from the present (two-sample paired Wilcoxon signed-rank test, p < 2.2 × 10−16). This was not the case for the B1 scenario (p = 0.26, two-sample paired Wilcoxon signed-rank test, and ). For the A1FI scenario, the CSR per grid cell was significantly lower than in the B1 scenario (a; two-sample paired Wilcoxon signed-rank test, p < 2.2 × 10−16), and individual CSR values per grid cell showed a higher variation (). Global average CSR for the B1 scenario remained similar to the present when the mean of all GCMs was considered (+0.3%), but there were pronounced differences among them ranging from +3.0 per cent (PCM) to −2.9 per cent (HadCM3; ). For the A1FI scenario, there resulted a pronounced decrease in global average CSR with a mean decrease among all GCMs of −9.4 per cent, ranging from −0.7 (PCM) to −20.0 per cent (HadCM3; ).
Figure 2. Modelled current global patterns of the capacity for species richness (CSR; species number per 110 × 110 km2) and future changes. (a) Modelled current patterns of CSR, (b) change in CSR under +1.8°C/B1 scenario for 2100, and (c) change (more ...)
Table 2. Summary results of future changes in the regional capacity for species richness (CSR; species number per 110 × 110 km2). (Presented are 18 combinations of four major IPCC emission scenarios (A1FI, A2, B1, B2) and five general circulation models (more ...)
Figure 3. Modelled changes in the capacity for species richness (CSR; species number per 110 × 110 km2) between today and the year 2100 under the +1.8°C/B1 scenario (blue) and the +4.0°C/A1FI scenario (red). (a) Global average CSR change (more ...)
Unlike the rather moderate changes in the global average future CSR, the projected changes in regional CSR at an individual grid cell basis are much more pronounced. Absolute changes in regional CSR considered independently from the direction of change are higher, the larger the expected temperature rise. For the B1 scenario, the average CSR change per cell reaches 15.3 per cent, and for the A1FI scenario, there is an average change of 30.9 per cent per grid cell, reaching 42 per cent in the most extreme HadCM3 circulation model ().
The uneven distribution of species richness around the globe is one of the most striking patterns in ecology and biogeography (Hawkins et al. 2003
; Ricklefs 2004
). According to our analysis, the global distribution of CSR will become profoundly more uneven than at present, as evidenced by an increase in the coefficient of variation in the year 2100 compared with today, calculated as the ratio of the standard deviation of all regional CSR values to the global mean CSR ().
We calculated the CSR for the year 2100 based on all 18 available combinations of IPCC scenarios and GCMs. Global CSR declined significantly in 13 of the 18 different models by 2100, on average by 4.9 per cent. To indicate the sensitivity of our results towards differences emerging from different GCMs, we calculated the direction of change and summed up the number of models indicating either increasing or decreasing CSR (d). Among all 18 models, 74 per cent of the land surface showed 100 per cent congruence in the direction of change. Inconsistent results were found in the transition zone between increasing and decreasing CSR, in particular in parts of the Amazon basin and central to southern Africa. The results indicate that independent from the magnitude of the expected future climate change, the direction of the calculated response in terms of CSR changes is similar in most parts of the world. The absolute changes in CSR, however, largely depend on the magnitude of climate change.
When averaged across the 40 industrialized countries listed in the Kyoto Protocol Annex B that are responsible for the highest per capita
emissions worldwide, the mean CSR for the year 2100 significantly increased by an average of 52 (B1) and 77 (A1FI) species per grid cell (mean CSR today: 594 species; two-sample paired Wilcoxon signed-rank test, p
< 2.2 × 10−16
). By contrast, the mean CSR decreased significantly by 64 (B1) to 186 (A1FI) species per grid cell (two-sample paired Wilcoxon signed-rank test, p
< 2.2 × 10−16
) in countries not listed as industrialized (mean CSR today: 1099 species). This apparent difference is mostly owing to the projected increase in CSR, owing to warming at higher latitudes, whereas CSR in most non-industrialized developing countries is projected to decrease owing to declining water availability. This discrepancy is alarming as the countries richest in plant biodiversity also are projected to experience the largest net loss in CSR. Moreover, it is inequitable that the countries being least responsible for the carbon dioxide concentration in the atmosphere are likely to be confronted with highest biodiversity threat owing to greenhouse gas-induced climate change. This is particularly worrying as the potential to develop climate change mitigation and adaptation strategies is much lower in these countries when compared with industrialized ones (IPCC 2007
For both the A1FI and the B1 scenario, a pronounced global discrepancy surfaced between regions exposed to either increasing or decreasing CSR (b
). Calculated across 13 major biomes in their current-day spatial location (excluding mangroves as an azonal system; Olson et al. 2001
), we found that by 2100 CSR shows the highest increase in tundras, followed by boreal forests, temperate coniferous forests, montane grasslands and shrublands, broadleaf and mixed forests and temperate grasslands (c
). In these systems, CSR might increase as a result of a relaxation from harsh thermal constraints, such as the occurrence or severity of frost or the duration of the thermal vegetation period, which all strongly limit plant distributions and richness (Sakai & Weiser 1973
; Woodward 1987
). On the other hand, a decrease in CSR is observed in biomes such as deserts and xeric shrublands, tropical and subtropical dry broadleaf forests, flooded grasslands and savannahs, tropical and subtropical grasslands, tropical and subtropical moist broadleaf forests, as well as in tropical and subtropical coniferous forests. The decrease in CSR in these areas can be explained by a shift of water balance to more negative values and resulting in an excess of drought tolerance levels for many species (compare Baltzer et al. 2008
; Engelbrecht et al. 2007
). If the Amazon rainforest is considered independently from African and Asian rainforests, it shows the most severe decrease in CSR compared with all other regions, with losses of approximately 30 (B1) to 50 per cent (A1FI). This corresponds to a potential dieback of Amazon forests by 2100 suggested by some GCMs (Cox et al. 2004
). Minor CSR changes are projected for temperate grasslands, savannahs and shrublands and in Mediterranean forests, woodlands and scrub. The low effect of climate change on CSR in Mediterranean regions may be explained by not resolving the seasonal distribution of precipitation in the GLM. The ranking of biomes differs slightly when absolute and relative changes in CSR are compared.
