Polymorphic Mountain Whitefish (Prosopium Williamsoni) in a Coastal Riverscape: Size Class Assemblages, Distribution, and Habitat Associations

Ecology of Freshwater Fish 2014 Published 2014. This article is a U.S. Government work and is in the public domain in the USA. ECOLOGY OF FRESHWATER FISH

Polymorphic mountain whitefish (Prosopium williamsoni) in a coastal riverscape: size class assemblages, distribution, and habitat associations

James C. Starr1,2, Christian E. Torgersen3 1School of Environmental and Forest Sciences, University of Washington, Seattle, WA, USA 2U.S. Geological Survey, Washington Water Science Center, Tacoma, WA, USA 3U.S. Geological Survey, Forest and Rangeland Ecosystem Science Center, Cascadia Field Station, School of Environmental and Forest Sciences, University of Washington, Seattle, WA, USA

Accepted for publication June 10, 2014

Abstract – We compared the assemblage structure, spatial distributions, and habitat associations of mountain whitefish (Prosopium williamsoni) morphotypes and size classes. We hypothesised that morphotypes would have different spatial distributions and would be associated with different habitat features based on feeding behaviour and diet. Spatially continuous sampling was conducted over a broad extent (29 km) in the Calawah River, WA (USA). Whitefish were enumerated via snorkelling in three size classes: small (10–29 cm), medium (30–49 cm), and large (≥50 cm). We identified morphotypes based on head and snout morphology: a pinocchio form that had an elongated snout and a normal form with a blunted snout. Large size classes of both morphotypes were distributed downstream of small and medium size classes, and normal whitefish were distributed downstream of pinocchio whitefish. Ordination of whitefish assemblages with nonmetric multidimensional scaling revealed that normal whitefish size classes were associated with higher gradient and depth, whereas pinocchio whitefish size classes were positively associated with pool area, distance upstream, and depth. Reach-scale generalised additive models indicated that normal whitefish relative density was associated with larger substrate size in downstream reaches (R2 = 0.64), and pinocchio whitefish were associated with greater stream depth in the reaches farther upstream (R2 = 0.87). These results suggest broad-scale spatial segregation (1–10 km), particularly between larger and more phenotypically extreme individuals. These results provide the first perspective on spatial distributions and habitat relationships of polymorphic mountain whitefish.

Key words: Mountain whitefish; resource polymorphism; river habitat; spatial patterns; generalised additive model

with their mouth, such as birds, amphibians, and Introduction fishes (Wimberger 1994). In fishes, morphological Fluvial mountain whitefish (Prosopium williamsoni) variation typically occurs in the head and mouth offer an example of trophic polymorphism among (Smith & Skúlason 1996). Fishes exhibit many exam- stream dwelling salmonids in temperate North Ameri- ples of polymorphisms across taxa, particularly can river systems. Trophic polymorphism is a form among species of cichlids in low latitudes (Klingen- of resource-based phenotypic diversification that berg et al. 2003), and in recently glaciated lakes in occurs when exploitation of under-utilised resources high latitudes (Robinson & Wilson 1994). The spatial necessitates specific morphological characteristics. structure of lacustrine environments (e.g. littoral, lim- Trophic polymorphisms are most common among netic, and profundal habitats) provides a template for animals that subdue, handle, and capture their prey diversification and has led to alternate phenotypes of

Correspondence: J. Starr, 338 Hunt Road, Port Angeles, WA 98363, USA. E-mail: [email protected] J. Starr is not currently employed by the University of Washington but was affiliated at the time of this research. doi: 10.1111/eff.12163 1 Starr & Torgersen various species. Examples include, but are not limited 2000). Observations on foraging tactics reported by to, lake whitefish (Coregonus clupeaformis), brook Troffe (2000) from the upper Fraser River, British char (Salvelinus fontinalis), three-spinned stickleback Columbia (Canada), indicated that the pinocchio form (Gasterosteus aculeatus), and Arctic char (Salvelinus expended approximately half of its time foraging for alpinus; Smith & Skúlason 1996). The occurrence of benthic invertebrates using the elongated snout to sympatric morphotypes may be maintained through overturn substrate, whereas the normal form fed pri- spatial segregation of discrete foraging habitats marily on drifting invertebrates. Whiteley (2007) (Ostberg et al. 2009). observed the greatest variation in snout morphology Mountain whitefish exhibit resource-based poly- in the largest fish (>47 cm), and the least amount of morphism in rivers, with one form that develops an variation in the smallest fish (<25 cm), and con- elongated and slightly upturned snout (McPhail & cluded that a stage-specific ontogenetic shift likely Troffe 1998), coined the ‘pinocchio’ form by Troffe occurs between the ages 2 and 3. This age range cor- (2000), which is distinguished from the normal form responds to shifts in diet (Pontius & Parker 1973) that does not develop an elongated snout and has an and patterns of habitat use (Northcote & Ennis 1994) evenly sloped head and blunt snout (Fig. 1). Moun- from juvenile to adult life stages. As an individual tain whitefish are widely distributed and abundant moves from shallow to deeper water, divergent devel- among temperate rivers in western North America opment of the snout occurs and leads to further spe- (Northcote & Ennis 1994), but polymorphic moun- cialisation as the individual matures. However, the tain whitefish have not been widely reported in the biotic interactions and environmental factors influ- literature. Morphological variation in mountain white- encing variation in snout development have not been fish was historically thought to be the result of sexual described. dimorphism, with males developing an elongated Quantifying the spatial distribution of alternate snout (Evermann 1893), but research has identified morphotypes in stream fishes in relation to stream that both sexes can have elongated snouts (Troffe habitat characteristics is important for understanding polymorphic populations (Davis & Pusey 2010). No studies have examined polymorphic fluvial whitefish within the context of habitat heterogeneity and size class assemblage structure. In addition, information on coastal populations of mountain whitefish is lack- ing in the published literature. The objectives of our (a) study were to (i) evaluate differences in the spatial distribution of whitefish morphotypes and size classes, (ii) investigate the assemblage structure of morphotypes and size classes and their relationships with aquatic habitat, and (iii) quantify the associations (b) between the relative density of morphotypes and longitudinal variation in aquatic habitat. We hypothesised that morphotypes would be associated with 1 cm distinctly different habitat features based on previous studies of feeding behaviour and diet and that these habitat differences would be associated with segregation of morphotypes at the reach scale. We define spatial segregation here as broad-scale differences in (c) the overall distribution of morphotypes across the entire stream length, although overlap at finer spatial scales may occur.

