Impacts of highway crossings on density of brook charr in streams

Authors

  • Marc Pépino,

    1. Centre de recherche sur les interactions bassins versants –écosystèmes aquatiques (RIVE) and Groupe de recherche interuniversitaire en limnologie et en environnement aquatique (GRIL), Université du Québec à Trois-Rivières, 3351 boul. des Forges, Trois-Rivières, Québec, G9A 5H7, Canada
    Search for more papers by this author
  • Marco A. Rodríguez,

    Corresponding author
      Correspondence author. E-mail: marco.rodriguez@uqtr.ca
    Search for more papers by this author
  • Pierre Magnan

    1. Centre de recherche sur les interactions bassins versants –écosystèmes aquatiques (RIVE) and Groupe de recherche interuniversitaire en limnologie et en environnement aquatique (GRIL), Université du Québec à Trois-Rivières, 3351 boul. des Forges, Trois-Rivières, Québec, G9A 5H7, Canada
    Search for more papers by this author

Correspondence author. E-mail: marco.rodriguez@uqtr.ca

Summary

1. Habitat degradation and fragmentation are growing concerns in ecology, yet distinguishing between the effects of these processes can be a challenging task. In riverine landscapes, roads impact fish populations mainly through: (i) restriction of passage of individuals (fragmentation) and (ii) reduction in habitat quality downstream by increases in sediment load (habitat degradation). These two processes can be differentiated during road construction projects.

2. This study examines the impacts of a highway expansion and the presence of existing highway crossings on local population density of brook charr Salvelinus fontinalis in stream reaches traversed by the highway. Density was estimated on three consecutive summers in 212 sections distributed among 36 streams. This extensive sampling design focused on replicated comparison of sites upstream and downstream of highway crossings and included different stages of highway construction (before, during and after) and types of highway crossing (low, intermediate and high passability). Mixed models were used to examine the impacts of the highway on population density.

3. Highway construction activities had no detectable effect on density. However, existing highway crossings appeared to have a strong effect on density, which differed markedly between upstream and downstream sites near highway crossings of intermediate and low passabilities.

4. A Markovian movement model yielded estimates of passability that were consistent with the classification of crossing types and provided evidence for the restriction of upstream movement at crossings as a plausible explanation for observed differences in local population density.

5.Synthesis and applications. Habitat fragmentation resulting from restriction of passage at highway crossings had markedly greater effects on local population density than short-term impacts arising from construction activities. The modelling approaches used in this study can be useful management tools for the conservation of mobile fish species in fragmented riverine landscapes.

Introduction

The reduction in connectivity caused by habitat fragmentation alters population dynamics and threatens the persistence of animal species (Fagan 2002; Fausch et al. 2002). Fragmentation per se, the breaking apart of formerly continuous habitat, must be distinguished from habitat loss, which reduces the amount of habitat remaining on the landscape; studies that confound these two processes can produce misleading results (Fahrig 2003).

Although distinguishing fragmentation from habitat loss is often difficult in practice, habitat loss is viewed as a more important driver of species distribution patterns than fragmentation in terrestrial landscapes (Fahrig 2003; Mortelliti et al. 2011). In riverine landscapes, these two processes can be differentiated during road construction projects. Roads are one of the main sources of anthropogenic fragmentation; they are associated with losses of both connectivity and habitat extent and can have major ecological impacts (Trombulak & Frissell 2000; Allan 2004; Wheeler, Angermeier & Rosenberger 2005). Some effects of roads such as alteration of physical and chemical environments, reduction in habitat connectivity and increased mortality for animal species occur both in aquatic and terrestrial ecosystems. However, effects such as mortality caused by sediments and pollutants can be spatially asymmetrical in riverine landscapes because they occur predominantly downstream of the road (Fig. S1, Supporting Information). Sedimentation (particularly during road construction, habitat degradation or loss) and obstructions to fish passage (road-crossing structures, fragmentation) are the two main environmental impacts leading to differences in fish density and distribution near roads (Wheeler, Angermeier & Rosenberger 2005). In rivers and streams, fish are subject to a range of impacts resulting from increased sediment loading from anthropogenic activity (Waters 1995; Kemp et al. 2011). Reduction in survival to emergence through the accumulation of fine sediments in the substrate has been well documented for fish species (reviewed in Kemp et al. 2011), including salmonids (Jensen et al. 2009; Guillemette et al. 2011). By reducing the abundance and availability of benthic invertebrates, fine sediment transport and deposition can also reduce the summer growth and survival of fish (Suttle et al. 2004; Harvey & Railsback 2009). Fish densities downstream of a road may therefore be reduced as a consequence of fine sediment loading. The presence of road crossings may also reduce the connectivity of the riverine landscape and thus be detrimental to viability of fish populations (Schick & Lindley 2007; Bouska & Paukert 2010). Fish may have difficulties moving through instream structures at road crossings during low summer flow (Warren & Pardew 1998) or spawning migrations (Belford & Gould 1989), and these barrier effects are often related to the type of structure built at the crossing (Warren & Pardew 1998).

