Advertisement

Regional adaptation of European beech (Fagus sylvatica) to drought in Central European conditions considering environmental suitability and economic implications

  • Lukas BaumbachEmail author
  • Aidin Niamir
  • Thomas Hickler
  • Rasoul Yousefpour
Original Article

Abstract

European beech (Fagus sylvatica) is a widespread deciduous tree in Europe, but may face significant distribution shifts due to expected increasing drought frequency under climate change. An alternative adaptation strategy for beech forests to improve drought tolerance and economic outcome consists in the admixtion of silver fir (Abies alba). To explore potentially suitable areas for mixing under future climate conditions, species distribution models (SDMs) represent a useful tool, but should be accompanied by economic analyses and uncertainty evaluations to serve as a solid decision basis for forest management. Therefore, in this study, we apply state-of-the-art SDMs, review uncertainties resulting from different modeling approaches, estimate the economic value of pure and mixed beech and fir stands, and discuss managerial implications of the results in Germany. Our model results projected widespread beech declines for Germany, while silver fir distributions remained largely constant. The degree of decline varied significantly between the investigated climate scenarios and resulted in associated economic losses between − 180 and − 4000 billion euros. With regard to the uncertain magnitude of climate change and the risk of high economic losses, we recommend an adaptation of beech forests in its projected hot spots of decline and find silver fir to be an environmentally suitable mixing species. The combination of ecological, economic, and uncertainty analyses used here represents a promising set of tools to evaluate climate change effects and assist in the regional adaptation of forests.

Keywords

Climate change Species distribution modeling Species mixing Adaptive management 

Introduction

Changing temperature and precipitation patterns resulting from climate change are expected to significantly alter the spatial distribution of tree species over the next decades. Already now, drought-induced tree mortality and species range shifts are reshaping forest ecosystems and posing serious challenges for forest management (Allen et al. 2010; Bonan 2008; Lindner et al. 2010). In Central Europe, first signs of these developments have become visible for the widespread European beech (Fagus sylvatica), which is showing declines at the southern edges and low altitudes of its current distribution range (Jump et al. 2006; Geßler et al. 2007). Without forest management intervention further beech declines of up to − 25% have been projected until the year 2100, while particularly oak species (e.g., Quercus cerris, Quercus robur, and Quercus petraea) could profit from these new conditions and spread through large parts of Central Europe (Hanewinkel et al. 2013; Spellmann et al. 2015). Due to their lower timber productivity (Pretzsch et al. 2013b; BMEL 2014), a large-scale shift towards oak would result in considerable economic losses for forest owners and the timber industry in Central Europe and lead to reduced carbon sequestration (Hanewinkel et al. 2013). This may particularly apply for Germany, where European beech represents one of the most important commercially used trees (BMEL 2014).

In response to this outlook, the transition of pure to mixed species stands is considered a viable adaptation strategy since the latter are more resistant against and may recover more rapidly from disturbances, reduce economic risk, increase stand structure diversity and biodiversity, and may show comparable or even higher yields than pure stands (Pretzsch et al. 2013a; Griess et al. 2012; Forrester 2014). Mixing experiments for beech have found a higher availability of water, due to interspecific complementarity, to be decisive for the enhanced performance of the mixed stands (Mölder and Leuschner 2014; Metz et al. 2016). This complementarity may be achieved through mixing species with different root systems, resulting in reduced competition, increased fine-rooting (Bolte et al. 2013), and in some cases even inter-specific facilitation (Kelty 2006). In this context, the deep-rooting silver fir (Abies alba) has been proposed as a compatible mixing species for the shallow-rooting beech, since it shares similar environmental requirements with beech (Ellenberg 1988), is moderately drought tolerant (Zang et al. 2014), and may lift and redistribute ground water to the topsoil in times of drought. This “hydraulic lift” or “hydraulic redistribution” has been observed across different ecosystems including mixed beech forests (Zapater et al. 2011; Neumann and Cardon 2012; Sardans and Peñuelas 2014) and may occur when the water gradient between roots and soil is high (e.g., during droughts), leading to water exudation from the roots (Fig. 1). Recent studies of mixed beech–silver fir stands point to a positive mixing effect and suggest the admixing of silver fir to beech as a potential silvicultural adaptation measure (Lebourgeois et al. 2013; Magh et al. 2017).
Fig. 1

Conceptual scheme of upwards hydraulic redistribution (Devi et al. 2016)

To investigate suitable areas for species-based adaptation, statistical species distribution models (SDMs) represent a practical tool, since they allow projecting species distributions and environmental suitability under future climate conditions at large spatial extents. While SDMs are already widely applied in studies on climate change effects on species (including beech, e.g., Serra-Diaz et al. 2012; Zimmermann et al. 2016, Dyderski et al. 2018), uncertainties about methodological choices and limits of model validity still represent major issues in this field (e.g., as described in Jimenez-Valverde et al. 2009; Van der Wal et al. 2009; Acevedo et al. 2012; Barbet-Massin et al. 2012; Cheaib et al. 2012). The adequateness of the resulting environmental suitability or climate risk maps to support forest management decisions may thus vary considerably as a result of the choice of the modeling approach or experimental setting. These uncertainties, however, are rarely communicated to decision makers.

Another shortcoming of many existing studies on species distributions is their confinement to ecological evaluation. In the field of climate adaptation of forests, many authors have thus called for an enhanced integrated use of the results of ecological models, e.g., with social, political, and economic sciences (Bolte et al. 2009; Yousefpour et al. 2012; Keenan 2015). Recently, researchers have been attracted to the economic valuation of potential future distributions of tree species (e.g., Hanewinkel et al. 2013). To draw practical recommendations from these findings, further research on regional and local scales is needed. At the same time, to be of relevance for decision-making, any such approach on the basis of modeling should be aimed at robustness. Here, we define this term as the stability of outcomes in an ensemble of different model settings. To this end, model projections should both consider external uncertainties (e.g., different climate scenarios) as well as internal uncertainties (e.g., different modeling scales or model algorithms) (Dormann 2007; Araújo and New 2007).

In this study, we investigate the environmental suitability of European beech and silver fir under climate change, consider economic implications, and evaluate adaptation strategies of beech forests in Germany. To this end, we apply an ensemble of multiple state-of-the-art SDMs, discuss the consequences of the choices of modeling approach and settings, distinguish uncertain from robust results, calculate the economic value of the modeled forest area, and discuss resulting implications for forest managers from an economic and managerial viewpoint.

Materials and methods

Study area

The study was carried out for the entire land area of Germany. To analyze modeling uncertainties resulting from fitting the model at the national scale in contrast to fitting it at the whole European range of the two species, one half of the SDMs were calibrated with national occurrence and environmental driver data and the other half with European occurrence and environmental driver data (also see “Species distribution modeling approach”). In detail, for the national-scale approach, the data was drawn from the area of Germany plus a 50 km buffer around the German borders. This decision was made (1) to avoid a fragmentation of ecosystems through administrative borders (e.g., between the Bavarian and Bohemian Forest) and (2) to cover edge effects around areas of species occurrence (also compare Acevedo et al. 2012). For the European-scale calibration on the other hand, the data was drawn from all of Europe excluding Russia (Fig. 2, panel a).
Fig. 2

Visualization of the EU-Forest species occurrence data set (Mauri et al. 2017; Strona et al. 2016) showing beech (green), fir (blue), and their mixed forest distributions (orange) for Europe (panel a) and Germany (panel b)

Data

Tree species occurrence data

The study was preceded by a comprehensive review of available species occurrence data for European beech and silver fir on a continental and national level. In the end, the EU-Forest dataset (Mauri et al. 2017; Strona et al. 2016; see Fig. 2) was considered the most appropriate due to its relatively narrow gridding (being composed of national data sets), large sample size, and widespread geographical coverage, which allows for an appropriate representation of environmentally suitable areas for both species. Since the data included species presences only and most of the modeling methods required a binary data structure, absence data needed to be generated. This step was directly implemented into the modeling procedure and is described in “Species distribution modeling approach.”

Environmental predictors

Predictors were selected based on ecological constraints of the two species known from literature. Although occurring in most parts of Germany, European beech favors a rather narrow range of climatic conditions. It is susceptible to high summer temperatures, waterlogging, and drought, which is mainly attributed to its shallow root system (Fang and Lechowicz 2006; Geßler et al. 2007; Hauck et al. 2015). Due to its sensitivity to waterlogging, beech requires easily penetrable soils with good drainage, which is why it is found more often on hillsides rather than in valleys (Houston Durrant et al. 2016). Beech tolerates a wide range of pH but prefers calcareous over acidic sites (Houston Durrant et al. 2016). Silver fir on the other hand is mostly restricted to the higher altitudes of Southern Germany. Like beech, fir tolerates a broad range of pH and prefers non-compacted soils (Mauri et al. 2016). In general, it favors cool and moist conditions but is not as sensitive to high summer temperatures as beech (Mauri et al. 2016).

To allow for projections over time, the selection of the climatic variables was further limited by data availability. The climatic predictors were additionally selected through a manual stepwise VIF (Variance Inflation Factor) calculation to avoid model overestimation due to multi-collinearity. During this procedure, for example, the minimum temperature of the coldest month was excluded in favor of the maximum temperature of the warmest month, since the former plays a stronger role in Eastern Europe and the latter is expected to have an increased effect on both species under climate change.

