Environmental filtering governs the spatial distribution of alien fishes in a large, human‐impacted Mediterranean river

To analyse the occurrence and abundance of native versus alien fish species in relation to climate, land use, hydrologic alteration and habitat fragmentation in a heavily invaded and human‐impacted riverine ecosystem. To test whether co‐occurrence patterns of native versus alien species are structured by environmental filtering or biotic associations.


| INTRODUC TI ON
Understanding species distribution patterns and causes of species co-occurrence is a major focus in ecology. A variety of factors, most prominently including abiotic environmental conditions and biotic interactions, are considered to determine where species occur at larger spatial scales, and why (Hutchinson, 1957;Kissling et al., 2012;Pollock et al., 2014). This topic is also a fundamental question in invasion ecology, in order to understand the proliferation of alien species and the often simultaneous decline of many native species and to implement suitable management measures. Whether introduced species colonize a certain habitat depends on the number of individuals introduced (i.e., propagule pressure, Lockwood, Cassey, & Blackburn, 2005, is driven by environmental filters, such as climate, that determine if propagules can survive (Theoharides & Dukes, 2007), and might be further limited by the dispersal abilities of species (e.g., Radinger et al., 2017;Stoll, Sundermann, Lorenz, Kail, & Haase, 2013). The subsequent establishment of a species might also be mediated by biotic interactions, such as competition or predation that affect population growth (Theoharides & Dukes, 2007). Particularly in a changing world, it is of great interest whether alien species are the direct drivers of the loss of native species through biotic interactions, or the passengers of decline caused by environmental change (e.g., habitat degradation) (Didham, Tylianakis, Gemmell, Rand, & Ewers, 2007;Didham, Tylianakis, Hutchison, Ewers, & Gemmell, 2005;Hermoso, Clavero, Blanco-Garrido, & Prenda, 2011;MacDougall & Turkington, 2005).
For example, Hermoso et al. (2011) and Light and Marchetti (2007) identified alien species as the primary driver of the decline of natives.
However, many other examples revealed environmental change as a leading cause of native species loss irrespective of the presence of alien species (e.g., Kominoski et al., 2018;MacDougall & Turkington, 2005). Both aspects are not mutually exclusive and there is increasing evidence that environmental degradation and invasive species interact and jointly govern biodiversity loss and the decline of many native species (Didham et al., 2007). For example, environmental change causing a loss of refugia habitats might enhance the predation risk of invasive on native species (Didham et al., 2007;Hermoso et al., 2011).
Identifying the leading causes of native species decline and the underlying processes related to the driver versus passenger hypotheses has major implications for environmental management. For example, invasive species eradication might be inefficient and unrewarded if native species loss is merely a result of environmental degradation and alien species have no effect on the decline of native species (Zavaleta, Hobbs, & Mooney, 2001). The complex interactions of abiotic and biotic determinants of species co-occurrence have also been reflected in the recent development and increasing application of so-called joint species distribution models (JSDM), which aim to discern whether larger-scale patterns of species co-occurrences can be attributed to similar environmental responses or biotic associations (D'Amen, Mod, Gotelli, & Guisan, 2018;Kissling et al., 2012;Pollock et al., 2014).
The specific role of anthropogenic barriers such as dams and weirs in this context is less known but generally considered equivocal. In general, barriers negatively impact many native fishes, especially those that routinely migrate to the sea or within the river network or those that are intolerant to habitat degradation and alterations of the natural flow regime (Aparicio et al., 2000;Benejam et al., 2016).
The specific environmental and biotic factors that shape the decline of native species and invasions by alien fishes of Mediterranean streams and rivers are unclear. Therefore, the main objectives of this study are (a) to analyse how the occurrence and abundance of alien fish species in the Ebro River are related to environmental variables describing climate, land use, hydrologic alteration, habitat fragmentation and river topography; and (b) to test whether co-occurrence patterns of native versus alien species are structured by environmental filtering or biotic associations using JSDM, which have been barely used for fish before. We hypothesized that: (a) as most introduced fishes in the Ebro River are warmwater species , the colonization of river habitats by alien species would be limited by temperature-related climate variables; (b) larger-scale variables (e.g., climate, topology) would determine the general occurrence patterns and range limits of alien species, whereas more localscale, habitat-related variables (e.g., hydrologic alteration) might determine their species richness and abundance; and (c) a combination of abiotic and biotic factors might structure both native and alien species occurrences.

