Importance of climate‐induced tree species composition changes in forecasting the amount of reachable habitat for forest birds

Bird and tree species distributions will shift along with future climatic conditions through direct effects related to species responses to bioclimatic variables. In addition, tree species composition changes should indirectly drive changes in forest bird habitats. Here, we predicted the effects of climate changes on the amount of reachable habitat for forest birds and teased apart the role of the indirect effects.


| INTRODUC TI ON
Climate changes drive species distribution through direct impacts related to species responses to climatic variations that constrain fitness and metapopulation processes (De Kort et al., 2020;Møller et al., 2010). Species' responses depend on their sensitivity to climatic conditions, which is exacerbated at range distribution limits (Devictor et al., 2008;Morin et al., 2007;Vissault et al., 2020). For example, thermal maximum or the temperature at the hot edge of the climate envelope was used as an indicator of birds' sensitivity to climate changes (Jiguet et al., 2007(Jiguet et al., , 2010. Sharp population declines have been observed for species with low thermal maximum, that is an affinity for cold temperatures (Jiguet et al., 2007(Jiguet et al., , 2010. Thus, the consequences of climate changes on a species are often seen through the lens of shifts in the species range distribution (e.g. Barbet-Massin et al., 2012). However, regional-scale species distribution can be impacted even within species range distribution because climatic changes can modify the distribution of ecosystems and thus alter species habitat availability and landscape connectivity (Ay et al., 2014;De Kort et al., 2020). Landscape connectivity is defined as the degree to which the landscape facilitates or impedes species movements between habitats (Taylor et al., 1993).
To understand these mechanisms, we need to account for land use change, understand how climate indirectly drives the distribution of dominant vegetation and clearly identify the cascading effects of climate changes on species dynamics (He et al., 2019). For example, our future climate will drive changes in tree species biogeographical distribution, local environmental conditions and forest management, and these processes will in turn drive changes in habitat availability and connectivity for forest-dwelling species (De Kort et al., 2020). Accounting for climatic and land use changes in modelling habitat availability and connectivity should enable us to understand the consequences of climate changes on metapopulations via direct effects (bird responses to bioclimatic variables through the direct occurrence-climate relationship) and indirect effects (climateinduced changes in tree species composition and their cascading effects on birds).
Tree species distributions are predicted to shift under future climatic conditions (Hanewinkel et al., 2013). A projection of the climatic envelop of European tree species have identified "winner" and "loser" species (Dyderski et al., 2018). However, large-scale predictions of the future distribution of individual tree species do not give enough information to forecast the spatial rearrangement of tree species when climatic conditions remain suitable (within species distribution range) for several tree species at the same location. For that purpose, it is necessary to account for local environmental conditions (soil, topography) and management trajectories related to socio-economic factors (Houballah et al., 2020). Even if options for forest management are constrained by climate changes in the temperate continental regions, several trajectories remain possible, such as favouring deciduous or coniferous trees (Lindner et al., 2010). Further investigations are needed to understand how climate-driven changes would translate into future forest stand composition. Studying a combination of climate change scenarios and land use change scenarios related to contrasted trajectories of forest management could fill this knowledge gap.
Climate-induced land use change could indirectly drive forest species distribution via changes in forest stand composition and structure (Ay et al., 2014). In the context of global changes, bird population trends are strongly correlated to habitat use, especially for specialist species, which can decline very rapidly when their habitat is degrading (Jiguet et al., 2007). Studies that forecast the future distribution of birds recognize the importance of accounting for dispersal but generally overlooked habitat connectivity, notably because the spatial scale of analysis is too large (Barbet-Massin et al., 2012;Devictor et al., 2008;Triviño et al., 2011). Indeed, species distribution models do not incorporate habitat connectivity, which depends on the area, quality and spatial configuration of the suitable habitats embedded in a more or less unsuitable environment, the so-called landscape matrix, which facilitates or impedes species dispersal among populations (Riva & Nielsen, 2020). To fill this gap, we need new methodological frameworks that integrate climate changes into landscape connectivity approaches (Costanza et al., 2019;De Kort et al., 2020;Keeley et al., 2018;Littlefield et al., 2019). To tackle this issue, species distribution models and landscape graphs could be combined (Duflot et al., 2018;Keeley et al., 2016). A landscape graph is a conceptual representation of a landscape where pairs of suitable habitats are connected by paths depicted in the landscape matrix.
Paths represent potential movements of individuals or gene fluxes between habitats (Urban et al., 2009). For example, they can be depicted by least-cost paths that minimize both Euclidean distance between habitats and matrix resistance to movements (Adriaensen et al., 2003). Matrix resistance can be derived from habitat suitability, using, for example a negative exponential transformation of suitability to strengthen the barrier effect of the least favourable areas (Keeley et al., 2016). While species distribution models only quantifies the amount of potential habitat, the amount of reachable habitat accounts for habitat area, quality (or mean suitability) and spatial configuration, matrix resistance and species dispersal distance into a single graph-based metric (Saura et al., 2011;Saura & Pascual-Hortal, 2007).
In this study, we evaluated how the future climate in 2050 combined with forest management trajectories would drive the K E Y W O R D S dispersal, ecological corridors, ecological niche modelling, ecological or habitat networks, forest management, fragmentation, global changes, habitat or patch dynamics, habitat suitability, landscape, multiscale amount of reachable habitat and isolated the specific role of climate-driven tree species composition change within the global effects of climate changes. More precisely, we combined two contrasted climate change scenarios (RCP 2.6 and 8.5) and two forest management scenarios (coniferous increase or deciduous increase), in a forest area with high compositional heterogeneity (662 km²). We selected five bird species with diverse ecological requirements in terms of tree species composition (Regulus regulus, Sitta europaea, Dendrocopos major, Certhia brachydactyla and

