Biodivers Conserv (2012) 21:2845–2863 DOI 10.1007/s10531-012-0341-z ORIGINAL PAPER

Iterative species distribution modelling and ground validation in endemism research: an Alpine jumping bristletail example Lukas J. Rinnhofer • Nu´ria Roura-Pascual • Wolfgang Arthofer • Thomas Dejaco • Barbara Thaler-Knoflach • Gregor A. Wachter • Erhard Christian • Florian M. Steiner • Birgit C. Schlick-Steiner

Received: 29 November 2011 / Accepted: 17 July 2012 / Published online: 4 August 2012 Ó Springer Science+Business Media B.V. 2012

Abstract Endemic species play an important role in conservation ecology. However, knowledge of the real distribution and ecology is still scarce for many endemics. The aims of this study were to predict the distribution of the short-range endemic Alpine jumping bristletail Machilis pallida; to evaluate the actual level of endemism via ground validation using an iterative approach for testing the models in field trips and increasing the quality of the prediction step by step; and to test the potential of species distribution modelling for increasing the knowledge about the ecological niche. Based on seven known locations of M. pallida, we used species distribution modelling via Maxent. After a set of seven field trips a new model was built if new locations were found. Three such iterations were performed to increase model quality. We discovered four new locations of M. pallida, increasing the area of known distribution from 470 to 4,890 km2. The distribution of M. pallida is thus wider than formerly known, but our results support Eastern Alpine endemism of the species. The knowledge about the ecological niche could be increased due to the newly found locations. Our study showcases the potential of the iterative approach of modelling and ground validation to evaluate the actual level of endemism and the ecological niche in Alpine species and beyond. Florian M. Steiner and Birgit C. Schlick-Steiner contributed equally to this work. L. J. Rinnhofer (&)  W. Arthofer  T. Dejaco  B. Thaler-Knoflach  G. A. Wachter  F. M. Steiner  B. C. Schlick-Steiner Molecular Ecology Group, Institute of Ecology, University of Innsbruck, Technikerstrasse 25, 6020 Innsbruck, Austria e-mail: [email protected] N. Roura-Pascual Departament de Cie`ncies Ambientals, Universitat de Girona, Campus Montilivi, 17071 Girona, Catalonia, Spain N. Roura-Pascual ` rea de Biodiversitat, Centre Tecnolo`gic Forestal de Catalunya, Ctra. St. Llorenc¸ de Morunys km 2, A 25280 Solsona, Catalonia, Spain E. Christian Institute of Zoology, University of Natural Resources and Life Sciences, 1180 Vienna, Austria

123

2846

Biodivers Conserv (2012) 21:2845–2863

Keywords Bristletails  Endemism  ENMTools  Maxent  Machilis pallida  Species distribution modelling

Introduction Endemism, the restriction of a taxon to a specific geographical area (Cowling 2001), has been recognised as a biogeographical pattern with important ramifications for research into evolution, ecology, and conservation. Endemic species are relevant to community organisation and ecosystem services (Hernandez et al. 2006; Papes and Gaubert 2007; Wanger et al. 2011) and the awareness of their contribution to global biodiversity is growing rapidly (Cassagne et al. 2006; Deharveng et al. 2000; Garzon et al. 2008). Sound knowledge about the geographical distribution and ecological demands of endemic species is central to biodiversity conservation (Myers et al. 2000). The European Alps are amongst the most important hotspots of endemism in Europe (Myers et al. 2000; Rabitsch and Essl 2009). Geographical isolation, climatic oscillations, low vagility and narrow ecological niches have been discussed as limiting the distribution of Alpine endemic species (Holdhaus 1954; Huemer 1998; Rabitsch and Essl 2009). However, often only a small number of populations are known for endemic species, which can either represent a true biological finding or a sampling artefact. In many cases, the true distribution of endemic species is not known and ecological knowledge is limited (Huemer 2011). This raises the question of how to uncover the actual distribution and the level of endemism of a given species. Species distribution modelling (SDM) offers the opportunity to estimate the distribution of a species in an effective way making it a widely used method for conservation science and biogeographical studies (Elith et al. 2006; Gormley et al. 2011; Newbold et al. 2009; Raxworthy et al. 2003; Rou´ra-Pascual and Suarez 2008; Steiner et al. 2008; Ward 2007). It combines spatial information with environmental variables (e.g., temperature, precipitation) and creates a model that displays areas of high similarity to the currently known locations of a species (Anderson et al. 2003; Peterson 2003). Species distribution models are useful for estimating the potential of species occurrence in poorly surveyed areas and for directing new surveys to places where species are likely to be found (Hernandez et al. 2008; Sa´nchez-Ferna´ndez et al. 2011). The template of the model rarely includes all the biotic and abiotic factors which shape the real distribution of a species. Entering more substantial information leads, of course, to a more accurate prediction of the possible range of a species, just as larger sample numbers do (Anderson et al. 2003; Pearson et al. 2007). Even good data, however, do not necessarily entail congruence of predicted and actual species ranges. SDM can hardly include limited means of dispersion, geographical barriers, long-term climatic effects (e.g., glaciations), or anthropogenic effects (Anderson et al. 2003; Arau´jo and Pearson 2005; Svenning and Skov 2004). The results of SDM thus require careful interpretation, but former studies have shown their high informative value (e.g., Hernandez et al. 2008; Rebelo and Jones 2010). The prediction of the potential distribution of endemic species is often impeded by the sparsity of records. Several studies have suggested that the SDM software Maxent (Phillips et al. 2006) performs well when modelling distributions of species with small sample sizes (\10 samples) compared to other software (Costa et al. 2010; Elith et al. 2006; Hernandez et al. 2006; Pearson et al. 2007; Williams et al. 2009; Wisz et al. 2008). However, to our

123

Biodivers Conserv (2012) 21:2845–2863

2847

knowledge, Maxent-based models have only infrequently been used for ground validation via actual sampling in areas predicted by the models (but see Jackson and Robertson 2011; Rebelo and Jones 2010; Williams et al. 2009). The objective of this study is to use an iterative modelling approach to examine the distribution and ecology of the Alpine, short-range endemic jumping bristletail Machilis pallida Janetschek, 1949, known from seven localities in the Tyrolean Alps prior to our study (Table 1). Jumping bristletails (Microcoryphia) are basal, wingless insects with about 480 species known worldwide (Sturm and Machida 2001). The Eastern Alpine jumping bristletails of the genus Machilis comprise 22 species, of which 17 are short-range endemics (T. Dejaco, unpubl. data), making them an excellent model system for endemism research. Biological knowledge about M. pallida is scarce. Based on the seven previously known records the species seems to prefer south-to-southeast facing slopes with scree of approx. 10 cm carbonate rocks at 2,000–2,600 m a.s.l. (Christian and Knoflach 2009; Janetschek 1951; Wachter et al. in press). The geographic range of M. pallida extends 65 km N–S/28 km E-W and covers just about 470 km2 (Table 1), rendering the species representative of putatively very narrow endemism. We test, as far as we know for the first time, Maxent’s potential as a tool for modelguided field work in endemism research; the highly heterogeneous, alpine area inhabited by M. pallida and the very small sample size challenge the task. The apparently small range raises questions of relevance to biogeography and conservation biology: (1) Is the real distribution of M. pallida wider than currently known? (2) Is M. pallida indeed endemic to the Eastern Alps? (3) Can SDM increase the knowledge about the ecological niche of endemic species in highly heterogeneous areas? We tackle these questions in an iterative process by creating new species distribution models after rounds of seven field trips, thus improving successive models by entering data from newly discovered populations.