| Study area and environmental variables
The study was conducted in the Ebro River basin, NE Spain, a large European river basin with a total catchment area of about 86,000 km 2 (Supporting Information Figure S1.1). The Ebro River flows from the Cantabrian and Pyrenean ranges to the Mediterranean Sea and discharges on average 452 m 3 /s at its mouth (Radinger, Alcaraz-Hernández, & García-Berthou, 2018). Climatic conditions and associated patterns of temperature and precipitation within the basin are very diverse with mean annual temperatures ranging from 0.8 to 16.2 °C and mean annual precipitation from >1,500 mm in the Pyrenees (with elevations >3,000 m a.s.l.) to <400 mm in the semiarid interior (see e.g., Radinger, Alcaraz-Hernández, et al., 2018). The river network used in subsequent analyses was obtained from the official hydrographic network (CHE, Confederación Hidrográfica del Ebro, http://iber.chebro.es/geoportal/) at a spatial scale of 1:50,000 and was partly complemented for some smaller tributaries from a river network at the 1:25,000 scale.
In the following, we refer to the main stem and lower main stem as reaches with catchment size >10 3 and >10 4 km 2 , respectively. Calculations of catchment characteristics were based on a European digital elevation model (EU-DEM version 1.1, https:// land.copernicus.eu/pan-european/satellite-derived-products/ eu-dem/eu-dem-v1.1/) at a spatial resolution of 50 × 50 m. As environmental predictors for all further analyses, we used 24 uncorrelated variables (|r| < 0.7; Dormann et al., 2013) related to topography, climate, land use, hydrologic alteration and network connectivity/fragmentation (Supporting Information Table   S1.1). Details on the calculation and selection of specific explanatory environmental variables are provided in the Supporting Information Appendix S1.   (Zajicek & Wolter, 2018).

| Fish data compilation
Due to the high efficiency of electrofishing and spatial coverage and intensity of the fish sampling, we assume that species detection probability is high. From the 614 sampling sites, some sites were dry or not accessible due to high flows (n = 29) or had no fish captures (n = 63) and thus were excluded, leaving 522 samplings for further analyses (Supporting Information Table S2.2). For 394 samples, species abundance and the survey extent was recorded, while for the other samples only information on species presence was available.
For further analyses, fish abundance data were standardized by the survey extent to catch per unit effort (CPUE, i.e., fish per 100 m fished length). For models based on occurrence-only information, the fish abundance data were transformed to presence/absence.

| Modelling framework
We modelled the richness of native and alien species and its association with environmental variables and the co-occurrence of native versus alien species using a combination of species distribution models: (a) a hurdle model-like approach involving boosted regression tree (BRT) models (Elith, Leathwick, & Hastie, 2008) to analyse which environmental variables are predominantly associated with the occurrence of alien species in the fish community; and (b) a joint species distribution model (JSDM; Pollock et al., 2014) to quantify the co-occurrence of (native and alien) fish species that can be attributed to shared environmental responses or to other ecological mechanisms such as biotic interactions.

