Ignoring biotic interactions overestimates climate change effects: The potential response of the spotted nutcracker to changes in climate and resource plants

Projecting future distributions of species under climate change remains a particular challenge for species that are trophically interacting. Interaction partners are often assumed to react differently to climate change, causing spatial mismatches in future distributions and increased extinction risks. We compare potential direct effects of climate change with combined direct and indirect effects mediated via food plants on a highly specialized bird species.


| INTRODUC TI ON
Interactions between species are a crucial component of resilient ecosystems, but they are often neglected in assessments of possible effects of climate change (Blois, Zarnetske, Fitzpatrick, & Finnegan, 2013;HilleRisLambers, Harsch, Ettinger, Ford, & Theobald, 2013). Species range shifts lead to major changes in ecosystem structure, function and species' interactions when interacting species respond differently to climate change (Walther et al., 2002), causing predominantly negative consequences for biodiversity (Blois et al., 2013;Böhning-Gaese & Lemoine, 2004). Shifts in geographic ranges in response to anthropogenic climate change such as upward shifts in tree line and northward shifts of range boundaries have already been observed (Chen, Hill, Ohlemüller, Roy, & Thomas, 2011;Lenoir, Gégout, Marquet, de Ruffray, & Brisse, 2008;Maggini et al., 2011). For interacting species, such range shifts could lead to a mismatch of the areas where each of the species is able to persist (Kissling et al., 2010;Schweiger et al., 2012). In mutualistic animal-plant interactions such range shifts have more negative impacts on animals' than on plants' survival (Schleuning et al., 2016;Stewart et al., 2015), which makes animal species that rely on specific plant resources especially vulnerable to climate change due to the combined direct and indirect effects of climate change.
The European spotted nutcracker (Nucifraga caryocatactes L. 1758) depends on three plants providing seed sources for its winter caches, which may render the bird especially vulnerable to indirect climate change effects via its food plants. Additionally, nutcrackers prefer mountain habitats which are suggested to be particularly at risk of extensive species losses under climate change (Kissling et al., 2012;Sekercioglu, Schneider, Fay, & Loarie, 2007). Already Gregory et al. (2009) found that the nutcracker is one of the bird species which most declined with global warming between 1980 and 2005.
Here, we analyse projected future changes in the distributional range of the nutcracker in Europe using species distribution models (SDMs). We compare projections based on direct effects of climate change with projections based on combined direct and indirect effects of climate change via potential occurrence changes of the three tree species as the nutcracker's main food sources.
Classical SDMs link occurrence data to abiotic environmental variables (Elith & Leathwick, 2009), assuming that biotic interactions are unimportant at large scales and equally strong across the species' range (Kissling et al., 2012;Pearson & Dawson, 2003). A common approach to integrate interacting species into SDMs is based on a secondary combination of separate distribution models. In previous studies the projected probabilities of occurrence are converted into presence-absence distribution maps where a threshold is used to determine which probability of occurrence is considered to be a presence. Only after this conversion is the distribution of the target species restricted to the distribution of its interaction partners (Schweiger et al., 2012;Schweiger, Settele, Kudrna, Klotz, & Kühn, 2008). Some studies, however, suggest that, as the rules of transformation from probabilities of occurrence to presence-absence data are an important source of model variability (Araújo, Whittaker, Ladle, & Erhard, 2005), the use of estimates of change in probabilities of occurrence might provide more robust forecasts (Araújo, Williams, & Fuller, 2002). Here, we combine the species' occurrence probabilities in a manner that uses actual probabilities of occurrence for both the separate projections of direct climate change effects and combined projections of direct and indirect climate change effects to give a more nuanced projection of future trends.

| MATERIAL S AND ME THODS
We projected current and future probabilities of occurrence of the nutcracker and its main food sources, Swiss stone pine and common hazel, using SDMs (Guisan & Thuiller, 2005;Guisan & Zimmermann, 2000;Willis et al., 2015).