Phylloscopus collybita).
First, we applied species distribution models to relate current focal bird and tree species occurrences to climatic variables at the national scale (France), which was largely enough to prevent spurious extrapolation when assessing future conditions at the regional scale.
Second, using other databases on bird and tree species available at the regional scale, we applied species distribution models to the same focal bird and tree species at this regional scale in order to account for topography and soil for trees and landscape compositional heterogeneity for birds (including forest land use, i.e. tree species composition).
Third, for trees, we combined national-scale and regional-scale distribution models to build future scenarios of tree species composition for all forest stands present in our study area. For birds, we combined national-scale and regional-scale models to quantify the amount of reachable habitat for birds through landscape graphs and equivalent connectivity EC (Saura et al., 2011).
Using the above modelling framework, our objectives were to (1) examine the direct, indirect and global effects of climate changes on birds using EC variations, and relate these effects to bird thermal maximum; (2) identify whether alternate forest management scenarios could mitigate the impact of climate changes on birds and thereby meet bird conservation objectives.

| Study area and model species
The study area (662 km²) was located in the northern part of the French central massif (latitude: 46.0022°, longitude: 3.8319°) and was composed of forests (57%), agricultural lands (37%), urban areas (5%), and streams and water bodies (1%; see Appendix S1). Altitude ranged from 298 to 1286 m and the forests were composed of coniferous stands (Pseudotsuga menziesii and Abies alba), deciduous stands (Fagus sylvatica and Quercus spp.) and mixed stands (A. alba and F. sylvatica or Quercus spp.; Table 1; see Appendix S1). Tree species composition of the stands was extracted from the attribute table of the vector layer BD Forêt ® v.2 map (2006) produced by the French National Institute of Geographic and Forest Information from orthophotos. The polygon composition specifies the dominant tree species if a single species occupies over 75% of the canopy cover (two oak species are grouped, i.e. Quercus spp. = Quercus petraea and Quercus robur), or indicates the tree species mixture if at least two species occupy over 25% of the canopy cover. We considered 77% of the forested area to be dynamic in our future projections, and excluded mixed-deciduous stands (for which the exact tree species composition was unknown) and other dominant trees that comprised less than 1% of the forested area. Nonforest areas (agricultural lands, urban areas, streams and water bodies) were considered stable over time.
We extracted bird data from the French Bird Protection League database for the year 2019. This database includes either presence-only bird points or bird abundance on 100 m-radius plots inventoried using songs and visual contacts. We used abundance data to calculate the degree of association between forest birds and five forest stand types according to an indicator value index (indval, Dufrêne & Legendre, 1997). We then selected four species-R. regulus, S. europaea, D. major and C. brachydactyla-as model species for forest stands dominated TA B L E 1 Tree species composition in 2020 and in 2050, according to the four scenarios. Areas are expressed in ha and as a percentage of the total forest area considered dynamic. Tree species composition is defined according to the dominant tree species based on relative tree cover. DEC 2.6: deciduous RCP 2.6, CON 2.6: coniferous RCP 2.6, DEC 8.5: deciduous RCP 8.5, CON 8. menziesii. Beyond the responses of birds associated with specific tree species, we also aimed to study a generalist species and included P.
collybita as a fifth model species. The indicator value showed no preferential association between this bird and specific tree species.

| National data
