Advertisement

Ecological Research

, Volume 33, Issue 1, pp 199–204 | Cite as

The effects of climate change on the distribution of European glass lizard Pseudopus apodus (PALLAS, 1775) in Eurasia

  • Reza Nasrabadi
  • Nasrullah Rastegar-Pouyani
  • Eskandar Rastegar Pouyani
  • Haji Gholi Kami
  • Ahmad Gharzi
  • Saeed Hosseinian Yousefkhani
Original Article
  • 182 Downloads

Abstract

The distribution area of Pseudopus apodus includes the Balkan, Crimean peninsulas, and Ciscaucasia region in Europe, and Asia Minor and the Middle East. This area has experienced a significant habitat loss and fragmentation because of human population growth, increased farming, logging and climate change. To estimate how climate change will affect the presumed future distribution of the studied species, we constructed the possible current distribution of the species and its potential environmental risk for future dispersion. We used an ensemble prediction to forecast the location and distribution of suitable habitats for P. apodus in present and future (i.e. 2070) based on 19 environmental variables. The results were consistent among models and indicated that there are two most important variables that affect distribution pattern of the species: temperature seasonality and precipitation seasonality. All of the models used in this study showed a significant AUC and TSS value. Based upon FDA and ensemble maps it is proposed here that species range will be extended to the east, in particular in higher altitude regions like Afghanistan, but its western range in Jordan will be shrunk. Comparison of the current distribution and future prediction reveals that suitable habitats of Pseudopus apodus will be shifted to higher elevations by 2070 and during this period the species is predicted to migrate from lowlands to higher elevations. Change in latitudinal range is also probable to find new suitable areas under predicted future climate scenarios.

Keywords

Biomod2 Reptiles Pseudopus apodus Distribution Biodiversity 

Introduction

Ecological niche models refer to an approach for predicting a species’ response to the global warming. These models use distribution data and statistical tools to forecast the range of a species (Araùjo et al. 2006).

The use of Species Distribution Models (SDMs), for mapping and monitoring animal range, has become increasingly important in the context of understanding climate changes and its ecological consequences (Miller 2010). SDMs provide detailed predictions of species’ range by linking the presence or abundance of species with environmental variables. As such, distribution models have provided researchers a new tool to explore diverse questions about species distributions (Nally and Fleishman 2004).

Global warming and increasing human activity such as human population growth, and increasing farming and logging have impacted temperate forests in Asia and Europe, resulting in forest degeneration and fragmentation. This has resulted in local extinctions or forced plants and animals to move to higher latitudes or elevations (Root et al. 2003; Walther et al. 2005).

In predictive modeling, applying single model based on similar data sample can have biases, high variability or outright inaccuracies that affect the reliability of its findings. Prediction of species distribution by different approaches give different results and many pitfalls are also presented, but the ensemble approach uses multiple algorithms to try to avoid any bias in models that could occur with a single modeling approach. (Marmion et al. 2009). In ensemble modeling by combining different models data, researchers can reduce the effects of those limitations and provide better information to predictive model. One of the approaches to ensemble forecasting of species distributions is BIOMOD2, which uses different types of statistical modeling methods aiming to maximize the predictive accuracy of current species distributions and future potential range. The most widely used modeling techniques in species predictions are bioclimatic envelope method (Surface Range Envelop, SRE), three regression methods (Generalized Linear Models, GLM; Generalized Additive Models, GAM; and Multivariate Adaptive Regression Splines, MARS), two classification methods (Classification Tree Analysis, CTA and Flexible Discriminant Analysis, FDA) and four machine learning methods (Generalized Boosting Model, GBM; Artificial Neural Networks, ANN; Random Forests, RF; and Maximum Entropy, MAXENT) (Thuiller 2016).

The European Glass Lizard (Pseudopus apodus) inhabits dry and well-vegetated rocky slopes traditionally cultivated areas and regions close to human settlements (Başoğlu and Baran 1977). The species lives in mesic places such as gardens, margin of rivers and under the leaves of trees and in grassland and shrubby vegetation near streams. Although this species is categorized as Least Concern on the IUCN Red List, it is listed as the Bern Convention due to habitat loss and degradation (Cox and Temple 2009).