| Data
We used presence-absence data of the European nutcracker from the second European Breeding Bird Atlas (Hagemeijer & Blair, 1997) obtained from the Global Biodiversity Information Facility (GBIF.org, 2016). The data for the atlas have been collected from 1972 to 1995 with a core fieldwork period from 1985 to 1988. The available spatial resolution is 0.5° × 0.5° latitude-longitude. A comparison with more current point observation data showed that although the EBCC data was collected more than 20 years ago it is still valid at the available resolution. We obtained plant occurrence data from chorological  Enescu, Houston Durrant, Rigo, & Caudullo, 2016 We extracted current climate data and projections of future conditions from WorldClim 1.4 (Hijmans, Cameron, Parra, Jones, & Jarvis, 2005). Current data represent the time period of 1960-1990. Future nutcracker distributions were projected for 2070 (average for 2061-2080). We used future climate databased on all four representative concentration pathways (RCP2.6, 4.5, 6.0 and 8.5) produced by general circulation models (GCMs) for climate change projections in the IPCC Fifth Assessment report (IPCC, 2013). We used the mean of two different GCMs, the MIROC5 (model for interdisciplinary research on climate) and the CCSM4.0 (community climate system model) (IPCC, 2014), to increase the validity of our results.
All species and climate data were rescaled to a spatial resolution of 0.5° × 0.5° latitude-longitude.

| Modelling
We used two approaches, Boosted Regression Trees (BRT) and Maximum Entropy (MaxEnt), to model distributions of the three focal species. By using two reliable modelling algorithms we intend to increase the robustness of our results. Compared to other existing SDM methods, BRTs provide outstanding predictive discrimination (Elith et al., 2006), while MaxEnt tends to have high predictive accuracy (Merow, Smith, & Silander, 2013). MaxEnt estimates environmental suitability for the species to occur, which we considered as equivalent to the probabilities of occurrence estimated by BRT for comparison. We will further refer to both model outputs as probabilities of occurrence, however, we acknowledge the initial difference between the models' intent. The nutcracker has exceptional dispersal ability and transports the plants' seeds across large distances (Didier, 2001), so we assumed unlimited dispersal for all species. All analyses were conducted in R 3.3.3 (R Core Team, 2017), using the package 'biomod2' 3.3-7 for building SDMs (Georges & Thuiller, 2013).
Model evaluation was based on several accuracy measures. The area under the curve (AUC) of receiver operating characteristic (ROC) is a measure of discrimination capacity between presences and absences where values >0.8 are considered good and values >0.9 excellent (Elith, Kearney, & Phillips, 2010;Pearce & Ferrier, 2000). The true skill statistic (TSS) indicates a prediction success rate where a model is considered to perform accurately if it scores >0.5 at least (Allouche, Tsoar, & Kadmon, 2006;Herkt et al., 2017).
Both measures are based on sensitivity representing the models' ability to correctly predict a known occurrence and specificity representing the ability to correctly predict a known absence (Liu, White, & Newell, 2011). We used a 5-fold systematic spatial blocking technique using the 'blockCV' package 1.0.0 (Valavi, Elith, Lahoz-Monfort, & Guillera-Arroita, 2019) based on Roberts et al. (2017).
We ran the models with different parameters for the three species and chose the final settings for each based on the highest evaluation scores achieved (see Table 1).

