Introduction

Bees are the main group of animal pollinators and ensure at least 35 % of global food production (Klein et al. 2003; Kremen et al. 2007; Klatt et al. 2014). The pollination of crops and wild plants by animals represent one of the ecosystem services that are at risk due to fast and growing anthropogenic changes (Steffan-Dewenter et al. 2005; Biesmeijer et al. 2006; Aizen and Harder 2009; Potts et al. 2010). The main factors determining recent environmental changes and, consequently, bee’s decline worldwide, are land use and climate change (Travis 2003; MEA 2005; Tylianakis et al. 2008). Jointly, these factors may cause extinctions, shifts in species ranges, and changes in species’ ecological and phenological events (Walther et al. 2002; Parmesan and Yohe 2003; Parmesan 2006; Tylianakis et al. 2008).

Population declines are well documented for honey bees as well as for bumblebees (genus Bombus), especially in Europe and North America (Williams and Osborne 2009; Bommarco et al. 2012; Colla et al. 2012; Bartomeus et al. 2013). Bumblebees are effective crop pollinators (Garratt et al. 2014) and the main factors responsible for the observed population declines are shortage of food sources and nesting sites due to habitat loss (Goulson et al. 2008). Pesticides (Thompson 2001; Whitehorn et al. 2012) and competition with introduced honeybees (Thomson 2004), or other bumblebee species (Morales et al. 2013), also exert a negative influence in bumblebee populations. A combination of these factors may also be a possible explanation for the observed decreases in Bombus abundance (Winfree et al. 2009; Bartomeus et al. 2013). Another reported factor that causes vulnerability in bumblebee populations is the low genetic diversity related to its social behaviour (Chapman and Bourke 2001; Ellis et al. 2006). However, the existence of a relationship between social behaviour traits and observed abundance declines of social insects is still controversial, since for some authors it exists (Winfree et al. 2009) and for others it does not (e.g. Bartomeus et al. 2013), or depends on environmental disturbance affecting the species (Williams et al. 2010).

So far, bumblebee declines in South America have gone unnoticed until Bombus bellicosus Smith, 1879 was reported to be locally extinct in the Brazilian state of Paraná, Southern Brazil (Martins and Melo 2009). Also Bombus dahlbomii Guérin, 1835 was reported as being replaced by the invasive species Bombus terrestris (Linnaeus, 1758) in Chile and Argentina (Morales et al. 2013). B. bellicosus is a surface ground-nesting bee, which requires only soil and plant detritus to build its nest (Varela 1992a, b). Females of B. bellicosus cover their nest cavity with a wax layer—possibly as a protection against cold temperatures. This is a unique trait among tropical and subtropical Bombus that links B. bellicosus to temperate and sub-temperate climates (Sakagami et al. 1967a). The original known distribution of B. bellicosus ranges from latitude 2446’S in southern Brazil up to Uruguay and northern Argentina (Martins and Melo 2009), always associated with native grasslands (Moure and Sakagami 1962).

Bombus bellicosus was the dominant native bee species in Paraná’s first plateau, southern Brazil, until the 1960s (Sakagami et al. 1967b; Sakagami and Laroca 1971). Since then, this region underwent intense land use by urbanization and/or agriculture, leading to a general local decrease in bee abundance and richness and to disappearance of B. bellicosus (Martins and Melo 2009; Martins et al. 2013). Despite the visible effect of intense land use in the past four decades, this species still survives in agricultural patches in more southern regions (Martins and Melo 2009). In Europe, land cover together with climate change influenced both richness and local abundance in bumblebees (Herrera et al. 2014). This leads us to question whether its disappearance in the northern limit of its distribution in southern Brazil was caused only by habitat loss or if climate change may have also played an important role in the species disappearance from its northernmost occurrence limit in Brazil. In South America, range shifts and species losses caused by climate change have been predicted for different groups: Atlantic forest trees (Colombo and Joly 2010), amphibians (Lemes and Loyola 2013), marsupials (Loyola et al. 2012) and moths (Ferro et al. 2014).