Methods

Study area 1 cm The South Fork (SF) Calawah and mainstem Cala- wah River flow westward for 34 and 18 km, respec- Fig. 1. Mountain whitefish morphotypes in the Quileute basin, WA (USA): pinocchio (a), intermediate (b), and normal (c). The tively, on the western side of the Olympic Peninsula thin white lines drawn over the snouts highlight differences in Washington State (USA; Fig. 2). They have a between morphotypes. combined drainage area of 187 km2 and range in

2 Polymorphic mountain whitefish

N Calawah River N.F. Calawah RiverHyas Creek Sitkum River Elk Creek

Upper S.F. Calawah River

042 81216 km

Fig. 2. Location of the Calawah River on the Olympic Peninsula in Washington State, USA. The thick grey line between black hash marks represents the sampled stream length in the mainstem Calawah River and South Fork Calawah River. elevation from sea level to 1143 m (De Cillis 1998). In addition to mountain whitefish, the Calawah The river channel is geomorphically stable and mod- River and its tributaries support populations of sum- erately confined by fluvial terraces in the lower main- mer and winter steelhead (Oncorhynchus mykiss), stem Calawah River but is confined by steep valley coastal cutthroat trout (O. clarkii clarkii), fall coho walls in the upper mainstem Calawah and SF Cala- salmon (O. kisutch), and fall and summer chinook wah rivers. The hydrograph is strongly influenced by salmon (O. tshawytscha). A small population of rainfall, with peak flows occurring in November and river-type sockeye salmon (O. nerka) spawns annu- December (Hook 2004). The upper 25 km of the SF ally in the SF Calawah. Non-salmonids include Calawah River are located within the Olympic Pacific lamprey (Entosphenus tridentata), western National Park (ONP), and the downstream portion of brook lamprey (Lampetra richardsonii), speckled the river, along with most of the Sitkum River, lies dace (Rhinichthys osculus), longnose dace (R. cata- on federally owned forestland (U.S. Department of ractae), and sculpin (Cottus spp.). Agriculture Forest Service), which is managed as late-successional forest reserve (De Cillis 1998; Hook Sampling 2004). The mainstem Calawah River flows through privately managed forests that are actively harvested We conducted spatially continuous snorkelling and on a 35–40-year rotation (Hook 2004). The riparian habitat surveys on 16–19 August 2010 during sum- forest is dominated by Sitka spruce (Picea sitchen- mer base flow (1.84 m3s1, USGS gauging station sis), red alder (Alnus rubra), bigleaf maple (Acer 12043000) in the mainstem Calawah and SF Calawah macrophyllum), western redcedar (Thuja plicata), and River. Sampling began within the boundaries of the Douglas fir (Pseudotsuga mensiesii) in the lowlands, ONP, 4 km upstream from the confluence of the SF with western hemlock (Tsuga heterophylla) and Calawah and Sitkum rivers (47°560N, 124°120W) and silver fir (Abies amabilis) at higher elevations (Smith ended approximately 2 km upstream from the conflu- 2000). ence of the mainstem Calawah River and Bogachiel

3 Starr & Torgersen

River (47°560N, 124°260W, Fig. 2). The total extent the count of pinocchio morphotypes. To avoid bias of the survey was approximately 29 km, and 225 associated with estimating fish size underwater (Edgar channel units were sampled; 8 units were <0.25 m et al. 2004), three size classes were used that were deep and could not be sampled. Two kilometres of easily distinguishable underwater: small (10–29 cm), the lower mainstem Calawah River were not sampled medium (30–49 cm), and large (≥50 cm). because the wetted channel was too wide (>40 m) In each sampled channel unit, a technician on the and deep (>5 m) for a single diver. bank measured data on length, wetted width, maxi- We used well-established methods for snorkelling mum depth, and channel slope, and the diver esti- in streams (Thurow 1994; Torgersen et al. 2006; mated substrate composition and instream cover Brenkman et al. 2012). Snorkelling offered an alter- visually. Maximum unit depth was measured with a native to electrofishing and provided a relatively stadia rod, and gradient was measured with a stadia accurate and efficient method for enumerating salmo- rod and clinometer. Length and width were measured nids in our study section, in which the channel was with a laser rangefinder (Impulse 200 LR). Substrate too small to sample with a boat-mounted electrofish- composition was visually estimated as a per cent of er, and yet too large to sample with a backpack elec- pebble, cobble, boulder, and bedrock. The dominant trofisher (Cunjack et al. 1988; Joyce & Hubert substrate of each channel unit was assigned a rank 2003). All surveys were conducted by the diver mov- based on median size: pebble (16–64 mm) = 1; cob- ing downstream, typically in a single pass adjacent to ble (64–256 mm) = 2; boulder (>256 mm) = 3; and the thalweg. In locations with high salmonid density similar estimates were averaged (Allan 1995). and instream cover, the diver made a second pass Bedrock substrate was converted to a binary variable and averaged the two counts. Habitat size and com- (i.e. present or absent) because it did not constitute a plexity, fish species and size, density, and the inten- significant portion of the streambed surface area sity of sampling effort may affect the efficiency of (i.e. <10%) and typically was covered by a lens of snorkelling estimates of fish abundance (Rodgers cobble or pebble. Divisions between channel units et al. 1992; Bayley & Dowling 1993). Our snorkel- were mapped with a global positioning system (GPS; ling surveys of whitefish distribution and abundance Garmin III). Field-measured length was corrected did not account for sampling efficiency using alterna- with map distances using GPS point data in ArcView tive methods to verify whitefish counts and sizes. GIS (version 9.1, ESRI 2004) and high-resolution Using one highly experienced diver, we maintained a digital orthographic imagery (USDA 2009). Channel consistent sampling bias that allowed us to evaluate units were classified as either a pool or nonpool. A patterns of distribution and relative abundance but pool was defined as a habitat unit with smooth sur- did not provide accurate estimates of total fish abun- face flow, a maximum depth that was at least 25% of dance (Hankin & Reeves 1988). bankfull depth, and length, or width that was at least Fish and habitat surveys began at 09:00 and were 10% of bankfull width (Montgomery et al. 1995). completed each day by 16:00 to avoid diurnal and Channel type was identified as pool-riffle, forced crepuscular movement patterns of salmonids (Woot- pool-riffle, or plane bed (Montgomery & Buffington ton 1998) and to maximise underwater visibility. 1997). During the fish surveys, the water was very clear, and the diver was able to observe fish across the Data analysis entire wetted width. Counts of whitefish morphotypes were recorded in each channel unit (i.e. pool and Data collected at the unit scale were binned into nonpool). In addition to head shape, pinocchio white- approximately 1-km reaches (n = 29) for analysis of fish were distinguished from normal whitefish by the distribution, assemblage structure, and habitat associ- presence of highly visible white scar tissue on the tip ations. Bin numbers started at the downstream end of of the snout (Fig. 1). This white scar tissue and elon- the Calawah River and continued to the upstream end gated snout was visible but less pronounced in the of the SF Calawah River. The length of each bin was smallest size class of whitefish, and fish <20 cm determined according to changes in channel morphol- were rare among both morphotypes. The white scar ogy, including major tributary junctions, and natural tissue and differences in snout morphology on the breaks in channel and unit type. Channel type was medium and large size classes of the pinocchio mor- calculated as a percentage of total bin length, and photype were easily distinguishable by the diver. Fish pool area was calculated as a percentage of total bin that were intermediate between pinocchio and normal area. Floodprone and bankfull width were measured in terms of snout length also displayed white scar tis- on high-resolution orthophotographs (USDA 2009). sue and an elongated snout (Fig. 1). Because these Bankfull width, floodprone width, valley width intermediate morphotype fish showed signs of devel- index, substrate score, and maximum unit depth were oping into the pinocchio form, they were included in averaged for each bin. Bin gradient was derived by

