Conuences and Land Cover As Agents of Change: Temporal Habitat Variability Modies the Movement and Assemblage Change of Headwater Fishes

Interspersed inputs of wood and sediment brings about morphological change at conuences and the extent to which these processes are modied by anthropogenic disturbance has ramications for stream sh assemblages. In this study, we use three functional groups of headwater shes to assess the inuences of conuence size and land cover on habitat stability, distance moved, movement rate, and assemblage change in a Gulf Coastal Plain drainage in the southeastern United States using a 2X2 design. Our results suggest that differences in habitat stability were described by a hydrogeomorphic gradient, and urban reaches characterized by a conuence size > 0.6 displayed the greatest habitat instability. Water-column specialists in urban reaches were more likely to move when habitat change was limited, whereas movement by this functional group in forested reaches was more likely in response to greater habitat instability. Therefore, the distance moved by water-column specialists was related to land cover. Assemblage change also occurred at a more constant rate in urban reaches in response to habitat instability. There was little evidence that conuence size inuenced movement or assemblage stability in these headwater assemblages. Our understanding of the extent to which land cover alters the geomorphic and ecological gradients associated with headwater conuences will be critical to ensure the conservation of sensitive species whose tness is dependent on the integrity of these habitats. In this study, we assessed the inuences of conuence size and land cover on habitat stability, the movement ecology of a functional guild, and on temporal assemblage change at four headwater conuences in a Gulf Coastal Plain dendritic river system. Hydrogeormorphic processes altered components of reach scale habitat stability, inuencing the rate of movement of water-column specialists, and promoting assemblage change at these four headwater conuences. Our analyses yielded ve primary ndings: (1) a hydrogeomorphic gradient (PC2) associated with differences in land cover altered habitat stability (2) along this gradient, urban reaches with a conuence size > 0.6 (i.e., Mixon’s Creek reaches) exhibited the greatest habitat instability, (3) differences in habitat complexity modied the PDMR of water column-specialists at the reach scale, (4) water column-specialists were more likely to make short or far moves as a consequence of upstream land cover and (5) absolute change in components of channel morphology (e.g., depth, bankfull width) regulated within-reach stream sh assemblage change. Our results highlight the inuence of conuence size and land cover on differences in habitat stability and assemblage change, and the movement rate of freshwater shes in headwaters. Because the movement and assemblage datasets used for this study were small (N = 51, N = 63); replication was low (9–18), reducing the power of analyses.


