To analyze the future composition of tree species and their contribution to different ecosystem services, we based our study on seven major native tree species in Austria: Abies alba, Fagus sylvatica, Larix decidua, Picea abies, Pinus sylvestris, Quercus robur and Quercus petraea, which cover in total around 84% of Austria’s managed and stocked forests (www.waldinventur.at). Since the two closely related white oak species share a similar distribution area in Austria, and are not discriminated in national forest inventory data, these two species were treated together as Quercus spp. Data from the occurrence and productivity (average increment) of these seven species was obtained from the Austrian forest inventory (period 2007–2009), which encompasses the entire federal territory and comprises 3141 permanent survey sites, each consisting of up to 4 plots.

To evaluate the possible contribution of NNT to forest ecosystem services, nine NNT were evaluated for their climatic suitability and their potential contribution to ecosystem services in Austria under climate change. These nine species were selected based on the following criteria: 1) species that are already well known, tested or established in Austria as forest species for forestry use (Abies grandis, Juglans nigra, Pseudotsuga menziesii, Quercus rubra, Thuja plicata), 2) species introduced to Austria as ornamental plants which are, however, spreading in forests due to their invasive characteristics (Fraxinus pennsylvanica, Robinia pseudoacacia), 3) species with worldwide importance as timber species but not yet present in Austrian forests (Pinus contorta, Pinus radiata), 4) availability of sufficient data to establish species distribution models and productivity as basis for an estimation of ecosystem services (all nine NNT). The level of knowledge in terms of their ecological characteristics, their invasive potential and cultivation in Central Europe varies widely for these nine species ( Supplementary Table S1 ).

To assess the suitability of the seven native tree species and their contribution for delivering ecosystem services in future forests, we used species distribution models (SDMs) that have been established previously by Chakraborty et al. (2021). As similar models were not available for the NNT of interest, the current and potential future distribution (probability of occurrence) of the nine NNT was estimated following the same modelling approach as for native species (Chakraborty et al., 2021). To reduce the uncertainty arising due to modelling algorithm, input data and assumptions therein a multi model ensemble approach was selected. This ensemble model included 10 different SDMs implemented through the Biomod2 platform (Thuiller et al., 2013). Assumptions for each models in the Biomod2 are also detailed in the Supplementary Material ( Supplementary Table S3 ).

10 modeling algorithms were selected: GLM (Generalized Linear Models), GAM (Generalized Additive Models), GBM (Generalized Boosted regression Models), CTA (Classification Tree Analysis), ANN (Artificial Neural Networks), SRE (Surface Range Envelop or BIOCLIM), FDA (Flexible Discriminant Analysis), MARS (Multivariate Adaptive Regression Spline), RF (Random Forest for classification and regression), and MAXENT. The SDMs were calibrated with the observed occurrence and bioclimatic variables of the NNT in their natural range and their introduced range in Europe ( Supplementary Table S2 ). The bioclimatic variables used to calibrate the SDMs were obtained from Worldclim database v2.0 (Fick and Hijmans, 2017). The occurrence data for the nine NNT at their native and introduced range in Europe was obtained from various sources such as national forest inventories, global biodiversity facilities, etc. as collected by the EU-COST Action NNEXT (Brus et al., 2019). The occurrence dataset for each target species was partitioned by splitting into 75% for model training and 25% for model evaluation. The dataset includes a total of 754, 413 occurrence records of the target species (i.e. presence locations). The assumptions and detail of input data are mentioned in Table S3 in the Supplementary material . With this approach the potential distribution of each species was estimated by each of the 10 SDMs in biomod2 in historical climate (1961–1990) and two future climate scenarios, RCP 4.5 and 8.5 for the periods 2041–2060, 2061–2080, and 2081–2100. Potential distribution in future scenarios is based on the mean of the 13 RCM projections available in Worldclim database v2.0 (Hijmans et al., 2005). Predicted probabilities from the individual models for each target species were ensembled as a consensus model which combined the median probability over the selected models with the true skill statistics threshold (TSS > 0.7). The predicted potential distributions for historical and future climates are available as GeoTiff rasters with a spatial resolution of 30 arcsec, which is roughly equivalent to 1x1 km or lower depending on latitude, in WGS 84 projection from ZENODO (Chakraborty et al., 2024). The Supplementary Material provides a more detailed account of the SDMs.

