Parapatric species and the implications for climate change studies: a case study on hares in Europe

Species data

The European distribution of Lepus spp. was extracted from The Atlas of European Mammals (Mitchell-Jones et al., 1999; Fig. 1). Information refers to UTM 50 × 50 km squares as they were the territorial units used for modelling purposes. Due to the fact that the sampling effort used to create the Atlas was not spatially homogeneous, we calculated the number of mammal species in each square as a proxy of sampling effort; for modelling purposes, we only considered the UTM squares in which at least one species has been reported (n = 2557). This filter excluded most of Eastern Europe which coincided with the most incomplete area identified by the Atlas authors (A.J. Mitchell-Jones, personal communication). In the study area, the most frequent hare species was L. europaeus (n = 1119 presences), followed by L. timidus (n = 532 presences) and finally the Iberian endemism L. granatensis (n = 118 presences).

Environmental data

The occurrence of the three hare species in each UTM square was modelled using 35 potential explanatory variables related to the following factors: spatial location (2 variables), topography (1 variable), climatology (15 variables) and land use (17 variables; see Table 1). These variables were chosen on the basis of availability at this scale and potential predictive power, and were assumed to be correlated with more explanatory factors.

Land use data came from Global Land Cover 2005, which is freely available at http://www.esa.int/esaEO/SEMXB7TTGOF_index_0.html. The map (~300 m spatial resolution) covers the entire planet, and its accuracy has been successfully validated (see Bicheron et al., 2008). Bioclimatic variables (for present and future times) and altitude (~1000 m spatial resolution) were obtained from the Worldclim project database (see Hijmans et al., 2005 for details). The models calibrated for the present period were projected into the future by replacing the current bioclimatic variables in the models with those expected according to the climate change scenario for the future period up to 2080 using the A2 emissions scenario (Nakicenovic et al., 2000). This scenario is defined as a world of strengthening regional cultural identities, with an emphasis on family values and local traditions, high population growth and less concern for rapid economic development. We used only one scenario because our main interest was to assess the changes in interspecific relationships between parapatric species due to climate, rather than to assess the effect of different global circulation models or emissions scenarios (see Real et al., 2010).

Modelling

We used an inductive approach to estimate the macroecological requirements of the species from the locations in which they occurred (Corsi et al., 2000). We modelled the occurrence of each species assuming that, after correcting for the sampling effort previously described, if a species was not observed within a UTM square, this was equivalent to the absence of the species. For each species, the model was calibrated using a 70% random sample of the data and evaluated against the remaining 30%. Firstly, to control for the increase in type I errors as the number of independent variables increased, we evaluated the false discovery rate (FDR; García, 2003) using the procedure proposed by Benjamini & Hochberg (1995), and only accepted variables that were significantly (P < 0.05) related to the species distribution under an FDR of q < 0.05. The selected variables were then used in a multiple logistic regression procedure (Hosmer & Lemeshow, 1989), and the final models were selected following a forwards–backwards stepwise procedure. Finally, to establish direct comparisons between models (species), the logistic probabilities were used to obtain favourability values using the function described by Real et al. (2006). The favourability function is a valuable tool to study biogeographical relationships between models whatever the proportion of presence records (sample prevalence) in the calibration datasets (e.g. see Real et al., 2009; Acevedo et al., 2010). This is due to the fact that a favourability value of 0.5 always corresponds to the same environmental threshold, thus the independence of these values in relation to species prevalence enables direct comparisons between models (species) built with different prevalences.

Sensitivity – the percentage of correctly predicted presences to the total number of presences –, specificity – the percentage of correctly predicted absences to the total number of absences –, and the area under the ROC curve (AUC) were estimated on the validation datasets to assess the discriminative capacity of the models (Fielding & Bell, 1997; but see Lobo et al., 2008). To calculate sensitivity and specificity a threshold of 0.5 was used as a cut-off for favourability values in all the models according to the favourability concept (Real et al., 2006). All statistical analyses were performed using spss 18 (spss Inc., Chicago, IL, USA) statistical software.