4 Polymorphic mountain whitefish multiplying unit-scale gradient by length, summing junction with vectors representing correlations across all units within the bin, and dividing by total between axis scores and a second matrix of environ- bin length. Counts of mountain whitefish were mental variables. We used a cut-off value of summed for each bin and standardised in two different P ≤ 0.10 for plotting vectors for environmental vari- ways. First, data were standardised by columns (per- ables in the ordination (Legendre & Legendre 1998). centage cumulative abundance) and plotted versus dis- Associations between the relative density of mor- tance upstream for comparison of spatial distributions photypes with respect to environmental variables in SigmaPlot, version 10.0 (SYSTAT 2006). Second, were quantified using a series of generalised additive abundance data were standardised as relative density models (GAMs) in R (R Core Development Team (Dr) to account for variation in bin length and to reduce 2009), with the ‘mgcv’ package (Wood 2006). The variance in the datasets (Brenkman et al. 2012). We default smoothing function, thin-plate penalised define relative density as the ratio between fish density regression splines (Wood 2003), was used to model within a sample unit (approximately 1-km bin) to the non-linear relationships with a Gaussian error distri- overall density across the entire sample area: bution and identity link function: fi dðl Þ¼b þ ð Þþ ð Þþ i 0 s1 x1i s2 x2i (2) li Dr ¼ (1) ft lt where E(yi) li, yi are the independent observations, d is a ‘link function’ (identity in this case), b0 is the where fi = number of fish per 1-km custom bin, intercept, and s1 is a smoothing function for the linear li = length (km) of custom bin, ft = total number of predictor x1 (Wood 2001). fish, and lt = total length (km) sampled. Positive and Generalised additive models have several advanta- negative values of relative density indicated densities ges over the commonly used generalised linear model of mountain whitefish that were above and below the when testing for species habitat relationships, particu- average density for the entire length of stream sam- larly when data are collected continuously over a pled, respectively (Brenkman et al. 2012). Relative large extent (Hastie & Tibshirani 1990). A GAM pro- density and mean reach-scale channel and floodprone vides a less restrictive non-linear approach to model- widths were transformed using the natural logarithm ling in which the data determine the shape of the (Zar 1984). relationship (vs. the limited response shapes in a Assemblage structure of normal and pinocchio parametric model) through a series of smoothing whitefish size classes was assessed with nonmetric splines (Hastie & Tibshirani 1990). This approach is multidimensional scaling (NMS) in the statistical particularly effective when the response of an organ- software R (R Core Development Team 2009), using ism to a habitat variable is not a simple linear or the ‘vegan’ package. NMS is a powerful nonparamet- curve–linear relationship. Smoothed estimates are ric ordination technique for analysis of community based on a weighted average of neighbouring obser- data that violates the assumption of multivariate nor- vations using a back-fitting algorithm (Hastie & Tib- mality (McCune & Grace 2002). NMS uses the rank shirani 1987). This approach is particularly useful order of distances to represent objects in multivariate when the form of the relationship between the space (Digby & Kempton 1987). We used the Bray- response and predictor variable is unknown. The Curtis distance coefficient, a distance measure com- form of the relationship may range from a straight monly applied to count data with zero values, and line to increasingly complex nonparametric curves. 10,000 iterations to avoid reaching a local minimum Prior to fitting models for each morphotype, we in stress (Legendre & Legendre 1998). Complete examined multicollinearity among independent vari- absence of all ‘species’ (i.e. row sum of observa- ables. Independent variables that were correlated tions = 0) cannot be calculated as a distance coeffi- (│r│ > 0.3) were excluded from analysis. We then cient in multivariate analysis. Therefore, we removed visually examined spatial patterns of physical habitat several rows (n = 5) from the multivariate data set, features in relation to patterns in the relative density which corresponded to bins 26–29 in the upper SF of whitefish. Physical habitat was modelled as a Calawah River, and bin 20 in the lower SF Calawah response to distance upstream using the thin-plate River. A two-dimensional ordination was used in the penalised regression spline smoother in the ‘mgcv’ analysis to simplify interpretation. Although addi- package (Wood 2006) in R (R Core Development tional dimensionality did result in a lower stress Team 2009). Resulting spatial patterns of physical value, this did not change the dominant gradients in habitat provided a context for understanding the rela- assemblage structure. Centroids of size classes and tionship between the relative density of whitefish and morphotypes in ordination space were plotted in con- distance upstream. Substrate score was modelled as a

5 Starr & Torgersen function of distance upstream and displayed a spatial from the best GAMs for normal and pinocchio white- pattern of peaks and troughs (i.e. larger and smaller fish were back-transformed and plotted versus dis- substrate size). However, the relationship in the tance upstream (x-axis) in SigmaPlot (SYSTAT GAM was only significant when the effect degrees of 2006) using locally weighted scatterplot smoothing freedom (i.e. ‘knots’) was >9, and the linear correla- (LOWESS; Trexler & Travis 1993) with a second- tion was not significant (R2 = 0.003, P > 0.1). Mean degree polynomial smoothing parameter. Observed maximum depth did not display a strong spatial pat- values were included in plots to evaluate how well tern (R2 = 0.05, P > 0.1) in the GAM. Therefore, we each GAM predicted the locations and magnitudes of assumed that including either substrate or depth in an peaks and troughs in the relative density of normal additive model with distance upstream would and pinocchio whitefish. not result in inflated variance due to the effects of co-linearity. Results We also examined spatial autocorrelation in the model residuals. Spatial autocorrelation is the ten- Longitudinal distribution of whitefish size classes dency for samples close together in space to be more similar than those far apart, which violates the Mountain whitefish abundance was dominated by assumption of independence in statistical analyses pinocchio whitefish (75% of all individuals), and (Legendre 1993). We used GAMs to demonstrate that nearly 50% of all whitefish counted were medium- distance upstream was significantly associated with sized pinocchio whitefish (Tables 1 and 2). No both normal and pinocchio whitefish relative density whitefish were observed in the SF Calawah River (P < 0.01). We then extracted the model residuals to upstream of the confluence with the Sitkum River. evaluate spatial autocorrelation using variograms Spatial distributions differed between pinocchio and (Palmer 2002) with the ‘variog’ function in the normal whitefish. Normal whitefish were distributed ‘geoR’ package in R (R Core Development Team farther downstream than pinocchio whitefish, and 2009). We assessed semi-variance at multiple lag large size classes of both morphotypes were distrib- intervals and distances and determined that the resid- uted farther downstream than small and medium uals did not display strong spatial dependence. We size classes (Fig. 3). Spatial patterns of small and included distance upstream as a covariate when quan- medium size classes of normal whitefish were simi- tifying associations between whitefish abundance and lar; 50% of small and medium normal whitefish other physical habitat characteristics. were in the lower 9.6 and 10.7 km of the study We analysed associations of whitefish relative den- area, respectively (Fig. 3). Fifty per cent of normal sity with each physical habitat variable individually, whitefish in the large size class were in the lower including distance upstream as the first covariate in 6.6 km of the study area. The cumulative abun- each model to de-trend spatial patterns. The variables dance of large, normal whitefish increased from that explained the most variance in each normal and 50% to 80% at 7.8 km. The upper extent of small pinocchio whitefish GAM were used in a model of and medium size classes (24.1 and 25.1, respec- the alternate morphotype to determine how relation- tively) was much farther upstream than the large ships with the same variables differed between mor- size class (16.4 km). photypes. We then used stepwise forward-variable Pinocchio whitefish displayed size class distribu- selection and analysis of deviance (ANODEV) with an tion patterns similar to those of normal whitefish approximate v2 test, (Hastie 1991) to select a suite of such that small- and medium-sized fish were distrib- physical habitat variables that were associated with uted upstream and larger fish were downstream. each morphotype. Fitted values and standard errors However, small and medium size classes of pinocchio

