Hydrologic variation influences stream fish assemblage dynamics through flow regime and drought

Hydrologic variation can play a major role in structuring stream fish assemblages and relationships between hydrology and biology are likely to be influenced by flow regime. We hypothesized that more variable flow regimes would have lower and more variable species richness, higher species turnover and lower assemblage stability, and greater abiotic environment-fish relationships than more stable flow regimes. We sampled habitats (pool, run, and riffle) in three Runoff/Intermittent Flashy streams (highly variable flow regime) and three Groundwater Flashy streams (less variable flow regime) seasonally (spring, early summer, summer and autumn) in 2002 (drought year) and 2003 (wet year). We used backpack electrofishing and three-pass removal techniques to estimate fish species richness, abundance and density. Fish species richness and abundance remained relatively stable within streams and across seasons, but densities changed substantially as a result of decreased habitat volume. Mixed model analysis showed weak response variable-habitat relationships with strong season effects in 2002, and stronger habitat relationships and no season effect in 2003, and flow regime was not important in structuring these relationships. Seasonal fish species turnover was significantly greater in 2002 than 2003, but did not differ between flow regimes. Fish assemblage stability was significantly lower in Runoff/Intermittent Flashy than Groundwater Flashy streams in 2002, but did not differ between flow regimes in 2003. Redundancy analysis showed fish species densities were well separated by flow regime in both years. Periodic and opportunistic species were characteristic of Runoff/Intermittent Flashy streams, whereas mainly equilibrium species were characteristic of Groundwater Flashy streams. We found that spatial and temporal variation in hydrology had a strong influence on fish assemblage dynamics in Ozark streams with lower assemblage stability and greater fluctuations in density in more hydrologically variable streams and years. Understanding relationships between fish assemblage structure and hydrologic variation is vital for conservation of fish biodiversity. Future work should consider addressing how alteration of hydrologic variation will affect biotic assemblages.


Study area
Study streams were in the Ozark Plateau, USA that is composed of the Ozark Highlands and the Boston Mountains (Fig. 1). In this region, stream substrate ranges from bedrock-boulder to gravel, and riffle-pool geomorphology is typical 23 . Land use in the Ozark Plateau is primarily forest dominated by oak and hickory and agriculture dominated by pasture and hay fields 23 . The Ozark Highlands has less elevation change 24,25 and more nutrientrich soils than the Boston Mountains contributing to higher percentage agricultural land in this region. Karst topography in the Ozark Highlands results in streams that are heavily spring-influenced, making them cooler with more stable flow regimes than Boston Mountain streams 22,25,26 .
We selected six headwater stream reaches: Bear (35° 46.1601′ N, 92° 48.0548′ W, second order, 34 sq km drainage), Cave (35° 51.4023′ N, 93° 1.1812′ W, second order, 32 sq km drainage), and Falling Water Creeks (35° 44.054′ N, 92° 56.014′ W, second order, 37 sq km drainage; Falling) in the runoff/intermittent flow regime, and Tomahawk (36° 3.888′ N, 92° 47.403′ W, third order, 20 sq km drainage), Water (36° 4.7519′ N, 92° 41.8947′ W, third order, 20 sq km drainage), and North Sylamore Creeks (35° 59.9641′ N, 92° 13.1154'W, fourth order, 120 sq km drainage, Sylamore) in the groundwater flow regime (Fig. 1). Streams were selected according to the following criteria: (1) wadeable for sampling, (2) similar stream width and depth, (3) sites had an intact riparian zone, and (4) sites were not affected by urban areas or dams. Reaches within the streams were selected on the basis of stream width and accessibility. Groundwater streams had significantly higher conductivity than runoff/ intermittent streams, indicating the influence of groundwater and the underlying limestone and dolomite in these streams. All sites were within the White River drainage basin. Although average annual rainfall in the area ranges 1.0-1.2 m year −127 , reduced flow is characteristic from June through November.