At the national scale, the occurrences of the tree species, previously described, were extracted from the data collected between 2005 and 2015 in 1-km squares by the French National Forest Inventory (Vidal et al., 2007), and occurrences of the five bird species models were extracted from the French National Atlas of Breeding Birds (2009)(2010)(2011)(2012)(2013) in a regular grid of 10-km squares (Issa & Muller, 2015).
We used the national scale to totally avoid spurious extrapolation when we predicted species response to 2050 climatic conditions currently inexistent in our study area.

| Species distribution models
The modelling framework that links species distribution and connectivity models to climate and land use changes is summarized in Figure 1. Species distribution models were fitted for each tree and bird species with generalized additive models (GAM; Wood, 2017).
Smoothing splines were limited to three degrees of freedom to avoid overfitting. Models were validated with the area under the curve (AUC) when using occurrence data, and AUC and the Continuous Boyce Index (CBI) when using presence-only data (Fielding & Bell, 1997;Hirzel et al., 2006). These indices were averaged over 20 replicates of a bootstrap cross-validation (80% and 20% partitioning for calibration and validation, respectively).
First, we related tree and bird species occurrence to climatic variables at the national scale to project species response according to 2050 climatic conditions. We selected four bioclimatic variables (1 km resolution at the equator), extracted from the Chelsa database (Table 2; Karger et al., 2017), that were not too much correlated to each other (Spearman correlation coefficient <0.7; Dormann et al., 2013). Bird occurrences in the 10-km squares were related to bioclimatic variables averaged over the 10-km squares ( Table 2; Issa & Muller, 2015). However, we then used this model to predict bird occurrences at the original resolution of the climatic data (1 km) under the hypothesis that the bird occurrence-climate relationship would not change between the two scales (1 km and 10 km). This assumption is reasonable for studying the variation in equivalent connectivity with time, but could lead to a loss of accuracy in the predictions of presence probability for each date.
Second, we fitted regional-scale models to account for local environmental conditions. We related tree species occurrences to topography and soil acidity, both of which were considered stable over time ( Information (see Appendix S2) for sampling bias correction and pseudo-absence selection. Bird occurrence was related to the proportion of each land use class at a 5-m resolution (including forest land use based on tree species composition). These landscape metrics were calculated in moving windows within a radius of 100 m in accordance with species home-range size ( Table 2; see Appendix S2).
Finally, for birds, the national and the regional-scale GAM were used to predict species occurrences in 2020, using an occurrence probability threshold following Liu et al. (2013). Species distribution models and GIS manipulations were performed with R software v.3.6.2 (R Core Team, 2021). . We used two scenarios, RCP 2.6 and RCP 8.5, which F I G U R E 1 Representation of the modelling framework connecting species distribution and landscape connectivity models to climate and land use change. Species distribution models that relate species occurrences with climatic conditions were built at the national scale (a, b), which was enough to totally prevent spurious extrapolation when assessing future (2050) climatic conditions. Regional-scale models refined the national-scale climatic models by accounting for topography and soil acidity for trees, and landscape factors for birds (a, b). Nationaland regional-scale models were combined according to distribution overlap (spatial intersection) and occurrence probabilities that were averaged (a, b). Land use scenarios where built at the regional scale according to the respective habitat suitability of tree species and specific decision-making rules for tree species changes between 2020 and 2050 (a; Appendix S1). Landscape connectivity models were also built at the regional scale (b). We evaluated the differences in equivalent connectivity between 2050 and 2020 (c, ΔEC ij , Equation 3 Land use scenarios 2050 (comparison of suitability for all trees)

