Modelling the distribution of Mustela nivalis and M. putorius in the Azores archipelago based on native and introduced ranges

Lucas Lamelas-López; Xosé Pardavila; Paulo A V Borges; Margarida Santos-Reis; Isabel R Amorim; Maria J Santos

doi:10.1371/journal.pone.0237216

. 2020 Aug 7;15(8):e0237216. doi: 10.1371/journal.pone.0237216

Modelling the distribution of Mustela nivalis and M. putorius in the Azores archipelago based on native and introduced ranges

Lucas Lamelas-López ^1,^*, Xosé Pardavila ², Paulo A V Borges ¹, Margarida Santos-Reis ³, Isabel R Amorim ⁴, Maria J Santos ^5,⁶

Editor: Ulrike Gertrud Munderloh⁷

PMCID: PMC7413552 PMID: 32764786

Abstract

The aims of this study were to predict the potential distribution of two introduced Mustelidae, Mustela nivalis and M. putorius in the Azores archipelago (Portugal), and evaluate the relative contribution of environmental factors from native and introduced ranges to predict species distribution ranges in oceanic islands. We developed two sets of Species Distribution Models using MaxEnt and distribution data from the native and introduced ranges of the species to project their potential distribution in the archipelago. We found differences in the predicted distributions for the models based on introduced and on native occurrences for both species, with different most important variables being selected. Climatic variables were most important for the introduced range models, while other groups of variables (i.e., human-disturbance) were included in the native-based models. Most of the islands of the Azorean archipelago were predicted to have suitable habitat for both species, even when not yet occupied. Our results showed that predicting the invaded range based on introduced range environmental conditions predicted a narrower range. These results highlight the difficulty to transfer models from native to introduced ranges across taxonomically related species, making it difficult to predict future invasions and range expansion.

Introduction

The deliberate or accidental introduction of non-native invasive species beyond their native range has been a consequence of human exploration and colonization [1]. The number of species introductions has largely increased in the last 200 years [2], and particularly in recent years due to global trade, transport and tourism [3]. Mammals were among the first organisms to be introduced by humans, either to provide food and transportation (e.g., livestock), company (e.g., pets), support for hunting activities [1, 4] and recently control of other invasive species [4]. The introduction and spread of non-native mammals, especially predators, has been a major cause of extinctions on oceanic islands worldwide and of significant changes in the composition and structure of ecological communities [5–9]. Invasive species especially threaten native biodiversity through predation, competition, disturbance, disease transmission and facilitation of other introduced species [6, 9, 10]. Most non-native mammal species were introduced on oceanic islands during European colonization [11] and colonial nesting seabirds are among the most negatively impacted native biota (e.g., [12]).

The Azores is an isolated oceanic archipelago in the North Atlantic where many mammal species have been introduced accidentally or deliberately since Portuguese colonization in the 15^th century [13–15]. The most widespread introduced mammal species in the Azores are rodents (house mouse—Mus musculus, rats—Rattus and R. norvegicus), rabbits (Oryctolagus cuniculus), cats (Felis catus), dogs (Canis familiaris), and livestock (cows, goats, sheep and pigs). These species occur in all (rodents and cats) or almost all (rabbits) the islands [15, 16], with known impacts on native birds [17–21]. It is known that some endemic terrestrial birds went extinct, probably associated to the arrival of humans and the introduction of non-native predators [22–24]. Two species of Mustelidae have also been introduced in the Azores during the islands’ colonization: the least weasel (Mustela nivalis) and the ferret (M. putorius) and both are classified as "invasive species" [25, 26]. M. putorius is more widespread, occurring in seven islands, and M. nivalis only occurs in two islands [14, 15], but both have impacted Azorean biodiversity, particularly through seabird’s predation [19, 27]. It is therefore important to understand the ecological requirements of these species on both native and introduced ranges, as information on their distribution patterns, habitat requirements and abundance might inform on their potential impact on native biodiversity.

Species distribution models (SDMs) estimate the relationship between species occurrence and the environmental variables of the occurrence sites, predicting species habitat suitability [28]. SDMs are widely used in biogeography, conservation biology and ecology [29] namely to predict the potential geographical distribution of invasive species (e.g., [30]). SDMs assume that a species range is in equilibrium with environmental conditions and that this equilibrium is achieved through maintaining (or conserving) ecological niche characteristics across space and time, i.e. niche conservatism [31]. Shea and Chesson [32] suggest that invasive species success is linked to how a species responds to the new environment and the degree of niche conservatism of an invasive species might determine which regions it can invade. If introduced species exhibit strong niche conservatism, they will only occupy regions with similar conditions to those of their native range restricting their invasion potential. If introduced species exhibit loose or lack of conservatism, then they will likely invade regions not similar to those in their native ranges and likely making them more successful invaders [30, 31].

In this study we tested the hypotheses that the more similar are the environmental conditions that predict native and invaded ranges the more restricted will be the invasion potential of a species, and that this process should be consistent in closely related species. We modeled the geographic distribution of M. nivalis and M. putorius in the Azorean introduced range based on presence data from and environmental variables that influence the distribution of both species in both native and introduced ranges. Our results allow us to gain a better understanding on how the invasion potential of closely related species is related to characteristics of the native range.

Material and methods

Study organisms

Mustela nivalis Linnaeus, 1766

The least weasel, M. nivalis is widely distributed in the Holarctic region. Its native range comprises much of Europe, northern Asia, northern Africa and northern North America [1, 33, 34]. The species has been introduced in many other areas including Australia, New Zealand, the Netherlands, and several islands in the Mediterranean Sea (Crete, Malta, Sicily, Sardinia, Corsica, Minorca and Mallorca, [34]) and in the Atlantic Ocean (São Miguel and Terceira in the Azores, São Tomé in the Gulf of Guinea, [1, 14, 25]). M. nivalis has been introduced by humans to control rodent and rabbit populations [14, 34], but its geographic origin and precise time of introduction, namely to the Azores, remain unknown [35]. M. nivalis uses a wide range of habitats, including mixed forests, farmlands and cultivated fields, grassy fields, meadows and hedgerows [1, 33, 34]. M. nivalis is a specialist predator of small mammals (especially rodents), but it is also able to alter its diet according to prey availability [33, 34]. M. nivalis may also consume small birds, bird's eggs, frogs, salamanders, fish, worms, beetles, carrion and lizards if food is scarce [1, 33, 34]. For example, in New Zealand mice account for a large portion of M. nivalis diet but native birds, invertebrates and reptiles are also consumed [36]. In the Azores, M. nivalis has been observed visiting Cory's shearwater (Calonectris diomedea borealis) nests during the reproduction period, suggesting egg and/or chick predation.

Mustela putorius Linnaeus, 1758

Ferrets are the domesticated form of the albino polecat. The two species interbreed with the western polecat (M. putorius), and hybrids are often indistinguishable in the wild, some authors do not consider them as two separate species (e.g., [37]) or consider ferrets a subspecies of M. putorius, M. putorius furo [25, 38, 39]. We will refer to this species throughout the text as M. putorius.

M. putorius is widespread in the western Palaearctic [40]. The native range comprises western Europe from the Mediterranean north to central Scandinavia and Finland, Great Britain (but absent from Ireland), and east to about central Kazakhstan, Russia, Romania, Hungary, Czechoslovakia, Yugoslavia, eastern China and Mongolia, South to the Himalayas [1, 40]. The introduced range of this species includes Australia, New Zealand, West Indies, Japan, several islands in Great Britain [1], Mediterranean islands (including Sicily and Sardinia, [26]), and Atlantic islands (Las Palmas in the Canary Islands, Flores, Faial, Pico, São Jorge, Terceira, São Miguel and Santa. Maria in the Azores, [15, 41]). M. putorius has been introduced by humans to control rabbit populations [1, 26, 40]. M. putorius occurs in a wide variety of habitats, namely lowland woods and riparian zones, forested and semi-forested areas near water sources, rural areas close to farms and villages, marshes and river valleys, agricultural land, forest edge and mosaic habitats [1, 26, 40]. M. putorius is a specialist predator feeding on small mammals (mainly rabbits), but its diet varies with food availability. M. putorius also preys on hares, possums, birds (occasionally domestic poultry), bird eggs, lizards, hedgehogs, frogs, carrion eels and invertebrates [1, 26, 40]. In its introduced range M. putorius threatens native wildlife as for example ground nesting and flightless birds in New Zealand [42] and the Scottish isles [43], and seabird populations in the Azores [27].

Study area

The study was conducted over two areas: the introduced range in the Azorean archipelago, and the native range in the European continent (Fig 1). Both species occupy Eurasian ranges, but as the geographic origin of the Azorean populations founders is still uncertain (but see [35]), we chose Europe assuming that it represents the environmental conditions that the species can use.

The two species were introduced in the Azores during the Portuguese colonization in the 15^th century [14]. At the time, the archipelago was covered by Laurissilva forests, but underwent severe modification post human settlement, mainly due to the replacement of native forest by crops and pastures for cattle, and the accidental or deliberate introduction of many plant species, mostly for agricultural and ornamental purposes [44, 45]. The new land-uses led to the extinction of numerous endemic species, particularly in the most disturbed islands [23, 46, 47]. Currently, the landscape is relatively similar in all islands of the archipelago, with urban and rural areas being concentrated near the coast, at the lowest elevations. At intermediate elevations, the dominant land cover types include crops, pasturelands, and exotic tree plantations of the Japanese cedar (Cryptomeria japonica) and the Australian cheesewood (Pittosporum undulatum). The native vegetation remnants are only found at the highest elevations and in the most inaccessible areas [48].

Species presence data

Species presence data were obtained from open published datasets, from standardized field sampling campaigns and from direct observations: (1) The species records in the native range were obtained from the Global Biodiversity Information Facility, an international network funded by the world`s governments that provides open access data ("GBIF.org"), "iNaturalist", a citizen science platform that generate biodiversity data from species occurrences ("iNaturalist.org"), "BioDiversity4All", a citizen science platform about Portuguese biodiversity occurrence data ("biodiversity4all.org"), and "Proyecto Turón", a Spanish project that relies on naturalists and researchers participation concerning the occurrence of M. putorius in the Iberian Peninsula ("proyectoturon.org"). We also collected data for Spain, including direct observations (non-systematic field observations by volunteer biologists) and systematic sampling campaigns with camera-traps conducted by the authors. We only used records from the last 10 years (2007–2017) and with an accuracy higher than 1 km. For open source occurrence data, records were filtered based on the accuracy information available, i.e., records with accuracy values equal or higher than 1000 meters, as well as records with no accuracy information were removed. However, presence records from these datasets may be affected by sampling bias, given that some places are more intensely sampled than other as detections are often spatially biased towards easily accessed areas and/or records differ across the countries (S1 File) [49, 50]. This spatial bias can lead biased results from the comparison of presence records with background data drawn at random from the entire region [50]. To account for these biases we followed the approach proposed by Phillips et al. [50], and chose background data exhibiting the same bias as the presence records (for more details see S1 File). We selected a total of 3,396 and 1,616 occurrence records of M. nivalis and M. putorius, respectively, for the native range (Europe).

(2) The introduced range records were obtained from recent direct observations (non-standardized field observations by volunteer biologists, 2007—present) and standardized camera-trapping campaigns across different habitat types conducted by the authors (2013–2015), in all islands where the species occur [15]. We obtained a total of 29 and 24 occurrence records of M. nivalis and M. putorius, respectively. Currently there are no other available records in the Azores that comply with our selection criteria, i.e., reliable, recent and with an accuracy higher than 1 km.

All records were scaled to 1km² cells for both study areas (i.e., one record per 1km² cell). All occurrence records used in the study (open source and field data records) were included in S2 File.

Environmental data

A set of candidate environmental variables was selected to test their ability to predict M. nivalis and M. putorius distributions. We selected (1) two topographic variables (e.g., slope) in order to describe the physical environment; (2) four climatic variables (e.g., annual temperature, precipitation seasonality) to describe the bioclimatic conditions; (3) a set of seven landscape level variables, including habitat type, vegetation structure (herbs, shrubs or trees), land use types, and landscape heterogeneity (e.g., number of patches); and (4) three human disturbance variables, related with human population (e.g., population density) and their activities (% of artificial areas, % of agricultural areas) (Table 1). The variables were selected according to species ecology. For example, the species occur from lowlands located at sea level to alpine areas upper 2,000 m.a.s.l. and in temperate and boreal areas (i.e., different climatic and topographic conditions). The species also inhabit in a wide variety of natural (deciduous and coniferous forests, scrublands) and human-associated habitats (agricultural areas, cultivated fields, near to villages) [34, 40]. Additionally, most of these variables have been previously used in other studies about carnivore distribution or habitat use (e.g., %urban_areas, in [51]; %forest_areas, %artificial_areas, %forest_coniferous, %forest_deciduous, in [52]; clim_bio1, clim_bio7, clim_bio12, clim_bio15, altimetry, slope and aspect, in [53]). Variables are described and their source provided in Table 1. For more details see S3 File. We conducted all spatial data processing using the ArcGIS software (ArcGIS 10 ArcMap v. 10.1). All variables were scaled to 1km² grids for both study areas.

Table 1. Description of the candidate variables to model M. nivalis and M. putorius distribution.

Variable	Description	Type	Original data source
*Topographic variables*
^*altimetry	Mean altimetry	Con	WorldClim 1.4
^*slope	Mean slope	Con	WorldClim 1.4
*Climatic variables*
clim_bio1	Annual mean temperature	Con	WorldClim 1.4
^*clim_bio7	Temperature annual range (difference between max. and min. temperatures)	Con	WorldClim 1.4
^*clim_bio12	Annual precipitation	Con	WorldClim 1.4
^*clim_bio15	Precipitation seasonality	Con	WorldClim 1.4
*Landscape variables*
^*landcover	Landcover map including 9 classes: (1) urban areas, (2) industrial areas, (3) agricultural areas, (4) livestock areas, (5) scrub and/or herbaceous vegetation associations, (6) forestry areas, (7) deciduous and mixed forests, (8) uncovered areas, and (9) wetlands and water bodies	Cat	CLC2006
^*%forest_areas	Percentage cover of coniferous, deciduous and mixed forests	Con	CLC2006
%forest_coniferous	Percentage cover of coniferous forests	Con	CLC2006
%forest_deciduous	Percentage cover of deciduous and mixed forests	Con	CLC2006
^*%scrub_herb	Percentage cover of scrub and/or herbaceous vegetation	Con	CLC2006
n_patches	Number of patches	Con	CLC2006
^*edge_density	Total edge length of patches	Con	CLC2006
*Human-disturbance variables*
^*%agricultural_areas	Percentage cover of agricultural areas (arable lands, permanent crops, pastures, heterogeneous agricultural areas)	Con	CLC2006
^*%artificial_areas	Percentage cover of artificial areas (urban, commercial and industrial areas)	Con	CLC2006
^*population_density	Density of population	Con	CIESIN 2015

Open in a new tab

* selected variables based on the correlation-test. Con—continuous variable, Cat—categorical variable.

To minimize model over-fitting because of potential correlation among predictor variables, we performed a pair-wise correlation analysis to identify and exclude highly correlated variables (r > 0.75), using the Spearman rank correlation coefficient. The excluded variables were clim_bio1, %deciduous, %coniferous and n_patches. For details about this analysis see S4 File. Correlation analyses were performed with R Studio software [54].

Species distribution modeling

Species distribution models were developed with MaxEnt v. 3.3. [55]. MaxEnt is a popular and widely used tool for species distribution modeling [56] with a predictive performance consistently among the highest performing methods [57, 58]. MaxEnt has been successfully used to produce models even from small data sets (similar to our sample size for introduced range; see Methods—Species presence data [59–61]), as those from rare or elusive species [59, 57]. These models have also been used to study invasive species, and despite the highlighted challenges inherent to modeling invaders [62], they have been shown to perform well in different phases of the invasion process (e.g., [63]). We have included a electronic supplementary material detailing the stepwise modeling process (S5 File).

Parameter configuration

MaxEnt requires a set of parameters to be specified by the user, namely test-training percentage (i.e., the percent of presence locations to be used for model development and for internal testing), number of background points, the form of the functional relationships (feature types in MaxEnt ‘language’), clamping (i.e., whether or not to constrain predictions within the range of variability of the input predictors), and regularization multiplier (i.e., to avoid over-fit of the response curves). However, there is no agreement in the literature on which set of parameter values to use in MaxEnt, and best practices suggest performing a preliminary sensitivity analysis on parameter performance for model selection (e.g., [56, 59]). Initially, we developed models to test the following parameter configurations and sets of variables: maximum number of iterations (500 and 1000), clamping (enabled, disabled), and regularization multiplier (0, 0.5, 1, 1.5, 2 and 2.5). We used 500 and 3,000 random background points for Azores and Europe models, respectively, 10 replicates, and selected the default feature type. All these parameter configurations were tested for both species (M. nivalis and M. putorius) and for both regions (Azores and Europe). We selected the best parameter configuration using the area under the curve of the receiver operating characteristic curve (hereafter AUC, [55, 56, 64]). The main advantage of this approach is that AUC provides a threshold-independent measure of model performance ([55, 56]). The AUC values vary from 0 to 1, where a value of 1 indicates perfect discrimination, a value of 0.50 indicates random predictive discrimination and values < 0.5 indicate performance worse than random [50]. However, although we used AUC to assess the parameter configuration performance and it is probably the most popular method to assess the accuracy of predictive distribution models [56] we have taken into account that exist a debate between scientists about its reliability (e.g., [65]). The parameter configuration used was: 10 runs, 500 and 3000 random background points for introduced and native ranges, respectively [66], 30% random test percentage, 1,000 maximum iterations, no clamping, auto features, and a regularization multiplier of 1.