Leasure et al. 22 classified natural flow regimes in the Interior Highlands and the streams in the present study were predicted to be Intermittent Flashy (Cave and Tomahawk), Runoff Flashy (Bear and Falling) and Groundwater Flashy (Water and Sylamore). Groundwater Flashy streams never dried completely and had less daily flow variability and frequency of low flow spells, and greater base flow index than runoff-dominated streams. Conversely, Runoff Flashy and Intermittent Flashy streams had greater daily flow variability and frequency of low flow spells, and much lower base flow index than Groundwater Flashy streams, as well as experiencing 2-15 days (Runoff Flashy) or 1-3 months (Intermittent Flashy) of drying per year. Drying in these systems is often complete but can also lead to residual pools with some standing water. Ground-truthing suggested that all of these classifications were appropriate with the exception of the Tomahawk Creek site, which appeared to be Groundwater Flashy based on abiotic and biotic variables. Unlike Cave Creek, in which some habitat units dried completely during summer and autumn 2002 and autumn 2003, no habitat units in Tomahawk Creek dried completely during 2002 or 2003. Given this and given that the Tomahawk Creek sampled stream segment is just upstream of a stream segment classified as Groundwater Flashy 22 , we reclassified this stream segment as Groundwater Flashy for the present study. www.nature.com/scientificreports/ per stream reach. We determined habitat types by qualitatively assessing depth and flow rate of the stream. Riffles were relatively fast flowing, shallower water with noticeable surface aeration. Runs consisted of fast to moderately flowing water with unbroken surface flow. Slow moving, deeper water was pool habitat. Prior to sampling, block nets were placed at the upstream and downstream ends of the entire stream reach to be sampled to prevent the potential immigration or emigration of fish within the reach. Additionally, each habitat unit was closed to fish escape just prior to sampling by placement of block nets across the downstream and upstream boundaries of the habitat unit. One person operated the backpack electrofisher along with a two-or three-person team of netters while all moved upstream through the habitat and removed fish they encountered. A second and third pass were made within each habitat unit with an equivalent effort. Fish collected from each pass were held in separate containers, identified to species, and measured (total length, TL) to the nearest 10 mm. All fish were then returned to the habitat unit from which they were collected and the process was repeated at the next adjacent upstream habitat unit. This study followed relevant guidelines and regulations. All the experimental protocols were approved by University of Arkansas Institutional Animal Care and Use Committee protocol #01027. Reporting in this manuscript follows the recommendations in the ARRIVE guidelines 28 .
Physical-chemical environmental data collection. Habitat  www.nature.com/scientificreports/ was then divided by the habitat unit area to derive species densities. For stream site estimates of abundance or density we summed the habitat unit estimates at a given site by season by year.
To determine how differences in capture probability by season would affect population estimates we ran simulations in the program MARK using the CAPTURE module. We used the Zippin removal estimator with constant capture probability over three sampling occasions (i.e., three pass removal sampling) and ran 1000 simulations for each combination of capture probability and population size. We varied capture probability from 0.10 to 0.90 and population size from 50 to 300 individuals.
Ordination was used to determine relationships between fish species densities and composition by flow regime and season. Redundancy analysis (RDA) was used to ordinate streams by seasons based on fish species relative densities in CANOCO 4.5 31 . Stream and season were used as explanatory variables and covariables 31 . We focused on streams for this analysis to examine fish species composition among streams, but we also examined influence of flow regime on fish assemblage structure. A linear model and RDA were appropriate because a preliminary detrended canonical correspondence analysis showed that gradient lengths were < 3 standard deviations 32 . Partial RDAs and Monte Carlo permutation tests 31 were used to test the null hypothesis that season and stream categories had no relationship to fish species densities.
We used three-way ANOVAs to examine the effect of flow regime, season and year on total fish density, total abundance, fish species richness and habitat volume. For the univariate analyses we were focused on habitat volume and drying, so for these analyses density was calculated as number of fish m −3 . When significant interactions were found, we used one-way ANOVA and Tukey's tests to examine differences among seasons separately for each flow regime.
We used mixed models to examine relationships between response variables species richness, total abundance and total density (number m −2 ) and fixed effects substrate size, habitat volume and current velocity, as well as season and flow regime. Habitat unit was nested in stream and treated as random variables. A separate mixed model analysis was done for each year. Each model contained fixed effects season and flow regime and all additive combinations of depth, substrate and current velocity were used in the candidate models, so six models per year were examined. Continuous predictor variables were standardized. Akaike Information Criterion corrected www.nature.com/scientificreports/ for small sample size (AICc) was used to select best models. Models within two AICc scores were considered equally plausible 33 . For this analysis packages "lme4" and "AICcmodavg" were used in R 34 . Seasonal species turnover rates and assemblage stability were calculated for each year using the package "codyn" 35 in R 34 . Species turnover is the proportion of species that differ between time points and assemblage stability is the temporal mean divided by the temporal standard deviation. Species turnover was compared among flow regime, season and year with three-way ANOVA. Effect of flow regime and year on assemblage stability were examined with two-way ANOVA. For all linear models, residuals indicated that total fish density, total abundance, habitat volume and assemblage stability required log 10 transformation to meet homogeneity of variance assumptions, whereas species richness and species turnover required no transformation. ANOVAs were conducted in SYSTAT 13 (Systat Software, San Jose, CA).
