Browse Subject Areas

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

Ecological Effects of the Invasive Giant Madagascar Day Gecko on Endemic Mauritian Geckos: Applications of Binomial-Mixture and Species Distribution Models

  • Steeves Buckland ,

    Affiliations School of Biological Sciences, University of Bristol, Bristol, United Kingdom, National Parks & Conservation Service, Reduit, Mauritius

  • Nik C. Cole,

    Affiliations School of Biological Sciences, University of Bristol, Bristol, United Kingdom, Durrell Wildlife Conservation Trust, Jersey, Channel Islands, Mauritian Wildlife Foundation, Vacoas, Mauritius

  • Jesús Aguirre-Gutiérrez,

    Affiliations Naturalis Biodiversity Center, Leiden, The Netherlands, Institute for Biodiversity and Ecosystems Dynamics (IBED), University of Amsterdam, Amsterdam, The Netherlands

  • Laura E. Gallagher,

    Affiliation National Parks & Conservation Service, Reduit, Mauritius

  • Sion M. Henshaw,

    Affiliation National Parks & Conservation Service, Reduit, Mauritius

  • Aurélien Besnard,

    Affiliation Laboratoire de Biogeographie et d'Ecologie des Vertébrés, Centre d'Ecologie Evolutive et Fonctionnelle, Montpellier, France

  • Rachel M. Tucker,

    Affiliations Mauritian Wildlife Foundation, Vacoas, Mauritius, Department of Animal and Plant Sciences, University of Sheffield, Sheffield, United Kingdom

  • Vishnu Bachraz,

    Affiliation National Parks & Conservation Service, Reduit, Mauritius

  • Kevin Ruhomaun,

    Affiliation National Parks & Conservation Service, Reduit, Mauritius

  • Stephen Harris

    Affiliation School of Biological Sciences, University of Bristol, Bristol, United Kingdom

Ecological Effects of the Invasive Giant Madagascar Day Gecko on Endemic Mauritian Geckos: Applications of Binomial-Mixture and Species Distribution Models

  • Steeves Buckland, 
  • Nik C. Cole, 
  • Jesús Aguirre-Gutiérrez, 
  • Laura E. Gallagher, 
  • Sion M. Henshaw, 
  • Aurélien Besnard, 
  • Rachel M. Tucker, 
  • Vishnu Bachraz, 
  • Kevin Ruhomaun, 
  • Stephen Harris


The invasion of the giant Madagascar day gecko Phelsuma grandis has increased the threats to the four endemic Mauritian day geckos (Phelsuma spp.) that have survived on mainland Mauritius. We had two main aims: (i) to predict the spatial distribution and overlap of P. grandis and the endemic geckos at a landscape level; and (ii) to investigate the effects of P. grandis on the abundance and risks of extinction of the endemic geckos at a local scale. An ensemble forecasting approach was used to predict the spatial distribution and overlap of P. grandis and the endemic geckos. We used hierarchical binomial mixture models and repeated visual estimate surveys to calculate the abundance of the endemic geckos in sites with and without P. grandis. The predicted range of each species varied from 85 km2 to 376 km2. Sixty percent of the predicted range of P. grandis overlapped with the combined predicted ranges of the four endemic geckos; 15% of the combined predicted ranges of the four endemic geckos overlapped with P. grandis. Levin's niche breadth varied from 0.140 to 0.652 between P. grandis and the four endemic geckos. The abundance of endemic geckos was 89% lower in sites with P. grandis compared to sites without P. grandis, and the endemic geckos had been extirpated at four of ten sites we surveyed with P. grandis. Species Distribution Modelling, together with the breadth metrics, predicted that P. grandis can partly share the equivalent niche with endemic species and survive in a range of environmental conditions. We provide strong evidence that smaller endemic geckos are unlikely to survive in sympatry with P. grandis. This is a cause of concern in both Mauritius and other countries with endemic species of Phelsuma.


Invasive alien species (IAS) are capable of establishing, dispersing and causing harm to indigenous species [1]. They can cause cascading effects in native ecosystems by disrupting trophic interactions and sharing ecological resources with native species [2][4]. Although IAS have led to the extinction of many endemic species [5], they are still being spread around the world. Oceanic islands are at greater risk [6], [7] as island endemics have evolved in the absence of predators and do not have anti-predator defence mechanisms [8]. So successful invaders encounter less competition from island endemics and, being genetically more adaptable, tend to expand their niches [9].

Reptile extinctions in Mauritius have mostly been due to the introduction of mammalian invaders, with 69% of its endemic reptile diversity lost since human colonisation at the end of the 16th century [10], [11]. Only one terrestrial skink (Gongylomorphus bojerii fontenayi, Macchabé skink) and four arboreal day geckos (Phelsuma cepediana, blue-tailed day gecko; Phelsuma guimbeaui, lowland forest day gecko; Phelsuma ornata, ornate day gecko; Phelsuma rosagularis, upland forest day gecko) survive on mainland Mauritius. While P. cepediana consists of three un-described species [11], they are treated as a single species for this analysis because they cannot be distinguished phenotypically with confidence. A further seven endemic species of reptile survive on seven offshore islets, one of which is Phelsuma guentheri (Günther's gecko), a large species of day gecko that co-existed with most, if not all, of the smaller species of Phelsuma on mainland Mauritius before the 1800s [12]. However, the recent introduction of Phelsuma grandis (giant Madagascar day gecko) is believed to threaten the four surviving smaller species of Phelsuma on mainland Mauritius. P. grandis was originally introduced in Baie du Tombeau (Figure 1) in the early 1990s through the pet trade and has since been deliberately moved elsewhere. This species can attain a length of 24–30 cm [13], nearly double the length of the four endemic geckos [14].