| Modelling interaction partners
We integrated the occurrence probabilities of the food plants into the occurrence probabilities of the nutcracker by calculating the combined probability of the species occurring together for each of our projections (see Figure 1). We standardized all projected occurrence probabilities, dividing the probabilities by the maximum in each data set, because the projected probability for the Swiss stone pine only reached a maximum of 0.5 (while others reach higher), but the projected current range fitted the current realized range. We estimated the combined probability of occurrence (P int ) of the bird (P Nc ) and all food plants (P Pc , P Ca ) as the product of each pair of interacting species' occurrence probabilities. We obtained probabilities of the nutcracker occurring together with each of the food plants separately. To estimate the combined effects of all food plants on the nutcracker's occurrence probability, we summed up all combined occurrence probabilities and divided the result by the maximum number of food plants occurring within one grid cell (P Pmax ): In addition to reflecting the probabilities of food sources to be present, this calculation gives more weight to those areas where more than one food source is available, which further increases the probability of the focal species to be able to persist even if one food source decreases.
For these calculations, we make the simplifying assumption that the three species' occurrences are mathematically independent from each other. We are aware that this is not the case and that a tight dependency between plant and bird occurrences implies lower combined probabilities of occurrence, thus the true probability of co-occurrence is greater than the calculated value for P int . To address this issue we decided to rescale all models before further analysis. We divided all future and current projections by the maximum probability of occurrence of their respective current climate projections, ensuring that all projected probabilities of occurrence for all models range between 0 and 1.
To assess the projected changes in occurrence probabilities in relation to the current range as well as to potential future ranges, we chose two subsets of our study extent in addition to the full European extent for the assessment. First, we chose a subset representing changes in all areas currently being or in the future becoming suitable for the nutcracker in Europe. This was done by selecting those grid cells with projected probabilities of occurrence in either one of the scenarios larger than the prevalence of the nutcracker in the EBCC dataset (similar to the approach for threshold selection by Liu, Berry, Dawson, and Pearson (2005)).
For the second subset, we restricted our resulting maps to areas currently inhabited by the nutcracker by selecting those grid cells with presences in the EBCC data set, representing changes in the current range. We calculated the differences between current and future projected probabilities of occurrence for both areas and compared the differences using a Wilcoxon rank sum test, including a Bonferroni correction for significant p-values to correct for multiple testing due to the number of grid cells included into the test for significance (Mangiafico, 2015).

| RE SULTS
All projections showed good performances, indicated by high AUC and TSS scores for the plant's projections and slightly lower performance scores for the nutcracker projections and the combined projections ( closely resembles that of the direct effects projection of the nutcracker, while the ROC score as well as both MaxEnt combined projection scores show slightly lower evaluation scores than the direct effects projections. All models show high accuracy regarding the projections of species' current ranges, however, MaxEnt projections tend to underestimate current ranges while BRT models slightly overestimate them at the range edges. The climate-only projections of the nutcracker show north-and upward shifts with stronger decreases in the southern range than increases in the north (Figure 2i,k) and an overall decreasing range size. A projected complete loss of lower mountain ranges as climatically suitable habitats in the future is accompanied by minor increases of occurrence probabilities at higher altitudes in the Alps (Figure 2e,g).
Similar to the nutcracker, the hazel is projected to strongly increase its occurrence probabilities north of its current range with decreased occurrence probabilities in southern range parts (Figure 3j,l).