However, information on many life-history features of B. bellicosus and on effects of environmental changes on its natural habitats (the Hutchinsonian shortfall; Cardoso et al. 2011) are lacking. Such absence of basic biological data regarding B. bellicosus occurs in parallel with a lack of broad scale quality data on distribution (the Wallacean shortfall; Whittaker et al. 2005), a knowledge gap observed for many other insect species (Diniz-Filho et al. 2010). The Wallacean shortfall is an undeniable obstacle hindering the assessment of insect species under conservation biogeography frameworks especially when we think of species conservation in broad scales (Whittaker et al. 2005; Diniz-Filho et al. 2010; Cardoso et al. 2011) and is an important motive for continuous field surveys to assess insect species distributions (Diniz-Filho et al. 2010).

A possible alternative to fully consider insect distributions and transcend the Wallacean shortfall in broad-scale studies is to use species distribution modelling (SDMs). Usually, these methods relate modelled occurrences of observed species with environmental variables of the known sampling sites to predict new areas suitable for its occurrence (Guisan and Zimmermann 2000; Kearney 2006). These techniques have been employed before to (1) determine insect species’ distributions and inform suitable areas for future surveys (Silva et al. 2013, 2014), (2) pinpoint modelled species rich areas uncovered by the available protected areas network (Nóbrega and De Marco 2011), and (3) to evaluate the effects of future climate changes on species distributions (Loyola et al. 2012; Lemes and Loyola 2013). Considering this context, we used SDMs to address the potential effects of climate change in distributions of B. bellicosus in southern Brazil considering both current and future climate change scenarios.

Materials and methods

Occurrence of Bombus bellicosus

We obtained a total of 303 records of B. bellicosus occurrence from (1) literature records; (2) the online datasets: CRIA’s Species Link (http://splink.cria.org.br), Global Biodiversity Information Facility (http://www.gbif.org), Discover Life Bee Species Guide (http://www.discover-life.org), World Checklist (Ascher and Pickering 2014); and (3) museum collections (see Supplementary Material). We used Google Earth (Google Inc. 2013) to acquire approximate geographical coordinates from downtown of each city near sites where B. bellicosus occurred, when lacking a precise geographical information. We assembled 171 unique occurrences to be used in the modelling procedures (grid cell resolution of 2.5 arc-min ≈ grid-cells with sizes of 0.041° in the tropics; see below). Figure 1 shows the spatial distribution of all unique B. bellicosus occurrences we obtained.

Fig. 1
figure 1

Bombus bellicosus geographical distribution. The map shows the around 170 points of occurrence for B. bellicosus in southern South America, sampled from literature and museum collections, which are used as input distribution for all the modelling approaches. Thick lines delimit South American countries and thin lines delimit provinces or states of each country

Full size image

We used all occurrences to model the potential distribution of B. bellicosus, including the sites mentioned by Martins and Melo (2009), where the species were observed after the early 1980s, for two reasons. At first, as we previously mentioned, Wallacean shortfalls severely affect insect distributional data elsewhere, especially in tropical regions (Diniz-Filho et al. 2010; Kamino et al. 2011). If we were to assume the occurrence records only sampled after 1980 were the most reliable distributional information on the species to infer its distribution (a progressive data quality measure, since Bombus queens usually live for less than 1 year; Michener 2007), nearly 80 % of the dated occurrences (~290 out of the total 303) would be missing (Figure S1A). Additionally, much of the species known distribution would be disregarded (Figure S1B). A higher number of occurrences may eventually add just a small amount, or even reduce, models’ accuracy (Stockwell and Peterson 2002). Still, we are dealing with a species potentially targeted for conservation purposes, so a higher number of known occurrences will provide a larger amount of areas which may be relevant for estimating its future distribution. The second reason for using all occurrences was that in preliminary tests we carried out, even if we removed the sites where the species is no longer observed, the same areas were still predicted as being suitable for the occurrence of B. bellicosus.