For the actual occurrence and species share in Austrian forests, the data from the four inventory plots per survey site were aggregated to the centroids of the sites. The mean values were used to aggregate the values for the tree species data, while the median was used to aggregate the slope and altitude of the respective survey site. This approach was taken to increase accuracy and precision. If none of the 15 tree species ( Table 1 ) was currently present at the centroid, then this centroid was neglected for further calculations. Out of a total of 3141 centroids, 117 were thus removed from the dataset. So, the models were applied to 3024 centroids. The calculations and graphical representations were executed in Python 3.11, using essential libraries such as Pandas, NumPy, GeoPandas, Rasterio, Matplotlib and Pickle.

To specify the potential tree species composition at the inventory centroids, the probability of occurrence, respectively the suitability under past (1961–1990) and future climate (2081–2100, RCP 4.5 and 8.5) were retrieved from the species distribution models of all native and non-native species. The continuous probability values of the ensemble SDMs were transferred into binary presence-absence predictions by applying species-specific mean cut-off values across the individual SDMs, as obtained from the biomod2 results as thresholds ( Table 1 , column 3). Species that showed suitability below this threshold at a given site were considered unsuitable and systematically assigned to a zero value. When calculating the climatic suitability for the native oak species, SDM data from the two individual species Q. petraea and Q. robur were used and aggregated to Quercus spp. in order to match to the data of species compositions and growth performance of the Austrian forest inventory which does not differentiate between these species. Consequently, for the future species projections at each centroid, those oak species was selected that had a higher suitability based on the established threshold.

Tree species that are occurring today according to the forest inventory but were found to be unsuitable for the RCP 4.5 or 8.5 scenario at a given centroid, were considered to be unable to regenerate and to become locally extinct in the long term. The vacated space at the respective centroid was either filled with other suitable tree species or left unstocked if none of the 15 studied native or non-native species could be identified as suitable. The potential planting of tree species at vacant sites was simulated according to five different species change scenarios ( Table 2 ).

To calculate the potential future tree species composition and the resulting contributions to the ecosystem services, we considered several tree species change scenarios ( Table 2 ). These scenarios are based upon the modelled probability of occurrence of the tested native and non-native species and their share within the forest inventory. We assume that tree species that were modelled to be not suitable (respectively absent) under future climate will get locally extinct throughout the course of the next decades. Depending on the share of this failing tree species in current forests, this may result in a loss of the considered ecosystem services.

The first scenario, referred to as ‘no adaptation’, represents the zero variant where no active adaptation of the species composition to climate change would occur. Thus, failing species could be replaced only by species which already occur today at the respective inventory centroid given that these species were found to be suitable. Otherwise, the centroid will be left unstocked. All other scenarios presume active adaptation by reforesting failing tree species either by suitable native tree species or by suitable NNT. In total four active adaptation scenarios were defined, which differ in terms of its consideration of NNT and in terms of replacing failing species either with the most suitable tree species or by replacing conifers preferably with suitable conifers and broadleaved trees preferably with suitable broadleaved trees. Centroids for which no climatically suitable species could be identified among the available tree species spectrum, would be left unstocked.

To depict the current composition of tree species, we calculated the frequencies of all native and non-native species across the inventory centroids. It is important to note that the present analysis focuses on seven stand-forming and most notable native species that cover in total around 84% of the stocked and managed forest area and present around 90% of the growing stocks in Austrian forests. Other native species are mostly minor, rare and scattered distributed species (i.e. Acer spec., Sorbus spec.) or face endangering diseases (e.g. Fraxinus spec, Ulmus spec.). Therefore, these species are not assumed to fully replace potentially climatically unsuitable stand-forming species and their contributions to ecosystem services. Also, this minor species lack sufficient data on potential distribution and growth potential and were not considered in the present analysis. To quantify the tree species diversity at a given inventory centroid, the tree species richness (number of tree species per plot) was calculated.