The final set of predictors included maximum temperature of the warmest month, precipitation of the warmest quarter, annual precipitation, soil pH, soil clay and sand content, soil depth to bedrock, elevation, slope, and aspect. The use of maximum temperatures was preferred over averages, since temperature extremes are expected to have greater influence on the range limits of beech (Zimmermann et al. 2009). In fact, extreme temperature events have already caused strong impacts on vegetation particularly over the last 2 decades and are projected to continue to do so in the future (e.g., Coumou and Rahmstorf 2012; Baumbach et al. 2017).

Current climate data were obtained from the CHELSA dataset Version 1.2 averaged over the years 1950–2000 (Karger et al. 2017). Soil data was retrieved from the ISRIC project SoilGrids250m (Hengl et al. 2017). The digital elevation model GTOPO30 (US Geological Service 1996) was used for elevation data and the derivation of aspect and slope. Projected data of future climate were obtained from the WorldClim database (Hijmans et al. 2005) averaged over the years 2041–2060. Altogether, nine different climate scenarios were considered, including the three Generalized Circulation Models (GCM) CCSM4, HadGEM2-AO, and MPI-ESM-LR, and the three Representative Concentration Pathways (RCP) RCP2.6 (low), RCP4.5 (intermediate), and RCP8.5 (high), representing an anthropogenic radiative forcing of 2.6, 4.5, and 8.5 W/m2, respectively.

The predictors were resampled to the same resolution (for practical reasons the CHELSA data resolution of ∼ 1 km2 was chosen), cropped to the same extent using ArcMap version 10.4.1 (ESRI 2011) and the R “raster” package (Hijmans 2016) in R version 3.2.4 (R Core Team 2017). Slope and aspect were calculated from the projected DEM in ArcMap.

To avoid overestimation of the models, all predictors were checked for multi-collinearity by calculating the Variance Inflation Factor (VIF). While there is an ongoing debate about the threshold above which a correction for multi-collinearity ought to be considered, a value greater than 10 is commonly seen as critical (Hair et al. 1995; Menard 1995). As all calculated VIF values were below 10, no further action was needed.

Species distribution modeling approach

Modeling method

The number of modeling methods applicable to species distribution modeling has been constantly growing over the last few decades. While there has been considerable praise for each of them (Elith et al. 2006; Drake et al. 2006; Duan et al. 2014; Merow et al. 2014), the selection of an appropriate method for studies on distribution modeling has become a difficult task. The mere step of model selection already represents a major source of uncertainty, even more so when the model is meant to be transferred over time. Some authors have thus recommended employing multiple modeling methods (i.e., a model ensemble) to increase model robustness (Guisan and Thuiller 2005; Araújo and New 2007).

To this end, we conducted a test run, in which we compared the performance of the six modeling methods Generalized Linear Models (GLM), Generalized Additive Models (GAM), Support Vector Machines (SVM), Boosted Regression Trees (BRT), Random Forests (RF), and Maximum Entropy (MaxEnt) for beech and fir. As a result, the methods GLM, GAM, and SVM outperformed the other methods (evaluation procedure as described in section 2.3.4) and were thus selected for this study (see Supplementary).

Pseudo-absence sampling

Because true absence data (i.e., recorded absences) was not available, a substitution with artificially sampled (or pseudo-) absences was necessary. In theory, pseudo-absence-sampling based on a distance from the environmental niche of a species is the most plausible strategy. In practice, however, the choice of an appropriate distance—whether environmental or geographic—plays a more crucial role (Lobo et al. 2010). Since the geographic distance approach can be applied more straightforward, in this study, we sampled pseudo-absence points randomly throughout the study area except for a buffer around the presence points. To determine an appropriate size of this buffer, the standard deviation of the distances between the presence points was calculated, which averaged to approximately 3 km. Based on this insight, a radius of the same distance was drawn around the presences, outside of which pseudo-absences were randomly sampled.

Model fitting

The models were created with the help of the R package “sdm” by Naimi and Araújo (2016). To account for uncertainties related to absence-sampling and the accuracy of the presence data, a Monte Carlo method was applied by running 20 iterations for each modeling method. During each iteration, presences were randomly sub-sampled from the presence-only data set and pseudo-absences were randomly sampled from the study area outside the buffer areas. For the German-scale-calibrated models, each iteration included 500 presences and 500 absences, while for the European models, these numbers were increase to 2500 each. Additionally, 40% of this data was set aside as test data during each iteration.

Model evaluation

The performance of the individual models was evaluated based on their Area Under the Receiver Operating Characteristics Curve (in short, AUC for Area Under the Curve) and model calibration. The AUC represents the probability value, for that a model correctly identifies a random test site as presence or absence and hence is an indicator for a model’s discrimination capacity (Phillips et al. 2009). In this study, we calculated the AUC using dependent test data (i.e., cross-validation by subsampling). Model calibration is a measure for the goodness of fit between the predicted and observed occurrences and delivers important information on the transferability of a model over space or time (Acevedo et al. 2016). Here, the calibration quality is inferred from the calibration plot and Miller’s calibration statistic (Miller et al. 1991). While in theory it is desirable to build a model that achieves both a high discrimination capacity and a high goodness of fit, in practice there is mostly a trade-off between both qualities (Lobo et al. 2010). The model evaluation in this study therefore is based on both measures.

Since the future climate in Germany is expected to exceed the range of the current climatic conditions at least in some regions (particularly in Southwest and Eastern Germany) and at least at one tail of the distribution (Zebisch et al. 2005), it is important to consider the consequences when interpreting the models (Fitzpatrick and Hargrove 2009). Here, this uncertainty was addressed by analyzing the current range of climatic parameters against their future versions for different scenarios. The results are presented as categorized maps which highlight areas where future climate parameters will exhibit values beyond the current calibration range. These areas with a “novel climate” depict regions with higher uncertainty regarding future prediction results.

Managerial implications

At least three findings with relevance for forest management may be derived from the distribution modeling results: (1) areas where beech environmental suitability may decline, (2) the mean occurrence probabilities of silver fir at sites where beech declines, and (3) areas where silver fir occurrence probability increases. The first highlights regions which are particularly susceptible to climate change. The second finding presents an overview of which areas would also be suitable for silver fir. There—on the assumption that the hypothesis of interspecific facilitation holds true—admixing silver fir to beech forests would represent a potential strategy to reduce the drought sensitivity of beech.

Economic valuation approach

While SDMs are useful for providing insights on the environmentally suitable areas for the different species, the resulting management options can only be fully assessed through an economic cost-benefit analysis. Basic to any management decision is a forest’s profitability over time, which can be expressed in terms of its Land Expectation Value (LEV). The LEV reflects a forest’s net present value over an infinite number of rotation cycles (Straka and Bullard 1996) and can be calculated using the equation of Faustmann (1849):
$$ LEV=\frac{\sum \limits_{t=1}^R\left({v}_t-{c}_t\right){q}^{R-t}}{q^R-1} $$
(1)
vtct = net revenue (value of harvested timber minus harvesting costs) at time t, q = discounting factor (1 + interest rate), and R = rotation time.

Possible forest transformation costs are not included in this equation, since for this study this calculation is not meant to depict a process but only compare different states of the forest. Furthermore, administrative costs were assumed to stay constant and thus do not appear in the equation. Our main results are based on an interest rate of 2%, representing a middle ground between existing studies (e.g., Dieter 2001; Yousefpour et al. 2010; Hanewinkel et al. 2013). To investigate the influence of this assumption, we additionally conducted a sensitivity analysis for different interest rates of 1, 2, and 3% (compare Hanewinkel et al. 2009).

For the calculation of average LEVs for beech and silver fir forests in Germany, a simulation of an exemplary beech, silver fir, and beech–silver fir mixed stand was conducted using the empirical individual-tree growth simulator BWINPro-S (Schröder et al. 2007). The simulator is a regional BWINPRO version modified by the Technical University of Dresden (Döbbeler et al. 2011) and has already proven a valuable tool in previous studies (see, e.g., Schröder et al. 2006; Yousefpour et al. 2010; Linkevicius 2014).

The simulations were run for stands of 1 ha each and were adjusted to represent average site quality in Germany. To this end, the data sets for all three stands (pure beech, pure fir, and mixed) were automatically created using averages of tree growth parameters (e.g., average productivity, heights, diameters, and stand density) from the third German National Forest Inventory (Thünen-Institut 2012). For the mixed stand, an arbitrary mixing ratio of 50:50 (referring to basal area) was assumed. The simulations of all stands were started at an age of 30, since no reliable information was available for younger trees. Other general simulation settings echoed those used by Wonsack (2016): competition index c66 = 6 (Nagel 1999), mortality via critical crown closure, length of one simulation cycle = 5 years, number of simulation cycles per simulation period = 2, and no automatic sorting or thinning. The simulation was continued in this manner until the stand was considered ready for harvesting, i.e., when the crop trees reached the predefined target diameter (see below). Simulation settings for harvesting costs were adopted from Wonsack (2016). Within the final step of monetarization, wood prices were based on 2016 annual reports by the State Forest Service of Baden-Württemberg (henceforth “ForstBW”) (Wonsack 2016). The stand was managed based on a business as usual management strategy (i.e., even-aged management), which was derived from the most current ForstBW guideline for forest development types (Wicht-Lückge et al. 2014). The pure and mixed beech stands were managed according to the recommendation for the type beech-conifer even-aged mixed forests and the pure fir stand was managed based on recommendations for Norway spruce (Picea abies) even-aged forests (for details see Supplementary).