Model selection

Model selection was performed by combining sets of candidate predictor variables as follows: topographic + climatic + landscape + human variables (n = 12), topographic + climatic + landscape variables (n = 9), topographic + climatic + human variables (n = 8), climatic + landscape + human variables (n = 10), topographic + climatic variables (n = 5), topographic + landscape variables (n = 6), climatic + landscape variables (n = 7), climatic + human variables (n = 6), landscape + human variables (n = 7), topographic variables only (n = 2), climatic variables only (n = 3), landscape variables only (n = 4) and human variables only (n = 3); For more details see S6 File. All these model combinations were created for both species (M. nivalis and M. putorius) and for both regions (Azores and Europe). Models were then selected based on their information content, as measured by the small-sample size corrected Akaike Information Criteria (AICc; [68]). We ranked the candidate models by their AICc, and computed the delta AICc, i.e., the difference in AICc from any given model to the model with the lowest AICc [68]. Models with ΔAICc ≤ 2 were selected. We calculated the Akaike weights to measure the weight of evidence for a given model to be the best model in each candidate model set [67, 68]. We calculated AICc using the ENMTools software (e.g., [28]).

Environmental variables contribution to the models

We obtained the relative contribution of the environmental variables to the geographic distribution of M. nivalis and M. putorius in their introduced and native areas from MaxEnt. Given that we obtained two equally performing top-models for M. putorius introduced range model, we averaged these models by calculating the weighted average of relative contributions of each variable based on their Akaike weight (e.g., [68]).

Habitat suitability for M. nivalis and M. putorius in the Azores

Species habitat suitability maps were generated by applying MaxEnt models to each cell of the Azorean archipelago map, obtaining a Habitat Suitability Index (HSI) that varied between 0.00 and 1.00 (e.g., [69]). We created a total of four suitability maps for the Azores, two maps per study species. One habitat suitability map was based on the top-models for the species introduced range and the other was based on the top-models for the species native range. Azorean distribution prediction models were projected from the Azorean islands where the species occur (introduced range) and from Europe (native range).

We used this HSI values to perform an additional evaluation of model outputs. We evaluated the outputs from the SDMs based on their fit to a subset of presence records obtained through camera-trapping surveys (see e.g., [70]). We obtained the HSI value of each cell for which we had field species records. Then, we calculated the proportion of occurrence records with HSI values higher than 0.75 (i.e., high habitat suitability), the proportion with HSI between 0.75 and 0.50 (high-medium habitat suitability), the proportion with HSI between 0.50 and 0.25 (medium-low habitat suitability), and the proportion with HSI lower than 0.25 (low habitat suitability). We performed a Chi² test with Yates correction to evaluated if the aforementioned values were significantly different than expected by chance (i.e. in relation to amount of cells available with HSI values higher than 0.75, between 0.5 and 0.75, between 0.25 and 0.5 and lower than 0.25), using R Studio software [54].

Additionally, due to the small sample size to model in the introduced area, to assess the reliability of the introduced models, we included maps of uncertainty in predictions, based on the introduced range by overlapping n-1 models (n = number of presence records). This allowed us to determine the consensus, i.e., how many times a given cell was predicted to be suitable (HSI > 0.5) for both study species (see S7 File).

Results

Model selection

M. nivalis and M. putorius top-models for both the native range in the Azores and the introduced range in Europe are shown in Table 2. For M. nivalis, the introduced range model included only climatic variables, and native range model included topographic, climatic, landscape and human variables (see Table 2). For M. putorius the introduced range model included climatic variables, and the native range model included topographic, climatic and human variables.

Table 2. Results of AIC-based model selection for the suitability of Mustela nivalis and M. putorius occurrence, in the introduced range (Azores) and in the native range (Europe).

	ΔAICc	w_i	К	−2 log (£)
*Mustela nivalis*
Introduced area
clim_bio7 + clim_bio12 + clim_bio15	0.00	0.65	4	353.80
Native area
altimetry + slope + clim_bio7 + clim_bio12 + clim_bio15 + landcover + %forest_areas + %scrub_herb + edge_density + %agricultural_areas + artificial_areas + population_density	0.00	0.99	80	62754.13
*Mustela putorius*
Introduced area
clim_bio7 + clim_bio12 + clim_bio15	0.00	0.54	5	359.20
Native area
altimetry + slope + clim_bio7 + clim_bio12 + clim_bio15 + %agricultural_areas + artificial_areas + population_density	0.00	0.99	68	29704.40

Open in a new tab

Top models are included (ΔAIC_c ≤ 2). ΔAIC_c AIC_c difference; w_i Akaike weight; К number of parameters; −2 log (£) −2 log-likelihood.

Predicted M. nivalis and M. putorius ranges in the Azores

We found marked differences in the potential distribution ranges for both species. The predicted distribution based on the introduced model for M. nivalis showed higher suitability in coastal areas of some islands (Terceira, Graciosa, Pico and São Jorge) and lower suitability inland. According to this model, the oriental islands of São Miguel and Santa Maria showed very low values of HSI. Contrarily, the native-based models showed higher HSI values inland (Fig 2). The predicted distribution for M. putorius showed the inverse pattern in the introduced-based models, with higher suitability towards the center of the islands and lower suitability in the coastal areas (Fig 3). Oriental islands also showed low suitability for M. putorius. In contrast, the native-based model for M. putorius showed lower HSI values inland.

Fig 2 — (a) Distribution map derived from SDM based on the introduced range; (b) distribution map derived from the SDM based on the native range. Black crosses indicate field data records. HSI, Habitat Suitability Index.

Fig 3 — (a) Distribution map derived from SDM based on the introduced range; (b) distribution map derived from the SDM based on the native range. Black crosses indicate field data records. HSI, Habitat Suitability Index.

Environmental correlates of island invaders

For M. nivalis, temperature annual range, annual precipitation and precipitation seasonality were included in introduced-based model and altimetry, slope, temperature annual range, annual precipitation, precipitation seasonality, land-use, cover of forest areas, cover of scrubs and herbaceous areas, edge density, cover of agricultural areas, cover of artificial areas and human population density were included in the native-based model (Table 2). Temperature annual range was the variable that showed a higher relative contribution to the introduced-based model (Table 3). For the native-based model the variables that showed higher relative contributions were altimetry, cover of forest areas and cover of artificial areas.

Table 3. Environmental variables contribution to the potential distribution of Mustela nivalis and M. putorius in introduced (Azores) and native (Europe) ranges.

		Mustela nivalis		Mustela putorius
		Introduced area	Native area	Introduced area	Native area
Topographic variables
	Altimetry	─	17.6	─	17.1
	Slope	─	10.6	─	5.6
Climatic variables
	clim_bio7	85.7	9.3	42.2	31.9
	clim_bio12	0.0	3.2	54.3	11.4
	clim_bio15	14.3	13.7	3.5	7.3
Landscape variables
	landcover	─	1.1	─	─
	%forest_areas	─	16.0	─	─
	%scrub_herbs	─	2.0	─	─
	edge_density	─	10.0	─	─
Human-disturbance variables
	%agricultural_areas	─	1.2	─	13.1
	%artificial_areas	─	14.9	─	0.2
	population_density	─	0.4	─	13.4

Open in a new tab

Values represent the weighted average of the relative contributions of each variable (%) based on top-models relative weights.

For M. putorius, the top-models derived from introduced and native range data also included different variables’ groups. The introduced-based model only included the climatic variables (temperature annual range, annual precipitation and precipitation seasonality) while in the native-based model the altimetry, slope, temperature annual range, annual precipitation, precipitation seasonality, cover of agricultural areas, cover of artificial areas and human population density variables were included in the model. Annual precipitation was the variable with higher relative contribution to the introduced-based model. Temperature annual range and altimetry were the variables with higher relative contribution to the native-based model (Table 3).

Validation of species models based on their fit to the field data, showed that the models performed well for both species, especially for native-based models (Table 4). More than 50% of the species records occurred in cells with HSI-values higher than 0.5 for all models (with exception of M. nivalis introduced-range model). The Chi² test revealed significant differences in the number of records for M. nivalis introduced-range model between observed and random values (Chi² = 44.096, df = 3, p-value<0.01). The other models showed no significant differences between observed and random values (M. nivalis native-range model: Chi² = 2.7308, df = 3, p-value = 0.435; M. putorius introduced-range model: Chi² = 5.4747, df = 3, p-value = 0.140; M. putorius native-range model: Chi² = 2.1504, df = 3, p-value = 0.542). Finally, the maps of uncertainty in prediction based on the introduced range showed similar patterns that the distribution models for the introduced range (see S7 File).

Table 4. Mustela nivalis and M. putorius species distribution model adjustment to the field data.

	Introduced-individuals records based SDM	Native-individuals records based SDM
*Mustela nivalis*	% records (n)	% records (n)
HSI > 0.75	14.28 (2)	57.14 (8)
0.5 < HSI < 0.75	28.57 (4)	14.28 (2)
0.25 < HSI < 0.5	35.71 (5)	7.43 (1)
HSI < 0.25	21.43 (3)	21.43 (3)
*Mustela putorius*
HSI > 0.75	9.09 (1)	54.55 (6)
0.5 < HSI < 0.75	63.64 (7)	27.27 (3)
0.25 < HSI < 0.5	27.27 (3)	18.18 (2)
HSI < 0.25	0 (0)	0 (0)

Open in a new tab

The percentage (%) and number of records (n) based on field data for four categories of Habitat Suitability Index (HSI): HSI > 0.75, 0.75 > HSI > 0.50, 0.50 > HSI > 0.25 and HSI < 0.25.

Discussion

We set out to predict for the first time the potential distribution of two introduced carnivores (M. nivalis and M. putorius) in the Azores, using species distribution models derived from native and introduced range records and environmental conditions. Wider distribution ranges based on native-based records, where species occupy most available habitats in all islands, compared to narrower ranges based on introduced-based records, suggest that both species are non-invasive in the Azores. Despite the fact that there is enough suitable habitat and that some invasive species on islands show a time lag before becoming invasive (e.g., [71]), given that both species were introduced to the Azores several centuries ago, they are not dominant and show limited distribution. The inclusion of different sets of variables in native and introduced ranges suggests that these species do not occupy habitats according to the species ecology in their native range, which in turn, according to our hypotheses, suggests that the archipelago provides potential for establishment and/or expansion, for both species.

Introduced and native range predictions

In general, native range models showed higher HSI values, and a more widespread distribution (Figs 2B and 3B) than the introduced range models (Figs 2A and 3A), for both species, which was consistent with our second hypothesis. M. nivalis introduced-range models showed higher suitability in coastal areas, while the native-range models showed also high suitability at intermediate elevations. M. putorius introduced range models showed higher suitability at intermediate elevations, while native-range models also included high suitability in coastal areas. These differences in the prediction maps for both species should be due to the climatic conditions at the introduced area, given that only climatic variables were included in the introduced-range models. Other authors also found differences in the climatic conditions invasive species withstand in native and introduced ranges (e.g., [72, 73]). This suggests that particular conditions of the introduced environment do not mimic those at the native range, i.e., variables that influence species distribution could be novel or differ between introduced and native ranges [73]. For island invasive species, island and mainland conditions may differ even more; for example, in the Azores altimetry and slope diversity influence the climate at very fine scales [74], and thus species distribution may respond to sudden changes in topographic complexity [48, 75, 76].

Moreover, the differences in the prediction maps between both ranges could also be due to ecological processes not included in the models as, for example, biotic interactions [31, 73, 77]. Biotic interactions among species are likely to be different on the native and introduced ranges, being that in the latter natural competitors/predators of introduced species are usually absent (e.g., [6]). In our case, M. erminea competes with M. nivalis [33] and Neovison vison probably competes with M. putorius [78] in their native range, but these competitors species are absent in the Azores. Due to this absence of predators we could expect a wider distribution range and set of environmental conditions in the introduced-based models. Curiously, the introduced range models showed a more restricted distribution for both species. This could be simply because of an effect of the much smaller sample size of the introduced range, which could affect AIC values when selecting the best models. Nonetheless, model comparison was performed within study areas, so the best models should have been selected in both cases. Although our native-area model was formed by a limited number of background points (see S1 File), the native-ranges generally comprise large continental areas, and the scale could affect predictions in narrower introduced ranges, as insular systems are [79]. These differences suggest that the transferability of models from native to introduced ranges needs to be performed with caution, as it can be greatly affected by sampling sizes and might miss important conditions only found in the introduced ranges. Further, model predictions to remote or inaccessible areas (e.g., top of mountain of Pico, located around 2,300 m.a.s.l.) should be carefully interpreted.

Mustela nivalis distribution

Although M. nivalis uses a wide variety of habitats [1, 25, 34], it prefers rural and agricultural areas [34], and habitats that provide protection against potential predators (e.g., raptors, [33]). Consequently, M. nivalis potential distribution maps showed high HSI values in low and middle elevation areas, where human activities and human associated-habitats (e.g., agriculture, rural areas, etc) are concentrated. Additionally, M. nivalis is a specialist predator feeding on small mammals, especially small-rodents, and habitat selection is usually determined by local prey distribution [33]. Given that in the Azores rodents are more common in human-associated habitats [14, 80], rodent abundance potentially explains the M. nivalis distribution patterns.

M. nivalis in the Azores only occurs in the most human populated islands of Terceira and São Miguel [14, 15], but the predicted potential distribution for the remaining islands showed, in general, high habitat suitability, comprising almost the entire area for the smaller islands, probably due to their extensive agricultural fields. Given that those smaller Azorean islands also hold house mice and rats, if M. nivalis were to be introduced, it would probably become widespread and abundant. In the larger islands (e.g., Pico island) the predicted suitability was again higher for urban and rural areas near agricultural areas. High elevation areas showed low suitability, suggesting that the native and most pristine ecosystems might remain free of M. nivalis or that the species might occur in low abundance. However, rodent populations also occur at higher elevations in the Azores [14, 80].

Mustela putorius distribution

M. putorius uses different habitat types [1, 26, 40]. In the introduced areas, M. putorius usually occurs in grasslands, scrubs, pasture-lands, agriculture areas and urban and suburban areas [1, 26]. Therefore, our results are in line with the known M. putorius preferred habitats. Potential distribution maps showed higher HSI values at intermediate elevations, which are dominated by grasslands, agricultural areas, semi-natural meadows and exotic tree plantations. However, native-based models showed high HSI values also at low elevations, where urban areas and other human activities are concentrated, habitats frequently occupied by this species according to the species ecology [40]. Consequently, the native-based model included human-disturbance variables. Additionally, M. putorius is a predator specialized mainly in lagomorphs [81–83], and rabbits in the Azores prefer agricultural areas, grasslands and semi-natural meadows located inland at intermediate elevations. This is consistent with M. putorius predicted distribution with higher HSI in areas where rabbits are expected to be more abundant. M. putorius in the Azores occurs in most islands, except for Corvo and Graciosa [15]. The predictive maps revealed high HSI values in inland areas for M. putorius free islands, which suggests that an eventual introduction would possibly result in the establishment of this species on those two islands. The highest elevation areas of the Azores also showed low HSI values for M. putorius, which suggests that M. putorius is absent or in lower abundances in the most pristine native forest areas of the archipelago.

Conclusion

SDMs are often used to predict the potential distribution of invasive species based on environmental conditions on their native range (e.g., [30]). However, factors that influence species distribution in the introduced range could be novel or differ from those in their native ranges [73]. The difference might even be starker when the introduced ranges include islands while the native ranges are continental areas. Invasive species SDMs can be useful for the management of biological invasions but a careful interpretation is necessary and must be based on ecological knowledge [63].

In the case of the Azores, M. nivalis and M. putorius distribution patterns are mainly associated with climatic variables and human-associated habitats. We found that islands that are currently free of these species provide highly suitable habitat, being therefore important to prevent species arrival and establishment on these islands. Future studies should investigate the distribution of the two introduced carnivores based on their diet knowledge. Furthermore, given the potential impact of M. nivalis and M. putorius on native insular biodiversity, our results on the potential distribution of these introduced predators in the Azores might have important conservation implications, namely concerning seabirds’ colonies. Although the real impact of these predators on seabirds in the archipelago is yet to be assessed, the few existing studies (e.g., [19, 27] suggest that weasels are potential threats to seabirds, as highly suitable areas for this predator overlap with seabird breeding areas. Modeling species invasions on islands is therefore crucial to understand invaders ecological requirements and consequences, with potential cascading effects to native fauna and ecosystems, and to decide on actionable management options.

Supporting information

S1 File. Selection of background points selected to model the native area conditions.

(DOCX)

Click here for additional data file.^{(522.5KB, docx)}

S2 File. Occurrence records of Mustela nivalis and M. putorius.

(DOCX)

Click here for additional data file.^{(436.9KB, docx)}

S3 File. Detailed explanation about obtained environmental data.

(DOCX)

Click here for additional data file.^{(12.2KB, docx)}

S4 File. Correlation analysis between variables in native and introduced ranges.

(DOCX)

Click here for additional data file.^{(25KB, docx)}

S5 File. Figure detailing the stepwise modeling process.

(DOCX)

Click here for additional data file.^{(82.6KB, docx)}

S6 File. Combination of sets of candidate predictor variables.

(DOCX)

Click here for additional data file.^{(13.2KB, docx)}

S7 File. Maps of uncertainty in predictions.

(DOCX)

Click here for additional data file.^{(98.7KB, docx)}

Acknowledgments

We gratefully acknowledge the following people, for support in GIS processing: Agustín Fernández and Filipe Fernandes; for providing species records and other data: Verónica Neves, Joel Bried, Luis Barcelos, Rémi Fontaine, Iván Salgado and the "Atlas de Mamíferos de Portugal" team, namely Joana Bencatel; for logistical support and field assistance: Jose Sarangollo, David Rodilla, María Olivo, Sophie Wallon and Luis Ansias; and for comments and suggestions on the manuscript: Marco Girardello, Pedro Cardoso and Artur Gil. Data on introduced range was a contribution to AZORESBIOPORTAL that supported the Open Access of this manuscript though the project ACORES-01-0145-FEDER-000072, financed by FEDER in 85% and by Azorean Public funds by 15% through Operational Program Azores 2020.