Figure 1. Locations of the ten sites with Phelsuma grandis (red) and 11 sites without Phelsuma grandis (green) used to estimate the abundance of endemic species of Phelsuma.

The black dot indicates Baie du Tombeau, the location where Phelsuma grandis was first introduced.

Although the export of P. grandis is controlled through CITES (Appendix II), it has also been introduced to Afghanistan [15], Florida [16], Hawaii [17] and Réunion [18]. Whether P. grandis has significant impacts in these localities is currently unknown, but it has been observed predating the endemic Phelsuma borbonica (gecko vert des hauts) and introduced Gehyra mutilata (stump-toed gecko) in Réunion [18]. In Mauritius, P. grandis has been observed to predate on geckos such as P. ornata and the introduced species G. mutilata, Hemidactylus frenatus (common house gecko) and Hemidactylus parvimaculatus (Sri Lankan house gecko).

Apart from the anecdotal sightings of predation and localised presence of P. grandis in Mauritius, our current knowledge of its ecological impact is limited. Therefore, we first investigated the potential distribution of the endemic species of Phelsuma in mainland Mauritius and P. grandis to determine whether they could share the same spatial distribution. We used Species Distribution Modelling (SDM) to predict the distribution of the endemic geckos and P. grandis, adjusted to their physical environment. SDM is widely used in studies of biogeography, global change ecology, conservation biology [19][22] and invasion biology [23][25], and is a valuable management tool [26]. We hypothesised that P. grandis is a “generalist” and so is able to occupy a broad climatic and habitat range which would overlap with the four endemic geckos. Second, we surveyed sites with and without P. grandis to determine their impact upon the number of endemic geckos. Abundance estimation in many ecological studies is based on counting the number of individuals seen without accounting for probability of detection of the species being studied [27]. Lack of detection can be affected by (i) site-level covariates such as behavioural changes in the presence of a competitor/predator and (ii) observation-level covariates such as climatic conditions, which might influence detection [27][29]. Hence, we used hierarchical binomial mixture modelling (BMM) [30] that implements a metapopulation approach to adjust abundance estimation to the probability of detection [31]. This enabled us to calculate the true abundance (hereafter, abundance) of geckos at each site. BMM has mainly been used in bird studies [32][33], but has also been used to estimate the abundance of amphibians [34][35] and mammals [36]. We hypothesised that P. grandis would have a negative impact on the abundance of the four endemic geckos. We then use this information to discuss the management of P. grandis in Mauritius and elsewhere.


Ethical statement

This study was approved by the University of Bristol's Ethical Review Committee (University Investigation Number UB/11/031) and the National Parks and Conservation Service, Ministry of the Agro-Industry, Mauritius.

Predicting the distribution and overlap of endemic geckos and P. grandis

Collecting presence data.

Articles were published in national newspapers requesting that members of the public report P. grandis sightings. We received more than 100 telephone calls giving locations of P. grandis across the island. Site visits were conducted to confirm their presence and geographical locations were recorded. Presence data on endemic species of Phelsuma were obtained through extensive field surveys in private and public forests between January and March 2010. Surveys were conducted along existing tracks between 07:00 and 19:00; any Phelsuma species seen were recorded. Opportunistic observations in urban areas were also collected between January 2010 and August 2012.

Environmental layers.

We obtained 19 bioclimatic variables from WorldClim ( They represented the annual trends in seasonality in Mauritius over the period 1950 to 2000 [37]. Elevation data were acquired from a Digital Elevation Model obtained from EOSDIS ( We derived topographic information such as aspect and slope from the Digital Elevation Model. We also obtained Normalised Difference Vegetation Index (NDVI) data from a 2008 Landsat ETM+ dataset (

All the variables were tested for multicollinearity using the pairwise Pearson's correlation test and only the most biologically meaningful ones with correlation values <0.7 were kept for further analyses. The nine retained variables were: BIO3 (isothermality), BIO4 (temperature seasonality), BIO7 (temperature annual range), BIO13 (precipitation of wettest period), BIO19 (precipitation of coldest quarter), aspect, elevation, NDVI and slope. As the bioclimatic variables had a resolution of 1 km2 and the other four layers had a spatial resolution of 0.0009 km2, we rescaled the five bioclimatic variables to 0.0009 km2 to take the advantage of the high resolution of the other layers to produce a map with 2,041,627 grid cells.

Model fitting and evaluation.

Prior to analysis, we removed duplicate sightings of each species at the same or different sites within a grid cell. From the original 950 records, 777 presence data remained (P. cepediana 265; P. grandis 74; P. guimbeaui 174; P. ornata 220; P. rosagularis 44). We selected five algorithms to predict distribution and construct our ensemble models [38]: (i) Maximum Entropy Modelling (MaxEnt) [39]; (ii) Generalised Boosted Model (GBM) [40]; (iii) Random Forest (RF) [41]; (iv) Generalised Linear Models (GLMs) [42]; and (v) Generalised Additive Model (GAM) [43]; for model specifications for these algorithms see [44]. We implemented this particular approach because it is more accurate than using a single algorithm [38].

We randomly divided the original dataset, using 80% to construct the models and 20% to validate their accuracy. We carried out 10 repetition runs to obtain robust estimates of the species distributions, and model accuracy [45] was tested by means of the Area Under the Curve (AUC) of a Receiver Operating Characteristic plot (ROC) [46]. The AUC is a threshold independent measure of accuracy to evaluate the performance of SDMs [38], [47], [48]. Absence data were needed to evaluate model performance for all the algorithms except MaxEnt. Since only presence data were collected, we randomly generated background pseudo-absences i.e. locations with no sightings of a particular species were selected at random and assigned as absent. The number of pseudo-absences per species was ten times the number of sites with presence data [49].