The LEV results of the stand simulations can be understood as optimum values, since the stands were grown in isolation and without the influence of disturbances. To determine the environmental suitability values corresponding to such an optimal LEV, we assumed all values above a certain model discrimination threshold as optimal (i.e., equal to 100% LEV). This threshold technique is commonly applied to binarize SDM outputs from occurrence probabilities into presence/absence. While different approaches for setting this threshold exist (e.g., arbitrary setting to 0.5, maximum kappa, sensitivity = specificity), we chose the maximum sum of sensitivity + specificity (as in Dyderski et al. 2018), which was calculated for each combination of modeling method and fitting scale separately. Any losses that occurred above this threshold were not accounted for, since the environmental suitability was considered still high enough for the species to not result in direct economic damage. In contrast, negative changes below the threshold were converted into direct economic losses, since the suitability was considered suboptimal and at the edge of the species’ niche. Following this, the LEV changes were calculated for each pixel i using this equation:
$$ {\varDelta LEV}_i={LEV}_{sim}\times {d}_i\times {A}_i $$
(2)

LEVsim = land expectation value from the stand simulation; di = difference of the environmental suitability value of the pixel i accounting for the discrimination threshold; Ai = area of the pixel i (corrected for latitude).

Results

All model results are presented and compared for the extent of Germany. However, to distinguish between the different calibration scales more easily, in the following we will refer to models trained with data from Germany + 50 km buffer as “German” models and to models trained with data from all of Europe as “European” models.

Model performance

With the exception of the European GLM for beech, all models achieved high AUCs (Table 1). In general, the models for fir resulted in higher AUC values than for beech. Among the fir models, the AUC values tended to be higher for the models trained with German data than for the models trained at the scale of Europe. This trend was inversed for beech, again with the exception of the European GLM model. On average, all models achieved a very good calibration with no significant differences between the species (Table 1). However, the German models showed considerable deviations between the single iterations for all methods, resulting in a high root-mean-square-error (RMSE) which sometimes amounted to double the values of the corresponding European models.
Table 1

Average Area Under the Curve (AUC), Miller’s calibration (CAL), and root-mean-square-error for the calibration (RMSE) for beech and fir for all modeling methods and model fitting scales (EU Europe, GER Germany)

 

GAM

GLM

SVM

AUC

CAL

RMSE

AUC

CAL

RMSE

AUC

CAL

RMSE

Beech EU

0.90

0.93

0.04

0.76

0.85

0.08

0.91

0.83

0.06

Beech GER

0.86

0.84

0.08

0.85

0.86

0.08

0.88

0.78

0.08

Fir EU

0.94

0.93

0.05

0.91

0.90

0.06

0.95

0.86

0.05

Fir GER

0.96

0.87

0.12

0.95

0.87

0.11

0.96

0.84

0.09

Overall, the model training was successful for all methods trained with European data. The standard deviations were low across all model iterations, which indicates a good methodological robustness (also see Supplementary). The high AUC values further indicate that most methods possess a comparably high discrimination capacity. Accordingly, for the current state of the environment, the models represent accurate classifiers for the presence or absence of the investigated species. Further, the measured calibration showed good results for all methods, which represents good conditions to transfer the models over time.

While the German models achieved even higher discrimination capacities, the increased RMSE of the calibration points to a lower capacity of the models to accurately predict the environmental suitability for new or unseen data. Transferring these models in time may thus result in higher uncertainty, especially for extreme climate scenarios.

Model projections for current conditions

According to the results of the German models, beech has high occurrence probabilities in Central and Southwest Germany and generally low environmental suitability in Northern Germany, with a few more suitable patches along the coastal areas of the North Sea and Baltic Sea (see Supplementary). All modeling methods largely agreed on this pattern with only small regional differences. The European models also showed a general division between north and south, yet the suitable area appeared less fragmented and extended also to Bavaria and the Northwestern German coast. Models calibrated at the European scale also yielded larger areas of high environmental suitability than models calibrated on national data. In addition, differences between the modeling methods were more pronounced for the European than for the German models, particularly with the GLM algorithm. The analysis of standard deviations between the projections for beech overall showed low values and did not reveal significant anomalies. Only small patches along the Central German Uplands and the Alpine foreland showed slightly higher, yet negligible deviations.

The German model results for silver fir revealed a more contrasted pattern with close to zero environmental suitability in Northern Germany and medium to high occurrence probabilities in Southern Germany. The latter mainly applied to the mountain ranges in Bavaria and Baden-Württemberg, most notably the Black Forest, Swabian Alps, Bavarian Forest, and the Alpine foreland. These patterns were consistent across all modeling methods. The European fir models on the other hand showed significantly larger suitable areas, extending from the Alpine Foreland to Central Germany with only few areas of lower occurrence probabilities in between (e.g., the Upper Rhine Valley and the Neckar Basin). Discrepancies between the modeling methods were smaller than for beech, yet the results of GAM and SVM again were more consistent compared to the GLM. The standard deviations were very low for the European fir models, while for the German fir models only the GAM showed slightly increased values for Southwest Germany.

Model projections for future conditions

Since the combinations of scenarios, modeling methods and different fitting scales produced a wide range of results, we focus on the robustness and uncertainties of the results in the following (for detailed results see Supplementary).

In terms of uncertainties, the results showed significant differences between the European and German models throughout all scenarios and for both species. While the core areas of beech’s current distribution remained at a largely constant environmental suitability, particularly the distribution edges showed diverse responses to a changed climate. For the case of the European models, all of them projected widespread linear declines of beech with increasing RCP strength. In contrast, a majority of the German models projected increasing extents of suitable areas for beech from RCP2.6 to RCP8.5. These mostly extended around the current core areas in Central and Southern Germany, but also expanded towards Northwestern Germany for the higher RCPs. However, the projections of the German models differed strongly between the single iterations, which goes in line with the aforementioned high calibration deviations. The European models on the other hand produced consistent projections throughout all iterations.

A view to the areas of “novel climate” (i.e., future climate conditions exceeding the range present in the model training areas) reveals that for the German-trained models large portions of the country fall under this definition. Particularly for the HadGEM2-AO scenarios and the CCSM4 RCP4.5 and 8.5, at least half of the area of Germany lies outside the calibration range of the German models. Consequently, their projections to these areas need to be seen as unreliable. Due to this circumstance, we excluded the German models from further analysis and continued only with the European models.

The robustness of the European model findings for the environmental suitability change of beech is analyzed in Fig. 3 (a). Both positive (green) and negative (red) changes are shown and ranked by robustness, where “medium” indicates common trends for two models and “high” indicates robust trends throughout all three models. Overall, the model projections agreed in their trends in 78–99% of the study area. According to this, negative trends of environmental suitability of beech are projected for most of Germany under all scenarios. Weak positive trends appeared for RCP2.6 in small areas along the North Sea coast and for RCP8.5 in larger areas of Lower Saxony for all but the HadGEM2-AO scenarios. Although the projected areas of decline thus had larger extents for the scenarios of lower radiative forcing, the degree of decline was still significantly higher under RCP8.5 for all GCMs. “Hot spots” of negative changes could be found in Southern and Central Germany, in particular within Palatinate, the Forest of Odes, the Franconian Alps, the Neckar basin, and the Thuringian Basin.
Fig. 3

Synthesis of the projections of the EU-trained models for all climate scenarios (CC = CCSM4, Had = HadGEM2-AO, MPI = MPI-ESM-LR, 2.6 = RCP2.6, 4.5 = RCP4.5, and 8.5 = RCP8.5) for the years 2041–2060. Panel a shows the direction of projected environmental suitability changes, ranked by robustness between all models. Panel b depicts the relative environmental suitability of silver fir within the highly robust areas of decline of panel a

At the same time, these regions also corresponded to a large extent with the projected suitable areas of fir. While the extent of these areas varied between the models, similar patterns appeared. For better comparability, we normalized and averaged the environmental suitability values for fir which are shown in Fig. 3 (b). Accordingly, the Black Forest and Swabian Alps showed the highest environmental suitability, followed by the Alpine foreland, Forest of Odes, Spessart, and Franconian Alps.

The general pattern of these results stayed constant over all climate scenarios. In particular, the CCSM4 and MPI-ESM-LR scenarios showed large similarities, while the HadGEM2-AO scenarios resulted in lower environmental suitabilities for silver fir with increasing RCP strength, but also larger uncertainty between the models.