Materials and methods Iterative modelling approach and ground validation To shed light into the ecology and distribution of M. pallida, we adopted an iterative approach which consisted of three rounds of SDM/ground-validation processes. The initial model (Model 1) was calibrated based on the seven published records of the species (Janetschek 1949, 1951; Wachter et al. in press) (Table 1). To spot new locations of M. pallida and test the quality of this initial model, we selected seven places within the predicted area of Model 1 for field sampling (Round 1). If new populations were found in the course of the field trips, a new model was created after finishing each round of seven field trips, adding the new localities to the previous occurrence data following the same procedure. We repeated this SDM/ground-validation process two additional times (Round 2 and 3), surveying a total of 21 new localities in the search of M. pallida. After the third round, the final distribution model was created using all records available to have an updated representation of the species’ potential distribution in the Alps. To select localities for sampling, we converted the model predictions to binary maps (suitable/unsuitable) using the lowest presence threshold (LPT; also known as minimum training presence), i.e., the lowest prediction value returned by Maxent for a location with observed presence of the species. Areas predicted above the LPT were identified as areas at least as suitable as those where the species had already been proved to occur

123

2848

Biodivers Conserv (2012) 21:2845–2863

Table 1 Known populations of M. pallida at the beginning of the study and localities sampled during successive field trips Geographical coordinates M. pallida locations known at the beginning of this study Ma¨uerlscharte 46.99°N, 11.53°E 47.10°N, 11.38°E

m a.s.l.

M. pallida

Rock type

2,325

x

Carbonate

2,280

x

Carbonate

Kesselspitze* La¨mpermahdspitze*

47.12°N, 11.38°E

2,220

x

Carbonate

Padasterjochhaus

47.08°N, 11.36°E

2,268

x

Carbonate

Obernberger Tribulaun

46.99°N, 11.39°E

2,012

x

Carbonate

Schlern, Refugio Molignon Schlern, Murmeltierhu¨tte

46.41°N, 11.65°E

2,112

x

Carbonate

46.51°N, 11.70°E

2,173

x

Carbonate

Round 1 Brenta

46.18°N, 10.87°E

2,220

x

Carbonate

Drei Zinnen

46.61°N, 12.30°E

2,370

x

Carbonate

Adamello

46.12°N, 10.63°E

2,475

Silicate

Val di Sole

46.35°N, 10.81°E

2,437

Silicate

Marmolada

46.42°N, 11.85°E

2,407

Carbonate

Cima d’Asta

46.18°N, 11.61°E

2,629

Silicate

Pale di San Martin

46.24°N, 11.88°E

2,287

Carbonate

Round 2 Nu¨rnberger Hu¨tte

47.00°N, 11.21°E

2,381

Seegrube

47.30°N, 11.38°E

1,950

x

Silicate Carbonate

Seekofel

46.67°N, 12.09°E

2,165

x

Carbonate

Penser Joch

46.35°N, 11.44°E

2,317

Silicate

Schwarzenstein

47.00°N, 11.89°E

2,420

Silicate

S-Charls

46.73°N, 10.35°E

2,182

Carbonate

Monte Frerone

45.95°N, 10.41°E

2,608

Silicate

Round 3 Dawinspitze

47.17°N, 10.30°E

2,494

Carbonate

Rofan

47.45°N, 11.78°E

2,024

Carbonate

Mojstrovka

46.43°N, 13.72°E

2,143

Carbonate

Il-Fourn

46.67°N, 10.23°E

2,158

Carbonate

Monte Baldo

45.72°N, 10.85°E

2,000

Carbonate

Lienzer Dolomiten

46.74°N, 12.68°E

2,370

Carbonate

Ortler

46.52°N, 10.56°E

2,257

Carbonate

00

* Locations placed in the same grid cell (resolution 30 ) and used as one location in the modelling process; x indicates a population of M. pallida

(Pearson et al. 2007). In selecting sampling sites within the predicted range, the following procedure was applied: Within the set of suitable areas, in each sampling round we first selected a geographic area without confirmed presence/absence of the species (e.g., the southernmost range) to potentially extend our knowledge on the distribution of M. pallida. Second, within this general geographical area, we used Google Earth (http://earth. google.com) and expert knowledge to identify suitable scree slopes among the predicted habitats. The formerly known populations of M. pallida occurred on scree slopes with an extent of about two to five kilometres. Thus, in order to avoid autocorrelation, each

123

Biodivers Conserv (2012) 21:2845–2863

2849

sampling location had to be 10 km off any known population and any location already sampled (with or without success); we considered a distance of 10 km as sufficient due to the low vagility of jumping bristletails and the heterogeneous landscape we investigated. When more than seven such locations were identified in the area selected, ease of access determined the final decision. Locations of both carbonate (e.g., limestone or dolomite) and silicate (e.g., granite) rock were sampled, the two predominant rock types of the Eastern Alps (Pfiffner 2010). Sampling was done by hand collection during up to 4 h (from 9 am to 1 pm) per site, moving from the bottom of the slope to the top to minimise disturbance. Specimens were immediately fixed in 96 % EtOH and stored individually in tubes at -20 °C at the Molecular Ecology Group, Institute of Ecology, University of Innsbruck. Maxent settings We used the SDM software Maxent 3.3.2 (Phillips et al. 2006) to estimate the potential range of the species. Maxent predicts the distribution of species using the principle of maximum entropy, which finds the probability that is the closest to uniform combining environmental data with presence localities and background records sampled from the overall study area (Phillips et al. 2006; Ward 2007). As recommended by Phillips et al. (2009, 2006), we used default settings. The Maxent default setting Remove duplicate presence record reduced the number of samples from seven to six because two localities were placed in the same grid cell (cell size approx. 1 km2) (Table 1). Linear features were used when sample size was smaller than 10 occurrence points, and linear and quadratic features when 10 or more occurrence points were available (Phillips et al. 2004). Due to the small number of occurrence points available, we could not split the data into test and training sets for model evaluation and we used the ‘‘leave-one-out’’ approach instead (cf. Pearson et al. 2007; see details below). For performance analysis of different variable and background combinations with the Jackknife test, we set the number of replicates equal to the total number of occurrence points and Type to Replicate. Selection of environmental variables The proper selection of environmental variables has been an issue in SDM (Beaumont et al. 2005; Kriticos and Randall 2001; Ward 2007). Because of the small set of initial occurrence localities and to avoid overfitting (Stockwell and Peterson 2002), we aimed at keeping the number of environmental variables small. Our initial set consisted of 21 climatic and terrain-related variables. Nineteen climatic variables with a resolution of 30 arc seconds were downloaded from the WorldClim database (Hijmans et al. 2005). The variable altitude was added from the NASA Shuttle Radar Topography Mission (SRTM) data set made available by the CGIAR Consortium for Spatial Information (available at: http://srtm.csi.cgiar.org). The variable surface area was included as it represents information about steepness and roughness of the terrain. Based on the altitude data, surface area was derived using the Arc GIS Surface Area Tool made available by Jenness (2004). This tool calculates surface area by generating eight 3-dimensional triangles connecting each 3000 -cell centrepoint with the centrepoints of the eight surrounding cells, then calculating and summing the area of the portions of each triangle that lies within the cell boundary (Jenness 2004). Sampling bias is known to potentially impact model quality (Elith and Graham 2009; Hortal et al. 2008; Pearson et al. 2007; Phillips et al. 2009; Rebelo and Jones 2010). Therefore we did not include rock type