We considered an AUC<0.7 as a poor model and an AUC>0.9 as a highly accurate model [46]. We generated 50 models per species (i.e. 250 models in total) and models with an AUC>0.7 were selected to construct an ensemble model [44]. The ensemble model corresponded to the median probability of occurrence across the selected models for each grid cell. The median value was chosen because it was less sensitive to outliers than the mean. We converted the continuous predictions (Figure S1) into presence-absence prediction maps (Figures 2 and 3) to carry out further analyses. For this conversion, we applied an optimal probability threshold that maximises the sensitivity and specificity of the created models [50], [51]. The presence/absence map was used to project the potential distribution and overlap between each endemic species and P. grandis. We also calculated the predicted distribution and overlap between the combined range of the four species of endemic geckos and P. grandis. Since we considered P. guimbeaui and P. rosagularis to be the species most at risk, with only 30 and two known subpopulations respectively, we calculated the distance between the predicted range of P. grandis and known subpopulation of P. guimbeaui and P. rosagularis.

Figure 2. The binary projection and overlap between Phelsuma grandis and four endemic species of Phelsuma.

Each map shows the overlap of P. grandis and one of the endemic species. On each map, grey indicates that no species of Phelsuma were predicted to be present, yellow shows the predicted range of Phelsuma grandis, blue the predicted range of that species of endemic Phelsuma, and red areas of predicted overlap between the two species. The first number in each heading is the % overlap of the predicted range of Phelsuma grandis with that species of endemic gecko and the second number is the % overlap of the endemic species' predicted range with that of Phelsuma grandis. The different maps suggest that Phelsuma cepediana (a) and Phelsuma ornata (c) will overlap more with Phelsuma grandis and thus could be at greater risk.

Figure 3. The binary projection and overlap between Phelsuma grandis and the combined predicted ranges of the four endemic species of Phelsuma on mainland Mauritius.

Grey indicates that no species of Phelsuma were predicted to be present, yellow shows the predicted range of Phelsuma grandis, blue the combined predicted range of all the species of endemic Phelsuma, and red areas of predicted overlap. The first number in the heading is the % overlap of the predicted range of Phelsuma grandis with the combined predicted ranges of the four species of endemic Phelsuma, and the second number is the overlap of the endemic species' combined predicted ranges with the predicted range of Phelsuma grandis.

Niche breadth.

We examined the environmental breadth indices to determine whether P. grandis had a narrow (specialist = 0) or wide (generalist = 1) niche. We used the binary prediction of presence and absence maps (Figures 2 and 3), modelled from the combined influence of all the environmental predictors and occurrence of each species, to estimate niche breadths. We used Levins' niche breadth metrics [52], where a value of 0 was equivalent to only one grid cell being suitable and a value of 1 was equivalent to all grid cells being suitable [53].

Data handling.

Data processing, model fitting and projection were conducted in R (R 3.0.1 Development Core Team 2013) BIOMOD2 package [44], while breadth metrics were estimated in ENMTools ( SDM predicted range and overlap estimation and GIS manipulations were carried out in ARCGIS 10.1 (ESRI, Redlands, California, USA) and Quantum GIS 1.8 (

Impact of P. grandis on the abundance of endemic geckos

Field surveys.

We conducted visual estimate surveys (VES) at ten sites where P. grandis were present and 11 sites where they were believed absent (Figure 1). We used two types of categorical habitat data: (i) building or non-building and (ii) vegetation type i.e. palm or non-palm. There were four building sites (two with, two without P. grandis), four non-palm sites (two with, two without P. grandis) and 13 palm sites (six with, seven without P. grandis). With the exception of one palm site, each P. grandis site was matched with a site without P. grandis that had comparable habitat characteristics in terms of area, number of trees and tree diameter at breast height (DBH). Building sites were human dwellings in residential areas along the south-east coast of Mauritius, with an average area of 176 m2 and 189 m2 for sites with and without P. grandis. Palm sites contained trees from the Arecaceae family. Each consisted of an isolated clump of four to 12 palm trees with a height less than 12 m and occupying an area from 50 m2 to 100 m2. Average basal tree coverage (based on DBH) was 0.26 m2 and 0.21 m2 in sites with and without P. grandis. Non-palm sites were isolated and covered an area of ∼400 m2 with a maximum of four trees up to 20 m high, mainly Terminalia arguna (terminalia) and Ficus benghalensis (banyan tree). The average basal tree coverage was 10.12 m2 and 10.00 m2 in sites with and without P. grandis.

To calculate the abundance of endemic geckos, one person slowly walked round each site, scanning every tree and wall surface, counting all the geckos seen and identifying them to species. We used features such as bite marks, size and general colour patterns to avoid repeat counts of the same gecko. This was repeated for 15 minutes each hour during peak activity periods from 08:00 to 11:00 and 14:00 to 18:00, i.e., six times in a day. All 21 sites were surveyed for one day between 16 and 23 May 2011. We assumed that the different time slots and minimum distance between sites (>100 m) would ensure that the sites would be temporally and spatially independent. Temperature was recorded at the start of each count using a Lutron Lm-8000 environmental meter (Lutron Electronic Enterprise Co. Ltd., Taipei, Taiwan). Cloud cover was estimated visually to the nearest 5%. Temperature and cloud cover were included as observation-level covariates to test their respective effects on detection probabilities. Site-level covariates, such as status (presence or absence) of P. grandis and habitat types were modelled to investigate their effects on abundance.

Data analyses.

