Turkish Journal of Zoology
Turk J Zool (2017) 41: 513-521 © TÜBİTAK doi:10.3906/zoo-1512-42
http://journals.tubitak.gov.tr/zoology/
Research Article
The potential effects of future climate change on suitable habitat for the Taiwan partridge (Arborophila crudigularis): an ensemble-based forecasting method 1,
2
3
Juncheng LEI *, Jun WU , Qingwei GUAN School of Geography and Planning, Gannan Normal University, Ganzhou, P.R. China 2 Nanjing Institute of Environmental Science, Ministry of Environmental Protection, Nanjing, P.R. China 3 College of Biology and the Environment, Nanjing Forestry University, Nanjing, P.R. China 1
Received: 15.12.2015
Accepted/Published Online: 10.10.2016
Final Version: 23.05.2017
Abstract: Climate change is considered to be one of the greatest threats to biodiversity in this century, especially for range-restricted island species. This study explored the potential effects of climate change on Arborophila crudigularis, a weak-flying endemic bird species in Taiwan. The potential effects of climate change on climatically suitable habitat for A. crudigularis were analyzed in biomod2 and ArcGIS software. Future climate change could increase the availability of climatically suitable habitat for A. crudigularis while decreasing the mean suitability for both the entire suitable area and the area with known presence records. By 2050 and 2080, climatically suitable habitat is expected to increase by an average of 4.57% and 5.18%, respectively; the mean suitability of the entire climatically suitable habitat is expected to decrease by 4.80% and 6.61%; and the mean suitability of known presence records is expected to decrease by 2.70% and 4.62%, respectively. Future climate change will not be disastrous for A. crudigularis in Taiwan. Future efforts to conserve this species should focus on northwestern Taiwan. Key words: Taiwan, species distribution model, biomod2, climate scenario, general circulation model, conservation
1. Introduction The Fifth Assessment Report of the Intergovernmental Panel on Climate Change (IPPC AR5) indicated that the annual global mean surface air temperature has increased by 0.85 °C during the past 100 years and is expected to increase by 0.3 °C to 4.8 °C by the end of this century (IPCC, 2014). Climate change, especially global warming, has caused and/or is changing many ecological phenomena and processes, e.g., species distributions and phenology (Thomas et al., 2004; Lenoir et al., 2008). Climate change could be one of the greatest threats to biodiversity in this century (Secretariat of the Convention on Biological Diversity, 2014). Learning about the potential effects of climate change on species and taking proactive measures are considered some of the most effective means for reducing the impacts of climate change on biodiversity (Sinclair et al., 2010; Hazen et al., 2013). Species distribution models (SDMs) are effective tools for exploring the potential distribution or suitable habitat of a given species under various climate conditions (Franklin, 2010; Conlisk et al., 2013). In predicting the potential suitable habitat for a given species under future climate scenarios, SDMs generally proceed by first identifying the environmental characteristics of each location where * Correspondence:
[email protected]
a species is known to be present under baseline climate conditions; this information is subsequently used to detect other areas that would possess similar characteristics under future climate scenarios (Diniz-Filho et al., 2009). SDMs have increasingly been applied to understand the conservation of endangered species and the management of alien species under future climate scenarios (Guisan and Thuiller, 2005; Hannah et al., 2013). The Taiwan partridge (Arborophila crudigularis) is a range-restricted bird that is endemic to the islands of Taiwan. Although the latest IUCN Red List (V 3.1) classified it in the Least Concern category (IUCN, 2015), learning about the potential effects of climate change on this species is still valuable for its conservation in the context of climate change, because A. crudigularis is a weak flier and cannot adapt to climate change quickly (Lu et al., 2012); the limited extent of the island does not provide enough opportunities for it to adjust to climate change by moving along latitudinal gradients (Chen et al., 2011); and its population trend is decreasing, and climate change may accelerate this process (Ko et al., 2012; IUCN, 2015). The primary objectives of this study are to determine the climate variables that exhibit the largest effect on the distribution range of A. crudigularis on a broad scale, to
513
LEI et al. / Turk J Zool model the potential climatically suitable habitat of the species under both current climate (baseline climate conditions) and future climate scenarios, and to explore the potential effects of climate change on climatically suitable habitat for this species. 2. Materials and methods 2.1. Species’ bioecology and presence records A. crudigularis is confined to the foothills and mountains of Taiwan, where it occurs mainly in broadleaved forest at 100–2300 m a.s.l. It favors thickets and damp undergrowth in evergreen broadleaved forest and is active on the ground in the daytime, while inhabiting trees at night. It feeds on earthworms, seeds, berries, seedlings, leaves, and insects (IUCN, 2015). In this study, presence records for A. crudigularis were obtained from the Global Biodiversity Information Facility (GBIF, http://www.gbif.org/) and the Taiwan Biodiversity Information Facility (TaiBIF, http://taibif.tw/). A total of 611 nonduplicate presence records were collected. For consistency with the spatial resolution of the predictor variables, the presence records were resampled to 2.5 arc minutes, leaving 1 presence record per 6.25 square arc minutes. A total of 280 presence records were selected for modeling (Figure 1).