Stand growth simulation

The results of the stand simulation after each simulation cycle are summarized in Table 2. The initial pure beech stand included 2108 trees. A pre-commercial thinning was conducted at the start of the simulation to remove dominant young trees. After 20 years, the top height of the stand exceeded 17 m. Consequently, 80 crop trees were marked and competitor trees removed. This release thinning was repeated after 10 years and continued at a reduced intensity of approximately 50% at the ages of 70 and 80. No further thinning was conducted thereafter, since tree heights exceeded the recommended range for release cutting. The marked crop trees reached the target diameter after 120 years and were harvested.
Table 2

Stand simulations for pure beech (B), mixed (M), and pure fir (F)

Age (years)

Harvest (m3)*

Net revenue (€)

Remaining (m3)*

Net value of the remaining stand (€)

B

M

F

B

M

F

B

M

F

B

M

F

30

16

0

29

− 105

0

− 416

76

114

110

− 843

− 1392

− 1265

40

0

94

77

0

1256

1152

140

95

137

− 1356

1103

3243

50

72

72

62

− 378

1763

2118

89

108

175

− 481

2905

7406

60

69

50

0

849

1092

0

102

149

267

1625

4835

13,184

70

38

0

0

895

0

0

133

229

377

4262

8124

19,580

80

31

0

0

901

0

0

168

320

494

4719

12,194

26,684

90

0

0

0

0

0

0

233

424

599

8171

17,986

32,744

100

0

0

691

0

0

0

309

531

0

11,457

23,450

38,528

110

0

624

0

0

396

0

15,030

27,900

120

494

0

0

18,884

*Volume refers to cubic meters of timber harvested

The beech–silver fir mixed simulation started with 1077 beech and 680 silver fir trees. After 10 years, crop trees were selected for both species (60 for beech, 100 for silver fir) and their nearby competitors removed. This treatment was continued over the next 20 years, after which both species entered the third stage of stock maintenance and thinning ceased. Silver fir reached its target diameter at the age of 90, beech followed at the age of 110. The stand was harvested thereafter. The pure silver fir stand comprised 1279 trees prior to the simulation. The initial stand already exceeded a top height of 12 m; hence, 200 crop trees were selected and released. This thinning practice was repeated after 10 years and with lower intensity after 20 years. The stock was then maintained and harvested after reaching the target diameter at an age of 100 years.

Based on the values of Table 2 and Eq. 1, at an interest rate of 2%, the land expectation value for the pure beech plantation amounted to 2450€/ha, for the mixed stand to 5318€/ha, and for the pure fir stand to 7422€/ha. The results for the sensitivity analysis showed a negative correlation between the assumed interest rate and the corresponding LEVs (1%, 3%): for pure beech 9659€/ha, 778€/ha; for pure fir 25,381€/ha, 2816€/ha; and for the mixed stand 29,592€/ha, 1677€/ha.

Economic valuation of the modeling results

All economic analyses are presented for an assumed interest rate of 2% and for the European model projections only. A synthesis of the economic valuation of the areas with a projected beech decline (applying the threshold rule described in Section 2.5) is presented in Fig. 4.
Fig. 4

Box plots depicting the range of changes (across all EU-trained SDM methods) of the land expectation value (i = 2%) between the years 1950–2000 and the years 2041–2060 by climate scenario (abbreviations as in Fig. 3). Only areas with a projected beech decline are considered

Across all scenarios, the range of projected economic losses spanned from − 180 to − 4340 billion euros. Overall, the projected economic losses were highest for the HadGEM2-AO scenarios, which outranked the impacts of the MPI-ESM-LR and CCSM4 scenarios even for its mildest scenario RCP2.6. At the same time, however, the HadGEM2-AO scenarios also resulted in the largest uncertainties between the individual modeling methods, increasing with RCP strength. The MPI-ESM-LR and CCSM4 scenarios ranked within a similar range.

While the RCP8.5 always caused significantly higher losses than the RCP2.6 across all GCMs, the RCP4.5 showed larger deviations and for each GCM positioned differently between the other RCPs.

Discussion

Simulated present-day environmental suitability

The present-day environmental suitability maps for beech closely reflect the actual distributions of beech in Germany (Thünen-Institut 2012). While the species is widespread over most parts of Germany, it is less common in Eastern Germany. Beyond ecological reasons, this circumstance is also due to the historic replacement of beech through Scots pine (Pinus sylvestris) plantations (Ellenberg 1988), which grow better on the mostly sandy soils in this region.

In the present-day simulations for silver fir, our models predicted significantly larger suitable areas than currently realized by the species. This discrepancy to a large part may be attributed to past land use and forest management decisions, which included the replacement of silver fir by beech or Norway spruce (Picea abies) (Kölling et al. 2011; Rösch 2015; Tinner et al. 2016). Although we utilized a large number of predictors for the demarcation of both species’ environmental niches, the historic human influence on their distribution range cannot be captured by classical SDMs. Palaeoecological analyses revealed that 6000–8000 years ago, mixed silver fir forests were abundant across Central Europe as far south as Central Italy at average summer temperatures of 22–25 °C (Tinner et al. 2013). Over time, these forests have been gradually driven back by increased browsing, stand transformation into broadleaf coppice (e.g., beech) after clearcutting, and more frequent fires due to cultural land use changes (Nocentini 2009). Additionally, high SO2 concentrations have contributed to the retreat of silver fir in the past (Elling et al. 2009). As a result, the fundamental climatic niche of silver fir is expected to exceed the niche currently realized by the species significantly (e.g., Tinner et al. 2013 found a realized niche of only 12–25% compared to the potential range).

Climate change effects on environmental suitability

The modeling results under the influence of climate change largely confirm existing projections for beech and fir distributions (compare Kramer et al. 2010; Kölling et al. 2011; Hanewinkel et al. 2013; Tinner et al. 2013; Saltré et al. 2015; Spellmann et al. 2015). While beech may suffer from higher temperatures and less summer precipitation in large parts of Germany, (e.g., Hauck et al. 2015), our projections also suggest significant regional disparities. According to our results, the largest decline may occur in South and Central Germany, while it may be less pronounced in East Germany. On the contrary, the Northwest may even show a higher environmental suitability for beech in the future, hence suggesting a possible northwards migration of the species (Kramer et al. 2010; Hickler et al. 2012; Hanewinkel et al. 2013). Meanwhile, the results for silver fir present promising settings for future forest management by staying largely constant. Only for the high RCP scenarios these conditions may be less favorable, which also lead to moderate declines of silver fir’s environmental suitability.

The impact of the RCPs did not always show a linear response from RCP2.6 via RCP4.5 to RCP8.5 in the economic valuation. Since this non-linearity only applied to the ranking of RCP4.5 between the other RCPs and differed between all GCMs, it is likely to stem from internal variabilities of the climate projections (Deser et al. 2012).

While our input climate data set “CHELSA” was specifically designed to improve the accuracy of species distribution models (Karger et al. 2017), uncertainties emanating from the underlying circulation models or the downscaling procedure also contribute to the “cascade of uncertainties” in the models (as described in Reyer 2013). Since we use climate data from the same source and based on the same downscaling technique both for current and future climate, this effect may be reduced in part.

From a physiological perspective, effects of changes in nitrogen cycling, deposition, and rising global CO2 concentrations may additionally affect species distributions through nitrogen deficiency, CO2 “fertilization,” and, for broad-leaved species, reduced stomatal conductance and transpiration (Leuzinger and Körner 2007; Hickler et al. 2015; Dannenmann et al. 2016). These factors are not accounted for in the SDMs; however, nitrogen deposition is unlikely to change significantly in the coming decades, and predictions for plant-physiological CO2 effects are highly uncertain and debated (Körner et al. 2005; Hickler et al. 2015). Equally, the development of forest pests under climate change represent an unknown variable (Goberville et al. 2016; Ramsfield et al. 2016), some of which thrive especially well on drought-stressed trees (e.g., bark beetle), whereas others favor moist conditions (e.g., Hymenoscyphus pseudoalbidus, a fungus responsible for ash dieback).

Modeling approach effects

Despite the relatively good performance of the models, the results showed large discrepancies between the different modeling approaches. In particular, the choice of the scale for model fitting (Europe, Germany) affected the extent of suitable areas and, more importantly, the transferability of models over time into the different climate change scenarios. In addition, the choice of the modeling method (GAM, GLM, SVM) influenced local patterns and the value range of environmental suitability.

The models calibrated on the whole species range include more provenances, with potentially rather different responses to environmental stress. A study of Czajkowski and Bolte (2006), for example, found differences in the water demand of seedlings from German and Polish beech provenances when exposed to drought. Such provenance- and region-specific responses might be better captured by the national-scale calibration. In turn, calibrating the models at a national scale also carries the drawback that beech populations occurring in warmer and drier climate than in Germany are not represented by the models. In our case, the lack of future climate analogues (particularly for maximum temperatures) in the national data used for model training caused the German models to perform poorly when projected to climate change scenarios (compare Fitzpatrick and Hargrove 2009). A solution to this problem lies in increasing the area used for training the models to also include occurrences from a broader environmental range. While this represents a convenient strategy to deal with situations of non-analog climate, it may also come at the cost of decreased discrimination capacities and a potential overestimation of the species adaptive capacity at the regional scale.

As mentioned earlier, general shortcomings of SDMs also apply in this study, most of which stem from their implicit assumption that species distributions are in an equilibrium state with the environment (Araújo and Pearson 2005; Dormann 2007). Accordingly, as applied here, the models assume that foresters only plant species where they would occur naturally and neglect the influence of deliberate species selection and land use change on the present species distributions. Additionally, SDMs are insensitive to range dynamics and do not consider migration limits of a species. Recent studies have thus discussed that projected species ranges from SDMs may be considerably more limited if coupled with dynamic range models (Zurell et al. 2016). While these limitations still represent a subject of ongoing debate about SDMs (e.g., Araújo & Guisan 2006; Sinclair et al. 2010; Dormann et al. 2012), we suggest that rather than searching for a “holy grail,” more attention should be paid to understanding and communicating the validity and shortcomings of the models.

