The role of connectivity in the interplay between climate change and the spread of alien fish in a large Mediterranean river

Understanding how global change and connectivity will jointly modify the distribution of riverine species is crucial for conservation biology and environmental management. However, little is known about the interaction between climate change and fragmentation and how movement barriers might impede native species from adjusting their distributions versus limit the further spread of alien species. In this study, we modelled the current and future distributions of 11 native and five alien fishes in the large and heavily fragmented Ebro River, located within the Mediterranean region, which has many freshwater endemics severely threatened by global change. We considered 10 climate change models and five modelling algorithms and assessed the effects of connectivity on the accessibility of future suitable habitats. Thereby, we identify most conflict‐prone river reaches, that is, where barriers pose a particular trade‐off between isolating and negatively impacting native species versus potentially reducing the risk of alien species spread. Our results projected upstream habitat shifts for the vast majority of the species. Climate change affected species differently, with alien species generally showing larger habitat gains compared to natives. Most pronounced distributional changes (i.e. losses of native species and gains of alien species) and compositional turnover might be expected in the lower and mid reaches of large tributaries of the Ebro River. The role of anthropogenic barriers in this context is often ambiguous but rather unfavourable, as they not only restrict native fishes but also alter stream habitats and flow conditions. However, with our spatial modelling framework, we could identify specific river reaches where the connectivity trade‐off in the context of climate change is particularly relevant. Overall, our findings emphasize the importance of the complex effects that climate change, riverine connectivity and alien species are expected to impose on river communities and the urgent need to adapt management strategies accordingly.

In addition, it has become clear that invasive alien species pose a further dominant threat to biodiversity worldwide that also interacts with climate change (Bellard, Jeschke, Leroy, & Mace, 2018;Ziska & Dukes, 2014). In fact, climate change has enabled alien species to expand into regions in which they previously could not survive and reproduce and has created new opportunities for them to become invasive (Bellard et al., 2018;Stachowicz, Terwin, Whitlatch, & Osman, 2002;Walther et al., 2009).
Fresh waters, which are among the most diverse ecosystems worldwide (Dudgeon et al., 2006;Lévêque, Oberdorff, Paugy, Stiassny, & Tedesco, 2008), are especially vulnerable and threatened by climate change and species invasions (Reid et al., 2019). For example, climate change affects populations and assemblages of riverine species via changes to temperature, flow conditions, sediment regimes and water quality (Palmer et al., 2008;Pletterbauer, Melcher, Ferreira, & Schmutz, 2015;Reid et al., 2019). Ultimately and in consequence of their physiological link to local climatic and flow conditions, freshwater fish species are particularly affected by climate change. For example, 33% of the European freshwater fishes, especially those in the Mediterranean, are already recognized as susceptible to climate change (Jarić, Lennox, Kalinkat, Cvijanović, & Radinger, 2019). Fish thermal habitats, in particular those of cold water river species, are commonly projected to shift to higher altitudes Isaak & Rieman, 2013;Pörtner & Farrell, 2008); however, specific spatial realizations of habitat shifts are further complicated by interactions with other anthropogenic drivers of global change such as land use change or habitat degradation and fragmentation by barriers (Grenouillet & Comte, 2014;Herrera-R et al., 2020;Radinger et al., 2016Radinger et al., , 2017. Regarding alien species in riverine ecosystems, to date more than 8,000 successful introductions of fish species into river basins outside their native range have been recorded worldwide (Leprieur et al., 2017;Tedesco et al., 2017). These introductions contribute to changes in fish species and functional diversity (Matsuzaki, Sasaki, & Akasaka, 2013;Toussaint et al., 2018;Villéger, Grenouillet, & Brosse, 2014). While aquatic invasions might have been understudied in the broader context of climate change (Bellard et al., 2018) in the past, assessing the effects of future climate change on the distribution of alien species constitutes an active area of research (Walther et al., 2009) with a growing number of case studies also for the aquatic realm (Bellard et al., 2018).