123

2850

Biodivers Conserv (2012) 21:2845–2863

Table 2 Environmental variables used for predicting the distribution of M. pallida in the Alps (Hijmans et al. 2005; Jenness 2004) and their contribution (%) as returned by Maxent Name

Abbreviation

Reason for elimination

Model 1

Model 2

Model 3

82.7

Temperature of warmest quarter

Bio10

–

82.6

79.9

Temperature of coldest quarter

Bio11

–

0

0.2

0

Precipitation of driest quarter

Bio16

–

0.2

0.1

0.8 15.1

Precipitation of wettest quarter

Bio17

–

15.2

17.3

Surface area

Surf

–

1.9

2.4

1.4

Annual mean temperature

Bio1

CORR

–

–

–

Mean diurnal range

Bio2

COMP

–

–

–

Isothermality

Bio3

COMP

–

–

–

Temperature seasonality

Bio4

COMP

–

–

–

Max temperature of warmest month

Bio5

CORR

–

–

–

Min temperature of coldest month

Bio6

CORR

–

–

–

Temperature annual range

Bio7

COMP

–

–

–

Mean temperature of wettest quarter

Bio8

COMP

–

–

–

Mean temperature of driest quarter

Bio9

COMP

–

–

–

Annual precipitation

Bio12

CORR

–

–

–

Precipitation of wettest month

Bio13

CORR

–

–

–

Precipitation of driest month

Bio14

CORR

–

–

–

Precipitation seasonality

Bio15

COMP

–

–

–

Precipitation of warmest quarter

Bio18

COMP

–

–

–

Precipitation of coldest quarter

Bio19

COMP

–

–

–

Altitude

Alt

CORR

–

–

–

Reasons for elimination of variables: COMP eliminated due to exclusion of complex variables, CORR eliminated due to results of correlation analysis (for details, see ‘‘Materials and methods’’)

as variable, because we suspected biased sampling behind the fact that all known locations were from carbonate rock. Even Riezler (1939) doubted the correlation between machilid occurrence and bedrock type. All variables were cut to the same extent as the three background areas at a spatial resolution of 30 arc seconds (approx. 1 km2). We then applied a two-step procedure to this initial pool of variables to retain only the most relevant ones (Table 2). Exclusion of complex variables Running a principal component analysis (PCA; calculated via the software PAST, Hammer et al. 2001) did not yield any statistically robust results for the extracted variables in our study area. In order to reduce complexity in the interpretation of the effect of abiotic factors on the model and opting for simple variables rather than compound ones, we decided to exclude climatic variables where there is a mix of temperature and precipitation (e.g., mean temperature of wettest quarter) and exclusively used variables describing one

123

Biodivers Conserv (2012) 21:2845–2863

2851

of the two (e.g., mean precipitation of driest quarter). Furthermore, we excluded variables which describe changes of temperature or precipitation over the annual cycle (e.g., temperature seasonality) because we considered absolute values as easier to interpret (Table 2). Correlation analysis To reduce redundancy within the remaining environmental variables, we used pairwise correlation values (Pearson Correlation Coefficient, calculated via STATISTICA V. 10, StatSoft Inc. 2010) of 20,000 randomly selected points of different backgrounds (Fig. 1). Due to the small area, the background for the Known Distribution showed intercorrelation among nearly all variables. Therefore, we used only the backgrounds AA and EA in the subsequent process. Seven variables showing a correlation [0.80 were eliminated for redundancy (Giovanelli et al. 2010) (Table 2). We opted to use surface area to represent topographical effects, because altitude correlated [0.80 with various temperature variables. Mean temperature of warmest quarter and mean temperature of coldest quarter correlated[0.80 but we decided to retain both of them, as each of the two values might set an ecological limit to M. pallida. Our final set of variables consisted of mean temperature of warmest quarter, mean temperature of coldest quarter, precipitation of wettest quarter, precipitation of driest quarter, and surface area (Table 2). Selection of background The selection of an appropriate background area has been a widely discussed topic in SDM (e.g., Elith et al. 2006; Phillips et al. 2009; VanDerWal et al. 2009) and so far no consensus optimal method has been found. Model performance changes when random pseudoabsence points are drawn from too small or too large a background area as compared to the actual distribution of a species. The contribution of each variable in the model differs with background size and therefore selection of background area can impact the modelling result (VanDerWal et al. 2009). In the absence of detailed knowledge on the real species

Fig. 1 Map of the study area with the different background areas used to calibrate the models. AA area of the Alps, covering the whole Alps from Nice to Vienna, EA Eastern Alps, covering the eastern Alps from the Rhein-Splu¨gen-Como line in the west to Vienna in the east, KD 1 known distribution of M. pallida prior to this study, KD 2 and KD3 known distribution of M. pallida after Round 1 and Round 2 of sampling, respectively. Brackets show the area of Maxent prediction in Fig. 2

123

2852

Biodivers Conserv (2012) 21:2845–2863

Table 3 Results of variable and background selection with jackknife and p value tests following Pearson et al. (2007) Variables

Background area

Success rate

p value

Model 1

Bio10, Bio11, Bio16, Bio17, Surf

Eastern Alps

0.833 (5 out of 6)

0.000023

Model 2

Bio10, Bio11, Bio16, Bio17, Surf

Eastern Alps

0.875 (7 out of 8)

0.000001

Model 3

Bio10, Bio11, Bio16, Bio17, Surf

Eastern Alps

0.900 (9 out of 10)

0.002297

Success rate shows the results of the jackknife test Bio10 temperature of warmest quarter, Bio11 temperature of coldest quarter, Bio16 precipitation of driest quarter, Bio17 precipitation of wettest quarter, Surf surface area

distribution in our study, we created three background areas (Fig. 1) to account for the effect of different backgrounds on the modelling process. First, we used the area spanned by all sample locations to represent the known distribution (KD) as background. When there were new populations added to the analysis during the iterative approach, we adapted the KD background accordingly. The second background covered the Eastern Alps (EA) from the Rhine-Splu¨gen-Como line (Schmo¨lzer 2001) in the west to the eastern end of the Alps in Vienna. The third background included the entire Alpine area (AA) from Grenoble in the west, to Nice in the south and Vienna in the north-east (Fig. 1). For small sample sizes, widely used performance criteria such as Area Under the Curve or Kappa statistics are not recommended (Boyce et al. 2002; Pearson et al. 2007) and we thus cross-validated models via jackknife tests, the so called ‘‘leave-one-out’’ method as described by Pearson et al. (2007), using the LPT (Kumar and Stohlgren 2009; Pearson et al. 2007) to evaluate the model performance by predicting the locality excluded from the modelling data. For cross validating, we used the variables selected for SDM and ran models with all three backgrounds. Maxent settings were left on default with the number of replicate runs for the jackknife validation equalling the number of locations available. We compared the results of each model via the p value test suggested by Pearson et al. (2007) and selected the combination of variables and background with the highest significance and lowest error-number for the final model. The selected background was then used for model calibration. The model then was projected on the whole area of the Alps for the final prediction of each model. In all three modelling iterations the highest-ranked model in the jackknife-validation included the five selected variables and the Eastern Alps as background. This combination had only one omission error (wrongly predicted known location) in each of the three modelling rounds (Table 3). Tests for differences among models We analysed Maxent models with niche identity and background similarity tests by the means of Ecological Niche Modelling (ENM) Tools (Warren et al. 2008), which implements various methods for quantifying niche overlap and background similarity. We evaluated if the predicted niche of M. pallida changed significantly over the three modelling rounds. Niche overlap values were calculated as Schoener’s D metric (Schoener 1968), comparing all three models among each other. D is a similarity index of niche-space for two models and compares values of each grid cell extracted from the Maxent prediction; ENMTools normalises the values of the Maxent models so that scores with the

