Anomalous droughts, not invasion, decrease persistence of native fishes in a desert river

Study area

We studied the upper Verde River, a major tributary of the Salt River (Gila River basin) in the US state of Arizona. The Verde River flows around 270 km from its headwaters in Big Chino Wash (Prescott National Forest) to its confluence with the Salt River near Phoenix. In 1994, the Rocky Mountain Research Station (USDA Forest Service) established seven monitoring sites from Sullivan Lake to Sycamore Creek (Rinne & Miller, 2006), which were sampled continuously (every April) until 2008. This upper part of the Verde River is relatively undisturbed, without any flow-altering dams or major diversions, and has been an ideal setting to study the abiotic and biotic factors that influence fish assemblage structure in desert rivers (Stefferud & Rinne, 1995). Sample sites ranged from 150 to 300 m in length and included a variety of aquatic habitats. Backpack electrofishing units were used to sample under debris, banks, and in riffles. For high- and low-gradient riffles and for glide-runs, shocking was conducted from upstream to downstream, and fish were collected into a 6-m, 3-mm mesh bag seine. Trammel nets (30 m in length and meshes of 13-, 40-, and 80-mm mesh arrays) were employed to sample in deep pools (>2 m). All fishes were counted and returned alive to the stream, and counts from different sampling methodologies were combined at each site and visit. Throughout the 105 recorded samples (7 stations × 15 years), a total of 16 fish species were detected, of which 6 were native and 10 were non-native. For the present work we dropped from the analyses 2 species with <2% occurrence (both non-native: fathead minnow Pimephales promelas, 1.9% occurrence; and largemouth bass Micropterus salmoides, 1.0% occurrence). This delivered a final matrix of 14 fish species (6 native, 8 non-native) × 105 samples. Daily mean discharge data for 45 complete years (from 1/1/1966 to 31/12/2010) were obtained from the US Geological Service (USGS) station 09504000 (Verde River near Clarkdale, AZ, USA).

Stochasticity in discharge variation

We used the Discrete Fast Fourier Transform (DFFT) to identify the characteristic signal of discharge variation in the Verde River, and the residuals or ‘anomalies’ from this long-term seasonal signal. We then used the maximum (high-flow) or minimum (low-flow) annual anomalies to construct hydrologic (abiotic) variables that would be considered as covariates in the multivariate time-series modeling. Hydrologic variation has been traditionally quantified using recurrence intervals (e.g., with Log-Pearson III distributions). However, such techniques focus on the magnitude of either high- or low-extremes (not both together) and ignore the seasonal occurrence (i.e., timing) of these events; a crucial aspect for ecological applications as biological impacts of high and low flows depend on seasonal base flows (Bunn & Arthington, 2002). Statistical tools borrowed from signal processing, such as the DFFT, provide a means for decomposing variation in discharge into the periodic (seasonal) and stochastic (interannual) components. This allows analyzing discharge series in a more ecologically meaningful way, as extreme forms of stochastic variation (but not predictable, periodic signals) cause ecological disturbance (Lande et al., 2003; Grossman & Sabo, 2010). A complete description of the method employed here is provided in Sabo & Post (2008). Briefly, the DFFT is applied to the time series and the characteristic signals (frequencies, phases, and amplitudes) of seasonal hydrologic variation (i.e., the long-term seasonal trend) are extracted from the power spectrum in the frequency domain. Observed daily events can then be compared to long-term seasonal trend, and residual events (observed – expected discharge) or discharge anomalies can be computed. Anomalies either higher (floods) or lower (droughts) than two standard deviations of the distribution of residual variation around the long-term seasonal can be regarded as catastrophic events (Sabo & Post, 2008). Because these events ‘carry’ information on their magnitude and timing (i.e., they are always of high magnitude and happen ‘unexpectedly’), their occurrence may be regarded as a proxy for ecological disturbance (Sabo & Post, 2008; Grossman & Sabo, 2010). In this study, we computed the long-term seasonal trend for the 45-year discharge series, and identified and extracted the highest- and lowest annual discharge anomaly for each year in the time series. This process yielded a new time series of annual maxima (or minima), which could then be used to index each year in terms of high- or low-flow variation for 1993–2007 (hereafter, discharge anomalies). We used these discharge anomalies as covariates in the subsequent multivariate time-series models of the fish community (Fig. 1).