To estimate the potential value of future species composition to provisioning ecosystem services, we used the tree species’ annual increment as productivity indicator ( Table 1 ). The mean annual increment has been obtained for the different species either on basis of the Austrian forest inventory for the native species or by mean values calculated from published studies of NNT ( Table 1 ). To obtain a dimensionless contrast among species, the annual volume increment has been standardized across the mean increment of all species. Although the actual productivity at a given centroid is strongly affected by local site conditions, tree age, management etc., our basic analysis assumed the annual increment to be equal across all centroids and the course of ongoing environmental changes.

The standardized increment value of the tree species apparent on the centroid were averaged and the centroids are categorized into three groups: high production (>0.78), moderate production (0.63–0.78), and low production (< 0.63). These categories were defined by the two native tree species with the highest increment (Abies alba and Picea abies) and the two with the lowest increment values (Pinus sylvestris and Larix decidua) ( Table 1 ).

P. abies is the most common species at the Austrian forest inventory now and will remain so in the future, regardless of the tree species change scenario ( Figure 2 ). The frequency of L. decidua and P. sylvestris on forest inventory plots is expected to decrease in the future compared to the present. The distribution area of native oaks (Q. petraea & Q. robur) in Austria is expected to expand strongly in the future, thus increasing their frequency in the total forest area. J. nigra and P. menziesii are the most frequent NNT. R. pseudoacacia is modelled to occur only on 0.26 - 0.5% of the forest area in the future. The NNT A. grandis, F. pennsylvanica, P. contorta, and Q. rubra only show low share in present forests and are also expected to occur only rarely in the future.

The predicted composition of tree species varies considerably according to the specific species change scenario and climate scenario in question ( Figure 2 ). For instance, Pinus radiata is present in the CC/BB scenarios in RCP 8.5, but not in any other combination of species change and climate scenario ( Figure 2 ). In contrast, for other tree species such as Picea abies and Pinus sylvestris, the decisive factor is the climate scenario. The highest tree species richness can be observed under current conditions, indicating that climate change both with and without active adaptation measures will result in a substantial decline of tree species richness ( Figure 3A ). It is also clear that the more extreme climate scenario RCP 8.5 has a more negative impact on tree species richness than the scenario RCP 4.5. Differences in the impact on different environmental zones are also clearly visible. The negative consequences for the Pannonian are the most prominent, while in contrast, the Continental Zone and the Alpine South exhibit a markedly reduced range of variation in the differences between the individual species change scenarios.

The mean share of NNT in Austrian forests is around 1% at present in de Alpine South and the continental zone, while in the Pannonian it nearly reaches 20% ( Figure 3B ). Without active adaptation or if adaptation is excluding NNT, the mean share of NNT may drop down in all three zones. As expected, including NNT in forest adaptation increases their mean share. The proportion of NNT would clearly be highest in Pannonian and could be over 60% under RCP 8.5.

Our results indicate that timber production would decrease without adaptation to climate change ( Figure 4 ). Areas that would be affected by a loss of production are located mainly in at the Pannonian zone and the Continental zone, but also within inner alpine valleys in western Austria, which are climatically characterized by dry continental climate ( Figure 4 ). If only native tree species are used for adaptation, the mean increment in the Alpine south and the Continental zone would remain similar as it is today or only change slightly, while it would clearly decrease in the Pannonian zone ( Figure 5 ). The map indicates that a limited number of study sites may exhibit a slight increase in timber production when native conifers are replaced with other native conifers and native broadleaves with native broadleaves (Native-CC/BB). However, the overall number of moderately and highly productive study sites appears to decline ( Table 4 ). If NNT are included as adaptation option, the mean production value of the Continental and Pannonian zone and the total number of high productive study sites for Austria would be higher than it is today ( Table 4 ). This would be especially beneficial in areas that already have low productivity today.