We assigned species to life-history strategies (Periodic, Opportunistic, Equilibrium) based on Robison and Buchanan 36 , Pflieger 37 , Winemiller 38 and Hoeinghaus et al. 39 . For species that were between categories, we used the approach of Hoeinghaus et al. 39 and provided combined categories (e.g., Opportunistic/Equilibrium). We determined characteristic species for ecoregion/flow regimes based on RDA and species relative densities. We defined characteristic species as those that were abundant in a particular flow regime.

Results
Hydrologic variation. Peak flows were greater in Bear Creek than Sylamore Creek despite a greater drainage area in Sylamore Creek, giving Bear Creek over twice the unit peak discharge as Sylamore Creek (    (Table 3). However, there were no significant relationships between water depth and fish species richness (Table 4), whereas total density showed a significant negative relationship with water depth ( Table 4). The best model for total fish abundance was the global model in 2002 (Table 3), but there were no significant relationships between the habitat variables and total abundance (Table 4). Season had a strong effect and was significant in all models in 2002, whereas flow regime was not significant for any model (Table 4). In 2003, the best model for fish species richness was the depth and current velocity model  www.nature.com/scientificreports/ (Table 3). Species richness was positively related to depth and negatively related to current velocity ( Table 4). The best model for total fish abundance was the depth model in 2003 (Table 3), with depth being positively related to total abundance ( Table 4). The best model for total fish density was the current velocity model in 2003 (Table 3), with velocity being negatively related to total density (Table 4). No response variable showed a significant relationship with season or flow regime in 2003 (Table 4). Species turnover (proportion of species differing between time points) varied by year and assemblage stability (temporal mean divided by temporal standard deviation) varied by year and flow regime. Species turnover showed no two-or three-way interactions among flow regime, season and year (ANOVA p > 0.05). There was a strong year effect with species turnover significantly greater in 2002 than 2003 (ANOVA p = 0.001, Fig. 5). Assemblage stability showed a marginal flow regime by year interaction (ANOVA p = 0.053; Fig. 6). Assemblage stability was significantly lower in 2002 than 2003 in Runoff/Intermittent Flashy streams (ANOVA p = 0.049), but did not differ between years in Groundwater Flashy streams (ANOVA p = 0.374; Fig. 6). In 2002, assemblage stability was significantly lower in Runoff/Intermittent Flashy than Groundwater Flashy streams (ANOVA p = 0.003), whereas in 2003 the two flow regimes did not differ significantly in assemblage stability (ANOVA p = 0.537; Fig. 6).
Redundancy analysis showed a significant relationship between fish species densities and explanatory variables in both 2002 (Monte Carlo p = 0.001) and 2003 (Monte Carlo p = 0.001). There was seasonal change in fish assemblage structure (species composition and relative density) that appeared to be driven largely by changes in fish species densities. Fish assemblages showed a separation between flow regimes in 2002 along RDA axis 2 and in 2003 along RDA axis 1 (Fig. 7). In both years, there were greater densities of central stonerollers, creek chubs, whitetail shiners and slender madtoms in Runoff/Intermittent Flashy streams and greater densities of banded sculpins, hornyhead chubs and lampreys in Groundwater Flashy streams (Fig. 7). Southern redbelly dace were abundant in Groundwater Flashy streams, but they were also found in high numbers in Cave Creek, an Intermittent Flashy stream. The other strong pattern found in the RDA ordinations was the seasonal shifts in fish species densities driven by strong increases in densities of the common fish species from spring through autumn, especially in Runoff/Intermittent Flashy streams. In 2002, autumn differed most from the other seasons, with particularly high relative fish densities in Bear and Cave Creeks (Fig. 7). In 2003, there was a strong change in fish species densities in Bear Creek along RDA axis 1 and Cave Creek along RDA axis 2 (Fig. 7).