Exploring long-term trajectories of fishes using multivariate time-series modeling

We assessed the effects of discharge anomalies and biotic interactions on the trajectories of fish using state-space versions of multivariate autoregressive models. Multivariate autoregressive (MAR) models have been used extensively in econometrics, and in ecology they have been mainly employed to quantify the effects of environmental drivers on species’ population growth rates and community dynamics (reviewed in Hampton et al., 2013). Multivariate autoregressive models rely on theory about the patterns of temporal correlation that emerge from environmental drivers and species interactions (Ives et al., 2003). Hence, instead of treating temporal correlation as a nuisance, MAR models use it to estimate effects of drivers vs. interactions on community dynamics. Despite being data-hungry, MAR models are less challenging than mechanistic models, as the latter (but not MAR) require ecophysiological parameters. Moreover, state-space versions of MAR models (i.e., MARSS models) allow inclusion of observation error in the data, which is advantageous because ignoring observation error can fundamentally change our inference about the underlying process (e.g., Knape & de Valpine, 2012). Our fish count data certainly contained observation error that could potentially bias the measured influence of abiotic and biotic drivers on community dynamics. Therefore, MARSS allowed us to separate the variation in the fish count data due to observation error from the variation due to true population fluctuations. This separation is possible because unlike true population fluctuations, observation error does not influence current or future abundances –it only affects our estimation of that abundance. We used the ‘MARSS’ R-package (Holmes et al., 2014), which provides support for fitting MARSS models with covariates (i.e., discharge anomalies) to multivariate time-series data (i.e., fish community abundances) via maximum likelihood, using an Expectation–Maximization algorithm. In the matrix form, a MARSS model takes the following form [Eqn 1 is the state process and Eqn 2 is the observation process]:

(1)

(2)

Data enter the model as y (with y_t being fish abundances modeled as a linear function of the ‘hidden’ or true abundances, x_t), and as c_t-1 (the lagged covariates, in our case discharge anomalies). In the state process [Eqn 1], B is an interaction matrix and models the effect of species on each other, C is the matrix whose elements describe the effect of each covariate on each species (hereafter discharge anomaly effects), and w is a matrix of the process error, with process errors at time t being multivariate normal with mean 0 and covariance matrix Q. In the observation process [Eqn 2], v is a vector of non-process errors, with observation errors at time t being multivariate normal with mean 0 and covariance matrix R.

We compared two different configurations of the C matrix. First, we allowed discharge anomaly effects to differ only between native- and non-native species (origin model in Fig. 1). This model was structured to have only four different values in the elements of the C matrix and hence allowed us to quantify how fishes of native and non-native origin differed (systematically) in their responses to discharge anomalies. Subsequently, we constructed a more complex MARSS model that allowed for species-specific discharge anomaly effects (species model in Fig. 1), thus obtaining as many discharge anomaly effects as the number of species × discharge anomalies. For both the origin and the species models, we compared several model structures, combining in each case two different configurations of the B matrix with two different configurations of the R matrix. In the B matrix [Eqn 1], we compared models allowing for biotic interactions (i.e., with all diagonal and off-diagonal coefficients being estimated) with models not allowing for them (using an identity B matrix, i.e., a diagonal matrix with 1's on the diagonal and 0's elsewhere). In the R matrix [Eqn 2], we compared models considering species-specific observation errors with models considering constant observation errors (i.e., equal across all species). For all models we used log-transformed abundances of the 14 fish species as variates, and the discharge anomalies as covariates. Discharge anomaly effects and biotic interaction strengths were assessed via 95% confidence intervals (1000 bootstrap samples).