We here present all data available on the distribution of Pseudopus apodus to describe its distribution range and use BIOMOD2 algorithm to predict its suitable habitats in the world. Dispersal range of this species has never been mapped and analyzed; therefore the achieved results from this model will be valuable for understanding the biogeography of European glass lizard and determination of their conservation status.

Materials and methods

Occurrence records and environmental data

Georeferenced distribution records for Pseudopus apodus were collected from the literature, publications (e.g. Anderson 1999; Šmíd et al. 2014) museum records and the GBIF website (www.gbif.org). The species occurrence data were also collected from our fieldwork on the Iranian Plateau during spring and summer 2013–2016. Collected specimens were deposited in the Sabzevar University Herpetological Collection (SUHC). In order to check the accuracy of the examined localities, downloaded samples from GBIF were matched with samples from various museums including Sabzevar University Herpetological Collection (SUHC), California academy of sciences Museum (CAS), Natural History Museum of Crete (NHMC), Natural History Museum of Denmark (NHMD), Zoological Research Museum Alexander Koenig (ZFMK). In total, 161 records were obtained from these resources as presence records and 192 random points were obtained as pseudo-absence records (www.geomidpoint.com/random/) (Table 1).
Table 1

Algorithms examined in this paper

Method

Class of model

Data

SRE

Surface Range Envelop is a correlative modeling tool that insert climatic parameters for any location

Only presence

MARS

Multivariate adaptive regression splines,provide an alternative regression-based method for fitting non-linear responses, using piecewise linear

Presence/absence

RF

A machine learning method which is a generally involve developing multiple models on different subsets of the data, Creates multiple boot-strapped regression trees and each tree is grown with a randomized subset of predictors, the results of which are averaged

Presence/absence

CTA

A machine-learning methods for constructing prediction models from data. The models are obtained by recursively partitioning the data space and fitting a simple prediction model within each partition, a classification method running a 50-fold cross-validation to select the best tradeoff between the number of leaves of the tree and the explained deviance

Presence/absence

FDA

FDA is a non-linear classification technique based on Fisher’s discriminant. The linear classification in feature space corresponds to a non-linear decision function in input space. The proposed regression techniques implement the idea of using nonlinear mappings to transform the input data into a new space in which again a linear regression is performed

Presence/absence

In total, 19 bioclimatic layers were downloaded for the current period (1950-2010) and future layers under the A1B scenario (a balanced emphasis on all energy sources in the next century, Based on this scenario the human population from 2000 to 2100 will increase as the amount of carbon dioxide from fossil fuels will increase (Nakicenovic et al. 2000). Increasing carbon dioxide causes a rise in Earth’s temperature) were obtained for 2070 from www.ccafs climate.org (Table 2). We used of the A1B emission scenario to characterize future climatic conditions in 2070. All layers were downloaded as 30 arc-second resolution and cropped for the study area (Fig. 1) using ArcGIS 10.3 (ESRI).
Table 2

Climatic variables used to develop the distribution model

BIO1

Annual mean temperature

BIO2

Mean diurnal range [mean of monthly (max temp–min temp)]

BIO3

Isothermality (BIO2/BIO7) × (100)

BIO4

Temperature seasonality (standard deviation × 100)

BIO5

Max temperature of warmest month

BIO6

Min temperature of coldest month

BIO7

Temperature annual range (BIO5–BIO6)

BIO8

Mean temperature of wettest quarter

BIO9

Mean temperature of driest quarter

BIO10

Mean temperature of warmest quarter

BIO11

Mean temperature of coldest quarter

BIO12

Annual precipitation

BIO13

Precipitation of wettest month

BIO14

Precipitation of driest month

BIO15

Precipitation seasonality (coefficient of variation)

BIO16

Precipitation of wettest quarter

BIO17

Precipitation of driest quarter

BIO18

Precipitation of warmest quarter

BIO19

Precipitation of coldest quarter

Fig. 1

The distribution points of Pseudopus apodus that were used in this study

We used the Bioclimate variables in Openmodeller v. 1.0.7 (Muñoz et al. 2011) to obtain numerical values for each environmental layer that is associated with our occurrence points (presence and pseudo-absence). To choose the variables with correlations < 0.75, the Pearson correlation coefficient was calculated using SPSS 16.0. Since the variables with correlation more than 0.75 would have the same results in the analysis, thus were eliminated from to prevent repetition. Finally, five variables with low correlation (< 0.75), BIO4 (Temperature Seasonality); BIO2 (Mean Diurnal Range); BIO8 (Mean Temperature of Wettest 86 Quarter); BIO15 (Precipitation seasonality); BIO12 (Annual precipitation) were chosen to run the models.

