Spatially Dependent Responses of a Large-River Fish Assemblage to Bank Stabilization and Side Channels

Sampling Design

The sampling design was a nested hierarchy. We divided the study area into five longitudinal segments (Figure 1). These segments were selected to include proximate locations of bank stabilization and unaltered banks as well as similar local geomorphology, braiding, and slope. Major tributary confluences and diversion dams were excluded from segments wherever possible. However, segment 2 included the Myers Diversion Dam to circumvent dividing this segment by both the Cartersville Diversion Dam and the confluence with the Bighorn River. Additionally, the Cartersville Diversion Dam impedes fish passage more than the Myers Diversion Dam (Helfrich et al. 1999).

Each site consisted of an upstream and downstream channel crossover and a main pool. We used stratified random sampling to select stabilized and reference alluvial and bluff sites (Figure 1) from braided and anabranching reaches (Boyd and Thatcher 2004) within segments. Alluvial sites were unconstrained laterally by bedrock bluffs, whereas bluff sites were constrained laterally by bedrock bluffs on one bank. We omitted all potential sites within 1.5 km of either a major tributary confluence or a diversion dam. Other exclusion criteria were related to having a unique feature; for example, one potential site was excluded because it was about twice as long as all the other potential sites.

Reference sites had no armoring on main pools or crossovers. Stabilized sites had at least 35% stabilization of outside-bend bank length in main pools. However, we made some exceptions to the reference site criteria where no other potential reference sites existed. In segment 1, sections of bank stabilization were present on the upstream crossover (34 m of stabilization; 7% of the upstream crossover banks) of one reference bluff site, the downstream crossover (188 m; 34%) of one reference bluff site, and the upstream crossover (220 m; 37%) and pool (30 m; 1%) of one reference alluvial site. Bank stabilization was present on one bank of the upstream crossover (79 m; 19%) of one reference bluff site in segment 3. We accounted for these exceptions to our reference site criteria by quantitatively assessing the lengths of bank stabilization and treating bank stabilization as a continuous rather than a categorical variable in our analyses of the potential effects of bank stabilization on the fish assemblage.

Fish Sampling

Fish sampling was conducted under late-summer and autumn base-flow conditions: September through early November 2009 and mid-August through mid-October in 2010 and 2011. Fish sampling occurred in mesohabitats: the inside bends of pools, outside bends of pools, and channel crossovers.

We sampled fish with five gears. Fyke nets, bag seines, and otter trawls were deployed to target small-bodied fishes, whereas electrofishing and trammel nets were employed to target large-bodied fishes. Nevertheless, large-bodied fishes were occasionally captured with fyke nets, bag seines, or otter trawls, and small-bodied fishes were occasionally captured with electrofishing or trammel nets. Otter trawls were only deployed in segments 4 and 5 because they effectively capture Sicklefin and Sturgeon chubs (Duncan et al. 2012) and the ranges of these species did not include segments 1–3 (Duncan et al. 2016). Fyke nets, bag seines, and boat electrofishing were used to sample fish along shorelines. Trammel nets and otter trawls were used to sample the fish in the deep portions of the channel in each mesohabitat.

Each gear deployment was a subsample, and the GPS coordinates of each subsample were recorded. We randomly selected the locations of fyke, seine, trammel net, and otter trawl subsamples by dividing each mesohabitat into equal longitudinal sections, then dividing each section into tenths and randomly selecting a starting location therein using the second digit in the seconds field of a digital watch. When sampling crossovers, we flipped a coin to decide which bank to sample. Fyke nets had two 1.2-m-wide and 0.6-m-high rectangular steel frames and two 0.6-m-diameter circular steel frames covered with 3-mm nylon mesh. Each fyke net had a 4.5-m-long lead and a 3-m-long cab and extended 7.5 m from the shoreline when deployed. Our fyke nets were identical to the “mini-fyke nets” described in detail in the upper Mississippi River system Long Term Resource Monitoring Program Procedures (Gutreuter et al. 1995). Three fyke nets were set at the shoreline in channel crossovers, inside bends, and outside bends. Seines were constructed of 6.4-mm mesh and were 9.1 m long and 1.8 m high with a centrally located cubic bag measuring 0.9 m on each side. Two 100-m-downstream seine hauls were made in wadable (less than 1.5 m deep) portions of channel crossovers and inside bends. Boat electrofishing was conducted with a Smith-Root Model 5.0 Generator Powered Pulsator using pulsed direct current at 60 pulses per second. One-pass upstream-to-downstream electrofishing was conducted along the entire shoreline of each mesohabitat where depths were 0.8 m or greater. Two 200-m trammel net and trawl subsamples were collected in each mesohabitat. Trammel nets were 38.1 m long with a 9.5-mm float line at the top and a 13.6-kg lead line at the bottom. Trammel nets had two panels; the inner panel was 2.4 m in height with 2.5-cm-bar mesh, and the outer panel was 1.8 m in height with 20.3-cm-bar mesh. Otter trawls were 4.9 m wide, 0.91 m tall, and 7.6 m long. The inner mesh size was 6.4 mm and the outer mesh size was 38.1 mm. Trawl doors were 76.2 cm × 38.1 cm and weighed 13.6 kg each. The trawl was towed downstream with two 30-m ropes. Otter trawls were identical to the “Model OT16” used for the Missouri River Pallid Sturgeon Population Assessment Program (Hamel et al. 2009).