Biological trait analyses

In addition to MARSS, we also explored whether differences in responses to discharge anomalies between the native and the non-native assemblage could be explained by differences in their biological traits. To this end, we described each fish species in terms of 38 traits using Fish Traits, a database of ecological and life-history traits of freshwater fishes of the United States (Frimpong & Angermeier, 2009). These traits describe the trophic ecology, body size, reproductive ecology, life history, and salinity and habitat preference of the present species (see Table S1 for more details). Replicating the recent approach by Buisson et al. (2013), we constructed a multidimensional functional space using Gower's distance for each species pair and a Principal Coordinate Analysis (PCoA). The orthogonal axes provided by the PCoA may be regarded as ‘synthetic functional traits summarizing a fish functional niche’ (Buisson et al., 2013). Then, the fish species coordinates in the space defined by the PCoA may be used to calculate functional indices. We aimed at obtaining the functional originality and the functional uniqueness of each species. Functional originality was defined as the Euclidean distance to the average position of the species pool (i.e., the centroid of the fish species coordinates in the space defined by the PCoA), with relatively higher distances to the centroid implying more functionally original fishes. Functional uniqueness was defined as the distance to the nearest neighbor in the functional space, with species falling relatively closer in the functional space being considered more redundant than those being isolated, and hence functionally unique (Buisson et al., 2013). For both functional originality and uniqueness, raw values were standardized by dividing them by the respective maximum value observed among the 14 species. Differences in functional originality and uniqueness between native and non-native fishes were tested using a GLM (1 fixed effect: origin). Finally, biological traits typifying the native vs. the non-native assemblage were identified by means of a similarity percentages (SIMPER) test (999 permutations, single layout: origin factor). Traits contributing highly and consistently to dissimilarity (i.e., those with dissimilarity/standard deviation >1) were considered to discriminate fish origins (native vs. the non-native species). Functional analyses were conducted with Primer v.6 software (Primer-E Ltd, UK) and the ‘vegan’ R-package (Oksanen et al., 2014).

Spatial patterns of the meta-community

Finally, the spatial extent of our fish data allowed us to examine temporal variation in spatial patterns of diversity. We used a beta diversity partitioning procedure to examine variation in the spatial patterns of the fish meta-community (i.e., the set of the seven studied local communities or monitoring sites). The ‘betapart’ R-package (Baselga & Orme, 2012) allowed us to quantify overall beta diversity (β_SOR) for each year, as well as the relative importance of the two additive components that contribute to β_SOR: nestedness (β_NES), which indicates ‘ordered’ species loss and gain, and turnover (β_SIM), which indicates species replacements (Baselga, 2010; Baselga & Orme, 2012). We carried out 15 analyses, one for each sampled year (1994–2008), on presence-absence matrices (14 fish species × 7 sites). Subsequently, we aimed at addressing how β_SOR, β_NES and β_SIM (variates) responded to discharge anomalies (covariates) with an additional MARSS model (Fig. 1), but one in which the variates were a matrix of beta diversity components rather than species abundances. We considered component-specific observation error and all possible interactions of discharge covariates on the beta diversity components (β_SOR, β_NES, and β_SIM). As with our analysis of the temporal dynamics of the global community, we were again interested in quantifying the discharge anomaly effects. We assessed these effects via 95% confidence intervals (obtained numerically after 1000 bootstrap samples). We refer to this model hereafter as beta diversity model (Fig. 1).

