1. Introduction

Many tree species face many environmental challenges due to the ongoing effects of climate change. As both average temperature and temperature extremes increase, and precipitation patterns become more erratic, forest areas are continuing to be negatively impacted [1–4]. This will have a large impact on forest services such as carbon sequestration [5,6], biodiversity [7–9] forest derived economic resources, [10–12] as well as the social, traditional and recreational values of forest [13,14]. While these negative impacts are well-documented, some studies indicate potential growth benefits under favourable conditions. Higher temperatures and longer growing seasons may enhance productivity where water and nutrient availability are sufficient, and climate envelope models project shifts in species distributions, including upward and northward expansion and increasing suitability for species such as oak and pine, while others such as Norway spruce may decline [15,16].

However, these effects are strongly context-dependent. For example, one study actually reported increased fine root biomass under warming at two nutrient-poor sites in Germany but emphasize the limited generalizability of their results [17]. Likewise, projections of increased productivity in boreal forests are accompanied by substantial uncertainties and rising risks from disturbances such as drought, fire, and insect outbreaks [18]. In addition, climate change is expected to intensify biological risks, including insect infestations, which can amplify forest damage.

Overall, the net effects of warming on forests remain region-specific and uncertain, reinforcing the need for integrated risk-based assessments. It is therefore urgent to better understand the vulnerability and the risks associated with current forest areas in relation to a changing climate and the many different factors that co-determine forest vulnerability and risk. These different factors have a varied distribution over space and time. Therefore, spatial and temporal variability need to be taken into account when developing climate adaptation strategies. It is also important to consider the interaction between these various factors. For example, areas with low soil water capacity are more prone to drought [19,20]. In order to have a better overview of all these adverse effects, there is a need for high resolution mapping of tree species’ climate change risks.

One method to estimate climate risks to ecosystems is to use a biophysical model, which can include either one or multiple climate hazards [21,22]. To include multiple hazards in this approach, one firstly requires information on the complex interrelationships between various selected hazard variables. Secondly, one also needs the data to be able to incorporate all the natural processes over a large area, at a high enough resolution to calculate the risks of each spatial unit.

Another approach is to use a more abstract and index-based approach, in which risk can be abstracted into various components relative to one another to be able to calculate the interactions between multiple hazards through simple addition [23,24]. In this approach, risk can be divided into different components consisting of hazard, vulnerability and exposure [25]. For forests, the two main types of hazards are climate variability: which can be divided between the historical variability of a certain location [26], and the increased variability caused by climate change [27,28]. For tree species within their native range, we can assume that they have previously adapted to the full spectrum of natural variability, as some species have been present in their current area since the end of the last ice age [29]. Therefore, it is anthropogenically-induced variability that poses the largest hazard. The IPCC defines vulnerability as “the propensity or predisposition to be adversely affected.” [30]. Here we can assume that each tree species has its own distinct vulnerability to changing climatic conditions. The final aspect, exposure refers to the presence of elements in an area in which hazard events may occur [31]. Which in this case would mean the location of the various forest stands.

To map out and visualise the spatial variation of our various risk factors, we can then plot them on to a risk map. Risk mapping has been used to document and explore the spatial aspect of various environmental factors, from pollutants [32–34] to flood risk [25,35,36] and forest fires [37–39]. One important factor to consider is how to combine various aspects of vulnerability and hazard together. The selection of which factors to include in the risk model, and the weighing of the various factors has a substantial influence on the end result. This risk analysis follows a qualitative approach. Bio-geophysical interactions are not taken into account. Instead the model is straightforward in assuming all factors have a proportional influence on risk than can be either positively correlated (additive) or negatively correlated (subtractive) to the overall risk factor.

This study aims to provide an initial overview of forest risk distribution throughout the country of Germany. We will calculate and map both vulnerability and exposure in various future climate scenarios to create an overview of which areas are more at risk for extreme climate events for four commercial tree species, and which forest locations are more exposed to these vulnerabilities and therefore more at risk. With the goal to assess which areas are the most at risk from climate change, how this varies between tree species and between different climate scenarios.