Modeling of species distribution

Ensemble modeling was chosen to model the species distribution. These models are based on the presence-absence records of species. We considered five modeling techniques, including Surface Range Envelop (SRE, Thuiller et al. 2016), Classification Tree Analysis (CTA; De’ath and Fabricius 2000), Flexible Discriminant Analysis (FDA; Hastie et al. 1994), Multivariate Adaptive Regression Splines (MARS; Friedman 1991), and random forest (RF; Breiman 2001) (Table 1). Species distribution modeling was performed using the BIOMOD2 platform for R.

SDMs were built for the species using pseudo-absence data and presence records to obtain a measure of the SDM accuracy. The most highly applied metric; the area under the curve (AUC) and True skill statistic (TSS) were applied to measure the accuracy of SDMs (Fourcade et al. 2013).

To get the best RUN, the runs were repeated three times and the amount of TSS (True skill statistic) was measured for each RUN. The highest TSS show the best RUN and the best RUN was selected (Table 3) to choose the best model out of it (each RUN includes 5 models) (Table 4) Model accuracy was evaluated using two thresholds: sensitivity (which represented the percentage of correct predictions of presence data) and specificity (which represented the percentage of correct predictions of absence data) (Fielding and Bell 1997) (Fig. 1). Mean of probabilities (MEAN), Coefficient of variation (CV), confidence interval superiority of probabilities(CISUP), median of probabilities (MEDIAN), committee averaging of probabilities (CA), and probabilities weighting mean (WMEAN) were used in ensemble model and the TSS value were measured (Table 5) to achieve the best ensemble model among them.
Table 3

TSS (True Skill Statistics) of all runs

 

Run1

Run2

Run3

Current

0.921

0.969

0.916

Future

0.921

1

0.895

Table 4

AUCs (Area Under the Curves) and TSSs (True Skill Statistics) of all of the models that were used in the study

 

Current

Future

AUC

TSS

AUC

TSS

SRE (surface range envelop)

0.827

0.825

0.836

0.828

CTA (classification tree analysis)

0.972

0.97

0.900

0.897

RF (andom forest)

0.999

0.99

0.971

0.98

MARS (multivariate adaptive regression splines)

0.999

0.99

0.954

0.94.7

FDA (flexible discriminant analysis)

0.997

0.99

0.992

0.99

Table 5

AUCs (Area Under the Curves) and TSSs (True Skill Statistics) of ensemble model that were used in the study

 

MEAN (mean of probabilities)

CV (Coefficient of variation)

CISUP (confidence interval superiority of probabilities)

MEDIAN (median of probabilities)

CA (committee averaging of probabilities)

WMEAN (weighting mean)

TSS

AUC

TSS

AUC

TSS

AUC

TSS

AUC

TSS

AUC

TSS

AUC

Current

0.95

0.99

0.95

0.98

0.960

0.99

0.95

0.99

0.95

0.099

Future

0.96

0.99

0.96

0.99

0.974

0.99

0.96

0.99

0.96

0.99

Results

Run2 was chosen as the best run in current and future models(Table 3) and the best model, according to maximum value AUC and TSS, was the FDA model in current and Future (Table 4).The most suitable region for P. apodus, according to the prediction of the ensemble model (Table 5 and Fig. 2), is changing during different time periods. The species range is predicted to be different by 2070 compared with the present time (Fig. 2). These results indicate the effect of climate change on the species range. The AUC and TSS values for the five models used in this study reached between 0.82 and 0.99 (Table 4).
Fig. 2

The species distribution prediction based upon ensemble models for presence and future. a Maps of the highly accurate model as FDA in current. b Maps of the highly accurate model as FDA in future. c Maps based upon averages among models created using the ensemble method in current, d Maps based upon averages among models created using the ensemble method in Future

The importance of contributed variables for the five models is presented in Table 5 and BIO4 and BIO12 received the highest scores. The result showed that climatic variables, such as Temperature Seasonality and Annual precipitation, are among the most important variables affecting the distribution of Pseudopus apodus.