| Modelling occurrence and relative abundance of alien fishes
We calculated BRT models using the package "dismo" (version 1.1-4, Hijmans, Phillips, Leathwick, & Elith, 2017) within the software R (version 3.4.3, R Core Team, 2017). BRT is a machine learning method that aggregates many simple single regression trees to a collective model of improved predictive performance (Elith et al., 2008). BRT models are considered to effectively select relevant variables, identify variable interactions and avoid overfitting (Elith et al., 2008;Hastie, Tibshirani, & Friedman, 2009) and thus they are increasingly applied in ecology (Radinger et al., 2016;Radinger, Wolter, & Kail, 2015). Here, we applied a hurdle model-like (Zeileis, Kleiber, & Jackman, 2008) approach, that is a two-part model framework that specifies one model for the general occurrence of alien species and another for species richness. Specifically, we first used the full set of sampling sites (n = 522) to analyse which environmental variables best discriminate between sites inhabited by at least one alien species versus sites without alien species (i.e., occurrence of any alien species, O A ). Secondly, we focused on sites with at least one alien species (n = 187) and modelled the relationship between the environmental variables and the richness of native (S N ) and alien species (S A ) and the relative share of alien species (%S A ) and alien individuals within a sample (%I A ).
For modelling O A , we applied a BRT model with a Bernoulli loss function (binomial model predicting presence vs. absence of any alien species); for S N and S A , we used a Poisson loss function; and for %S A and %I A , we considered a Laplace loss function that minimizes absolute errors. For each response variable, we fitted a BRT model including an automatized stepwise variable selection (Radinger, Alcaraz-Hernández, et al., 2018;Radinger et al., 2016) and a 10-fold cross-validation. Cross-validation works by randomly splitting rows of the full dataset into K = 10 equally sized folds, repeatedly training the model on K − 1 folds and testing on the remaining fold. Tree complexity was set to 5 and learning rate to 0.0025 for all models to achieve the recommended number of more than 1,000 trees (Elith et al., 2008). To assess the model quality of the binomial O A model, we calculated the mean and standard error (SE) of the AUC (area under the receiver operating characteristic curve) over all 10 folds.
For the models of S N , S A , %S A and %I A , we calculated the mean and SE of the correlation coefficient r between fitted and raw values. The relative importance (%, VI) of each predictor variable in the final BRT model was quantified based on the number of times each variable was used for splitting, weighted by the squared improvement at each split and averaged over all trees (Elith et al., 2008;Radinger, Alcaraz-Hernández, et al., 2018).

| Modelling the co-occurrence of native versus alien species
We applied a joint species distribution model (JSDM) to investigate our dataset for patterns of species co-occurrence with a specific focus on native versus alien species. JSDM quantify the leftover residual correlation (Rho) among multiple species after accounting for the correlation that can be attributed to shared environmental responses (EnvRho; Pollock et al., 2014;Warton et al., 2015). The basic idea of JDSM is that species that co-occur more frequently than expected given the environment (positive Rho) might either indicate some sort of dependence (e.g., facilitation) or may simply point to missing relevant environmental variables, while negative values of Rho may indicate some sort of negative species association (e.g., competitive exclusion; Pollock et al., 2014). Specifically, we fitted a JSDM using the code provided by Pollock et al. (2014), which is a hierarchical multivariate probit regression model that relates the environmental variables (Supporting Information Table   S1.1) to a binary response variable (presence/absence) involving latent variables (Pollock et al., 2014). Further details on the JSDM are described in Pollock et al. (2014). The JSDM was fitted for the subset of the species with a minimum prevalence of 0.05 (≥27 presence records in the 522 sampling sites; Supporting Information Table S2.2) using the Markov chain Monte Carlo (MCMC) algorithm from the software JAGS (version 4.3.0) via the R package "R2jags" (version 0.5-7, Su & Yajima, 2015). In total, we calculated five chains of 200,000 iterations with a burn-in of the first 50,000 iterations and a thinning rate of 20 to reach model convergence, retaining 7,500 samples for further analyses. Model convergence was informally assessed for all relevant parameters using diagnostic trace plots and the Gelman-Rubin convergence diagnostic (Gelman & Rubin, 1992). To avoid potential overfitting, we calculated a preliminary JSDM and excluded all environmental variables that were insignificant with a 95%-highest posterior density interval (HPDI) of the Beta coefficients including zero for all species in the final JSDM run. The JSDM yields two matrices, with the first providing pairwise environmental correlations, EnvRho, and the second providing pairwise residual correlation among species occurrences, Rho, after accounting for their shared environmental response (Bar-Massada & Belmaker, 2017). We calculated the median and the 95% HPDI of EnvRho and Rho from the posterior distributions of the environmental and residual correlation matrix. Correlations were considered significant when the HPDI did not include zero (Zurell, Pollock, & Thuiller, 2018). We applied permutation tests (asymptotic general independence test; R package "coin," version 1.2-2, Hothorn, Hornik, Wiel, & Zeileis, 2008) and pairwise permutation post hoc tests with a p-value adjustment that controls the false discovery rate (FDR, Hochberg & Benjamini, 1995; R package "rcompanion," version 1.11.1, Mangiafico, 2017) to test for differences in the distribution of the environmental and residual correlations among pairs of native and alien species.