2.2. Model range According to Pearson and Dawson (2003), the dominant environmental factor determining the distribution range of a species is scale-dependent; at regional (>200 km) or coarser scales, climate is the main determinant of species’ distributions. Taking into account that the Taiwanese islands are not large enough, we included not only the islands of Taiwan but also nearby areas of mainland China as the model range, including southern China, southwestern China, and central China (Figure 1; for details regarding these regions, please refer to Zhang, 1999). Given that A. crudigularis is a weak flier, we restricted the analysis of the potential effects of climate change on suitable habitats of the species to the Taiwan islands. 2.3. Predictor variables We selected 19 bioclimatic variables as preliminary predictor variables for both baseline climate and future climate scenarios (for details about the 19 bioclimatic variables, please refer to Hijmans et al., 2005). The baseline climate conditions data were acquired from Worldclim (http://www.worldclim.org/). The spatial resolution of the bioclimatic variables used in this study was 2.5 arc minutes. All baseline climate condition data layers were extracted at the spatial scale of the model. To avoid including highly correlated variables, we tested all variables for pairwise
Figure 1. Modeled range and presence records for Arborophila crudigularis.
514
LEI et al. / Turk J Zool correlations using Spearman’s correlation coefficient. If the absolute value of the correlation coefficient was ≥0.7, the variables were considered redundant (Olson et al., 2014). Five bioclimatic variables were selected: annual mean temperature (AMT), mean diurnal range (MDR), temperature seasonality (TS), annual precipitation (AP), and precipitation of warmest quarter (PWQ). The climate scenario data used in this study were acquired from the CGIAR Research Program on Climate Change, Agriculture and Food Security (CCAFS, http:// www.ccafs-climate.org/). Climate projections for 2050 and 2080 were derived from two general circulation models (GCMs; cccma_canesm2 and csiro_mk3_6_0, which are referred to as cccma and csiro, respectively) and two Special Reports on Emission Scenarios (SRES; RCP2.6 and RCP8.0). The resolution of the bioclimatic variables for the future climate scenarios was also 2.5 arc minutes. 2.4. Species distribution models The climatically suitable habitat of A. crudigularis was modeled by running 9 different SDMs using the biomod2 software package on the R platform (https://cran.r-project. org/web/packages/biomod2/index.html). Pseudo-absence records were chosen twice from outside the suitable range predicted by the surface range envelope (SRE) model; each time, 2000 pseudo-absences were randomly selected (Barbet-Massin et al., 2013). The total weight of the pseudo-absence data was set equal to the total weight of the presence data. To evaluate the performance of a species distribution model, we randomly split the species records into 2 parts, in which 75% of the records were used to calibrate the models, while the remaining 25% were used for evaluation. Model performance was assessed with the area under the receiver operating characteristic curve (AUC), Cohen’s kappa statistic (kappa), and the true skill statistic (TSS) (Cohen, 1960; Fielding and Bell, 1997; Allouche et al., 2006). We repeated the data split 3 times and calculated the average AUC, kappa, and TSS of the cross-validations. Finally, we used the AUC score to weight the corresponding model’s continuous output for baseline climate and future climate scenario; only those models with an AUC score exceeding 0.8 were included in the consensus estimate (D’Amen et al., 2011). 2.5. Analysis of the effects of climate change The continuous layers were converted to binary using the minimum presence threshold to analyze the climate change effects on the habitat of A. crudigularis (Giovanelli et al., 2010; Rödder and Engler, 2011). We calculated the suitable habitat area under both baseline climate conditions and future climate scenarios. To illustrate the spatial change in suitable habitat, we overlaid the suitable habitat for baseline climate conditions with the suitable habitat determined for each future climate scenario based on the map algebra principle in ArcGIS (V9.3). To