The pace of global change may actually exceed the capacity of many fishes to track suitable habitats Radinger et al., 2017). Whether species are able to track global change depends on their species-specific responses to climatic conditions, the spatial extent of projected habitat shifts and the capabilities of species to track these shifts by means of dispersal (Radinger et al., 2017). Species dispersal is particularly limited in heavily fragmented landscapes such as river ecosystems, where anthropogenic barriers to movement (e.g. weirs, dams) might disconnect the only route between two locations and the related loss of connectivity may prevent fish from keeping pace with future environmental changes (Radinger et al., 2017;Radinger, Hölker, Hölker, Horký, Slavík, & Wolter, 2018). Barriers to movement substantially impact riverine fishes (Gido, Whitney, Perkin, & Turner, 2016), for example, by disconnecting them from spawning sites (e.g. Carvajal-Quintero et al., 2017;Fagan, 2002;Jager, Chandler, Lepla, & Winkle, 2001), or by preventing recolonization from other river reaches (e.g. Lasne, Sabatié, Jeannot, & Cucherousset, 2015;Radinger & Wolter, 2015).
An important dilemma is that while barriers may negatively impact native species of conservation concern and may hinder from adjusting their distributions in response to global change, restoring river connectivity may also enable the spread of alien species (Fausch, Rieman, Dunham, Young, & Peterson, 2009;Kirk, Rosswog, Ressel, & Wissinger, 2018;Rahel & McLaughlin, 2018). Impoundments formed by dams are also more susceptible to higher propagule pressure and dams are likely to alter the flow regime in a way that may benefit alien species (Gido, Propst, Olden, & Bestgen, 2013;Johnson, Olden, & Vander Zanden, 2008;Liew, Tan, & Yeo, 2016). This potential conflict also affects river management where the objective of restoring connectivity to achieve ecological targets might be in apparent conflict with the objective to limit the spread of alien species.
While previous studies investigated distributions of freshwater fish in response to climate change, including alien species (e.g. Britton, Cucherousset, Davies, Godard, & Copp, 2010;Buisson, Thuiller, Lek, Lim, & Grenouillet, 2008;Fletcher, Gillingham, Britton, Blanchet, & Gozlan, 2016), as well as the combined effects of fragmentation and climate change (e.g. Kano et al., 2016;Radinger et al., 2017), no study has so far explicitly compared the combined effects of dams and climate change on the distribution of native and alien species.
Consequently, there is an urgent need to better understand the interactive effects of climate change and the spread of alien species, especially in the context of riverine connectivity and the antithetical role of movement barriers. To address this question, we focus on the Ebro River, Spain, which is highly fragmented by anthropogenic barriers (Radinger, Alcaraz-Hernández, Alcaraz-Hernández, & García-Berthou, 2018), and located within the Mediterranean region, where many endemic species are particularly threatened by future climate change (Jarić et al., 2019;Maceda-Veiga, 2013). The Ebro River is characterized by many Iberian endemic fish species but also a large number of successful alien fish introductions (Almeida, Alcaraz-Hernández, Merciai, Benejam, & García-Berthou, 2017;. The main objectives of this study are: (a) to model the distributions of native and alien fishes in a large Mediterranean river for current and future climate scenarios, (b) to assess the effects of connectivity on the accessibility of future suitable habitats and compare native and alien species, and thereby (c) to identify most conflict-prone river reaches (spread risk of alien species vs. isolation of native species).
We expect species to modify their ranges in response to climate change in particular in upstream direction, but more importantly that alien species would benefit from climate change in contrast to many native species in the study area. Moreover, we expect that barriers impede species from tracking suitable habitats in particular those species that show largest climate change induced habitat shifts.