Environmental data, principal component variables, and modelling procedures

A summary of all methods of analysis and input data are presented in Fig. 2. For the current scenario, we used the 19 environmental variables available from the WorldClim’s dataset (http://www.worldclim.org/current, Hijmans et al. 2005). We used a Principal Components Analysis (PCA) on a correlation matrix, to derive 19 principal components (PCs hereon), from which the first seven were used as our final environmental layers (98 % of all original climatic variation; Table S1). We obtained the same 19 variables for the year 2080 for three different Atmosphere–Ocean General Circulation Models (AOGCMs; CCCMA-CGM2, CSIRO-MK2.0, and UKMO-HADCM3) from CIAT (http://ccafs-climate.org) considering both optimistic and pessimistic emission scenarios (B2a and A2a, respectively). Then, we projected the linear combinations of all 19 PCs obtained for the current scenario into each different future AOGCM scenario. We also used the first seven PCs obtained for each future climatic scenario in the modelling procedures to predict the future distribution of B. bellicosus. Using PCs to predict species distribution is recommended to reduce collinearity among environmental variables, and to avoid model overfitting, which may result in biologically unreliable species potential distributions (Jiménez-Valverde et al. 2011; Serra et al. 2012; Silva et al. 2014).

Fig. 2
figure 2

Flowchart representing stages of analysis in this study. The flowchart represent the stages of analysis, methods employed (algorithms), source of climatic data, and data inputs used in the distribution modelling for Bombus bellicosus current and future distribution under climate change

Full size image

Given the uncertainty involved with SDMs (Barry and Elith 2006; Diniz-Filho et al. 2009; Rocchini et al. 2011), we evaluated B. bellicosus’ distributions using three different modelling algorithms: (1) Maximum Entropy (Phillips et al. 2006; Phillips and Dudik 2008) implemented in Maxent (Phillips et al. 2006); (2) Support Vector Machines (SVM; Schölkopf et al. 2001; Tax and Duin 2004); and (3) Mahalanobis distance (MAHAL; Farber and Kadmon 2003) implemented in open modeller desktop (Muñoz et al. 2011). The later algorithm is simpler and usually needs presence-only data to predict the potential distribution of the targeted species. Meanwhile, Maxent and SVM are artificial intelligence algorithms that are more complex and tend to better predict the known/true distribution of the target species (Rangel and Loyola 2012).

We used 50 subsets of occurrences of B. bellicosus, divided into 70 % training and 30 % testing subsets to, respectively, produce the potential distributions and evaluate them, considering the current and all future climatic scenarios. B. bellicosus is a potentially targeted species for future practical conservation actions. Thus, we considered the ROC threshold to cut the modelled suitability matrices into presence/absence maps. This threshold balances both omission and commission errors and assures higher prediction rates than other thresholds usually considered in modelling species distribution (Liu et al. 2005; Barbet-Massin et al. 2012). We used true skilled statistics (TSS hereafter; Allouche et al. 2006), a threshold-dependent statistics which varies from −1 to 1, to assess models’ performance. Negative or near zero TSS values represent distributions which were not better than a random distribution, while values near one represent a perfect agreement between the observed and the modelled species’ distribution in current climatic scenarios (Allouche et al. 2006).

We produced a total of 1,050 distribution maps for B. bellicosus, for both current (n = 150) and future scenarios (n = 900, 450 for each). We used both frequency maps (Araújo and New 2007) and mean ensemble distributions of all 50 potential distributions obtained in each climatic scenario (current, optimistic and pessimistic future climatic conditions) to represent the species’ final potential distributions in each combination of algorithm and AOGCM used. Additionally, we also show mean consensus distribution for the species in each one of the three climatic scenarios considered (current, optimistic, and pessimistic future). This ensemble method is considered one of the most reliable to determine the potential distributions of species from different modelling algorithms (Marmion et al. 2009).