In summary, the aforementioned limitations illustrate the uncertainties researchers are faced with when working with projections in species distribution modeling. In the end, the choices of the model calibration scale, the modeling algorithm and climate scenarios all show important effects on the modeling results. By investigating multiple versions of each of these factors, this study exemplifies a diverse modeling approach and as such recommends that other species distribution modelers do not rely on a single method. Provided that SDMs are carefully evaluated, selected, and interpreted as an ensemble, useful practical guidance for forest managers can be derived from their projections. This includes the discrimination between environmentally suitable and less suitable areas for a species under present and future climate. From the comparison of the present range of the species with its expected future suitability in the same region, we may then derive its sensitivity towards long-term climate change. Coupling the resulting environmental suitability maps with economic valuation then further allows for an estimation of the financial risk of a business-as-usual scenario.

Economic valuation

Although losses of environmental suitability were found for all climate scenarios—being most extensive for the low and most intensive for the high RCPs—the economic analysis allows for a slightly modified interpretation. Since most of the losses in RCP 2.6 were small and occurred in areas with high environmental suitability, they fell outside our definition for economic losses. In consequence, only small areas (mainly in the Rhine Valley) showed economically accountable losses under the low end climate scenarios, while these areas almost constantly widened to also cover larger parts in Central and South Germany under higher RCPs.

Overall, the results of the economic analysis are consistent with those of Hanewinkel et al. (2009) and Albrecht et al. (2013), considering no planting costs were included in the simulation. It should be emphasized, however, that the results showed to be very dependent on the assumed interest rate. When interest rates are high, the LEV decreases when the rotation time is kept constant, since an early harvest would be more lucrative in this case (compare Hanewinkel et al. 2009). At a low interest rate on the other hand, the influence of rotation time decreases and the LEV increases. Additionally, variables like harvesting costs and timber prices influence the results to a certain degree. Their future development, however, may depend on market demand, labor costs, advances in technology, etc. and represents another source of uncertainty.

While the LEV of the simulated 50–50 mixture stand ranked almost exactly between the LEV of both pure stands, it needs to be emphasized that BWINPro-S only takes into account very basic interactions between the individual trees within a stand. It therefore cannot represent a realistic mixing effect. The corresponding LEV of the mixed stand should thus be understood as a very rough approximation. On the other hand, no direct negative mixing effect caused by canopy competition could be found for the simulation. In any case, further field experiments (e.g., Lebourgeois et al. 2013; Magh et al. 2017) are required to gain a better understanding of the underlying processes, which could then be implemented to refine growth simulations. Moreover, detailed ecological-economic approaches are needed to couple climate sensitive forest process based models with the here used economic calculus (LEV) to take into account changes in growth, productivity, and quality of forests under changing environmental conditions over time (Yousefpour et al. 2012).

Implications for forest management

The analysis of areas with a projected beech decline unveiled a significant threat to the forest economy in Germany, potentially resulting in high economic losses. The large differences between the projected impacts of the climate scenarios, however, show that implications for forest managers may drastically change depending on the degree of climate change. Although even the RCP2.6 scenarios resulted in extensive losses of environmental suitability, these may only entail small-scale economic losses and not convince forest managers to adapt their existing strategies. Within the RCP4.5 scenarios, the suitability changes varied the most and thus represent the scenario pathway linked to the largest uncertainties for management. On average, higher losses were projected than for RCP2.6, pointing to higher risks of economic failure. Finally, all RCP8.5 scenarios resulted in significant losses both under an ecological and economic perspective and would demand for urgent adaptation in large portions of Germany’s beech forests.

Within the decision framework for adaptation presented by Yousefpour et al. (2017), the high uncertainty of the magnitude of climate change impacts coupled with the risk of high potential costs falls under the recommendation for robust adaptation strategies if economic losses are to be minimized (risk averse). Under most scenarios, silver fir showed medium to high occurrence probabilities in the areas of the strongest projected beech decline in South and Central Germany. Its admixtion to beech in these areas may hence be considered a robust adaptation strategy. While an adaptive transformation of beech into beech–silver fir mixed forests may carry high initial costs such as plantation costs or young growth tending of silver fir against ungulate browsing (e.g., Häsler and Senn 2012) these may be interpreted as opportunity costs for sustaining beech in the respective areas.

A continuation of business-as-usual management may on the other hand not yet result in immediate economic losses (since our results only refer to an average between 2041 and 2060) and thus appear a more profitable strategy under low-end climate scenarios compared to the aforementioned costly transition of pure to mixed stands. However, as has been impressively shown in 2018, recurring episodic droughts may affect unadapted forests much sooner than projected long-term climatic changes and hence exacerbate the need for adaptation. More importantly, considering the results of the stand simulation, admixing silver fir to pure beech stands may not only reduce the projected economic losses under a business-as-usual scenario but even increase their economic value due to the higher LEV of silver fir. Since regeneration costs for silver fir would mainly comprise protection against ungulate browsing in the juvenile stage (chemical treatment ~ 300–600€/ha, fence ~ 1800–4400€/ha; Staatsbetrieb Sachsenforst 2016), the benefits of mixing would overall still exceed the expected transformation costs. Therefore, the transformation of beech into mixed forests with silver fir can be recommended as adaptation strategy under all investigated magnitudes of climate change (also see Yousefpour et al. 2017).

An optimal performance and LEV of these mixed forests may yet depend on site-specific biotic interactions and the mixing ratio between the two species (Mina et al. 2018). Pretzsch et al. (2013a) and Lebourgeois et al. (2013), for example, found significant differences in biotic interactions and inter-specific complementarity depending on the availability of resources, which were higher at poor than at rich sites. Also, while our stand simulation applied a 50–50 mixing ratio, a share of 50% may already be considered the upper limit for silver fir. Under the current forest management regimes of Baden-Wurttemberg, silver fir is generally considered a tree species for admixtion (i.e., not central to silvicultural land use) and would usually cover shares of 20–50% in a beech-coniferous forest (Wicht-Lückge et al. 2014).

Conclusion

This study investigated current and potential future distributions of European beech (Fagus sylvatica) and silver fir (Abies alba) for nine climate scenarios applying various state-of-the-art modeling approaches. We projected distribution changes under climate change, assessed the land expectation value of beech and fir and discussed the implications of the results for forest management. Throughout all scenarios, the models generally agreed on a decline of beech’s environmental suitability in South and Central Germany. As our economic analyses showed, the projected declines may also result in substantial economic losses. At the same time, silver fir was projected to largely resemble its current distribution and may thus represent a potential admixing species to adapt beech forests to increasing drought frequency under climate change.

Our findings emphasize the need for interdisciplinary studies addressing both ecological and economic implications of climate change. In this context, species distribution models may be valuable tools, given that robust modeling approaches are pursued and uncertainties communicated to the decision-makers. This specifically applies to regional models, which may not be suitable to project environmental suitability of a species under a non-analog climate.

Notes

Funding information

This work has been financially supported by the Federal Ministry of Food and Agriculture (BMEL) as part of the project “BuTaKli: Beech–Silver Fir Mixed Forests as an Adaptation of Commercial Forests to Climate Change Extreme Events.”

Compliance with ethical standards

Conflict of interest

The authors declare that they have no conflict of interest.