| Study area and environmental variables
The study area comprises the whole Ebro River basin, NE Spain, with a total catchment area of about 86,000 km 2 . The Ebro River flows from the Cantabrian and Pyrenean Mountains to the Mediterranean Sea and discharges on average 452 m 3 /s at its mouth (Solans & Poff, 2013). The Ebro River and its tributaries are heavily regulated by about 300 large dams and over 2,100 smaller sized weirs, with the largest densities of dams, especially in the central and upper parts (Radinger, Alcaraz-Hernández, et al., 2018). For our analysis, we used a river network obtained from the official hydrographic network (CHE, Confederación Hidrográfica del Ebro, http://iber.chebro.es/geopo rtal/) at a spatial scale of 1:50,000 that has been complemented for some smaller tributaries from a river network at the 1:25,000 scale. The river network was segmented into 250 m long river reaches (n = 50,948) that represent the spatial units for all subsequent models and analyses.
Environmental data used in the models comprised of 15 moderately to weakly correlated variables (|r| < .7; Dormann et al., 2013) related to climate, land use, hydrological alteration, network connectivity/fragmentation and general basin topography/topology (Table 1).
Collinearity among variables was reduced using hierarchical clustering based on the correlation matrix (Dormann et al., 2013;Harrell, 2015) and selecting the best univariate predictor (i.e. ecologically more meaningful) of each cluster of two or more collinear variables.
As baseline climatic characterization of the Ebro River catchment, we generated three bioclimatic variables related to air temperature and precipitation patterns for the period 1992-2015 calculated from minimum and maximum temperature and precipitation data of the E-OBS data set (version 20.0e, spatial resolution: 0.1 degrees) of the European Climate Assessment & Dataset project (ECAD; Cornes, van der Schrier, van den Besselaar, & Jones, 2018; see Supplementary Material S1 for details). Bioclimatic variables were calculated using the 'biovars' function of the R-package 'dismo' (version 1.1-4, Hijmans, Phillips, Leathwick, & Elith, 2017) and were spatially resampled using bilinear interpolation to correspond to the modelling grid with a spatial resolution of 50 × 50 m. For precipitation-related bioclimatic variables, we calculated average values for the catchment upstream of each river reach. Land use information was extracted from a raster map based on CORINE Land Cover 2000 level-3 classes (spatial resolution of 250 × 250 m; Büttner, Feranec, & Jaffrain, 2002). We calculated average shares of each of five aggregated thematic land use classes (built-up, arable, permanent crops, grassland and forest) for the catchment upstream of each river reach. To characterize hydrological flow alteration, we calculated the distance to the next upstream dam and the share of reservoir surface area in the catchment upstream of each reach. These two variables were identified in a previous study as relevant correlates of hydrologic alteration in the Ebro basin, with 'distance to the next dam' being particularly related to alterations of summer flow conditions (Radinger, Alcaraz-Hernández, et al., 2018). To describe network connectivity and fragmentation by dams, we calculated each reach's network closeness centrality using the Python igraph library TA B L E 1 Descriptive statistics of the 15 environmental variables used in the analyses (version 0.7.1, Csardi & Nepusz, 2006) and the fragment size (i.e. length of unfragmented river habitats) respectively. Closeness centrality of a reach is a measure of its centrality in the entire network (Freeman, 1978) and is an indicator of how well a reach is connected to all other reaches given the river network geometry. Network centrality has been identified as a significant predictor of α-diversity and community similarity in aquatic insect communities of river networks, and reflects dispersal pathways along the dendritic landscape structure and environmental characteristics that are linked to the network position (Altermatt, Seymour, & Martinez, 2013 As future climatic conditions, we considered a range of regional- and CLMcom (Climate Limited-area Modelling Community; see Table S1.1). For the study catchment and across the selected GCM-RCM, annual mean temperatures are projected to generally increase a median of 1.9°C; precipitation projections are spatially variable with median changes of 0.9 mm in the future scenario ( Figure S1.1).