Discharge anomalies overview

The long-term seasonal trend extracted with the DFFT (red line in Fig. 2a) showed low basal flows and a seasonal peak in late winter. High flow anomalies generally occurred in two distinct periods (summer monsoons and winter frontal storms), whereas low-flow anomalies were clustered around later winter, between mid-January to mid-April (Fig. 2b). A long drought (9 years, 1984–1992) immediately preceded our 15-year study (shaded area in Fig. 2c; years 1993–2007). Our 15-year period encompassed both wet (1993–1995, 2004–2005) and dry periods (1996–2003, 2006–2007), and started with the most extreme high flow event observed across the entire series (in 1993). Overall, this 15-year period encompassed a total of four catastrophic high flow events and nine low-flow events that were either immediately below or above the catastrophic threshold.

Discharge anomaly effects on the long-term trajectories of fishes

The native and the non-native assemblages demonstrated large fluctuations in abundance and a general negative co-variation (Figure S1). The best (lowest AICc) origin model (the model where the effect of environmental drivers is determined by whether the species is native or non-native; Fig. 1) was the one considering species-specific observation errors and origin-specific biotic interactions (i.e., allowing effects of non-native on native species and vice versa, as well as effects within the native and within the non-native assemblage; model 1 in Table 1). The support for this model was higher than that observed for the best species model (the effect of environmental drivers is determined by species regardless of whether it is native or non-native; Fig. 1), implying that fish responses to discharge anomalies could be safely summarized as native vs. non-native responses. For the species models, species-specific observation errors were also better supported, but the model not allowing species-specific biotic interactions received virtually the same support as the one allowing species-specific biotic interactions (AICc = 1104.2 and 1104.6, respectively; models 5 and 6 in Table 1). To assess if selecting one of these species models would change our inference about the drivers, we compared the discharge anomaly effects estimated by this pair of models (Figure S2). Both models 5 and 6 provided a similar inference about the drivers (Figure S2). Therefore, for the sake of simplicity we selected model 6 as the best species model.

After bootstrapping the discharge anomaly effects estimated by the origin model (model 1), we observed that native fishes responded positively to high-, and negatively to low-flow anomalies (Fig. 3a). In contrast, anomalies did not significantly influence long-term abundances of non-native fishes (Fig. 3a). Biotic interaction strengths between the native and the non-native assemblages were significant in both directions (Fig. 3b), but much closer to zero than the discharge anomaly effects. By bootstrapping the discharge anomaly effects of the species model (model 6) we were able to identify which were the species most and least affected by discharge anomalies. Three native species responded significantly and positively to high flow anomalies (the desert sucker Catostomus clarki, the sonora sucker Catostomus insignis, and the roundtail chub Gila robusta), with this same set fishes responding negatively to low-flow anomalies (Fig. 4). As observed in the origin model, most of the non-native species of the species model appeared to be less responsive. The only exception was the green sunfish Lepomis cyanellus, which was significantly favored by droughts (Fig. 4).

Biological traits of the native vs. the non-native assemblage

To test if differences in the responses to discharge anomalies between the native and the non-native assemblage (as observed in the species and origin models) could be explained by an adaptive suite of traits, we contrasted assemblages based on their functional originality and uniqueness, and on their typifying traits. Functional originality did not differ between native and non-native fishes (GLM; F_1,12 = 2.232, P = 0.161; Table 2), but functional uniqueness was significantly higher in native species (GLM; F_1,12 = 4.951, P = 0.046; Table 2). When investigating individual traits we identified 9 strategies (describing body size, habitat preference, life history and trophic ecology of the fishes) that typified native relative to non-native species (Table 3).

Discharge anomaly effects on the spatial patterns of the meta-community

Beta diversity partitioning showed that the turnover component (β_SIM) dominated over the nestedness component (β_NES) in explaining overall beta diversity (β_SOR) (mean ± SD: β_SOR, 0.438 ± 0.092; β_SIM: 0.248 ± 0.129; β_NES: 0.190 ± 0.094). More importantly, the relative importance of these components was not constant across the 15-year period: despite turnover generally dominating, in some years spatial dissimilarity was largely or totally explained by the nestedness component (Figure S3). The MARSS beta diversity model showed that discharge anomalies played a significant role in explaining transitions between beta diversity patterns dominated by nestedness vs. turnover, with high flow anomalies diminishing, and low-flow anomalies increasing, the relative contribution of nestedness (Fig. 5). This result implies that after anomalous droughts (but not after anomalous floods), beta diversity was largely explained by variation in species richness among local communities.