Opportunistic, Opportunistic/Equilibrium and Equilibrium species were characteristic of both stream types. Periodic and opportunistic species were characteristic of Runoff/Intermittent Flashy streams, whereas mainly

Discussion
Hydrology appeared to play a major role in structuring fish assemblages of Runoff/Intermittent Flashy and Groundwater Flashy streams and the patterns largely followed expectations from ecological theory. In the case of Runoff/Intermittent Flashy streams, highly variable discharge with high flows in spring and extensive stream drying in summer led to strong seasonal changes in fish species densities with reduced assemblage stability but this was dependent on hydrologic variability. In Groundwater Flashy streams, more seasonally stable flows led to less extreme seasonal shifts in fish species densities and greater assemblage stability. Others have shown strong effects of drought on stream fish assemblage dynamics 14,40,41 . Numerous studies have examined the effects of drought and temporal variation related to hydrology on fish assemblage dynamics in streams [42][43][44] , but few have explicitly examined the role of differing flow regimes in this process. Fish assemblage structure differed among hydrologically variable and hydrologically stable streams in Wisconsin and Minnesota 5 . Warfe et al. 8 examined three intermittent and three perennial streams in Tasmania and found that they differed in abiotic and biotic variables including productivity and macroinvertebrate assemblage structure, whereas fish assemblages were not strong indicators of flow regime. However, they suggested this was largely due to low diversity and abundance of fish in these systems, which were much lower than species richness and fish abundance in the present study. www.nature.com/scientificreports/ Another study found that fish assemblage stability (annual variation of fish biomass) was less variable in more species rich communities and was not associated with hydrology 45 . We found that spatial and temporal variation in hydrology appeared to have a strong influence on fish assemblage dynamics in Ozark streams with lower assemblage stability and greater fluctuations in density in more hydrologically variable streams. Differences in fish assemblages among flow regimes appeared to be responsible for driving some of the patterns we saw in relationships among season and fish assemblage structure. In Runoff/Intermittent Flashy streams mainly periodic and opportunistic species were driving the strong seasonal shifts in density, whereas in Groundwater Flashy streams a set of opportunistic, equilibrium and periodic species influenced the patterns of fish assemblage structure. These differences among seasons in fish assemblage structure were largely due to changes in fish density over time. In Runoff/Intermittent Flashy streams these changes in density were largely influenced by changes in habitat volume. Therefore, it appears that Runoff/Intermittent Flashy streams have a strong concentration effect due to drought and stream drying, whereas Groundwater Flashy streams do not.
This type of concentration effect and boom and bust cycles are common in intermittent streams and are related to stream drying and fish movement. Fish may move among habitats and stream reaches when flows are high and then may be trapped when flows recede 44 . Drying and loss of habitat volume can then produce a concentration effect leading to high fish densities. This can also lead to episodic isolation and habitat fragmentation that can prevent dispersal and rescue effects 46,47 . However, intense drying can lead to substantial fish mortality, including the extreme of extirpation of all individuals within the habitat unit 4,44 . It is also likely that recruitment affected the seasonal density changes in these systems as most of the fish species are spring and summer spawners. In our study streams drying and a concentration effect were important especially during the drought year. However, it is unclear the role that fish movement and recruitment may have played in shaping these patterns and this is a potential avenue for future study.
We found fewer environment-fish relationships during a drought year. Fewer environment-fish relationships during drought could be due to increased biotic interactions. As streams dry during drought, predation often increases and competition is also likely to intensify 48 . Ludlam and Magoulick 49 found that seasonal drought led to increased effects of fish and crayfish consumers on periphyton biomass in Ozark streams. These increased biotic interactions could lead to reductions in importance of abiotic factors. As expected, we found fish species richness was positively related to water depth during a non-drought year, but this was not the case during a drought year. This suggests that strong drying may negate some environment-fish relationships. However, previous research in Groundwater Flashy streams in the Ozark Highlands found that there were more fish assemblage-environment relationships during a drought year than a wet year, but more macroinvertebrate-environment relationships during the wet year than drought year 17 . It is likely that assemblage-environment relationships are context dependent, influenced by factors such as taxon, flow regime, and drought frequency and intensity.