| D ISCUSS I ON
The integration of biotic interactions into projections of species ranges is widely acknowledged to play a crucial role in estimating species' occurrence probabilities and responses to climate change (Araújo & Luoto, 2007). In our study, climate-only projections show high probabilities of occurrence throughout the current range, while projections that combine direct climate change effects and indirect effects mediated via the resource plants give higher probabilities of occurrence in areas where population densities of the nutcracker are reported to be highest (Bauer & Glutz von Blotzheim, 1993). Trophic interactions seem to influence the species' occurrence not only at small scales, but throughout the projected range, as suggested by Wisz et al. (2013). In our case, however, the resource plants appear not to restrict the bird's range but its population density or probability of occurrence, while climatic factors define its range limits.   (Mattes, 1982), which is reflected in our combined models' results that give lower probabilities of occurrence there. Accordingly, the concentration of projected climate-driven range losses in lower mountain ranges leads in sum to less negative change in combined than in climate-only projections. The nutcracker's core area with highest population densities in the Alps is projected to remain stable under climate change both for the nutcracker and its food plants.
The nutcracker's core area in the Alps relies on the occurrence of the Swiss stone pine. Currently the birds are dispersing stone pines above and beyond the plants' current range (Didier, 2001;Kharuk et al., 2009), providing them with the possibility to shift their ranges with climate change. These events however appear to be rare and might only occur frequently during mast seeding years . Previous studies found significant variation in stone pine growth driven by climatic factors (Carrer, Nola, Eduard, Motta, & Urbinati, 2007;Holtmeier & Broll, 2007), highlighting possible effects of climate change on individual trees. Today's seedlings are growing under climatic conditions that are probably different to those in the future , and as the stone pine needs 40-60 years to mature and produce seeds (Ulber, Gugerli, & Bozic, 2004) our projections looking ~50 years into the future give only rough estimates of climate change effects.
Our projections suggest that the stone pine might be able to persist in its current range under climate change, providing a safe core area for the nutcracker. In contrast to our results, a small-scaled future trends analysis for the Swiss stone pine by Casalegno, Amatulli, Camia, Nelson, and Pekkarinen (2010) suggests heavy losses of suitable areas. The difference to our results could be driven by differences in the spatial scales of the analyses. Mountainous areas show high levels of climatic variation across small spatial scales where increases in large-scale average temperatures can be overruled by local microclimate and topography (Holtmeier & Broll, 2005). Such small-scale effects could not be captured by the coarse scale of our study, which calls for more detailed investigations that integrate direct and indirect effects at different levels of spatial scale.
Besides the changing temperature and precipitation regimes assessed in our study, other factors affected by climate change limit plant growth as well (Anfodillo et al., 1998;Wieser et al., 2009).
Additional biotic factors apart from the trophic interactions analysed in our study influence species ranges under climate change (Rosenzweig et al., 2007;Wieser et al., 2009). Under warming conditions invading deciduous trees have an advantage and outcompete slow-growing pine species (Fisichelli, Frelich, & Reich, 2014;Gimmi et al., 2010). Understorey vegetation and seed predation have been shown to be more influential on stone pine establishment than climatic factors . Also, there are unknown effects of pathogenic fungi and insects which depend on stand density as well as climatic conditions . The species are currently in a state where neither pathogens nor competitors pose a threat for the Swiss stone pine, but changing climate and phenology could lead to unknown effects.
Although we find some direct and indirect influences of climate on the distribution of the stone pine, its current distribution is largely shaped by historic human influences (Casalegno et al., 2010;Didier, 2001). The nutcracker's occurrence data show presences further north than its food sources. A likely reason for this discrepancy could be plantations of Siberian stone pine (Pinus sibirica) in Finland and Sweden (Bauer & Glutz von Blotzheim, 1993) which the nutcracker also uses as a food source. Some authors argue that the development of forest structures at higher altitudes does not depend on bioclimatic factors altogether (Holtmeier & Broll, 2007), but on forest management practices that have shaped alpine ecosystems (Motta & Lingua, 2005). The recent establishment of the Swiss stone pine at higher altitudes might merely be a re-establishment where currently decreasing human pressure has previously suppressed forest growth (Wieser et al., 2009). These discussions emphasise that human land use, especially forest management and grazing regimes, will play a fundamental role in the conservation of plant habitats (Casalegno et al., 2010), and thus also determine the abilities of associated animal species to persist under climate change.
The consideration of biotic interactions is crucial for the success-    and

| CON CLUS ION
an increasing proportion of Swiss stone pine at higher altitudes in the Alps (Didier, 2001;Schröter et al., 2005) could offer new habitats in the future. Vice versa, the nutcracker provides its food plants with 'extraordinary colonization capacity' (Didier, 2001), giving them a high potential for quick responses to climate warming. Increasing seed production due to increasing summer temperature might shorten the interval between good seed years (Holtmeier & Broll, 2007) and increases tree growth (Motta & Nola, 2001), offering further potentially positive effects of climate change. However, shifts from boreal to temperate vegetation in lower mountain ranges and southern latitudes might lead to range decreases as well  and effects of climate change on tree pathogens are yet unknown. Furthermore, in recent decades human land use has been shown to be the most influential factor in determining species ranges, emphasizing the need for conservation strategies that consider these multiple anthropogenic effects. Overall, the case of the European spotted nutcracker and its resource plants suggests that considering trophic interactions in modelling species responses to climate change might help reduce overestimations of potential range losses, thus stressing that species interactions should be taken into account in future models of climate change effects as well as in conservation strategies.

ACK N OWLED G EM ENTS
We would like to thank the members of the Senckenberg