Discharge anomalies, not non-native species, determine trajectories of native fish

The notion that seasonal flooding has differential effects on native and non-native freshwater faunas has a long history dating to Seegrist & Gard (1972), who reported that the negative covariation between native and introduced trout (Salmonidae) in Sagehen Creek (California, USA) was explained by alterations to the natural flow regime. There, seasonal winter floods favored native salmonid life histories (spring spawners) but not non-native life histories (fall spawners); whereas minor spring floods destroyed native salmonid eggs, thereby enhancing survival of young non-native salmonids. Since then, several studies have demonstrated different responses of native and non-native fishes to seasonal flows, especially in other arid-land rivers in the American Southwest (Tyus & Saunders, 2000; Eby et al., 2003; Propst et al., 2008; Gido & Propst, 2012; Gido et al., 2013). These studies generally show opposite effects of streamflow features on natives and non-natives, and/or significant antagonistic interactions of the non-native assemblage (or some particular non-native species) on the natives.

Our results enrich the findings of these previous studies in at least two ways. First, our DFFT-MARSS framework allowed us to explicitly measure the relative effects of biotic and abiotic drivers of variance in stream fish abundances and we observed only weak biotic effects on the long-term trajectory of the fish community as a whole. Species interactions (captured by the B matrix) were non-significant in the species model, and negative but much less important than discharge anomaly effects (in the C matrix) when considering the global effects of the non-native on the native assemblage in the origin model. Second, the effects of discharge variation on native and non-native fishes were different but surprisingly not opposite in sign. In contrast to strong negative effects of low-flow anomalies on native species, the non-native assemblage was largely unresponsive to high- or low-flow variation (as measured by anomalies). Hence, abundances of native species varied due to significant environmental variation (high- and low-flows) and in spite of a relatively more constant baseline of non-natives.

The context of our results is also important and differs from previous work that has drawn different conclusions. First, our system experiences large flow variation. Disturbance is known to alter competitive interactions among species (Byers, 2002). Therefore, species interactions may indeed be relevant in more stable rivers, such as desert rivers with managed flows (Gido & Propst, 2012; Gido et al., 2013) or rivers under hydroclimates with lower catastrophic variation (i.e., Mediterranean rather than arid and semiarid Koppen climate types; Seegrist & Gard, 1972; Bernardo et al., 2003). Second, the effects of non-native species on native species’ persistence may strongly depend on the identity of the non-native species (Bunn & Arthington, 2002; Pusey et al., 2006). Gido & Propst (2012) found that the channel catfish Ictalurus punctatus was included in the top-ranked models predicting native abundances in the San Juan River (New Mexico and Utah, USA), but in our study, this predator was extremely scarce, as also was the flathead catfish Pylodictis olivaris (both represented <0.1% of the global abundance; Table 2). Although small-bodied non-native fishes like the red shiner Cyprinella lutrensis (by far the most abundant species in our study) have been found to prey on larvae of native fishes (Ruppert et al., 1993; Brandenburg & Gido, 1999), these interactions did not influence our native assemblage. Overall, our results suggest that even when biotic interactions are potentially relevant, faunal communities in highly variable environments may be driven predominantly by environmental variation (abiotic paradigm).