demonstrate the potential effects of climate change on this species more clearly, we compared the mean suitability of the entire climatically suitable habitat and that of presence records under baseline climate conditions and future climate scenarios. 3. Results 3.1. Model performance The AUC, kappa, and TSS scores for all model runs exceeded 0.8 (Figure 2). The AUC, kappa, and TSS scores for the ensemble model were 1.000, 0.992, and 0.999, respectively. This result indicates that all of the models performed well and could be used to predict suitable habitat for A. crudigularis under future climate scenarios. 3.2. Importance of environmental variables Six of the 9 models (i.e. GLM, GBM, GAM, CTA, RF, and MAXENT) indicate that temperature seasonality is the most important variable to define the distribution range of A. crudigularis at a broad scale; 3 models (ANN, FDA, and MARS) indicate that annual precipitation is the most important variable. In the ANN, FDA, and MARS models, temperature seasonality is the second most important variable. The mean importance of each predictor variable for the distribution of A. crudigularis indicates that temperature seasonality is the most important variable, while annual precipitation is second in importance (Table). 3.3. Suitable habitat under baseline climate conditions Under baseline climate conditions, the climatically suitable habitat extent for A. crudigularis slightly exceeded the extent of the known distribution. All of the main island of Taiwan, with the exception of the western area, was predicted to be a climatically suitable habitat for A. crudigularis. The total suitable habitat was 29,904 km2. 3.4. Potential effects of climate change 3.4.1. Change in extent of suitable habitat Under the RCP2.6 emission scenario, the cccma model predicted an increase in climatically suitable habitat from 2000 to 2050, followed by a decrease from 2050 to 2080; the csiro model predicted a gradual increase over both time periods. Although the two models differed in their response, climatically suitable habitat under the 4 future climate scenarios was predicted to be larger than under baseline climate conditions. By 2080, climatically suitable habitat was predicted to increase by 2.6% (cccma) and 7.4% (csiro) (Figure 3). The cccma and csiro predictions under the RCP8.5 emission scenario were the same as under the RCP2.6 emission scenario. The climatically suitable habitat extent under the 4 future climate scenarios was predicted to be larger than under baseline climate conditions. The climatically suitable habitat was predicted to increase by 0.5% (cccma) and 10.2% (csiro) by 2080 (Figure 3).
515
LEI et al. / Turk J Zool
Figure 2. Performance of each model for predicting the suitable habitat for Arborophila crudigularis. GLM: Generalized linear model; GBM: generalized boosting model; GAM: generalized additive model; CTA: classification tree analysis; ANN: artificial neural network; FDA: flexible discriminant analysis; MARS: multiple adaptive regression splines; RF: random forest; MAXENT: maximum entropy model.
516
LEI et al. / Turk J Zool Table. Importance of each predictor variable for the distribution of Arborophila crudigularis. AMT
MDR
TS
AP
PWQ
GLM
0.06
0.50
0.72
0.03
0.29
GBM
0.00
0.25
0.59
0.46
0.00
GAM
0.07
0.32
0.65
0.11
0.41
CTA
0.01
0.31
0.65
0.60
0.02
ANN
0.03
0.06
0.45
0.68
0.05
FDA
0.01
0.04
0.41
0.86
0.03
MARS
0.01
0.21
0.30
0.65
0.08
RF
0.00
0.10
0.28
0.18
0.01
MAXENT
0.00
0.21
0.58
0.13
0.10
Average
0.02
0.22
0.52
0.41
0.11
GLM: Generalized linear model; GBM: generalized boosting model; GAM: generalized additive model; CTA: classification tree analysis; ANN: artificial neural network; FDA: flexible discriminant analysis; MARS: multiple adaptive regression splines; RF: random forest; MAXENT: maximum entropy model. AMT: annual mean temperature; MDR: mean diurnal range; TS: temperature seasonality; AP: annual precipitation; PWQ: precipitation of warmest quarter.
335 330 325 320 315 310 305 300 295 290 285 280
RCP2.6
be lower than under baseline climate conditions. By 2080, the mean suitability was predicted to decrease by 3.1% (cccma) and 3.8% (csiro) (Figure 4). Under the RCP8.5 emission scenario, the cccma and csiro models both resulted in gradual decreases in the mean suitability. By 2080, the mean suitability was predicted to decrease by 9.4% (cccma) and 10.2% (csiro) (Figure 4).