The highway between Quebec and Saguenay cities (Quebec, Canada) was built in 1948. A major construction project, undertaken during the period 2006–2012, widened the highway from two to four traffic lanes. The mean width of the highway’s ‘zone of influence’, which includes the traffic lanes and all areas required for road security and maintenance, such as ditches and additional strips of land, increased from 30 to 120 m. The measures implemented to reduce sedimentation during construction included the use of erosion control mattresses to stabilise stream banks, gravel filters placed along construction ditches to promote sedimentation and geotextile curtains to limit dispersal of suspended particulate matter. Brook charr Salvelinus fontinalis, which is found in almost all of the streams crossed by the highway, can be affected both by sediment loading (Guillemette et al. 2011) and facilities for fish passage at road crossings (Belford & Gould 1989). This system therefore provides a suitable model for evaluating the influence of habitat degradation and fragmentation on brook charr populations.

Environmental impacts often occur at different spatio-temporal scales, creating challenges for study design and statistical analysis (Stewart-Oaten & Bence 2001). One of the main difficulties is to isolate human-made disturbance from natural variation of the studied phenomenon (Walters, Collie & Webb 1988). The before-after-control-impact (BACI) approach and its derivates have been the most commonly used sampling designs in environmental impact assessments (Stewart-Oaten & Bence 2001). However, BACI designs have been criticised for lacking replication and not discriminating effectively between variation within and between control and treatment units (Murtaugh 2000), and for not taking into account the correlation structure of nested observations (McDonald, Erickson & McDonald 2000). When more than a single impacted site is available for inclusion in the study, multiple BACI (MBACI) designs, which compare a group of impacted sites to a group of control sites, can be used to address the issue of replication (Downes et al. 2002; Angeler & Moreno 2007).

The present study aims to investigate the potential impacts of highway construction (habitat degradation through sedimentation) and the presence of existing highway crossings (fragmentation through barrier effects) on brook charr density. Our analysis is based on spatially extensive sampling, over a 3-year period, of brook charr density at sites upstream and downstream of the highway. The major strengths of our design are that (i) treatment temporal trajectories have adequate replication; (ii) the initiation of construction treatments is staggered in time (the ‘staircase’ design of Walters, Collie & Webb 1988), thus controlling for time–treatment interactions; (iii) intra-group correlations arising from nested sampling are accounted for by random effects in mixed model analyses.

Specifically, we tested whether differences between local population densities at upstream and downstream sites were affected by two treatments: stage of highway construction and type of highway crossing. We tested the hypotheses that (i) construction effects operating over the time scale covered by the study (3 years or shorter) cause differences in density to increase following the initiation of construction; (ii) if existing highway crossings act as long-term barriers to passage, differences in density will be linked to structural features that determine passability of the crossing. We also developed a Markovian random walk model allowing for differences in barrier passability and directional bias in fish movements and examined its behaviour at equilibrium to examine whether restriction of passage at crossings can explain observed differences in local density.

Materials and methods

Study system and data collection

The study was conducted along a 157-km stretch of the highway in the Jacques-Cartier National Park and the Laurentides Wildlife Reserve, located on the Laurentian Plateau at altitudes between 190 and 820 m (Fig. 1). The area has a humid continental climate with harsh winters (mean annual temperature: 0·3 °C; annual snowfall: 639 cm). Vegetation cover is continuous boreal forest dominated by balsam fir Abies balsamea and white birch Betula papyrifera. Watershed geology is largely homogeneous and consists of a metamorphic basement (gneiss) with intrusive rocks (mainly mangerites); stream formation is primarily by glacial deposits and outwash. The annual discharge regime has a dominant peak in the spring at snowmelt and seasonal lows in late summer. The highway is the only conspicuous source of human disturbance on the aquatic environment.