Table 1. Abundance and relative density of mountain whitefish (Prosopium williamsoni) by morphotype and size class in the Calawah River. See Equation 1 for the calculation of relative density.

Morphotype Size class Length (cm) Raw abundance Per cent abundance Mean relative density (SD)

Normal mountain whitefish Small 10–29 283 9 0.968 1.396 Medium 30–49 469 14 0.975 1.167 Large ≥50 70 2 0.986 1.935 All size classes 822 25 0.974 1.192 Pinocchio mountain whitefish Small 10–29 566 17 0.965 1.420 Medium 30–49 1542 47 0.980 0.979 Large ≥50 348 11 0.989 0.907 All size classes 2456 75 0.978 1.004

6 Polymorphic mountain whitefish

Table 2. Summary statistics of physical habitat characteristics in the lower and upper mainstem Calawah River, and the lower and upper South Fork (SF) Calawah River.

Lower mainstem (n = 10) Upper mainstem (n = 6) Lower SF (n = 9) Upper SF (n = 4)

Habitat characteristics Mean Range SD Mean Range SD Mean Range SD Mean Range SD

Length (km) 1.0 0.86–1.19 0.12 1.0 0.88–1.09 0.09 1.02 0.87–1.26 0.11 0.94 0.75–1.06 0.15 Wetted width (m) 33 26–42 4.2 29 22–38 6.3 20 16–25 2.7 9.8 8–12 1.7 Bankfull width (m) 47 38–59 5.5 40 31–45 4.9 39 22–53 10.1 17.5 15–21 2.9 Channel type (%) Pool-riffle 52 0–100 39 44 0–100 50 37 0–100 39 27 0–54 25 Forced pool-riffle 30 0–100 33 17 0–100 41 41 0–100 40 57 13–100 37 Plane bed 17 0–54 23 40 0–100 49 22 0–100 44 16 0–44 21 Gradient (%) 1.2 0.02–1.5 0.04 1.2 0.09–1.5 0.03 1.2 0.08–2.2 0.05 1.1 1–1.4 0.02 Pool area (%) 35 0–76 23 37 24–54 12 29 0–66 29 23 11–44 16 Mean max. depth (m) 1.45 0.94–2.05 0.35 1.5 0.77–2.12 0.49 1.25 0.54–1.80 0.46 1.05 0.74–1.33 0.26 Substrate score 2.3 1.7–2.7 0.33 2.2 1.5–2.75 0.41 2.3 1.8–2.9 0.40 2.2 2.0–2.4 0.16 Valley width index 1.54 1.32–1.78 0.13 1.53 1.30–1.84 0.19 1.56 1.31–2.05 0.22 1.52 1.49–1.57 0.03

Sections were sampled at the unit scale and data were binned into approximately 1-km reaches (N = 29) using major tributary junctions, and natural breaks in channel and unit type. Substrate score corresponds to pebble, cobble, and boulder (i.e. scores of 1, 2, and 3, respectively). Valley width index is the ratio of flood-prone width to channel width. Cumulative abundance (%)

Fig. 3. Cumulative abundance of mountain whitefish morphotypes and size classes Distance upstream (km) versus distance upstream. whitefish were distributed farther upstream than these Size class assemblage structure size classes of normal whitefish. Approximately 50% Multivariate ordination with NMS provided evidence of the small and medium pinocchio whitefish were of segregation among size classes in both morpho- distributed in the lower 13.1 and 12.8 km, respec- types (Fig. 4). Centroids of normal whitefish size tively. Small pinocchio whitefish reached 78% in classes were grouped in the lower left quadrant of the cumulative abundance at 14.9 km, and medium ordination plot. Small and medium size classes of pinocchio whitefish reached 76% in cumulative normal whitefish were associated with both reach abundance at 15.2 km. Pinocchio whitefish in both gradient and mean maximum depth, whereas the small and medium size classes extended 25 km large size class was associated with reach gradient. upstream to the confluence with the SF Calawah All three size classes of normal whitefish were inver- River and Sitkum River. The distribution of the large sely associated with distance upstream, and the large pinocchio whitefish was similar to the distribution of size class had the strongest inverse relationship. medium-sized normal whitefish. Large pinocchio Pinocchio whitefish size classes were grouped in the whitefish reached 51%, 75% and 100% in cumula- upper two quadrants of the ordination plot. Small tive abundance at 9.0, 14.1 and 25 km upstream, and medium size classes of pinocchio whitefish were respectively.