| Projection of future tree species composition
Regional tree suitability map Combination of regional maps (habitat suitability averaging) (b) Species distribution models and landscape graph for birds (c) Comparison of the amount of reachable habitat for birds (d) Isolating the indirect effects of climate changes on bird (due to changes in tree species composition) (a) Species distribution models for trees TA B L E 2 Description of variables used to relate tree and bird species with climate at the national scale, and topography and soil acidity (for trees) or landscape factors (for birds) at the regional scale respectively predict an increase of 1.1°C and 2.1°C in mean annual temperature by 2050 in the study area. In addition, we tested two different forest management scenarios, which predict overall changes in tree species composition in 2050 but no change in total forest area. The "coniferous" scenario predicts an increase in the share of pure coniferous stands and reflects intensive management devoted to the production of construction and industrial wood. The "deciduous" scenario predicts an increase in the share of pure deciduous stands and foresees an adaptation of the wood sector towards wood energy and a decrease in the introduced species P. menziesii.
In both scenarios, the current mixed coniferous-deciduous stands were considered to be stable over time, as long as it was consistent with the future tree species distributions. We combined the two climate scenarios with the two forest management scenarios and compared them to the current situation. We labelled the scenarios by merging climate and management scenario codes (e.g. "deciduous 2.6"). We also considered an additional third scenario that consisted

| Landscape connectivity modelling for birds
For birds, we defined a habitat suitability index (HSI), at a 5-m resolution, as the mean occurrence probability from the national and the regional model predictions combined. In the landscape matrix, resistance to dispersal was derived from the HSI with a negative exponential transformation, following Keeley et al. (2016): The parameter c, that determines the shape of the curve, was fixed to four considering an intermediate strength of barrier effect of the least favourable areas between habitats. A variation of this parameter rarely results in significant differences in model outputs (Meyer et al., 2020;Zeller et al., 2018). The resistance of degraded forests was arbitrarily fixed to the 95 th quantile of resistance values in forests. Landscape planar graphs were created with the Graphab v.2.5 software (Foltête et al., 2012) from the habitat map defined from an optimal threshold applied on the HSI map (Liu et al., 2013), the resistance maps derived from Equation (1) where n is the number of habitats, a k and a l are the habitat area weighted by quality (or mean HSI) of patches k and l, and p * kl is the maximum product (or dispersal) probability between the pair of habitats k and l. The maximum product probability was calculated from all the possible paths p kl = e − d kl , where d kl is the cost-weighted distance of the path between habitat pair k and l, and is a parameter fixed according to species dispersal distance to control the decay of dispersal probability with increasing resistance distance. If patches k and l are directly connected without an intermediary patch, the maximum product probability is simply the cost-weighted distance of the leastcost path (p * kl = p kl ). We compared equivalent connectivity (EC ij ) in 2050 to the current equivalent connectivity (EC i2020 ) for species i as follows: The index Δ EC ij was qualitatively compared to species thermal maximum extracted from Jiguet et al. (2007) while the links between Δ EC ij and changes in landscape indices (habitat area, quality and mean cost distance of the least-cost paths) were depicted in (Appendix S4). (1)