950
RCP8.5
EN RCP2.6
EN RCP8.5
PR RCP2.6
PR RCP8.5
900 Suitability
Suitable habitat area (100 km 2)
3.4.2. Change in the mean suitability for entire suitable habitat Under the RCP2.6 emission scenario, the cccma model predicted that the mean suitability would decrease from 2000 to 2050 and increase from 2050 to 2080; the csiro analysis indicated that the mean suitability would decrease gradually. Although the models differed in their projections of future climate, both models predicted that the mean suitability under future climate scenarios would
850 800 750 700
2000 Baseline
2050
2080 cccma Climate condition
2050
2080 csiro
Figure 3. Suitable habitat for Arborophila crudigularis under baseline climate conditions and future climate scenarios. cccma and csiro represent two general circulation models; RCP2.6 and RCP8.5 represent two greenhouse gas emission scenarios.
2000 Baseline
2050
2080 cccma Climate condition
2050
2080 csiro
Figure 4. Mean suitability for Arborophila crudigularis under baseline climate conditions and future climate scenarios. cccma and csiro represent two general circulation models; RCP2.6, and RCP8.5 represent two greenhouse gas emission scenarios; EN is entire suitable habitat; PR is presence records.
517
LEI et al. / Turk J Zool 3.4.3. Change in mean suitability for presence records Under each climate change scenario, the mean suitability analysis based on presence records produced the same results as when all habitat information was included; however, the predicted decrease in amplitude was smaller. Under the RCP2.6 emission scenario, the mean suitability was predicted to decrease by 1.9% (cccma) and 1.8% (csiro) by 2080; under the RCP8.5 emission scenario, the mean suitability was predicted to decrease by 8.7% (cccma) and 6.1% (csiro) (Figure 4). 3.4.4. Spatial distribution of suitable habitat Under the RCP2.6 emission scenario, the 2 climate scenarios simulated with the cccma model projected gains and losses in suitable habitat in western Taiwan; an increase was predicted in the southwest. For the 2 climate scenarios simulated with the csiro model, habitat increased in western Taiwan (Figure 5). Under the RCP8.5 emission scenario, the 2 climate scenarios simulated with the cccma model projected habitat gains in southwestern Taiwan and habitat losses in northwestern Taiwan. For the 2 climate scenarios simulated with the csiro model, habitat increased in Taiwan (Figure 6). 4. Discussion Based on the climate scenarios applied in our study, the climate of Taiwan will become warmer and rainier; on average, by 2050 and 2080, the annual mean temperature will increase by 1.9 °C for the RCP2.6 emission scenario and 2.6 °C for the RCP8.5 emission scenario; moreover, the annual precipitation will increase by 178 mm for the RCP2.6 emission scenario and 202 mm for the RCP8.5 emission scenario. In the case of temperature seasonality, it will decrease in the future. In our study, the climatically suitable habitat area under different climate scenarios highly correlated with the temperature seasonality (Spearman’s r = –0.81, R2 = 0.008), which can confirm the finding of our study about the most important variable in defining the distribution range of A. crudigularis at a broad scale. In addition, Spearman’s coefficient was less than 0, which indicates that A. crudigularis prefers an environment with little fluctuation in temperature seasonality. Even though the changing climate could lead to an increase in suitable habitat for A. crudigularis, some currently occupied areas in northwestern Taiwan may lose suitable habitat under some climate scenarios, particularly the two scenarios simulated with the cccma model under the RCP8.5 emission scenario (Figure 5 and 6). Therefore, we suggest implementing long-term monitoring of the effects of climate change on this species in the aforementioned area. A. crudigularis is considered to be well represented in national parks, nature reserves, and wildlife sanctuaries in
518
Taiwan (IUCN, 2015). According to our findings, future climate change will not have major consequences for this species. Consequently, attention should be paid to other potential threats to the species, such as habitat losses due to deforestation and conversion to agricultural land (IUCN, 2015). In addition to direct effects on the species, climate change could have indirect effects, e.g., promoting the invasion of alien species and changing vegetation types (Cramer et al., 2001; Hellmann et al., 2008). In future analyses of the impacts of climate change on this species, we should focus not only on the direct effects but also on the indirect effects. The changes in suitable habitat predicted by our study differ from the findings of Ko et al. (2012), who predicted that