Assessing the exposure of forest habitat types to projected climate change—Implications for Bavarian protected areas

Abstract Aim Due to their longevity and structure, forest ecosystems are particularly affected by climate change with consequences for their biodiversity, functioning, and services to mankind. In the European Union (EU), natural and seminatural forests are protected by the Habitats Directive and the Natura 2000 network. This study aimed to assess the exposure of three legally defined forest habitat types to climate change, namely (a) Tilio‐Acerion forests of slopes, screes, and ravines (9180*), (b) bog woodlands (91D0*), and (c) alluvial forests with Alnus glutinosa and Fraxinus excelsior (91E0*). We analyzed possible changes in their Bavarian distribution, including their potential future coverage by Natura 2000 sites. We hypothesized that protected areas (PAs) with larger elevational ranges will remain suitable for the forests as they allow for altitudinal distribution shifts. Methods To estimate changes in range size and coverage by PAs, we combined correlative species distribution models (SDMs) with spatial analyses. Ensembles of SDM‐algorithms were applied to two climate change scenarios (RCP4.5 and RCP8.5) of the HadGEM2‐ES model for the period 2061–2080. Results Our results revealed that bog woodlands experience the highest range losses (>2/3) and lowest PA coverage (max. 15% of sites with suitable conditions). Tilio‐Acerion forests exhibit opposing trends depending on the scenario, while alluvial forests are less exposed to climatic changes. As expected, the impacts of climate change are more pronounced under the “business as usual” scenario (RCP8.5). Additionally, PAs in flat landscapes are more likely to lose environmental suitability for currently established forest habitat types. Main conclusions Based on these findings, we advocate the expansion of the Natura 2000 network particularly in consideration of elevational gradients, connectivity, and projected climatic suitability. Nonclimatic stressors on forest ecosystems, especially bog woodlands, should be decreased and climate change mitigation efforts enhanced. We recommend transferring the approach to other habitat types and regions.