References

  1. Acevedo P, Jiménez-Valverde A, Lobo JM, Real R (2012) Delimiting the geographical background in species distribution modelling. J Biogeogr 39(8):1383–1390.  https://doi.org/10.1111/j.1365-2699.2012.02713.x Google Scholar
  2. Acevedo P, Jiménez-Valverde A, Aragón P, Niamir A (2016) New Development in the Study of Species Distribution. In: Mateo R, Arroyo B, Garcia JT (eds) Current trends in wildlife research. Springer, Cham, pp 151–175.  https://doi.org/10.1007/978-3-319-27912-1_7 Google Scholar
  3. Albrecht A, Fortin M, Kohnle U, Ningre F (2013) Ein Sturmschadensmodul für BWinProBW: Konzept und erste Ergebnisse waldbaulicher Variantenstudien. In: DVFFA. Sektion Ertragskunde: Beiträge zur Jahrestagung, pp 75–83Google Scholar
  4. Allen CD, Macalady AK, Chenchouni H, Bachelet D, McDowell N, Vennetier M, Kitzberger T, Rigling A, Breshears DD, Hogg EH, Gonzalez P, Fensham R, Zhang Z, Castro J, Demidova N, Lim JH, Allard G, Running SW, Semerci A, Cobb N (2010) A global overview of drought and heat-induced tree mortality reveals emerging climate change risks for forests. For Ecol Manag 259(4):660–684.  https://doi.org/10.1016/j.foreco.2009.09.001 Google Scholar
  5. Araújo MB, New M (2007) Ensemble forecasting of species distributions. Trends Ecol Evol 22(1):42–47.  https://doi.org/10.1016/j.tree.2006.09.010 Google Scholar
  6. Araújo MB, Pearson RG (2005) Equilibrium of species’ distributions with climate. Ecography 28:693–695.  https://doi.org/10.1111/j.2005.0906-7590.04253.x Google Scholar
  7. Araújo MB, Guisan A (2006) Five (or so) challenges for species distribution modelling. J Biogeography 33(10):1677–1688.  https://doi.org/10.1111/j.1365-2699.2006.01584.x
  8. Barbet-Massin M, Jiguet F, Albert CH, Thuiller W (2012) Selecting pseudo-absences for species distribution models: how, where and how many? Methods Ecol Evol 3(2):327–338.  https://doi.org/10.1111/j.2041-210x.2011.00172.x Google Scholar
  9. Baumbach L, Siegmund JF, Mittermeier M, Donner RV (2017) Impacts of temperature extremes on European vegetation during the growing season. Biogeosciences 14:4891–4903.  https://doi.org/10.5194/bg-14-4891-2017 Google Scholar
  10. BMEL (2014) The forests in Germany - selected results of the third national forest inventory. Federal Ministry of Food and Agriculture (BMEL), BerlinGoogle Scholar
  11. Bolte A, Ammer C, Löf M, Madsen P, Nabuurs GJ, Schall P, Spathelf P, Rock J (2009) Adaptive forest management in Central Europe: climate change impacts, strategies and integrative concept. Scand J For Res 24(6):473–482.  https://doi.org/10.1080/02827580903418224 Google Scholar
  12. Bolte A, Kämpf F, Hilbrig L (2013) Space sequestration below ground in old-growth spruce-beech forests – signs for facilitation? Front Plant Sci 4:322.  https://doi.org/10.3389/fpls.2013.00322 Google Scholar
  13. Bonan GB (2008) Forests and climate change: forcings, feedbacks, and the climate benefits of forests. Science 320(5882):1444–1449.  https://doi.org/10.1126/science.1155121 Google Scholar
  14. Cheaib A, Badeau V, Boe J, Chuine I, Delire C, Dufrêne E, François C, Gritti ES, Legay M, Pagé C, Thuiller W, Viovy N, Leadley P (2012) Climate change impacts on tree ranges: model intercomparison facilitates understanding and quantification of uncertainty. Ecol Lett 15:533–544.  https://doi.org/10.1111/j.1461-0248.2012.01764.x Google Scholar
  15. Coumou D, Rahmstorf S (2012) A decade of weather extremes. Nat Clim Chang 2(7):491–496.  https://doi.org/10.1038/nclimate1452 Google Scholar
  16. Czajkowski T, Bolte A (2006) Unterschiedliche Reaktion deutscher und polnischer Herkünfte der Buche (Fagus sylvatica L.) auf Trockenheit. Allgemeine Jagd- und Forstzeitung 177(2):30–40Google Scholar
  17. Dannenmann M, Bimüller C, Gschwendtner S, Leberecht M, Tejedor J, Bilela S, Gasche R, Hanewinkel M, Baltensweiler A, Kögel-Knabner I, Polle A, Schloter M, Simon J, Rennenberg H (2016) Climate change impairs nitrogen cycling in European beech forests. PLoS One 11(7):e0158823.  https://doi.org/10.1371/journal.pone.0158823 Google Scholar
  18. Deser C, Phillips A, Bourdette V, Teng H (2012) Uncertainty in climate change projections: the role of internal variability. Clim Dyn 38(3–4):527–546.  https://doi.org/10.1007/s00382-010-0977-x Google Scholar
  19. Devi S, Angrish R, Madaan S, Toky OP, Arya SS (2016) Sinker root system in trees with emphasis on soil profile. In: Cloudhary DK, Varma A, Tuteja N (eds) Plant-microbe interaction: an approach to sustainable agriculture. Springer, Singapore, pp 463–474.  https://doi.org/10.1007/978-981-10-2854-0_21 Google Scholar
  20. Dieter M (2001) Land expectation values for spruce and beech calculated with Monte Carlo modelling techniques. Forest Policy Econ 2(2):157–166.  https://doi.org/10.1016/s1389-9341(01)00045-4 Google Scholar
  21. Döbbeler H, Albert M, Schmidt M, Nagel J, Schröder J (2011) BWINPro. Programm zur Bestandesanalyse und Prognose. Handbuch zur gemeinsamen Version von BWINPro und BWINPro-S. Version 6.3. Nordwestdeutsche Forstliche Versuchsanstalt und TU DresdenGoogle Scholar
  22. Dormann CF (2007) Promising the future? Global change projections of species distributions. Basic Appl Ecol 8(5):387–397.  https://doi.org/10.1016/j.baae.2006.11.001 Google Scholar
  23. Dormann CF, Schymanski SJ, Cabral J, Chuine I, Graham C, Hartig F, Kearney M, Morin X, Römermann C, Schröder B, Singer A (2012) Correlation and process in species distribution models: bridging a dichotomy. J Biogeogr 39:2119–2131.  https://doi.org/10.1111/j.1365-2699.2011.02659.x Google Scholar
  24. Drake JM, Randin C, Guisan A (2006) Modelling ecological niches with support vector machines. J Appl Ecol 43(3):424–432.  https://doi.org/10.1111/j.1365-2664.2006.01141.x Google Scholar
  25. Duan RY, Kong XQ, Huang MY, Fan WY, Wang ZG (2014) The predictive performance and stability of six species distribution models. PLoS One 9(11):e112764.  https://doi.org/10.1371/journal.pone.0112764 Google Scholar
  26. Dyderski MK, Paź S, Frelich LE, Jagodziński AM (2018) How much does climate change threaten European forest tree species distributions? Glob Chang Biol 24:1150–1163.  https://doi.org/10.1111/gcb.13925 Google Scholar
  27. Elith J, Graham CH, Anderson RP, Dudík M, Ferrier S, Guisan A, Hijmans RJ, Huettmann F, Leathwick JR, Lehmann A, Li J, Lohmann LG, Loiselle BA, Manion G, Moritz C, Nakamura M, Nakazawa Y, Overton JM, Townsend Peterson A, Phillips SJ, Richardson K, Scachetti-Pereira R, Schapire RE, Soberón J, Williams S, Wisz MS, Zimmermann NE (2006) Novel methods improve prediction of species’ distributions from occurrence data. Ecography 29(2):129–151.  https://doi.org/10.1111/j.2006.0906-7590.04596.x Google Scholar
  28. Ellenberg H (1988) Vegetation ecology of Central Europe. Cambridge University Press, Cambridge, New York, New Rochelle, Melbourne, SydneyGoogle Scholar
  29. Elling W, Dittmar C, Pfaffelmoser K, Rötzer T (2009) Dendroecological assessment of the complex causes of decline and recovery of the growth of silver fir (Abies alba Mill.) in Southern Germany. For Ecol Manag 257(4):1175–1187.  https://doi.org/10.1016/j.foreco.2008.10.014 Google Scholar
  30. ESRI (2011) ArcGIS Desktop: Release 10. Environmental Systems Research Institute (ESRI), RedlandsGoogle Scholar
  31. Fang J, Lechowicz MJ (2006) Climatic limits for the present distribution of beech (Fagus L.) species in the world. J Biogeogr 33(10):1804–1819.  https://doi.org/10.1111/j.1365-2699.2006.01533.x Google Scholar
  32. Faustmann M (1849) Calculation of the value which forest land and immature stands possess for forestry. Reprinted in: Journal of Forest Economics 1(1):7–44Google Scholar
  33. Fitzpatrick MC, Hargrove WW (2009) The projection of species distribution models and the problem of non-analog climate. Biodivers Conserv 18(8):2255–2261.  https://doi.org/10.1007/s10531-009-9584-8 Google Scholar
  34. Forrester DI (2014) The spatial and temporal dynamics of species interactions in mixed-species forests: from pattern to process. For Ecol Manag 312:282–292.  https://doi.org/10.1016/j.foreco.2013.10.003 Google Scholar
  35. Geßler A, Keitel C, Kreuzwieser J, Matyssek R, Seiler W, Rennenberg H (2007) Potential risks for European beech (Fagus sylvatica L.) in a changing climate. Trees 21(1):1–11.  https://doi.org/10.1007/s00468-006-0107-x Google Scholar
  36. Goberville E, Hautekèete NC, Kirby RR, Piquot Y, Luczak C, Beaugrand G (2016) Climate change and the ash dieback crisis. Sci Rep 6:35303.  https://doi.org/10.1007/s00468-006-0107-x Google Scholar
  37. Griess VC, Acevedo R, Härtl F, Staupendahl K, Knoke T (2012) Does mixing tree species enhance stand resistance against natural hazards? A case study for spruce. For Ecol Manag 267:284–296.  https://doi.org/10.1016/j.foreco.2011.11.035 Google Scholar