2. Methodology

This study was performed for the territory of the Federal Republic of Germany. Within the boundaries of the German state lie a variety of landscapes from the loam and clay dominated northern European plain [40,41], stretching to the central hills and low mountains of the so-called “Mittelgebirge” to the foothills and first peaks of the Alps. Climatically the country lies on the transitional boundary of a temperate oceanic climate (located mostly in the northern and western pars of the country), to a continental climate (located more in the south and east) [42]. All these landscape and climatic variations amount to a significant variation in tree species risk.

The central hills of Germany, starting from the Harz mountains in the north, and bounded by the Danube river to the south contain most of the Forest area of Germany. Theses forested hills have long served an important role in forestry and recreation. These areas are also more vulnerable due to climate change, as they have an intrinsically lower soil water capacity [43,44] as compared to the more clay based soils of northwestern Germany. The northeastern plains are heavily forested, mostly with pine, as a result of forest plantations established during the era of the German Democratic Republic. However, these are some of the driest sandy soils in the German northern European plain [45] with these areas being particularly prone to summer droughts [46].

The methodological approach follows three main steps. First, climate variables, both historcal and future projections were processed to represent relevant hazard components. Second, species-specific climate envelopes were quantified and combined with these hazards and other geographical data to derive a composite risk index. Third, the resulting index was mapped spatially across Germany to assess relative climate-related risk for each species. The workflow is designed to provide a comparative, spatially explicit measure of risk rather than absolute predictions of tree mortality. In the following subsections we address data acquisition and processing for each of these steps.

In this study, “risk” is defined as a relative index combining climate-related hazards and species-specific sensitivities. It does not represent a direct probability of mortality, but rather indicates the relative degree to which a species may be adversely affected by projected climatic conditions compared to other regions or species.

2.1. Climate variables

The climate envelopes were obtained via the methodology adapted from [47]. Following this methodology, we used ERA5 reanalysis data [48] to calculate for 19 bioclimatic variables. We categorized these variables into three categories (Table 1). We then weighted each variable according to the results of the original PCA performed in [47] with the final weight being a summary of the NMDS1 and NMDS2 axes.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. The 19 Bioclimatic variables used in this study, their category and weight.

https://doi.org/10.1371/journal.pclm.0000931.t001

Next, we used Europe wide species occurrence data, for the four tree species, Norway spruce (Picea abies), Scots pine (Pinus sylvestris), European beech (Fagus sylvatica) and English oak (Quercus robur). These four species will be referred to as simply “Spruce,” “Pine,” Beech” and “Oak” for simplicity. These were selected based on being the most dominant tree species in Germany with regards to forest cover [49]. We then transformed the occurrence data for these species into presence-absence data in order to feed them into a Species Distribution Model (SDM) where we assumed a unimodal distribution along all 19 bioclimatic variables. After we obtained the results from the SDM, we calculated climate envelopes per species and per bioclimatic variables. The climate envelope is defined based on the upper and lower limits of of the SDM for each species and BIO factor, based on an occurrence probability threshold of 0.1.

To evaluate the robustness of climate envelope exceedance estimates to uncertainty in the species-specific climatic envelopes, we conducted a parametric sensitivity analysis. For each of the four tree species and both climate scenarios (SSP245 and SSP585), exceedance maps were recalculated after systematically modifying the lower and/or upper climatic tolerance bounds by 10%, as a standardized local deviation that is sufficiently large to exceed numerical noise while remaining within a biologically plausible parameter space. This yielded six uncertainty adjusted scenarios in addition to the unadjusted baseline scenario. From each scenario raster, 500 spatially consistent grid cells were randomly sampled to ensure direct comparability across species, scenarios, and sensitivity scenarios. Sensitivity was quantified as the deviation of adjusted exceedance values from the baseline at each grid cell, and mean absolute deviations were used to assess the magnitude of change. Elasticity was calculated as the proportional change in exceedance relative to the imposed 10% alteration of tolerance limits, thereby providing a standardized measure of model responsiveness. See the equations 1 and 2. This framework allowed us to determine whether projected climate risks were robust to plausible uncertainty in tolerance thresholds and to identify which species and tolerance bounds (lower vs. upper) exerted the strongest influence on exceedance patterns under contrasting climate scenarios.