Modelled CSR values provide insights into the potential of an area to host a certain number of species. Thus, future CSR projections represent a first baseline risk assessment of the global distribution of plant diversity in the face of climate change. Similar to environmental niche modelling, we employ the covariation of environmental variables and species richness in space to derive temporal predictions (i.e. ‘space-for-time’ substitution; La Sorte et al. 2009
). An obvious limitation of this approach is that it does not provide direct information about possible range expansions, contractions or extinctions. While the modelled projections account for particular aspects of future climate change, they do not address the complexity of species interactions, potential additional environmental constraints and changes in the non-climatic environmental variables that were not included in the model. Moreover, it is yet unclear how climate-richness relationships vary over time, and whether the same relationships will hold under future climate conditions. Another uncertainty of our approach comes from novel future climate conditions and climatic extremes (Williams et al. 2007
The considered timespan of roughly one century appears too short to trigger substantial speciation events for vascular plants. Short-term changes in local species composition and richness should therefore mostly come about owing to species colonizing from other areas and arise from local extinctions. There is evidence that most species tend to keep their ecological preferences when colonizing new habitats (Crisp et al. 2009
). For this reason, some regions may lack the appropriate number of suitable species to fill the provided habitat space. Future climate conditions equivalent to current conditions will in many cases be beyond reach owing to geographical distance or may be even non-existent (Williams et al. 2007
). The risks of climate change-induced range shifts are multiplied in transformed and fragmented landscapes that provide little accessible space, reduced migration routes and little flexibility for the persistence of disadvantaged native species (Walther et al. 2002
; Svenning & Skov 2004
). On the other hand, the spatial patterning of landscape features and environmental variables at different spatial scales can also have a stabilizing effect on species distributions. Many species may be able to persist in small pockets of suitable conditions, e.g. in valleys or gallery forests with still suitable meso- and microscale conditions, even when the overall broad-scale climate conditions are getting harsh and unsuitable.
The rate at which climate is projected to change and at which species displacement is induced in many cases may exceed the velocity at which new arriving species and functional communities are able to establish (Hector et al. 1999
). Changes in species composition require time for dispersal and recruitment success of invasive species as well as displacement of formerly native species confronted with unfavourable conditions. Disturbances and catastrophic events (Pounds et al. 2005
) as well as complex biotic interactions (Pearson & Dawson 2003
) can further influence the velocity of this process. Ecosystem changes are not likely to appear gradually but are connected to thresholds and tipping points (Scholze et al. 2006
). In terms of species richness, this can result in timespans with relative stability followed by a cascade of local extinction events. However, as the current occurrence of species represents their realized niches that can be considerably smaller than their fundamental ones (Araújo & Pearson 2005
), some species ranges could be considerably more resilient to changing climate conditions than expected. Hence, the eventual achievement of equilibrium between local CSR and realized species richness is subject to interacting factors related to the resilience capacity of individual species and communities (Leemans & Eickhout 2004
While the negative impacts of a climate-change-induced reduction in regional CSR on global biodiversity and ecosystem functions are apparent, perils of increasing CSR are less obvious at first glance. From a human perspective, an increase in CSR may even be associated with some positive effects such as higher agricultural productivity, higher carbon storage and a wider range of options to manage ecosystem services in some parts of the world (Leemans & Eickhout 2004
). On the other hand, a fast increase in CSR beyond the potential for adaptation by established ecosystems may signal high prevalence of species invasions and an extensive replacement of native floras by widespread and competitive species immigrating from elsewhere (Scholze et al. 2006
; Woodward & Kelly 2008
). Paradoxically, an increase in CSR can thereby even cause an intermediate decrease in the absolute species numbers within many regions. Especially, species adapted to harsh environmental conditions may be particularly vulnerable if the climate becomes more favourable for generalists. In this respect, endemic species may get more threatened, as many of them evolved under long-term stable climatic conditions (Jansson 2003
; Linder 2008
). As a consequence, future climate change may trigger the reallocation of the global pool of existing species. Competitive generalist species will get more abundant and widespread at the expense of specialists that will get more rare and range-restricted or even go extinct, resulting in biotic homogenization (White & Kerr 2007
; La Sorte et al. 2009
). Altogether, this may alter ecological interactions. Although newly arriving species may fill in some of the ecological functions of disappearing species, there is a high risk that ecosystem functions and services may be impaired (Schröter et al. 2005
Our results indicate that the consequences of climate change for plant distributions differ dramatically between the two examined IPCC scenarios, B1 and A1FI. Hence, a precautionary principle dictates that an immediate implementation and continuous further improvement of mitigation strategies are necessary to minimize negative impacts on biodiversity, environmental functionality and sustainable human development. In addition, our results reinforce the necessity of regionalized adaptation strategies in regions with either expected increase or decrease in CSR to minimize the negative impacts of climate change.