  38. Guisan A, Thuiller W (2005) Predicting species distribution: offering more than simple habitat models. Ecol Lett 8(9):993–1009.  https://doi.org/10.1111/j.1461-0248.2005.00792.x Google Scholar
  39. Hair JF, Anderson RE, Tatham RL, Black WC (1995) Multivariate data analysis with readings, 3rd edn. MacMillan, New YorkGoogle Scholar
  40. Hanewinkel M, Hummel S, Cullmann DA (2009) Modelling and economic evaluation of forest biome shifts under climate change in Southwest Germany. For Ecol Manag 259(4):710–719.  https://doi.org/10.1016/j.foreco.2009.08.021 Google Scholar
  41. Hanewinkel M, Cullmann DA, Schelhaas MJ, Nabuurs GJ, Zimmermann NE (2013) Climate change may cause severe loss in the economic value of European forest land. Nat Clim Chang 3(3):203–207.  https://doi.org/10.1038/nclimate1687 Google Scholar
  42. Häsler H, Senn J (2012) Ungulate browsing on European silver fir Abies alba: the role of occasions, food shortage and diet preferences. Wildl Biol 18(1):67–42.  https://doi.org/10.2981/09-013 Google Scholar
  43. Hengl T, de Jesus JM, Heuvelink GB, Gonzalez MR, Kilibarda M, Blagotić A, Shangguan W, Wright MN, Geng X, Bauer-Marschallinger B, Guevara MA, Vargas R, MacMillan RA, Batjes NH, Leenaars JGB, Ribeiro E, Wheeler I, Mantel S, Kempen B (2017) SoilGrids250m: global gridded soil information based on machine learning. PLoS One 12(2):e0169748.  https://doi.org/10.1371/journal.pone.0169748 Google Scholar
  44. Hickler T, Vohland K, Feehan J, Miller PA, Smith B, Costa L, Giesecke T, Fronzek S, Carter TR, Cramer W, Kühn I, Sykes MT (2012) Projecting the future distribution of European potential natural vegetation zones with a generalized, tree-species based dynamic vegetation model. Glob Ecol Biogeogr 21:50–63.  https://doi.org/10.1111/j.1466-8238.2010.00613.x Google Scholar
  45. Hickler T, Rammig A, Werner C (2015) Modelling CO2 impacts on forest productivity. Curr Forest Rep 1(2):69–80.  https://doi.org/10.1007/s40725-015-0014-8 Google Scholar
  46. Hijmans RJ (2016) Raster: geographic data analysis and modeling. R package version 2.5–8Google Scholar
  47. Hijmans RJ, Cameron SE, Parra JL, Jones PG, Jarvis A (2005) Very high resolution interpolated climate surfaces for global land areas. Int J Climatol 25(15):1965–1978.  https://doi.org/10.1002/joc.1276 Google Scholar
  48. Houston Durrant T, de Rigo D, Caudullo G (2016) Fagus sylvatica and other beeches in Europe: distribution, habitat, usage and threats. In: San-Miguel-Ayanz J, de Rigo D, Caudullo G, Houston Durrant T, Mauri A (eds) European atlas of Forest tree species. Publ, Off. EU, pp 94–95Google Scholar
  49. Jimenez-Valverde A, Lobo JM, Hortal J (2009) The effect of prevalence and its interaction with sample size on the reliability of species distribution models. Commun Ecol 10(2):196–205.  https://doi.org/10.1556/comec.10.2009.2.9 Google Scholar
  50. Jump AS, Hunt JM, Penuelas J (2006) Rapid climate change-related growth decline at the southern range edge of Fagus sylvatica. Glob Chang Biol 12(11):2163–2174.  https://doi.org/10.1111/j.1365-2486.2006.01250.x Google Scholar
  51. Karger DN, Conrad O, Böhner J, Kawohl T, Kreft H, Soria-Auza RW, Zimmermann NE, Linder HP, Kessler M (2017) Climatologies at high resolution for the earth’s land surface areas. Sci Data 4:170122.  https://doi.org/10.1038/sdata.2017.122 Google Scholar
  52. Keenan RJ (2015) Climate change impacts and adaptation in forest management: a review. Ann For Sci 72(2):145–167.  https://doi.org/10.1007/s13595-014-0446-5 Google Scholar
  53. Kelty MJ (2006) The role of species mixtures in plantation forestry. For Ecol Manag 233(2):195–204.  https://doi.org/10.1016/j.foreco.2006.05.011 Google Scholar
  54. Kölling C, Falk W, Walentowski H (2011) Standörtliche Anbaumöglichkeiten der Tanne (Abies alba und Abies grandis) in Bayern. LWF Wissen 66:11–19Google Scholar
  55. Körner C, Asshoff R, Bignucolo O, Hattenschwiler S, Keel SG, Pelaez-Riedl S, Pepin S, Siegwolf RTW, Zotz G (2005) Carbon flux and growth in mature deciduous forest trees exposed to elevated CO2. Science 309(5739):1360–1362.  https://doi.org/10.1126/science.1113977 Google Scholar
  56. Kramer K, Degen B, Buschbom J, Hickler T, Thuiller W, Sykes MT, de Winter W (2010) Modelling exploration of the future of European beech (Fagus sylvatica L.) under climate change: range, abundance, genetic diversity and adaptive response. For Ecol Manag 259(11):2213–2222.  https://doi.org/10.1016/j.foreco.2009.12.023 Google Scholar
  57. Lebourgeois F, Gomez N, Pinto P, Mérian P (2013) Mixed stands reduce Abies alba treering sensitivity to summer drought in the Vosges mountains, Western Europe. For Ecol Manag 303:61–71.  https://doi.org/10.1016/j.foreco.2013.04.003 Google Scholar
  58. Leuzinger S, Körner C (2007) Water savings in mature deciduous forest trees under elevated CO2. Glob Chang Biol 13:2498–2508.  https://doi.org/10.1111/j.1365-2486.2007.01467.x Google Scholar
  59. Lindner M, Maroschek M, Netherer S, Kremer A, Barbati A, Garcia-Gonzalo J, Seidl R, Delzon S, Corona P, Kolström M, Lexer MJ, Marchetti M (2010) Climate change impacts, adaptive capacity, and vulnerability of European forest ecosystems. For Ecol Manag 259(4):698–709.  https://doi.org/10.1016/j.foreco.2009.09.023 Google Scholar
  60. Linkevicius E (2014) Single tree level simulator for Lithuanian pine forests. Dissertation, Technische Universität Dresden, Institute of Forest Growth and Forest Computer SciencesGoogle Scholar
  61. Lobo JM, Jiménez-Valverde A, Hortal J (2010) The uncertain nature of absences and their importance in species distribution modelling. Ecography 33(1):103–114.  https://doi.org/10.1111/j.1600-0587.2009.06039.x Google Scholar
  62. Magh RK, Grün M, Knothe VE, Stubenazy T, Tejedor J, Dannenmann M, Rennenberg H (2017) Silver-fir (Abies alba MILL.) neighbors improve water relations of European beech (Fagus sylvatica L.), but do not affect N nutrition. Trees 32(1):337–348.  https://doi.org/10.1007/s00468-017-1557-z Google Scholar
  63. Mauri A, de Rigo D, Caudullo G (2016) Abies alba in Europe: distribution, habitat, usage and threats. In: San-Miguel-Ayanz J, de Rigo D, Caudullo G, Houston Durrant T, Mauri A (eds) European atlas of Forest tree species. Publ. Off. EU, pp 48–49Google Scholar
  64. Mauri A, Strona G, San-Miguel-Ayanz J (2017) EU-Forest, a high-resolution tree occurrence dataset for Europe. Sci Data 4:160123.  https://doi.org/10.1038/sdata.2016.123 Google Scholar
  65. Menard S (1995) Applied logistic regression analysis: Sage university series on quantitative applications in the social sciences. Sage, Thousand OaksGoogle Scholar
  66. Merow C, Smith MJ, Edwards TC, Guisan A, McMahon SM, Normand S, Thuiller W, Wüest RO, Zimmermann NE, Elith J (2014) What do we gain from simplicity versus complexity in species distribution models? Ecography 37:1267–1281.  https://doi.org/10.1111/ecog.00845 Google Scholar
  67. Metz J, Annighöfer P, Schall P, Zimmermann J, Kahl T, Schulze ED, Ammer C (2016) Site-adapted admixed tree species reduce drought susceptibility of mature European beech. Glob Chang Biol 22(2):903–920.  https://doi.org/10.1111/gcb.13113 Google Scholar
  68. Miller ME, Hui SL, Tierney WM (1991) Validation techniques for logistic regression models. Stat Med 10(8):1213–1226.  https://doi.org/10.1002/sim.4780100805 Google Scholar
  69. Mina M, Del Río M, Huber MO, Thürig E, Rohner B (2018) The symmetry of competitive interactions in mixed spruce, silver fir and European beech forests. J Veg Sci 29:775–787.  https://doi.org/10.1111/jvs.12664 Google Scholar
  70. Mölder I, Leuschner C (2014) European beech grows better and is less drought sensitive in mixed than in pure stands: tree neighbourhood effects on radial increment. Trees 28:777–792.  https://doi.org/10.1007/s00468-014-0991-4 Google Scholar
  71. Nagel J (1999) Konzeptionelle Überlegungen zum schrittweisen Aufbau eines waldwachstumskundlichen Simulationssystems für Nordwestdeutschland. Schriften aus der Forstlichen Fakultät der Universität Göttingen und der Nieders. Forstl. Versuchsanstalt, Band 128. J.D. Sauerländer's Verlag, Frankfurt a.M.Google Scholar
  72. Naimi B, Araújo MB (2016) sdm: a reproducible and extensible R platform for species distribution modelling. Ecography 39:368–375.  https://doi.org/10.1111/ecog.01881 Google Scholar
  73. Neumann RB, Cardon ZG (2012) The magnitude of hydraulic redistribution by plant roots: a review and synthesis of empirical and modeling studies. New Phytol 194:337–352.  https://doi.org/10.1111/j.1469-8137.2012.04088.x Google Scholar
  74. Nocentini S (2009) Structure and management of beech (Fagus salvatica L.) forests in Italy. iForest 2:105–113.  https://doi.org/10.3832/ifor0499-002 Google Scholar