Finally, we assessed the IUCN’s World Database on the Protected Areas website (http://protectedplanet.net/) to obtain the South American reserve network (categories ranging from I to IV ≈ strict reserves). Then, we overlaid the South American network of protected areas (PA hereon) onto the potential distributions of B. bellicosus to evaluate its current and future conservation vulnerability under different future climatic scenarios.

Results

The modelled distributions for B. bellicosus had good prediction rates overall. The TSS values generally reached values higher than 0.70 (0.788 ± 0.038; mean ± SD), which indicates a good model fit (MAX 0.798 ± 0.054; MAHAL 0.779 ± 0.023; SVM 0.787 ± 0.028). In the current scenario, regions extending from southern Brazil, Uruguay, and southeastern Paraguay, up to central Argentina were always predicted as suitable for B. bellicosus occurrence. This suitability occurred regardless of the inherent uncertainty and variance of both the SDMs and the AOGCM models considered, as shown in the frequency maps and consensual distributions (Fig. 3). The more restricted distribution obtained for the current scenario was mainly determined by the consensual distribution obtained with the SVM algorithm (Fig. 3).

Fig. 3
figure 3

Bombus bellicosus current and future distribution in southern South America. The figure shows three different modelling approaches based on the modelling algorithms: Mahalanobis distance, Maxent and SVM. Frequency and consensus maps are shown. Current and future scenarios were based on the 19 environmental variables from the WorldClim dataset. The future climate modelling variables were based on different Atmosphere Ocean General Circulation Models (AOGCMs) for the years 2080 (CCCMA-CGM2, CSIRO-MK2.0, and UKMO-HADCM3), considering both optimistic and pessimistic emission scenarios (B2a and A2a, respectively)

Full size image

Our resulting distribution of B. bellicosus for the future scenarios showed a clear trend despite intrinsic differences of the considered algorithms (AOGCMs, and the emission scenarios): the suitable climatic conditions observed for B. bellicosus in the present will retreat southwards, regardless of the climate scenario and algorithm considered (Fig. 3). We observed the same trend in the consensual distributions produced by each combination of modelling algorithm and climatic scenario. According to these combinations, B. bellicosus may occupy areas ranging from the southern Brazilian state of Rio Grande do Sul, Uruguay, and central/south Argentina. Nonetheless, we also predicted the occurrence of a disaggregated patch of suitable landscapes for southern Brazilian states (Paraná, Santa Catarina, and Rio Grande do Sul), where grid cells had average altitudes of 675 ± 305 (mean ± SD) meters above sea level, at least for the distributions produced by both Maxent and MAHAL. The models generated for the future distributions with SVM algorithms did not produce such fractioned distribution patch in southern Brazil. When we consider the ensemble distributions, suitable areas for B. bellicosus near the Andean ridge may also be observed, although in a smaller frequency for the distributions of B. bellicosus produced by the current climatic scenario (Fig. 3). Considering the current, pessimistic and optimistic future potential distributions obtained with all different modelling algorithms, B. bellicosus is expected to occur in some of the available South American protected areas (Fig. 4). Nonetheless, southern protected areas may provide better refuges than northern ones in both future and potential climatic scenarios.

Fig. 4
figure 4

Bombus bellicosus’ current and future distributions overlapped with South American protected areas network (IUCN). The three distribution maps represents the consensual distribution all maps obtained by all three algorithms models (Mahalanobis, Maxent and SVM) in each different (Current, Pessimistic, and Optimistic) climatic scenario. The numeric scale correspond to the consensual distribution obtained by each modelling algorithm and climatic scenario. Future distributions were obtained from the mean consensus from the predicted distributions by each modelling algorithm and on the three of AOGCM for both pessimist (A2a) and optimistic (B2a) emission scenarios. Grey polygons refer to the available South American protected areas network (categories I–IV ≈ strict reserves) from the IUCN’s World Database on the protected areas