7 Starr & Torgersen

Fig. 4. Non-metric multidimensional scaling (NMS) ordination of mountain whitefish morphotypes and size classes. Black diamonds represent 1-km bins in species space. Small, medium, and large open circles (normal) and open squares (pinocchio) indicate the centroids of size classes. Vectors indicate the direction and relative correlation of environmental variables with respect to ordination axes. Only vectors with P ≤ 0.10 and R2 > 0.30 are plotted. positively associated with per cent pool area and size (Fig. 5d; R2 = 0.59). Pinocchio whitefish rela- mean maximum depth. Large pinocchio whitefish tive density was best explained by distance upstream were positively associated with distance upstream. and mean maximum depth (Fig. 5g,h; R2 = 0.87). All three size classes of pinocchio whitefish were Pinocchio whitefish displayed a strong positive more closely associated with distance upstream than curve–linear relationship with mean maximum depth, normal whitefish. The grouping of size classes by with higher densities in reaches with a mean maxi- morphotype indicated that combining size classes mum depth ≥1.2 m. In contrast, there was a weak was appropriate for analyses of habitat associations. positive linear relationship (R2 = 0.47) between normal whitefish relative density and mean maximum depth (Fig. 5f), after accounting for the effect of dis- Habitat associations with the relative density of tance upstream (Fig. 5e). Mean maximum depth morphotypes (d.f. = 2.9, P < 0.001) and distance upstream Generalised additive models indicated that variation (d.f. = 7.3, P < 0.001) in the GAM for pinocchio in the relative density of normal whitefish was best whitefish resulted in 67% and 24% reductions in explained by distance upstream, substrate size (an deviance, respectively. Outliers were omitted from increase in substrate score is synonymous with an Fig. 5c (x = 13, y = 8.8, and x = 15, y = 12.7), increase in size), and a linear coefficient for the Fig. 5d (x = 3.2, y = 4.9), Fig. 5g (x = 15, y = 5.8), occurrence of bedrock substrate (Fig. 5a,b; and Fig. 5h (x = 2, y = 7.7). No other variables were R2 = 0.64). The association between normal white- significantly associated with pinocchio whitefish rela- fish relative density and substrate, given the effects tive density. of distance upstream and bedrock occurrence, was Plots of fitted and observed whitefish relative den- positive up to a substrate score of 2.4 (analogous to sities versus distance upstream indicated that the reaches dominated by boulder and boulder/cobble GAM for normal whitefish accurately predicted the mix); at substrate scores >2.4, there was an inverse locations of peaks and troughs in whitefish relative relationship between substrate size and normal white- density (Fig. 6a). However, this model was not able fish relative density. Distance upstream (d.f. = 4.9, to predict the magnitudes of peaks and troughs in rel- P<0.001), substrate size (d.f. = 4.1, P = 0.02), and ative density. For example, the three highest peaks in bedrock occurrence (d.f. = 1, P = 0.02) resulted in observed normal whitefish relative density exceeded 46%, 14% and 18% reductions in deviance, respec- the standard error interval. In contrast, the pinocchio tively. No other variables were significantly associ- whitefish GAM accurately predicted both the loca- ated with the relative density of normal whitefish. tions and the magnitudes of peaks and troughs in rel- The relative density of pinocchio whitefish was ative density (Fig. 6b). Observed values for positively associated with distance upstream to pinocchio whitefish relative density were generally approximately 13 km and was negatively associated within the standard error intervals, and the second with distance upstream to 29 km (Fig. 5c). However, highest observed peak in pinocchio whitefish relative there was no significant association with substrate density matched the fitted value.

8 Polymorphic mountain whitefish

(a) (b)

P = 0.008 P = 0.059 Relative density

Substrate score (c) (d)

P = 0.505 P = 0.001 Relative density

Substrate score (e) (f) P = 0.032 P = 0.011 Relative density

(g) (h) Mean max. depth (m) P = 0.002

P < 0.001 Relative density

Distance upstream (km) Mean max. depth (m) Fig. 5. Thin-plate regression splines of modelled relative density of normal (a, b, e, f) and pinocchio (c, d, g, h) whitefish with respect to the additive effects of distance upstream and substrate score, and distance upstream and mean maximum depth. Substrate score corresponds to pebble, cobble, and boulder (i.e. scores of 1, 2, and 3, respectively). Solid circles represent partial raw residuals, dashed lines indicate 2 standard errors, and P-values refer to the significance of the smoothed parameter.

9 Starr & Torgersen

(a) previous studies on the spatial distribution of polymorphic whitefish in rivers, our results corroborate other reports that mountain whitefish are generally found lower in the watershed and in mainstem habitats (Platts 1979; Meyer et al. 2009). Smaller tributaries may not provide suitable habitat conditions, such as adequate depth and cover for mountain whitefish (Sigler 1951). However, McPhail & Troffe (1998) and Meyer et al. Relative density (2009) found that mountain whitefish occupied smaller streams (5–10 m wetted width). The species also has been observed in small tributaries (mean wetted width <10 m) within the Calawah River and Hoh River drainages during late winter and spring (J.C. Starr & J.R. McMillan, unpublished data). Meyer et al. (2009) (b) speculated that whitefish in the southern portion of their range typically use larger streams, whereas whitefish in the northern portion of their range generally are found in smaller streams. Coastal mountain whitefish may move into tributaries after they spawn in the fall to seek refuge during periods of higher flow. We observed a longitudinal pattern of larger fish downstream and smaller fish upstream in both mor- Relative density photypes (Fig. 3). Similar patterns with large individuals downstream have been documented for other species (Power 1984; Welcomme 1985). Our results suggest that large mountain whitefish, regardless of morphotype, may require a greater volume of habitat Distance upstream (km) during summer base flow conditions, perhaps to bal- ance the trade-offs between predation risk and forag- Fig. 6. Longitudinal variation in modelled and observed relative density of normal (a) and pinocchio (b) mountain whitefish. The ing opportunities. solid lines are locally weighted scatterplot smoothing (LOWESS) Small and medium size classes of both morpho- of the fitted values, dashed lines represent a LOWESS smooth of types had similar spatial distributions that were dis- 2 standard errors, and open circles are observed values. tinctly different from the large size classes of normal and pinocchio whitefish, which were distributed farther downstream. We observed differences between Discussion small and medium versus large size classes that were Our results showed that pinocchio and normal moun- similar to the observations by Whiteley (2007), who tain whitefish morphotypes had different spatial dis- found that morphological and dietary variation was tributions that were associated with differences in greatest in larger individuals (>47 cm), and least in aquatic habitat. Overall distribution patterns sug- smaller individuals (<25 cm). Although the patterns gested that there was segregation of whitefish mor- in cumulative abundance among size classes corre- photypes at a broad spatial scale (1–10 km). It is sponded with this spectrum of size-related morpho- important to note that previous studies of segregation logical variation, the relative abundances did not. If typically focus on finer spatial scales (Fausch & dietary variation is less in smaller individuals, we White 1981; Gibson & Erkinaro 2009). Our study would expect that the abundance of smaller normal focused on overall patterns of segregation across and whitefish would be greater in this size class. How- entire stream length. ever, the abundance of small (10–29 cm) pinocchio whitefish was twice that of small normal whitefish. It is possible that juvenile mountain whitefish in low- Longitudinal distribution elevation coastal river systems, specifically on the Normal whitefish were more abundant in the lower Olympic Peninsula, undergo more rapid growth as Calawah River, and pinocchio morphotype fish were juveniles and make a shift in feeding habits earlier in more abundant in the upper Calawah River and the their ontogeny (McHugh 1942). Currently, there is lower SF Calawah River (Fig. 3). No whitefish were limited information on growth and maturation of found in the upper SF Calawah River where mean wet- coastal mountain whitefish populations, and more ted width was <10 m (Table 2). Although there are no research in this area is needed.