The emergence of hierarchical models has decreased the dependence on labour-intensive mark recapture in the estimation of population parameters such as abundance and occupancy [54]. Simulation studies have demonstrated that BMM was robust in the estimation of abundance [55], but dependent upon meeting key assumptions i.e. that the population is closed during the survey and that the sites are spatially and temporally independent [56].

We developed a BMM using two probability distributions:-where Ni = the abundance of geckos in site i; dist = either Poisson, negative binomial or zero-inflated Poisson distribution; λ = mean gecko abundance; Yij = number of geckos recorded at site i during survey j; pij = probability of detecting any gecko at site i during survey j; i = site 1,2,…21; and j = the number of the survey 1,2…6.

An integrated likelihood method as implemented in the unmarked R package [57] was used to calculate the above variables. We found the most parsimonious model by an information-theoretic approach using the corrected Akaike Information Criterion (AICc) in the R package AICcmodavg [58]. The AICc was used because the sample size (n) divided by the number of parameters (K) was less than 40 [59]. Model adequacy of the global model was tested with a parametric bootstrapping chi-squared goodness of fit. Depending on the model adequacy, a Poisson, negative binomial or zero-inflated Poisson probability distribution was used to determine abundance. Only the top model was used to predict the average detection probability and abundance under different covariate conditions.


Species distribution modelling and niche breadths

Only 9 of the 250 SDMs i.e. 7 GLMs, 1 MaxEnt and 1 GAM had an AUC<0.7 and were therefore excluded from the ensemble model. Four ensemble models were considered as excellent with AUC>0.9 and one as a good model with AUC>0.8 (Table 1). P. cepediana had the largest predicted suitable range of 376 km2 (Table 1, Figure 2a). P. grandis had a predicted suitable range of 161 km2 in the centre and along the west to the north-east coast of Mauritius (Table 1, Figure 3). P. guimbeaui had a predicted suitable range of 186 km2 on the west side of the island (Table 1, Figure 2b). P. ornata had a predicted suitable range of 342 km2 along the coast, particularly from the south-west to north-west (Table 1, Figure 2c). P. rosagularis had the smallest predicted suitable range, only 85 km2, mainly in south-west Mauritius (Table 1, Figure 2d).

Table 1. Ensemble model evaluation results, showing the Area Under the Curve (AUC) of the median ensemble model for the five species of Phelsuma.

Between 4% and 37% of the predicted suitable range of P. grandis overlapped with each endemic species, whereas 7% to 18% of the predicted suitable range of each endemic species overlapped with P. grandis (Figure 2). For P. grandis, 60% of its predicted range overlapped with the combined area of suitability for all four endemic species of Phelsuma, whereas 15% of the predicted ranges of all the four endemic species overlapped with P. grandis (Figure 3). Of the 32 known subpopulations of P. guimbeaui and P. rosagularis, 13 were within the predicted range of P. grandis, four within 50 m, five within 200 m, five within 500 m, three within 1000 m and two within 1500 m.

Niche breadth for the five species of Phelsuma varied from 0.140 to 0.652. P. guimbeaui and P. rosagularis had a narrow specialist niche breadth metric, whereas P. cepediana, P. grandis and P. ornata had a moderate to broad generalist niche breadth metric (Table 2).

Table 2. Levins' niche breadth metric, where restricted to specific environmental conditions (specialist) = 0 and able to exploit a wide range of environmental conditions (generalist) = 1.

Effect of P. grandis on the abundance of endemic geckos

We recorded three endemic species, P. cepediana, P. guimbeaui and P. ornata, during visual surveys of the 21 sites. Since endemic geckos were sympatric, and of similar sizes and highly detectable colours, we combined their counts and compared abundance adjusted to imperfect detection probability between sites with and without P. grandis. On 126 surveys (21 sites each surveyed 6 times), we recorded 32 sightings of endemic geckos and 346 of P. grandis in the invaded sites, and 441 sightings of endemic geckos and no P. grandis in the control sites. We only detected endemic geckos in six of the ten sites with P. grandis.

We used three different statistical distributions to test for over-dispersion. The Poisson distribution showed signs of over-dispersion in the global model and was not used in model fitting. The negative binomial and zero-inflated Poisson distributions passed the goodness of fit test and adequately fitted the abundance model; the zero-inflated Poisson distribution for abundance was selected because it had smaller confidence intervals.

The lowest AICc model (AICc weight of 76%) was chosen as the best fitting model (Table 3). The selected model showed that the abundance of endemic geckos was affected by the presence of P. grandis; the abundance of endemic geckos was 13.3 (95% confidence interval 7.0–19.9) and 1.6 (95% confidence interval 0.0–3.2) in sites without and with P. grandis respectively (Figure 4). This represented a decline of 89% of endemic geckos in P. grandis sites.

Figure 4. The mean endemic gecko abundance (with 95% confidence intervals) in sites with and without Phelsuma grandis.

Sites without Phelsuma grandis had a high abundance of endemic geckos, those with Phelsuma grandis a low abundance of endemic geckos. N = 10 sites with Phelsuma grandis, N = 11 sites without Phelsuma grandis.

Table 3. Model selection results. Abundance was modelled with habitat and status as site-level covariates.

Probability of detection was affected by habitat and cloud cover (Table 3). Probability of detection was similar in the three habitats, with the highest detection in building sites (0.530, 95% confidence interval 0.413–0.644), followed by palm (0.465, 95% confidence interval 0.375–0.557) and non-palm sites (0.374, 95% confidence interval 0.284–0.437) (Figure 5a). This is not surprising: building sites had the least vertical diversity, making it relatively easy to spot geckos. The higher detectability between palm versus non-palm sites can be explained by the simple structure of palms with heights less than 12 m, compared to the complex structure of branching non-palm trees with heights up to 20 m. An increase in cloud cover had a negative effect on probability of detection in the three habitat types because the geckos do not bask in cloudy conditions, and so are less likely to be detected as they find refuge in optimal thermo-regulatory spots such as between leaves (Figure 5b).