| Isolating the indirect effects of climate changes on birds due to changes in tree species composition
For the bird species, AUC indices ranged from 0.75 to 0.83 for the GAM accounting for climatic conditions at the national scale. At the regional scale, AUC and CBI indices ranged from 0.61 to 0.76, and from 0.57 to 0.82, respectively. AUC and CBI indices were slightly low for P. collybita and C. brachydactyla (0.63 and 0.77 and 0.61 and 0.57, respectively). For R. regulus, the Δ EC ij showed that the indirect effects of climate changes were more often preponderant, with positive or negative effects depending on the scenario ( Figure 2). The direct effects were always slightly negative due to shrinkage in the climatic envelope in future projections, while the indirect and the full effects were either positive or negative depending on the scenario (Figures 2, 3). For P. collybita and S. europaea, the climatic envelope was fully favourable and stable over time, indicating that the effects of climate changes were entirely explained by the indirect effects due to changes in tree species composition (Figure 2, see Appendix S1, S10). For D. major and C.
brachydactyla, direct effects were preponderant and positive: the climatic envelope expanded in future projections (Figure 2). These direct effects were especially strong for C. brachydactyla as mixed forest areas decreased ( Figure 3, Table 1). Δ EC ij showed opposite direct and indirect effects of climate changes for C. brachydactyla, depending on the scenario (Figure 2). The predominance of the direct effects of climate changes were positively correlated with species thermal maximum (Figure 2).

| Projection of tree species composition in 2050
For tree species, AUC indices ranged from 0.75 to 0.88 for the GAM

| Variations in equivalent connectivity among scenarios
For P. collybita and R. Regulus, EC increased slightly under the two RCP 2.6 scenarios and decreased slightly under the two RCP 8.5 scenarios, compared to the current situation (−39% ≤ ΔEC ij ≤52%, Figure 4). For C. brachydactyla, EC increased for all the scenarios, but much more for RCP 8.5 (295% ≤ ΔEC ij ≤737%, Figure 4).
For D. major, EC also increased, but more so for the RCP 2.6 scenarios (138% ≤ ΔEC ij ≤ 159%, Figure 4). For S. europaea, ΔEC ij varied with management, contrary to the other species: ΔEC ij was higher in the "deciduous 2.6" scenario than in the "coniferous 2.6" scenario ( Figure 4). See Supporting Information for details on tree species distribution and responses to climatic conditions (Appendix S8), landscape factors (Appendix S9) and their combination (Appendix S10).