| Occurrence and relative abundance of alien species
Of the 522 fish community samples available, 187 had one or more alien species (Figure 1). The most frequent alien species in the Ebro basin were bleak (n = 113 presence records), common carp (n = 109) and Wels catfish (n = 42). By contrast, the three most frequent native species were Ebro barbel (n = 299), Ebro nase (n = 247) and Pyrenean gudgeon (n = 227), which are all endemic to the Iberian Peninsula.
Along the upstream-downstream gradient native species richness was hump-shaped, with only a few native species inhabiting the headwaters, higher richness in the larger tributaries and upper main stem with catchment sizes up to about 10 4 km 2 and followed by a strong decline in the lower main stem (Figure 2 and Supporting Information Figure S3.4). In the most downstream reaches (downstream of the Flix dam, Supporting Information Figure S1.1), native species richness was increasing again. By contrast, alien species were only present in the lower main stem parts of the Ebro River (catchment >10,000 km 2 ) and increased in downstream direction ( Figures 1 and 2). The number and share of alien species per site was on average 0.78 species (range = 0-9, IQR = 1-0) and 16% (range = 0%-100%, IQR = 25%-0%), respectively. The number of native species was weakly positively correlated with the number of alien species (Spearman correlation, r S = 0.12, p < 0.01, Figure 2).
The occurrence of alien fishes, O A , in the Ebro River could be modelled with high accuracy, as indicated by a cross-validation AUC of 0.90 (SE = 0.01, see Supporting Information Table S4.3). Ten out of 24 variables were selected as important in the final O A model, that is discriminating between sites inhabited by at least one alien species versus sites without alien species while other predictor variables were less important and not selected (Table 1). Specifically, the size of the upstream catchment (44% VI) was the by far most influential variable (Figure 3), followed by the minimum temperature of the coldest month (9% VI) and arable land use (7% VI). Sites inhabited by alien species were generally located more downstream in the Ebro River main stem, had upstream catchments >10 3 km 2 (Figure 2 Table S4.3). The final models of S N and S A were less complex and selected 11 and 7 environmental variables, respectively, compared to the models of %S A and %I A , which selected ≥19 variables (Table 1).
The three most important variables related to the number of native species, S N , were network closeness centrality (22.3% VI), forest land use (19.1% VI) and upstream basin size (9.8% VI). Specifically, sites with more native species were generally less centrally located within the network, more upstream and had higher shares of forest (>30%) in their catchment. By contrast, the richness of alien species, S A , was predominantly related to annual mean temperature (41.3% VI) followed by upstream basin size (15.5% VI) and minimum temperature of the coldest month (12.5% VI). Temperature constituted a limiting factor for alien species, that is sites with higher S A were located in regions with a mean annual temperature >15.5°C and a minimum temperature of the coldest month >−0.5°C (Supporting Information Figure S4.5). Moreover, S A steadily increased with catchment size and was relatively lower in 15-25 km but higher in 25-60 km distance downstream a dam.