All of the models used in this study reached an AUC value above 0.8, indicating good overall prediction accuracy (Table 4). The accuracy of RF, MARS, FDA, in the current model, was the highest (AUC = 0.99) followed by CTA (AUC = 0.97) and SRE (AUC = 0.82). In future model, the accuracy of FDA was the highest (AUC = 0.99) followed by RF (AUC = 0.97), MARS (AUC = 0.95), CTA (AUC = 0.90) and SRE (AUC = 0.83).

The ensemble model showed that the most important climatic variable; in current climatic conditions and in those anticipated for the year 2070, for prediction of this species range are Temperature Seasonality and Annual precipitation. According to comparison of either ensemble models, the suitable area for the species at present is plains and valleys, but this lizard migrates to the mountains by 2070. Based on FDA and ensemble maps, it is proposed here that species range will be extended to the east; in particular in higher altitude regions like Afghanistan, otherwise its western range in Jordan will be shrunk.

Discussion

Our study predictions illustrate the spatial distribution of Pseudopus apodus at present time and 2070. According to Ensemble and FDA models, the suitable range for this taxon extends from the Balkan Peninsula to Anatolia, the Caucasus, Central Asia, and the Middle East at present (Fig. 2 a, c). According to A1B scenario, precipitation level in plains and lowlands will be decreased, therefore P. apodus tends to migrate to higher elevations in 2070. Range expansion in Iran and Afghanistan is also predicted by the same scenario but its western range in Jordan will be shrunk (Fig. 2 b, d).

The present distribution of any type of organism is generally attributed to a set of historical and ecological factors (Monge-Nájera 2008). Many studies have recently attempted to develop models and methods to predict the potential effects of future climate changes on biodiversity (e.g. Dawson et al. 2011; McMahon et al. 2011; Bellard et al. 2012). Our results show the main factors, influencing the presence of the European Glass lizard, are temperature and Precipitation. This confirmed the assumption that temperature and precipitation are the most important factors affecting the distribution of this species. According to the both models (current and future), the current distribution range of P. apodus is restricted in the plains and valleys, but expanding to higher altitude from the current to 2070, an assumption which is supported by high AUC and TSS value (Tables 4, 5).

Consequently, temperature seasonality (BIO4) and Precipitation Seasonality (BIO15) have the most effect on species distribution prediction (Table 6). According to A1B scenario, the distribution will be affected by climate change, because increasing CO2 concentration will cause an increase in the global temperature, which results in lowering the precipitation level. High temperature has been defined as one of the most important features for the species’ presence (Wilms and Böhme 2007). In Europe, a physiological niche model based on thermal performance curves for sprint speed suggesting a positive effect of climate warming on lizards (Deutsch et al. 2008). There is strong evidence indicating that global temperature and rainfall patterns have been changing as a consequence of human activities in recent decades, and this trend is likely to continue for many coming decades (IPCC 2013; Solomon 2007; Zhu et al. 2010).
Table 6

Importance of contributed variables in models used in this study

Model name

SRE1

CTA2

RF3

MARS4

FDA5

Currentnt

Future

Currentnt

Future

Currentnt

Future

Currentnt

Future

Currentnt

Future

BIO2

0.145

0.059

0.000

0.000

0.018

0.018

0.150

0.205

0.175

0.075

BIO4

0.256

0.288

0.357

0.327

0.244

0.136

0.767

0.570

0.732

0.364

BIO8

0.255

0.163

0.501

0.000

0.283

0.066

0.001

0.001

0.013

0.025

BIO12

0.293

0.295

0.452

0.735

0.348

0.264

0.359

0.313

0.293

0.309

BIO15

0.257

0.413

0.000

0.616

0.013

0.268

0.344

0.591

0.468

0.525

1 surface range envelop, 2 classification tree analysis, 3 random forest, 4 Multivariate adaptive regression splines, 5 flexible discriminant analysis