2.1.1. Future climate scenarios.

We then included future climate scenarios from seven CMIP6 models for both the SSP245 and SSP585 scenarios (Table 2). SSP scenarios are a combination of a social narrative with an associated amount of radiative forcing [50]. For SSP245 this means a combination of the second pathway, the so-called “Middle of the road” scenario regarding global development, and a global forcing level of 4.5 W/m², and for SSP585 the worst-case so-called “Fossil fuel development” scenario with a global forcing level of 8.5 W/m². We obtained monthly climate data of total precipitation, mean temperature at 2-metre elevation above ground, maximum 2-metre temperature and minimum 2-metre temperature for the entirety of Germany. The temporal range of our data was from 2025 until the year 2100. Using these scenarios, we calculated the same 19 bioclimatic variables for each year.

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. List of the seven included CMIP6 scenarios and their associated spatial resolutions.

https://doi.org/10.1371/journal.pclm.0000931.t002

Uncertainty in projected climate scenarios was quantified per grid cell across all climate models and projection years separately for each climate scenario (SSP245 and SSP585) for both temperature and precipitation. For annual mean temperature (BIO1), uncertainty was expressed as the standard deviation (SD) of the ensemble projections. Because temperature (°C) is measured on an interval scale and can assume negative values, relative dispersion metrics such as the coefficient of variation (CV) are unstable and difficult to interpret. Therefore, absolute variability (SD) was used. For annual precipitation (BIO12), uncertainty was expressed as the coefficient of variation (CV = SD / mean). As precipitation is strictly positive and measured on a ratio scale, the CV provides a scale-independent and spatially comparable measure of relative uncertainty.

2.3. Interpolation and calculation

The following section describes the mathematical formulation of the risk index. Readers primarily interested in the conceptual approach may focus on the structure of the index, while the detailed formulation provides transparency on the weighting and aggregation steps.

As the CMIP6 future climate data consist of discrete points, we interpolated the yearly climate data over the entire extent of the territory of the Federal Republic of Germany. To find the most appropriate interpolation method, we compared IDW (Inverse Distance Weighting) to Kriging. IDW assumes that similarity of an unknown point increases with distance to a known point, then applies as a distance reverse function to interpolate between data points [54]. Kriging, uses a variogram to weigh nearby points and interpolate the missing data [55]. We found the kriging spatial interpolation method provided the best results. The IDW interpolation was prone to over-smooth variables leading an underestimation of highs and an overestimation of lows. This is similar to results found in previous studies comparing interpolation methods for climate data [56–58]. The interpolation was performed over the geographical area between the following coordinates: North: 55°, West: 5.5°, South: 47.1° and East: 15.5°. Prior to interpolation, observations with identical spatial coordinates were aggregated by averaging their values to avoid zero-distance pairs during variogram estimation. Ordinary kriging was then performed on a regular prediction grid, ensuring consistent distance calculations. An empirical semivariogram was calculated from the aggregated data and fitted with a spherical model using a robust weighted least-squares approach. Initial variogram parameters were derived from the spatial extent and variance of the input data, and alternative model structures were tested if convergence failed. Kriging predictions were generated using an adaptive neighborhood definition to ensure numerical stability across varying point densities. For the temperature variables, interpolated temperature fields were subsequently adjusted by subtracting a resampled temperature anomaly raster.