Droughts decimate the native assemblage and lead to ‘functional debt’

The term lentification has been recently coined to describe the widely observed transformation of rivers from lotic to lentic habitats (Sabater, 2008). The fact that native species are severely affected by anomalous droughts in our system is very relevant when considering the impacts of lentification on biodiversity. In the Southwest United States, the equivalent of nearly 76% of streamflow is being currently appropriated by humans, and this figure could increase if demographic projections that predict a doubling of the super-region's population within this century come true (Sabo et al., 2010a). More importantly, in the American Southwest high human water appropriation is interacting with a transition to a more arid climate (Seager et al., 2007; Karl et al., 2009). Our study suggests that if frequency and magnitude of droughts increase in southwestern rivers and anomalous droughts (due to high intensity or aseasonal timing) become the new normal, functionally homogenized fish communities with low prevalence of native members are to be expected.

We found that the native, but unexpectedly not the non-native assemblage, responded to discharge anomalies. The general lack of response of non-natives apparently contradicts previous studies suggesting that natural flow regimes should provide environmental resistance to the spread of non-native species (Kiernan et al., 2012; Gido et al., 2013; Mims & Olden, 2013). Those views are based on the ‘natural flow regime’ concept (Poff et al., 1997), which implies that non-native organisms that evolved in systems with different suites of abiotic features should not find suitable conditions in systems with natural regimes (Lytle & Poff, 2004). In our study, only the green sunfish Lepomis cyanellus benefitted from the anomalous droughts that were negatively associated with native abundances, and no non-natives were reduced in abundance by the floods that were positively associated with native abundance. Whereas the positive response of green sunfish to anomalous droughts is in agreement with its high preference for slow current conditions and lacustrine habitats (Frimpong & Angermeier, 2009; Gibson et al., 2014; Pool & Olden, 2014), non-natives not responding to anomalous floods is a more intriguing outcome. We offer two explanations for the observed lack of response by non-natives: (i) catastrophic floods (4 events during the studied period and 6 preceding it) were perhaps not intense enough to influence the non-native assemblage; or (ii) non-natives indeed responded (negatively) to flow variation on the short-term but recolonized quickly, so that analyses of long-term trajectories would not reflect such responses. Historical flow data suggest that the first hypothesis is highly unlikely, since our fish time series started with the most extreme floods recorded throughout the whole 45-year period (including a 100-year recurrence event), which dramatically altered stream channel morphology and the aquatic habitats of the upper Verde River (Rinne & Stefferud, 1998). In contrast, the second hypothesis seems plausible as other studies have reported that non-native fishes in arid-land rivers are highly resilient. Pool & Olden (2014) found that an experimental flood from Alamo Dam (Bill Williams River basin, western Arizona) elicited a short-term reduction in the abundance of non-native species in sites downstream to the dam, but the pre-flood composition (similar to that of the upper Verde River) was recovered shortly after the event. Given the temporal resolution we employed (yearly spaced, with a variable lag of several months between the high flow event and the spring fish sampling), it is likely that we missed such short-term responses. This suggests that flow management may be able to enhance the native components of the assemblage and boost the ratio of native:non-native abundance, but mechanical removal may need to be invoked to manage the non-native assemblage in the long term.

As expected, biological traits offered a mechanistic explanation for the effects of hydrology on the relative persistence of native and non-native fishes. Traits conferring resilience from or resistance against flood disturbance typified the native assemblage. These traits included: habitat preference for moderate current and not for lacustrine systems, long spawning seasons and short longevities, or trophic preference for detritus and not for large macroinvertebrates and other vertebrates. Similarly, a study that paired flow-regulated and free-flowing rivers across the United States found that fish assemblages in free-flowing locations, relative to assemblages below dams, were characterized by a relatively lower proportion of equilibrium species (species with strategies favored in more stable, predictable environments) (Mims & Olden, 2013). Our analyses show that trait combinations of the native component were also more unique (sensu Buisson et al., 2013). This result contributes to the notion that generally extirpations do not occur randomly, but elicit disproportionate functional homogenization (Pool & Olden, 2012). Of the three drought-sensitive native species of our study, the Sonora sucker Catostomus insignis and the Roundtail chub Gila robusta fall under the threatened IUCN category, with the latter additionally being a candidate endangered species under the Endangered Species Act criteria. We observed both species were, at the same time, very functionally unique (≥ 0.9 of a possible 1, Table 2). Functional homogenization had previously been attributed to the contributions of non-native species on the assemblage as a whole (Rahel, 2002; Pool & Olden, 2012). Here, we demonstrate that anomalous droughts may induce a ‘functional debt’ whereby trait combinations disappearing as a consequence of the extirpation of drought-sensitive native species will probably not be offset by the trait combinations of the remaining drought-resistant non-natives.