Introduction
Movement is integral to species survival because it in uences colonization, gene ow, and resource use. The movement of a species is a function of its probability of transition between habitat patches (Levins 1966;Wiens 2002). Habitat fragmentation limits the movement of species by disrupting the connectivity and structure of habitat patches (Brown et al. 2001;Fagan 2002). Because organisms move to complete their life histories, even a small degree of fragmentation across an ecosystem can be detrimental to their survival, reproduction and growth. In dendritic river systems (Campbell- Grant et al. 2007), random habitat fragmentation increases the number of patch fragments, variance in patch size (Fagan 2002) and distance between patches, thus diminishing movement of species (Roberts and Angermeier 2007).
Con uences are areas of active geomorphic activity (Rice et al. 2001; Benda et al. 2004). Tributary junctions result in increases in the supply of water, sediment, and wood to the mainstem, thus mainstem channel responses should be greater in magnitude close to or directly downstream of con uences (Benda et al. 2004). Channel widening and declines in stream gradient often occur upstream of con uences as a consequence of greater wood and sediment storage, thus facilitating an increase in the magnitude of ow-related disturbances (Benda and Dune 1997). The magnitude of geomorphic change at con uences in headwater drainages is characterized by high ow events, which transport sediment and wood (Johnson and Rodine, 1984;Hogan et al. 1998). High ow events scour sediment and transport wood downstream, forming depositional areas such as gravel bars and alluvial fans (Benda and Cundy 1990; Benda et al. 2003). Sediment deposition at con uences induces predictable, localized, geomorphic responses in mainstem channels (Benda et al. 2004). For example, decreasing sediment transport at con uences should facilitate reductions in upstream channel gradient and substrate size, while increasing channel meandering and oodplain width in the mainstem. Such changes are counterbalanced in the downstream reach of the mainstem with increases to channel gradient, channel width, substrate size, pool depth, and bar occurrence. Because the supply of sediment is spatially interspersed, depositional areas at con uences will expand and contract in response to temporal variation in the hydrology (Benda et al. 2003). Thus, the upstream and downstream spatial extent of these habitats' in uence on con uences should uctuate with time.
Con uence size (i.e., the ratio between tributary size and mainstem size) and network geometry contribute substantially to the extent of geomorphology at con uences. Discharge-related morphological changes (i.e., channel width, depth), which scale to the size of the tributary in relation to the mainstem, occur at con uences where the ratio between tributary size and main stem size approaches 0.6 to 0.7 (Rhoads 1987). Local network geometry describes the angle at which tributaries intersect the mainstem. Kilometer-scale variation of tributary effects in the mainstem can be described by local network geometry (Benda et al. 2004). In headwaters, Benda and Cundy (1990) demonstrated how con uence angles may regulate variability of depositional events, where con uence angles greater than 70º should engender high ow deposition, whereas deposition would be less likely at con uences with more acute angles.
Heightened habitat heterogeneity may encourage landscape complementation or supplementation by stream shes at con uences because patches contain non-substitutable resources (food, spawning habitat) which are in close proximity to one another (Dunning et al. 1992). Thus, temporal assemblage turnover at con uences is often reduced ( For example, Cathcart et al. (2015) noted that sh movement patterns at a desert con uence on a large river varied substantially by species as a consequence of sensitivity to seasonal variation, and variability in the hydrologic regime.
Urban and agricultural land cover fragments streams, and habitat fragmentation at small spatiotemporal scales within these systems is intimately tied to land cover at coarser spatiotemporal scales (Wang et al. 2001;Allan 2004;Leal et al. 2016). Thus, broad anthropogenic impacts reduce the connectivity, stability, and diversity of habitat patches at smaller scales (Padgham and Webb 2010). Anthropogenic disturbance disrupts gene ow (Stow et al. 2001) while also degrading regional emergent properties such as species richness (Perkins and Gido 2012) by the elimination of movement pathways. However, the extent to which land cover modi es the role of con uences as agents of habitat change, and therefore, altering stream sh movement and assemblage change, is poorly understood. patterns of movement in stream shes. However, it is unclear whether alterations to these abiotic factors as a consequence of land cover and con uence size would yield similar effects on stream sh movement.
In this study, we examine habitat structure, movement, and stream sh assemblages at four headwater con uences that differed in size and surrounding land cover. We used mark-recapture methods to assess movement rate and assemblage change at the reach scale. We applied an information-criterion approach to test which habitat characteristics best described movement and assemblage change across con uence size or land cover factors. We predicted that 1) urban reaches characterized by a con uence size > 0.6 would exhibit the greatest habitat instability, 2) the distance moved by movers would vary as a consequence of land cover and con uence size, 3) movement rate and assemblage change would be associated with distinct reach scale components of habitat stability mediated by land cover and con uence size, and 4) stream shes in urban reaches will exhibit a more rapid response (e.g., movement, assemblage change) to habitat instability.

Study System and Design
We conducted our study in the Pascagoula River drainage, which is positioned within the Southern Pine Plains and Hills ecoregion in the Gulf Coastal Plain province, and is the largest unimpounded river system in the contiguous United States (Dynesius & Nilsson, 1994;Hupp, 2000). Streams that drain this region tend to be either alluvial or blackwater systems and are characterized by low gradients and gravel, sand and clay substrates (Hupp, 2000).
Each of our four sites was located at the con uence of two headwater streams. Of the two upstream reaches at each site, we identi ed the reach with the larger upstream drainage area as the mainstem (Fig. 1A). We then delineated three contiguous 100 m reaches containing alternating ri e and pool patches. We established two factor levels for land cover (urban, forested) and con uence size (> 0.60, < 0.60) for each of the four sites. Con uence size ratios (CSR) are calculated by dividing the drainage area of the smaller stream by the drainage area of the mainstem; a threshold of > 0.6 represents the lowest value at which consistent geomorphic change may be observed because of con uence size (Benda et al. 2004). Two sites (Garraway and Mixon's Creeks) had large (> 0.6) CSR and two (Priest and Sweetwater Creeks) had small CSR. We measured CSR for each site using the National Hydrography Data Plus (NHDPlus) (USEPA and USGS 2016) set in ArcGIS (version 10.0, ESRI). Sites were only chosen if their con uence angle was < 60º to diminish local geomorphic effects that were a symptom of network geometry. The 2016 national land cover dataset (NLCD, https://www.mrlc.gov/data) was used to estimate the proportions of urban and forested land cover within each site's upstream catchment. Our urban sites (Priest and Mixon's Creeks, > 50% urban land cover) were located within the city limits of Hattiesburg, MS, and our forested sites (Garraway and Sweetwater Creeks, 100% forested land cover) were located within De Soto National Forest (Fig. 1B).