Full size image

Discussion

Climate change may be a key factor for the southward retraction of B. bellicosus distribution under future climate change scenarios, as shown by our potential models. The current PA network in South America will be able to protect at least some segments of B. bellicosus potential future distributions. Models of future distribution for other bee species also show the loss of suitable areas in future climate change scenarios (Giannini et al. 2012). Climate change in the next 50 years is pointed out as a major factor in forecasts of distribution range changes for Atlantic forest trees (Colombo and Joly 2010), amphibians (Lemes and Loyola 2013) and marsupials (Loyola et al. 2012). For South American moths, climate change will force range shifts and reduce the conservation effectiveness of some protected areas, even leading 4 % of all analysed species to extinction (Ferro et al. 2014). The bee fauna (bumblebees included) from this region, as they are also dependent on the same environmental variables, may also show similar patterns upon future climate change scenarios.

According to our potential distribution models, the center of the original distribution of B. bellicosus, near Buenos Aires, Argentina (Abrahamovich et al. 2004), will be still suitable for the species in the future. Declines on bumblebee populations on the border and its persistence in areas closer to the center of their distribution was also observed in other species (Williams et al. 2007, 2009). Many studies show the vulnerability of species in the boundaries of their distributions, especially regarding climate change (Parmesan 2006). Climate is the main factor determining the boundaries of species distributions. It may decrease the fitness of marginal populations (Hoffmann and Blows 1994), and limit the bees’ ability to adapt to potential climate change scenarios (Bridle and Vines 2007). However, some intrinsic biological traits related to B. bellicosus may also determine how it will behave under future climate changes.

The presence of a wax cover in nests of B. bellicosus, unlike other tropical and subtropical species, is a biological trait that links this species to temperate climates (Sakagami et al. 1967a; Hines et al. 2007), and allows their nests to maintain higher temperatures, especially during cold seasons. However, under future climate change scenarios and increasing environmental temperatures, this wax layer may eventually cause severe temperature increases inside nests of B. bellicosus. Temperature is a very important environmental characteristic for the development of all insect species (Chown and Terblanche 2006). High temperatures negatively affect the development of insect broods (Polak and Tomkins 2013). In the future, both vitality and strength of nests of B. bellicosus may decrease, given the expected temperature increases under future scenarios of climate change.

There is solid evidence in the literature that some invertebrate species may not cope with plant species phenological changes (Root et al. 2003; Hegland et al. 2009). Consequently, the available resources available for pollinator insect species during the most important period of their developmental stages may eventually decrease under climate warming, causing considerable problems to insect pollinators (Durant et al. 2007), especially for specialized plant-pollinators interactions (Schweiger et al. 2012). In some cases, plant species phenological advances may not cause significant problems to the insect that are dependent on them (Bartomeus et al. 2011). However, the role of plant species phenological changes on the future distribution of B. bellicosus may be of a smaller importance, as it is a generalist species (Arbulo et al. 2011) which visits more than 60 different plant species (Martins and Melo 2009). Consequently, eventual phenological changes of South American plant species undergoing climate change, as already reported elsewhere (Root et al. 2003; Parmesan 2006), will probably not affect the persistence of B. bellicosus in the considered area.

Some studies also question the association between climate change and declines in bumblebee populations (Williams and Osborne 2009) and link population declines mainly to habitat loss (Fitzpatrick et al. 2006). Other key factors, such as introduced species, pesticides and pathogens, are also claimed to have an important effect on bumblebee populations (Goulson et al. 2008). All these factors cannot be discarded in the local extinction of B. bellicosus, especially because intense land use and the increase of urban matrixes have been proved to influence the observed population decline for many bee species, including B. bellicosus in southern Brazil (Martins et al. 2013). Land use change is continuously growing in the areas occupied by B. bellicosus in the last four decades due to strong agricultural expansion. Such factors will continue to affect the bee fauna in the region studied unless conservation actions take place (Overbeck et al. 2007; WWF 2013). An increasing urban matrix is usually a deleterious factor for bees, although some surviving species may still persist this habitat that is inhospitable for many bee species (Winfree et al. 2007; Banaszak-Cibicka and Żmihorski 2011). Bee survival in urban environments (and other anthropogenic landscapes) depends on the availability of feeding and nesting resources within the bee’s dispersal ability (Kremen et al. 2007).