The three most important variables for %S A were upstream basin size (10.7% VI), distance from the next upstream dam (8.6% VI), F I G U R E 1 Total species richness and relative abundance of alien species (%S A ) and its spatial distribution in the Ebro River catchment. The size of circles is proportional to the observed total species richness; the colour-gradient green to red indicates an increasing share of alien species (%S A ) and minimum temperature of the coldest month (7.9% VI). The %S A steadily increased with catchment size and was elevated in 25-60 km downstream of dams (Supporting Information Figure S4.5). The three most important variables for %I A were distance from the next upstream dam (10.2% VI), sinuosity (7.9% VI), and mean annual temperature (7.6% VI). The %I A increased with distance from dams and higher %I A was found in river reaches of low sinuosity and with mean annual temperatures >14.5°C (Supporting Information Figure S4.5).

| Co-occurrence of native and alien species
The JSDM revealed that co-occurrence patterns of the 17 most fre-

| D ISCUSS I ON
Our initial hypothesis that the distributional range of alien, mainly warmer-water fish species in the study area is largely co-determined by temperature-related climate variables was fully supported by our results. We observed contrasting patterns of native versus alien F I G U R E 2 Relationship between number of native (S N ) and alien species (S A ) per site and catchment size. Solid line illustrates a smoothed conditional mean based on a fitted generalized additive model (GAM); corresponding 95% confidence interval in grey. The relationship S N -S A is presented as a bubble plot (circles are proportional to the number of sampling sites) substrates and temperature (Allan & Castillo, 2007). Besides catchment size also the minimum temperature during the cold winter period and the mean annual temperature themselves appeared as decisive environmental predictors of the occurrence and richness of alien fish species in our study. As most fish are ectothermic organisms, this suggests physiological mechanisms related to lower temperature limits (e.g., for spawning or survival) which might restrict these alien species (Portner & Farrell, 2008). For example, suitable temperatures during the spawning period for the alien cyprinid species carp and bleak are considered 16-22 and 17-28°C, respectively (Mann, 1996). For the nonnative Eastern mosquitofish (Gambusia holbrooki), Pen and Potter (1991) Alcaraz-Hernández, et al., 2018) is favouring alien species at the expense of locally adapted native species (Aparicio et al., 2000;Bunn & Arthington, 2002;Hermoso et al., 2011;Poff et al., 2007). In addition, dams can also favour alien TA B L E 1 Relative importance of environmental predictor variables (% variable importance, VI) selected in the boosted regression tree models of S N (richness of native species), S A (richness of alien species), %S A (share of alien species) and %I A (share of alien individuals)  (11) Built-up land use 4.9 (9) 3.95 (10) 8.09 (5) 6.02 (4) 5.19 (6) Forest land use 19.14 (2) 3.18 (15) 4.41 (14) Sinuousity 4.94 (9) 5.73 (7) 5.84 (5) 7.87 (2) Arable land use 7.47 (3) 6.13 (8) 4.33 (10) 5.04 (9) Mean temperature of driest quarter 3.81 (11) 7.38 (6) 5.28 (8) 3.49 (17) Temperature seasonality 8.97 (4) 3.42 (14) 5.07 (7) Number of barriers upstream 7.67 (5) 3.53 (13) 4.86 (10) Grassland land use 5.08 (8) 5.6 (6) 4.32 (15) Annual precipitation 5.51 (6) (22) Impounded river length within 10 km upstream 0.46 (23)

Number of reservoirs within 10 km upstream
Note. Variables are displayed in descending order based on the sum of VI across the models. The importance rank of each variable is provided in parenthesis. Variables without information were not selected in the final model. species through the created impoundments (Johnson et al., 2008;Liew et al., 2016). Hydrologic impacts of dams on native versus alien fish largely depend on the complex interaction of species-specific habitat requirements with the magnitude and type of flow regulation (Gido, Propst, Olden, & Bestgen, 2013). Interestingly, while Batalla, Gómez, and Kondolf (2004) and Radinger, Alcaraz-Hernández, et al. (2018) revealed that hydrologic alteration in the Ebro basin is most pronounced in downstream proximity to dams, our results revealed highest numbers of alien species in 25-60 km distance downstream of dams. The ultimate reason for this downstream shifted effect remains unclear, but might be related to the so-called "rhithralization" effect (Hohensinner, Hauer, & Muhar, 2018). By that, reaches immediately downstream of dams may show relatively elevated flow velocities and larger grain sizes of the substrate, which might benefit the native rheophylic, lithophylic species (e.g., Ebro barbel) rather than the limnophilic alien species (e.g., largemouth black bass  Table S2.2 for species codes) in the Ebro River catchment. Colours of the correlation heatmaps indicate negative (red) to positive (blue) correlations; asterisks indicate significant correlations (95% HPDI did not include zero). Only significant correlations are displayed in the network diagrams; bold lines indicate strong correlations |r| > 0.7. Alien species displayed in bold font in the network diagram.  (Schmutz & Moog, 2018). Cumulative effects of dams generally cause a deficit of coarse sediments, increased rates of sedimentation of smaller grain sizes and consequently decreased water turbidity (Schmutz & Moog, 2018), which benefits most alien and disfavours the native species in the Ebro River.