Assessing relationships between parapatric species

An inherent quality of favourability values is that they can be regarded as the degree of membership in the fuzzy set of sites whose environmental conditions are favourable to the species (Robertson et al., 2004; Real et al., 2006). Thus, fuzzy logic operations can be used to compare different models. This is an advantage of the favourability function over other SDM techniques when the aim of the study is to combine models for different species, scenarios, etc. (see Estrada et al., 2008; Acevedo et al., 2010, 2011). The biogeographical relationships between two species can be assessed using the fuzzy overlap index (FOvI; see Acevedo et al., 2010), i.e. the ratio between the degree to which the study area is favourable to the two studied species simultaneously and the degree to which it is favourable for either species (Dubois & Prade, 1980; Kunchenva, 2001). This index varies from 0 (no overlap in favourability) to 1 (complete overlap in favourability). The FOvI can be decomposed into absolute local overlap values (FOvI-L) that represent the contribution of each locality (UTM square) to the FOvI. Thus, the FOvI-L shows the spatial location of the areas where spatial overlap between species is expected to occur (Acevedo et al., 2010).

Trends on species favourability were assessed across the range of FOvI-L values for each pair of parapatric species (L. granatensis/L. europaeus and L. europaeus/L. timidus) using the procedure described by Acevedo et al. (2010). Briefly, FOvI-L values were divided into 10 intervals (0.1 width), and mean favourability values at each interval were calculated for each pair of species. Throughout the gradient defined by FOvI-L, and consistent with the favourableness-severity hypothesis (Richerson & Lum, 1980), it can be assumed that competition between species increases and competitive exclusion decreases as FOvI-L increases. Subsequently, we divided the curve into fixed intervals: FOvI-L < 0.2 (areas that were unfavourable for at least one species) and FOvI-L > 0.8 (areas simultaneously highly favourable to the two species). According to the favourableness-severity hypothesis, the area with 0.2 <  FOvI-L < 0.8 is where biotic interactions could limit species occurrence (Acevedo et al., 2010).

Assessing changes in distribution patterns between climatic scenarios

We used the fuzzy logic indices described in Real et al. (2010) to calculate the increases in favourability (I), the favourability overlap (O), favourability maintenance (M) and the predicted shift in favourability (S) between present-future climate scenarios (date₁–date₂ respectively) for each hare species and between parapatric species interaction:

where cX is the cardinality of the X fuzzy set, i.e. the sum of all squares' favourability and Min is the minimum value. The fuzzy intersection is the minimum value between the favourability of two scenarios and the fuzzy union the maximum value between them (Zadeh, 1965). These indices are useful tools to describe the magnitude and direction of the changes in distribution patterns between two scenarios such as those driven by climate (Real et al., 2010) or land use changes (Acevedo et al., 2011). The FOvI-L can be also included in fuzzy logic operations due to its conceptual characteristics, and thus variations can be also assessed in the distribution pattern of the FOvI-L when models are projected into a future climate change scenario.

The methodological approach

A key step in SDM involves taking into account biotic interaction; thus, changes in climate may not only directly alter the distribution of a species, but also indirectly alter it through affecting the distribution of other interactive species (Meier et al., 2011). Our study offers a new perspective on the role of interspecific interactions on shaping future distribution ranges in response to climate change. The methodological approach, previously described by Acevedo et al. (2010), allowed us: (i) to determine the areas where the probability of competition between species is higher; and assuming equal population status for the species involved in the interaction; (ii) to infer a directional hypothesis on the role of each species; and (iii) to explore the spatial shifts in species interactions pattern under different scenarios, that is, to assess interspecific interactions and how they would evolve under climate change scenarios (see Klanderud & Totland, 2005). It is an improvement over other approaches previously used in the context of SDM and climate change assessment which are mainly based on including the distribution data of the interacting species as predictors during the modelling processes.