Using SDMs to support decisions regarding species conservation in the future climate change scenarios is not a simple task, mainly because the resulting potential distributions (based on the current knowledge on species distribution) are impossible to be truly validated (Pearson and Dawson 2003). However, SDMs are still one of the best alternatives to discuss species conservation under future climate change (Hannah et al. 2007; Heller and Zavaleta 2009; Guisan et al. 2013). The current PA network from South America will be able to protect small disconnected portions of the distribution of B. bellicosus under both the current and the future climate change scenarios. Nonetheless, and given the disaggregated predicted distributions of B. bellicosus in southern Brazil, remnant vegetation and PAs located within this region may eventually be stepping-stones for B. bellicosus populations (which is fundamental in cases of climate change; Hannah et al. 2007). Stepping-stones facilitate southward dispersal, considering the future scenarios of climate change. The establishment of several new protected areas can contribute to future efforts of B. bellicosus conservation, although the current PA network will be able to protect some portions of the distribution of B. bellicosus. Protected areas on the grasslands of southern South America, which are the main habitat for B. bellicosus, are fundamental for the conservation of this pollinator species in particular. However, these are one of the most neglected habits in that region, often endangered by large scale livestock and crops (Overbeck et al. 2007).

The current presence of the species in a protected area does not necessarily mean it will be protected in the future, although the conservation biogeography framework considers a given species to be protected even if the smallest portion of its distribution is covered by a protected area (Rodrigues et al. 2004). The current effectiveness of a PA in protecting B. bellicosus (and other species as well) may vary, especially because such areas were generally created without any ecological criteria. Additionally, when we consider the future climate change scenarios, the current network of protected areas will not be enough to protect species (Hannah et al. 2007), even in an optimistic scenario. Considering this situation, future PA implementation needs to be based on ecological theories rather than only on the stakeholders’ will and/or landscape scenic beauty (Hannah et al. 2007).

An additional factor which may potentially affect B. bellicosus and other South American bumblebees in the future is the competition for resources with B. terrestris, as already recently reported in Argentina (Morales et al. 2013). B. terrestris is a very competitive species (Goulson 2003), introduced in many regions of the world to pollinate crops, including in Chile, where it began its spread towards other South American countries, and to consequently replace both a previously introduced and a native bumblebee species (Morales et al. 2013). There is a chance that this species may continuously spread to southern and southeastern Brazil, and cause negative effects on other native bumblebees species (Saraiva et al. 2013).

In this paper we showed the potential distribution of B. bellicosus, under current and future scenarios of climate change, and discussed the potential influence of climate in its disappearance in some regions of Southern Brazil, where it was once the most abundant native species (40 years ago). However, the general absence of quality biological and ecological aspects of B. bellicosus as well as information regarding its wild populations are still major drawbacks for its efficient current and future protection. Besides its undeniable value in conservation predictions, SDMs results rarely become practical conservation decisions, mainly due to a gap in the communication between researchers (who design the models) and decision makers (Guisan et al. 2013). We hope that the results presented here may, eventually, leave the theoretical grounds and serve both conservationists and stakeholders as practical bridges towards the future protection of B. bellicosus in South America, especially given the potential adverse effects of climate changes for this species. Increasing the area and representativeness of protected areas, especially of the grasslands which serve as habitat for B. bellicosus, may have a major importance in the conservation of this and other species of Latin American pollinators in the near and potentially threatening future.