10 Polymorphic mountain whitefish

size classes were combined in a model of relative den- Size class assemblage structure sity, normal whitefish were associated with larger sub- We found that whitefish morphotypes occurred in strate (Fig. 5b). In our study, it is possible that deeper, distinct assemblages based on size class. For exam- fast-water habitat, combined with large boulder sub- ple, small and medium size classes occurred together, strate in the lower Calawah River provided better for- whereas the large size class was distinct from other aging opportunities for drifting invertebrates preferred size classes in ordination space (Fig. 4). The positive by normal whitefish (sensu Troffe 2000; Whiteley association between depth and the small and medium 2007). However, distance upstream, substrate, and size classes of both morphotypes provided further bedrock occurrence explained only 64% of the varia- evidence of fine-scale spatial overlap between white- tion in relative density, suggesting that other variables, fish morphotypes in deep-water habitat (i.e. pools). which we did not consider (e.g. temperature, primary These results suggest that habitat use becomes more production, and aquatic invertebrate communities), specialised in larger individuals that are more pheno- may explain additional longitudinal variation in the typically extreme, i.e. have more pronounced, elon- abundance of normal whitefish. The weak linear rela- gated snouts. These results are similar to the findings tionship between normal whitefish relative density and by Whiteley (2007), in which larger individuals depth (Fig. 5f) suggests that the normal morphotype is (>47 cm) displayed greater morphological variation associated with large substrate in deep-water habitats that was associated with significant differences in that provides refuge from both flow and predation for diet. Whiteley (2007) concluded that this stage-spe- optimal feeding on drifting aquatic invertebrates. cific ontogenetic shift occurs at approximately Pinocchio whitefish relative density was strongly asso- 3 years of age (i.e. 20–25 cm in length). There may ciated with mean maximum depth (Fig. 5h). Mountain be a significant dietary shift between juvenile and whitefish have been shown to be associated with adult life stages (Pontius & Parker 1973) that corre- greater depth (DosSantos 1985; Torgersen et al. 2006), sponds with a change in habitat use from shallow, and it has been speculated in other studies that deep marginal habitat to deeper and faster habitat in the pools provide adequate cover for the species (Sigler main channel (Northcote & Ennis 1994). In our 1951). The pinocchio whitefish GAM fitted peaks in study, the medium size class (30–49 cm) was the the relative density located in the upper Calawah River most abundant size class. The small size class of and a second in the SF Calawah River. Bedrock out- whitefish (10–29 cm) was dominated by individuals croppings and forced pool-riffle channel morphology that were 25–29 cm in length and had the head shape dominate these river segments (Table 2), creating deep and white scar tissue on the tip of the snout charac- scour pools that provide slower velocities, cover from teristic of the pinocchio morphotype. Morphological predators, and small-diameter substrate. Deep pools variation appeared less extreme in fish <25 cm, and with small-diameter substrate in low-gradient reaches whitefish <20 cm were extremely rare. Benjamin may provide the habitat conditions that pinocchio et al. (2014) showed that the majority of juvenile whitefish require to overturn substrate and forage for whitefish move downstream in their first year of life. benthic invertebrates (Troffe 2000). We observed indi- Thus, our study area may not encompass the down- vidual pinocchio whitefish using their snout to over- stream extent of juvenile whitefish in the Quileute turn pebbles and small cobbles, but we did not basin. Our results support the findings by Pontius & quantify the extent of this behaviour. Parker (1973) that mountain whitefish undergo a Feeding habits of mountain whitefish morphotypes niche shift later in ontogeny (20–24 cm) and the con- may partially explain their patterns of habitat associa- clusions by Whiteley (2007) that this shift leads to tion (Northcote & Ennis 1994). The species is often phenotypic diversification. Although the positive associated with greater stream width, lower gradient, association between gradient and large normal white- deep pool habitat (Sigler 1951; Meyer et al. 2009), fish may be a departure from the typical relationship and small substrate size (DosSantos 1985). They also between whitefish and depth, this association does have been associated with higher-gradient shallow suggest that normal whitefish may be using high-gra- habitats (e.g. runs and riffles; Torgersen et al. 2006). dient reaches for their high rates of invertebrate prey Thompson & Davies (1976) found that Sheep River delivery (Troffe 2000). mountain whitefish occupied positions in the water column that were 2 to 10 cm from the stream bed while they were feeding on drifting invertebrates. Associations between whitefish relative density and None of these whitefish were observed feeding on aquatic habitat benthic invertebrates, although gravel or sand Generalised additive models enabled us to quantita- occurred in the stomach contents of 54% of the fish tively assess spatial associations between mountain examined. In contrast, DosSantos (1985) found that whitefish relative density and aquatic habitat. When all mountain whitefish in the Kootenai River fed