Figure 1.

 Map showing the position of the 37 study reaches along the 73/175 highway (Quebec, Canada). The intensity of shading is proportional to altitude.

Brook charr density (100−1 m−2) was estimated in 3 consecutive years (2006-2008) at 212 stream sections distributed among 37 reaches in 36 streams (Strahler order 1–3, median length = 3·0 km and median slope = 4·4%) crossed by the highway (Fig. 2). The portion of total stream length spanned by the reaches ranged from 0·6 to 77·6% (median = 22·8%). Reaches were randomly assigned a rank (1–37) and were visited in the sequence determined by rank order from mid-June to late August in 2006 (37 reaches, 206 sections), 2007 (36 reaches, 198 sections) and 2008 (34 reaches, 187 sections). Brook charr were sampled by single-pass electrofishing (Smith-Root D-15, Vancouver, WA, USA) in an upstream direction within open stream sections (Jones & Stockwell 1995). Voltage and waveform were adjusted as required to account for variation in water conductivity. Stream width (m) was measured at five transects spaced equally along each section. Sections were 25 m in length (mean section width = 3·9 m and mean section area = 96·2 m2). Habitat management during the construction period (nine reaches) involved restoration of natural pool and riffle habitats within the first 100 m upstream or downstream of the highway. No major modifications to crossings (e.g. concrete aprons and baffles) were present in any of the reaches. Pools immediately downstream of the highway were avoided when selecting the study sections. Captured fish were counted, measured (total length, TL) and released at their section of origin.

Figure 2.

 Schematic representation of the distribution of 212 stream sections (vertical tick marks) among 37 reaches (horizontal lines). Reaches are identified by milepost distance from south (km 64·70) to north (km 221·60). Distances from the highway are provided along the bottom scale for the upstream (negative values) and downstream (positive) sections of each reach. Strahler stream order is provided for each reach.

Brook charr were present in all 37 reaches and accounted for 91·9% (10 653 individuals) of all fish captured. Longnose dace Rhinichthys cataractae (4·5%), Atlantic salmon Salmo salar (1·8%), white sucker Catostomus commersoni (0·3%), longnose sucker Catostomus catostomus (0·1%) and small unidentified cyprinids (1·4%) were also present in the streams. Brook charr were relatively small (TL: median = 65 mm; 1st quartile = 50 mm; 3rd quartile = 96 mm) and young (54·9% in the young-of-the-year age group).

Statistical analyses

We used linear mixed models to account for the hierarchical structure of the sampling design, which had sections nested within reaches and repeated observations nested within sections (Wagner, Hayes & Bremigan 2006; Deschênes & Rodríguez 2007). All mixed models had the following general structure:

image

where Ysrt is brook charr density transformed as loge (+ 1), the Xjsrt is p variables associated with treatment effects and spatial location as described below, αi (= 0,…, 3) and βj (= 1, …, p) are coefficients and the remaining indices represent section (s), reach (r) and sampling year (t). The first five terms on the right-hand side of the equation represent the systematic component of the model. Year2007 and Year2008 are categorical covariates coded as (−1, −1), (1, 0) and (0, 1) for years 2006, 2007 and 2008, respectively. Width is a continuous covariate, mean stream width, transformed as log(x) and standardised. The last three terms on the right-hand side of the equation represent the random component of the model. The usr and vr terms represent random effects for section and reach, respectively, and εsrt is random error. All random terms are assumed to follow a normal distribution with zero mean and variance to be estimated.

Two types of treatment effect were considered: stage of highway construction and type of highway crossing. Three alternative classification schemes based on the durations of both construction and recovery after construction and denoted here as C1, C2 and C3, were used to characterise the stage of construction of individual reaches (Fig. S2, Supporting Information). Type of highway crossing (H) was classified according to passability as high (21 reaches), intermediate (10 reaches) or low (6 reaches) (Fig. 3). To assess passability, we modified a classification system previously used by Love & Taylor (2003) and Poplar-Jeffers et al. (2009) to classify culverts in trout streams. The modifications we introduced account for features specific to our study system (e.g. presence of bridges and spillway design). Specifically, we (i) used the presence of a bridge at the crossing, rather than occurrence of streambed substrate in a culvert, as our first criterion for branching; (ii) did not use descriptors of the culvert inlet or channel width and (iii) assumed that control structures (spillways with natural substrate) ensured high passability; the unmodified classification assumes that weirs and baffles lead to intermediate passability. Dummy indicators were used to code both construction (reference category: ‘Before construction’) and highway crossing (reference category: ‘High passability’). Reaches 122·39 and 212·16, initially classified as having intermediate passability, were classified as highly passable after modifications from highway construction activities in 2008 and 2007, respectively. Spatial location of sections was represented by two variables: position relative to the highway, P, coded as a binary indicator: upstream (0) or downstream (1) and distance from the highway, D (km, negative distances upstream and positive distances downstream from the highway).