For reasons of space, only the maps for RCP 8.5 are presented here ( Figure 4 ). The map for RCP 4.5 (see Supplementary Figure S5 ) exhibit a comparable pattern to that observed in RCP 8.5, albeit to a lesser extent.

The criteria for the selection of study sites at which the avalanche and rockfall control is analyzed (see methods section) result in the study sites being limited to the Alpine South and the Continental Zone. For reasons of space, only the maps for RCP 8.5 are presented here ( Figures 6 , 7 ). The maps for RCP 4.5 ( Supplementary Figures S6, 7 ) exhibit a comparable pattern to that observed in RCP 8.5, albeit to a lesser extent.

Due to the topographical diversity in Austria, the forest types vary naturally greatly among the Alpine south, the Continental and Pannonian zone. Our findings indicate that the dry lowland in Austria’s east, which is part of the Pannonian zone, will particularly face the consequences of the changing climate. For many decades, coniferous tree species were planted for forest management there. These include Pinus sylvestris, Picea abies, Larix decidua and Abies alba. Our predictions indicate that these conifers, along with Fagus sylvatica, will no longer thrive in the future due to unsuitable climatic conditions in the Pannonian zone. As a result, the tree diversity and timber yield of the forests in the region will be drastically reduced. If native tree species are unable to migrate to higher altitudes, as is the case in the lowlands of Austria, they are likely to face extinction in the region (Jandl et al., 2019). One way to counteract the loss of timber production in lowlands (Continental and Pannonian zone) with future climate could be the use of NNT (Brang et al., 2016) and our results show potential benefits of such an approach. Especially, lowland areas with low rainfall, like the Pannonian zone, could benefit of a combination of native species and NNT ( Figure 3 ).

As shown in Figure 2 , the good climatic suitability of native oaks is evident under future climatic conditions. The two native species, in particular Q. petraea, are species of the colline and montane stages (Aas, 2014a, 2014b). This makes them the ideal native species for the management of lowland forests. The non-native red oak (Q. rubra) is not a suitable replacement species in these forests, as it is less suited to the climate and could also cause conservation problems (Dreßel and Jäger, 2002; Vor et al., 2015). Therefore, the native oaks should clearly be favored over the non-native Q. rubra. To achieve a balanced mix of tree species, also other tree species should be cultivated next to native oaks in lowland forests. One of the most popular tree species for mixing with oaks is the native species hornbeam (Carpinus betulus) (Ruhm et al., 2021; Schönauer and Schüler, 2021). However, due to the anticipated reduction in the suitable range for hornbeam (Varol et al., 2022), this combination may not be possible in the future. According to our results, tree species such as Pinus sylvestris or P. radiata could be suitable mixed tree species for oak forests.

According to the results, a decrease in the proportion of Black locust (Robinia pseudoacacia) can be expected in both climatic scenarios. This decrease may be caused by the models used here, as the study sites only represent forest area and not open land. While climatic suitability is a main driver of a species’ distribution other factors such as its competitive strength (e.g. rapid growth and change in soil chemistry due to nitrogen deposition) are not considered in our calculations. Black locust is highly competitive in open land (Huntley, 1990) and considered invasive in many European countries, including Austria (Essl and Rabitsch, 2002). Due to its additional good adaptation to drought, black locust is expected to thrive in Austria’s changing climate (Mantovani et al., 2014; Vítková et al., 2017; Nicolescu et al., 2020). Contrary to our results, an increase in the occurrence of R. pseudoacacia in Austria can therefore be expected in the future, especially in open land (Kleinbauer et al., 2010; Li et al., 2014). R. pseudoacacia is known for its tendency to colonize open land (Huntley, 1990). The data only refers to Austrian forests and do not reflect this occurrence. Therefore, it can be assumed that the black locust is more widespread in Austria today than the data shows. Nevertheless, the reliability of the National forest inventory data is remarkable for tree species of low abundance. The data show a meaningful correlation with actual occurrences, underlining the credibility and applicability of the method even for less common tree species.