Data Availability

All relevant data are within the manuscript and its Supporting Information files.

Funding Statement

LLL (SFRH/BD/115022/2016) and PAVB and MSR (UID/BIA/00329/2019) and IAR (SFRH/BPD/102804/2014) were supported by the Fundação para a Ciência e Tecnologia - FCT. MJS was supported by the University Research Priority Program in Global Change and Biodiversity from the University of Zürich.

References

1. Long JL. Introduced mammals of the world: their history, distribution and influence. Csiro Publishing; Collingwood, Victoria, Australia; 2003. [Google Scholar]
2.Mack RN, Simberloff D, Mark Lonsdale W, Evans H, Clout M, Bazzaz FA. Biotic invasions: causes, epidemiology, global consequences, and control. Ecol Appl. 2000; 10(3): 689–710. [Google Scholar]
3.Seebens H, Blackburn TM, Dyer EE, Genovesi P, Hulme PE, Jeschke JM, et al. Global rise in emerging alien species results from increased accessibility of new source pools. PNAS. 2018; 201719429. [DOI] [PMC free article] [PubMed] [Google Scholar]
4.Clout MN, Russell JC. The invasion ecology of mammals: a global perspective. Wildlife Res. 2008; 35(3): 180–184. [Google Scholar]
5.Ceballos G, Brown JH. Global patterns of mammalian diversity, endemism, and endangerment. Conserv Biol. 1995; 9(3): 559–568. [Google Scholar]
6.Courchamp F, Chapuis JL, Pascal M. Mammal invaders on islands: impact, control and control impact. Biol Rev. 2003; 78(3): 347–383. 10.1017/s1464793102006061 [DOI] [PubMed] [Google Scholar]
7.Towns DR. Invasive mammal eradications: the value of an ecosystem view. Anim Conserv. 2011; 14(5): 466–467. [Google Scholar]
8.Bellard C, Cassey P, Blackburn TM. Alien species as a driver of recent extinctions. Biol Lett. 2016; 12(2): 20150623 10.1098/rsbl.2015.0623 [DOI] [PMC free article] [PubMed] [Google Scholar]
9.Doherty TS, Glen AS, Nimmo DG, Ritchie EG, Dickman CR. Invasive predators and global biodiversity loss. PNAS. 2016; 113: 11261–11265. 10.1073/pnas.1602480113 [DOI] [PMC free article] [PubMed] [Google Scholar]
10.Wyatt KB, Campos PF, Gilbert MTP, Kolokotronis SO, Hynes WH, DeSalle R, et al. Historical mammal extinction on Christmas Island (Indian Ocean) correlates with introduced infectious disease. PLoS One. 2008; 3(11): e3602 10.1371/journal.pone.0003602 [DOI] [PMC free article] [PubMed] [Google Scholar]
11.Vitousek PM. Diversity and biological invasions of oceanic islands. In: Wilson E.O., Peter F.M. (Eds.) Biodiversity. National Academy of Sciences, Washington, DC; 1988. [Google Scholar]
12.Bicknell TW, Reid JB, Votier SC. Probable predation of Leach’s Storm‐petrel Oceanodroma leucorhoa eggs by St Kilda Field Mice Apodemus sylvaticus hirtensis. Bird Study. 2009; 56(3): 419–422. [Google Scholar]
13.Frutuoso G. Saudades da terra. Instituto Cultural de Ponta Delgada, Ponta Delgada. Portugal. 1561.
14.Mathias MDL, Ramalhinho MG, Santos-Reis M, Petrucci-Fonseca F, Libois R, Fons R, et al. Mammals from the Azores islands (Portugal): an updated overview. Mammalia. 1998; 62(3): 397–408. [Google Scholar]
15.Borges PAV, Costa A, Cunha R, Gabriel R, Gonçalves V, Martins AF, et al. A list of the terrestrial and marine biota from the Azores. Princípia, Cascais, Azores; 2010. [Google Scholar]
16.Bencatel J, Sabino-Marques H, Álvares F, Moura AE, Barbosa AM. Atlas de Mamíferos de Portugal. 2ª Edition. University of Évora, Évora; 2019. [Google Scholar]
17.Monteiro LR, Ramos JA, Furness RW. Past and present status and conservation of the seabirds breeding in the Azores archipelago. Biol Conserv. 1996; 78(3): 319−328. [Google Scholar]
18.Bried J, Magalhães MC, Bolton M, Neves VC, Bell E, Pereira JC, et al. Seabird habitat restoration on Praia Islet, Azores archipelago. Ecol Restor. 2009; 27(1): 27–36. [Google Scholar]
19.Neves VC, Panagiotakopoulos S, Ratcliffe N. Predation on roseate tern eggs by European starlings in the Azores. Arquipelago–Life and Marine Sciences. 2011; 28: 15−23. [Google Scholar]
20.Hervías S, Henriques A, Oliveira N, Pipa T, Cowen H, Ramos JA, et al. Studying the effects of multiple invasive mammals on Cory’s shearwater nest survival. Biol Invasions. 2013; 15(1): 143–15521. [Google Scholar]
21.Hervías S, Oppel S, Medina FM, Pipa T, Díez A, Ramos JA, et al. Assessing the impact of introduced cats on island biodiversity by combining dietary and movement analysis. J Zool. 2014; 292(1): 39–47. [Google Scholar]
22.Rando JC, Alcover JA, Olson SL, Pieper H. A new species of extinct scops owl (Aves: Strigiformes: Strigidae: Otus) from São Miguel Island (Azores Archipelago, North Atlantic Ocean). Zootaxa. 2013; 3647(2): 343–357. [DOI] [PubMed] [Google Scholar]
23.Alcover JA, Pieper H, Pereira F, Rando JC. Five new extinct species of rails (Aves: Gruiformes: Rallidae) from the Macaronesian Islands (North Atlantic Ocean). Zootaxa. 2015; 4057(2): 151−190. 10.11646/zootaxa.4057.2.1 [DOI] [PubMed] [Google Scholar]
24.Rando JC, Pieper H, Olson SL, Pereira F, Alcover JA. A new extinct species of large bullfinch (Aves: Fringillidae: Pyrrhula) from Graciosa Island (Azores, North Atlantic Ocean). Zootaxa. 2017; 4282(3): 567–583. [Google Scholar]
25.Global Invasive Species Database. Species profile: Mustela nivalis. Available from http://www.iucngisd.org/gisd/speciesname/Mustela+nivalis; 2018a.
26.Global Invasive Species Database. Species profile: Mustela furo. Available from http://www.iucngisd.org/gisd/speciesname/Mustela+furo; 2018b.
27.Pitta-Groz M, Pereira JC, Silva AG. Invasive alien species as the main threat to Azores seabird populations. In: Proceedings of Workshop on invasive alien species on European islands and evolutionary isolated ecosystems. Horta, Azores, Portugal: 10 October 2002. Convention on the conservation of European wildlife and natural habitats; 2002. [Google Scholar]
28.Warren DL, Seifert SN. Ecological niche modeling in Maxent: the importance of model complexity and the performance of model selection criteria. Ecol Appl. 2011; 21(2): 335–342. 10.1890/10-1171.1 [DOI] [PubMed] [Google Scholar]
29.Elith J, Leathwick JR. Species distribution models: ecological explanation and prediction across space and time. Annu Rev Ecol Evol S. 2009; 40: 677–697. [Google Scholar]
30.Welk E, Schubert K, Hoffmann MH. Present and potential distribution of invasive garlic mustard (Alliaria petiolata) in North America. Divers Distrib. 2002; 8(4): 219–233. [Google Scholar]
31.Wiens JJ, Graham CH. Niche conservatism: integrating evolution, ecology, and conservation biology. Annu Rev Ecol Evol S. 2005; 36: 519–539. [Google Scholar]
32.Shea K, Chesson P. Community ecology theory as a framework for biological invasions. Trends Ecol Evol. 2002; 17(4): 170–176. [Google Scholar]
33.Sheffield SR, King CM. Mustela nivalis. Mammalian Species. 1994; 454: 1−10. [Google Scholar]
34.McDonald RA, Abramov AV, Stubbe M, Herrero J, Maran T, Tikhonov A, et al. Mustela nivalis. The IUCN Red List of Threatened Species 2016; 2019.
35.Rodrigues M, Bos AR, Schembri PJ, de Lima RF, Lymberakis P, Parpal L, et al. Origin and introduction history of the least weasel (Mustela nivalis) on Mediterranean and Atlantic islands inferred from genetic data. Biol Invasions. 2017; 19(1): 399–421. [Google Scholar]
36.King CM. Weasel. In: King C.M.(Ed.), The Handbook of New Zealand Mammals. Oxford University Press, Auckland; 2005. [Google Scholar]
37.Corbett GB, Ovenden D. The Mammals of Britain and Europe. Collins, London; 1980. [Google Scholar]
38.King CM. The Handbook of New Zealand Mammals, 1st Edition. Oxford University Press, Auckland; 1990. [Google Scholar]
39.Costa M, Fernandes C, Birks JDS, Kitchener AC, Santos-Reis M, Bruford MW. The genetic legacy of the 19th century decline of the British polecat: evidence for extensive introgression from feral ferrets. Mol Ecol. 2013; 22(20): 5130−5147. 10.1111/mec.12456 [DOI] [PubMed] [Google Scholar]
40.Skumatov D, Abramov AV, Herrero J, Kitchener A, Maran T, Kranz A, et al. Mustela putorius. The IUCN Red List of Threatened Species 2016; 2016. [Google Scholar]
41.Arechavaleta M, Rodríguez S, Zurita N, García A. Lista de especies silvestres de Canarias. Hongos, plantas y animales terrestres. Gobierno de Canarias, Canarias; 2009.
42.Clapperton BK. Advances in New Zealand mammalogy 1990–2000: Feral ferret. J Roy Soc New Zeal. 2001; 31(1): 185–203. [Google Scholar]
43.Lever C. Naturalized mammals of the world. Longman Science & Technology, London, UK; 1985. [Google Scholar]
44.Gaspar C, Borges PA, Gaston KJ. Diversity and distribution of arthropods in native forests of the Azores archipelago. Arquipélago–Life and Marine Sciences. 2008; 25: 1–30. [Google Scholar]
45.Triantis KA, Borges PA, Ladle RJ, Hortal J, Cardoso P, Gaspar C, et al. Extinction debt on oceanic islands. Ecography. 2010; 33(2): 285–294. [Google Scholar]
46.Cardoso P, Rigal F, Fattorini S, Terzopoulou S, Borges PA. Integrating landscape disturbance and indicator species in conservation studies. PLoS One. 2013; 8(5): e63294 10.1371/journal.pone.0063294 [DOI] [PMC free article] [PubMed] [Google Scholar]
47.Terzopoulou S, Rigal F, Whittaker RJ, Borges PA, Triantis KA. Drivers of extinction: the case of Azorean beetles. Biol Lett. 2015; 11(6): 20150273 10.1098/rsbl.2015.0273 [DOI] [PMC free article] [PubMed] [Google Scholar]
48.Borges PAV, Lobo JM, Azevedo EB, Gaspar C, Melo C, Nunes LV. Invasibility and species richness of island endemic arthropods: a general model of endemic vs. exotic species. J Biogeogr. 2006; 33: 169–187. [Google Scholar]
49.Hortal J, Jiménez‐Valverde A, Gómez JF, Lobo JM, Baselga A. Historical bias in biodiversity inventories affects the observed environmental niche of the species. Oikos. 2008; 117(6): 847–858. [Google Scholar]
50.Phillips SJ, Dudík M, Elith J, Graham CH, Lehmann A, Leathwick J, et al. Sample selection bias and presence‐only distribution models: implications for background and pseudo‐absence data. Ecol Appl. 2009; 19(1): 181–197. 10.1890/07-2153.1 [DOI] [PubMed] [Google Scholar]
51.Ordeñana MA, Crooks KR, Boydston EE, Fisher RN, Lyren LM, Siudyla S, et al. Effects of urbanization on carnivore species distribution and richness. J Mammal. 2010; 91(6): 1322–1331. [Google Scholar]
52.Long RA, Donovan TM, MacKay P, Zielinski WJ, Buzas JS. Predicting carnivore occurrence with noninvasive surveys and occupancy modeling. Landsc Ecol. 2011; 26(3): 327–340. [Google Scholar]
53.Cuyckens GAE, Perovic PG, Cristobal L. How are wetlands and biological interactions related to carnivore distributions at high altitude?. J Arid Environ. 2015; 115: 14–18. [Google Scholar]
54.Team RStudio. RStudio: Integrated Development for R. RStudio, Inc., Boston, MA; 2015. URL http://www.rstudio.com/ [Google Scholar]
55.Phillips SJ, Anderson RP, Schapire RE. Maximum entropy modeling of species geographic distributions. Ecol Modell, 2006; 190(3–4): 231–259. [Google Scholar]
56.Merow C, Smith MJ, Silander JA. A practical guide to MaxEnt for modeling species’ distributions: what it does, and why inputs and settings matter. Ecography. 2013; 36(10): 1058–1069. [Google Scholar]
57.Elith J, Graham CH, Anderson RP, Dudík M, Ferrier S, Guisan A, et al. Novel methods improve prediction of species' distributions from occurrence data. Ecography. 2006; 29: 129–151. [Google Scholar]
58.Elith J, Phillips SJ, Hastie T, Dudík M, Chee YE, Yates CJ. A statistical explanation of MaxEnt for ecologists. Divers Distrib. 2011; 17(1): 43–57. [Google Scholar]
59.Hernandez PA, Graham CH, Master LL, Albert DL. The effect of sample size and species characteristics on performance of different species distribution modeling methods. Ecography. 2006; 29(5): 773–785. [Google Scholar]
60.Bean WT, Stafford R, Brashares JS. The effects of small sample size and sample bias on threshold selection and accuracy assessment of species distribution models. Ecography. 2012; 35(3): 250–258. [Google Scholar]
61.Proosdij AS, Sosef MS, Wieringa JJ, Raes N. Minimum required number of specimen records to develop accurate species distribution models. Ecography. 2016; 39(6): 542–552. [Google Scholar]
62.Gallien L, Douzet R, Pratte S, Zimmermann NE, Thuiller W. Invasive species distribution models–how violating the equilibrium assumption can create new insights. Global Ecol Biogeogr. 2012; 21(11): 1126–1136. [Google Scholar]
63.Václavík T, Kupfer JA, Meentemeyer RK. Accounting for multi‐scale spatial autocorrelation improves performance of invasive species distribution modelling (iSDM). J Biogeogr. 2012; 39(1): 42–55. [Google Scholar]
64.McPherson JM, Jetz W, Rogers DJ. The effects of species’ range sizes on the accuracy of distribution models: ecological phenomenon or statistical artefact?. J Appl Ecol. 2004; 41(5): 811–823. [Google Scholar]
65.Lobo JM, Jiménez‐Valverde A, Real R. AUC: a misleading measure of the performance of predictive distribution models. Global Ecol Biogeogr. 2008; 17(2): 145–151. [Google Scholar]
66.Barbet‐Massin M, Jiguet F, Albert CH, Thuiller W. Selecting pseudo‐absences for species distribution models: how, where and how many?. Methods Ecol Evol. 2012; 3(2): 327–338. [Google Scholar]
67.Burnham KP, Anderson DR. Model selection and multimodel inference: a practical information-theoretic approach. Springer Science & Business Media; 2003. [Google Scholar]
68.Burnham KP, Anderson DR. Multimodel inference: understanding AIC and BIC in model selection. Sociol Methods Res. 2004; 33(2): 261–304. [Google Scholar]
69.Kuemmerle T, Perzanowski K, Chaskovskyy O, Ostapowicz K, Halada L, Bashta AT, et al. European bison habitat in the Carpathian Mountains. Biol Conserv. 2010; 143(4): 908–916. [Google Scholar]
70.Araújo MB, Anderson RP, Barbosa AM, Beale CM, Dormann, C. F., Early, R., et al. Standards for distribution models in biodiversity assessments. Sci Adv. 2019; 5(1): eaat4858 10.1126/sciadv.aat4858 [DOI] [PMC free article] [PubMed] [Google Scholar]
71.Rilov G, Benayahu Y, Gasith A. Prolonged lag in population outbreak of an invasive mussel: a shifting-habitat model. Biol Invasions. 2004; 6(3): 347–364. [Google Scholar]
72.Broennimann O, Treier UA, Müller‐Schärer H, Thuiller W, Peterson AT, Guisan A. Evidence of climatic niche shift during biological invasion. Ecol Lett. 2007; 10(8): 701–709. 10.1111/j.1461-0248.2007.01060.x [DOI] [PubMed] [Google Scholar]
73.Fitzpatrick MC, Weltzin JF, Sanders NJ, Dunn RR. The biogeography of prediction error: why does the introduced range of the fire ant over‐predict its native range?. Global Ecol Biogeogr. 2007; 16(1): 24–33. [Google Scholar]
74.Azevedo EB, Pereira LS, Itier B. Modelling the local climate in island environments: Water balance applications. Agric Water Manag. 1999; 40: 393–403. [Google Scholar]
75.Hortal J, Borges PAV, Jiménez-Valverde A, Azevedo EB, Silva L. Assessing the areas under risk of invasion within islands through potential distribution modelling: the case of Pittosporum undulatum in São Miguel, Azores. J Nat Conserv. 2010; 18: 247–257. [Google Scholar]
76.Aparício BA, Cascalho J, Cruz MJ, Borges PA, Azevedo EB, Elias RB, et al. Assessing the landscape functional connectivity using movement maps: a case study with endemic Azorean insects. J Insect Conserv. 2018; 1–9. [Google Scholar]
77.Fitzpatrick MC, Weltzin JF. Ecological niche models and the geography of biological invasions: a review and a novel application. In Invasive Plants: Ecological and Agricultural Aspects. Birkhäuser Basel. Switzerland; 2005.
78.Kauhala K. Introduced carnivores in Europe with special reference to central and northern Europe. Wildlife Biol. 1996; 2(3): 197–204. [Google Scholar]
79.Dungan J, Perry J, Dale M, Legendre P, Citron-Pousty S, Fortin M, et al. A balanced view of scale in spatial statistical analysis. Ecography. 2002; 25: 626–640. [Google Scholar]
80.Collares-Pereira M, Mathias ML, Santos-Reis M, Ramalhinho MG, Duarte-Rodrigues P. Rodents and Leptospira transmission risk in Terceira island (Azores). Eur J Epidemiol. 2000; 16(12): 1151–1157. 10.1023/a:1010916132497 [DOI] [PubMed] [Google Scholar]
81.Smith GP, Ragg JR, Moller H, Waldrup KA. Diet of feral ferrets (Mustela furo) from pastoral habitats in Otago and Southland, New Zealand. New Zeal J Zool. 1995; 22(4): 363–369. [Google Scholar]
82.Ragg JR. Intraspecific and seasonal differences in the diet of feral ferrets (Mustela furo) in a pastoral habitat, East Otago, New Zealand. New Zeal J Ecol. 1998; 113–119. [Google Scholar]
83.Santos MJ, Matos HM, Baltazar C, Grilo C, Santos-Reis M. Is polecat (Mustela putorius) diet affected by “mediterraneity”?. Mammalian Biology-Zeitschrift für Säugetierkunde. 2009; 74(6): 448–455. [Google Scholar]

PLoS One. doi: 10.1371/journal.pone.0237216.r001

Decision Letter 0

Ulrike Gertrud Munderloh

14 Jan 2020

PONE-D-19-33731

Modelling the distribution of Mustela nivalis and M. putorius in the Azores archipelago based on native and introduced ranges

PLOS ONE

Dear Mr Lamelas-López,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

While reviewers agree that this is an important study, there are major concerns about study design and data analysis.