Figure 3.

 Decision tree for classification of highway crossings, modified from Love & Taylor (2003) and Poplar-Jeffers et al. (2009). Each highway crossing structure is classified by passability as high, intermediate or low.

We built a sequence of increasingly complex models including various additive combinations of treatment and location variables, as well as interactions between these variables (Table 1). The interaction terms between position and treatment indicate whether differences between local population densities at upstream and downstream sites are influenced by the treatment; such interactions are therefore viewed as instrumental in detecting environmental impacts (Underwood 1994). The set of models considered allows for the detection of a broad variety of potential effects on densities. Model comparison was based on Akaike Information Criterion adjusted for sample size, AICc (Burnham & Anderson 2002). Models were ranked using ΔAICc, the difference in AICc between a candidate model and the model with the lowest (best) AICc. Parameter estimation was based on a full maximum likelihood procedure. All analyses were carried out in the R environment (R Development Core Team 2010; nlme package, v. 3.1-96).

Table 1.   Comparison of 12 candidate models differing in their systematic component. The deviance, number of model parameters (K), adjusted Akaike Information Criterion (AICc) and difference in AICc relative to the best-fitting model (ΔAICc) are presented. Model terms are C3, stage of highway construction (‘before’, ‘during’ or ‘after’ construction; Fig. S2c, Supporting Information); H, type of highway crossing (‘low’, ‘intermediate’ or ‘high’ passability; Fig. 3; P, position relative to the highway (‘upstream’ or ‘downstream’); D, distance from the highway
ModelTerms in systematic component of model*DevianceKAICcΔAICc
  1. *All models include a constant term (α0), categorical covariates coding for year, mean stream width (m, standardised after loge-transform) and random terms for stream, section and error.

 1Constant (α0), year and stream width15007151556
 2P14818149738
 3D14928150950
 4P + D14779149637
 5P + D + P × D147610149637
 6P + C3 + P × C3147212149738
 7P + C3 + P × C3 + D146813149536
 8P + C3 + P × C3 + D + P × D146614149536
 9P + H + P × H14421214667
10P + H + P × H + D14401314678
11P + H + P × H + D + P × D14321414612
12P + H + P × H + D + P × D + H × D14261614590

We used a Markovian random walk model (Appendix S1, Supporting Information) to assess whether longitudinal fish movements and responses to barriers were a plausible explanation for the spatial patterns in density revealed by the linear mixed models. The model applies to a fish population in an idealised stream reach separated into upstream and downstream portions by a potential barrier located at the mid-point of the reach. Our approach extends the simple random walk by allowing barrier effects and directional bias to influence movements. Key model assumptions are as follows: (i) movement behaviours, including responses to a barrier, are density-independent, constant in time and identical across individuals; (ii) population size and spatial distribution are at equilibrium; (iii) fish follow a random walk that may be biased towards upstream or downstream movement; (iv) barriers do not hinder downstream movement but may affect upstream movement of fish that encounter them. We used a least-squares procedure to fit the Markovian model to density estimates derived from the linear mixed models (Appendix S1, Supporting Information). This procedure yielded point estimates and standard errors for four parameters that defined the transition matrix of the Markov chain: barrier passabilities for the three types of crossing (high, intermediate and low) and probability of downstream movement.

Results

The comparison of models incorporating different classification schemes to represent stage of construction (Fig. S2, Supporting Information) indicated that scheme C3 outperformed schemes C1 and C2; therefore, the results for models accounting for the stage of construction are presented only for models based on scheme C3. Model fit was improved by inclusion of variables representing position relative to the highway and distance from the highway (Table 1), but further inclusion of stage of construction (scheme C3; models 6–8) led only to marginal improvement in fit. The best fits were obtained for models including position relative to the highway, distance from the highway and type of crossing. More specifically, the best-fitting model in the candidate set (model 12) included interactions between position and type of crossing, position and distance, and type of crossing and distance (Tables 1 and 2). Graphical comparison of observed densities with those estimated from model 12 revealed that the model provided a reasonable fit and showed no obvious departures from the statistical assumptions of linearity and constant residual variance (Fig. S3, Supporting Information).