| Isolating the indirect effects of climate changes on birds due to changes in tree species composition
We provided the first study investigating the importance of climateinduced changes in forest-tree species distribution and their cascading effects on birds via landscape processes. We showed that climate changes can drive birds' amount of reachable habitat, mostly by indirect effects due to changes in tree species composition, depending on species and scenario. For example, climate changes only indirectly impact species with an intermediate thermal maximum (P. collybita and S. europaea) because the climatic envelopes for these two species were actually stable. However, habitat area, quality and spatial configuration, and matrix resistance did change after we accounted for changes in climate-induced tree species composition.
For species with a high thermal maximum (D. major and C. brachydactyla), future climatic conditions have positive direct effects. Low negative indirect effects could be offset by strongly positive direct effects, leading to an increase in the amount of reachable habitat for the full effects of climate changes (C. brachydactyla). Both direct and indirect effects can also be at play for species with a low thermal maximum (R. regulus). As a consequence, the importance of the indirect effects of climate changes must be accounted for in projection scenarios and conservation management to evaluate strategies that F I G U R E 2 (a) Indirect, direct and full effects of climate changes on the variation in equivalent connectivity calculated on landscape graphs for the five focal forest birds. For each species, the thermal maxima or the temperature at the hot edge of the European climate envelope is indicated (Jiguet et al., 2007). The indirect effects are related to climatic-induced tree species composition changes and their cascading effects on birds. The direct effects are related to bird responses to bioclimatic variables through the direct occurrence-climate relationship. The metric ΔEC ij is expressed as a percentage and compares the equivalent connectivity of a given species i and a given scenario j to the current equivalent connectivity (2050 vs. 2020). Minimum, 25 th quartile, median, 75 th quartile and maximum ΔEC ij values are represented by boxes and whiskers. (b) Corresponding bird climatic envelopes variations between 2020 and 2050 according to the scenarios RCP 2.6 and 8.5 (study area boundaries are in black and the species climatic envelope in green). The relative importance of the direct and indirect effects of climate changes (a) depends on temporal changes in the climatic envelopes (b). For example, a stable climatic envelope implied only indirect effects of climate via changes in tree species composition (Phylloscopus collybita and Sitta europaea). However, the metric EC ij not only depended on the climatic envelope but also on management scenario ("deciduous" and "coniferous"), habitat area, quality and matrix resistance, which are not depicted here. See Material and Methods and Equation 3  Our results also suggested that the relative importance of the di- modelling studies were conducted at regional scales and consistently showed the importance of land use among other drivers (Ay et al., 2014;Bonnot et al., 2017;McRae et al., 2008). On the contrary, Barbet-Massin et al. (2012), who aimed to encompass the full realized niche of bird species, concluded that a direct impact of climate changes was preponderant at the European scale. These results highlight the importance of assessing the processes governing species response to global changes at different scales (Sirami et al., 2017). At the large scale, changes in species distribution could be explained by expanding or contracting climatic envelopes at distribution range limits (Morin et al., 2007;Vissault et al., 2020). At the regional scale (within the climatic envelope), the role of changes in habitat area, quality and spatial configuration and matrix resistance must be taken into account (Riva & Nielsen, 2020). Our study demonstrates the importance of disentangling tree and forest bird responses to climate changes. We highlight a strong potential effect of changes in climate-induced tree species composition on the amount of reachable habitat for forest birds, when climatic conditions remain suitable.

| Projection of tree species composition in 2050
Forested areas with P. menziesii increased for the RCP 2.6 scenario but decreased for RCP 8.5, while forested areas with A. alba decreased in both cases. Quercus spp. and F. sylvatica responded positively to climate changes in the study area. These findings F I G U R E 3 Example of landscape planar graphs for two bird species (Regulus regulus and Certhia brachydactyla) showing the current and future habitat patches (DEC 2.6 and DEC 8.5) (in green) and the corresponding least-cost paths between pairs of habitat patches (in black). The landscape graphs of the "deciduous" and "coniferous" scenarios were close for these species; only the "deciduous" scenario is depicted here (i.e. DEC 2.6, Appendix S10 for full results are concordant with the "winner" and "loser species" reported at a larger scale by Dyderski et al. (2018). In our study, land use projections showed that the influence of forest management on tree species composition will be strongly constrained by climate change, in accordance with previous findings (Dyderski et al., 2018;Hanewinkel et al., 2013). The proportion of the different tree species varied across management scenarios when climate change intensity was moderate (RCP 2.6 scenario), but this was no longer true with more severe climate changes (RCP 8.5 scenario).
Degraded forest areas drastically increased with severe climate changes (RCP 8.5 scenario). This result indicates that the tree species currently found in most of the study area are likely to suffer from the consequences of global warming (Taccoen et al., 2019).
Here, we did not consider the introduction of allochtonous tree species for two reasons: first, we were aiming above all to assess the future of the current tree species and second, the role of allochtonous species in offsetting climate changes and improving biodiversity conservation is being largely debated (Bremer & Farley, 2010;Quine & Humphrey, 2010). Finally, our results suggest that the heterogeneity of the current tree species composition could drastically decrease in the future, because few tree species responded positively to climate changes.