| INTRODUC TI ON
Besides timber and fuel production, forests are providing a series of ecosystem services for human well-being and societal interests (Brockerhoff et al., 2017). Forest ecosystems play a crucial rule, inter alia, for carbon sequestration, balancing climatic extremes and maintaining biodiversity, explaining their importance in nature conservation and for protected area networks.
In the European Union (EU), natural and seminatural forest habitat types are listed in the Habitats Directive since 1992 (European Council, 1992). In this Directive, standardized "habitat types" are defined by characteristic species assemblages and abiotic conditions. One instrument, which arose from this EU legislation, is the Natura 2000 network. It was set up as a Europe-wide network of conservation sites, which were designated under the Habitats Directive and the Birds Directive from 1979 (European Council, 1992). Covering more than one-fifth of the EU territory, Natura 2000 is declared the largest protected area network across the globe (EEA, 2015).
However, concerns have been issued whether static protected area (PA) networks will represent the species and habitats of conservation interest under projected climate change (e.g., Alagador, Cerdeira, & Araújo, 2016;Hannah, 2008;Kujala, Araújo, Thuiller, & Cabeza, 2011). According to a study by Araújo, Alagador, Cabeza, Nogués-Bravo, and Thuiller (2011), more than half of the assessed species from the EU Habitats and Birds Directives will lose suitable climatic conditions within current European PAs until 2080. Onefifth of the listed habitats is graded as threatened by climate change according to the countries' reports (Evans, 2012).
Within Annex I of the EU Habitats Directive, they are categorized as priority (indicated by asterisk in their codes). Priority natural habitat types are classified as "in danger of disappearance," which translates into enhanced conservation responsibilities for the member states (European Council, 1992). Therefore, it is of outstanding importance to investigate menaces to their persistence.
Climate change represents a major threat to the selected forest habitat types as warming combined with drought is expected to cause water stress in plants (Breshears et al., 2013). All types of forest in this study are soil moisture dependent (EC, 2013)-either by their vicinity to rivers and shallow aquifers (alluvial forests), their substrate and precipitation regime (bog woodlands) or their shady topography and reduced evapotranspiration (Tilio-Acerion forests). The restriction of these forests to specific site conditions renders them especially vulnerable to climate change, as they might not find the required conditions in places with future climatic suitability.
The study was carried out for the federal state of Bavaria in Germany. The responsibilities for nature conservation legislation and implementation, including of EU Directives, lie at this administrative level. Consequently, Natura 2000 sites are designated by regional authorities and the habitat types monitored at this scale.
The example of Bavaria was chosen because of its extensive area of woodland (about 2.6 million hectares, approx. 37% of the total area of Bavaria) ranging across a diversity of landscapes and elevation (BMEL, 2015). Bavaria has designated 11.4% of its territory to Natura 2000 sites (LfU, 2018). The individual Natura 2000 sites vary significantly in their spatial extent-from very local features to national parks, former military areas, and vast mountain ranges.
Bavaria is located between continental and alpine biogeographic regions of Europe and represents a transition zone of contrasting future rainfall trends Kovats et al., 2014; implications for forests, as they would limit the water available to plants during the vegetation period. Regarding the future temperature development, regional climate models predict increases of varying magnitude (e.g., Jacob et al., 2014). In the Alpine foothills, for example, summer warming is expected to exceed 4°C in comparison with 1971-2000 by the end of this century under an extreme scenario .
To analyze climate change threats to the selected forest habitat types, this study built on ensembles of correlative species distribution models (SDMs), which are widely used in climate change impact research and conservation planning (e.g., Meller et al., 2014;Summers, Bryan, Crossman, & Meyer, 2012). For conservation purposes, it is crucial to communicate the multitude of possible trajectories to avoid decision-making based on a singular, uncertain model projection. Ensembles of model forecasts are capable of illustrating the range and trend of projections, which vary across algorithms, model setting, global circulation models, or climate change scenarios (Araújo & New, 2007).
Using this methodology, we aimed to: (a) predict changes in the habitat types' range size and distribution, (b) estimate their potential future coverage by PAs, (c) and examine the linkage between elevational ranges within PAs and their projected environmental suitability for the considered forest habitat types. Previous studies have documented distribution shifts of tree species along elevational gradients (both upward and disparate directions) under warming climate (e.g., Morin et al., 2018;Rabasa et al., 2013). Therefore, we hypothesized that PAs with larger elevational ranges are more likely to remain hosts of currently established forests as they allow for such distribution shifts. The analyses of this study were designed to enable comparisons between habitat types with regard to the impacts of climate change in order to detect eventual needs for management and adaptation strategies.

| Distribution data and environmental predictors
The three habitat types, which are defined in the EU Habitats

| Modeling
Ensembles of species distribution models (SDMs) were created with the "biomod2"-package Modeling took place at the 30-arc-seconds resolution of the utilized climatic variables. After masking out the 10 × 10 km occurrence cells, pseudo-absences were selected randomly within the EU following the 50% prevalence approach by Liu, Berry, Dawson, and Pearson (2005).
Input datasets were split into train (70%) and test data (30%) for the evaluation of the models' performances. In a cross-validation process, TA B L E 1 Selected environmental predictors for forest habitat types utilized in correlative species distribution models the weighted mean of the ensembles for further analysis. Using the threshold that maximizes the sum of sensitivity and specificity (Liu, White, & Newell, 2013), the occurrence probabilities for each habitat type were converted into binary information.

| Range change, coverage analysis, and elevational gradient statistics
Based on the modeling results, range changes of the habitat types were estimated for each RCP scenario by comparing the projected future distribution with reported present-day occurrences ("BIOMOD_RangeSize"-function in "biomod2"-package). For this purpose, the binary model outputs (30 arc-seconds) were aggregated to the resolution of the original distribution data (10 × 10 km). The estimated changes in range size for the study region served as one criterion for the determination of the habitat types' exposure levels. • loss: PA hosts habitat type at present but potentially not in future; • occupied stable: PA hosts habitat type at present and potentially also in future; • gain: PA does not host habitat type at present but potentially in future.
The changing environmental suitability of the PAs was then tested against the elevational range found inside of the corresponding conservation site. The zonal statistics tool in ArcGIS 10.5 was used to assign elevation ranges to the individual PAs based on a digital elevation map (EEA, 2017). Kruskal-Wallis rank-sum tests (R package "stats" 3.5.1) compared the previously defined "changing suitability" categories of the PAs with regard to this elevation range attribute.
Post hoc tests were carried out with the "kruskalmc"-function of the R package "pgirmess" version 1.6.9 (Giraudoux, Antonietti, Beale, Pleydell, & Treglia, 2018). Wilcoxon tests (R package "stats") were executed to furthermore test whether the elevational range interferes with the future environmental suitability of PAs independent of their current occupation by the habitat types.