Figure 5. Effects of habitat type and cloud cover in each habitat type on detection probability.

Figure 5a shows that the detection probability (with 95% confidence intervals) was similar in the three habitat types, with detection probability slightly higher in building sites than palm followed by non palm sites. Figure 5b shows a general decrease in detection probability (with 95% confidence intervals indicated by broken lines) with an increase in cloud cover in the three habitat types. Blue indicates building sites, black palm sites and red non-palm sites.


We used SDMs combined with BMM to investigate the ecological impacts of P. grandis on the endemic Phelsuma community in mainland Mauritius. The large AUC values for the five ensemble models suggest that the predictive maps had high statistical robustness. Though SDMs can give an accurate prediction of the spatial distributions of IAS [24], interpretation of the results need to be treated with caution. Model projections assumed complete random dispersal. However, as gecko movements are restricted by habitat isolation and fragmentation, the predictive maps of all five species could be overestimates [60]. Certainly, for both P. guimbeaui and P. rosagularis, there are very few known subpopulations within their predicted range. Some studies advocate the use of variables such as species interactions [61], species traits [62] or “natural history” [63] to build SDMs. However, little information was available for these variables, especially for P. grandis. BMM model performance using the zero-inflated Poisson distribution showed that our models were robust and had a good fit to the data. We used BMM to account for imperfect detection in the estimation of the abundance of geckos [34], [36], [57], [64], since this can lead to erroneous conclusions [27], especially when invasive and native species co-occur [65]. Phelsuma spp. tend to partition their habitats along an axis to reduce competition [66], [67], and so one possible response would be for the endemic species to shift along an axis following invasion by P. grandis. However, we did not observe any shift in habitat selection by endemic geckos in the presence of P. grandis; there were significantly fewer endemic geckos in the presence of P. grandis, and endemic geckos were not even detected at four of the ten sites with P. grandis. Despite being a snapshot survey, we believe that the same results would have been observed throughout the year.

Characteristics of a generalist invader and the potential threat of invasion

The predictive overlap maps and a relatively large niche breadth suggest that P. grandis has the typical attributes of a generalist invader [68], with the ability to persist in a large range of environmental conditions. The predictive maps also suggest that the two commonest species, P. cepediana and P. ornata, are the most threatened by P. grandis. However, the distance between the predicted range of P. grandis and known subpopulations of P. guimbeaui and P. rosagularis were relatively small, suggesting that they could be under more immediate threat from P. grandis invasion. Both these endemic species had narrow Levins' niche breadths typical of specialists; they also occupy small restricted ranges such that the potential threat posed by P. grandis enhances their vulnerability to extinction [7].

With a relatively large niche breadth, we expected P. grandis to have a bigger predicted distribution. For example, the niche breadth of P. ornata was 0.356, with a predicted range of 342 km2, compared to a niche breadth of 0.497 and a predicted range of 161 km2 for P. grandis. One plausible explanation for this apparent disparity is that our presence data do not reflect the full extent of the niche suitability for P. grandis, which has not had time to invade all the available niches in the 20 years since its introduction. For instance, in its native range in Madagascar (approximately 1000 km to the west of Mauritius), P. grandis has been recorded at elevations up to 900 m [15]. This suggests that the whole altitudinal range of Mauritius, which has a maximum altitude of 828 m, is vulnerable to invasion. However, we had no records of P. grandis above 700 m in Mauritius. When an IAS first arrives, it usually utilises habitats similar to its native range [69] and subsequently expands and invades new niches [9], [70]. We suspect that eventually P. grandis will have a larger range in Mauritius than we have predicted from its early spread and pattern of habitat selection, and so there will be more extensive overlap with the ranges of the endemic species of Phelsuma.

The threat to endemic geckos

There was a dramatic decline or total absence of endemic species of Phelsuma in the presence of P. grandis, suggesting that extirpation follows the arrival of P. grandis. Observations by local residents near the study sites suggest that P. grandis took less than 12 years to cause the disappearance of the endemic species. Similarly, reports from the public in Baie du Tombeau indicate that, in the two decades since its release, P. grandis has colonised the entire 1.8 km2 suburban region and no P. cepediana, P. guimbeaui or P. ornata have been seen in this area since 2009.

How P. grandis is leading to the extirpation of endemic Phelsuma populations is unclear, although competitive exclusion and predation are believed to be key drivers of species extinction [71], [72]. Field observations suggest that P. grandis shares temporal and spatial niches with the endemic geckos and so there is potential for competition and predation. Being larger than the endemic geckos, P. grandis is likely to consume larger prey items and thus relax competition for food [73], although day geckos frequently feed on nectar, pollen [74] and tree saps, and so there may be competition with P. grandis for these food resources [75]. We often observed aggressive behaviour from P. grandis towards endemic geckos at inflorescences and tree-sap foraging sites, suggesting competitive displacement at important food resources. A decrease in egg production [76], [77] and/or reduced mating success [77] can occur in the presence of a predator or competitor, leading to a decline in reproductive output and increasing the risk of extinction. Understanding the underlying causes of how P. grandis leads to the decline and local loss of endemic Phelsuma populations will be key to quantifying the long-term impact and potentially managing systems to mitigate the threats.

P. grandis has also been introduced in nearby Réunion. Two endemic species of day gecko, P. borbonica and P. inexpectata (Manapany day gecko), are both threatened by the arrival of P. grandis [18]. P. grandis is widely available in the pet trade and there is the risk of further invasions in countries such as the Comores and Seychelles that have their own endemic species of Phelsuma.