Study Period
Initially, we marked sh in August of 2017 and May of 2018; subsequent mark-recapture events occurred at 4 to 6 week intervals (7 marking events total) and repeated visits to reaches within a site were treated as replicates. We attempted to conduct six recapture events at all four sites across two eld seasons; two in the fall of 2017 and four in the springsummer of 2018. However, ooding in the fall of 2017, resulted in a bridge failure that restricted access to one site. We replaced this site (small CSR, forested land cover) with a new site prior to beginning sampling in the summer of 2018.
Thus, we conducted six recapture events at three sites, but only four recapture events were performed at the fourth site.
Preliminary analysis suggested that there was not a signi cant effect of site, despite the unbalanced design (Table S1).

Mark-Recapture Methods
At each patch we made three passes using a 4.8 mm mesh seine and a ETS (Electro shing Systems) Badger 1 backpack electro shing unit using pulsed DC (range: 100 to 250 V). All darters > 40 mm standard length (SL), all sun sh > 80 mm in SL, and all shiners > 40 mm in SL were marked. We gave all sh patch, batch-speci c marks using visual implant elastomer (VIE; Northwest Marine Technology, Inc., Shaw Island, WA, USA). Small bodied minnows (Bangs et al. 2013;Neufeld et al. 2015), darters (Roberts and Angermeier 2004), and centrarchids (Laux et al. 2007) exhibit high retention of VIE tags over extended periods. We used three colors and six body locations to produce patch + visit marks.
Prior to marking, we anesthetized sh with tricaine methanesulfonate (range = 60-100 mg/L). We marked sh to identify their initial patch of capture. During consecutive mark-recapture events, any sh that had moved received an additional mark speci c to the new patch occupied, and all sh were released back to where they were captured. After marking, we held sh for 30 minutes to allow for recovery from anesthesia. Following recovery, we released sh in low velocity microhabitats near the midpoint of the patch from which they were sampled. We used the midpoint distance of the original patch of capture and patch of recapture to establish the beginning and end points of the total distance moved by a sh for all pairs of patches. Because recapture rates of benthic specialists and structure specialists were variable among sites (Table 1), we only analyzed differences in distance moved and movement rate of water-column specialists. Our rst recapture event of the 2018 eld season allowed us to estimate the percentage of shes recaptured between our two eld seasons. At the end of the 2018 eld season, we performed broader sampling (1-2 river km) for each site to identify any individuals that may have moved outside of the study-reaches. Habitat Sampling Following each sampling event, we collected habitat data along 10 evenly spaced transects perpendicular to ow within each of the three reaches at a site. At one meter intervals along each transect we measured depth (m), current velocity (m·s ¹), substrate size (modi ed Wentworth scale, six categories; Cummins, 1962), and the presence (binary variable) of woody structure. We also measured the wetted and bankfull widths (m) for each transect. We calculated mean, absolute reach-scale differences to quantify the degree of change in each habitat variable between sampling events. Reach scale means for each habitat variable were estimated for all sampling trips, and then absolute differences between consecutive trips were calculated for each variable. We then divided the absolute mean differences by the number of days between sampling events to adjust for differences in time. These absolute mean differences were then regarded as reach scale components of habitat stability and incorporated into our models of movement rate and assemblage change.