Table 2.   Parameter estimates for the best overall model (Table 1: model 12) of brook charr density. The dependent variable is brook charr density, transformed as loge (+ 1). Estimates and confidence intervals for the systematic and random components are presented. The σ2 terms represent variances of random terms for error (inline image), section (inline image) and reach (inline image)
 Estimate95% confidence intervals
LowerUpper
Systematic component
 Constant (α0)2·7572·2793·235
 Year
  Year20070·1270·0550·198
  Year20080·1360·0630·209
 Mean stream width (loge (x); standardised)−0·557−0·703−0·411
 Position (reference: upstream)0·023−0·4460·491
 Highway crossing (reference: high passability)
  Intermediate passability−1·260−1·902−0·618
  Low passability−2·021−3·127−0·916
 Distance (km)0·913−0·0981·925
 Distance × Position−1·244−2·6020·113
 Distance × Highway crossing
  Distance × Intermediate passability−1·726−3·6660·214
  Distance × Low passability−2·152−3·964−0·340
 Position × Highway crossing
  Position × Intermediate passability1·6660·8282·504
  Position × Low passability2·8791·8663·893
Random component
inline image0·6110·5690·657
inline image0·5150·4350·609
inline image1·0260·7901·331

Parameter estimates from model 12 (Table 2) were used to display graphically the joint effect of position, type of highway crossing and distance from the highway on density (Fig. 4). Density seemed mostly unaffected by position and distance from the highway for highway crossings having high passability (Fig. 4). In contrast, markedly different relationships between density and distance from the highway were found on either side of the highway for crossings of intermediate and low passability, leading to sharp discontinuities in estimated densities near the highway crossing (Fig. 4). In the vicinity of the highway (distance c. 0 m), the ratio of downstream to upstream densities was estimated as c. 7 for crossings of intermediate passability and c. 34 for crossings of low passability. Averaged over the upstream and downstream portions of the reach (0·8 km to either side of the crossing; Fig. 4), the ratio of downstream to upstream densities was estimated as c. 2 for crossings of intermediate passability and c. 6 for crossings of low passability.

Figure 4.

 Estimated fish density (solid lines) and 95% confidence intervals (shaded areas) as a function of distance from the highway and passability type (high, intermediate and low; Fig. 3), for the best-fitting linear mixed model (Table 1: model 12). Estimated fish density derived from the Markovian movement model (Appendix S1, Supplementary Information) is also shown (broken lines). Distances from the highway are provided along the bottom scale for upstream (negative values) and downstream (positive) sections.

The Markovian random walk model was a good fit to density estimates from the linear mixed model (Fig. 4). Parameter estimates (standard error in parentheses) for barrier passabilities for the three types of crossing (low, kL; intermediate, kI and high, kH) (Appendix S1, Supporting Information) were kL = 0·043 (0·0026), kI = 0·186 (0·0044) and kH = 0·645 (0·0122). These estimates indicate a decline in probability of passage for all crossing types; reduction in passability relative to a perfectly permeable barrier (= 1; Rodríguez 2010) ranged from c. 35% for crossings of high passability to >95% for crossings of low passability. The estimate for the probability of downstream movement, pd = 0·488 (0·0002) pointed to an overall directional bias towards upstream movement. The Markov model applies to an equilibrium situation (as described by the stationary distribution of the chain), and so this estimate represents a long-term value that averages over seasonal and yearly variations. Directional bias was required for the Markov model to generate the observed spatial discontinuities in density. Simulations for different values of k showed that the observed curvilinear gradients in density were not generated from a spatially homogeneous initial distribution in the absence of directional bias, that is, when pd = 0·5.

Discussion

Effect of highway construction

We did not detect an effect of highway construction activities on brook charr density downstream of the highway. Increased sedimentation is the most common impact affecting stream fish following road construction (Trombulak & Frissell 2000; Wheeler, Angermeier & Rosenberger 2005). The hydrological network can be particularly altered at intersections with roads, resulting in higher peak flows and sedimentation in downstream reaches (Jones et al. 2000). Sediment loading arising from road construction has been shown to be related to the stage of construction mostly along the first kilometre downstream of the source of sediments (Lachance et al. 2008). Highway construction activities were the only known supplemental source of sediments in our study, and sediment-induced turbidity was apparent in some reaches, yet the models incorporating the stage of highway construction did not reveal any substantial impact of sediment loading on brook charr densities. Newcombe & MacDonald (1991) suggested that both intensity and duration of exposure to suspended sediments must be known to predict the impacts of sedimentation on aquatic ecosystems. However, intensity of sediment loading alone is a poor predictor of impacts of suspended sediment (Newcombe & MacDonald 1991), and accurate quantification of sediment loading in situ is costly and time-consuming.