Clearly, the critical issue is not to determine whether biotic interactions have effects at biogeographical scales, but to quantify their magnitude (e.g. Leathwick & Austin, 2001). This key question is difficult to answer mainly because there is no validation data available by which to assess the predictions of the models under climate change conditions (Araújo et al., 2005). It is also often very difficult to distinguish spatial patterns generated by interspecific interactions from those caused by abiotic causes, historical factors or dispersal barriers (Wiens, 1989). With our approach two extreme situations for each species can be determined, one in which no effects of interspecific interaction were considered when assessing shifts in the species distribution area due to climate changes, and another in which it was assumed that the species was totally excluded from potentially competitive exclusion areas (see also Araújo & Luoto, 2007). In other words, it presents a range between no effects and the full effects of interspecific interactions. The real situation is likely to be between these extremes, although the exact outcome cannot be determined. The approach used allows the identification and mapping of the most probable areas for competitive exclusion, and so would be of use when designing subsequent studies on biotic interactions at local scales (Anderson et al., 2002; Jiménez-Valverde et al., 2007). Even when the real distribution of the species in the future remains unknown, with this approach a more detailed assessment can be made of the expected species distribution in response to change, based on the directional hypothesis about the role of each species and, when available, on previous knowledge of species' relationships obtained from local studies. This is the case for hare species in Europe (see below), although this may not be possible for every species.

On the hare species in Europe

Our results on the effects of climate change on L. granatensis and L. europaeus distributions indicate slight northward shifts in their ranges by 2080 according to the A2 emissions scenario. However, this situation does not apply to L. timidus, as the predictions suggest that its range will undergo notable reductions because of climate changes (Fig. 2). First, these results are consistent with studies which have documented differences in sensitivity to climate change in relation with the ecoregion inhabited by the species; i.e. species from the Boreo-Alpine region were more sensitive to climate change than those inhabiting more temperate regions (Thuiller et al., 2005). Thus, species occurring in colder regions should be affected by climate change because of a loss of suitable habitat. Given this context, and consistent with previous studies, the future range of L. timidus in Europe is highly dependent on climate (Jannson & Pehrson, 2007; Anderson et al., 2009). When interspecific interactions were considered the expected effects of climate change on these species were even more pronounced (Table 3); the real situation probably lies between these extremes, but unfortunately this cannot be determined at present as no validation data exist regarding the future.

We can interpret the predicted interactions based on previous knowledge about the ecology of the studied species. On the one hand, L. granatensis is expected to have some advantage over L. europaeus in their contact area as it is more favourable to the former (Fig. 3A). In addition, data obtained from population monitoring has described an increasing population trend for L. granatensis, whereas L. europaeus is declining in their contact area (Gortázar et al., 2007), thus enhancing the potential advantage of L. granatensis over L. europaeus. In this context, the effects of biotic interactions on L. granatensis could be considered negligible when the models are projected onto future scenarios. Therefore, the future distribution of L. europaeus is predicted to be negatively affected by L. graatensis, although it should be noted that L. europaeus has been suggested as a competitor able to exclude L. timidus in the borders of distribution (Thulin, 2003; and references therein). Thus, it is expected that L. europaeus would display an intermediate pattern between the reported extremes (Table 3), as even if the viability of the southern European populations from the Iberian Peninsula may be compromised, it is not expected that those of the north will be constrained by competition. Nevertheless, the last interpretation is not directly supported by our analysis; when the population status of both species was expected to be similar, then L. timidus seemed to be favoured over L. europaeus in their contact area (Fig. 3B). In our opinion, a plausible explanation would be related to unbalanced densities between these species when they co-occur; L. timidus is usually found at lower densities than L. europaeus (Jannson & Pehrson, 2007). Even if a territory is more favourable to L. timidus, differences in densities could drive a situation in which this species is disadvantaged compared with L. europaeus when resources become limited (Thulin, 2003). However, in addition to resources, exclusion mediated by hybridization (see Rhymer & Simberloff, 1996) can occur when species differ in density, especially in highly unbalanced situations. This was also suggested as a potential factor mediating the L. europaeus/L. timidus interactions (Thulin, 2003; but see Jansson et al., 2007), and even explaining the ancestral local extinction of L. timidus in the Iberian Peninsula due to displacement by L. granatensis (e.g. Melo-Ferreira et al., 2007). These range replacements with hybridization have also important implications on the genetic composition of the involved species, as gene introgression should predominantly occur from the resident into the invading species (Currat et al., 2008). Finally, future predictions for the distribution of L. timidus cannot be viewed with optimism; it is expected that by 2080 the distribution of this species will be reduced to 30% of its current range (72% under the most optimistic predictions) due to climate changes. According to our results, the future of this species will be compromised by climate change especially when biotic interactions with L. europaeus are taken into account. Thus, we suggest that climate change should be included among the factors to be monitored when addressing the conservation of L. timidus (Smith & Johnston, 2008).