Our results point to catchment land use as another important correlate for fish in the Ebro River with native species richness being positively associated with forest land use and negatively with agricultural land use. This supports previous findings showing that sensitive (often lithophylic) fish are positively associated with forest land use and react negatively with increasing shares of agriculture and urban land use (Almeida et al., 2017;Trautwein, Schinegger, & Schmutz, 2012). Land use acts through numerous mechanisms, with fish most likely reacting to both direct (e.g., nutrients, toxins) and indirect (e.g., sedimentation, hydrologic alteration) effects (Hering et al., 2006;Strayer et al., 2003;Trautwein et al., 2012). Furthermore, we note that arable and urban land use might also be considered indicative of human presence, which may in turn elevate propagule pressure of introduced alien species, especially for species that are intentionally stocked. Other studies also suggest complex interactions of climate and land use jointly affecting species distributions (Oliver & Morecroft, 2014;Radinger et al., 2016).
Given the characteristic dendritic structure of river networks (Fagan, 2002;Grant, Lowe, & Fagan, 2007), the upstream-downstream gradient of species richness observed in this study is also closely linked to the connectivity of sites within the network.
Specifically, the main stem is more centrally located and better connected than peripheral headwater streams, which is commonly associated with greater local species richness at downstream sites and lower richness at headwater sites (Altermatt, Seymour, & Martinez, 2013;Muneepeerakul et al., 2008;Seymour, Fronhofer, & Altermatt, 2015 (Aparicio, Pintor, Durán, & Carmona-Catot, 2012;Clavero & Hermoso, 2015;Elvira, Almodóvar, & Lobón-Cerviá, 1991;López, Gázquez, Olmo-Vidal, Aprahamian, & Gisbert, 2007). Therefore, dams located in critical sections of the river network (e.g., the topologically highly connected main stem) can cause severe upstream effects especially on migratory species (Brevé, Buijse, Kroes, Wanningen, & Vriese, 2014;Segurado, Branco, & Ferreira, 2013). Fragmentation by barriers (dams and weirs), as investigated in our analysis, played only a subordinate role to explain fish richness and abundance (see also Branco, Segurado, Santos, Pinheiro, & Ferreira, 2012). However, we used fragmentation indices (e.g., numbers of barriers in proximity to a site and fragment size), whereas explicitly modelling fish dispersal (Radinger et al., 2017;Radinger, Hölker, et al., 2018;Radinger, Kail, & Wolter, 2014; or analysing time-series of species distribution patterns might be more appropriate in explaining if and to what degree barriers impede dispersal. Although isolation by dams may prove beneficial for native biota by blocking the entry of invasive species (Fausch et al., 2009;Liermann, Nilsson, Robertson, & Ng, 2012;McLaughlin et al., 2013), the distribution of many alien fish in the Ebro River is the result of potentially multiple introduc- in another Mediterranean river basin (Guadiana River, Spain) suggesting that biotic interactions with invasive species are the leading driver of the decline of native freshwater fish assemblages (Hermoso et al., 2011). Whether these differences are related to the different geographical, ecological context or to methodological differences remains unclear and requires further investigation, also with a particular focus on the fast-moving advances in JSDMs (D'Amen et al., 2018;Hui, 2016;Kissling et al., 2012;Ovaskainen, Abrego, Halme, & Dunson, 2016;Pollock et al., 2014;Tikhonov, Abrego, Dunson, & Ovaskainen, 2017;Warton et al., 2015). We found limited evidence for biotic-driven co-occurrence patterns in fish communities in the Ebro River as indicated by weak positive or negative pairwise species correlations after accounting for shared environmental responses ( Figure 4). However, several facilitative interactions between fish as detected by our model seem unlikely, but positive residual correlations between fishes could have been due to a shared response to an abiotic variable that was not considered (Pollock et al., 2014; but see also Zurell et al., 2018). For example, the strongest positive residual correlation was detected between Ebro nase and Ebro barbel, that is two lithophilic species, indicating that our JSDM might lack information on bottom substrates.