  75. Phillips SJ, Dudík M, Elith J, Graham CH, Lehmann A, Leathwick J, Ferrier S (2009) Sample selection bias and presence-only distribution models: implications for background and pseudo-absence data. Ecol Appl 19(1):181–197.  https://doi.org/10.1890/07-2153.1 Google Scholar
  76. Pretzsch H, Schütze G, Uhl E (2013a) Resistance of European tree species to drought stress in mixed versus pure forests: evidence of stress release by inter-specific facilitation. Plant Biol 15(3):483–495.  https://doi.org/10.1111/j.1438-8677.2012.00670.x Google Scholar
  77. Pretzsch H, Bielak K, Block J, Bruchwald A, Dieler J, Ehrhart HP, Kohnle U, Nagel J, Spellmann H, Zasada M, Zingg A (2013b) Productivity of mixed versus pure stands of oak (Quercus petraea (Matt.) Liebl. and Quercus robur L.) and European beech (Fagus sylvatica L.) along an ecological gradient. Eur J For Res 132:263–280.  https://doi.org/10.1007/s10342-012-0673-y Google Scholar
  78. R Core Team (2017) R: a language and environment for statistical computing. R Foundation for Statistical Computing, ViennaGoogle Scholar
  79. Ramsfield TD, Bentz BJ, Faccoli M, Jactel H, Brockerhoff EG (2016) Forest health in a changing world: effects of globalization and climate change on forest insect and pathogen impacts. Forestry 89:245–252.  https://doi.org/10.1093/forestry/cpw018 Google Scholar
  80. Reyer C (2013) The cascade of uncertainty in modeling forest ecosystem responses to environmental change and the challenge of sustainable resource management. Dissertation, Humboldt University BerlinGoogle Scholar
  81. Rösch M (2015) Nationalpark – Natur – Weißtanne – Fichte. Sechs Jahrtausende Wald und Mensch im Nordschwarzwald. Denkmalpflege Baden-Württemberg 44(3):154–159Google Scholar
  82. Saltré F, Duputié A, Gaucherel C, Chuine I (2015) How climate, migration ability and habitat fragmentation affect the projected future distribution of European beech. Glob Chang Biol 21(2):897–910.  https://doi.org/10.1111/gcb.12771 Google Scholar
  83. Sardans J, Peñuelas J (2014) Hydraulic redistribution by plants and nutrient stoichioimetry: shifts under global change. Ecohydrology 7:1–20.  https://doi.org/10.1002/eco.1459 Google Scholar
  84. Schröder J, Röhle H, Eisenhauer D, Brand S (2006) Zum Jugendwachstum der Eiche unter Kiefernschirm in Sachsen. Forstarchiv 77(6):195–202Google Scholar
  85. Schröder J, Röhle H, Gerold D, Münder K (2007) Modeling individual-tree growth in stands under forest conversion in East Germany. Eur J For Res 126(3):459–472.  https://doi.org/10.1007/s10342-006-0167-x Google Scholar
  86. Serra-Diaz JM, Ninyerola M, Lloret F (2012) Coexistence of Abies alba (Mill.) - Fagus sylvatica (L.) and climate change impact in the Iberian Peninsula: a climatic-niche perspective approach. Flora 207(1):10–18.  https://doi.org/10.1016/j.flora.2011.10.002 Google Scholar
  87. Sinclair SJ, White MD, Newell GR (2010) How useful are species distribution models for managing biodiversity under future climates? Ecol Soc 15(1):8.  https://doi.org/10.5751/es-03089-150108 Google Scholar
  88. Spellmann H, Meesenburg H, Schmidt M, Nagel RV, Sutmöller J, Albert M (2015) Klimaanapassung ist Vorsorge für den Wald. ProWald November 2015:4–10Google Scholar
  89. Staatsbetrieb Sachsenforst (2016) Walderneuerung und Erstaufforstung: Hinweise für Waldbesitzer. Staatsbetrieb Sachsenforst, PirnaGoogle Scholar
  90. Straka TJ, Bullard SH (1996) Land expectation value calculation in timberland valuation. Appraisal J 64:399–405Google Scholar
  91. Strona G, Mauri A, San-Miguel-Ayanz J (2016) A high-resolution pan-European tree occurrence dataset. figshare.  https://doi.org/10.6084/m9.figshare.c.3288407
  92. Thünen-Institut (2012) Third national forest inventory database. https://bwi.info. Accessed 27 July 2017
  93. Tinner W, Colombaroli D, Heiri O, Henne PD, Steinacher M, Untenecker J, Vescovi E, Allen JRM, Carraro G, Conedera M, Joos F, Lotter AF, Luterbacher J, Samartin S, Valsecchi V (2013) The past ecology of Abies alba provides new perspectives on future responses of silver fir forests to global warming. Ecol Monogr 83(4):419–439.  https://doi.org/10.1890/12-2231.1 Google Scholar
  94. Tinner W, Conedera M, Bugmann H, Colombaroli D, Gobet E, Vescovi E, Heiri O, Joos F, Luterbacher J, La Mantia T, Pasta S, Untenecker J, Henne PD (2016) Europäische Wälder unter wärmeren Klimabedingungen - Neue Erkenntnisse aus Paläoökologie und dynamischer Vegetationsmodellierung. AFZ-DerWald 71(18):45–49Google Scholar
  95. US Geological Service (1996) GTOPO30: Global 30 arc-seconds digital elevation model. https://lta.cr.usgs.gov/GTOPO30. Accessed 21 February 2017
  96. Van der Wal J, Shoo LP, Graham C, Williams SE (2009) Selecting pseudo-absence data for presence-only distribution modeling: how far should you stray from what you know? Ecol Model 220:589–594.  https://doi.org/10.1016/j.ecolmodel.2008.11.010 Google Scholar
  97. Wicht-Lückge G, Biewald G, Göckel C, Jacob A, Kilian M, Kohnle U, Michiels HG, Nagel J, Schabel A, Schmalfuß N (2014) Richtlinie landesweiter Waldentwicklungstypen. Landesbetrieb Forst Baden-Württemberg, Ministerium für Ländlichen Raum und Verbraucherschutz Baden-Württemberg, StuttgartGoogle Scholar
  98. Wonsack D (2016) Überführung von gleichaltrigen in ungleichaltrige Fichtenwälder im Mathislewald. Master’s thesis, Albert-Ludwigs-Universität-Freiburg, Professur für Forstökonomie und ForstplanungGoogle Scholar
  99. Yousefpour R, Hanewinkel M, Le Moguédec G (2010) Evaluating the suitability of management strategies of pure Norway spruce forests in the Black Forest area of Southwest Germany for adaptation to or mitigation of climate change. Environ Manag 45(2):387–402.  https://doi.org/10.1007/s00267-009-9409-2 Google Scholar
  100. Yousefpour R, Jacobsen JB, Thorsen BJ, Meilby H, Hanewinkel M, Oehler K (2012) A review of decision-making approaches to handle uncertainty and risk in adaptive forest management under climate change. Ann For Sci 69(1):1–15.  https://doi.org/10.1007/s13595-011-0153-4 Google Scholar
  101. Yousefpour R, Augustynczik AL, Hanewinkel M (2017) Pertinence of reactive, active, and robust adaptation strategies in forest management under climate change. Ann For Sci 74(2):40.  https://doi.org/10.1007/s13595-017-0640-3 Google Scholar
  102. Zang C, Hartl-Meier C, Dittmar C, Rothe A, Menzel A (2014) Patterns of drought tolerance in major European temperate forest trees: climatic drivers and levels of variability. Glob Chang Biol 20(12):3767–3779.  https://doi.org/10.1111/gcb.12637 Google Scholar
  103. Zapater M, Hossann C, Bréda N, Bréchet C, Bonal D, Granier A (2011) Evidence of hydraulic lift in a young beech and oak mixed forest using 18O soil water labelling. Trees 25:885–894.  https://doi.org/10.1007/s00468-011-0563-9 Google Scholar
  104. Zebisch M, Grothmann T, Schröter D, Hasse C, Fritsch U, Cramer W (2005) Climate change in Germany. In: Federal Environmental Agency Germany/Umweltbundesamt (ed.) vulnerability and adaptation of climate sensitive sectors / Klimawandel in Deutschland–Vulnerabilität und Anpassungsstrategien klimasensitiver Systeme, report 201. Federal Environment Agency Germany, Dessau, pp 41–253Google Scholar
  105. Zimmermann NE, Yoccoz NG, Edwards Jr TC, Meier ES, Thuiller W, Guisan A, Schmatz DR, Pearman PB (2009) Climatic extremes improve predictions of spatial patterns of tree species. PNAS 106(Supplement 2):19723–19728.  https://doi.org/10.1073/pnas.0901643106
  106. Zimmermann J, Hauck M, Dulamsuren C, Leuschner C (2015) Climate warming-related growth decline affects Fagus sylvatica, but not other broad-leaved tree species in Central European mixed forests. Ecosystems 18(4):560–572.  https://doi.org/10.1007/s10021-015-9849-x Google Scholar
  107. Zimmermann NE, Normand S, Psomas A (2016) PorTree final report. A project funded by the BAFU-WSL program on “forests and climate change” in Switzerland. Swiss Federal Research Institute (WSL), Birmensdorf.  https://doi.org/10.3929/ethz-a-010689681 Google Scholar
  108. Zurell D, Thuiller W, Pagel J, Cabral JS, Münkemüller T, Gravel D, Dullinger S, Normand S, Schiffers KH, Moore KA, Zimmermann NE (2016) Benchmarking novel approaches for modelling species range dynamics. Glob Chang Biol 22:2651–2664.  https://doi.org/10.1111/gcb.13251 Google Scholar

Copyright information

© Springer-Verlag GmbH Germany, part of Springer Nature 2019

Authors and Affiliations

  1. 1.Institute of Forestry Economics and Forest PlanningAlbert-Ludwigs-Universität FreiburgFreiburg im BreisgauGermany
  2. 2.Senckenberg Biodiversity and Climate Research Centre (BiK-F)Frankfurt am MainGermany

Personalised recommendations