We then compared each raster to the upper and lower envelope limits. We calculated the yearly envelope exceedance for each bioclimatic variable and for every model year. Envelope exceedances allow us to integrate the uncertainties created by the spread of results from the entire ensemble of climate models used in this study. An envelope exceedance is calculated using the values of the future climate scenario for a particular BIO-factor and the climate envelope of the tree species for that same BIO-factor. Here, the number of times that a climate variable is outside the bounds of a species’ climate envelope is counted and turned in to a climate exceedance number, per tree species and per BIO-factor. For example, an exceedance value of 0.2 for the factor BIO1 for Spruce, signifies that in 20% of all years, of all CMIP6 scenarios and their associated runs, the yearly average annual temperature has surpassed the climate envelope, and thus the maximum or minimum tolerance levels for Spruce. Thus, the higher the climate exceedance value, the more the models agree on the BIO-factor exceeding the known tolerance levels of the tree species. We assigned a value of -1 if the exceedance was below the lower envelope range, a value of 0 if the exceedance was within the climate envelope and a 1 if the value was above the climate envelope. For each bioclimatic variable we then summarized the SSP245 and SSP585 exceedances over the geographical study area, and divided the result by the total amount of years. The result is raster with the values (-1,1) a value from negative one to one. Risk was calculated across the entire study area, including regions where a given species is not currently present. This approach allows for the assessment of potential future suitability and emerging risks under changing climatic conditions, but results should be interpreted as indicative rather than predictive of actual species occurrence.

For each individual species, we collected the 19 climate exceedance maps together with the usable field capacity and combine the results to create the final climate risk maps. These weighed exceedances, together with the usable field capacity data were combined using the following equation for calculating the climate risk value R:

With the temperature factors being an average of the following envelope exceedances multiplied by their weight ():

The precipitation factors being an average of:

The variation factors being an average of:

The adjusted field capacity was calculated using the following method:

A value of 350 was used as this represents the highest value in the “very high” field capacity class, then rescaled so that 350 represents 1 and 0, -1.

We then calculated the associated risk maps for each tree species and for the SSP245 and SSP585 scenarios. The resolution of the risk map was unified using the resolution of the field capacity map for the entire area of Germany. The forest area was overlaid over the risk maps using the Global Forest Cover 2020 dataset by the Joint Research Centre [59].

3. Results

For all four tree species, we can see a clear shift in the distribution to higher values on average (Fig 1). The largest relative increases in risk level between the two scenarios are for beech and oak, the species with the highest risk levels overall is pine, followed by spruce. The SSP245 scenarios show around 30% of all oak 60% of all beech and 75% of all spruce and pine have a risk level higher than 0.6. In the SSP585 scenario this rises to 65% for oak, 75% for beech, 80% for spruce and almost 90% for pine.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Risk level distributions Risk level distributions for the four selected tree species across two SSP scenarios.

The left bar represents the percentage of the total surface area under different risk levels for the SSP245 scenario and the right bar the same for the SSP585 scenario.

https://doi.org/10.1371/journal.pclm.0000931.g001

The spatial distribution of risk varies greatly between the four tree species included in this study. For beech (Fig 2), in the SSP245 scenario, a large part of Germany is still considered low risk relative to other tree species. These include the lowlands around the northern Rhine, Ruhr and Danube rivers, as well as the prominent “Großes Bruch”, a former marshland area in the federal states of Saxony-Anhalt and Lower Saxony to the north of the Harz mountains. In the SSP245 scenario the risk factor for these lowlands remains relatively low, while the upland regions as well as most of the northern European plain see a significant increase in risk factor.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Risk map for Beech (Fagus sylvatica) Risk map over the area of Germany for Beech in both the SSP245 and SSP585 scenarios.