Environmental variation drives meta-community diversity patterns

Investigating beta diversity patterns is key to understanding meta-community dynamics and the relative importance of deterministic vs. stochastic processes (Chase & Myers, 2011). Beta diversity partitioning has lately become a popular tool in community ecology because it allows disentangling the relative contributions of variation in species richness and in species identities to overall composition dissimilarity (Baselga, 2010; Baselga & Orme, 2012; Legendre, 2014). Although beta diversity (β_SOR) and its additive components (nestedness β_NES, and turnover β_SIM) are usually studied across space (e.g., Gossner et al., 2013; Nascimbene et al., 2013; Escoriza & Ruhí, 2014), some studies have examined them through time to assess global change impacts on the spatial patterns of biodiversity (Albouy et al., 2012; Angeler, 2013). To our knowledge our study is the first to integrate environmental covariates (here discharge anomalies) with beta diversity and its components (as variates) in the context of stochastic modeling of time series (beta diversity model). This analysis revealed that different types of hydrological extremes lead to different meta-community patterns, with anomalous droughts increasing, and anomalous floods decreasing, the relative importance of nestedness (β_NES).

Nestedness is widespread in freshwater biotas (Heino, 2011; Batzer & Ruhí, 2013; Tornés & Ruhí, 2013) and habitat suitability is known to influence this structure (Heino & Muotka, 2005; McAbendroth et al., 2005). Therefore, results from the beta diversity model were not only expected but also congruent with the origin and species models, as low flows differentially harming the native assemblage should ‘naturally’ generate a nested meta-community (see schematization in Fig. 6). In this way, the unresponsive non-native assemblage constituted a core set of taxa present in virtually every local community, whereas some drought-sensitive species disappeared from the most physically adverse sites after anomalous droughts. Differential extinction and selective environmental tolerances are well-known determinants of nestedness (Ulrich et al., 2009). Our study exemplifies these mechanisms and further shows that successive nestedness- and turnover-dominated scenarios may be explained by low resistance, but high resilience, of part of the assemblage (sensu Connell & Sousa, 1983). Overall, seasonal floods are essential to maintain mixed assemblages of native and non-native fishes as only by preserving meta-community dynamics, the drought-sensitive assemblage can bounce back to pre-drought conditions after this disturbance.

Implications of the current findings

In this study we quantified the stochastic component of discharge and integrated this variation directly into community time-series models. Our results suggest that stochastic variation can predict community composition and its spatial variability even in very flashy environments, offering an alternative but effective way to understand how hydroclimatic anomalies and invasive species can affect long-term trajectories of native riverine fauna. Climate and invasion are two key components of global change (Staudt et al., 2013), and well recognized drivers of freshwater fish extinctions (Xenopoulos et al., 2005; Rahel & Olden, 2008). Disentangling their effects on native assemblages is crucial to anticipating the ecological impacts of new hydrological regimes. Although stream and riparian ecosystems in the American Southwest are subject to frequent hydrological disturbance (Grimm et al., 1997; Sponseller et al., 2010), our results suggest that anomalous droughts (in timing and/or magnitude) are impairing fish communities both taxonomically and functionally. As drought severity increases with projected climate change and groundwater depletion, native biodiversity losses are to be expected in desert rivers of the American Southwest.