| Fish data
The native fish community of the Ebro River basin is characterized by

| Species distribution models
We calibrated species distribution models (SDMs) of 16 fishes in the Ebro using the 'biomod2' framework (version 3.4.6; Thuiller, 2003; Thuiller, Georges, Engler, & Breiner, 2020) and five modelling algorithms that are commonly used in fish distribution models (e.g. Grenouillet & Comte, 2014;Markovic et al., 2014), including regression methods (generalized linear models, generalized additive models) and machine learning methods (boosted regression trees models, random forest, maximum entropy modelling). For all modelling algorithms, we used pseudo-absences selected from an ecologically informed background data set as observed absences might be affected by more than the lack of suitability related to our set of environmental variables. For example, observed absences might be affected by imperfect sampling, unconsidered environmental variables, dispersal and colonization processes or biotic interactions that prevent the presence at a given location (Lobo, Jiménez-Valverde, & Hortal, 2010).
Thus, we used an approach of generating pseudo-absence data from outside the environmental domain where the species is present (Jiménez-Valverde, Lobo, & Hortal, 2008). More specifically, we followed an idea proposed by Chapman, Pescott, Roy, and Tanner (2019): as background data we combined both unsuitable and accessible river reaches for each species to account for species' environmental tolerances and potential dispersal constrains respectively. For alien species, we defined the accessible background area as all river reaches within a distance of 50 river km from known presence points but not further than the next upstream dam; for native species, we considered all river reaches as potentially accessible assuming no effects of anthropogenic barriers and dispersal constrains in an evolutionary timeframe. For each species, we defined the environmentally unsuitable background domain using a simple range envelope model (using the 'sre' function from the R-package 'biomod2', Thuiller et al., 2020) based on known presence records and the 15 environmental variables. The range envelope is defined by minimum and maximum percentile values of the environmental conditions of the locations where a species is found. River reaches with an environmental variable falling outside this envelope were considered to be included in the data set of the environmentally unsuitable background domain.
Fitting SDMs for each of the 16 species comprised 10 replicate model runs of each of the five modelling algorithms. For each replicate model run, we used three pseudo-absence data sets of 5,000 randomly sampled points from the background domain of each species to reduce sampling variation. These settings were based on a preliminary analyses of the data set which revealed that further increasing the size and number the pseudo-absence data sets and replicate model runs did not improve overall model quality while being computationally cumbersome. We complemented the pseudo-absence data set during model configuration (function 'BIOMOD_FormatingData' of the R-package 'biomod2') with true absences (Table 2), that is, reaches that have been sampled but where the target fish species was not detected.
The modelling resulted in 150 SDMs (5 × 10 × 3) for each of the 16 species. For each replicate model run, the species occurrence data were randomly split into a training (70%) and testing set (30%). The models were fitted by balancing the weight of presences and pseudoabsences such that prevalence was 0.5, meaning that the presences will have the same importance as the absences in the calibration process of the models (Barbet-Massin, Jiguet, Albert, Thuiller et al., 2020). SDMs were fitted using the default settings in 'biomod2' (Thuiller et al., 2020). Withhold testing data were then used to assess the models' predictive performance for each run and pseudo-absence data set. We then calculated the mean and standard deviation (SD) of the area under the curve (AUC), and the mean sensitivity (proportion of correctly predicted presences) and specificity (proportion of correctly predicted absences) for each species over all model runs to evaluate model quality and performance.
We used the fitted species-specific SDMs to map habitat suitabilities of single river reaches across the entire Ebro River network for the baseline and the 10 future GCM-RCM projections. Therefore, the 150 species-specific models were merged to a single ensemble model for each species and modelling algorithm (i.e. averaged over 10 replicate model runs and three pseudo-absence data sets; five models per species) using the models' AUC as a proportional weighing factor (Marmion, Parviainen, Luoto, Heikkinen, & Thuiller, 2009).
Single models of poor performance (AUC < 0.70) were excluded from ensemble model building.