123

Biodivers Conserv (2012) 21:2845–2863

2853

geographic space sum to 1 (Warren et al. 2010). A result of D = 0 means that models have no overlap and D = 1 that models represent identical niches (Warren et al. 2008). We applied a niche identity test to search for differences between model predictions and random samples of data. Identity tests pool together the occurrence points of two given models and use an equivalent number of randomly selected points from the study area (i.e., background points) to randomise two new samples of localities that can be considered as null model predictions. By this means, 100 pseudoreplicates were created. The niche overlap between these two sets of null model predictions was then compared with the overlap of the original model predictions. Niche identity test was run as a one-tailed test (Warren et al. 2008). Background similarity tests were run to determine whether the differences of the three model predictions derived from the iterative process can be explained by the environmental background or if the model predictions are more similar than expected based on the background available. These tests compare all occurrence points from model A to the background from model B with the same number of randomly chosen background points as occurence points included in the model (Warren et al. 2008). One hundred pseudoreplicates were run for each comparison. Instead of two or more different species as in McCormack et al. (2010) and Peterson (2011), we used the predicted model distribution from each round to compare change over the iterative process of modelling and ground validation. We used data from the KD area (see ‘‘Selection of background’’) of each model as environmental background for all analysis in ENMTools. The background similarity test was treated as a two-tailed test (Warren et al. 2008). To examine change in the ecological niche of M. pallida with the discovery of additional locations, we also conducted PCA using the values of the environmental variables considered in the modelling exercise (see Table 2) for all localities with confirmed presence of the species. In a bivariate plot of the first two PCA axes, we drew the convex hulls that included all records of the species at each iteration round. The analysis of the axis loadings allowed identifying the contribution of variables to change in the species’ ecological niche. As in the selection of environmental variables, this analysis was conducted using the software PAST. Species identification Specimens were identified at the species level using the morphology-based key for European Microcoryphia by Palissa (1964). Diagnostic morphological characters for M. pallida are its pigmentation on the first and second segment of the palpus maxillaris, lack of pigmentation on the legs, an overall light, greyish appearance, gonapophyse VIII consisting of 48–56 segments, and a digging-claws-bearing ovipositor which projects beyond the tip of styli IX. To confirm species identification, mitochondrial DNA was sequenced for all individuals. A volume of 1–2 mm3 of abdominal muscle tissue was excised from each specimen and extracted with the GenEluteTM Mammalian Genomic DNA Miniprep kit (Sigma-Aldrich St. Louis, MO, USA). An approx. 1,000 bp fragment of the mitochondrial cytochrome C oxidase I (COI) gene was amplified in a PCR containing each 0.2 lM Machilis-specific forward and reverse primers (MachF1: 50 -ACAAAYCATAAAGATATTGG-30 ; MachR3: 50 TCTATTCCGTGAAGGGTTGC-30 ) (Dejaco et al. 2012), 0.5 U MyTaq polymerase (Bioline, UK), the buffer provided with the enzyme, and 1 ll of DNA template in a total volume of 20 ll. The cycling protocol included following steps: 95 °C—2 min; 359 (95 °C—40 s; 52 °C—40 s; 72 °C—60 s); 72 °C—10 min. Amplification was verified by agarose gel electrophoresis and amplicons were purified using the PeqGOLD Cycle Pure kit

123

2854

Biodivers Conserv (2012) 21:2845–2863

(Peqlab, Germany) following the instructions of the manufacturer. Sanger sequencing was performed by a commercial provider using the forward PCR primer. Sequences were edited with BioEdit 7.0.9 (Hall 1999) and aligned with Clustal W (Larkin et al. 2007). MEGA5 (Tamura et al. 2011) was used to calculate pairwise distances.

Results Iterative modelling and ground validation The results of Model 1 with an LPT of 0.54 suggested a wide distribution of M. pallida over the Eastern Alps with a concentration of the highest values in the areas of Adamello, Brenta and the Dolomites (Fig. 2). Also a small area in the Western Alps (combined approx. 300 km2 in the Pennine and Bernese Alps, Switzerland) was predicted with values above the LPT (data not shown). We decided to focus our sampling effort in the first round of the iterative process on areas closer to the already known populations and did not consider the Western Alpine area for ground validation, but focused on field trips to the southern prediction areas of Model 1. Two of the seven field trips based on Model 1 were successful: M. pallida was found at the Brenta group and the Drei Zinnen (Table 1). The distribution predicted by Model 2 (into which the two new locations were included) covered about the same area as in Model 1 but was more fragmented and had an LPT of 0.39 (Fig. 2). The area predicted as suitable in the Western Alps had the same extent but lower predicted values than in Model 1. Thus, we again decided not to sample this area. Sampling in this second round was directed to the northern part of the predicted distribution. Another seven field trips were conducted and again at two locations, Seekofel and Seegrube, M. pallida was collected (Table 1). Model 3 was run with now 10 occurrence points. Again, the predicted distribution area was similar to the former models but an even higher fragmentation of the predicted area occurred. The predicted values for the Western Alps were now below the LPT (0.34) (Fig. 2). In the final seven field trips, we searched for M. pallida in the outer ranges of the predicted area, east and west of the known distribution area, but no further populations were retrieved (Table 1). Four out of 21 field trips were successful and we extended the known range of M. pallida by 135 km (107 %) in the north–south and by 110 km (290 %) in the east–west extension. The total area increased about tenfold to 4,890 km2. Fourteen and seven field trips were conducted at carbonate-rock and silicate-rock locations, respectively (Table 1). All newly discovered populations were found at carbonate-rock locations, on rock scree slopes at an altitude between 1,950 m and 2,370 m a.s.l. (Table 1). Based on the analysis of variable contribution as given by Maxent, mean temperature of the warmest quarter had the highest contribution in all three models, followed by precipitation of the driest quarter; the contribution of mean temperature of the coldest quarter, precipitation of the wettest quarter and surface area combined was just about 2 % (Table 2). In the PCA of the environmental variables from successfully sampled locations, the first two principal components together explained 98.1 % (93.9 and 4.2 %, respectively) of the variation in the data (Table 4). The first and most influential component represented a precipitation gradient (loading of ?0.64 with precipitation of the driest quarter, Bio16, and of ?0.72 with precipitation of the wettest quarter, Bio17), and the second component a temperature-precipitation gradient (loading of ?0.67 with temperature of the warmest quarter, Bio10, and of ?0.61 with precipitation of the driest quarter, Bio16) (Table 4). The position of the locations in the PCA plot suggested that the locations

123

Biodivers Conserv (2012) 21:2845–2863

2855

Fig. 2 Predicted distribution of M. pallida in the Eastern Alps for the three models derived from the iterative approach (a–c). Overview d shows the locations known prior to this study and all 21 locations sampled here. Only values above the Lowest Presence Threshold (Model 1: 0.56; Model 2: 0.45; Model 3: 0.34) are depicted. As there were no new findings in the sampling round 3, Model 3 also represents the final model. Warmer colours show higher prediction values, cooler colours represent lower prediction values. Grey areas indicate carbonate rock in the Eastern Alps. See text for details

Brenta in the south and Seegrube in the north added relevant information about the ecological niche of M. pallida, as these locations scored far from other ones (Fig. 3). Niche identity and background tests The overlap between the model predictions was quite high (Model 1 vs. Model 2, D = 0.84; Model 2 vs. Model 3, D = 0.78; and Model 1 vs. Model 3, D = 0.65, Fig. 4a). The niche identity tests were highly significant (p \ 0.01) and indicated that model predictions differed from randomly created pseudoreplicates (Fig. 4a). The background tests rejected the null hypotheses that the models were more different (p \ 0.05) than expected based on their environmental background in all instances except Model 1 versus Model 3 (p [ 0.05) (Fig. 4b–d).