Analysis of Habitat Stability
We used principal components analysis (PCA) to reduce the dimensionality of our reach scale components of habitat stability. We retained PCA scores for the rst two axes (PC1 and PC2) as descriptors of habitat stability. To test our prediction that urban reaches characterized by a CSR > 0.6 (i.e., Mixon's Creek reaches) would exhibit the greatest habitat instability, we performed one-way ANOVAs on the reach PC1 and PC2 scores with stream as a main effect. We then corrected for multiple comparisons using pairwise t-tests and the Bonferroni method.

Analysis of Distance Moved
We examined differences in movement as a consequence of distance moved, land cover and CSR. We treated distance as a factor with three levels (0 to 30 m = short; 30 to 60 m = intermediate; >60 m = far) and summed the number of movers within each group for each site (n = 11). Movers were pooled into groups regardless of directionality (i.e., upstream and downstream moves were pooled). We used Poisson regression to test our hypotheses that the number of movers would vary as a consequence of distance travelled, land cover or con uence size. Because samples within and among groups were uneven, site was included as a random effect. We corrected for multiple comparisons using Tukey's HSD test.

Estimation of Movement Rates and Assemblage Change
We estimated proportional daily movement rates (PDMR) and Bray-Curtis distances of dissimilarity as metrics of movement rate and assemblage change for reaches between sampling events. Sample sizes for estimates of movement (N = 54) and assemblage change (N = 63) differed because we did not recapture any individuals in some reaches during a sampling event. Thus, for a given sampling event, reaches in which we did not recapture any individuals were subsequently dropped from our movement dataset prior to any analyses. We estimated PDMR using the formula M *R − ¹ D − 1 , where M is the number of sh that moved, R is the total number of recaptures within a reach, and D is the number of days since the rst marking period of the eld season (Warren and Pardew 1998). We estimated PDMR separately for 2017 and 2018 movement data because low recapture rates (1.6-3.9%, see Table 1) suggested that mark loss, emigration, or mortality (electro shing or natural) were high between years. To estimate assemblage change, we calculated the relative abundance of each functional group within all reaches during each sampling event and then estimated reach scale, Bray-Curtis distances of dissimilarity between sampling events and across all functional groups.

Analysis of Reach Scale, Movement Rates and Assemblage Change
We modeled movement and assemblage change using Beta-regression and Zero-In ated Beta regression models respectively, because, preliminary analyses suggested that movement rates and Bray-Curtis distances were not normally distributed. As an expansion of generalized linear models, Beta regression models can evaluate the effects of explanatory variables for proportional and probability data falling within the interval (0, 1) by modeling zeros in a dataset as a function of a two-state process (Liu and Eugenio 2018). One of the two states, the zero state, may be de ned as the probability of an event being so low that it cannot be readily differentiated from zero. The second state, the normal state, includes both zeros and continuous values falling within the interval (0, 1) (Liu and Eugenio 2018). We assessed whether the addition of "site" as a random intercept term improved model t. For both response variables, random-intercept, null models were ranked higher than null models (PDMR: w = 0.76; Bray-Curtis Distance: w = 0.74) justifying the inclusion of a random intercept term. All models converged. Prior to modeling, we tested for the correlation between covariates. Any two covariates that had a Pearson correlation greater than the absolute value of 0.6, were not included in the same model. We standardized all habitat variables by subtracting the mean and dividing by twice the standard deviation.
We constructed both reach and multi-scale models (Table S2). We included six predictors in our reach scale models to identify the abiotic components of habitat stability that best explained variation in movement rates and Bray-Curtis distances. In total, we constructed 23 models to test our hypotheses on movement rate and assemblage change at the reach scale. Multi-scale model sets were con gured so that both coarse scale factors, CSR and land cover, could be integrated into our best reach scale models as both additive and interactive effects. Our multi-scale model sets also featured the best reach scale model without land cover or CSR included as additive or interactive effects. Finally, within our multi-scale model sets, both response variables were modeled directly as a consequence of the singular effects of either land cover or CSR or as a consequence of the interaction between these two factors. We included a null model and global model in both model sets. For model sets, we only incorporated two-way interactions. We used Akaike's Information Criterion for small sample sizes (AIC C ) to assess the quality of competing models (Burnham and Anderson 2002). We only used reach scale models with w > 0.10 to construct multi-scale models.