Finally, each species-and algorithm-specific ensemble probability map for each climate scenario (baseline + 10 GCM-RCM; 55 maps per species) was transformed into a presence/absence map using a threshold provided by 'biomod2' that minimizes the absolute difference between specificity (true negative rate) and sensitivity (true positive rate), which is similar to the sensitivity-specificity sum maximization approach (described e.g. in Liu, Berry, Dawson, & Pearson, 2005).

TA B L E 2
Sixteen modelled fish species in the Ebro River basin and corresponding model performances. Species are characterized as native (N) or alien (A) to the Ebro River basin. Model performance is provided as mean and standard deviation (SD) of the area under the curve (cross-validation AUC) and mean sensitivity and specificity across all model runs and algorithms for a species

| Analysis of species range shifts and connectivity
To evaluate species range shifts, we first calculated spatial descriptors of each species' habitat range for the modelled baseline and future GCM-RCM scenarios and each modelling algorithm describing: (a) the centre of their distribution, (b) the lower and upper range limits, and (c) the overall extent of their suitable habitats (adapted from . The centre of a species distribution along the upstream-downstream gradient was defined as the mean Strahler stream order (H Str ) across all river reaches (total n = ~51k) that were projected suitable under a given climate scenario. The lower (H lwr ) and upper (H upr ) range limits were defined as the 2.5% and 97.5% percentiles of the distance from the source (km) of all suitably projected river reaches Quinn, Gaston, & Arnold, 1996 To assess projected changes in fish species composition, we cal- Olden & Poff, 2003) and quantified as: where S imm is the number of newly recorded species in a reach, S ext is the number of species lost from a reach and S tot is the number of species across both samples (baseline and future scenario). The SER r quantifies the gross change in species composition and ranges between 0 (all species persist, identical species composition) and 1 (all species are exchanged, no species in common; Hillebrand et al., 2018). The SER r was calculated for each future scenario (GCM-RCM and modelling algorithm) and then averaged across models. To assess the relationship between temporal β-diversity and change in α-diversity (i.e. species richness), we calculated Pearson correlations between SER r and the absolute values of projected (native and alien) species net gain across all river reaches.
To evaluate potential effects of barriers, we determined all river reaches that are considered suitable for a given species but are con-

| RE SULTS
The distributions of 16 fish species in the Ebro River network were successfully modelled with a mean (across replicate model runs and algorithms) cross-validated AUC across species of 0.85 (Table 2). Of all considered species, G. holbrooki, S. glanis and A. alburnus yielded the best performing models. The SDMs projected future changes in habitat range descriptors across the modelled species (Figure 1; Supporting Information S2).
The centre of species distributions described by mean Strahler stream order H Str ranged between 1.7 (B. haasi) and 5.2 (S. glanis) with an overall median across all species of 2.9 (CI = 2.7-3.2) for the baseline scenario ( This shift in mean stream order corresponds to an upstream shift of 16% of a species distributional area by one stream order. Over all modelling algorithms and GCM-RCM, alien species exhibited larger upstream shifts of their distribution centres compared to native species (Table 3). Differences in ΔH Str between alien and native species were statistically clear (AGIT; Z = −12.51, p < .001). When restricted connectivity caused by dams was considered, the share of suitable but concurrently inaccessible habitat, %H inaccess , was projected to increase from 4.4% (median across all species and models; CI = 2.1-6.1%) for the baseline scenario to 8.6% (CI = 7.6-9.3%) for the future scenarios (Figure 2; Tables S2.1 and S2.2). Median Δ%H inaccess across species, modelling algorithms and GCM-RCM was 2.7% (CI = 2.2-3.1%). Differences in Δ%H inaccess between alien and native species were statistically clear (AGIT; Z = 6.40, p < .001; Table 3). Largest shares of suitable but F I G U R E 1 Changes in the upper and lower range limits between the baseline and future climate change scenario for native (blue) and alien (red) fish species. Representation of range shift metrics is based on    (Table S3.2), and thus associated with observed differences between alien and native species while ΔH lwr contributed particularly to the second dimension. Across the entire Ebro basin, mean SER r of fish communities between the baseline and the projected future scenario was 0.38 (Figure 4; Figure S4.1). High species turnover indicated by an SER r > 0.5 was detected for 20% of the river reaches, particularly in the down-and midstream reaches of larger Ebro tributaries.