We found periodic and opportunistic species were characteristic of Runoff/Intermittent Flashy streams, whereas mainly equilibrium species were characteristic of Groundwater Flashy streams. Poff and Allan 5 found that fish assemblages from hydrologically variable sites tended to be generalist feeders, tolerant of silt and slowvelocity species associated with headwater streams. Olden and Kennard 13 examined how southeastern USA and eastern Australia fish life history strategies related to hydrologic variability and productivity and found that opportunistic strategists were associated positively and periodic strategists were associated negatively with hydrologic variability. McManamay et al. 50 related fish traits to hydrologic classes across the conterminous USA and found that periodic strategists were associated with stable, predictable flows and opportunistic strategists were associated with intermittent, variable flows. Our findings agree with others in that opportunistic strategists  www.nature.com/scientificreports/ were associated with hydrologic variability (i.e. Runoff/Intermittent Flashy streams) and equilibrium strategists were associated with hydrologic stability (i.e. Groundwater Flashy streams). However, in contrast to previous studies, we also found periodic strategists to be associated with hydrologically variable streams. Periodic strategists produce large clutches and have high adult survival during periods of suboptimal conditions 12 . Therefore, it may not be surprising that these species were associated with Runoff/Intermittent Flashy streams that had flooding in spring and strong drying during summer and autumn. Additionally, we did a simple descriptive examination of traits and did not account for phylogeny, so it is possible that some of the differences among studies relates to phylogeny. However, this seems unlikely given that we found similar patterns to others for two of three life history strategies. Unexpectedly, fish species richness was generally temporally stable and both species richness and seasonal turnover did not differ between flow regimes. However, seasonal species turnover varied strongly by year and appeared strongly affected by temporal variation in hydrology. In particular, the drought year significantly increased species turnover. Fish assemblage stability was also reduced in the drought year in Runoff/Intermittent Flashy streams and was lower in Runoff/Intermittent Flashy than Groundwater Flashy streams in the drought year. Therefore, although drought and flow regime may not have strong effects on fish species richness, species composition and assemblage stability appear to be affected by drought. Based on previous work examining survival and movement in similar systems, we suspect that increased mortality during drought may lead to species extirpations, but emigration may also play a role 51 , and both of these are likely to lead to increased species turnover and reduced assemblage stability. In Mediterranean streams, Magalhaes et al. 52 found that multi-year droughts had little effect on fish species richness, composition and rank abundances, but did affect individual species abundances. Others have suggested that fish assemblages recover rapidly from drought 4,52,53 . In experimental manipulations of drought and fragmentation, Driver and Hoeinghaus 54 found that fish survivorship and species diversity were affected little at the experimental unit level, but effects were found at the habitat level. This suggests that effects of drought are likely to be dependent on spatial and temporal scale. Future research would be required to determine mechanisms behind these changes in fish assemblage structure.
Differences in capture probabilities and species detection probabilities among seasons and streams could potentially explain some changes in fish density. However, our species detection probabilities were high and relatively stable and our simulations showed that changes from 0.3 to 0.9 in capture probabilities among sampling periods will lead to increased precision of estimates for population sizes 50-300, but estimates should be unbiased. The only exception to this could be Falling Water during late summer 2002, where we may have slightly underestimated abundances at this time. However, this should not have had a noticeable effect on the overall estimates for Runoff/Intermittent Flashy streams. Therefore, changes in fish densities and species richness over time in some of our study streams were real and not simply the result of changes in species detection or capture probabilities.
It is likely that the fish assemblage dynamics we observed could be influenced by factors that alter stream hydrology, such as climate change or water withdrawals. Water withdrawals could lead to increased flow  www.nature.com/scientificreports/ variability, frequency of low flows and zero flow days, and fall rate, while decreasing base flow and minimum 30-day mean flow (Leasure and Magoulick unpublished data). Future work addressing how alteration of hydrologic variation will affect fish and other biotic assemblages would be worthwhile.

Conclusions
We found that spatial and temporal variation in hydrology had a strong influence on fish assemblage dynamics in Ozark streams with lower assemblage stability and greater fluctuations in density in more hydrologically variable streams and years. We found fewer environment-fish relationships during seasonal drying in a drought year. Fish species life history strategies generally followed expected patterns, but periodic strategists were unexpectedly associated with more hydrologically variable systems. Understanding relationships between fish assemblage structure and hydrologic variation is vital for conservation of fish biodiversity. Future work should consider addressing how alteration of hydrologic variation will affect biotic assemblages.