Based on the predicted maps, the distributional range of P. apodus will be significantly affected by climate change and human impact (including human population growth, deforestation, agricultural expansion), especially by the year 2070 compared to the current situation. There are generally two spatial responses for wild animals affected by global warming and habitat destruction and fragmentation: shifting their distributions to higher elevations and shifting their distributions northward within the North Hemisphere. This kind of latitudinal or altitudinal expansions has been reported in mobile animals and in plants (Parmesan 2006). A recent global study mentioned that 19.1% of the world’s terrestrial reptiles are threatened (Bo¨hm et al. 2013). However, Sinervo et al. (2010) mentioned that local extinctions may reach 39% and at a global scale 20% species loss by 2080. A global study suggests that terrestrial vertebrate ectotherms seem to be more sensitive to climatic cooling than warming, showing an increase in potential range for a large number of species under the latter scenarios. (Arau´jo et al. 2006). Therefore, it can be excluded P. apodus will expands its habitats to suitable areas in mountainous regions in near future (Fig. 2b, d). The maps in Fig. 2a, b show the standard deviations from exact predictions in the FDA models and illustrate the biases in the FDA algorithm. This bias can be resolved by using ensemble modeling that integrates different models to reach the best unbiased model (Fig. 2c, d). The output of ensemble modeling was better than the FDA method in predicting distributions, because ensemble predictions were made according to five models, but the FDA shows the highest accuracy model with higher AUC and TSS values than others. The ensemble predictions for future climate scenarios were used by different authors (Buisson et al. 2010). Prediction of species distribution by different approaches reached different results and many pitfalls are present among them, but the ensemble approach uses multiple algorithms to try to avoid any bias in models that could occur with a single modeling approach. Ensemble approaches can improve the accuracy of species distribution prediction (Marmion et al. 2009).

Notes

Acknowledgements

We wish to thank Dr. Ali Reza Keykhosravi for his help and support during this study.