11 Starr & Torgersen disproportionately on chironomids, and underwater multiple spatial scales. Broad-scale spatial segrega- observations revealed that these fish were using their tion of phenotypically extreme individuals (i.e. those snout to overturn smaller substrate. Differences in with more pronounced, elongated snouts) and the dif- habitat relationships and feeding behaviour described fering habitat associations of the two morphotypes in the aforementioned studies may be explained in may have the ecological function of reducing intra- part by specific polymorphic associations with habitat specific competition and maintaining phenotypic unique to a given basin or subbasin. diversity. The degree of reproductive isolation (spatial or temporal) between the two morphotypes is unknown. Because mountain whitefish are broadcast Ecological importance and implications of polymorphism spawners, reproductive isolation between morpho- in mountain whitefish types may be very limited (Whiteley 2007). Addi- Across their range, mountain whitefish play an tional investigation on reproductive behaviour is important role in stream ecosystems, but their ecol- needed to determine the degree of spatial and tempo- ogy is still not fully understood. Fluvial mountain ral isolation between morphotypes. Condition factor whitefish spend their entire lives in the freshwater (i.e. lipid levels), growth, movement, and survival of environment and have been reported to live up to the two morphotypes at juvenile and adult life stages 29 years (McHugh 1942). Consequently, mountain also require further examination to elucidate the whitefish have been identified as a potential indicator physiological trade-offs of phenotypic diversification. species of riverine habitat conditions and water qual- Spatially explicit data on basin-wide abundance col- ity (McPhail & Troffe 1998). Studies indicate that lected in conjunction with detailed information on mountain whitefish exhibit homing behaviour associ- movement, growth, and survival may be required to ated with summer rearing and fall spawning locations identify the mechanisms that maintain the phenotypic and that they are capable of extensive movement plasticity found in mountain whitefish. Coastal moun- within a stream network (Baxter 2002; Benjamin tain whitefish constitute an important component of et al. 2014) and, thus, may contribute substantially to the biomass and ecological complexity of river sys- nutrient transport within a watershed (Lance & Baxter tems in the Pacific Northwest. Polymorphism enables 2011). Movement by mountain whitefish throughout mountain whitefish to diversify its food base, and this stream networks during different life stages suggests provides a potential explanation for why mountain that they may require a watershed-scale approach to whitefish are often more abundant than sympatric habitat conservation that has been applied to managed salmonids in rivers across their range. stocks of anadromous salmonids (Roni et al. 2002). Mountain whitefish have been studied extensively in interior rivers (e.g. east of the Cascade Mountains Acknowledgements and in the Rocky Mountains) in the USA and Can- We thank Walter A. (Peter) Starr for generous moral and ada, yet comparatively little is known about their financial support to J.C.S. throughout this project. The U.S. ecology in coastal watersheds of the USA. The Geological Survey (USGS) Western Fisheries Research Center Olympic Peninsula in western Washington offers a also provided funding. We are grateful to Aaron Ruesch [pre- unique example of coastal mountain whitefish popu- viously with the University of Washington (UW), School of lations. Whiteley et al. (2006) found that whitefish Environmental and Forest Sciences] for his assistance with from the Hoh River basin, which flows into the data collection. Samuel Brenkman (Olympic National Park, Pacific ocean on the western side of the Olympic Chief Fisheries Biologist), and the USGS Forest and Range- Peninsula, were genetically related to the coastal pop- land Ecosystem Science Center (FRESC) Cascadia Field Sta- ulation of mountain whitefish in the lower Fraser tion provided essential survey equipment and field support. We also thank Susan Bolton (UW School of Environmental River in British Columbia, Canada. However, white- and Forest Sciences), Thomas Quinn (UW School of Aquatic fish from the North Fork Skokomish River that flows and Fishery Sciences), and the Quinn Lab group for essential into the Hood Canal on the eastern side of the Olym- feedback and constructive criticism. We thank Jeffrey Duda pic Peninsula were more closely related to popula- (Western Fisheries Research Center, Research Ecologist) and tions west of the Cascade Mountains in Oregon, two anonymous reviewers for their comments that greatly Washington, and British Columbia. Beyond range- improved the manuscript. Any use of trade, product, or firm wide genetic assessments, there have been no other names is for descriptive purposes only and does not imply studies of mountain whitefish in the coastal regions endorsement by the U.S. Government. of Washington, Oregon, and northern California. Our results provide an important perspective on References coastal mountain whitefish ecology, including the spatial distributions of morphotypes, size class Allan, J.D. 1995. Stream ecology: structure and function of assemblages, and associations with aquatic habitat at running waters. New York: Chapman and Hall.

12 Polymorphic mountain whitefish

Baxter, C.V. 2002. Fish movement and assemblage dynamics Hastie, T.J. & Tibshirani, R. 1987. Generalized additive mod- in a Pacific Northwest riverscape. Corvallis: Doctoral disser- els: some applications. Journal of the American Statistical tation, Oregon State University. Association 82: 371–386. Bayley, P.B. & Dowling, D.C. 1993. The effects of habitat in Hastie, T.J. & Tibshirani, R. 1990. Generalized additive mod- biasing fish abundance and species richness estimates when els. Monographs on Statistics and Applied Probability 43. using various sampling methods in streams. Polskie Archi- New York: Chapman and Hall. wum Hydrobiologii 40: 5–14. Hook, A. 2004. WRIA 20: technical assessment level I water Benjamin, J.R., Wetzel, L.A., Martens, K.D., Larsen, K. & quality and habitat. Forks: University of Washington Olym- Connolly, P.J. 2014. Spatio-temporal variability in move- pic Natural Resources Center. ment, age, and growth of mountain whitefish (Prosopium Joyce, M.P. & Hubert, W.A. 2003. Snorkeling as an alterna- williamsoni) in a river network based upon PIT tagging and tive to depletion electrofishing for assessing cutthroat trout otolith chemistry. Canadian Journal of Fisheries and Aquatic and brown trout in stream pools. Journal of Freshwater Ecol- Sciences 71: 131–140. ogy 18: 215–222. Brenkman, S.J., Duda, J.J., Torgersen, C.E., Welty, E., Pess, Klingenberg, C.P., Barluenga, M. & Meyer, A. 2003. Body G.R., Peters, R. & McHenry, M.L. 2012. A riverscape per- shape variation in cichlid fishes of the Amphilophus citrinel- spective of pacific salmonids and aquatic habitats prior to lus species complex. Biological Journal of the Linnean Soci- large-scale dam removal in the Elwha River, Washington, ety 80: 397–408. USA. Fisheries Management and Ecology 19: 36–53. Lance, M.J. & Baxter, C.V. 2011. Abundance, production, Cunjack, R.A., Randall, R.G. & Chadwick, E.M.P. 1988. and tissue composition of mountain whitefish (Prosopium Snorkeling versus electrofishing: a comparison of census williamsoni) in a central Idaho wilderness stream. Northwest techniques in Atlantic salmon rivers. Canadian Naturalist Science 85: 445–454. 115: 89–93. Legendre, P. 1993. Spatial autocorrelation: trouble or new par- Davis, A.M. & Pusey, B.J. 2010. Trophic polymorphism and adigm? Ecology 74: 1659–1673. water clarity in northern Australian Scortum (Pisces: Tera- Legendre, P. & Legendre, L. 1998. Numerical ecology, 2nd pontidae). Ecology of Freshwater Fish 19: 638–643. English edition. Amsterdam: Elsevier. De Cillis, P. 1998. Fish habitat. Sitkum and South Fork Cala- McCune, B. & Grace, J.B. 2002. Analysis of ecological com- wah watershed analysis. Olympia, WA: Olympic National munities. Gleneden Beach, OR: MJM Software Design. Forest. McHugh, J.L. 1942. Growth of the Rocky Mountain white- Digby, P.G.N. & Kempton, R.A. 1987. Multivariate analysis fish. Journal of the Fisheries Resources Board of Canada 5: of ecological communities. In: Usher, M.B. & Rosenzweig, 337–343. M.L., eds. Population and community biology. New York: McPhail, J.D. & Troffe, P.M. 1998. The mountain whitefish Chapman and Hall, pp. 149–175. (Prosopium williamsoni): a potential indicator species for DosSantos, J.M. 1985. Comparative food habits and habitat the Fraser system. Vancouver: Report DOE FRAP 1998-16 selection of mountain whitefish and rainbow trout in the prepared for Environment Canada, Aquatic and Atmospheric Kootenai River, Montana. Bozeman: Master’s of Science Sciences Division. thesis, Montana State University. Meyer, K.A., Elle, F.S. & Lamansky, J.A. Jr 2009. Edgar, G.J., Barret, N.S. & Morton, A.J. 2004. Biases associ- Environmental factors related to the distribution, abundance, ated with the use of underwater visual census techniques to and life history characteristics of mountain whitefish in quantify the density and size-structure of fish populations. Idaho. North American Journal of Fisheries Management Journal of Experimental Marine Biology and Ecology 308: 29: 753–767. 269–290. Montgomery, D.R. & Buffington, J.M. 1997. Channel-reach ESRI. 2004. ArcGIS desktop: release 9.1. Redlands, CA: morphology in mountain drainage basins. Geological Soci- Environmental Systems Research Institute. ety of America Bulletin 109: 596–611. Evermann, B.W. 1893. A reconnaissance of the streams and Montgomery, D.R., Buffington, J.M., Smith, R.D., Schimdt, lakes of western Montana and northwestern Wyoming. Bul- K.M. & Pess, G.R. 1995. Pool spacing in forest channels. letin of the United States Fish Commission XI: 3–60. Water Resources Research 31: 1097–1105. Fausch, K.D. & White, R.J. 1981. Competition between brook Northcote, T.G. & Ennis, G.L. 1994. Mountain whitefish biol- trout (Salvelinus fontinalis) and brown trout (Salmo trutta) ogy and habitat use in relation to compensation and for positions in a Michigan stream. Canadian Journal of improvement possibilities. Reviews in Fisheries and Science Fisheries and Aquatic Sciences 38: 1220–1227. 2: 347–371. Gibson, R.J. & Erkinaro, J. 2009. The influence of water Ostberg, C.O., Pavlov, S.D. & Hauser, L. 2009. Evolution- depths and inter-specific interactions on cover responses of ary relationships among sympatric life history forms of juvenile Atlantic salmon. Ecology of Freshwater Fish 18: Dally Varden inhabiting the landlocked Kronotsky Lake, 629–639. Kamchatka, and the neighboring anadromous population. Hankin, D.G. & Reeves, G.H. 1988. Estimating total fish Transactions of the American Fisheries Society 138: abundance and total habitat area in small streams based on 1–14. visual estimation methods. Canadian Journal of Fisheries Palmer, M.W. 2002. Scale detection using semivariograms and Aquatic Sciences 45: 834–844. and autocorrelograms. In: Gergel, S.E. & Turner, M.G., Hastie, T.J. 1991. Generalized additive models. In: Chambers, eds. Learning landscape ecology: a practical guide to J.M. & Hastie, T.J., eds. Statistical models in S. New York: concepts and techniques. New York: Springer, pp. 129– Chapman and Hall, pp. 249–307. 144.