| Range change of forest habitat types under climate change in Bavaria, Germany
The three studied forest habitat types face very different perspectives until the 2070s. Bog woodlands react most sensitively to expected abiotic changes. Up to 94% of their small Bavarian range are projected to be lost until the second half of this century ( Table 2).
Irrespective of the scenario, <10% of the entire study region will provide suitable environmental conditions for bog woodlands.
Potential refugia are situated in the Alps and their foothills ( Figure 2).
For RCP4.5, sites in eastern Bavaria additionally remain suitable or emerge as appropriate locations.
For the Tilio-Acerion forests, the spatial response varies considerably between the two climate change scenarios. Under RCP4.5, the model projects an overall range expansion (36% relative to current range size), constituted by both losses of currently occupied cells (34%) and gains of new suitable area (70%) ( Table 2).
In contrast, the proportion of newly available areas under RCP8.5 is noticeably lower (35% relative to current range size), while the TA B L E 2 Projected range change of Tilio-Acerion forests of slopes, screes, and ravines (9180* Note: Change analysis compared observed current with projected future occupation of cells (10 × 10 km). "Total suitable grid cells" refers to environmentally suitable proportion of Bavaria's territory. Percentages for "loss," "gain," "net change" are relative to current range size. The future distribution of the habitat types was modeled as ensembles of correlative species distribution models combining GAM, GLM, GBM and RF. Climate change scenarios RCP4.5 and RCP8.5 were considered for the HadGEM2-ES model for 2061-2080.
projected losses amount to 67% of currently occupied grid cells.
As a consequence, Tilio-Acerion forests will experience an overall range contraction under RCP8.5 to about two-third of their current range size within Bavaria. Spatially, the distribution of this habitat type will shift east-and southward (Figure 2) according to the model. While the observed presences in central and northern Bavaria could still be occupied under RCP4.5, they are likely to disappear under RCP8.5.
The alluvial forests with A. glutinosa and F. excelsior prove to be less exposed. For both climate change scenarios, this habitat type is projected to not shrink. Focussing on the fundamental environmental suitability of space rather than the exact locations of watercourses, the alluvial forests could slightly expand their range under RCP4.5 (9% gain relative to current range size) ( Table 2).

| Future coverage of forest habitat types in the Bavarian Natura 2000 network
One important criterion when assessing the endangerment of forest types is their coverage by PAs. The intersection between modeled future distribution of the forest habitat types and the locations of Bavarian Natura 2000 sites revealed distinct differences in potential representation (

| Elevational range as predictor for future suitability of protected areas
In addition to the exposure assessments for the forest habitat types, it was analyzed whether larger elevational ranges inside of conservation sites favor their future role as refuge for endangered habitat types. The statistical results (Figure 3, Appendix S3) indicate that PAs differ significantly in their future potential for hosting two out of three habitat types (bog woodlands and Tilio-Acerion forests) depending on the elevational range found inside of them. Partially confirming the initial assumption, PAs with larger elevational ranges are more likely to maintain environmental suitability for these two habitat types for both climate change scenarios. For the widespread alluvial forests such a pattern could not be found.