suitable habitat for A. crudigularis would decrease under future climates. Several factors explain this difference. MAXENT was the only model used in Ko et al.’s study, while an ensemble forecasting method was used in our study. MAXENT can perform well under many conditions; however, recent studies have demonstrated that discrepancies among different models can be very large, particularly when the models are used to project species’ distributions under future climate scenarios (Thuiller, 2004; Pearson et al., 2006). Ensemble forecasting techniques, such as biomod2 and ModEco, are considered to be good solutions to this problem because they provide a more robust prediction (Thuiller et al., 2009; Guo and Liu, 2010). The annual mean temperature was the only climate variable in Ko et al.’s study, while our study added 4 additional climate variables. In our study, temperature seasonality and annual precipitation were the dominant climate variables for defining the range of A. crudigularis on a broad scale; the annual mean temperature was the least important. The climate scenario data for Ko et al.’s study were derived from IPCC AR4 A2 and B2 emission scenarios, while the climate scenario data applied in our study were derived from IPCC AR5 RCP2.6 and RCP8.5 emission scenarios. Finally, the modeled range for Ko et al.’s study was Taiwan; we selected not only Taiwan but also nearby parts of China because our preliminary results indicated that the climate variables performed poorly if the scale was limited to the Taiwanese islands. Traditionally, analyses of the potential effects of climate change on species have focused on threatened species; other species have rarely been studied (Dawson et al., 2011). Although A. crudigularis is classified in the Least Concern category in the latest IUCN Red List, knowing the potential effects of climate change could be invaluable for future conservation efforts if the population decreases in the future. Additionally, although A. crudigularis is an endemic species, it could be considered a common species in Taiwan, given its wide distribution on the main island.
LEI et al. / Turk J Zool
Figure 5. Changes in suitable habitat for Arborophila crudigularis under the RCP2.6 emission scenario. cccma and csiro represent two general circulation models.
The potential effects of climate change on this species may be representative of other widely distributed communities on the main island of Taiwan. Generally, future climate changes will not be a disaster for A. crudigularis. Although the mean suitability was
predicted to decrease slightly for both the entire suitable habitat and the area with known presence records, the climatically suitable habitat area was predicted to increase slightly for each climate scenario relative to baseline climate conditions.
519
LEI et al. / Turk J Zool
Figure 6. Changes in suitable habitat for Arborophila crudigularis under the RCP8.5 emission scenario. cccma and csiro represent two general circulation models.
Acknowledgments The authors are grateful to Tiago S Vasconcelos for suggestions and to Junwei Wang for graph plotting. This
520
work received financial support from the Grant Program of Clean Development Mechanism in China (1213114).
LEI et al. / Turk J Zool References Allouche O, Tsoar A, Kadmon R (2006). Assessing the accuracy of species distribution models: prevalence, kappa and the true skill statistic (TSS). J Appl Ecol 43: 1223-1232.
Hellmann JJ, Byers JE, Bierwagen BG, Dukes JS (2008). Five potential consequences of climate change for invasive species. Conserv Biol 22: 534-543.
Barbet-Massin M, Rome Q, Muller F, Perrard A, Villemant C, Jiguet F (2013). Climate change increases the risk of invasion by the yellow-legged hornet. Biol Conserv 157: 4-10.
Hijmans RJ, Cameron SE, Parra JL, Jones PG, Jarvis A (2005). Very high resolution interpolated climate surfaces for global land areas. Int J Climatol 25: 1965-1978.
Chen IC, Hill JK, Ohlemüller R, Roy DB, Thomas CD (2011). Rapid range shifts of species associated with high levels of climate warming. Science 333: 1024-1026.
IPCC (2014). Climate Change 2014: Synthesis Report. Geneva, Switzerland: IPCC.
Cohen J (1960). A coefficient of agreement for nominal scales. Educ Psychol Meas 20: 37-46. Conlisk E, Syphard AD, Franklin J, Flint L, Flint A, Regan H (2013). Uncertainty in assessing the impacts of global change with coupled dynamic species distribution and population models. Glob Change Boil 19: 858-869. Cramer W, Bondeau A, Woodward FI, Prentice IC, Betts RA, Brovkin V, Cox PM, Fisher V, Foley JA, Friend AD (2001). Global response of terrestrial ecosystem structure and function to CO2 and climate change: results from six dynamic global vegetation models. Glob Change Boil 7: 357-373.