Besides their potential to provide ecosystem services in commercial forests, NNT can pose several risks. NNT can become invasive, spreading from planting sites into areas of high ecological value and sometimes causing irreversible damage to biodiversity and related ecosystem services (Richardson et al., 2000; Richardson and Rejmánek, 2011; Castro-Díez et al., 2019). The recommended use of site-specific risk assessments helps to detect and evaluate the risk response and intensity under certain site conditions (Bindewald et al., 2021; Lapin et al., 2023).

Douglas fir (P. menziesii) is one of the most used NNT in Europe. It has no known negative impacts on Central European forest communities except on two (Luzulo-Quercetum petraeae HILITZER 1932 and Deschampsia flexuosa-Acer pseudoplatanus - Association) (Bauhus et al., 2017; Bindewald and Michiels, 2016). Western Redcedar (T. plicata), which can grow in a range of habitats (Antos et al., 2016), has been categorized as invasive in the UK (Richardson and Rejmánek, 2004) and potentially invasive in Belgium (Fanal et al., 2021). Unlike the Douglas fir and the Western Redcedar, the grand fir (A. grandis) does not appear to have the potential to invade any habitats in Central Europe (Spellmann et al., 2015). The Monterey pine (P. radiata), one of the most frequently cultivated species of pine (McDonald and Laacke, 1990), is already considered invasive in some countries (Higgins and Richardson, 1998; Lindenmayer and McCarthy, 2001; Williams and Wardle, 2009) and the lodgepole pine (P. contorta), is widely regarded as the most invasive Pinus species in the southern hemisphere (Davis et al., 2019). Due to its ability to allelopathize (Rietveld, 1983), black walnut (J. nigra) can have a negative effect on the native species composition of the understory (Nicolescu et al., 2020). And the last of the nine non-native species, the green ash (F. pennsylvanica), is so ecologically adaptable that it can take advantage of almost any opportunity to invade, with floodplain habitats being particularly favorable (Drescher and Prots, 2016).

In the case of T. plicata, P. contorta, Q. rubra and F. pennsylvanica, both the results of the modelling carried out here and the potential invasive risks speak for themselves. These species are not suitable as future tree species in the face of climate change for either the Alpine south, the Continental or the Pannonian zone. Due to its high risk of invasion, R. pseudoacacia should also not be encouraged in forestry. Forestry utilization can be considered for the NNT species P. menziesii, A. grandis, P. radiata and J. nigra. However, it is important to take a number of management measures into account when cultivating NNT for forestry purposes (Marinšek et al., 2022). For instance, NNT should only be cultivated in a mixture with native species.

It is becoming clear that forest management will need to adapt to climate change in order to counteract potentially declining timber production in the future. The continued use of native species such as Norway spruce (P. abies) can maintain production levels in Alpine South of Austria at least comparable to current levels. However, the situation in the Pannonian zone and partly in the Continental zone is different ( Figure 4 ). Due to climate change and resulting calamities (e.g. bark beetle outbreaks or storm events), there is a growing need to adapt species selection in forestry. The current dominance and the expected decrease of Norway spruce - known for its high yield – is thus one driver of lower future timber production (Hanewinkel et al., 2010). Especially, secondary pure stands of Norway spruce planted in lowlands beyond its native habitat (Spiecker, 2004).