The high degree of shared environmental responses among alien fish (compared to native vs. alien species, Figure 5) suggests strong environmental filtering and that alien fish occur in distinct downstream habitats largely affected by river regulation. This is also supported by the fact that the native Iberian fish fauna has very few limnophilic species, that is species that prefer stagnant waters and could thus colonize novel ecosystems such as reservoirs. Thus, our results might also be viewed in the context of Darwin's naturalization conundrum, which postulates the importance of functional distinctiveness of alien species from native species to avoid competition and, at the same time, the importance of shared similarity to pass environmental filters and successfully establish (Diez, Sullivan, Hulme, Edwards, & Duncan, 2008;Marx, Giblin, Dunwiddie, & Tank, 2016;Thuiller et al., 2010). Although in our study, we did not measure functional or phylogenetic relatedness, differences in functional traits between alien and native species (e.g., related to limnophily, spawning substrate, temperature) have been shown elsewhere ; Vila-Gispert, . Moreover, the alien fish that have been introduced in the Ebro River are from genera without close native relatives (congeners, see Supporting Information Table S2.2).
Although our results suggest that alien species are rather passengers of ecological changes than the leading driver, we acknowledge that habitat degradation and alien species might complexly interact and jointly govern the decline of native species (Didham et al., 2007). Thus, effective conservation of native fishes in the Ebro River should focus on both the restoration of habitats and the natural hydrologic regime and the control of invasive species, particularly the prevention of further introductions.
There are some limitations that might affect our modelling results. Despite our efforts to minimize multicollinearity and to select methods that are relatively robust to it, we acknowledge the general intercorrelated nature of environmental variables.  Thorp, Thoms, & Delong, 2006). Furthermore, we note that our study is limited by not considering temporal dynamics, that is changes in species assemblages or environmental conditions over time which might have also influenced our model results. Our analysis represents a snapshot based on aggregated fish samplings over multiple years and assuming constant land use, climate, number of barriers etc. over that time. However, we acknowledge that the introduction, establishment and spread of alien species are complex time-dynamic processes in a changing environment and involve potential interactions with native species that might also be variable in time. Consequently, we strongly encourage further studies to use comprehensive time-series datasets documenting biological invasions (potentially also including numbers of propagules and failed introductions) and to investigate the dynamics of species invasions and potential biotic interactions with native species over time (Kuczynski & Grenouillet, 2018).
In summary, our results revealed the proliferation of alien fishes in the Ebro River and the simultaneous decline of many native fish species to be more strongly correlated with environmental rather than biotic associations. Particularly, the distributional limits of alien fishes were strongly related to larger-scale temperature-related climatic factors. Additionally, local-scale factors related to habitat degradation and hydrologic alteration were found to be important correlates of the spatial distribution of many alien fishes. However, detecting biotic effects (e.g., predation, competition) causing the proliferation of alien fishes in our study proves challenging because ( Food and Environment and the many people involved in the surveys for providing the fish data. We also thank four anonymous reviewers for their helpful comments.

DATA ACCE SS I B I LIT Y
Species and environmental data, and BRT and JSDM models used in this study are available as csv files and R scripts from the figshare repository: https://doi.org/10.6084/m9.figshare.6530588.