With the ever-increasing numbers of invasive species, it is important to decide whether an IAS needs to be eradicated. Ideally, this decision needs to be made before the IAS is well-established. While the eradication of an IAS can help in the restoration of native ecosystems [78], eradication can also cause more damage [79], especially when an IAS has established key ecological functions [9]. However, since the majority of alien species will never be eradicated, alternative management strategies may be more appropriate in many circumstances [80].

P. grandis is a relatively recent introduction to Mauritius. Our data suggest that it is a generalist capable of invading a diversity of Mauritian habitats with dramatic impacts on the endemic Phelsuma community. Currently, P. grandis is localised in Mauritius, and mostly occurs in private gardens and plantations. However, this poses additional risks since it could be spread by accidental anthropogenic transportation. P. grandis is also being actively moved to new locations by locals to control other introduced geckos that are considered to be a messy and noisy nuisance in houses. It is now critical to identify the mechanisms whereby P. grandis leads to the loss of local populations of endemic Phelsuma to limit further impacts, and to decide if, and how, an eradication programme could be undertaken before P. grandis becomes better established in Mauritius. Our data also highlight the importance of banning the importation of P. grandis to other countries, particularly those with vulnerable populations of Phelsuma and other endemic geckos.

Supporting Information

Figure S1.

The continuous probability of occurrence of the five species of Phelsuma using the ensemble model with the highest probability of suitability indicated by red and the lowest by grey.



We thank Mannickchand Puttoo from the National Parks and Conservation Service, Ministry of the Agro-Industry, Mauritius for support and encouragement; Terry Burke and colleagues of the Molecular Ecology Laboratory of Sheffield University for their support; the Mauritian Wildlife Foundation for constant logistical support; J.C. (Koos) Biesmeijer and Leon Marshall from Naturalis Biodiversity Center, The Netherlands, for help with the SDM analyses; Ralph Budzinski and Mickael Sanchez for all their helpful information; and the National Parks and Conservation Service for permission to undertake this research.

Author Contributions

Conceived and designed the experiments: SB, NCC, SH, VB, LEG, SMH, KR, RMT. Analyzed the data: AB, BMM, SB, JAG, SDM. Wrote the manuscript: SB SH NCC. Data collection: SB NCC LEG SMH RMT.