Under consideration of restricted connectivity, mainly lower order headwater reaches of many Ebro tributaries (stream order 1-3) were identified as suitable but inaccessible due to dams, especially for native species (Figure 5). By contrast, for alien species, the mainly central reaches (stream order 3-5) of the Rivers Segre, Jalón, Guadalope and Matarraña as well as the upper Ebro main stem were identified as potentially suitable in future but where dams might restrict their spread from downstream reaches ( Figure 5).

| Climate-related habitat shifts of native and alien fishes
This study is aimed at improving our understanding of the joint effects of climate change and restricted river connectivity on the distributions of native and alien fish species in a large Mediterranean river. In general, the extent of projected climate change-induced habitat shifts in the Ebro River was highly variable among species, with approximately 70% of the modelled species projected to gain habitats and 30% to rather loose climate suitable habitats. Similar complex patterns of habitat shifts related to climate change have been reported in previous studies of freshwater fish (e.g. Grenouillet & Comte, 2014;Markovic, Freyhof, & Wolter, 2012;Radinger et al., 2017) and other species groups (e.g. Domisch, Jähnig, & Haase, 2011;Fulton, 2011;Tayleur et al., 2016). For example, 47% of fishes in the European River Elbe (Radinger et al., 2017) and 63% of fishes in French rivers (Grenouillet & Comte, 2014) were modelled to undergo or have already exhibited range expansions, respectively, while the other species showed rather range losses. We note that such differences in climate-induced habitat shifts might result, for example, from differences in climate change velocities over time and for different geographic areas (e.g. Loarie et al., 2009) or differences in speciesspecific ecological requirements (e.g. Graham & Harrod, 2009).
Previous studies suggested that in particular the interaction of climate change and other anthropogenic drivers contribute to the complex spatial patterns of species responses towards ongoing global change (Heino, Virkkala, & Toivonen, 2009;Olden et al., 2010;Radinger et al., 2016). As hypothesized, species' suitable habitats shifted in upstream direction. This is in good agreement with many previous studies that projected or observed shifts of fish thermal habitats towards higher altitudes and in upstream direction in response to warming climates Isaak & Rieman, 2013;Pörtner & Farrell, 2008), although such studies are often limited Large upstream shifts were projected for the group of alien species (e.g. G. holbrooki, S. glanis and C. carpio), which showed distinct patterns of climate change responses compared to native species (Figures 1 and 3). This suggests that alien species in the Ebro River, which are currently dominating the main stem , might further spread into the upstream tributaries in response to changing climates. In general, our results indicated spatially variable patterns of species gains and losses with gains often dominating over species losses-a pattern that has previously been detected in other river systems (e.g. Buisson & Grenouillet, 2009;Radinger et al., 2016). As a consequence of the projected habitat shifts, species assemblages are predicted to change with most pronounced species turnover (i.e. temporal βdiversity) in the down-and midstream reaches of larger Ebro tributaries where the current community comprising several native species is expected to convert to an increasingly alien-dominated community. Surprisingly, species gains and losses as measured by net gain were only weakly associated with SER r . A lack of coherence between the temporal change in assemblage α-diversity (e.g. number of species) and temporal β-diversity (i.e. change in species composition) has been noted in previous studies (e.g. Magurran et al., 2018). Thus, even at river reaches with little evidence for pronounced changes in species richness, the compositional effects might be much more prevalent, hence emphasizing the importance of adequate measures of temporal species turnover to capture the consequences of climate change induced species shifts. Such changes might be even more pronounced at the level of local abundances of single species (Shimadzu, Dornelas, & Magurran, 2015) and might express in changes of the interaction among and within species within these new communities (Montoya & Raffaelli, 2010;Walther, 2010 Climate-related range shifts of alien species are increasingly studied, with partly variable outcomes (reviewed by Bellard et al., 2018;Smith et al., 2012). The studies focusing on alien fish suggest that their establishment or spread would be enhanced by increasing temperatures, particularly those of the relatively warm water alien species, such as C. carpio, S. glanis or Micropterus spp. (e.g. Bae, Murphy, & García-Berthou, 2018;Britton et al., 2010;Sharma & Jackson, 2008). However, we note that alien species responses to climate change are likely to be complex, potentially involving other interrelated human-mediated environmental changes such as habitat impairment or hydrological alteration (Britton et al., 2010).