We would appreciate receiving your revised manuscript by Feb 28 2020 11:59PM. When you are ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter.

To enhance the reproducibility of your results, we recommend that if applicable you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. For instructions see: http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). This letter should be uploaded as separate file and labeled 'Response to Reviewers'.
A marked-up copy of your manuscript that highlights changes made to the original version. This file should be uploaded as separate file and labeled 'Revised Manuscript with Track Changes'.
An unmarked version of your revised paper without tracked changes. This file should be uploaded as separate file and labeled 'Manuscript'.

Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out.

We look forward to receiving your revised manuscript.

Kind regards,

Ulrike Gertrud Munderloh, Ph.D.

Academic Editor

PLOS ONE

Journal Requirements:

When submitting your revision, we need you to address these additional requirements.

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at

http://www.journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and http://www.journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

2. Our internal editors have looked over your manuscript and determined that it is within the scope of our Biodiversity Conservation Call for Papers. This collection of papers is headed by a team of Guest Editors for PLOS ONE (https://collections.plos.org/s/biodiversity). The Collection will encompass a diverse range of research articles on biodiversity conservation, including management of invasive species. Additional information can be found on our announcement page: https://collections.plos.org/s/biodiversity

If you would like your manuscript to be considered for this collection, please let us know in your cover letter and we will ensure that your paper is treated as if you were responding to this call. If you would prefer to remove your manuscript from collection consideration, please specify this in the cover letter

3. In your Methods section, please provide additional location information of the study area, including geographic coordinates for the data set if available.

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: No

Reviewer #2: Partly

**********

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: No

Reviewer #2: No

**********

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: No

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

**********

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: This paper models habitat suitability in the Azores archipelago for two introduced mustelid species, Mustela nivalis and Mustela putorius. For each species, the authors fit two models: (i) one model fitted to a dataset of detections collected over the whole Europe (native range), and (ii) one model fitted to a dataset of detections collected only in the Azores islands (introduced range). For each species, the authors then compare the maps predicted by the two models (introduced and native), and show that the maps obtained with the two datasets are different. The map predicted using data collected over the whole Europe does not correspond to the map predicted using data collected in the Azores. The authors conclude that this difference indicates that species distribution models should be carefully interpreted when used to predict the range of a species in an area based on data collected in another area.

As a whole, I found this paper well written. Their aim is interesting and I fully endorse their conclusion, i.e. that "factors that influence species distribution in the introduced range could be novel or differ from those in their native ranges (...). The difference might even be starker when the introduced ranges include islands while the native ranges are continental areas. Invasive species SDMs can be useful for the management of biological invasions but a careful interpretation is necessary and must be based on ecological knowledge".

However I have concerns related to the statistical methods used in their paper. In addition, I think that this paper should contain more details on the approach used to fit the models, possibly given in supplementary data. It is presently difficult to judge of the validity of the analyses based on the information provided in the paper.

My main concerns are:

****

My first main concern is related the difference of scales of their two models (native and introduced). The concept of scale is an essential concept in ecology, and we should not expect to obtain the same results at different scales. For example, on a large scale, a savannah ungulate species might select areas close to water during the dry season, but on a small scale rarely use it rarely of the abundance of nearby predators. Thus large scale SDM might predict a high HSI for areas close to water, whereas small-scale SDM might predict low HSI for such areas. Similarly, I do not expect that differences between the distribution of a species over a whole continent with a very large diversity of climates, landscapes, human densities, etc. and the expected distribution in a set of small islands with a comparatively homogeneous climate and landscape to be automatically caused by the fact that the species is native in one place and introduced in another. This difference of scale may for example explain why "native" models predict wider range than "introduced" models. In my opinion, it would be more sensible to compare the Azores to areas with similar size and environmental characteristics (or at least as similar as possible) where the species is natively present.

****

Another important concern is the use of a very heterogeneous detection dataset to fit the models. This dataset was built from various datasources including mainly GBIF and citizen science programs. As rightly noted by the authors, such data are characterized by numerous collection bias. Thus, species detection are generally more numerous in areas where there are more observers. For example, considering the Terceira island in Fig. 2 and 3, it is clear that most detections occur in places where there are most people (Angra do Heroismo on the Eastern side, Praia da Vitoria on the southern side). The same is true for data collected in mainland Europe.

Thus, any model not accounting for these bias will be biased and predict high suitability indices in areas where there are both many observers and many animals. The only possible approach to SDM with such data implies a model of the spatial distribution of the survey effort.

The authors acknowledge this point and indicate "presence records from these datasets may be affected by sampling bias, because they are often spatially biased toward easily accessed areas. This spatial bias can lead environmental bias given that the background data are usually drawn at random from the entire region [48]. Consequently, and according to Phillips et al. [48], we chose background data exhibiting the same bias as the presence records." However, the authors do not detail how they chose such biased background data.

It seems that the authors have used the *target group* method of Phillips et al., as they cite this paper. This approach rely on the definition of a broad set of many species, chosen to represent the specimen collection or observation activities of collectors of the target species (so-called "target groups"). Is this indeed the method used here ? If this is the case, what are the species chosen for this target group ? Are the data selected from the same source?

Moreover, the Azorean data were "cleaned" by selecting one record per squared km "to avoid a potential bias of spatially auto-correlated presences". I do not see how the method of Phillips et al. could be used with such a treatment: either (i) the same "cleaning" treatment is carried out on the background data (keeping one background point per squared km) and in this case, we loose the bias correction in the model (since the greater probability to find a detection at a given place is no longer reflected by the amount of background points at this place), or (ii) the same "cleaning" treatment is not carried out on the background data, and these data "overcorrect" the bias where the species is detected (since the presence data have a "cleaning" treatment that the background data do not have, and the species can only be detected once at a given place).

I would like to see a map of the distribution of the points of other species used as background data, to have an idea of the spatial distribution of the survey effort.

This is an important aspect for a SDM, as the fitting of SDM to such a biased dataset will focus only on parts of geographic space that contain presence samples. As noted by Phillips et al. "predictions into unsampled areas, especially those with conditions outside the range observed in sampled areas, should be treated with strong caution." This is an important warning given the aim of the paper: given the small sample size of data available in the Azores island, I fear that there are areas that are not sampled at all, and this map would allow to identify which predictions should not be trusted at all.

****

Other methodological concerns are related to the small sample size of the data set used for the "introduced models":

* Note that because the "introduced" model is fitted on a very small dataset (less than 30 detections), the number of variables in the model will necessarily be much smaller than the number of variables in the "native" model based on a much larger dataset (the sample size has a very strong effect on the number of variables that the AIC will select). This difference of sample sizes makes difficult the comparison of the structure of the two separate models (in particular, it is difficult to conclude anything concerning the fact that a particular variable is present in the "native" model and not in the "introduced" model).

* Visual comparisons of maps in figure 2 and 3 are actually simply visual comparisons of point predictions by models. The prediction uncertainty is not accounted for in theses comparisons. Given the small sample size for "introduced" models, and given that some areas in the Azores islands seem to be scarcely sampled, this will probably lead to weak conclusions at these places.

*****

Detailed comments follow:

* Line 241--247. The description of the models is not clear. Would it be possible to present them in a table ? Presently, it is very difficult to understand which models have been compared. For example, the second model presented here is: "topographic + climatic + landscape variables (n=9) + topographic + climatic + human variables (n=8)". Here, "Topographic + climatic" is present twice ? I think that a more detailed explanation is required.

* Table 2: for the model "M. putorius, introduced": what is the difference between the two models since both include the same variables?

* line 282 : "For M. nivalis, one model was selected for both the native and the introduced range". Do the authors mean "one model was selected for each dataset" ? The selected model is indeed not the same for the two species.

* line 281--290. It is not necessary to include the Akaike weights both in the table and in the text.

* Table 4: there are several problems with this table, for the "introduced" models: the sum of percentage for M. putorius is >100%, and for Mustela nivalis, > 50% of the records are characterized with HSI<0.5, contrarily to what is written in the text.

Reviewer #2: GENERAL COMMENTS

This a well-written and potentially interesting study modelling two species’ distributions based on native and (insular) introduced ranges. However, some parts of the methods need further explanation or tuning, and the Discussion also needs improvements. It is also not clear where the species occurrence data collected by the authors are made publicly available.

Figure 1 shows a substantial problem of survey bias, with some countries presenting a very high and other countries a very low density of occurrence records, without this reflecting the actual occurrence patterns of the analysed species. France and Switzerland are very clear examples of countries providing wildly insufficient data in this case, but even other countries like Spain and Portugal are clearly under-represented in the analysed dataset, compared to other data sources (at broader spatial resolutions) such as the national mammal atlases, which show that these species occur in many more areas than were used in the native range models. Moreover, this bias is not related to accessible areas, and it is unclear how the Maxent bias analysis deals with it.

Other parts of the methods also need better explanation and justification, such as the limited variations in model parameters that were chosen for sensitivity analysis, as well as the limited method for model evaluation and selection. Recent specific literature on good modelling practices should be used for better tuning of model parameters.

Figures 2 and 3 show very different (sometimes nearly opposite) predictions from native-range and introduced-range models. This needs to be explored and explained more clearly – and separately – in the Discussion.

SPECIFIC COMMENTS

Ln 20: I would add “archipelago” after “Azores”, as this is a relevant biogeographic feature which may not be immediately obvious to all readers.

Ln 24-26: The current phrasing is confusing as to whether the “differences” are between species or between models. I suggest rephrasing to something like “We found differences in the predicted distributions of models based on introduced and native occurrences for both M. nivalis and M. putorius in the Azores”.

Ln 32: It its not clear what “this” refers to.

Ln 62: The Azores are more than one island, so “island’s ” should be “islands’ ”.

Ln 55-59: This paragraph refers common mammals in the Azores, but most of the provided references are focused on birds. I miss a reference to the atlas of Portuguese mammals, which seems to be mentioned in acknowledgments but is not among the cited literature.

Ln 69: “distributions” should be “distribution”.

Ln 78: Overly long sentence. I would add a comma after “conservatism”.

Ln111: Replace the final comma with a full stop and start a new sentence.

Ln 144: “deliberately” should be “deliberate”.

Ln 145: “or accidentally” is redundant in this sentence.

Ln 163: Were these direct observations and sampling campaigns conducted across the modelled range, or centered on particular regions/countries? It would also be interesting to mention how many points were added by the authors to the data available in the open published datasets, and if these data were added to these (or to which) public platforms.

Ln 164: How was this record accuracy assessed? Namely, which columns of the public databases were used for this filtering, and with which values?

Ln 166: In this case, the bias was clearly not (only) towards accessible areas, but also reflected the habits of different countries in uploading occurrence records to the analysed databases. France and Switzerland, for example, have a lot of accessible areas but a tiny portion of records.

Ln 168-169: More details are necessary on how this biased background sample was generated exactly, and a map of these background points should be included either in the article or the supplementary files.

Ln 169: I would start a new paragraph at “The introduced...”

Ln 173-174: Was this elimination of records within the same km2 really done only for the Azores? In that case, why (and why use “independent” also in Ln 177 for Europe)? Also, this elimination does not avoid (nor should it) spatial autocorrelation in the presences, but rather in the survey effort.

Ln 177 and 269: Remove “the” before “Europe”.

Ln 204-206: Which were the normal and the non-normal variables? Was Pearson’s coefficient only used when both of the variables in a pair were normally distributed? In any case, this seems like quite an unequal treatment of different variables, as the two coefficients have visibly different power. Do you have a justification or a reference for this procedure?

Ln 223-225: The literature has evolved quite a bit since this reference of 2006, and a few more papers on “best practices” are available nowadays – including Araújo et al. 2019, which is in the reference list although I could not find it cited in the text; and others such as Sofaer et al. 2019 (https://academic.oup.com/bioscience/article/69/7/544/5505326). Also, two important references on Maxent modelling in particular are Elith et al. 2010 (https://onlinelibrary.wiley.com/doi/full/10.1111/j.1472-4642.2010.00725.x) and Merow et al. 2013 (https://onlinelibrary.wiley.com/doi/10.1111/j.1600-0587.2013.07872.x). This should be used to choose appropriate parameters sensibly, rather than just testing limited arbitrary choices.

Ln 226-227: How are 25 and 30% “random test percentages”? Do you mean percentages of 25 and 30% of random test records? I also find this testing of two such similar percentages quite limited, and I find it surprising that a larger proportion of records left out for testing produced apparently better models. This may have to do with both percentages being so similar, and/or with the limited model evaluation procedure (see comment further down).

Ln 228-229: The number of background points is among the main factors affecting model quality, so how did you select this particular number of points, and (especially) why was this parameter not subjected to the sensitivity analysis?

Ln 230: What are the “total background points?”

Ln 234-235: The AUC is not the best metric (especially if used alone) to choose between models, as it has relevant known problems (e.g. Lobo et al. 2008, https://onlinelibrary.wiley.com/doi/abs/10.1111/j.1466-8238.2007.00358.x). Also, the AUC assesses the capacity of models to distinguish between presence and absence records, whereas the data and modelling algorithm in this paper imply presence and background (not absence, nor pseudo-absence) records. Even the reference that the authors cite for “best practices” (Ln 224-25) says that “multiple evaluation measures are necessary to determine accuracy of models produced with presence-only data”.

Ln 235-236: The AUC actually varies between 0 (not 0.5) and 1; 0 means perfect discrimination but backwards classification, and it can happen when models are applied to external sets. Also, 1 does not mean “perfect accuracy”, but perfect discrimination (with correct classification). Accuracy would imply the correctness of the exact continuous values predicted by the model, whereas the AUC only assesses to which extent a model gives higher values (regardless of how much higher) to presence than to absence localities.

Ln 236-239: This sentence is misleading, as some of the parameters mentioned (e.g. number of background points, type of features) were not subjected to sensitivity analysis, so you cannot include them in the “best parameter configuration”. Also, was this configuration selected for all four models tested (M. nivalis and M. putorius in Azores and Europe)?

Ln 271-272: This is only a “validation” if the same field data were not used in model building.

Ln 285-290: Here it is not clear what these model “weights” refer to, and why a weight of 0.86 is better than a weight of 0.99.

Ln 291: “probability of […] occurrence” is not provided by a presence-background model such as Maxent. Do you mean “suitability for […] occurrence”?

Ln 338-341: I don’t see how this is an “ecological” approach. Also, what is relevant is if the proportion of records in areas with suitability >0.5 is higher that expected by chance, given the amount of occurrence records and the amount of pixels available with suitability >0.5. These numbers alone do not prove anything without some assessment of significance, e.g. with a test of equal or given proportions.

Ln 367: Add “more” before “widespread”.

Ln 372-374: This is interesting, but it conflicts with the finding that the introduced range is actually more restricted than would be predicted by the native range, even without these competitors in the introduced range. The whole Discussion should separate and interpret more clearly the results of the different types of models.

Figures 2 and 3: I’d suggest switching to a colourblind-friendly colour scale, as red and green are indistinguishable for a significant fraction of the potential readers.

**********

6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files to be viewed.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at figures@plos.org. Please note that Supporting Information files do not need this step.

PLoS One. 2020 Aug 7;15(8):e0237216. doi: 10.1371/journal.pone.0237216.r002

Author response to Decision Letter 0

4 Mar 2020

University of Azores,

Angra do Heroísmo, Portugal

04 March 2020

Dear Editor of PLOS ONE

We are pleased to submit the revised version of the manuscript entitled “Modelling the distribution of Mustela nivalis and M. putorius based on native and introduced ranges” for consideration for publication in PLOS ONE. We greatly appreciate the depth of the comments provided by yourself and the reviewers, which we believe enriched and improved the manuscript. Below we detail our response to every comment we have received, how we have incorporated it in the manuscript and the changes we implemented in the new version of the manuscript. We believe that the outcome is a much improved version of the manuscript. Also, we would like considered our manuscript for "Biodiversity Conservation Call for papers" collection.