Table 4 Loadings of the first two axes of the principal component analysis based on environmentalvariable values for all presence localities Variables

PC1 (93.9 %)

PC2 (4.2 %)

Bio10

-0.1840

0.6739

Bio11

-0.1665

0.0296

Bio16

0.6439

0.6066

Bio17

0.7238

-0.3001

-0.00566

-0.0107

Surface Area

Bio10 temperature of warmest quarter, Bio11: temperature of coldest quarter, Bio16 precipitation of driest quarter, Bio17 precipitation of wettest quarter, Surf surface area

123

2856

Biodivers Conserv (2012) 21:2845–2863

Fig. 3 Representation of the range of environmental conditions suitable to M. pallida in the Eastern Alps by means of a principal component analysis applied to the variables used in Maxent modelling. Blue polygon shows all locations of M. pallida before the study (Model 1). Orange polygon includes all locations of Model 2. Green polygon shows all locations of Model 3 (no additional locations were detected in Round 3, see text for details). (Color figure online)

Molecular analysis The morphological identification of the specimens from the newly discovered populations at Brenta, Drei Zinnen, Seekofel and Seegrube was confirmed by the molecular analysis. In total, three haplotypes, HT1 to HT3, were identified: The specimens from Seegrube and Seekofel had exclusively HT1, specimens from Brenta exclusively HT2; the specimens from Drei Zinnen showed HT1 and HT3. The haplotypes differed among each other by one mutational event.

Discussion This study demonstrates that a very low sample number, little knowledge on the species’ ecology, and mountainous and thus highly heterogeneous environments are not an impediment for predicting the distribution of a species with confidence. We discovered four new populations of M. pallida in the Eastern Alps using the modelling software Maxent (Fig. 2). We did not test any other modelling approach but our results corroborate previous studies that show that Maxent performs well under conditions similar to ours (Elith et al. 2006; Costa et al. 2010; Hernandez et al. 2006; Kumar et al. 2009; Pearson et al. 2007).

123

Biodivers Conserv (2012) 21:2845–2863

2857

Fig. 4 Niche identity and background tests comparing the predicted distributions of M. pallida derived from the iterative approach (see Fig. 2) using ENMTools (Warren et al. 2008). a Results of the identity tests (bars) and the Schoener’s D values (arrows Dx,y = D for comparison of Model x vs. Model y). b–d Background tests of the three pairwise model comparisons. As there were no new findings in the third sampling round, Model 3 also represents the final model. ** Significant at p \ 0.01; * significant at p \ 0.05

In the light of our three initial questions, the results of our study provide the following new insights on M. pallida: (1)

Distribution range The discovery of four new populations, and the considerable extension of the known area of distribution show that M. pallida indeed has a significantly wider range than previously thought. This geographical dimension in gaining additional information is similar to that in the few previous Maxent studies using ground validation (e.g., Jackson and Robertson 2011; Rebelo and Jones 2010; Williams et al. 2009), lending, in general, credit to the underlying models. However, in 17 out of 21 field excursions no population of M. pallida was discovered. One reason may be that M. pallida simply escaped our attention here and there. We consider this not to be very likely, because we captured the first specimens within the first hour of search at all positive locations and because weather conditions were equally suitable during all field trips. Furthermore, the failure rate might partly be due to what appears now as a miscarried bias correction: the intentional exclusion of the modelling variable ‘‘lithology’’. Since sampling silicate-rock locations was always unsuccessful, we now think that focusing on carbonate-rock locations within the predicted distribution would have increased the rate of positive inspections. We also expect that lithology as additional modelling variable would lead to a higher fragmentation, i.e., a more precise prediction at the local scale, but not to a change in the overall range of distribution since M. pallida is limited to the Eastern Alps. Our decision to sample at least 10 km away from known locations might have reduced the success rate, but is an unavoidable prerequisite to detect new parts of the distribution range with previously undetermined environmental conditions. Eventually, the high failure rate

123

2858

(2)

(3)

Biodivers Conserv (2012) 21:2845–2863

may at least partly reveal a punctually scattered pattern of distribution. Recent findings indicate that the species has survived the last Pleistocene glaciation on few ice-free inneralpine spots, so-called nunataks (Wachter et al. in press). The limited dispersal ability due to winglessness and the orographic isolation of suitable high-mountain habitats may have impeded the postglacial colonisation of potential habitats (cf. Hirzel and Le Lay 2008; Rebelo and Jones 2010). Endemic state The unsuccessful field trips, especially those to the eastern and western borders of the predicted distribution range, support the hypothesis of M. pallida being endemic to the Eastern Alps. But can we definitely exclude occurrence in the Western Alps? Model 1 and Model 2 predicted areas in the Western Alps in Switzerland. In the first two rounds of field trips we set aside the possible Swiss area by reason of its small size, the wide separation, and time economy. Model 3 predicted no Western Alpine area anymore. Definite proof of confinement to the Eastern Alps is as difficult as any proof of absence (Jime´nez-Valverde et al. 2008; Lobo et al. 2010), but Eastern Alpine endemism of M. pallida is supported by the fact that Models 1 and 2 predicted no carbonate-rock area above 2,000 m a.s.l. in the Western Alps. The Eastern Alps are a hotspot of endemism (Rabitsch and Essl 2009; Scho¨nswetter et al. 2005) and a distinct biogeographical unit. Niklfeld (1998) reported that two-thirds of the about 400 Alpine-endemic plant species are only found in the Eastern Alps. Huemer (1998) identified 221 butterfly species endemic to the Alps of which the majority occurs either in the Western or the Eastern Alps. Also for other regions re-analysis based on SDM confirmed the confinement of species to previously established biogeographical regions at a fine spatial resolution (e.g., Glor and Warren 2011: Martens Line). SDM and knowledge about the ecological niche All newly discovered locations of M. pallida were in carbonate rock scree, suggesting that lithology is of major influence on the occurrence of M. pallida. But what about the environmental variables in our models? ENM analyses of the ecological niche (Warren et al. 2008) revealed an increase of knowledge through SDM-guided inspections in the field (Fig. 4). Godsoe (2010), Hawlitschek et al. (2011) and Peterson (2011) argued for a careful interpretation of the background test and a focusing on the biological rather than the statistical significance. Calculated overlap values are high for all three model comparisons, but the highly significant results of the niche identity test reject the null hypotheses of niche identity (Fig. 4a). Additionally, as more locations were added to the analysis, overlap between the models decreased. This indicates a change of prediction during the iterative process. The null hypothesis of niche similarity was rejected for two of the three model comparisons. This shows that Model 1 and Model 2 as well as Model 2 and Model 3 are more similar than expected based on the environmental background available (Fig. 4b, c). But there is no significance in niche similarity of one way for comparing Model 1 with Model 3 (Fig. 4d). This indicates an unexpected prediction of Model 3, if based on the background of Model 1. There has been an increase of knowledge on the niche of M. pallida in the course of the modelling rounds as also discernible in the PCA (Fig. 3). The species occupies drier locations (such as Brenta) than previously thought based on published records.