  1. 1. Mack RN, Simberloff D, Lonsdale WM, Evans H, Clout M, et al. (2000) Biotic invasions: causes, epidemiology, global consequences, and control. Ecol Appl 10: 689–710
  2. 2. O'Dowd DJ, Green PT, Lake PS (2003) Invasional ‘meltdown’ on an oceanic island. Ecol Lett 6: 812–817
  3. 3. Charles H, Dukes JS (2007) Impacts of invasive species on ecosystem services. In: Nentwig DW, editor. Biological invasions. Berlin: Springer. pp. 217–237.
  4. 4. Kurle CM, Croll DA, Tershy BR (2008) Introduced rats indirectly change marine rocky intertidal communities from algae- to invertebrate-dominated. Proc Natl Acad Sci USA 105: 3800–3804
  5. 5. Clavero M, García-Berthou E (2005) Invasive species are a leading cause of animal extinctions. Trends Ecol Evol 20: 110
  6. 6. Case TJ, Bolger DT (1991) The role of introduced species in shaping the distribution and abundance of island reptiles. Evol Ecol 5: 272–290.
  7. 7. Clavero M, Brotons L, Pons P, Sol D (2009) Prominent role of invasive species in avian biodiversity loss. Biol Conserv 142: 2043–2049
  8. 8. Atkinson IAE (2001) Introduced mammals and models for restoration. Biol Conserv 99: 81–96
  9. 9. Pearson DE (2009) Biological invasions on oceanic islands: implications for island ecosystems and avifauna. In: Chae HY, Choi CY, Nam HY, editors. Seabirds in danger: invasive species and conservation of island ecosystem, pp. 3–16. Available: Accessed 16 June 2013.
  10. 10. Arnold EN (2000) Using fossils and phylogenies to understand evolution of reptile communities on islands. Bonn Zool Monogr 46: 309–324.
  11. 11. Austin JJ, Arnold EN, Jones CG (2004) Reconstructing an island radiation using ancient and recent DNA: the extinct and living day geckos (Phelsuma) of the Mascarene islands. Mol Phylogenet Evol 31: 109–122
  12. 12. Cole N. (2009) A field guide to the reptiles and amphibians of Mauritius. Vacoas, Mauritius: Mauritian Wildlife Foundation. 81 p.
  13. 13. Glaw F, Vences M (2007) Field guide to the amphibians and reptiles of Madagascar. Köln: Vences & Glaw. 436 p.
  14. 14. Vinson J, Vinson J-M (1969) The saurian fauna of the Mascarene Islands. Mauritius Inst Bull 4: 203–320.
  15. 15. Ratsoavina F, Glaw F, Rakotondrazafy NA (2011) Phelsuma grandis. IUCN Red List of Threatened Species. Version 2012.2. Available: Assessed 25 April 2013.
  16. 16. Krysko KL, Hooper AN, Sheehy CM (2003) The Madagascar giant day gecko, Phelsuma madagascariensis grandis Gray 1870 (Sauria: Gekkodinae): a new established species in Florida. Flo Scient 66: 222–225.
  17. 17. Goldberg SR, Bursey CR, Kraus F (2010) Helminth records for the Madagascan giant day gecko, Phelsuma grandis (Gekkonidae) from Hawaii. Occas Pap Bernice P Bishop Mus 108: 49–52.
  18. 18. Sanchez M, Gandar A (2010) Etat des lieux de la population introduite à Manapany-les-Bains du grand gecko vert malgache, Phelsuma grandis Gray 1870. Reunion: Association Nature Océan Indien unpublished report. 26 p.
  19. 19. Carvalho SB, Brito JC, Pressey RL, Crespo E, Possingham HP (2010) Simulating the effects of using different types of species distribution data in reserve selection. Biol Conserv 143: 426–438
  20. 20. Doko T, Fukui H, Kooiman A, Toxopeus AG, Ichinose T, et al. (2011) Identifying habitat patches and potential ecological corridors for remnant Asiatic black bear (Ursus thibetanus japonicus) populations in Japan. Ecol Model 222: 748–761
  21. 21. Bosso L, Rebelo H, Garonna AP, Russo D (2013) Modelling geographic distribution and detecting conservation gaps in Italy for the threatened beetle Rosalia alpina. J Nat Conserv 21: 72–80
  22. 22. Virkkala R, Heikkinen RK, Fronzek S, Leikola N (2013) Climate change, northern birds of conservation concern and matching the hotspots of habitat suitability with the reserve network. PLoS One 8(5): e63376
  23. 23. Beaumont LJ, Gallagher RV, Thuiller W, Downey PO, Leishman MR, et al. (2009) Different climatic envelopes among invasive populations may lead to underestimations of current and future biological invasions. Divers Distrib 15: 409–420
  24. 24. Ficetola GF, Thuiller W, Padoa-Schioppa E (2009) From introduction to the establishment of alien species: bioclimatic differences between presence and reproduction localities in the slider turtle. Divers Distrib 15: 108–116
  25. 25. Ficetola GF, Maiorano L, Falcucci A, Dendoncker N, Boitani L, et al. (2010) Knowing the past to predict the future: land-use change and the distribution of invasive bullfrogs. Global Change Biol 16: 528–537
  26. 26. Gallien L, Münkemüller T, Albert CH, Boulangeat I, Thuiller W (2010) Predicting potential distributions of invasive species: where to go from here? Divers Distrib 16: 331–342
  27. 27. Kéry M (2010) Introduction to WinBUGS for ecologists: a Bayesian approach to regression, ANOVA, mixed models and related analyses. Amsterdam: Elsevier. 302 p.
  28. 28. MacKenzie DI, Nichols JD, Lachman GB, Droege S, Royle JA, et al. (2002) Estimating site occupancy rates when detection probabilities are less than one. Ecology 83: 2248–2255
  29. 29. De Wan AA, Sullivan PJ, Lembo AJ, Smith CR, Maerz JC, et al. (2009) Using occupancy models of forest breeding birds to prioritize conservation planning. Biol Conserv 142: 982–991
  30. 30. Royle JA (2004) N-mixture models for estimating population size from spatially replicated counts. Biometrics 60: 108–115
  31. 31. Kéry M, Royle JA, Plattner M, Dorazio RM (2009) Species richness and occupancy estimation in communities subject to temporary emigration. Ecology 90: 1279–1290
  32. 32. Royle JA, Dawson DK, Bates S (2004) Modeling abundance effects in distance sampling. Ecology 85: 1591–1597
  33. 33. Kéry M (2008) Estimating abundance from bird counts: binomial mixture models uncover complex covariate relationships. Auk 125: 336–245
  34. 34. Dodd CK, Dorazio RM (2004) Using counts to simultaneously estimate abundance and detection probabilities in a salamander community. Herpetologica 60: 468–478
  35. 35. McKenny HC, Keeton WS, Donovan TM (2006) Effects of structural complexity enhancement on eastern red-backed salamander (Plethodon cinereus) populations in northern hardwood forests. Forest Ecol Manage 230: 186–196
  36. 36. Graves TA, Kendall KC, Royle JA, Stetz JB, Macleod AC (2011) Linking landscape characteristics to local grizzly bear abundance using multiple detection methods in a hierarchical model. Anim Conserv 14: 652–664
  37. 37. 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