Looking forward to your assessment,

Kind regards

Lucas Lamelas-López, on behalf of all the co-authors.

ASSOCIATE EDITOR COMMENTS TO AUTHOR:

While reviewers agree that this is an important study, there are major concerns about study design and data analysis

Authors’ Response: We have taken both your and the reviewers comments very seriously and incorporated them in the updated version of the manuscript.

JOURNAL REQUIREMENTS:

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at

Authors’ Response: We have revised the style requirements and the file names.

Authors’ Response: Thanks for your suggestion. We would like considered our manuscript for Biodiversity Conservation collection.

3. In your Methods section, please provide additional location information of the study area, including geographic coordinates for the data set if available.

Authors’ Response: We have created a new supplementary material (S2 File), including the geographic coordinates of the occurrence records, according your suggestion and the reviewers comments.

REVIEW COMMENTS TO THE AUTHOR:

REVIEWER #1

GENERAL COMMENTS:

This paper models habitat suitability in the Azores archipelago for two introduced mustelid species, Mustela nivalis and Mustela putorius. For each species, the authors fit two models: (i) one model fitted to a dataset of detections collected over the whole Europe (native range), and (ii) one model fitted to a dataset of detections collected only in the Azores islands (introduced range). For each species, the authors then compare the maps predicted by the two models (introduced and native), and show that the maps obtained with the two datasets are different. The map predicted using data collected over the whole Europe does not correspond to the map predicted using data collected in the Azores. The authors conclude that this difference indicates that species distribution models should be carefully interpreted when used to predict the range of a species in an area based on data collected in another area.

Authors’ Response: We appreciate the reviewers’ interest on the aim of our research. We are grateful for your constructive comments and suggestions that helped us to improve the manuscript quality. Below we answer to each of them individually.

Authors’ Response: Thank you for your concerns regarding the statistical analyses. We agree that there was restricted description of the approach used to fit the models and to amend this we have now included new supplementary materials (Files S1, S2, S5, S6 and S7) that detail the approach step-by-step. We believe that with this information you will be able to assess the validity of the analyses and results presented in the paper. We also included eight new references suggested by the reviewers to support our Methods and Discussion.

My main concerns are:

****

Authors’ Response: Thank you for your concern and example on how the effect of scale could affect our results. However, we believe this effect is not affecting the model outputs because we used the same spatial resolution and the same data sources for both study areas; therefore, if one factor at a smaller spatial scale is missed it will be missed on both analyses. The only factor that we anticipate to be affected by the different study area sizes is prevalence (i.e. the fraction of the study area that the species can occupy), but this is compensated for by having a similar sampling density for both areas.

Similarly, I do not expect that differences between the distribution of a species over a whole continent with a very large diversity of climates, landscapes, human densities, etc. and the expected distribution in a set of small islands with a comparatively homogeneous climate and landscape to be automatically caused by the fact that the species is native in one place and introduced in another. This difference of scale may for example explain why "native" models predict wider range than "introduced" models. In my opinion, it would be more sensible to compare the Azores to areas with similar size and environmental characteristics (or at least as similar as possible) where the species is natively present.

Authors’ Response: Thank you for your interesting suggestion. We took it into consideration and we concluded that it would be difficult to justify selecting sub areas within continental Europe that would match the size of the Azores to conduct the analysis as suggested. This is because there could be an even bigger effect due to this selection than the potential effects of modeling a smaller area using information from a larger area. Further, we believe that this suggestion is not in line with other approaches to model invasive species ranges. Several authors have also modeled invasive species range from information retrieved from the native range (e.g., Broennimann et al., 2007; Fitzpatrick et al., 2007; Rödder et al., 2009; Bidinger et al., 2012); This is because invasive species are yet to be in equilibrium with their environment in the invaded ranges thus becoming even more important to include a broader set of parameter ranges from the native range to predict areas with potential for invasion.

Bidinger K, Lötters S, Rödder D, Veith M. Species distribution models for the alien invasive Asian Harlequin ladybird (Harmonia axyridis). J Appl Entomol. 2012; 136(12): 109–123.

Broennimann O, Treier UA, Müller‐Schärer H, Thuiller W, Peterson AT, Guisan A. Evidence of climatic niche shift during biological invasion. Ecol Lett. 2007; 10(8): 701−709.

Fitzpatrick MC, Weltzin JF, Sanders NJ, Dunn RR. The biogeography of prediction error: why does the introduced range of the fire ant over‐predict its native range?. Global Ecol Biogeogr. 2007; 16(1): 24−33.

Rödder D, Schmidtlein S, Veith M, Lötters S. Alien invasive slider turtle in unpredicted habitat: a matter of niche shift or of predictors studied? PLOS ONE. 2009; 4(11).

****

Another important concern is the use of a very heterogeneous detection dataset to fit the models. This dataset was built from various datasources including mainly GBIF and citizen science programs.

Authors’ Response: Indeed, but this is also the case of GBIF and other datasets commonly used in species distribution models.

As rightly noted by the authors, such data are characterized by numerous collection bias. Thus, species detection are generally more numerous in areas where there are more observers. For example, considering the Terceira island in Fig. 2 and 3, it is clear that most detections occur in places where there are most people (Angra do Heroismo on the Eastern side, Praia da Vitoria on the southern side). The same is true for data collected in mainland Europe.

Authors’ Response: We have noted that SDM datasets are often affected by several bias due to differences in numbers of observers, experience, location, detectability, date, season, etc. Most of these biases resulting from opportunistic sampling may influence SDM performance and predictability. This is not an unknown or unresolved problem in SDM, and to account for these sampling biases we followed the suggestions by Phillips et al. (2009) to chose background data exhibiting the same spatial bias as the presence records. We have included two new Supplementary Material files with all occurrence records used in the study (S2 File), including information about the source (Open Source or Field Data); S1 File contains the methodology used to create the biased background points (see for more details).

In the case of the introduced-range records, most of them were collected from a set of standardized sampling campaigns across all island habitats, therefore without a bias towards urban areas. The authors were involved in these sampling campaigns (e.g., Mathias et al., 1998; Collares-Pereira et al., 2000; Lamelas-López & Salgado, in press).

Collares-Pereira M, Mathias ML, Santos-Reis M, Ramalhinho MG, Duarte-Rodrigues P. Rodents and Leptospira transmission risk in Terceira island (Azores). Eur J Epidemiol. 2000; 16(12): 1151−1157.

Lamelas-López L &Salgado I (In press). Applying camera-trapping to detect and monitor introduced mammal species on oceanic islands. Oryx.

Mathias MDL, Ramalhinho MG, Santos-Reis M, Petrucci-Fonseca F, Libois R, Fons R, et al. Mammals from the Azores islands (Portugal): an updated overview. Mammalia. 1998; 62(3): 397−408.

Phillips SJ, Dudík M, Elith J, Graham CH, Lehmann A, Leathwick J, Ferrier S. Sample selection bias and presence‐only distribution models: implications for background and pseudo‐absence data. Ecol Appl. 2009; 19(1): 181−197.

Authors response: We have included a new supplementary material containing the methodology used to create the biased background points (See S1 File). The chosen species were our two target invasive mammals, Mustela putorius and Mustela nivalis.

Are the data selected from the same source?

Authors response: Yes. The data was selected by mimicking the sampling bias in the presence data. We included a new ESM which contains the explanation on how this dataset was created and a map with the biased background points used to model the native area (see S1 File - Figure S1).

Authors response: We performed two corrections (i) sample cleaning, and (ii) spatial biases. Sampling cleaning refers to selecting a grid cell only once to avoid pseudo-replication due to high occurrence within that grid cell or the same animal being recorded more than once within the cell. This is common practice to avoid duplicate cells that inflate model performance while not adding explanatory power. This practice is common in SDM and any other statistical analyses that require independent (or near-independent) data as input. The second correction was the spatial biases because of open data, as addressed in the previous point. The Phillips et al. (2009) method was applied after the trimming to 1km cells being represented only once so it was not affected by this data quality control method.

I would like to see a map of the distribution of the points of other species used as background data, to have an idea of the spatial distribution of the survey effort.

Authors Response: We provide this in a new electronic supplementary material (S1 File).

Authors Response: We agree with your point and have acted carefully in our modelling approach to avoid predictions into no-sampled areas. As mentioned above all the island habitats were sampled and we used quality control techniques to avoid pseudo-replication of data and spatial biases. We included this information in the new version of the manuscript. In addition we created uncertainty in prediction maps (S7 File) based on the introduced range by overlapping n-1 models (n= number of presence records). This allows us to determine the consensus, i.e., how many times a given cell was predicted to be suitable (HSI>0.5) for both study species. These two approaches together should give the context of the results and the reliability of the predictions. This is also quite unique in many SDM studies that often do not produce such consensus map that depicts uncertainty.

****

Other methodological concerns are related to the small sample size of the data set used for the "introduced models":

Authors’ Response: Indeed the difference in sample sizes could affect how AIC makes model selection. Nonetheless these comparisons are done for each study area individually, i.e., comparing models with similar samples sizes with each other. Thus, we believe that this does not affect which variables would be selected and how similar are they between the island and the continental study area. We would argue they are two independent processes of variable selection that we then compare, having in mind the sampling size effects when making claims of the similarity of the variables for each study area.

Authors’ Response: Although we have a small dataset for the introduced area, some studies revealed that MaxEnt performs well in these cases (e.g., around 25 records as in our case; Hernandez et al., 2006; Bean et al., 2012; Proosdij et al., 2016). Further the data we use comes from sampling sessions that covered all the habitats in the island, reflecting the total existing data for the Azores, in fact currently there are no more available records in the Azores that comply with our selection criteria. However, the model predictions to remote or inaccessible areas were interpreted with caution (see Discussion section for more details).

Bean WT, Stafford R, Brashares JS. The effects of small sample size and sample bias on threshold selection and accuracy assessment of species distribution models. Ecography. 2012; 35(3): 250−258.

Hernandez PA, Graham CH, Master LL, Albert DL. The effect of sample size and species characteristics on performance of different species distribution modeling methods. Ecography. 2006; 29(5): 773−785.

Proosdij AS, Sosef MS, Wieringa JJ, Raes N. Minimum required number of specimen records to develop accurate species distribution models. Ecography. 2016; 39(6): 542−552.

*****

SPECIFIC COMMENTS:

Authors’ Response: We agree with your comment. We have included a new supplementary material (S6 File) to clarify this. In the example that you indicate, we made a mistake: we should be included a comma between (n=9) and + topographic, because are two different sets of variables. We have corrected this, thank you.

* Table 2: for the model "M. putorius, introduced": what is the difference between the two models since both include the same variables?

Authors’ Response: We have corrected this Table, including the referred comment.

Authors’ Response: We have simplified this text according to comments from both reviewers (Results - Model selection).

* line 281--290. It is not necessary to include the Akaike weights both in the table and in the text.

Authors’ Response: Thank you for the suggestion, we removed the Akaike weight from the text.

Authors’ Response: You are right. We have corrected Table 4 values and included in the manuscript the particular case of M. nivalis introduced-range HSI values.

REVIEWER #2

GENERAL COMMENTS:

Authors’ Response: Thank you for your suggestions and appreciation for our findings. We appreciate your highly constructive comments that helped us to improve the quality of the manuscript, particularly the Methods and Discussion section. Additionally, we have included new five supplementary files and eight new references to support these sections.

Authors’ Response: Thanks for the comment. We agree with you that occurrence records from open source data are often affected by sampling bias. Phillips et al. (2009) proposed to select background data exhibiting the same bias as the presence data. For example, if the presence data are taken of a determinate portion of the study area, then the background data should be taken from the same areas. We have included a new supplementary material describing how we did this background-data bias process (S1 File). Additionally, we re-wrote the sentence about records bias.

Authors’ Response: We have improved the Methods - Parameter configuration section, and included some relevant references (see the answers to the specific comments below).

Authors’ Response: We agree with your comment and, accordingly, we have revised and, whenever needed, modified the discussion to better emphasize the results from the different analytical approaches and by explaining more clearly our interpretation of the obtained results, especially the surprising ones.

SPECIFIC COMMENTS:

Ln 20: I would add “archipelago” after “Azores”, as this is a relevant biogeographic feature which may not be immediately obvious to all readers.

Authors’ Response: We added “archipelago” after “Azores".

Authors’ Response: We changed the sentence according to your suggestion.

Ln 32: It its not clear what “this” refers to.

Authors’ Response: We removed this sentence from the abstract in the new version of the manuscript.

Ln 62: The Azores are more than one island, so “island’s ” should be “islands’ ”

Authors’ Response: We changed "island’s" to "islands’"

Authors’ Response: We included the suggested reference in the manuscript.

Ln 69: “distributions” should be “distribution”.

Authors’ Response: We changed the text accordingly.

Ln 78: Overly long sentence. I would add a comma after “conservatism”.

Authors’ Response: We changed the text accordingly.

Ln111: Replace the final comma with a full stop and start a new sentence.

Authors’ Response: Done.

Ln 144: “deliberately” should be “deliberate”.

Authors’ Response: We have changed the text accordingly.

Ln 145: “or accidentally” is redundant in this sentence.

Authors’ Response: We removed "or accidentally" from the text.

Authors’ Response: We specified the country where we collected the data, according your comment. Also, we included a new supplementary material (S2 File) with all presence records used in the study. We detailed if the record was obtained from open source database or was collected by the authors.

Ln 164: How was this record accuracy assessed? Namely, which columns of the public databases were used for this filtering, and with which values?

Authors’ Response: Most of the records were obtained from the GBIF database. We downloaded the records as a .csv file from the GBIF platform. Then we filtered the accuracy of the record from "coordinateUncertaintyInMeters" column. The records with values equal or higher than 1000 meters were removed. Also, we removed the records without information in this column. We clarify this in the Methods section - Species presence data.

Authors’ Response: We have included in the new version this irregular facilitation/uploading of the occurrence records (Methods section - Species presence data).

Authors’ Response: We created a new Supplementary Material (S1 File), explaining how were selected the biased background points, according your comment.

Ln 169: I would start a new paragraph at “The introduced...”

Authors’ Response: We have changed the text accordingly.

Authors’ Response: You are right. We have selected only one record per 1km2 in both ranges. We have re-written the Methods - Species presence data Section to include this.

Ln 177 and 269: Remove “the” before “Europe”.

Authors’ Response: Done.

Authors’ Response: Thanks for your comment. We newly performed a normality test and detected a mistake in the correlation analysis because all variables are non-normally distributed; so we repeated the correlation analysis and consequently used the Spearman rank coefficient (see S4 File for the new results).

Ln 223-225: The literature has evolved quite a bit since this reference of 2006, and a few more papers on “best practices” are available nowadays – including Araújo et al. 2019, which is in the reference list although I could not find it cited in the text; and others such as Sofaer et al. 2019. Also, two important references on Maxent modelling in particular are Elith et al. 2010 and Merow et al. 2013. This should be used to choose appropriate parameters sensibly, rather than just testing limited arbitrary choices.

Authors’ Response: Thanks for this constructive comment. We have included the suggested references in the text.

Authors’ Response: We agree with you in that to test two percentages very similar is quite limited. For this reason we removed this aspect of the testing process.

Authors’ Response: We have selected around 30% of the total background points for each area (1,800 for Azores and 10,000 for Europe). Some authors recommend the use a large number of background points (e.g. 10,000); however, for our case we chose only 30% of these values and chose to run models 10 times, according Barbet-Massin et al. 2012. We used also this lower number of background points as a trade-off between the number of different combinations of models and the time for model runs.

Barbet‐Massin M, Jiguet F, Albert CH, Thuiller W. Selecting pseudo‐absences for species distribution models: how, where and how many?. Methods Ecol Evol. 2012; 3(2): 327−338.

Ln 230: What are the “total background points?”

Authors’ Response: We have removed this sentence in the new version of the manuscript.

Ln 234-235: The AUC is not the best metric (especially if used alone) to choose between models, as it has relevant known problems (e.g. Lobo et al. 2008). Also, the AUC assesses the capacity of models to distinguish between presence and absence records, whereas the data and modelling algorithm in this paper imply presence and background (not absence, nor pseudo-absence) records. Even the reference that the authors cite for “best practices” (Ln 224-25) says that “multiple evaluation measures are necessary to determine accuracy of models produced with presence-only data”.

Authors’ Response: We used AUC only to test the different parameter configuration performance (see Methods - Parameter configuration). Then, when parameter configuration was selected, we performed model selection using AICc (see Methods - Model selection).

Authors’ Response: You are right. We have corrected the text accordingly.

Authors’ Response: We re-wrote the sentence. Yes, this configuration was used for all four models tested.

Ln 271-272: This is only a “validation” if the same field data were not used in model building.

Authors’ Response: We have used a subset of presence data randomly selected to perform the validation. We have clarified this in the main text.

Ln 285-290: Here it is not clear what these model “weights” refer to, and why a weight of 0.86 is better than a weight of 0.99.

Authors’ Response: We have modified the text in the new version of the manuscript (Results - Model selection).

Ln 291: “probability of […] occurrence” is not provided by a presence-background model such as Maxent. Do you mean “suitability for […] occurrence”?

Authors’ Response: We have replaced "probability" by "suitability".

Authors’ Response: Thanks for your suggestion. We have included a Chi2 test to evaluate if the HSI values were significantly different than expected by chance, as suggested.

Ln 367: Add “more” before “widespread”.

Authors’ Response: Done.

Figures 2 and 3: I’d suggest switching to a colourblind-friendly colour scale, as red and green are indistinguishable for a significant fraction of the potential readers.

Authors’ Response: Thanks for your suggestion. We have changed the scale colour of the Figures 2 and 3.

PLoS One. doi: 10.1371/journal.pone.0237216.r003

Decision Letter 1

Ulrike Gertrud Munderloh

17 Apr 2020

Your manuscript has the potential to provide useful information. However, it is important that you carefully read and respond to the concerns and detailed explanations from Reviewer 1.