Different hypotheses could explain why highway construction activities had no apparent impact on brook charr density. First, measures taken to mitigate sediment loading into the streams during construction may have sufficed to prevent any impact on brook charr density. Secondly, most analyses of the effect of sediment on salmonids are from laboratory or field experiments focusing on growth and survival of juveniles (Suttle et al. 2004), or field studies focusing on embryonic stages (Guillemette et al. 2011). However, the responses at the population level might be different from those at embryonic and juvenile stages. For example, Curry & MacNeill (2004) showed that density of brook charr did not decrease in response to sedimentation, although survival to emergence was reduced.

In contrast to the long-term effects of highway crossings, which in our study system have been in place since 1948, the effects of suspended sediments can be short-lived if the stream has sufficient power to flush the material rapidly. Sub-lethal short-term responses of fish to increased sediment loads include behavioural avoidance (Scrivener, Brown & Andersen 1994), increased movement (Bergstedt & Bergersen 1997) and changes in physiology, foraging and growth (Harvey & Railsback 2009). Our design limits the detection of construction effects to those manifested within at most a 3-year period. Longer-lived latent effects of deposited sediment (Harvey & Railsback 2009) may therefore have gone undetected.

Effect of highway crossings

Brook charr populations were affected by highway crossings, as shown by differences in local density between upstream and downstream sites in the vicinity of highway crossings of intermediate and low passabilities. The slope of highway crossings and the presence of an outlet drop appear to be the most important predictors of passability (Love & Taylor 2003; Poplar-Jeffers et al. 2009). These two factors are often cited as causes of obstruction to free movement of fish and are widely considered in management policies (Warren & Pardew 1998; Poplar-Jeffers et al. 2009).

It seems unlikely that the observed density gradients were driven by differences in local habitat (e.g. through congregation of fish in plunge pools below the crossings), because habitats near the crossings were in a relatively natural state and fish were not collected in or very near to pools immediately downstream of crossings (Materials and methods: study system and data collection). Furthermore, larger pools tended to be associated with highly passable crossings such as wide bridges.

The Markov model produced results that were consistent with the observed patterns in density and provided insight into potential processes (barrier effects coupled with upstream movement) that could parsimoniously explain these patterns. The Markov model also provided quantitative estimates of passability that were consistent with the classification of crossing types and pointed to a decline in probability of passage for all crossing types. The apparent reduction in passability at crossings classified as highly passable is perhaps surprising in view of the known upstream swimming ability of brook charr (Adams, Frissell & Rieman 2000). However, lack of an attempt to pass a structure may result not from inability to pass, but from lack of motivation to pass after encountering the barrier (Kemp & O’Hanley 2010). Furthermore, swimming ability is strongly size-dependent and small brook charr (< 100 mm) may move less frequently and be inhibited by obstacles more readily than larger fish (Adams, Frissell & Rieman 2000). Brook charr populations in our study comprised mostly small juvenile fish (77% had TL < 100 mm); our results may therefore not generalise to populations of larger individuals.

Both the relative surplus (downstream) and deficit (upstream) of density near the crossing and the directional movement bias quantified by the Markov model are consistent with the notion that reduced passability affected primarily upstream movements (Morita, Yamamoto & Hoshino 2000). Telemetric tracking and length-frequency distributions indicate that in this system, spawners move readily between lakes and streams for at least some of the streams and suggest that the extent to which reproductive strategies involve migration, partial migration or residency may vary as a function of hydrological features such as distance from the stream spawning sites to the nearest lake. If the directional bias indicated by the Markov model truly reflects a stable, long-term pattern of movement, it would raise the intriguing possibility that brook charr populations in these small streams are at least partly sustained or supplemented by immigration from downstream source populations. A plausible scenario is that longitudinal differences in juvenile density arise when crossings restrict the upstream migration of spawners originating from downstream sources, and the spatial distribution of juvenile fish subsequently reflects the distribution of spawners during the spawning period. Such residual effects of spawning site location on the distribution of juvenile fish are well documented in stream salmonids (Hudy et al. 2010; Tentelier & Piou 2011). Alternatively, longitudinal differences in density could be generated primarily by upstream movement of young-of-the-year fish. Young brook charr sometimes show preferential upstream movement during the summer (Adams, Frissell & Rieman 2000; Peterson & Fausch 2003). However, mark–recapture trials conducted in four of the study streams showed no evidence of directional bias in brook charr movements during the summer (M. Pépino, M.A. Rodríguez & P. Magnan; unpublished data).