  38. 38. Aguirre-Gutiérrez J, Carvalheiro LG, Polce C, van Loon EE, Raes N, et al. (2013) Fit-for-purpose: species distribution model performance depends on evaluation criteria - Dutch hoverflies as a case study. PloS One 8(5): e63708
  39. 39. Phillips SJ, Anderson RP, Schapire RE (2006) Maximum entropy modeling of species geographic distributions. Ecol Model 190: 231–259
  40. 40. Ridgeway G (1999) The state of boosting. Comp Sci Stat 31: 172–181.
  41. 41. Breiman L (2001) Random forests. Mach Learn 45: 5–32.
  42. 42. McCullagh P, Nelder JA (1989) Generalized linear models. London: Chapman and Hall. 511 p.
  43. 43. Hastie T, Tibshirani R (1999) Generalized additive models. London: Chapman and Hall. 335 p.
  44. 44. Thuiller W, Lafourcade B, Engler R, Araújo MB (2009) BIOMOD - a platform for ensemble forecasting of species distributions. Ecography 32: 369–373
  45. 45. Barbet-Massin M, Jiguet F, Albert CH, Thuiller W (2012) Selecting pseudo-absences for species distribution models: how, where and how many? Methods Ecol Evol 3: 327–338
  46. 46. Swets JA (1988) Measuring the accuracy of diagnostic systems. Science 240: 1285–1293
  47. 47. 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.
  48. 48. Thuiller W, Lavorel S, Araújo MB (2005) Niche properties and geographical extent as predictors of species sensitivity to climate change. Global Ecol Biogeogr 14: 347–357
  49. 49. Thuiller W, Georges D, Engler R (2013) Package “biomod2”. Species distribution modeling within an ensemble forecasting framework. Available: Accessed 25 August 2013.
  50. 50. Jiménez-Valverde A, Lobo JM (2007) Threshold criteria for conversion of probability of species presence to either-or presence-absence. Acta Oecol 31: 361–369
  51. 51. Liu C, White M, Newell G (2013) Selecting thresholds for the prediction of species occurrence with presence-only data. J Biogeogr 40: 778–789
  52. 52. Levins R (1968) Evolution in changing environments: some theoretical explorations. Princeton: Princeton University Press. 132 p.
  53. 53. Pike DA (2013) Climate influences the global distribution of sea turtle nesting. Global Ecol Biogeogr 22: 555–566
  54. 54. Royle JA, Dorazio RM (2006) Hierarchical models of animal abundance and occurrence. J Agric Biol Envir Stat 11: 249–263
  55. 55. Dorazio RM (2007) On the choice of statistical models for estimating occurrence and extinction from animal surveys. Ecology 88: 2773–2782
  56. 56. Mazerolle MJ, Bailey LL, Kendall WL, Royle JA, Converse SJ, et al. (2007) Making great leaps forward: accounting for detectability in herpetological field studies. J Herpetol 41: 672–689.
  57. 57. Fiske I, Chandler R, Miller D, Royle A, Kéry M (2010) Models for data from unmarked animals. R package version 0.8–6. Available: Accessed 29 June 2013.
  58. 58. Mazerolle MJ (2011) AICcmodavg-package. R Package Version 131. Available: Accessed 29 June 2013.
  59. 59. Burnham KP, Anderson DR (2002) Model selection and multimodel inference: a practical information-theoretic approach. New York: Springer. 488 p.
  60. 60. Sinclair SJ, White MD, Newell GR (2010) How useful are species distribution models for managing biodiversity under future climates? Ecol Soc 15(1): 8
  61. 61. Araújo MB, Luoto M (2007) The importance of biotic interactions for modelling species distributions under climate change. Global Ecol Biogeogr 16: 743–753
  62. 62. Evangelista PH, Kumar S, Stohlgren TJ, Jarnevich CS, Crall AW, et al. (2008) Modelling invasion for a habitat generalist and a specialist plant species. Divers Distrib 14: 808–817
  63. 63. Rödder D, Schmidtlein S, Veith M, Lötters S (2009) Alien invasive slider turtle in unpredicted habitat: a matter of niche shift or of predictors studied? PLoS One 4(11): e7843
  64. 64. Chandler RB, King DI (2011) Habitat quality and habitat selection of golden-winged warblers in Costa Rica: an application of hierarchical models for open populations. J Appl Ecol 48: 1038–1047
  65. 65. Cayuela H, Besnard A, Joly P (2013) Multi-event models reveal the absence of interaction between an invasive frog and a native endangered amphibian. Biol Invasions 15: 2001–2012
  66. 66. Harmon LJ, Harmon LL, Jones CG (2007) Competition and community structure in diurnal arboreal geckos (genus Phelsuma) in the Indian Ocean. Oikos 116: 1863–1878
  67. 67. Noble T, Bunbury N, Kaiser-Bunbury CN, Bell DJ (2011) Ecology and co-existence of two endemic day gecko (Phelsuma) species in Seychelles native palm forest. J Zool 283: 73–80
  68. 68. Williamson MH (1996) Biological invasions. London: Chapman and Hall. 244 p.
  69. 69. Bomford M, Kraus F, Barry SC, Lawrence E (2009) Predicting establishment success for alien reptiles and amphibians: a role for climate matching. Biol Invasions 11: 713–724
  70. 70. Fitzpatrick MC, Weltzin JF, Sanders NJ, Dunn RR (2007) The biogeography of prediction error: why does the introduced range of the fire ant over-predict its native range? Global Ecol Biogeogr 16: 24–33
  71. 71. Losos JB (2000) Ecological character displacement and the study of adaptation. Proc Natl Acad Sci USA 97: 5693–5695
  72. 72. Schoener TW, Spiller DA, Losos JB (2001) Predators increase the risk of catastrophic extinction of prey populations. Nature 412: 183–186
  73. 73. Pianka ER, Pianka HD (1976) Comparative ecology of twelve species of nocturnal lizards (Gekkonidae) in the Western Australian desert. Copeia 1976: 125–142
  74. 74. Hansen DM, Kiesbüy HC, Jones CG, Müller CB (2007) Positive indirect interactions between neighboring plant species via a lizard pollinator. Am Nat 169: 534–542
  75. 75. Krysko KL, Hooper AN (2006) Phelsuma madagascariensis grandis (Madagascar giant day gecko). Nectarivory; potential pollination. Herpetol Rev 37: 226.
  76. 76. Rummel JD, Roughgarden J (1985) Effects of reduced perch-height separation on competition between two Anolis lizards. Ecology 66: 430–444
  77. 77. Schoener TW, Spiller DA, Losos JB (2002) Predation on a common Anolis lizard: can the food-web effects of a devastating predator be reversed? Ecol Monogr 72: 383–407
  78. 78. McGeoch MA, Butchart SHM, Spear D, Marais E, Kleynhans EJ, et al. (2010) Global indicators of biological invasion: species numbers, biodiversity impact and policy responses. Divers Distrib 16: 95–108
  79. 79. Bergstrom DM, Lucieer A, Kiefer K, Wasley J, Belbin L, et al. (2009) Indirect effects of invasive species removal devastate World Heritage Island. J Appl Ecol 46: 73–81
  80. 80. Carroll SP (2011) Conciliation biology: the eco-evolutionary management of permanently invaded biotic systems. Evol Appl 4: 184–199