Dear Mr Lamelas-López,

We would appreciate receiving your revised manuscript by Jun 01 2020 11:59PM. When you are ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter.

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). This letter should be uploaded as separate file and labeled 'Response to Reviewers'.
A marked-up copy of your manuscript that highlights changes made to the original version. This file should be uploaded as separate file and labeled 'Revised Manuscript with Track Changes'.
An unmarked version of your revised paper without tracked changes. This file should be uploaded as separate file and labeled 'Manuscript'.

We look forward to receiving your revised manuscript.

Kind regards,

Ulrike Gertrud Munderloh, Ph.D.

Academic Editor

PLOS ONE

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #1: (No Response)

Reviewer #2: All comments have been addressed

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

Reviewer #1: No

Reviewer #2: Yes

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: No

Reviewer #2: Yes

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

Reviewer #1: Yes

Reviewer #2: Yes

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

Reviewer #1: Yes

Reviewer #2: Yes

**********

6. Review Comments to the Author

Reviewer #1: This paper is a review of a previous paper submitted to PLOS One. Its aim is to model habitat suitability in the Azores archipelago for two introduced mustelid species, Mustela nivalis and Mustela putorius. For each species, the authors fit two models: (i) one model fitted to a dataset of detections collected over the whole Europe (native range), and (ii) one model fitted to a dataset of detections collected only in the Azores islands (introduced range). For each species, the authors then compare the maps predicted by the two models (introduced and native), and show that the maps obtained with the two datasets are different. The map predicted using data collected over the whole Europe does not correspond to the map predicted using data collected in the Azores. The authors conclude that this difference indicates that species distribution models should be carefully interpreted when used to predict the range of a species in an area based on data collected in another area.

In my review of the previous manuscript, I expressed several major concerns related to the methods used to reach the aim, as well as several minor comments. Many of my comments have been satisfactorily taken into account. However, I think that we disagree on my main comment: I think that the comparison between native and introduced models carried out in this paper does not allow to draw conclusions on the ability to predict introduced range from the native range, because of the very different scales of study for the two areas.

It is possible that I was not clear enough, so that I explain more clearly below why I think that the metholodogy in this paper is very problematic. Moreover, the more precise description that the authors now give of the methodology used to correct for sampling bias raises another major methodological problem.

I describe these two problems below.

******

** On the concept of scale.

In my previous review, my main criticism was related to the difference of geographical scales between the model fitted in the native range and the model fitted in the Azores archipelago. I expressed it in my previous review by noting that results may vary strongly between scales, and comparing a model fitted using data collected over continental scale and a model collected on a higly local scale is meaningless, as the two models are necessarily describing very different processes, so that they are necessarily returning different results.

To this criticism, the authors replied: "we believe this effect is not affecting the model outputs because we used the same spatial resolution and the same data sources for both study areas."

Actually, the concept of scale in Ecology is more general than just the resolution of the study (Dungan et al. 2002): the concept of scale involves many aspects (grain, lag, support, etc.). But in this study, the most problematic aspect is the extent of the study area, rather than its resolution. Dungan et al. (2002) illustrates clearly how the results of a study might change strongly when the scale changes. Another reference here -- a seminal one actually -- is Johnson (1980), who defines in the discussion four orders of habitat selection (first order: geographical range, second order: distribution of animals in a region, third order: distribution of animal locations within their home ranges, fourth order: selection of food items at the locations). The processes and preferences at a scale are not necessarily the same than those at another. In the present study, the authors compare first order selection (whole Europe) with second order selection (distribution within the islands). Many other authors have stressed the importance of this scale and variability of the results according to the scale considered (e.g. Pearce and Boyce 2006, Soberon and Peterson 2005).

More concretely, here, any analysis of environmental variables classically used for SDM on the continental scale in Europe will mainly show large environmental patterns. For example, on a continental scale, the effect of elevation on the presence of a species will be driven by the differences between mountainous areas (the Alps) and other areas. That is, by the differences of weather, climate, vegetation, snow cover, etc. between Alpine and non-Alpine climates. Of course, other climatic variables will also account for it to some extent, but no variable will synthetize the difference between non-Alpine and Alpine climate as efficiently as the elevation (by definition !). Therefore, any effect of the elevation in SDM models at this scale will summarize the difference of environments (vegetation, climate, etc.) between Alpine and non-Alpine environments. Therefore, if -- for example -- a species is absent or rarer in the Alps than in the rest of the continent, it will be because the Alpine climate is less suitable for the species than non-Alpine climates (whatever the reason, e.g. too much snow in winter, rocky soils, etc.). This will be synthetized by a negative effect of the elevation in the SDM.

On the other hand, the elevation, in the Azores archipelago, has a very different meaning. In these islands, high elevation areas are not characterized by a "more Alpine climate" than other areas. It has a very different meaning. In these islands, for example, there is probably a very strong correlation between elevation and the distance to the sea. So that elevation here rather represents this distance (e.g. further from the sea = less urban, etc. -- maybe a better habitat for the species?). Thus the meaning of elevation for the species is not the same, just because we do not work at the same scale. Of course, my example here describes a fictious species (I do not know what are the selection patterns for weasel and ferret at this scale), but it illustrates why results are expected to vary between two scales, even within the native range (e.g. the results will not be the same if we compare a model fitted on the whole Europe and an model fitted e.g. on similar sized areas in continental Portugal).

The same is true for e.g. climatic variables. The effect of these variables on a continental scale will likely reflect the differences in densities between the different climates in Europe. If a species is less abundant in e.g. areas with continental climates (with less precipitation), then the effect of clim_bio12 in a SDM on a continental scale will mainly represent this large scale pattern. At the scale of the Azores Islands, this variable will not have the same meaning, as there is no continental climate in the Azores islands. Etc.

If the aim is to predict the introduced range from data collected in the native range, it would make much more sense to compare this archipelago with areas *of similar sizes* located in similar climates, with a similar elevation range, etc.

I suggested this approach in my previous review, but the authors replied:

"We concluded that it would be difficult to justify selecting sub areas within continental Europe that would match the size of the Azores to conduct the analysis as suggested. This is because there could be an even bigger effect due to this selection than the potential effects of modeling a smaller area using information from a larger area."

I do not understand why it would be difficult to justify it. I do not think that there would be a bigger effect in this selection. The authors have a very precise description of their study area according to numerous environmental variables (climate, elevation, etc.). It would be easy to select randomly numerous places of similar sizes in Europe. The standardized environmental variables (i.e. minus the mean and divided by the standard deviation) then each define a dimension in a multidimensional space (one dimension is the average temperature, one is the elevation, on is the precipitation, etc.). Each one of these random areas would define a point in this space. The Azores archipelago also defines a point in this space. Then, we can select a sample of -- say -- 10 random places with the smallest Euclidean distance in this multidimensional space to Azores. The resulting sample will be a random and objective sample of by areas taken in the native range with sizes and environmental conditions similar to the target area. This would be a better approach in my opinion: if the aim is to infer the effect of a factor on a process by comparing two sets of areas -- one with the factor and one without -- it is better to design the study so that only that factor varies and the other variables are identical. Here, to try to find one or several study areas in the native range with sizes and conditions similar to those of the Azores archipelago, with only "introduced/native" differing. On the other hand, as exemplified above, ignoring the effect of scale will lead to erroneous conclusions.

The authors further note:

"Further, we believe that this suggestion is not in line with other approaches to model invasive species ranges. Several authors have also modeled invasive species range from information retrieved from the native range (e.g., Broennimann et al., 2007; Fitzpatrick et al., 2007; Rödder et al., 2009; Bidinger et al., 2012); This is because invasive species are yet to be in equilibrium with their environment in the invaded ranges thus becoming even more important to include a broader set of parameter ranges from the native range to predict areas with potential for invasion."

However, none of these works compare areas with so large size differences. Broenniman et al. (2007) predict the distribution of spotted knapweed in North America with a model fitted with data collected in its native range in western Europe, two areas covering similar sizes. Fitzpatrick et al. (2007) compare the distribution of red ants in tropical south America and tropical north America, again at similar scales and in similar climates. Rödder et al. (2009) study the slider turtle in their native range (North America), and the invasive range is also on a continental scale. Finally, Bidinger et al. (2012) studies the distribution of Harlequin ladybird in their native range in eastern Asia, and in their introduced range (area of similar size in Europe).

I fully endorse the aim of the present study. From a conservation as well as ecological point of view, it is essential to find a way to predict the introduced range from the native range of invasive species. I agree on the importance of understanding how the introduced range might differ from the native range to identify how a species adapt to a new environment. I do not disagree on the aim, but on the method. The problem of scale is not a minor one in ecology. Comparing two areas of very different sizes amounts to compare a process at two very different scales. Therefore, different results are expected, even if there is no difference between native and introduced ranges.

******

** On the correction of sampling effort.

The authors rightly explained that the biased data collection characterizing the dataset would lead to to biased inference if it was not taken into account. They use the "target group" method of Philips et al. (2009) to collect a biased sample of background points exhibiting the same bias as the presence records. The authors did not describe clearly their approach in the previous version of the manuscript. It is now more clearly described.

The "target group" approach of Philips et al. aims at distinguishing whether the absence of detection of the focus species at one place is caused by the absence of the species itself or by the absence of data collection. The idea is to define a "target group" containing many species for which we think that the collection or observation activities of collectors is similar to those of the focus species. For example, to model the SDM of a particular bird species in a citizen science program, it would make sense to use the whole set of bird species studied in the citizen science program as the target group used to select background points. The hope is that, at any point where data collection for the focus species has occurred, at least one species of the target group was present and reported by the same data collection (even if the focus species itself is not). So that the presence of a species of the target group and the absence of the focus species ensures that the absence of reported detection for this focus species is likely due to the actual absence of the species.

This method allows to correct -- to some extent, it is impossible to completely account for all the bias in such contexts -- the collection bias because the target group is usually made of a large number of species, and at least one is supposed to be present where data collection has occurred. In the paper of Philips, for example, the target group contains from 7 to 52 species.

In the present paper, the target group was defined by just the ferret and the weasel. This is a small target group! It means that any area unsuitable for both species, but where data collection has actually occurred will not be present in the data. The background sample will be defined by the set of habitat conditions allowing the presence of the weasel and/or the ferret (well, a biased sample of it, but this is the aim of the target group method to obtain such a biased sample of background points). So that modelling the presence of the weasel (resp. ferret) in a set of locations defined so that either the ferret or the weasel are present, is not a model of the species distribution. It is a model of the niche difference between the two species: the model predicts the probability of presence of one species given that at least one of the two species is present.

So that the "native" model describes the niche differences between ferret and weasel modelled at the scale of the continent species, whereas the "introduced" model describes theses differences at the scale of the archipelago. In other words, the models do not focus on the potential distribution of the species as indicated in the paper, but on the differences between the two species. Which may be of interest, but is not the actual aim of the study.

******

** References

Dungan, J.; Perry, J.; Dale, M.; Legendre, P.; Citron-Pousty, S.; Fortin, M.; Jakomulska, A.; Miriti, M. & Rosenberg, M. 2002. A balanced view of scale in spatial statistical analysis. Ecography 25, 626-640

Johnson, D. 1980. The comparison of usage and availability measurements for evaluating resource preference. Ecology 61, 65-71

Levin, S. 1992. The problem of pattern and scale in Ecology. Ecology 73, 1943-1967

Soberon, J. & Peterson, A. 2005. Interpretation of models of fundamental ecological niches and species' distributional areas. Biodiversity Informatics 2, 1-10

Pearce, J. & Boyce, M. 2006. Modelling distribution and abundance with presence-only data. Journal of Applied Ecology 43, 405-412

Elith, J. & Leathwick, J. R. 2009. Species distribution models: ecological explanation and prediction across space and time. Annual Review of Ecology, Evolution, and Systematics 40, 677

Reviewer #2: The authors have addressed the concerns expressed in the previous review. The explanation and map provided in Appendix S1 make the procedure much clearer and more transparent, and the dataset provided in appendix S2 adds significant value and usefulness to the manuscript. All other supplementary materials also help make the methodology clearer. Although not everything was done the way I would have done it myself, I am generally satisfied with the current version of the manuscript and I believe the authors have appropriately explained and defended their work.

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

PLoS One. 2020 Aug 7;15(8):e0237216. doi: 10.1371/journal.pone.0237216.r004

Author response to Decision Letter 1

6 Jul 2020

Response to Reviewers

Reviewer #1:

This paper is a review of a previous paper submitted to PLOS One. Its aim is to model habitat suitability in the Azores archipelago for two introduced mustelid species, Mustela nivalis and Mustela putorius. For each species, the authors fit two models: (i) one model fitted to a dataset of detections collected over the whole Europe (native range), and (ii) one model fitted to a dataset of detections collected only in the Azores islands (introduced range). For each species, the authors then compare the maps predicted by the two models (introduced and native), and show that the maps obtained with the two datasets are different. The map predicted using data collected over the whole Europe does not correspond to the map predicted using data collected in the Azores. The authors conclude that this difference indicates that species distribution models should be carefully interpreted when used to predict the range of a species in an area based on data collected in another area.

It is possible that I was not clear enough, so that I explain more clearly below why I think that the methodology in this paper is very problematic. Moreover, the more precise description that the authors now give of the methodology used to correct for sampling bias raises another major methodological problem.

I describe these two problems below.

Authors: We are grateful for your detailed review. You have raised two major concerns regarding the methodology: (i) scale and (ii) bias correction. In summary we believe these concerns are not fully warranted because of the way we designed our study, the comparison between islands and continents, the predictor variables selected, and the methods chosen. In general, we followed methodologies, software and analysis described as best practices used in SDM (Araujo et al. 2019).

Indeed, we agree that SDM transferability should be better studied, which is perhaps extremely important when using these tools for, for example, studies that project range expansions or considerable range contractions – changing dramatically spatial extent of predictor beyond the information range, and invasive species. One way to ameliorate the spatial extent effect could be by having similar sampling intensities, or iteratively transfer models across spatial locations or points in time. This would make for very nice studies, but much beyond our simple analysis. Below we provide much more detail to our reasoning for the issue of scale and bias correction.

Araújo MB, Anderson RP, Barbosa AM, Beale CM, Dormann, C. F., Early, R., et al. Standards for distribution models in biodiversity assessments. Sci Adv. 2019; 5(1): eaat4858.

******

** On the concept of scale.

In my previous review, my main criticism was related to the difference of geographical scales between the model fitted in the native range and the model fitted in the Azores archipelago. I expressed it in my previous review by noting that results may vary strongly between scales, and comparing a model fitted using data collected over continental scale and a model collected on a highly local scale is meaningless, as the two models are necessarily describing very different processes, so that they are necessarily returning different results.

Authors: Thank you for your comment, this is indeed a major aspect to consider in any SDM study. Soberon (2007) illustrated that species respond to environmental conditions like habitat type, food resources, etc at local scales, and to environmental conditions like bioclimate and topography at larger scales and that in order to fully understand factors that drive species ranges we would need to derive nested models at these scales. Unfortunately, the literature on SDM has yet to get to this stage but many examples are now emerging that combine both bioclimatic and habitat variables and show that there is an interplay of the importance of the two (Santos et al. 2017). This is further complicated when dealing with invasive species which likely change both associations with habitat and bioclimatic parameters in the invaded range – that is what makes them invasive in the first place. Further, as for our case, there is a complication because we are studying islands. The spatial extent of the invaded area is spatially confined because these are islands, and therefore any conversion of this spatial extent to a continuous continental surface would be arbitrary, and add spurious edge effects for the continental models which would not be present in the islands. To account for all these aspects we carefully designed our study so that:

1. We included the same predictor variables in both the islands and the continental area. These include both bioclimatic variables – to represent processes at large scales, and habitat variables – to represent processes at local scales. In the text: ”We selected (1) two topographic variables (e.g., slope) in order to describe the physical environment; (2) four climatic variables (e.g., annual temperature, precipitation seasonality) to describe the bioclimatic conditions; (3) a set of seven landscape level variables, including habitat type, vegetation structure (herbs, shrubs or trees), land use types, and landscape heterogeneity (e.g., number of patches); and (4) three human disturbance variables, related with human population (e.g., population density) and their activities (% of artificial areas, % of agricultural areas).”

2. We collected the same variables in the islands and the continent. To represent the continent, we did not analyze any continuous surface, but rather chose to select 10,000 isolated cells, randomly selected from whole Europe.

3. We experimented with the scenario you proposed to restrict the native-area to a small portion of the native range of both species that would match the spatial extent of the Azores. We were confronted with several choices that we were not comfortable making – (i) where to seed the area? (ii) Which shape should it take? (iii) are we inducing arbitrarious edges? (iv) if we created 2,250 km2 areas, how do we then justify the differences between the models? (v) how do we control for very uneven sample size within the Azores, like continental areas? Additionally, the particular environmental conditions of the Macaronesian archipelagos, and particularly of the Azores, are not "comparable" with a continental area with a similar extent. Therefore, we opted to sample from the native-range and used 10,000 isolated cells.

4. We are unaware of where do the species invade from, so this would also limit selecting certain areas within the European range.

5. Of course it is possible that mismatched spatial extents of study areas affect model transferability, however this is an aspect seldom analyzed in SDM studies. Transferability studies are few and far in between, and there are yet no clear guidelines on how to do them properly because indeed we are transferring models developed with a given range of a variable to areas where the range of values the variable can take is different – this is the case for invasive species. Further studies should do an experiment on this topics.

Soberón J. Grinnellian and Eltonian niches and geographic distributions of species. Ecol Lett. 2007; 10(12): 1115−1123.

Santos MJ, Smith AB, Thorne JH. et al. The relative influence of change in habitat and climate on elevation range limits in small mammals in Yosemite National Park, California, U.S.A. Clim Chang Responses. 2017; 4: 7.