Density estimates derived from one-pass electrofishing are potentially subject to various sampling biases. In the present study, the use of random effects at the reach and section levels in the mixed models presumably helped to control for nuisance effects of unmeasured covariates on density (Gilks et al. 1993). Habitat features that influence the efficacy of electrofishing, such as stream width, conductivity, stream gradient and fish size distribution (Hense, Martin & Petty 2010), were relatively homogeneous within reaches (Table S1, Supporting Information), suggesting that variation in capture efficiency among sites did not unduly affect observed longitudinal patterns in density.

Management implications and conclusions

Restriction of fish passage at highway crossings and concomitant declines in stream connectivity (fragmentation) had markedly greater effects on local population densities than short-term impacts arising from construction activities (habitat degradation). These results contrast with the findings of many studies in terrestrial landscapes showing that habitat loss, rather than habitat fragmentation per se, is a main driver of distribution patterns (Mortelliti et al. 2011). Our results support the notion that highway crossings can contribute to fragmentation of the riverine landscape, with potential impacts on population persistence of stream fish (Letcher et al. 2007).

Density differences upstream and downstream of crossings may have implications for individual fitness of fish near the crossings. At densities similar to those in our study system, individual growth and energy acquisition for a population of brook charr have been shown to be density-dependent during warm periods (Utz & Hartman 2009). Sharp density differences were generated by the Markov model through processes that do not depend on local habitat quality. This result adds to a long list of caveats that density may be a misleading indicator of habitat quality (Van Horne 1983) and illustrates that explicit consideration of movement can sometimes help to avoid this pitfall (Bélanger & Rodríguez 2002). In this context, development of dispersal models that account for fragmentation in riverine landscapes (e.g. Schick & Lindley 2007; Rodríguez 2010) can contribute to a better understanding of biological processes underlying the observed spatial patterns (McIntire & Fajardo 2009).

Small tributaries are key rearing and spawning habitats for most salmonid species (Curry, Sparks & Van De Sande 2002); therefore, estimating habitat suitability upstream from barriers to determine the amount of habitat rendered unavailable by restriction of passage is an important goal for wildlife managers (Poplar-Jeffers et al. 2009). The decline of local densities upstream of crossings of intermediate and low passabilities provides a quantitative estimate of habitat loss resulting from reduction in connectivity. Our approach is therefore complementary to molecular approaches used to assess impacts of barriers (Wofford, Gresswell & Banks 2005; Griffiths et al. 2009), which highlight the genetic isolation of populations and provide information on changes in effective population size rather than local population density.

Explicit modelling of barriers to passage is a valuable management tool for the conservation of mobile fish species in fragmented riverine landscapes. Managers can use the modelling approaches proposed in this study to evaluate the impact of different crossing structures and the uncertainty associated with the resulting estimates. Estimates of impact derived from the models can inform decisions about construction of new barriers and prioritization for mitigation schemes or barrier removal. We emphasise that the evaluation of alternative structures needs not be based on discrete crossing types, because mixed models allow for a more nuanced characterisation of crossings by combining multiple descriptors, both continuous and discrete.

Acknowledgements

We thank Y. Paradis for invaluable logistic and field support and S. Beaulac and G. Pépin for field assistance. Y. Bédard and M. Lafrance (Ministère des Transports du Québec, MTQ) and the Société des établissements de plein air du Québec provided logistic support. The editor, J. Dunham, S. Wenger and an anonymous reviewer provided thoughtful and constructive suggestions. This research was financed primarily by the MTQ. Additional funding came from grants from the Natural Sciences and Engineering Research Council of Canada to P.M. and M.A.R. and a grant from the Canada Research Chair in Freshwater Ecology Program to P.M. M.P. was supported by the MTQ and a fellowship from the Fondation de la faune du Québec.

Ancillary