Climate change and the effect on species distribution in alpine areas Endemic species are likely to be affected by changes in the niche they prefer (Thuiller et al. 2005). Based on the results of the contributions of variables to the predicted

123

Biodivers Conserv (2012) 21:2845–2863

2859

distribution, temperature plays an important role in the distribution of M. pallida (Table 2). All populations of M. pallida are located at altitudes above 1,900 m a.s.l. Being animals with low vagility, jumping bristletails may not be able to shift their distribution latitudinally when impacted by climate change. Inhabiting the highest elevations in the area of its populations, the possibility for M. pallida to move upwards to cooler regions will be limited. For other species with similar characteristics, the possibility of mountain-top extinction has been discussed (Colwell et al. 2008; Sandel et al. 2011). Further study into the local availability of microhabitats (cf. Scherrer and Ko¨rner 2011), as well as into the species’ life cycle and potential evolutionary adaption to temperature changes will be needed, however. In any case, M. pallida would be well suitable for monitoring the effects of climate change on high-altitude biodiversity to gain further information for conservation aspects. Iterative approach of species distribution modelling and ground validation The need for species distribution models on an iterative basis was stated recently (Costa et al. 2010; Williams et al. 2009). Our results, along with those of Jackson and Robertson (2011), show that this approach can indeed be effective in discovering new populations of species and in improving model quality. The ground validation process in our study did not result in paramount changes in the range of the predicted distribution area, but after Model 1 every additional set of samples made the models produce a more fragmented and thus probably a more precise prospecting area (Fig. 2). The originally predicted area in the Western Alps vanished in Model 3, and only the improved models predicted a northern extension of the area, which led to the discovery of M. pallida at Seegrube. The iterative approach might be especially important when the ecological niche is poorly outlined (i.e., possibly more poorly outlined than in our case) by the initial set of records. Since guidelines for model-guided iterative field work do not exist, we can only speculate about the implications of methodological details. A range of decisions has to be made. We used, for example, the LPT for converting model predictions into binary maps. However, the LPT tends to decrease with an increasing number of sample locations used for modelling. When very few locations are available at the outset of a study (cf. Pearson et al. 2007), a threshold lower than the LPT would increase the prospective area and might, at least in early iteration rounds, enhance the chance to discover populations in environments which differ more distinctly from the known ones. If so, our knowledge about the geographic distribution and the ecological niche would escalate. On the other hand, a lower threshold may entail fruitless field work by displaying sites where the focal species is absent. It could be wise not to predetermine the number of iteration rounds. For instance, running a new model after every detection of a population would update the knowledge about the species’ distribution for the next field trip. Considerations of practicability might become limiting here, in that, for example, it might be more efficient to target multiple sites in one journey before proceeding with the species identification. Another open question pertains to the inclusion of general geographic areas into earlier or later rounds. It is hard to balance the potential discovery of populations that provide maximum increase of knowledge against the risk of unsuccessful field trips. In our study it was tempting to search at the Western Alps in Rounds 1 and 2, given the confinement of M. pallida records to the Eastern Alps at that time. Retrospectively, our decision to postpone sampling the Western Alps was fortunate, but in other instances a more risky decision might accelerate the success of the study.

123

2860

Biodivers Conserv (2012) 21:2845–2863

We cannot reliably assess to what extent our experiences with M. pallida apply to other species. Certainly the particularities of the organism must be considered as well as conditions of the environment. Our approach is promising for the considerable number of Alpine species of uncertain endemicity (cf. Huemer 2011). It could be applied without major changes to other Alpine endemics with limited capability of dispersion, such as many other machilids (Sturm and Machida 2001), most Megabunus harvestmen (Martens 1978), and several Mughiphantes spiders (Komposch 2009). A longer minimum distance between known populations and a newly prospected site appears advisable for vagile organisms in more homogenous landscapes. But generally we believe that after the appropriate methodological adaptations, ENM-guided field work is worth being tested in any organism with few known populations, unless factors that cannot be included in the model evidently limit the distribution of the species. This restriction pertains to species involved in strong biotic interactions such as specialised herbivory (e.g., Odegaard 2006; Samways and Lu 2007), mutualism (e.g., Mehdiabadi and Schultz 2010; Orivel and Leroy 2011), parasitism (e.g., Buschinger 2009), and mimicry (e.g., Jackson and Nelson 2012), as well as to species whose populations have been strongly diminished by human influence (e.g., Fritz and Unso¨ld 2011). Some researcher bias along the decision-making process appears inevitable. In such cases (cf. Schlick-Steiner et al. 2010) careful documentation of the rationale behind any decision is indispensable for traceability and later meta-analyses. Notwithstanding that the current methodological uncertainties entailed many intuitive decisions, the results of our study open a broad prospect for model-guided field work.

Conclusion Our results show that Maxent is a suitable tool to study the distribution of endemic species with a low number of records in a heterogeneous area such as the Alps. The iterative approach of modelling and ground validation increased the knowledge of the distribution and the ecological niche of the jumping bristletail M. pallida. Thus the method appears promising with regard to biogeography and conservation biology of species in the Alps and beyond. Acknowledgments We thank Thomas Beswick, Paul Donovan and Dominik Rinnhofer for help with field work, Clemens Folterbauer for assistance in the molecular lab, Edward Middleton and Gary Muir for linguistic revision, and an anonymous reviewer for valuable comments on the manuscript. We also thank Tyrolean, South Tyrolean and Swiss authorities for granting sampling permissions and the Federal Institute for Geosciences and Natural Resources in Hannover, Germany, for giving access to geological data. The University of Innsbruck covered LR’s travel costs; NR-P was supported by project Consolider Montes (CSD2008-00040) granted by the Spanish Ministry of Research, Science and Innovation (MICINN), TD by the association in aid of South Tyroleans at the University of Innsbruck, GAW by project 32.06/138020 granted by the Autonomous Province South Tyrol, and FMS by project P 23949-B17 granted by the Austrian Science Fund (FWF).

References Anderson RP, Lew D, Peterson AT (2003) Evaluating predictive models of species’ distributions: criteria for selecting optimal models. Ecol Model 162:211–232 Arau´jo MB, Pearson RG (2005) Equilibrium of species’ distributions with climate. Ecography 28:693–695