Base map for this and all subsequent maps: Administrative boundaries from the Federal Agency for Cartography and Geodesy (BKG), dataset “NUTS regions 1:250,000 (status 31.12.)”, available at https://gdz.bkg.bund.de/index.php/default/nuts-gebiete-1-250-000-stand-31-12-nuts250-31-12.html. Licensed under the “Datenlizenz Deutschland – Namensnennung – Version 2.0” (https://www.govdata.de/dl-de/by-2-0).

https://doi.org/10.1371/journal.pclm.0000931.g002

For oak (Fig 3), the SSP245 scenario shows almost no areas with a significant risk factor. We can see that this changes greatly for the SSP585 scenario, with most of northern Germany, as well as the western and central regions around the States of North-Rhine-Westphalia, Rhineland-Palatinate and Hesse showing an increased risk factor. Both pine (Fig 4) and spruce (Fig 5) already show high risk factors overal among the alpine and central areas, with only the lowlands north and south of the Harz mountains on the border with Lower Saxony and Saxony-Anhalt having a low risk factor. The SSP585 provides clearly higher risk values for spruce and pine, than the deciduous tree species, with the highest risk values approaching 1, seen in North-western Germany in the SSP585 scenario for spruce.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Risk map for Oak (Quercus robur) Risk map over the area of Germany for Oak in both the SSP245 and SSP585 scenarios.

https://doi.org/10.1371/journal.pclm.0000931.g003

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. Risk map for Pine (Pinus sylvestris) Risk map over the area of Germany for Pine in both the SSP245 and SSP585 scenarios.

https://doi.org/10.1371/journal.pclm.0000931.g004

Download:

• PNG
  larger image
• TIFF
  original image

Fig 5. Risk map for Spruce (Picea abies) Risk map over the area of Germany for Spruce in both the SSP245 and SSP585 scenarios.

https://doi.org/10.1371/journal.pclm.0000931.g005

3.1. Uncertainties

Ensemble uncertainty in annual mean temperature (BIO1), expressed as the inter-model standard deviation, ranged from 0.96 to 1.96 °C under SSP245 and increased to 1.82–2.63 °C under SSP585. Thus, temperature uncertainty was consistently higher under the high-emission scenario, indicating greater model divergence with increasing radiative forcing.

For annual precipitation (BIO12), relative uncertainty (coefficient of variation) ranged from 14% to 25% under SSP245 and from 15% to 25% under SSP585. In contrast to temperature, precipitation uncertainty showed only minor differences between scenarios, with spatial variability in ensemble spread reaching up to 25% of the mean annual precipitation. These results are consistent with the overal uncertainties found for the models in the CMIP6 project, and do not indicate that our study area represents any increased likelihood of climate model uncertainties [60].

Sensitivity to reductions and increases in upper climatic tolerance limits was consistently higher than to lower-bound adjustments across all species (Fig 6). With reductions of the upper climate envelope limits leading to an average increase of between 0.1 and 0.22 in the risk factor depending on climate scenario and species and increases in the upper envelope limit leading to an average decrease in risk factor of between -0.11 and -0.03. Changing both envelope limits leads to a similar but slightly more moderated effect than changing only upper limits, and changing lower limits had a negligible effect for all tree species and scenarios.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 6. Elasticity barplot Uncertainties as absolute difference from the baseline scenario for each of the four tree species and two SSP scenarios, with a change of 10% in the upper, lower and both the upper and the lower climate envelope bounds.

https://doi.org/10.1371/journal.pclm.0000931.g006

Under both SSP scenarios, mean elasticity exceeded 2.0 for beech and 1.75 for the -10% upper bound sensitivity scenario, and 1.5 for beech and 1.5 for spruce in the upper and lower bound -10% scenario, indicating a disproportionate increase of exceedance risk under the narrowed envelopes. For all species, mean elasticity was most pronounced in the changes made to the upper envelope bound (Fig 7).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 7. Uncertainty boxplot Elasticity of the various sensitivity scenarios each of the four tree species and two SSP scenarios.