13 Starr & Torgersen

Platts, W.S. 1979. Relationships among stream order, fish Torgersen, C.E., Baxter, C.V., Li, H.W. & McIntosh, B.A. populations and aquatic geomorphology in an Idaho river 2006. Landscape influences on longitudinal patterns of river drainage. Fisheries 4: 5–9. fishes: spatially continuous analysis of fish-habitat relation- Pontius, R.W. & Parker, M. 1973. Food habits of the moun- ships. In: Hughes, R., Wang, L. & Wofford, J.E., eds. Influ- tain whitefish, Prosopium williamsoni (Girard). Transactions ences of landscapes on stream habitats and biological of the American Fisheries Society 102: 764–773. assemblages. Bethesda, MD: American Fisheries Society, Power, M.E. 1984. Depth distribution of armored catfish: pp. 473–492. predator induced resource avoidance? Ecology 65: 523– Trexler, J.C. & Travis, J. 1993. Nontraditional regression 528. analysis. Ecology 74: 1629–1637. R Core Development Team. 2009. R: A language and envi- Troffe, P.M. 2000. Fluvial mountain whitefish (Prosopium ronment for statistical computing. R Foundation for Statisti- williamsoni) in the Upper Fraser River: a morphological, cal Computing, Vienna, Austria. Available at: http://www. behavioural, and genetic comparison of foraging forms. R-project.org/. Vancouver: Masters of Sciences, The University of British Robinson, B.W. & Wilson, D.S. 1994. Character release and Columbia. 68 pp. displacement in fishes: a neglected literature. American Nat- USDA. 2009. Compressed county mosaic 1-m color ortho- uralist 144: 596–627. photo for Clallam County. Salt Lake City, UT: USDA-FSA Rodgers, J.D., Solazzi, M.F., Johnson, S.L. & Buckman, M.A. Aerial Photography Field Office, FSA National Agriculture 1992. Comparison of three techniques to estimate juvenile Imagery Program (NAIP). coho salmon populations in small streams. North American Welcomme, R.L. 1985. River fishes. FAO technical paper Journal of Fisheries Management 12: 79–86. 262. Rome, Italy: Food and Agriculture Organization of the Roni, P., Beechie, T.J., Bilby, R.E., Leonettie, F.E., Pollock, United Nations. M.M. & Pess, G.R. 2002. A review of stream restoration Whiteley, A.R. 2007. Trophic polymorphism in a riverine fish: techniques and a hierarchical strategy for prioritizing restora- morphological, dietary and genetic analysis of mountain tion in Pacific Northwest watersheds. North American Jour- whitefish. Biological Journal of the Linnean Society 92: nal of Fisheries Management 22: 1–20. 253–267. Sigler, W.F. 1951. The life history and management of Proso- Whiteley, A.R., Spruel, P. & Allendorf, F.W. 2006. Can com- pium williamsoni, in the Logan River, Utah. Utah State mon species provide valuable information for conservation? Agricultural College, Bulletin 347, Logan. Also available Molecular Ecology 15: 2767–2786. online at: http://digitalcommons.usa.edu/uaes_bulletins/308. Wimberger, P.H. 1994. Trophic polymorphisms, plasticity, Smith, C.J. 2000. Salmon and steelhead habitat limiting fac- and speciation in vertebrates. In: Stouder, D.J., Fresh, K.L. tors in the north coastal streams of WRIA 20. Lacey, WA: & Feller, R.J., eds. Advances in fish foraging theory and Washington State Conservation Commission. ecology. Columbia, SC: University of South Carolina Press, Smith, T.B. & Skúlason, S. 1996. Evolutionary significance of pp. 19–43. resource polymorphism in fishes, amphibians, and birds. Wood, S.N. 2001. mgcv: GAMs and generalized ridge regres- Annual Review of Ecology and Systematics 27: 111–113. sion for R. R-NEWS 1: 20–25. SYSTAT. 2006. SigmaPlot, version 10.0. Richmond, CA: Wood, S.N. 2003. Thin-plate regression splines. Journal of the SYSTAT Software Inc. Royal Statistical Society: Series B (Statistical Methodology) Thompson, G.E. & Davies, R.W. 1976. Observations on the 65: 95–114. age, growth, reproduction, and feeding of mountain white- Wood, S.N. 2006. Generalized additive models: an introduc- fish (Prosopium williamsoni) in the Sheep River, Alberta. tion with R. New York: Chapman and Hall/CRC. Transactions of the American Fisheries Society 105: 209– Wootton, R.J. 1998. Ecology of teleost fishes. London: Klu- 219. wer Academic Publications. Thurow, R.F. 1994. Underwater methods for study of salmo- Zar, J.H. 1984. Biostatistical analysis, 2nd edn. Englewood nids in the Intermountain West United States Forest Service. Cliffs, NJ: Prentice Hall. General Technical Report INT-GTR-307.