We used MS-222 to anesthetize fish prior to handling. All fish greater than 30 mm total length (TL) were identified to species and counted. Catostomids, cyprinids, and centrarchids less than 30 mm TL and ictalurids less than 20 mm TL were identified to family, counted, and treated as separate taxa in subsequent analyses. Fish that we were unable to identify to family (0.01% of our total catch) were omitted from subsequent analyses. We captured 45 species of fish (Table 1) and 94,504 fish in total. We captured 72,917 fish with fyke nets, 11,667 with seines, 7,212 by electrofishing, 2,121 with trammel nets, and 587 with otter trawls (see Supplementary Figure S.A.1 available in the the online version of this article).

Spatial Analyses

We used ESRI ArcGIS 10.0 and existing digitizations of bank stabilization, flow lines (Yellowstone River Corridor Resource Clearinghouse 2013), and channel margins (T. Thatcher, DTM Consulting, unpublished data) for all spatial analyses. We examined the longitudinal variation in bank stabilization and side channels in the Yellowstone River to ensure that the selected sites reflected a diverse sample (Figure 2).

We calculated the spatial center of each subsample (“sampling point”) from its GPS coordinates and established its position in the channel network. The spatial scales at which bank stabilization and side channels influenced fish assemblages were unknown. Specifically, we were unsure how far up- and downstream from a sampling point to summarize bank stabilization and side-channel length to capture any potential effect on the fish assemblage. Therefore, we incorporated 12 different channel network distances (“spatial scales”: 0, 50, 100, 200, 400, 800, 1,200, 1,600, 2,000, 2,400, 2,800, and 3,200 m) into our analyses. We used ArcGIS Network Analyst to extract the network of main and side channels for each distance upstream and downstream of each sampling point (Figure 3). We then applied a 200-m lateral buffer to each extracted network and summed the bank stabilization and side-channel bank lengths within each buffer. We chose the range from 0 to 3,200 m based on the assumption that side channels and bank stabilization more than 3,200 m from the sampling point would not have an effect on the fish assemblage at the sampling point.

We standardized the lengths of bank stabilization and side-channel banks for each subsample for each spatial scale by calculating bank-stabilization and side-channel proportions. Each bank-stabilization proportion was calculated by dividing the length of bank stabilization by the main-channel bank length. Each side-channel proportion was calculated by dividing the length of side-channel banks by the length of all banks. The bank-stabilization and side-channel proportions for all subsamples of each gear type within each mesohabitat were averaged and resulted in one bank-stabilization proportion and one side-channel proportion for each combination of mesohabitat, gear type, and spatial scale.

Statistical Analyses

We examined the relationships of individual fish species and fish assemblage structure to the bank-stabilization and side-channel proportions. Catch per unit effort (CPUE) was calculated for each species captured by each gear type. Trammel net, otter trawl, and seine haul CPUEs were calculated by dividing catch by the distance sampled. Electrofishing CPUE was calculated by dividing catch by the time electrofished. Fyke net CPUE was calculated by dividing catch by the set duration. The CPUE for each mesohabitat × gear combination was averaged for each site because our sampling design was balanced at the mesohabitat level. All statistical analyses were conducted in R 3.0.1 (R Development Core Team 2013).