| Variation in equivalent connectivity among scenarios
Here, we illustrated the potential of simulation studies to evaluate

| Study limitations
Our study does have several limitations. First, though our predictions were generally accurate, model validity was rather low in some cases, especially when relating P. menziesii to topography and soil acidity. This is not surprising given that the species was introduced to France in the middle of the 19 th century and has only been inten- largely balance the predictions of the regional-scale model.
Second, we assumed that selecting bird species associated with the main tree species in our study area would be a suitable way to assess the response of forest birds in general. Focal species are commonly used in landscape connectivity analyses but they should be carefully selected, for example to cover a range of habitat requirements and species traits (Lalechère & Bergès, 2021;Meurant et al., 2018). In our study, we considered a wide range of maximal dispersal distances (from 5.0 km for R. regulus to 28.0 km for D. major). We included the species most sensitive to climate warming (R. regulus) according to Jiguet et al. (2007) and covered a wide range of thermal preferences (72% of the range of values defined by the authors for a large set of species). Here, we acknowledged the importance of species thermal maximum, previously highlighted for bird communities (Jiguet et al., 2007(Jiguet et al., , 2010, thanks to the wide range of species thermal maximum considered. Third, climatic conditions could also be related to metabolic, physiological and behavioural adaptive responses triggered to maintain thermoregulation (Møller et al., 2010). For example, fitness is affected when individuals stop regular activities for the benefit of survival and at the expense of reproduction (Wingfield et al., 2017).
Accounting for these drivers could be a promising approach to better understand the consequences of climate change.
Fourth, we did not account for forest succession while dynamic process-based models offer the opportunity to simulate tree growth and competition and to dissociate forest structure and composition (Bonnot et al., 2017;McRae et al., 2008). However, a large amount of knowledge, dependent on the study area, must be accumulated to calibrate such models. We are unaware of any study coupling dynamic process-based models with landscape connectivity models but this could be a promising way to better account for species interactions in mixed stands.
Fifth, the equivalent connectivity index compared two static independent landscape graphs (in 2020 and 2050) to provide information about the actual and future need for conservation actions, but did not evaluate metapopulation dynamics over time. We point out that the equivalent connectivity index can be incorporated into discrete time metapopulation models to evaluate species dynamics and the speed of colonization and extinction processes over time (Lalechère et al., 2018).

| CON CLUS ION
Finally, species distribution modelling has improved our knowledge of the relationships between climate and land use changes, while integrating climate changes into landscape connectivity models is still in its infancy (Costanza & Terando, 2019;De Kort et al., 2020;Keeley et al., 2018;Littlefield et al., 2017). We believe that the innovative framework we have provided here, linking species distribution and landscape connectivity models to climate and land use changes, is a promising approach to finely evaluate the interactive effects of global changes. In this study, we assessed the importance of climate-induced tree species composition change in habitat connectivity for common forest bird species with different thermal maximum. We highlighted the risk of a decrease in landscape heterogeneity and an increase in forest degradation. Our simulations underlined that the decision by forest managers to promote deciduous trees could maximize landscape connectivity for several forest birds. Further investigations are required to refine the complex interactive effects of climate and land use changes on forest species at regional scale, in particular in less heterogeneous forest landscapes or in agricultural or urbanized landscapes.

ACK N OWLED G EM ENTS
This work was supported by the FEDER project "Trame verte forestière" (contract n°RA0017232). We are grateful to Isabelle Boulangeat, Frederic Jiguet and Emmanuel Véricel (Ligue de Protection des Oiseaux, LPO AuRA -2019) for providing us with species data.

CO N FLI C T O F I NTE R E S T
The authors declare that they have not known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are openly available in dryad at https://doi.org/10.5061/dryad.6hdr7 sr23.