https://doi.org/10.1371/journal.pclm.0000931.g007

3.2. Tree species distribution

Next, we compared the distribution of risk to the current species distribution. Using the EU tree species distribution map by the European Forest Institute [61]. We overlaid the land cover area of the four tree species with the risk distribution of the two climate scenarios. In all species we can clearly see that the proportion of forest area within the total area of a high risk level varies per SSP scenario (Fig 8). For beech, a proportional shift between the high and low risk areas can be seen between the SSP245 and SSP585 scenarios, low risk areas where the species is currently present represent less than 10 000 km² for SSP245. And less than 2 000 km² for SSP585. This trend is less so in oak, but oak is present less than half as frequently as beech, so the low risk oak forests are similar in area to low risk beech forests. Both scenarios show a low amount of low risk pine and spruce forests, and even the total suitable area declines dramatically, in an SSP245 scenario, around 100 000 km² of land area will be suitable for pine, 120 000 km² for spruce. While in an SSP585 scenario this will be around 60 000 km² and 70 000 km² respectively.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 8. Land area diveded by risk Total land area of Germany divided by risk level and presence of the four selected tree species.

https://doi.org/10.1371/journal.pclm.0000931.g008

3.3. NDVI comparison

Finally, the risk factor was compared with an NDVI difference raster of the Harz region. Most regions showed a decrease in NDVI, directly resulting from the mass die-offs of spruce forests between the years 2018 and 2022. As spruce represents the majority of the forest area in the Harz region [62], we chose to compare the data with the spruce risk map. Risk was divided in to categories (Fig 9), and the NDVI per category was averaged.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 9. NDVI by risk group The change in NDVI (Normalized Difference Vegetation Index) between the years 2017 and 2022 by risk level categories for the Harz region.

Shown here is the SSP245 scenario.

https://doi.org/10.1371/journal.pclm.0000931.g009

Differences in NDVI change were analysed across discrete risk classes using analysis of variance (ANOVA). A correlation-based approach was not applied, as the risk variable is not continuous but grouped into categories derived from the index structure, which includes discretized components (e.g., field capacity). The aim was therefore to test for differences between groups rather than a linear relationship. Differences in the change in NDVI among risk groups were evaluated using both parametric and non-parametric approaches. A one-way ANOVA indicated a significant effect of risk group (F_4,17046 = 51.04, p < 0.001). Given departures from normality, results were corroborated using a Kruskal–Wallis test, which likewise revealed significant differences among groups ( = 297.47, p < 0.001).

Post-hoc comparisons (Tukey HSD) revealed statistically significant differences between several risk classes; however, the magnitude of these differences was small (generally < 0.015). The relationship between risk class and NDVI change was not monotonic. In particular, intermediate risk levels (0.3–0.6) tended to show slightly lower values compared to low-risk classes, whereas the highest risk class (0.6–1) exhibited marginally higher values relative to several other groups. Overall, these results indicate that while differences between risk classes are detectable, the effect size is limited and does not support a simple linear relationship between increasing risk and NDVI change.

4. Discussion

As our results show, there is a large disparity between the two SSP-scenarios, with all species showing a significant increase in high risk areas. For many years, spruce has been considered a species with a difficult future, due to its lack of adaptation to high temperatures and low rainfall in the summer, and deciduous species such as beech and oak have been considered viable alternatives [63,64]. But our study shows that, in fact, compared to spruce, the other three major tree species will also show significant risk in the future, and are in no way guaranteed to be long term substitutes for maintaining healthy forest cover. The sensitivity and elasticity analysis show that the upper bound of the climate envelope is most sensitive for all species and scenarios, this is most likely due to the outsized effect of high temperatures on tree species mortality.

For the two coniferous tree species in this study, there will be very few areas of suitability left, under both of the two climate change scenarios in this study. The most pessimistic, the SSP585 scenario will almost certainly lead to the widespread destruction of Germany’s coniferous forest area. Spruce and pine make up almost 48% of the forest area of Germany, with oak and beech making up another 25% [49]. With a majority of the forest area in Germany facing increased risk of mortality in the coming decades, strategies need to be developed on how to cope with an increase in future high mortality occurrences in the German forest sector. This study focussed on the four most widely dispersed tree species in Germany, these four species were selected as they had the best distribution data in our species presence dataset. However, it would be of great importance to expand this approach to include other, less common but more drought and heat adapted species in order to assess their suitability to aid in maintaining forest cover in the face of future climate change.