Depth and Velocity Profiling

Each site was surveyed with an Acoustic Doppler Current Profiler (ADCP) during base flow in 2011 to determine whether depths and velocities differed significantly between stabilized and reference pools. Profiles were conducted along evenly spaced transects that varied in number as a function of pool length (<550 m, two transects; 550–850 m, three transects; 850–1,150 m, four transects; 1,150–1,450 m, five transects; and >1,450 m, six transects).

We determined whether main-channel bank stabilization was present within 50 m upstream or downstream of each ADCP transect to locate partially stabilized pools because some “reference” pools had small amounts of bank stabilization. Partially stabilized pools were classified as stabilized because even short lengths of bank stabilization can cause local changes in pool depth or current velocity. We calculated the means and 95th percentiles of depth and velocity of each pool and regressed these against bank stabilization and river segment using ordinary least-squares regression. We used the 95th percentiles of depth and velocity instead of the true maxima to avoid reporting overestimations resulting from potentially spurious ADCP readings. Nevertheless, we refer to the 95th percentile as the “maximum” in this text.

Individual Species Responses

Bank stabilization and side channels influenced the relative abundances of many fish species. Individual species responses to bank stabilization and side channels generally varied according to site geomorphology and sampling method (Figure 4). The responses of Flathead Chub to bank stabilization varied according to sampling method in alluvial sites; the relative abundances of Flathead Chub captured with fyke nets and seines increased with the bank-stabilization proportion, but the relative abundances of those caught by electrofishing and trammel nets did not differ. Only White Suckers captured with electrofishing and trammel nets responded to bank stabilization consistently across site geomorphologies, their relative abundances decreasing with bank stabilization at both alluvial and bluff sites. Goldeye responses to side channels and Goldeye, Shorthead Redhorse, and Longnose Sucker responses to bank stabilization varied according to site geomorphology when captured with electrofishing and trammel nets. Moreover, Goldeye responses to bank stabilization and side channels were opposite at alluvial sites; relative abundances increased with the bank-stabilization proportion and decreased with the side-channel proportion.

Assemblage Subset Responses

Bank stabilization and side channels influenced the fish assemblage structure of many assemblage subsets (Table 2), although the R² values for these model terms indicate that their explanatory power was modest. For example, the terms for site and year generally explained a greater portion of the assemblage structure than the terms for either bank stabilization or side channels (compare the R² values in Table 2). Nevertheless, the eigenanalysis of the NMDSs generally confirmed the perMANOVA results for the bank-stabilization and side-channel terms. The influences of bank stabilization and side channels on the fish assemblage subsets differed, as indicated by the differing and often opposing eigenvector directions for the bank-stabilization and side-channel proportions (Figures 5, S.B.1). Where eigenvector directionalities were generally longitudinally consistent, species CPUE correlations with side channels were more frequent and consistent than those with bank stabilization (Table S.B.1).

Spatial Scale Dependence

Clear patterns in scale dependence emerged from the relative abundance models of single species (Figure 6). Modeling each species at the optimum spatial scale resulted in substantial improvements in adjusted R² values; for example, the adjusted R² for modeling White Sucker relative abundances in bluff habitats (Figure 6, bottom right) improved fourfold (from 0.15 to 0.57) simply as a result of measuring the bank-stabilization and side-channel proportions at appropriate spatial scales. In contrast with the models of assemblage structure, the scale-dependent distributions in explanatory power attributable to bank stabilization and side channels were generally unimodal for the single-species models.

The spatial scales of measurement for bank stabilization and side channels influenced the explanatory power of the models of assemblage structure (Figure 7), but to a lesser degree than in the single-species models. The patterns of scale dependence of explanatory power were bimodal for side channels in the electrofishing subset at alluvial and bluff sites and for bank stabilization in the trammel net subset at alluvial sites. The results for the remaining assemblage subsets were unimodal or lacked a clear pattern. Additionally, although changes in overall R² values across spatial scales were small (Figure 7), these changes are attributable solely to the scale dependence of bank stabilization and side channels; other model variables are not scale dependent. Changes in overall R² can be considered relative to the R² of model terms for bank stabilization and side channels (0.02 to 0.10; Table 2). Thus, a 0.01 increase in overall R² equates to a 10–50% improvement in the R² of bank stabilization and side channels (i.e., 0.01 is 50% of 0.02 and 10% of 0.10).