Many forest managers are now using a balanced mixture of Norway spruce and other (broadleaved) tree species when replanting secondary Norway spruce forests (Russ, 2019) (Hanewinkel et al., 2010). The use of both native and NNT can sustain production levels. In some cases, this combined approach can even lead to increased production compared to current levels. Specifically, if coniferous tree species are preferred as a replacement for declining conifers (CC/BB), a substantial increase in production (increment) compared to today can be expected in the Continental zone. The reason for this higher productivity can be explained by the change in species composition. Our results indicate that Norway spruce trees at secondary sites, where they might grow suboptimal, are being replaced by more productive and better suited species (A. alba, P. radiata und P. menziesii). Anyhow, an extended plantation non-native trees such as P. menziesii or P. radiata will also require adaptations in the Austrian timber industries as they provide different wood qualities and assortments and might therefore be suitable for other timber applications (Huber et al., 2023). However, reforestation action in the coming decades will only deliver sawnwood by the end of the century, which gives sufficient times for wood industries to update its technologies as innovations may take 25–30 years to reach the market (Teischinger and Müller, 2010). Besides wood qualities, the consideration of forest growth and productivity is of utmost importance, as the value chain from the European forestry industry and wood sector totals around 1.1 trillion euros (EU 27, Norway, Switzerland, United Kingdom). In total, the industry thus secures around 17.5 million jobs across Europe (Econmove and Economica, 2023). Therefore, the forest and its resources, specifically wood, play a significant role in the domestic economy.

Another important ecosystem service related to forest productivity is the forest carbon sink, i.e. the sequestration of carbon from the atmosphere, which was recently found to be strongly reduced by the loss and future lack of native tree species (Wessely et al., 2024). Although, we have not quantified the forest carbon sink explicitly, the improved productivity of mixtures with native and non-native conifers, would also reveal the highest carbon sink and thus help to improve or continue the mitigation of climate change (Lorenz and Lal, 2010).

The protective function is of great importance throughout the Alps. This importance is illustrated here by the example of Austria. In Austria, almost every fourth citizen benefits from the function of protective forests (BML, 2023). In total 1.6 million hectares, or 42% of Austria’s forest area, are classified as protective forests (BML, 2023). Over 25% of these forests are directly protecting infrastructure and settlements against geohazards such as avalanches and rockfalls (Berger et al., 2013). Our results show that the selection of appropriate tree species depends on the primary protective function of the forest and varies by objective, i.e. whether protection from avalanches or rockfall are more likely to occur. It is also worth mentioning that that mountain forests are widely recognized as the most effective, cost-efficient, and aesthetically pleasing form for protecting against natural hazards (European Observatory of Mountain Forests, 2000).

Modelling and visualizing future forest development in the face of climate change is crucial for forest managers and decision makers to develop adaptation strategies. However, it is important to interpret the modelling results objectively, as each approach is based on specific assumptions and development scenarios. The study was conducted based on a climate change scenario that assumes a temperature increase of up to 4.3°C by the end of the century (RCP 4.5 and 8.5). Any deviation from this temperature increase assumption could render different results. The greatest source of uncertainty in our study is therefore the uncertainty surrounding our future climate. We were unable to consider other RCP models in this study, as the SDMs we used are currently only available for RCP 4.5 and RCP 8.5.

Another limitation of our study is the use of tree species distribution models for the end of the century in conjunction with current forest stands. This assumes that the current forest stands will remain relatively constant over the next few decades. However, it is uncertain whether this assumption is correct and depends on various factors, such as the current age distribution of forests, current harvesting regimes, and national and European policy decisions. In recent decades, there has been a continuous increase in forest area and stock in Austria (SChadauer, 1996; BMNT, 2018). The first signs of stock and carbon saturation have only been observed in the last decade. Therefore, it is unclear whether assuming a constant forest stock for the next 50 years is realistic. To clarify this question, a modelling study based on empirical forest growth models that takes all of these variables into account would be necessary.

It is also important to note that our analysis does not consider local site characteristics such as exposition or soil type, as these were not part of the SDMs we used. Sustainable forest management adapted to local conditions could enable the survival of tree species at their climatic limits, even if they are currently classified as non-viable in our models (e.g. Norway spruce on north-facing slopes at lower elevations). Therefore, is it important to consider the potential benefits of adapting forest management practices to local conditions. As the introduction of NNT is not without controversy, we further recommend that a site-specific risk assessment should always be carried out before introducing them (e.g. Bindewald et al., 2021).