To this criticism, the authors replied: "we believe this effect is not affecting the model outputs because we used the same spatial resolution and the same data sources for both study areas."

Actually, the concept of scale in Ecology is more general than just the resolution of the study (Dungan et al. 2002): the concept of scale involves many aspects (grain, lag, support, etc.).

Authors: We agree and see above our comment to this. Indeed what we are talking about is the extent of the study area.

But in this study, the most problematic aspect is the extent of the study area, rather than its resolution. Dungan et al. (2002) illustrates clearly how the results of a study might change strongly when the scale changes. Another reference here -- a seminal one actually -- is Johnson (1980), who defines in the discussion four orders of habitat selection (first order: geographical range, second order: distribution of animals in a region, third order: distribution of animal locations within their home ranges, fourth order: selection of food items at the locations). The processes and preferences at a scale are not necessarily the same than those at another. In the present study, the authors compare first order selection (whole Europe) with second order selection (distribution within the islands). Many other authors have stressed the importance of this scale and variability of the results according to the scale considered (e.g. Pearce and Boyce 2006, Soberon and Peterson 2005).

Authors: Thank you for your comment, this is indeed a major aspect to consider in any SDM study. As mentioned above, we took the approach from Soberon (2007), which illustrated that species respond to environmental conditions like habitat type, food resources, etc at local scales, and to environmental conditions like bioclimate and topography at larger scales and that in order to fully understand factors that drive species ranges we would need to derive nested models at these scales. Unfortunately, the literature on SDM has yet to get to this stage but many examples are now emerging that combine both bioclimatic and habitat variables and show that there is an interplay of the importance of the two (Santos et al. 2017). This is further complicated when dealing with invasive species which likely change both associations with habitat and bioclimatic parameters in the invaded range – that is what makes them invasive in the first place.

More specifically, our selected native area is not strictly a continental area, as we are sampling 10,000 isolated cells within Europe (Fig R1 in "Response to Reviewers" Letter; see ESM1 for more details). Therefore, the extent (i.e. the total area) used during the modeling process is similar between regions (extent of introduced range: 2,250 background points, extent of native range: 3,000 background points). As the native area is formed by isolated cells obtained from a continental area, the scale is not necessarily a problem. To address Johnson (1980) orders of habitat selection, we chose variables that represent processes at each of the orders: (1) First order: geographical range – we use bioclimatic variables; (2) Second order: distribution of animals in a region – we use habitat and topographic variables; (3) Third order: distribution of animal locations within their home ranges – vegetation structure (herbs, shrubs or trees), human presence; (4) Fourth order: selection of food items at the locations – vegetation structure.

Soberón J. Grinnellian and Eltonian niches and geographic distributions of species. Ecol Lett. 2007; 10(12): 1115−1123.

Ferrier S, Watson G, Pearce J, Drielsma M. Extended statistical approaches to modelling spatial pattern in biodiversity in northeast New South Wales. I. Species-level modelling. Biodivers Conserv. 2002; 11(12): 2275−2307.

Authors: Indeed, interpretation of the effect of elevation is different for the Alps and the Azores, but we do not believe this is linked to scale but rather we are talking about the second order from Johnson (1980) and the local conditions in Soberon’s work or the geographic factors that determine species distributions in mountains from Grinnell or Humboldt. Indeed, we are dealing with mountains in islands and not continental mountains. We carefully interpret the coefficients (direction and magnitude) in light of the context in which they occur. Often the proxy elevation is used to represent the variety of conditions that occur with zonations in mountains (Schrodt et al. 2019). The specific mechanisms by which individuals and species respond to these are seldom not represented in SDMs. This gets back to the use of SDM to predict distribution, habitat selection, or individual movements – and in the ecological literature for each of these processes there are different types of models. SDM, as the name states, refers to geographical distributions and some authors even argue they should not be used to infer habitat use or animal movements as inferred from Johnson (1980) orders of habitat selection across scales.

Schrodt F, Santos MJ, Bailey JJ, Field R. Challenges and opportunities for biogeography—What can we still learn from von Humboldt?. J Biogeogr. 2019; 46(8): 1631−1642.

Authors: Indeed, interpretation of the effect of any variable is different due to the context and in the discussion we state how we interpret these effects. This is true for any SDM study.

Authors: Thank you for the suggestion. As mentioned above we experimented with it, and our conclusion was that the criteria to select areas of similar extent to that of the Azores, located in similar climates and with a similar elevation range would be extremely arbitrary. First, selecting random 2,250 km2 areas in a continent that is continuous would suffer from many of the issues you mention above as to why we should do it. Second, there are no similar climates to the Azores in continental Europe, and third, as in your explanation, the Alps and the mountains of the Azores are different in the way they affect species distributions therefore not warranting making these choices. Therefore we chose to represent our native area by non-connected cells, i.e., sampling on the native range.

I suggested this approach in my previous review, but the authors replied:

Authors: Thanks for your suggestion. Although we respectfully disagree, we tried to perform the suggested analysis. First, we created a polygon with similar extension to the Azores (2,250 km2; 75x30 cells). After that, we randomly generated 100 polygons across whole Europe (see Fig R2 in "Response to Reviewers" Letter). The polygons were generated in areas with presence data (i.e. all polygons possess occurrence-records of the target species). We calculated the mean value of the bioclimatic variables for each polygon. Means of the bioclimatic variables were also calculated for the Azores. Finally, we performed a PCA analysis and selected the polygons with the smallest Euclidean distance in this multidimensional space to Azores (see Fig R3 in "Response to Reviewers" Letter).

However, due to the large extension of Europe and the disperse spatial location of the occurrence records, the polygon with most occurrence data had only 30 records (our native area, i.e., Europe, has more than 1,500 occurrence records).

The results of the PCA showed that the polygons with more similar conditions to the Azores had less than 10 occurrence records (polygons ID 10 and 9, see Fig R3). These polygons were located in the occidental area of the Iberian peninsula, near to the coast line.

In conclusion, although with the analysis suggested by the reviewer the extent of the study was similar to the Azores (2,250 km2 vs 2,300 km2), in our study the extent of the native area used to perform the models was also similar (3,000; i.e.10,000 total background points, from which MaxEnt randomly selected 30%). However, the number of presence data used to run the models is dramatically lower using 75x30 polygons in comparison with our selected native-area.

The authors further note:

Authors: At the moment we have conducted three analysis:

1. Analysis with 10,000 cell across Europe and the Azores ─ in the original manuscript;

2. Additional analysis of the occidental part of the Iberian peninsula as the native-area (comprising mainly the north-west of Spain and Portugal) and the Azores. (not in the manuscript). We selected this area because: (1) it comprises a large area to where both species are native, thus including most relevant ecological requirements of the study species; and (2) it is the closest native distribution area for the study species. Now, the results of the PCA (Fig R3) showed that this native area has similar environmental conditions to the Azores. We received criticism for this approach because it is unknown where the original introduced individuals originated from, as they easily could have originated from another region in their large native range. By restricting the study to a small portion of the native range of both species the authors are not capturing the full range of environments these species can inhabit and thus will impact the model predictions and comparisons with the invaded range model;

3. We attempted the proposed approach of generating 2,250 km2 areas and select those with closer climatic properties to the Azores, but as explained above we ran into limitations of sample sizes

So we are facing a trade-off, of restricting analyses to randomly selected areas of the same size of the Azores (as mentioned above, a bit difficult to justify their choice) and have small samples, or run models sampling from the native range in a similar area as the Azores (10,000 cells is a similar number of cells to those used for the Azores models). Therefore, selecting a smaller area, the number of native records would be dramatically lower and would not reflect the native range. This is the reason why other reply was "we believe that this suggestion is not in line with other approaches to model invasive species ranges [...] This is because invasive species are yet to be in equilibrium with their environment in the invaded ranges thus becoming even more important to include a broader set of parameter ranges from the native range to predict areas with potential for invasion."

******

** On the correction of sampling effort.

The authors rightly explained that the biased data collection characterizing the dataset would lead to biased inference if it was not taken into account. They use the "target group" method of Philips et al. (2009) to collect a biased sample of background points exhibiting the same bias as the presence records. The authors did not describe clearly their approach in the previous version of the manuscript. It is now more clearly described.

Authors: We did not use exactly the target-group method proposed by Phillips et al. 2009. We only based our biased background data on the reasoning of Phillips et al. 2009 about the necessity to account for bias in the selection of the background points according to presence data, but we did not considered a target-group species to select presence-absence data. We used the MaxEnt software to perform the models, which takes as input presence-only data and a set of environmental variables across a user-defined area, in our case, 10,000 isolated background points across whole Europe.

We clarified this in the new version of the ESM1 File. Now it reads: "Phillips et al. 2009, proposed to select background data that exhibits the same bias as the presence data. For example, if the presence data are taken from a determined portion of the study area, then the background data should be taken from the same areas (Ferrier et al., 2002; Phillips et al., 2009). Following this reasoning, we (i) created a grid comprising all native area (i.e., Europe), (ii) randomly selected 10,000 cells within <10 km of the species presence cells, and (iii) used 3,000 random background points from this selection to run the native-based models in MaxEnt."

******

** References

Johnson, D. 1980. The comparison of usage and availability measurements for evaluating resource preference. Ecology 61, 65-71

Levin, S. 1992. The problem of pattern and scale in Ecology. Ecology 73, 1943-1967

Soberon, J. & Peterson, A. 2005. Interpretation of models of fundamental ecological niches and species' distributional areas. Biodiversity Informatics 2, 1-10

Pearce, J. & Boyce, M. 2006. Modelling distribution and abundance with presence-only data. Journal of Applied Ecology 43, 405-412

Elith, J. & Leathwick, J. R. 2009. Species distribution models: ecological explanation and prediction across space and time. Annual Review of Ecology, Evolution, and Systematics 40, 677

Reviewer #2:

The authors have addressed the concerns expressed in the previous review. The explanation and map provided in Appendix S1 make the procedure much clearer and more transparent, and the dataset provided in appendix S2 adds significant value and usefulness to the manuscript. All other supplementary materials also help make the methodology clearer. Although not everything was done the way I would have done it myself, I am generally satisfied with the current version of the manuscript and I believe the authors have appropriately explained and defended their work.

Authors: Thanks for your comment. We are pleased that you are satisfied with the manuscript.

Attachment

Submitted filename: Response to Reviewers.docx

Click here for additional data file.^{(755.4KB, docx)}

PLoS One. doi: 10.1371/journal.pone.0237216.r005

Decision Letter 2

Ulrike Gertrud Munderloh

23 Jul 2020

Modelling the distribution of Mustela nivalis and M. putorius in the Azores archipelago based on native and introduced ranges

PONE-D-19-33731R2

Dear Dr. Lamelas-López,

We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at http://www.editorialmanager.com/pone/, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

Kind regards,

Ulrike Gertrud Munderloh, Ph.D.

Academic Editor

PLOS ONE

Additional Editor Comments (optional):

Reviewers' comments:

PLoS One. doi: 10.1371/journal.pone.0237216.r006

Acceptance letter

Ulrike Gertrud Munderloh

27 Jul 2020

PONE-D-19-33731R2

Modelling the distribution of Mustela nivalis and M. putorius in the Azores archipelago based on native and introduced ranges

Dear Dr. Lamelas-López:

I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

If we can help with anything else, please email us at plosone@plos.org.

Thank you for submitting your work to PLOS ONE and supporting open access.

Kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Dr. Ulrike Gertrud Munderloh

Academic Editor

PLOS ONE

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

S1 File. Selection of background points selected to model the native area conditions.

(DOCX)

Click here for additional data file.^{(522.5KB, docx)}

S2 File. Occurrence records of Mustela nivalis and M. putorius.

(DOCX)

Click here for additional data file.^{(436.9KB, docx)}

S3 File. Detailed explanation about obtained environmental data.

(DOCX)

Click here for additional data file.^{(12.2KB, docx)}

S4 File. Correlation analysis between variables in native and introduced ranges.

(DOCX)

Click here for additional data file.^{(25KB, docx)}

S5 File. Figure detailing the stepwise modeling process.

(DOCX)

Click here for additional data file.^{(82.6KB, docx)}

S6 File. Combination of sets of candidate predictor variables.

(DOCX)

Click here for additional data file.^{(13.2KB, docx)}

S7 File. Maps of uncertainty in predictions.

(DOCX)

Click here for additional data file.^{(98.7KB, docx)}

Attachment

Submitted filename: Response to Reviewers.docx

Click here for additional data file.^{(755.4KB, docx)}

Data Availability Statement

All relevant data are within the manuscript and its Supporting Information files.

[pone.0237216.ref001] 1. Long JL. Introduced mammals of the world: their history, distribution and influence. Csiro Publishing; Collingwood, Victoria, Australia; 2003. [Google Scholar]

[pone.0237216.ref002] 2.Mack RN, Simberloff D, Mark Lonsdale W, Evans H, Clout M, Bazzaz FA. Biotic invasions: causes, epidemiology, global consequences, and control. Ecol Appl. 2000; 10(3): 689–710. [Google Scholar]

[pone.0237216.ref003] 3.Seebens H, Blackburn TM, Dyer EE, Genovesi P, Hulme PE, Jeschke JM, et al. Global rise in emerging alien species results from increased accessibility of new source pools. PNAS. 2018; 201719429. [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0237216.ref004] 4.Clout MN, Russell JC. The invasion ecology of mammals: a global perspective. Wildlife Res. 2008; 35(3): 180–184. [Google Scholar]

[pone.0237216.ref005] 5.Ceballos G, Brown JH. Global patterns of mammalian diversity, endemism, and endangerment. Conserv Biol. 1995; 9(3): 559–568. [Google Scholar]

[pone.0237216.ref006] 6.Courchamp F, Chapuis JL, Pascal M. Mammal invaders on islands: impact, control and control impact. Biol Rev. 2003; 78(3): 347–383. 10.1017/s1464793102006061 [DOI] [PubMed] [Google Scholar]

[pone.0237216.ref007] 7.Towns DR. Invasive mammal eradications: the value of an ecosystem view. Anim Conserv. 2011; 14(5): 466–467. [Google Scholar]

[pone.0237216.ref008] 8.Bellard C, Cassey P, Blackburn TM. Alien species as a driver of recent extinctions. Biol Lett. 2016; 12(2): 20150623 10.1098/rsbl.2015.0623 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0237216.ref009] 9.Doherty TS, Glen AS, Nimmo DG, Ritchie EG, Dickman CR. Invasive predators and global biodiversity loss. PNAS. 2016; 113: 11261–11265. 10.1073/pnas.1602480113 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0237216.ref010] 10.Wyatt KB, Campos PF, Gilbert MTP, Kolokotronis SO, Hynes WH, DeSalle R, et al. Historical mammal extinction on Christmas Island (Indian Ocean) correlates with introduced infectious disease. PLoS One. 2008; 3(11): e3602 10.1371/journal.pone.0003602 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0237216.ref011] 11.Vitousek PM. Diversity and biological invasions of oceanic islands. In: Wilson E.O., Peter F.M. (Eds.) Biodiversity. National Academy of Sciences, Washington, DC; 1988. [Google Scholar]

[pone.0237216.ref012] 12.Bicknell TW, Reid JB, Votier SC. Probable predation of Leach’s Storm‐petrel Oceanodroma leucorhoa eggs by St Kilda Field Mice Apodemus sylvaticus hirtensis. Bird Study. 2009; 56(3): 419–422. [Google Scholar]

[pone.0237216.ref013] 13.Frutuoso G. Saudades da terra. Instituto Cultural de Ponta Delgada, Ponta Delgada. Portugal. 1561.

[pone.0237216.ref014] 14.Mathias MDL, Ramalhinho MG, Santos-Reis M, Petrucci-Fonseca F, Libois R, Fons R, et al. Mammals from the Azores islands (Portugal): an updated overview. Mammalia. 1998; 62(3): 397–408. [Google Scholar]

[pone.0237216.ref015] 15.Borges PAV, Costa A, Cunha R, Gabriel R, Gonçalves V, Martins AF, et al. A list of the terrestrial and marine biota from the Azores. Princípia, Cascais, Azores; 2010. [Google Scholar]

[pone.0237216.ref016] 16.Bencatel J, Sabino-Marques H, Álvares F, Moura AE, Barbosa AM. Atlas de Mamíferos de Portugal. 2ª Edition. University of Évora, Évora; 2019. [Google Scholar]

[pone.0237216.ref017] 17.Monteiro LR, Ramos JA, Furness RW. Past and present status and conservation of the seabirds breeding in the Azores archipelago. Biol Conserv. 1996; 78(3): 319−328. [Google Scholar]

[pone.0237216.ref018] 18.Bried J, Magalhães MC, Bolton M, Neves VC, Bell E, Pereira JC, et al. Seabird habitat restoration on Praia Islet, Azores archipelago. Ecol Restor. 2009; 27(1): 27–36. [Google Scholar]

[pone.0237216.ref019] 19.Neves VC, Panagiotakopoulos S, Ratcliffe N. Predation on roseate tern eggs by European starlings in the Azores. Arquipelago–Life and Marine Sciences. 2011; 28: 15−23. [Google Scholar]

[pone.0237216.ref020] 20.Hervías S, Henriques A, Oliveira N, Pipa T, Cowen H, Ramos JA, et al. Studying the effects of multiple invasive mammals on Cory’s shearwater nest survival. Biol Invasions. 2013; 15(1): 143–15521. [Google Scholar]

[pone.0237216.ref021] 21.Hervías S, Oppel S, Medina FM, Pipa T, Díez A, Ramos JA, et al. Assessing the impact of introduced cats on island biodiversity by combining dietary and movement analysis. J Zool. 2014; 292(1): 39–47. [Google Scholar]

[pone.0237216.ref022] 22.Rando JC, Alcover JA, Olson SL, Pieper H. A new species of extinct scops owl (Aves: Strigiformes: Strigidae: Otus) from São Miguel Island (Azores Archipelago, North Atlantic Ocean). Zootaxa. 2013; 3647(2): 343–357. [DOI] [PubMed] [Google Scholar]

[pone.0237216.ref023] 23.Alcover JA, Pieper H, Pereira F, Rando JC. Five new extinct species of rails (Aves: Gruiformes: Rallidae) from the Macaronesian Islands (North Atlantic Ocean). Zootaxa. 2015; 4057(2): 151−190. 10.11646/zootaxa.4057.2.1 [DOI] [PubMed] [Google Scholar]

[pone.0237216.ref024] 24.Rando JC, Pieper H, Olson SL, Pereira F, Alcover JA. A new extinct species of large bullfinch (Aves: Fringillidae: Pyrrhula) from Graciosa Island (Azores, North Atlantic Ocean). Zootaxa. 2017; 4282(3): 567–583. [Google Scholar]

[pone.0237216.ref025] 25.Global Invasive Species Database. Species profile: Mustela nivalis. Available from http://www.iucngisd.org/gisd/speciesname/Mustela+nivalis; 2018a.