Depths and Velocities in Stabilized and Reference Pools

Depths differed between reference and stabilized alluvial pools (Figure S.C.1a), but velocities did not (Figure S.C.2a). Stabilized alluvial pools were deeper (mean depths: t_{5, 14} = 2.18, P = 0.047; maximum depths: t_{5, 14} = 3.12, P = 0.008) than reference alluvial pools. The mean depths of stabilized alluvial pools were 0.41 m greater than those of reference pools (95% confidence interval [CI] = 0.01–0.81 m), and the maximum depths were 1.26 m greater (95% CI = 0.39–2.13 m). The mean velocities (linear regression: t_{5, 14} = −0.12, P = 0.907) and maximum velocities (t_{5, 14} = −0.46, P = 0.654) were not different between stabilized and reference alluvial pools.

Neither depths nor velocities differed between reference and stabilized bluff pools (Figures S.C.1b, S.C.2b). The mean depths (t_{3, 8} = 1.93, P = 0.089) and maximum depths (t_{3, 8} = 1.58, P = 0.153) were not different between stabilized and reference bluff pools. The mean velocities (t_{3, 8} = 0.58, P = 0.577) and maximum velocities (t_{3, 8} = 0.20, P = 0.851) were not different between reference and stabilized bluff pools.

Bank Stabilization

Bank stabilization influenced fish habitat (i.e., pool depth) and fish habitat use differently at alluvial and bluff sites. Bank stabilization probably caused the deepening of alluvial pools by attenuating lateral erosion and accelerating bed degradation, whereas bedrock bluffs constrained lateral erosion regardless of bank stabilization. The trammel net subset may have responded directly to such changes in depth, because the assemblage structure of this subset differed according to the bank-stabilization proportion at alluvial but not bluff sites (Table 2). At alluvial sites, the catches of Shovelnose Sturgeons and Channel Catfish were positively correlated with bank stabilization in the electrofishing subset but negatively correlated in the trammel net subset (Figure 5; Supplement B). This apparent contradiction in responses to stabilization may have been caused by fish shifting their habitat use from deepened portions of the channel near the thalweg (which were sampled with trammel nets) to shorelines (which were sampled with electrofishing). For instance, Shovelnose Sturgeons most often use depths less than 2.5 m in the Yellowstone River (Bramblett and White 2001), and Channel Catfish may have exploited riprap for cover or spawning habitat (McMahon and Terrell 1982).

The effects of bank stabilization can be indirect and vary according to pool geomorphology, fish species, and life history stage. Stabilized alluvial pools were deeper than reference pools, which probably resulted in smaller shallow, slow-current velocity (SSCV; sensu Bowen et al. 2003) patches along inside bend margins. These decreases in SSCV patch size reduced suitable habitat for small fish that use shallow habitats for refuge from aquatic predators and slow current velocities to avoid downstream displacement. Thus, the reduction in SSCV patch size may have increased the densities of some small fish, contributing to the differences in the fyke net and seine assemblage subsets between stabilized and reference alluvial pools. For example, the fyke net and seine relative abundances of small Flathead Chub were greater in inside bend and channel crossover shorelines of stabilized alluvial pools than in reference alluvial pools. However, the relative abundances of large Flathead Chub captured using electrofishing and trammel nets did not differ according to stabilization extent (Figure 4a).

Bank stabilization may be detrimental to the persistence of fish, including fishes that were positively associated with bank stabilization during base flow, because bank stabilization can reduce floodplain connectivity (Florsheim et al. 2008), which is critical during high-water periods (Sheaffer and Nickum 1986b; Galat et al. 1998). Although the relative abundances of small Flathead Chub were positively correlated with bank stabilization during base flow, the abundances of Flathead Chub may be limited by seasonally inundated floodplains (Reinhold et al. 2016). For example, Flathead Chub were extirpated from the unimpounded upper Mississippi River concomitant with the anthropogenic disconnection of its floodplain (Barko et al. 2004b). Therefore, if the extent of bank stabilization and floodplain disconnection exceeds threshold levels, fishes that depend on floodplain habitats will probably experience population declines.