Recapture rates
Over the duration of the study period, we marked 2459 sh; the total sum of sh marked varied between sites  (Table 1). Proportionally, across all sites, 61% (152 individuals), of water-column specialists were recaptured in pools, 63% (37 individuals) of benthic specialists were recaptured in ri es, and 93.6% (103 individuals) of structure specialists were recaptured in pools.

Habitat stability
The rst two components from the PCA explained 50% of the variance in reach scale components of habitat stability across time (Fig. S1, Table S3).

Modeling of PDMR
For our reach scale model set, the best model (w = 0.30, pseudo R² = 0.18) explained the variation in PDMR of watercolumn specialists as the outcome of a signi cant, negative interaction (beta = -1.57 ± 0.50, p = 0.003) between absolute mean difference in current velocity and absolute mean difference in woody structure (Table 3). Our multi-scale modeling results suggested the addition of land cover or CSR did not increase model t. Table 3 Top reach and multi scale models of assemblage change (Bray-Curtis Distances) and proportional daily movement rate (PDMR) of water-column specialists sampled at four headwater con uences in the Pascagoula River basin, MS, USA.
Values are shown for the number of parameters (K), AIC C , ∆AIC C and model weights (w i ). Intercept-only (i.e., null) models are designated by periods in place of covariates Creek reaches rapidly increased in relation to changes in current velocity (Fig. 3a). Oppositely, for reaches in which the absolute change in woody structure availability was large (i.e., > 0.001% change), the PDMR of water-column specialists generally appeared to decline within Garraway Creek and Sweetwater Creek reaches, but stayed relatively constant within Priest Creek reaches (Fig. 3b).

Modeling of Bray-Curtis Distances
Two models of Bray-Curtis distance were associated with w > 0.10. The best model had moderate support (w = 0.31) and suggested that reach scale Bray-Curtis distances were best explained by a signi cant interaction (beta = 0.24 ± 0.07, p = 0.003) between absolute mean difference in depth and absolute mean difference in bankfull width (Table 3). Multiscale modeling results suggested the addition of CSR did increase model t, though the increase in t was nonsigni cant (beta = -0.16 ± 0.12, p > 0.05). The second best model was weakly supported (w = 0.23), and suggested that reach scale, Bray-Curtis distances were best explained by a signi cant, interaction (beta = 0.16 ± 0.06, p = 0.007) between absolute mean difference in depth and absolute mean difference in woody structure (Table 3). Multi-scale modeling results suggested the addition of CSR in this model did increase model t (Table 3), though the increase in t was nonsigni cant (beta = -0.13 ± 0.12, p > 0.05).
The plotting of predicted values suggests that, for reaches in which the absolute change in bankfull width was small (i.e., approximately < 3% change), reach scale Bray-Curtis distances generally declined within Garraway Creek and Sweetwater Creek reaches (Fig. 4a), while remaining static within Priest Creek reaches. Oppositely, for reaches in which the absolute change in depth was large (i.e., > 3% change), reach scale Bray-Curtis distances generally appeared to increase within Priest Creek and Garraway Creek reaches, remained fairly constant in Mixon's Creek reaches, and declined within Sweetwater Creek reaches (Fig. 4b). Despite the inclusion of con uence size in the best multi-scale model, plots of predicted values do not support any effect of this predictor on assemblage change.

Discussion
In this study, we assessed the in uences of con uence size and land cover on habitat stability, the movement ecology of a functional guild, and on temporal assemblage change at four headwater con uences in a Gulf Coastal Plain dendritic river system. Hydrogeormorphic processes altered components of reach scale habitat stability, in uencing the rate of movement of water-column specialists, and promoting assemblage change at these four headwater con uences. Our analyses yielded ve primary ndings: (1) a hydrogeomorphic gradient (PC2) associated with differences in land cover altered habitat stability (2) along this gradient, urban reaches with a con uence size > 0.6 (i.e., Mixon's Creek reaches) exhibited the greatest habitat instability, (3) differences in habitat complexity modi ed the PDMR of water columnspecialists at the reach scale, (4) water column-specialists were more likely to make short or far moves as a consequence of upstream land cover and (5) absolute change in components of channel morphology (e.g., depth, bankfull width) regulated within-reach stream sh assemblage change. Our results highlight the in uence of con uence size and land cover on differences in habitat stability and assemblage change, and the movement rate of freshwater shes in headwaters. Because the movement and assemblage datasets used for this study were small (N = 51, N = 63); replication was low (9-18), reducing the power of analyses.