123

Biodivers Conserv (2012) 21:2845–2863

2861

Beaumont LJ, Hughes L, Poulsen M (2005) Predicting species distributions: use of climatic parameters in BIOCLIM and its impact on predictions of species’ current and future distributions. Ecol Model 186:251–270 Boyce MS, Vernier PR, Nielsen SE, Schmiegelow FKA (2002) Evaluating resource selection functions. Ecol Model 157:281–300 Buschinger A (2009) Social parasitism among ants: a review (Hymenoptera: Formicidae). Myrmecol News 12:219–235 Cassagne N, Gauquelin T, Bal-Serin M, Gers C (2006) Endemic Collembola, privileged bioindicators of forest management. Pedobiol 50:127–134 Christian E, Knoflach B (2009) Jumping bristletails (Archaeognatha) in Austria: current knowledge and gaps. In: Tajovsky´ K, Schlaghamersky´ J, Pizˇl V (eds) Contributions to soil zoology in Central Europe ˇ eske´ Budeˇjovice, pp 9–12 III. ISB BC AS CR, C Colwell R, Brehm G, Cardelu´s C, Gilman A, Longino JT (2008) Global warming, elevational range shifts, and lowland biotic attrition in the wet tropics. Science 322:258–261 Costa GC, Nogueira C, Machado RB, Colli GR (2010) Sampling bias and the use of ecological niche modeling in conservation planning: a field evaluation in a biodiversity hotspot. Biodivers Conserv 19:883–899 Cowling RM (2001) Endemism. In: Simon AL (ed) Encyclopedia of biodiversity. Academic Press, San Diego, pp 497–507 Deharveng L, Dalens H, Drugmand D, Simon-Benito JC, Da Gama MM, Sousa P, Gers C, Bedo´s A (2000) Endemism mapping and biodiversity conservation in Western Europe: an arthropod perspective. Belg J Entomol 2:59–75 Dejaco T, Arthofer W, Sheets DH, Moder K, Thaler-Knoflach B, Christian E, Mendes L, Schlick-Steiner BC, Steiner FM (2012) A toolbox for integrative species delimitation in Machilis jumping bristletails (Microcoryphia: Machilidae). Zool Anz. doi:10.1016/j.jcz.2011.12.005 Elith J, Graham CH (2009) Do they? How do they? WHY do they differ? On finding reasons for differing performances of species distribution models. Ecography 32:66–77 Elith J, Graham CH, Anderson RP et al (2006) Novel methods improve prediction of species’ distributions from occurrence data. Ecography 29:129–151 Fritz J, Unso¨ld M (2011) Artenschutz und Forschung fu¨r einen historischen Schweizer Vogel: der Waldrapp im Aufwind. Wildbiol Int 5(17):1–15 Garzon BM, Sanchez De Dios R, Sainz Ollero H (2008) Effects of climate change on the distribution of Iberian tree species. App Veg Sci 11:169–178 Giovanelli JGR, de Siqueira MF, Haddad CFB, Alexandrino J (2010) Modeling a spatially restricted distribution in the Neotropics: how the size of calibration area affects the performance of five presenceonly methods. Ecol Model 221:215–224 Glor RE, Warren D (2011) Testing ecological explanations for biogeographic boundaries. Evolution 65:673–683 Godsoe W (2010) Regional variation exaggerates ecological divergence in niche models. Syst Biol 59:298–306 Gormley AM, Forsyth DM, Griffioen P, Lindeman M, Ramsey DS, Scroggie MP, Woodford L (2011) Using presence-only and presence-absence data to estimate the current and potential distributions of established invasive species. J Appl Ecol 48:25–34 Hall TA (1999) BioEdit: a user-friendly biological sequence alignment editor and analysis program for Windows 95/98/NT. Nucleic Acids Symp Ser 41:95–98 Hammer Ø, Harper DAT, Ryan PD (2001) PAST: Paleontological Statistics Software Package for education and data analysis. Palaeontol Electron 4(1):9 Hawlitschek O, Porch N, Hendrich L, Balke M (2011) Ecological Niche Modelling and nDNA sequencing support a new, morphologically cryptic beetle species unveiled by DNA barcoding. PLoS One 6:e16662 Hernandez PA, Graham CH, Master LL, Albert DL (2006) The effect of sample size and species characteristics on performance of different species distribution modeling methods. Ecography 29:773–785 Hernandez PA, Franke I, Herzog SK, Pacheco V, Paniagua L, Quintana HL, Soto A, Swenson JJ, Tovar C, Valqui TH, Vargas J, Young BE (2008) Predicting species distributions in poorly-studied landscapes. Biodivers Conserv 17:1353–1366 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 Hirzel AH, Le Lay G (2008) Habitat suitability modelling and niche theory. J Appl Ecol 45:1372–1381 Holdhaus K (1954) Die Spuren der Eiszeit in der Tierwelt Europas. Universita¨tsverlag Wagner, Innsbruck Hortal J, Jime´nez-Valverde A, Go´mez JF, Lobo JM, Baselga A (2008) Historical bias in biodiversity inventories affects the observed environmental niche of the species. Oikos 117:847–858 ¨ berblick (Lepidoptera). Stapfia 55:229–256 Huemer P (1998) Endemische Schmetterlinge der Alpen—ein U

123

2862

Biodivers Conserv (2012) 21:2845–2863

Huemer P (2011) Pseudo-endemism and cryptic diversity in Lepidoptera—case studies from the Alps and the Abruzzi. eco.mont 3:11–18 Jackson RR, Nelson XJ (2012) Specialized exploitation of ants (Hymenoptera: Formicidae) by spiders (Araneae). Myrmecol News 17:33–49 Jackson CR, Robertson MP (2011) Predicting the potential distribution of an endangered cryptic subterranean mammal from few occurrence records. J Nat Conserv 19:87–94 Janetschek H (1949) Beitrag zur Kenntnis der Felsenspringer (Thysanura, Machilidae) Nordtirols. Vero¨ffentl Mus Ferdinandeum, Innsbruck 26/29:147–165 ¨ ber Borstenschwa¨nze Su¨dtirols, besonders des Schlerngebietes (Apterygota, ThyJanetschek H (1951) U sanura). Der Schlern 7–8:321–329 Jenness JS (2004) Calculating landscape surface area from digital elevation models. Wildl Soc Bull 32:829–839 Jime´nez-Valverde A, Gomez JF, Lobo JM, Baselga A, Hortal J (2008) Challenging species distribution models: the case of Maculinea nausithous in the Iberian Peninsula. Ann Zool Fenn 45:200–210 ¨ stKomposch C (2009) Araneae (Spinnen). In: Rabitsch W, Essl F (eds) Endemiten—Kostbarkeiten in O erreichs Pflanzen- und Tierwelt. Naturwissenschaftlicher Verein fu¨r Ka¨rnten und Umweltbundesamt GmBH, Klagenfurt and Vienna, pp 408–463 Kriticos D, Randall RP (2001) A comparison of systems to analyze potential weed distributions. In: Groves RH, Panetta FD, Virtue JG (eds) Weed risk assessment. CSIRO Publishing, Collingwood, pp 61–79 Kumar S, Stohlgren TJ (2009) Maxent modeling for predicting suitable habitat for threatened and endangered tree Canacomyrica monticola in New Caledonia. J Ecol Nat Environ 1:094–098 Larkin MA, Blackshields G, Brown NP, Chenna R, McGettigan PA, McWilliam H, Valentin F, Wallace IM, Wilm A, Lopez R, Thompson JD, Gibson TJ, Higgins DG (2007) Clustal W and Clustal X version 2.0. Bioinformatics 23:2947–2948 Lobo JM, Jime´nez-Valverde A, Hortal J (2010) The uncertain nature of absences and their importance in species distribution modelling. Ecography 33:103–114 Martens J (1978) Spinnentiere, Arachnida: Weberknechte, Opiliones. Die Tierwelt Deutschlands. G. Fischer Verlag, Jena McCormack JE, Zellmer AJ, Knowles LL (2010) Does niche divergence accompany allopatric divergence in Aphelocoma jays as predicted under ecological speciation? Insights from tests with niche models. Evolution 65:184–202 Mehdiabadi NJ, Schultz TR (2010) Natural history and phylogeny of the fungus-farming ants (Hymenoptera: Formicidae: Myrmicinae: Attini). Myrmecol News 13:37–55 Myers N, Mittermeier RA, Mittermeier CG, da Fonseca GA, Kent J (2000) Biodiversity hotspots for conservation priorities. Nature 403:853–858 Newbold T, Reader T, Zalat S, El-Gabbas A, Gilbert F (2009) Effect of characteristics of butterfly species on the accuracy of distribution models in an arid environment. Biodivers Conserv 18:3629–3641 Niklfeld H (1998) Mapping the flora of Austria and the eastern Alps. Rev Valdotaine Hist Nat 51:53–62 Odegaard F (2006) Host specificity, alpha- and beta-diversity of phytophagous beetles in two tropical forests in Panama. Biodivers Conserv 15:83–105 Orivel J, Leroy C (2011) The diversity and ecology of ant gardens (Hymenoptera: Formicidae; Spermatophyta: Angiospermae). Myrmecol News 14:73–85 Palissa A (1964) Apterygota, Urinsekten. Die Tierwelt Mitteleuropas. Quelle & Meyer, Leipzig Papes M, Gaubert P (2007) Modelling ecological niches from low numbers of occurrences: assessment of the conservation status of poorly known viverrids (Mammalia, Carnivora) across two continents. Divers Distrib 13:890–902 Pearson RG, Raxworthy CJ, Nakamura M, Peterson AT (2007) Predicting species distributions from small numbers of occurrence records: a test case using cryptic geckos in Madagascar. J Biogeogr 34:102–117 Peterson AT (2003) Predicting the geography of species’ invasions via ecological niche modeling. Q Rev Biol 78:419–433 Peterson AT (2011) Ecological niche conservatism: a time-structured review of evidence. J Biogeogr 38:817–827 Pfiffner OA (2010) Geologie der Alpen, 2nd edn. UTB/Haupt Verlag, Stuttgart Phillips SJ, Dudik M, Shapire RE (2004) A maximum entropy approach to species distribution modeling. In: Greiner R, Schuurmans D (eds) Proceedings of the 21st international conference on machine learning. ACM Press Banff, Canada, pp 655–662 Phillips SJ, Anderson RP, Schapire RE (2006) Maximum entropy modeling of species geographic distributions. Ecol Model 190:231–259 Phillips SJ, Dudik M, Elith J, Graham CH, Lehmann A, Leathwick J, Ferrier S (2009) Sample selection bias and presence-only distribution models: implications for background and pseudo-absence data. Ecol Appl 19:181–197