There was much overlap between the species compositions of stabilized and reference sites, which was concordant with the results of similar investigations from other large rivers (Barko et al. 2004a; White et al. 2010; Schloesser et al. 2012). Additionally, the relative abundances of some species were higher at stabilized sites than at reference sites, whereas other species exhibited the opposite trend. This variability in individual species responses was consistent with the findings of similar investigations in the Kansas River (White et al. 2010) and the upper Mississippi River (Madejczyk et al. 1998). However, unlike investigations in other large rivers, our study was not confounded by the influences of main-stem impoundment, dredging, channelization, locks and dams, or a largely disconnected floodplain. Such differences complicate direct comparison of our study with many other studies. For example, in our study Shorthead Redhorse used stabilized alluvial sites less than reference alluvial sites, whereas in the Mississippi River Moxostoma species preferentially occupied wing dikes (Madejczyk et al. 1998), which are not present in our study area. However, in the heavily stabilized upper Mississippi River, Moxostoma species may have been concentrated at wing dikes because wing dikes offered some of the only slow-velocity habitat (Barko et al. 2004a, 2004b). In our study area, fish could access slow-velocity habitats in both the main stem and side channels (Bowen et al. 2003; Reinhold et al. 2016).

Side Channels

Consistent with our hypothesis, side channels influenced fish assemblages (Table 2). The catches of more species were positively correlated with side-channel availability than were negatively correlated with it (Figures 5, S.B.1; Table S.B.1). We speculate that one cause of positive correlations is proximal food sources in side channels; side channels had higher densities of zooplankton (Bothar 1981), macroinvertebrates (Eckblad et al. 1984; Sheaffer and Nickum 1986a), and larval fish (Eckblad et al. 1984) than main channels in other large rivers. Moreover, main-channel habitats with connected side channels had higher densities of zooplankton (Bothar 1981) and larval fish (Eckblad et al. 1984; Sheaffer and Nickum 1986b) than main-channel habitats without connected side channels. Positive correlations may also be caused by proximal habitat heterogeneity for spawning and rearing habitat, as the densities of larval (Eckblad et al. 1984) and juvenile fish (Eckblad et al. 1984; Sheaffer and Nickum 1986b) were greater in side channels than in main channels. Thus, proximal side channels may increase local recruitment; in our study area, the total catches of small-bodied fish were higher in river bends with side channels than in river bends without side channels (Reinhold et al. 2016). Proximal food sources may also cause negative correlations if fish concentrate in side channels, where foraging efficiency is maximized. Also potentially contributing to negative correlations is the use of side channels by small fish as refuges from predation by larger fish (sensu Jackson et al. 2001). Although our speculations on the effects of side-channel availability assume movement between side and main channels, the spatial and temporal details of lateral fish movement in the Yellowstone River are not known and merit future study.

Spatial Scale Dependence

The influences of bank stabilization and side channels on fish were spatially scale dependent, but the patterns in this scale dependence were clearer and more robust for the relative abundance models than for the assemblage subset models (Figure 6 versus Figure 7). Perhaps the scale dependence was clearer in the relative abundance models because they are inherently simpler—focusing only on one species—than the assemblage subset models, which incorporate data from several species. Thus, the variability and weak patterns in the assemblage subset models suggest that the species within each subset are responding to bank stabilization and side channels at different scales.

Bank stabilization may increase fish species richness and diversity at finer spatial scales by creating novel bank habitat, but it may also influence fish at coarser spatial scales by altering normal riverine function (Schmetterling et al. 2001; White et al. 2010). Our results broadly support this idea; the bimodal distribution in the scale dependence of some of the assemblage subsets indicated that at least two processes structured those subsets—one at finer scales and another at coarser scales. We speculate that introducing novel substrates (riprap) and increasing pool depth contributed to assemblage structure at finer scales and that channel simplification or land use practices associated with bank stabilization contributed to assemblage structure at coarser scales (Figures 6, 7) .

Concluding Remarks

The influence of bank stabilization on the fish assemblages in our study area depended on the natural riverine template and how bank stabilization altered the physical habitat therein. Fish assemblage subsets responded differently based on longitudinal river position, local geomorphology, and access to side channels. Additionally, the influences of bank stabilization generally differed from, and often opposed, the influences of side channels on the fish assemblage subsets. Therefore, conservation or restoration of side channels may provide an appropriate mitigation strategy for bank stabilization. Although the shifts in the fish assemblage associated with bank stabilization did not result in completely different assemblage structures at stabilized and reference sites, shifts in the fish assemblage were widespread, as bank stabilization was present in many lower Yellowstone River reaches (Figure 2; Boyd and Thatcher 2004). Moreover, we suspect that the present fish assemblage reflects the incremental changes in fish habitat that have resulted from the installation of bank-stabilization structures and the loss of side channels concomitant with increased land use and the development of the Yellowstone River basin.