| Model evaluation
Evaluating the quality of the models, projections for the reference period of 1970-2000 were mostly over-predictive in comparison with the currently observed distributions within Bavaria (compare Figure 1 and Appendix S4). On European scale, the ensemble models were capable to project the range of the forest habitat types adequately (Table S5.1 and Figure S1.1). Comparing expert-based ecology descriptions for the key tree species (e.g., EC, 2013; IUCN, 2018) with generated response curves (Appendix S6) and variable importance rankings (Table S7.1) permitted an additional assessment of the models' plausibility. Known differences of the forest types with regard to frost sensitivity, for example, were captured well by the models. The corresponding variables "minimum temperature of the coldest month" or "mean temperature of the coldest quarter" were crucial to explain current distribution patterns of all three habitat types. While bog woodlands tolerate cold temperatures, which is emphasized by their range expansion toward Northern Europe ( Figure S1.1), Tilio-Acerion forests and alluvial forests react sensitively toward frost. The different ecological requirements also serve to reason the reaction of the individual habitat types to climate change.  (2014), distribution shifts of habitat types are less likely than the contraction of their ranges. In our study, bog woodlands are most threatened, followed by Tilio-Acerion forests of slopes, screes, and ravines. The alluvial forests with A. glutinosa and F. excelsior seem to be remarkably less exposed to climatic changes.

| Exposure differences among forest habitat types
For bog woodlands, projects by the German Federal Agency for Nature Conservation (Bittner & Beierkuhnlein, 2014)  Note: Based on intersection of observed current and projected future distribution of habitat types with protected area (PA) polygons. For detailed description on models, see Section 2 and caption of Table 2. Change classes of PAs: "unoccupied stable" (no current host and no future host of habitat type), "loss" (current host of habitat type but no future host), "occupied stable" (current and future host of habitat type) or "gain" (no current host but future host of habitat type).
which are characteristic to bog woodlands, other studies projected vast declines for their European range (Takolander, Hickler, Meller, & Cabeza, 2019) and their climate envelopes' accordance with the climatic conditions of Bavaria (Kölling & Zimmermann, 2007). Based on the identified high importance of cold minimum temperatures for this habitat type, we conclude that expected milder temperatures cause the projected negative range trends of bog woodlands.
Seasonal decreases in precipitation might be a further threat to bog woodlands, as they depend on specific soil water conditions (EC, 2013). Due to their low pH values and restrictions in nutrient availability (see Figure S6.2; Table S7.1), bog woodlands are additionally narrowed to sites that will not evolve within short time scales.
Focusing on the tree species of Tilio-Acerion forests, a trade-off between the thermophilic character of T. cordata and T. platyphyllos (Ellenberg & Leuschner, 2010) Kölling and Zimmermann (2007), as well as Bittner and Beierkuhnlein (2014), therefore potentially originate in differences of climate change scenarios, climate models, time period, and utilized variables. These studies found the region as a stable host for Tilio-Acerion forests and identified a low susceptibility of key tree species to climate change.
However, model results for single species cannot be translated directly into the development of the corresponding habitat type.

| Implications for nature conservation
This study demonstrated that all examined forest habitat types face less favorable conditions under more intense climatic changes Spatially, conservation gaps might form in the east and south of the study region. Especially, the Alpine foothills and Alps serve as refugia to the two more threatened habitat types.
In the selection process for new Natura 2000 sites, connectivity, coherence, area, redundancy, and climate change concerns need to be considered (Hannah, 2008). Connecting corridors or stepping stones should link recent occurrences with regions of projected future suitability and allow for genetic exchange between populations to foster adaptation (e.g., Keeley et al., 2018;Nuñez et al., 2013). As elevational ranges demonstrably influence the future potential of PAs to host certain habitat types, we add the inclusion of elevational ranges to these recommendations. Other researchers have already mentioned diversity of topography and "topoclimate" as necessary criteria (Heller et al., 2015;Nadeau, Fuller, & Rosenblatt, 2015). PAs flexible in space and time represent another progressive conservation concept (Bull, Suttle, Singh, & Milner-Gulland, 2013;Hannah, 2008 Paletto, 2018;Kati et al., 2015;Paletto et al., 2019). They suggest to enable participatory approaches which involve local stakeholders, for example, foresters and land owners.
Particularly for tree species and forest habitats, assisted colonization is an option worth considering (e.g., Williams & Dumroese, 2013). Kreyling et al. (2011) summarize the advantages and risks connected to this conservation concept. In a forest context, this would include the support of better-adapted tree species or provenances and increasing overall heterogeneity and diversity. With assisted migration, the barriers imposed by landscape fragmentation and limited dispersal abilities of tree species could be reduced. The improvement of the management of Natura 2000 sites plays another important role in safeguarding the priority habitat types (Geyer, Kreft, Jeltsch, & Ibisch, 2017).
Conservation measures for the here considered forest habitat types include, for example, the restoration of natural hydrological conditions and the removal of exotic plants (Hughes, del Tánago, & Mountford, 2012;Stiftung Naturschutzfonds Brandenburg, n.d.).
In the context of changing environmental suitability for tar- Questions arise whether the EU Habitats Directive provides the means to adapt conservation strategies and the protected area network as described above (Cliquet, 2014
A major obstacle lies in the availability of data: Firstly, monitoring data for EU habitat types is only available at a 10 × 10 km resolution. However, Bavarian Natura 2000 sites are often smaller.
Secondly, distribution data from non-EU countries would be useful, as nature does not conform with political boundaries. In addition, habitat types are interpreted differently across countries and cannot be described by the mere sum of their constituting species (Berry, 2012;Bittner et al., 2011;Evans, 2012