| Effects of restricted connectivity
Our results re-emphasize that restricted longitudinal connectivity caused by dams is impacting fishes' abilities to keep pace with future climate-induced habitat shifts as indicated in previous studies (e.g. Radinger, Hölker, et al., 2018). The share of suitable habitats in the Ebro River that is inaccessible due to dams is likely to increase under future climate scenarios and varies among species ( Figure 2). However, differences between alien and native species were rather small. Substantial impacts of barriers on river fishes' abilities to track future habitat shifts have been reported previously with larger bodied species particularly being impacted by physical barriers while smaller species being rather restricted by their dispersal ability (Radinger et al., 2017;Radinger, Hölker, et al., 2018). Not surprisingly, species with projected large shifts in our study were exceptionally affected by dams. For example, for species such as S.
fluviatilis or S. laietanus, more than 30% of their projected suitable habitats in future are inaccessible, that is, located upstream of dams relative to their current occurrence.
Our results illustrate the dilemma of restoring longitudinal connectivity and the antithetical role of movement barriers in areas prone to alien species: Dams in the Ebro River clearly impact native species by fragmenting and isolating their habitats and preventing them from adjusting their distributions in response to global change.
However, enhanced river connectivity may also foster the further spread of alien species. This dilemma has been repeatedly warned in previous work (Fausch et al., 2009;Kirk et al., 2018;Rahel & McLaughlin, 2018). With the spatially explicit modelling framework, we could for the first time reveal the spatially variable impacts of dams on native versus alien species that can help to prioritize river connectivity management (see Figure 5). Thereby, we identified river reaches where restricted connectivity mainly negatively affects native species versus river reaches that are likely conflict-prone in regard to the spread of alien species given full river connectivity.
In this context, we also note recent efforts in assessing the selective fragmentation of aquatic systems, such as approaches to allow desirable species pass upstream while preventing or greatly reducing passage by undesirable, for example, alien species (Rahel & McLaughlin, 2018). However, completely selective fish passage systems are rather unlikely or methodologically or economically currently barely feasible (Rahel & McLaughlin, 2018) and their applicability in systems with larger numbers of alien species, such as the Ebro River, remains unclear.
The potential impediment of the active spread of alien species by barriers should not hide the manifold detrimental effects of dams on fish (Fuller, Doyle, & Strayer, 2015;Olden, 2016), of which many even promote the spread of alien species. In fact, the establishment of alien fishes is often facilitated by the changed flow and habitat conditions that result from damming rivers (Gido et al., 2013;Johnson et al., 2008;Liew et al., 2016;. Consequently, given the connectivity between reservoirs and other streams of the same catchment, reservoirs might act as 'stepping-stone' habitats and thereby contribute to an increased likelihood of the continued spread of alien species (Johnson et al., 2008;Liew et al., 2016). In this vein, dam reoperation to mimic the natural flow regime (instead of dam removal) has gained increasing interest among practitioners (Poff, 2018).