[pone.0237216.ref026] 26.Global Invasive Species Database. Species profile: Mustela furo. Available from http://www.iucngisd.org/gisd/speciesname/Mustela+furo; 2018b.

[pone.0237216.ref027] 27.Pitta-Groz M, Pereira JC, Silva AG. Invasive alien species as the main threat to Azores seabird populations. In: Proceedings of Workshop on invasive alien species on European islands and evolutionary isolated ecosystems. Horta, Azores, Portugal: 10 October 2002. Convention on the conservation of European wildlife and natural habitats; 2002. [Google Scholar]

[pone.0237216.ref028] 28.Warren DL, Seifert SN. Ecological niche modeling in Maxent: the importance of model complexity and the performance of model selection criteria. Ecol Appl. 2011; 21(2): 335–342. 10.1890/10-1171.1 [DOI] [PubMed] [Google Scholar]

[pone.0237216.ref029] 29.Elith J, Leathwick JR. Species distribution models: ecological explanation and prediction across space and time. Annu Rev Ecol Evol S. 2009; 40: 677–697. [Google Scholar]

[pone.0237216.ref030] 30.Welk E, Schubert K, Hoffmann MH. Present and potential distribution of invasive garlic mustard (Alliaria petiolata) in North America. Divers Distrib. 2002; 8(4): 219–233. [Google Scholar]

[pone.0237216.ref031] 31.Wiens JJ, Graham CH. Niche conservatism: integrating evolution, ecology, and conservation biology. Annu Rev Ecol Evol S. 2005; 36: 519–539. [Google Scholar]

[pone.0237216.ref032] 32.Shea K, Chesson P. Community ecology theory as a framework for biological invasions. Trends Ecol Evol. 2002; 17(4): 170–176. [Google Scholar]

[pone.0237216.ref033] 33.Sheffield SR, King CM. Mustela nivalis. Mammalian Species. 1994; 454: 1−10. [Google Scholar]

[pone.0237216.ref034] 34.McDonald RA, Abramov AV, Stubbe M, Herrero J, Maran T, Tikhonov A, et al. Mustela nivalis. The IUCN Red List of Threatened Species 2016; 2019.

[pone.0237216.ref035] 35.Rodrigues M, Bos AR, Schembri PJ, de Lima RF, Lymberakis P, Parpal L, et al. Origin and introduction history of the least weasel (Mustela nivalis) on Mediterranean and Atlantic islands inferred from genetic data. Biol Invasions. 2017; 19(1): 399–421. [Google Scholar]

[pone.0237216.ref036] 36.King CM. Weasel. In: King C.M.(Ed.), The Handbook of New Zealand Mammals. Oxford University Press, Auckland; 2005. [Google Scholar]

[pone.0237216.ref037] 37.Corbett GB, Ovenden D. The Mammals of Britain and Europe. Collins, London; 1980. [Google Scholar]

[pone.0237216.ref038] 38.King CM. The Handbook of New Zealand Mammals, 1st Edition. Oxford University Press, Auckland; 1990. [Google Scholar]

[pone.0237216.ref039] 39.Costa M, Fernandes C, Birks JDS, Kitchener AC, Santos-Reis M, Bruford MW. The genetic legacy of the 19th century decline of the British polecat: evidence for extensive introgression from feral ferrets. Mol Ecol. 2013; 22(20): 5130−5147. 10.1111/mec.12456 [DOI] [PubMed] [Google Scholar]

[pone.0237216.ref040] 40.Skumatov D, Abramov AV, Herrero J, Kitchener A, Maran T, Kranz A, et al. Mustela putorius. The IUCN Red List of Threatened Species 2016; 2016. [Google Scholar]

[pone.0237216.ref041] 41.Arechavaleta M, Rodríguez S, Zurita N, García A. Lista de especies silvestres de Canarias. Hongos, plantas y animales terrestres. Gobierno de Canarias, Canarias; 2009.

[pone.0237216.ref042] 42.Clapperton BK. Advances in New Zealand mammalogy 1990–2000: Feral ferret. J Roy Soc New Zeal. 2001; 31(1): 185–203. [Google Scholar]

[pone.0237216.ref043] 43.Lever C. Naturalized mammals of the world. Longman Science & Technology, London, UK; 1985. [Google Scholar]

[pone.0237216.ref044] 44.Gaspar C, Borges PA, Gaston KJ. Diversity and distribution of arthropods in native forests of the Azores archipelago. Arquipélago–Life and Marine Sciences. 2008; 25: 1–30. [Google Scholar]

[pone.0237216.ref045] 45.Triantis KA, Borges PA, Ladle RJ, Hortal J, Cardoso P, Gaspar C, et al. Extinction debt on oceanic islands. Ecography. 2010; 33(2): 285–294. [Google Scholar]

[pone.0237216.ref046] 46.Cardoso P, Rigal F, Fattorini S, Terzopoulou S, Borges PA. Integrating landscape disturbance and indicator species in conservation studies. PLoS One. 2013; 8(5): e63294 10.1371/journal.pone.0063294 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0237216.ref047] 47.Terzopoulou S, Rigal F, Whittaker RJ, Borges PA, Triantis KA. Drivers of extinction: the case of Azorean beetles. Biol Lett. 2015; 11(6): 20150273 10.1098/rsbl.2015.0273 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0237216.ref048] 48.Borges PAV, Lobo JM, Azevedo EB, Gaspar C, Melo C, Nunes LV. Invasibility and species richness of island endemic arthropods: a general model of endemic vs. exotic species. J Biogeogr. 2006; 33: 169–187. [Google Scholar]

[pone.0237216.ref049] 49.Hortal J, Jiménez‐Valverde A, Gómez JF, Lobo JM, Baselga A. Historical bias in biodiversity inventories affects the observed environmental niche of the species. Oikos. 2008; 117(6): 847–858. [Google Scholar]

[pone.0237216.ref050] 50.Phillips SJ, Dudík M, Elith J, Graham CH, Lehmann A, Leathwick J, et al. Sample selection bias and presence‐only distribution models: implications for background and pseudo‐absence data. Ecol Appl. 2009; 19(1): 181–197. 10.1890/07-2153.1 [DOI] [PubMed] [Google Scholar]

[pone.0237216.ref051] 51.Ordeñana MA, Crooks KR, Boydston EE, Fisher RN, Lyren LM, Siudyla S, et al. Effects of urbanization on carnivore species distribution and richness. J Mammal. 2010; 91(6): 1322–1331. [Google Scholar]

[pone.0237216.ref052] 52.Long RA, Donovan TM, MacKay P, Zielinski WJ, Buzas JS. Predicting carnivore occurrence with noninvasive surveys and occupancy modeling. Landsc Ecol. 2011; 26(3): 327–340. [Google Scholar]

[pone.0237216.ref053] 53.Cuyckens GAE, Perovic PG, Cristobal L. How are wetlands and biological interactions related to carnivore distributions at high altitude?. J Arid Environ. 2015; 115: 14–18. [Google Scholar]

[pone.0237216.ref054] 54.Team RStudio. RStudio: Integrated Development for R. RStudio, Inc., Boston, MA; 2015. URL http://www.rstudio.com/ [Google Scholar]

[pone.0237216.ref055] 55.Phillips SJ, Anderson RP, Schapire RE. Maximum entropy modeling of species geographic distributions. Ecol Modell, 2006; 190(3–4): 231–259. [Google Scholar]

[pone.0237216.ref056] 56.Merow C, Smith MJ, Silander JA. A practical guide to MaxEnt for modeling species’ distributions: what it does, and why inputs and settings matter. Ecography. 2013; 36(10): 1058–1069. [Google Scholar]

[pone.0237216.ref057] 57.Elith J, Graham CH, Anderson RP, Dudík M, Ferrier S, Guisan A, et al. Novel methods improve prediction of species' distributions from occurrence data. Ecography. 2006; 29: 129–151. [Google Scholar]

[pone.0237216.ref058] 58.Elith J, Phillips SJ, Hastie T, Dudík M, Chee YE, Yates CJ. A statistical explanation of MaxEnt for ecologists. Divers Distrib. 2011; 17(1): 43–57. [Google Scholar]

[pone.0237216.ref059] 59.Hernandez PA, Graham CH, Master LL, Albert DL. The effect of sample size and species characteristics on performance of different species distribution modeling methods. Ecography. 2006; 29(5): 773–785. [Google Scholar]

[pone.0237216.ref060] 60.Bean WT, Stafford R, Brashares JS. The effects of small sample size and sample bias on threshold selection and accuracy assessment of species distribution models. Ecography. 2012; 35(3): 250–258. [Google Scholar]

[pone.0237216.ref061] 61.Proosdij AS, Sosef MS, Wieringa JJ, Raes N. Minimum required number of specimen records to develop accurate species distribution models. Ecography. 2016; 39(6): 542–552. [Google Scholar]

[pone.0237216.ref062] 62.Gallien L, Douzet R, Pratte S, Zimmermann NE, Thuiller W. Invasive species distribution models–how violating the equilibrium assumption can create new insights. Global Ecol Biogeogr. 2012; 21(11): 1126–1136. [Google Scholar]

[pone.0237216.ref063] 63.Václavík T, Kupfer JA, Meentemeyer RK. Accounting for multi‐scale spatial autocorrelation improves performance of invasive species distribution modelling (iSDM). J Biogeogr. 2012; 39(1): 42–55. [Google Scholar]

[pone.0237216.ref064] 64.McPherson JM, Jetz W, Rogers DJ. The effects of species’ range sizes on the accuracy of distribution models: ecological phenomenon or statistical artefact?. J Appl Ecol. 2004; 41(5): 811–823. [Google Scholar]

[pone.0237216.ref065] 65.Lobo JM, Jiménez‐Valverde A, Real R. AUC: a misleading measure of the performance of predictive distribution models. Global Ecol Biogeogr. 2008; 17(2): 145–151. [Google Scholar]

[pone.0237216.ref066] 66.Barbet‐Massin M, Jiguet F, Albert CH, Thuiller W. Selecting pseudo‐absences for species distribution models: how, where and how many?. Methods Ecol Evol. 2012; 3(2): 327–338. [Google Scholar]

[pone.0237216.ref067] 67.Burnham KP, Anderson DR. Model selection and multimodel inference: a practical information-theoretic approach. Springer Science & Business Media; 2003. [Google Scholar]

[pone.0237216.ref068] 68.Burnham KP, Anderson DR. Multimodel inference: understanding AIC and BIC in model selection. Sociol Methods Res. 2004; 33(2): 261–304. [Google Scholar]

[pone.0237216.ref069] 69.Kuemmerle T, Perzanowski K, Chaskovskyy O, Ostapowicz K, Halada L, Bashta AT, et al. European bison habitat in the Carpathian Mountains. Biol Conserv. 2010; 143(4): 908–916. [Google Scholar]

[pone.0237216.ref070] 70.Araújo MB, Anderson RP, Barbosa AM, Beale CM, Dormann, C. F., Early, R., et al. Standards for distribution models in biodiversity assessments. Sci Adv. 2019; 5(1): eaat4858 10.1126/sciadv.aat4858 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0237216.ref071] 71.Rilov G, Benayahu Y, Gasith A. Prolonged lag in population outbreak of an invasive mussel: a shifting-habitat model. Biol Invasions. 2004; 6(3): 347–364. [Google Scholar]

[pone.0237216.ref072] 72.Broennimann O, Treier UA, Müller‐Schärer H, Thuiller W, Peterson AT, Guisan A. Evidence of climatic niche shift during biological invasion. Ecol Lett. 2007; 10(8): 701–709. 10.1111/j.1461-0248.2007.01060.x [DOI] [PubMed] [Google Scholar]

[pone.0237216.ref073] 73.Fitzpatrick MC, Weltzin JF, Sanders NJ, Dunn RR. The biogeography of prediction error: why does the introduced range of the fire ant over‐predict its native range?. Global Ecol Biogeogr. 2007; 16(1): 24–33. [Google Scholar]

[pone.0237216.ref074] 74.Azevedo EB, Pereira LS, Itier B. Modelling the local climate in island environments: Water balance applications. Agric Water Manag. 1999; 40: 393–403. [Google Scholar]

[pone.0237216.ref075] 75.Hortal J, Borges PAV, Jiménez-Valverde A, Azevedo EB, Silva L. Assessing the areas under risk of invasion within islands through potential distribution modelling: the case of Pittosporum undulatum in São Miguel, Azores. J Nat Conserv. 2010; 18: 247–257. [Google Scholar]

[pone.0237216.ref076] 76.Aparício BA, Cascalho J, Cruz MJ, Borges PA, Azevedo EB, Elias RB, et al. Assessing the landscape functional connectivity using movement maps: a case study with endemic Azorean insects. J Insect Conserv. 2018; 1–9. [Google Scholar]

[pone.0237216.ref077] 77.Fitzpatrick MC, Weltzin JF. Ecological niche models and the geography of biological invasions: a review and a novel application. In Invasive Plants: Ecological and Agricultural Aspects. Birkhäuser Basel. Switzerland; 2005.

[pone.0237216.ref078] 78.Kauhala K. Introduced carnivores in Europe with special reference to central and northern Europe. Wildlife Biol. 1996; 2(3): 197–204. [Google Scholar]

[pone.0237216.ref079] 79.Dungan J, Perry J, Dale M, Legendre P, Citron-Pousty S, Fortin M, et al. A balanced view of scale in spatial statistical analysis. Ecography. 2002; 25: 626–640. [Google Scholar]

[pone.0237216.ref080] 80.Collares-Pereira M, Mathias ML, Santos-Reis M, Ramalhinho MG, Duarte-Rodrigues P. Rodents and Leptospira transmission risk in Terceira island (Azores). Eur J Epidemiol. 2000; 16(12): 1151–1157. 10.1023/a:1010916132497 [DOI] [PubMed] [Google Scholar]

[pone.0237216.ref081] 81.Smith GP, Ragg JR, Moller H, Waldrup KA. Diet of feral ferrets (Mustela furo) from pastoral habitats in Otago and Southland, New Zealand. New Zeal J Zool. 1995; 22(4): 363–369. [Google Scholar]

[pone.0237216.ref082] 82.Ragg JR. Intraspecific and seasonal differences in the diet of feral ferrets (Mustela furo) in a pastoral habitat, East Otago, New Zealand. New Zeal J Ecol. 1998; 113–119. [Google Scholar]

[pone.0237216.ref083] 83.Santos MJ, Matos HM, Baltazar C, Grilo C, Santos-Reis M. Is polecat (Mustela putorius) diet affected by “mediterraneity”?. Mammalian Biology-Zeitschrift für Säugetierkunde. 2009; 74(6): 448–455. [Google Scholar]

PERMALINK

Modelling the distribution of Mustela nivalis and M. putorius in the Azores archipelago based on native and introduced ranges

Lucas Lamelas-López

Xosé Pardavila

Paulo A V Borges

Margarida Santos-Reis

Isabel R Amorim

Maria J Santos

Roles

Abstract

Introduction

Material and methods

Study organisms

Mustela nivalis Linnaeus, 1766

Mustela putorius Linnaeus, 1758

Study area

Fig 1. Selected areas for the study of the factors affecting the distribution range of Mustela nivalis and M. putorius.

Species presence data

Environmental data

Table 1. Description of the candidate variables to model M. nivalis and M. putorius distribution.

Species distribution modeling

Parameter configuration

Model selection

Environmental variables contribution to the models

Habitat suitability for M. nivalis and M. putorius in the Azores

Results

Model selection

Table 2. Results of AIC-based model selection for the suitability of Mustela nivalis and M. putorius occurrence, in the introduced range (Azores) and in the native range (Europe).

Predicted M. nivalis and M. putorius ranges in the Azores

Fig 2. Mustela nivalis potential distribution map for the Azores.

Fig 3. Mustela putorius potential distribution map for the Azores.

Environmental correlates of island invaders

Table 3. Environmental variables contribution to the potential distribution of Mustela nivalis and M. putorius in introduced (Azores) and native (Europe) ranges.

Table 4. Mustela nivalis and M. putorius species distribution model adjustment to the field data.

Discussion

Introduced and native range predictions

Mustela nivalis distribution

Mustela putorius distribution

Conclusion

Supporting information

Acknowledgments

Data Availability

Funding Statement

References

Decision Letter 0

Ulrike Gertrud Munderloh

Roles

Author response to Decision Letter 0

Decision Letter 1

Ulrike Gertrud Munderloh

Roles

Author response to Decision Letter 1

Decision Letter 2

Ulrike Gertrud Munderloh

Roles

Acceptance letter

Ulrike Gertrud Munderloh

Roles

Associated Data

Supplementary Materials

Data Availability Statement

ACTIONS

PERMALINK

RESOURCES

Similar articles

Cited by other articles

Links to NCBI Databases