One limitation of this study concerns model validation. While validation is essential for substantiating the results presented here, only a preliminary assessment was possible within the scope of this work, the resulting validation only showed a comparatively small effect size between risk classes a more comprehensive validation is therefore required. Because the risk values derived in this study are based on future climate scenarios, robust validation against observed forest mortality will require the calculation of risk values using historical climate data corresponding to periods of documented decline. Implementing such retrospective analyses constitutes an important next step in the further development of this approach. The analysis also only focuses on climate-related variables directly linked to temperature and precipitation. Other important disturbance agents, such as wildfire, windthrow, or pests, are not explicitly modelled, but are indirectly reflected where they are climate-driven (e.g., drought-related stress). These factors represent important avenues for future model extensions.

Our methodology is grounded in the IPCC definition of risk and uses this framework for the development of the risk maps, which include hazards, vulnerability and exposure. Accordingly, the resulting risk values should be interpreted within this conceptual context. The vulnerability of each of the four tree species, defined through their climatic envelopes in relation to changes in the 19 bioclimatic variables, was found to be substantial, as envelope exceedances occur for all species under all scenarios considered. Hazard is represented by increases in temperature and greater irregularity in precipitation, both of which are evident under the two SSP scenarios. Exposure is likewise high, as forest stands are frequently located in areas characterized by poor soils with low moisture-holding capacity.

These findings are consistent with broader European trends, where warming is driving shifts in species composition and disturbance regimes [7,65,66]. Climate envelope projections indicate declining suitability for Norway spruce and, to a lesser extent, Scots pine across large parts of Central Europe, while oak and pine are expected to expand under warmer and drier conditions, particularly at higher elevations [16,67]. However, such shifts are accompanied by increasing risks from drought, fire, and biotic disturbances, including insect outbreaks, which can amplify tree mortality and accelerate forest die-back. This would result in elevated levels of dead wood, which may locally enhance biodiversity, for example by supporting organisms dependant on dead and decaying wood and increasing habitat availability for certain bird species [68], but increases in dead wood also indicate large-scale ecosystem stress. Consequently, future forest dynamics are likely to involve both losses in currently dominant species and gains in more drought-tolerant species, resulting in novel tree-species composition. From a management perspective, this underscores the importance of promoting structural and species diversity, enhancing soil water retention, and facilitating adaptive transitions rather than relying on single-species replacement strategies.

5. Conclusions

As a way forward, all the three components of risk have to be addressed. Since vulnerability depends on specific species traits, one way to decrease vulnerability and increase forest resilience would be to increase forest diversity. This includes increasing diversity of structure, age and species [69–74]. To address the factor of exposure, we argue to increase diversity of location in this list. Forest resilience would therefore greatly be improved if forests were located in a greater variety of locations. Historically forests in Europe were largely planted or allowed to persist in marginal lands unsuitable for agriculture [75]. These locations are now more at risk from climate change.

A large scale migration of forests away from marginal locations would not be feasibly however, as this would be an issue of conflicts with the availability of arable land. Therefore this diversification would largely be small scale, and main efforts would have to be focused on improving the situation on currently forested land through re-wetting and increasing soil moisture content where possible. This leads to a re-evaluation of the concept of nature restoration. As risk increases, restoration efforts can no longer focus on restoring forests on the basis of past conditions. But must increase adaptation to be able to deal with new climatic occurrences beyond what was experienced in the past.

Acknowledgments

We thank Günay Najafzade, for calculating the NDVI data used in this study and for her valuable contribution through her Master’s thesis entitled “Spatial heterogeneity of forest mortality in the Harz region, Germany.”