In this study, we considered only along-channel movement of fish and how dams restrict species spread. However, previous studies clearly showed the potential relevance of human-mediated dispersal and multiple introductions in allowing alien fish to cross barriers and reach future suitable but inaccessible regions (e.g. Bullock et al., 2018;Johansson et al., 2018). In fact, impoundments created by dams are considered important introduction points for several alien species. Human-mediated dispersal of alien fish across barriers such as deliberate releases of (ornamental) fish (e.g. goldfish) from aquaria or bait buckets (e.g. bleak) not only counteract measures to prevent the further invasion of alien species but might also blur the spatial patterns of restricted species spread found in this study.
Here, effective fish monitoring programmes (Radinger, Britton, et al., 2019) to early detect new or secondary introductions of alien species in reservoirs and adjacent river reaches are essential.
There are some limitations that may influence our modelling results. In this study, we only considered the larger dams in the Ebro River catchment (approx. 300) that constitute impassable barriers for the upstream movement of fish. However, we acknowledge that in addition, more than 2,100 smaller sized weirs (Radinger, Alcaraz-Hernández, et al., 2018) fragment the river network and of which many potentially restrict fish movement and alter habitat and flow conditions of the up-and downstream reaches. Thus, our results might be considered rather conservative estimates in regard to habitat fragmentation and the accessibility of suitable habitats.
However, here we neither modelled species abundances nor did we consider temporal dynamics and biotic interactions between native and alien species as drivers of distribution patterns. A previous study of the same area, however, indicated that biotic interactions were of less importance compared to environmental factors in shaping joint species distributions .
We acknowledge that the distributions of spreading alien species are by definition in a non-equilibrium state and are structured by both the species' environmental tolerances and dispersal constraints (Chapman et al., 2019). This might limit SDMs due to their assumption of equilibrium that is likely to be violated during the invasion process (Gallien, Douzet, Pratte, Zimmermann, & Thuiller, 2012). In our models, we address this concern following an approach proposed by Chapman et al. (2019) that uses prior knowledge of species' requirements to define the unsuitable background for modelling, while also considering accessibility.
Nevertheless, we explicitly note that (a) our SDMs were calibrated using a spatially restricted data set that does potentially not capture the full range of environmental conditions found across species (native) ranges, that is, distributional information outside the good spatial coverage of sampling sites, we cannot fully exclude any bias in our models resulting from the fish sampling data that were originally collected for monitoring programmes. This all could have influenced the estimation of species response curves and our modelling results on species-specific shifts of their ranges and upper and lower limits (Thuiller, Brotons, Araújo, & Lavorel, 2004).
Thus, we encourage future studies to investigate the rarely considered effects of climate change on the transport, introduction success and spread of alien species (Bellard et al., 2018), how climate change shapes biotic interactions and how this in turn will affect native species communities.
In summary, our results predict variable responses to climate change among fishes in the Ebro River, with the majority of species gaining habitats and shifting their range in upstream direction. A distinct upstream spread was especially projected for several alien fish species and most pronounced distributional changes (i.e. losses of native species and gains of alien species) and compositional changes might be expected in the lower and mid reaches of larger tributaries.
The role of anthropogenic barriers in this context is highly uncertain but rather unfavourable, as they not only restrict native fishes from tracking shifting habitats but also alter stream habitats and flow conditions. Consequently, effective conservation of fishes in the Ebro River should focus on the restoration of habitats and the natural flow regime, improvements of connectivity for native species and the control of alien species, particularly the prevention of further introductions . To this end, we could identify several river reaches where the connectivity trade-off in the context of climate change is particularly relevant. These first important insights should be further complemented by mechanistic modelling approaches (Tonkin et al., 2019) and accompanied by continued monitoring efforts (Radinger, Britton, et al., 2019) to further improve our understanding and forecasting of the often interrelated effects that climate and ecosystem change, riverine connectivity and alien species are expected to impose on river fish communities.

DATA AVA I L A B I L I T Y S TAT E M E N T
Environmental and fish data that support the findings of this Emili García-Berthou https://orcid.org/0000-0001-8412-741X