Effect of land cover on habitat stability
Our results indicated that land cover facilitated habitat change among all reaches more so than con uence size.
Differences in habitat change between sites was associated with differences in the magnitude of geomorphic variation (e.g., changes in bankfull width, current velocity, Fig. S1, Table S3). Priest Creek and Mixon's Creek reaches were characterized by greater bankfull widths, greater variation in wetted width, diminished availability of woody structure, shallower depths, and greater variance in current velocity (Table S4). These differences were reduced in Garraway Creek and Sweetwater Creek reaches. Our results also suggested that reach-scale Priest Creek habitat was not more stable than reach-scale Mixon's Creek habitat, or that that reach-scale Sweetwater Creek habitat was not more stable than reach-scale Garraway Creek habitat, reducing the probability that habitat stability was in uenced by CSR. In small, coastal plain streams, limited impervious cover (< 11%) may increase ashiness, and limit the in uence of base ow on Our results indicate that there was little variation in mean sediment size during the study period across all sites, but urban drainages were typi ed by greater variability in active channel width (Table S5). Utz and Hildebrand (2011) suggested that the stability of sediment structure in small, urban Coastal Plain streams may be related to bed material. Coarse substrate particles, which are often in great abundance in Piedmont and montane drainages, are rarely present in Coastal Plain streams.

Effect of con uence size on habitat stability
Our prediction that urban reaches characterized by a con uence size > 0.6 would exhibit the greatest amount of habitat instability was supported. In relation to other sites, Mixon's Creek reaches were associated with much greater variability in bankfull width, wetted width, and substrate size (Table S4). Furthermore, mean percent occurrence of woody structure at this site was effectively zero for the duration of the study period for all study reaches (Table S4).
Excluding trends associated with depth, reach scale habitat parameter means for Garraway Creek and Mixon's Creek (i.e., our two sites with con uence sizes > 0.6) are in disagreement with the predictions of Benda et al. (2004). Annual mean depth for mainstem downstream reaches at both sites was consistently greater than the mean depth of mainstem upstream reaches at both sites. However, mean bankfull width was greater in the mainstem downstream reaches at both sites, and generally, substrate size was greater in the mainstem upstream reach at both sites. Finally, in opposition to the predictions of Benda et al. (2004), the percent occurrence of woody structure was generally greater in mainstem downstream reaches at both sites.
Effect of habitat stability on the movement rate of water-column specialists Differences in the availability of woody structure altered the size of the effect of hydrologic (e.g., current velocity) variation on movement rates of water-column specialists. Overall, Garraway Creek and Sweetwater Creek reaches were characterized by a much higher percentage of woody structure throughout the study period than Priest Creek and Mixon's Creek reaches (Table S4). Our prediction that stream shes would exhibit greater movement as a consequence of habitat stability in urban reaches was supported. For reaches in which absolute change in woody structure availability was small (approximately 0.001% absolute change), in relation to growing absolute change in current velocity (Fig. 3a), movement rates of water-column specialists appeared to increase within and among Mixon's Creek and Priest Creek reaches. Conversely, for reaches in which absolute change in woody structure availability was large (> 0.001% absolute change), movement rates seemed to decline in response to cumulative absolute changes in current velocity. Movement rates seemed to decline steeply within and among Garraway Creek and Sweetwater Creek reaches, but decreased moderately within and among Priest Creek reaches (Fig. 3b). Because