References

  1. Anderson S (1999) The lizards of Iran. Society for the study of Amphibians and Reptiles Contributions to Herpetology. Ithaca, New YorkGoogle Scholar
  2. Araújo MB, Thuiller W, Pearson RG (2006) Climate warming and the decline of amphibians and reptiles in Europe. J Biogeogr 33:1712–1728CrossRefGoogle Scholar
  3. Başoğlu M, Baran I (1977) The reptiles of Turkey. Ilker MatbaasiGoogle Scholar
  4. Bellard C, Bertelsmeier C, Leadley P, Thuiller W, Courchamp F (2012) Impacts of climate change on the future of biodiversity. Ecol Lett 15:365–377CrossRefPubMedPubMedCentralGoogle Scholar
  5. Böhm M et al (2013) The conservation status of the world’s reptiles. Biol Conserv 157:372–385CrossRefGoogle Scholar
  6. Breiman L (2001) Random forests. Mach Learn 45:5–32CrossRefGoogle Scholar
  7. Buisson L, Thuiller W, Casajus N, Lek S, Grenouillet G (2010) Uncertainty in ensemble forecasting of species distribution. Glob Change Biol 16:1145–1157CrossRefGoogle Scholar
  8. Cox NA, Temple HJ (2009) European red list of reptiles. vol 333.957094 E89e. IUCN, Gland (Suiza)Google Scholar
  9. Dawson TP, Jackson ST, House JI, Prentice IC, Mace GM (2011) Beyond predictions: biodiversity conservation in a changing climate. Science 332:53–58CrossRefPubMedGoogle Scholar
  10. De’ath G, Fabricius KE (2000) Classification and regression trees: a powerful yet simple technique for ecological data analysis. Ecology 81:3178–3192CrossRefGoogle Scholar
  11. Deutsch CA, Tewksbury JJ, Huey RB, Sheldon KS, Ghalambor CK, Haak DC, Martin PR (2008) Impacts of climate warming on terrestrial ectotherms across latitude. Proc Natl Acad Sci USA 105:6668–6672CrossRefPubMedPubMedCentralGoogle Scholar
  12. Fielding AH, Bell JF (1997) A review of methods for the assessment of prediction errors in conservation presence/absence models. Environ Conserv 24:38–49CrossRefGoogle Scholar
  13. Fourcade Y, Engler JO, Besnard AG, Rödder D, Secondi J (2013) Confronting expert-based and modelled distributions for species with uncertain conservation status: a case study from the corncrake (Crex crex). Biol Conserv 167:161–171CrossRefGoogle Scholar
  14. Friedman JH (1991) Multivariate adaptive regression splines. Ann Stat 1:1–67CrossRefGoogle Scholar
  15. Hastie T, Tibshirani R, Buja A (1994) Flexible discriminant analysis by optimal scoring. J Am Stat Assoc 89:1255–1270CrossRefGoogle Scholar
  16. Marmion M, Parviainen M, Luoto M, Heikkinen RK, Thuiller W (2009) Evaluation of consensus methods in predictive species distribution modeling. Divers Distributions 15:59–69CrossRefGoogle Scholar
  17. McMahon SM et al (2011) Improving assessment and modelling of climate change impacts on global terrestrial biodiversity. Trends Ecol Evol 26:249–259CrossRefPubMedGoogle Scholar
  18. Miller J (2010) Species distribution modeling. Geogr Compass 4:490–509CrossRefGoogle Scholar
  19. Monge-Nájera J (2008) Ecological biogeography: a review with emphasis on conservation and the neutral model. Gayana 72:102–112Google Scholar
  20. Muñoz DSME et al (2011) openModeller: a generic approach to species’ potential distribution modeling. GeoInformatica 15:111–135CrossRefGoogle Scholar
  21. Nakicenovic N, Alcamo J, Grubler A, Riahi K, Roehrl R, Rogner H-H, Victor N (2000) Special Report on Emissions Scenarios (SRES), A Special Report of Working Group III of the Intergovernmental Panel on Climate Change. Cambridge University PressGoogle Scholar
  22. Nally RM, Fleishman E (2004) A successful predictive model of species richness based on indicator species. Conserv Biol 18:646–654CrossRefGoogle Scholar
  23. Parmesan C (2006) Ecological and evolutionary responses to recent climate change. Annu Rev Ecol Evol Syst 37:637–669CrossRefGoogle Scholar
  24. Root TL, Price JT, Hall KR, Schneider SH, Rosenzweig C, Pounds JA (2003) Fingerprints of global warming on wild animals and plants. Nature 421:57–60CrossRefPubMedGoogle Scholar
  25. Sinervo B et al (2010) Erosion of lizard diversity by climate change and altered thermal niches. Science 328:894–899CrossRefPubMedGoogle Scholar
  26. Šmíd J, Moravec J, Kodym P, Kratochvíl L, Yousefkhani SSH, Frynta D (2014) Annotated checklist and distribution of the lizards of Iran. Zootaxa 3855:1–97CrossRefPubMedGoogle Scholar
  27. Solomon S (2007) Climate change 2007-the physical science basis: working group I contribution to the fourth assessment report of the IPCC vol 4. Cambridge University PressGoogle Scholar
  28. Thuiller W, Georges D, Engler R, Breiner F, Georges MD, Thuiller CW (2016) Package ‘biomod2’Google Scholar
  29. Walther G-R, Berger S, Sykes MT (2005) An ecological ‘footprint’ of climate change. Proc R Soc B 272:1427–1432CrossRefPubMedPubMedCentralGoogle Scholar
  30. Wilms T, Bohme W (2007) A review of the taxonomy of the spinytailed lizards of Arabia (Reptilia: Agamidae: Leiolepidinae: Uromastyx). Fauna Arabia 23:435–468Google Scholar
  31. Zhu L, Zhan X, Wu H, Zhang S, Meng T, Bruford M, Wei F (2010) Drastic reduction of the smallest and most isolated giant panda population: implications for conservation. Conserv Biol 24:1299–1306CrossRefPubMedGoogle Scholar

Copyright information

© The Ecological Society of Japan 2017

Authors and Affiliations

  1. 1.Department of Biology, Faculty of SciencesRazi UniversityKermanshahIran
  2. 2.Department of Biology, Faculty of SciencesHakim Sabzevari UniversitySabzevarIran
  3. 3.Department of Biology, Faculty of SciencesGolestan UniversityGorganIran
  4. 4.Young Researchers and Elite ClubIslamic Azad UniversityShirvanIran

Personalised recommendations