| Outlook
Building on the here modeled future environmental suitability of Bavaria for the considered forest habitat types, precise locations for complementary PAs can be identified in a next step by including mask layers for land cover and watercourses. This step will be particularly crucial to increase the reliability of suitability maps for the alluvial forests. We advocate to consider additional climate models and scenarios in subsequent ensemble modeling studies to further reduce uncertainties related to the climate development itself (Buisson, Thuiller, Casajus, Lek, & Grenouillet, 2010;Peterson et al., 2018).
Moving ahead methodologically, the monitoring of habitats from the EU Directive needs to be conducted at a better resolution. To further improve the quality of input data, research on nonequilibrium states of the observed distributions of European forests should be strengthened (García-Valdés, Zavala, Araújo, & Purves, 2013). In order to advance the models themselves, we call for the incorporation of three additional factors: the competition by alien species, scenarios on forest management options, and the dispersal abilities of the habitat types' constituting species. For the latter, supplementary research is necessary to fully comprehend the translation of species' dispersal into the more complex process of habitat shifts.
Moreover, management, such as assisted colonization, enables the establishment of trees even in naturally "unreachable" grid cells.
Expanding the analysis to a continental scale could provide insights into the overall perspectives of the forests, detect large-scale refugia for the most threatened bog woodlands, and determine Bavaria's role in conserving the selected habitat types. Comparable research is needed to improve the understanding of climate change impacts on other EU habitat types, especially the prioritized ones.
Ultimately, this research addressed essential knowledge gaps regarding the future conservation status of protected EU forest habitat types and the risks they face under climate change. Combining range change analyses based on correlative SDMs with estimates on the coverage by PAs, different levels of exposure of three moisture-dependent habitats toward climatic changes were identified.
The study therefore offers a methodology for conservation-oriented research questions in the face of climate change. For the investigated habitat types (Tilio-Acerion forests of slopes, screes, and ravines; bog woodlands; alluvial forests with A. glutinosa and F. excelsior), first suggestions for conservation strategies were derived. By the second half of the century, practitioners will be confronted with altered climatic conditions of currently established PAs.
Environmental suitability maps and exposure comparisons of conservation targets can support them by allocating limited resources to most threatened biota and improving the Bavarian Natura 2000 network under here identified criteria. As the EU law requires favorable conservation statuses for all listed natural habitat types (European Council, 1992), we advocate the evaluation of future impacts on protected habitats to initiate informed conservation strategies.

ACK N OWLED G M ENTS
This project has received funding from the European Union's

CO N FLI C T O F I NTE R E S T
The authors declare no conflict of interest.

AUTH O R CO NTR I B UTI O N S
While all authors conceptualized the study, Claudia Steinacker conducted the analysis, interpreted the results, and led the writing process. Carl Beierkuhnlein and Anja Jaeschke supervised this study and reviewed the manuscript.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data used for this study were derived from publicly available datasets. The sources are listed in Appendix S1.2 as well as in the reference list. The R code, produced ensemble models, range change maps, and protected area shapefiles can be accessed under https :// doi.org/10.5281/zenodo.3532892 of the Zenodo Repository.