Effect of land cover on distances travelled by water-column specialists
Our nding that land cover appears to have more strongly in uenced the distances travelled by water-column specialists adds further credence to our argument that the movement rate of this functional group was more pronounced in urban reaches due to reduced habitat complexity and greater variability in channel morphology. Longer distance moves were more often detected within among Mixon's Creek and Priest Creek reaches as opposed to Sweetwater and Garraway Creek reaches. Thus, it is not surprising that 1) the greatest observed distance moved by a water-column specialist (113 m) and benthic specialist (97 m) were both recorded within Mixon's Creek reaches, and 2) all observed moves > 90 m occurred within Priest Creek (n = 1) or Mixon's Creek (n = 3) reaches. Because we did not recapture any sun sh within Garraway Creek or Sweetwater Creek reaches, we do not report a maximum distance travelled by an individual within this functional group.
Effect of habitat stability on assemblage change The temporal and spatial scales at which eld studies are implemented often makes disentangling the relationship between assemblage change and habitat stability di cult. Despite this, we did detect some patterns that we believe are worth highlighting. Despite the inclusion of CSR in the best multi-scale model of assemblage change, plots of predicted values do not support a strong con uence size effect. For reaches in which changes in bankfull width were minimized (.e.g., Fig. 4a), assemblages in Sweetwater Creek and Garraway Creek reaches were more likely to remain stable through time. In opposition, assemblage change among Priest Creek and Mixon's Creek reaches remained fairly constant (approximately 0.007-0.0085, Fig. 4a). Therefore, our prediction that stream sh assemblages would exhibit a more rapid response in urban reaches was supported.  (Table S5). For reaches in which absolute change in bankfull width was large, assemblage differences were more likely to decline in Sweetwater Creek reaches as opposed, to Priest Creek and Mixon's Creek reaches (Fig. 4b). In opposition, assemblage differences were ampli ed in Garraway Creek reaches which were subject to an increase in absolute change in bankfull width (Fig. 5B). The dichotomy in observed differences between these two sites in relation to greater variation in channel morphology is interesting and deserves further research; con uence size may have contributed to these discrepancies in assemblage change.

Conclusions
We assumed that there would be an interaction between con uence size and land cover, and thus, urban reaches with a larger con uence size would exhibit the greatest variability in habitat and movement. Geomorphic characteristics were most variable at Mixon's Creek, subjecting water-column specialists to greater geomorphic change. Thus individuals in these reaches were more likely to move greater distances. However, there was no statistical evidence that land cover or con uence size contributed to substantial variation in movement rate or assemblage change.
We used a 2 X 2 design to assess the effects of con uence size and land cover on stream sh movement and knowledge have attempted to connect local geomorphic change at con uences to variation in reach scale movement in a natural setting. Thus our study provides baseline data for future studies further investigating the relationship between con uences and land cover on movement and assemblage change at the reach scale. A key weakness of our study was our reduced sample sizes; however, to address our stated objectives, a smaller number of sites was deemed necessary to effectively deploy a movement study using mark-recaptures methods. Future studies should consider the use of other tagging methods (e.g., passive integrated transponders; acoustic tags) to increase the number of replicates (e.g., marking of individuals), but also allowing for long-term monitoring to better understand how con uences and land cover may in uence movement and assemblage change in headwaters.

Declarations
Competing Interests The authors declare there are no competing interests.

Con ict of Interest Statement
The authors have no con ict of interest regarding the submission of this manuscript.

Funding Statement
Funding was provided by the Department of Agriculture (USDA) US Forest Service and The University of Southern Mississippi Department of Biological Sciences.

Availability of Data
The data that support the ndings of this study are available from FigShare    Frequency distributions for three movement categories, regardless of directionality (i.e., upstream or downstream) in relation to distance moved for each mark-recapture site in the Pascagoula River basin, MS.

Figure 3
Plot of predicted values from zero-in ated Beta regression as a result of an interaction between absolute mean differences in woody structure and current velocity on the proportional daily movement rate (PDMR) of water-column specialists. To ease interpretation of our results, we split our dataset in half, and converted woody structure into a factor with two levels (a: small change = ≤ 0.001%; b: large change = > 0.001%) to identify differences in PDMR as a consequence of the interaction between the two habitat variables. The number of plotted points along each regression represents the number of reaches associated with that level of change in woody structure for each site. *No predicted estimates of PDMR for Mixon's Creek were associated with a large change in woody structure. Dashed boxes represent 95% con dence limits for each regression Figure 4 Plot of predicted values from zero-in ated Beta regression as a result of an interaction between absolute mean differences in bankfull width and depth on Bray-Curtis distances. To ease interpretation of our results, we split our dataset in half, and converted bankfull width into a categorical factor with two levels (a: small change = ≤ 3%; b: large change = > 3%). to identify differences in Bray-Curtis distance as a consequence of the interaction between the two habitat variables.