IUCN (2015). The IUCN Red List of Threatened Species. Geneva, Switzerland: IPCC. Ko CY, Root TL, Lin SH, Schneider SH, Lee PF (2012). Global change projections for Taiwan island birds: linking current and future distributions. Nat Conserv 2: 21-40. Lenoir J, Gégout J, Marquet P, De Ruffray P, Brisse H (2008). A significant upward shift in plant species optimum elevation during the 20th century. Science 320: 1768-1771. Lu N, Jing Y, Lloyd H, Sun YH (2012). Assessing the distributions and potential risks from climate change for the Sichuan Jay (Perisoreus internigrans). Condor 114: 365-376.
D’Amen M, Bombi P, Pearman PB, Schmatz DR, Zimmermann NE, Bologna MA (2011). Will climate change reduce the efficacy of protected areas for amphibian conservation in Italy? Biol Conserv 144: 989-997.
Olson LE, Sauder JD, Albrecht NM, Vinkey RS, Cushman SA, Schwartz MK (2014). Modeling the effects of dispersal and patch size on predicted fisher (Pekania [Martes] pennanti) distribution in the US Rocky Mountains. Biol Conserv 169: 89-98.
Dawson TP, Jackson ST, House JI, Prentice IC, Mace GM (2011). Beyond predictions: biodiversity conservation in a changing climate. Science 332: 53-58.
Pearson RG, Dawson TP (2003). Predicting the impacts of climate change on the distribution of species: are bioclimate envelope models useful? Global Ecol Biogeogr 12: 361-371.
Diniz-Filho JAF, Mauricio Bini L, Fernando Rangel T, Loyola RD, Hof C, Nogués-Bravo D, Araújo MB (2009). Partitioning and mapping uncertainties in ensembles of forecasts of species turnover under climate change. Ecography 32: 897-906.
Pearson RG, Thuiller W, Araújo MB, Martinez-Meyer E, Brotons L, McClean C, Miles L, Segurado P, Dawson TP, Lees DC (2006). Model-based uncertainty in species range prediction. J Biogeogr 33: 1704-1711.
Fielding AH, Bell JF (1997). A review of methods for the assessment of prediction errors in conservation presence/absence models. Environ Conserv 24: 38-49.
Rödder D, Engler J (2011). Quantitative metrics of overlaps in Grinnellian niches: advances and possible drawbacks. Global Ecol Biogeogr 20: 915-927.
Franklin J (2010). Moving beyond static species distribution models in support of conservation biogeography. Divers Distrib 16: 321-330.
Secretariat of the Convention on Biological Diversity (2014). Global Biodiversity Outlook 4. Montreal, Canada: Secretariat of the Convention on Biological Diversity.
Giovanelli G, de Siqueira MF, Haddad CF, Alexandrino J (2010). Modeling a spatially restricted distribution in the Neotropics: how the size of calibration area affects the performance of five presence-only methods. Ecol Model 221: 215-224. Guisan A, Thuiller W (2005). Predicting species distribution: offering more than simple habitat models. Ecol Lett 8: 993-1009. Guo Q, Liu Y (2010). ModEco: an integrated software package for ecological niche modeling. Ecography 33: 637-642. Hannah L, Roehrdanz PR, Ikegami M, Shepard AV, Shaw MR, Tabor G, Zhi L, Marquet PA, Hijmans RJ (2013). Climate change, wine, and conservation. P Natl Acad Sci USA 110: 6907-6912. Hazen EL, Jorgensen S, Rykaczewski RR, Bograd SJ, Foley DG, Jonsen ID, Shaffer SA, Dunne JP, Costa DP, Crowder LB et al. (2013). Predicted habitat shifts of Pacific top predators in a changing climate. Nat Clim Change 3: 234-238.
Sinclair SJ, White MD, Newell GR (2010). How useful are species distribution models for managing biodiversity under future climates? Ecol Soc 15: 8. Thomas CD, Cameron A, Green RE, Bakkenes M, Beaumont LJ, Collingham YC, Erasmus BF, De Siqueira MF, Grainger A, Hannah L et al. (2004). Extinction risk from climate change. Nature 427: 145-148. Thuiller W (2004). Patterns and uncertainties of species’ range shifts under climate change. Glob Change Biol 10: 2020-2027. Thuiller W, Lafourcade B, Engler R, Araújo MB (2009). BIOMOD–a platform for ensemble forecasting of species distributions. Ecography 32: 369-373. Zhang RZ (1999). Zoogeography of China. Beijing, China: Science Press (in Chinese).
521