123

Biodivers Conserv (2012) 21:2845–2863

2863

¨ sterreichs Pflanzen- und Tierwelt. NaturwisRabitsch W, Essl F (2009) Endemiten—Kostbarkeiten in O senschaftlicher Verein fu¨r Ka¨rnten and Umweltbundesamt GmbH, Klagenfurt and Vienna Raxworthy CJ, Martı´nez-Meyer E, Horning N, Nussbaum RA, Schneider GE, Ortega-Huerta MA, Peterson AT (2003) Predicting distributions of known and unknown reptile species in Madagascar. Nature 426:837–841 Rebelo H, Jones G (2010) Ground validation of presence-only modelling with rare species: a case study on barbastelles Barbastella barbastellus (Chiroptera: Vespertilionidae). J Appl Ecol 47:410–420 ¨ ber Machiliden Nordtirols. Vero¨ffentl Mus Ferdinandeum Innsbruck 19:192–267 Riezler H (1939) U Rou´ra-Pascual N, Suarez A (2008) The utility of species distribution models to predict the spread of invasive ants (Hymenoptera: Formicidae) and to anticipate changes in their ranges in the face of global climate change. Myrmecol News 11:67–77 Samways MJ, Lu SS (2007) Key traits in a threatened butterfly and its common sibling: implications for conservation. Biodivers Conserv 16(14):4095–4107 Sa´nchez-Ferna´ndez D, Lobo JM, Abella´n P, Milla´n A (2011) How to identify future sampling areas when information is biased and scarce: an example using predictive models for species richness of Iberian water beetles. J Nat Conserv 19:54–59 Sandel B, Arge L, Dalsgaard B, Davies RG, Gaston KJ, Sutherland WJ, Svenning J-C (2011) The influence of late Quaternary climate-change velocity on species endemism. Science 334:660–664 Scherrer D, Ko¨rner C (2011) Topographically controlled thermal-habitat differentiation buffers alpine plant diversity against climate warming. J Biogeogr 38:406–416 Schlick-Steiner BC, Steiner FM, Seifert B, Stauffer C, Christian E, Crozier RH (2010) Integrative taxonomy: a multisource approach to exploring biodiversity. Annu Rev Entomol 55:421–438 Schmo¨lzer K (2001) Wo liegt die Grenze zwischen Ost- und Westalpen? Zur Frage der Verteilung biogeographischer Arealgrenzen im Alpenraum. Gredleriana 1:227–242 Schoener TW (1968) The Anolis lizards of Bimini: resource partitioning in a complex fauna. Ecology 49:704–726 Scho¨nswetter P, Stehlik I, Holderegger R, Tribsch A (2005) Molecular evidence for glacial refugia of mountain plants in the European Alps. Mol Ecol 14:3547–3555 StatSoft Inc (2010) STATISTICA (data analysis software system). StatSoft Inc., Tulsa Steiner FM, Schlick-Steiner BC, VanDerWal J, Reuther KD, Christian E, Stauffer C, Suarez AV, Williams SE, Crozier RH (2008) Combined modelling of distribution and niche in invasion biology: a case study of two invasive Tetramorium ant species. Divers Distrib 14:538–545 Stockwell DRB, Peterson AT (2002) Effects of sample size on accuracy of species distribution models. Ecol Model 148:1–13 Sturm H, Machida R (2001) Archaeognatha. de Gruyter, Berlin Svenning JC, Skov F (2004) Limited filling of the potential range in European tree species. Ecol Lett 7:565–573 Tamura K, Peterson D, Peterson N, Stecher G, Nei M, Kumar S (2011) MEGA5: Molecular Evolutionary Genetics Analysis using maximum likelihood, evolutionary distance, and maximum parsimony methods. Mol Biol Evol 28:2731–2739 Thuiller W, Lavorel S, Arau´jo MB (2005) Niche properties and geographical extent as predictors of species sensitivity to climate change. Global Ecol Biogeogr 14:347–357 VanDerWal J, Shoo LP, Graham C, Williams SE (2009) Selecting pseudo-absence data for presence-only distribution modeling: how far should you stray from what you know? Ecol Model 220:589–594 Wachter GA, Arthofer W, Dejaco T, Rinnhofer LJ, Steiner FM, Schlick-Steiner BC (in press) Pleistocene survival on central Alpine nunataks: genetic evidence from the jumping bristletail Machilis pallida. Mol Ecol Wanger TC, Wielgoss AC, Motzke I, Clough Y, Brook BW, Sodhi NS, Tscharntke T (2011) Endemic predators, invasive prey and native diversity. Proc R Soc B 278:690–694 Ward DF (2007) Modelling the potential geographic distribution of invasive ant species in New Zealand. Biol Invasions 9:723–735 Warren DL, Glor RE, Turelli M (2008) Environmental niche equivalency versus quantitative approaches to niche evolution. Evolution 62:2868–2883 Warren DL, Glor RE, Turelli M (2010) ENMTools: a toolbox for comparative studies of environmental niche models. Ecography 33:607–611 Williams JN, Seo C, Thorne J, Nelson JK, Erwin S, O’brien JM, Schwartz MW (2009) Using species distribution models to predict new occurrences for rare plants. Divers Distrib 15:565–576 Wisz MS, Hijmans RJ, Li J, Peterson AT, Graham CH, Guisan A (2008) Effects of sample size on the performance of species distribution models. Divers Distrib 14:763–773

123