Effects of subsidies from spawning chum and pink salmon on juvenile coho salmon body size and migration timing

Organisms transporting nutrients from highly productive ecosystems can subsidize food webs and alter ecosystem processes. For example, the carcasses and eggs of migratory Pacific salmon (Oncorhynchus spp.) provide a high-quality food source that could potentially benefit other species of salmon rearing in fresh water. We investigated relationships between spawning chum (O. keta) and pink (O. gorbuscha) salmon density, and the body size and age of juvenile coho salmon (O. kisutch) in 17 streams on the central coast of British Columbia, Canada. Chum salmon density was the most consistently important and positive correlate of coho body size, in comparison with pink salmon density, juvenile coho salmon density, and numerous characteristics of habitats. This was shown by comparisons both among and within streams, and between sites above and below natural barriers to spawning chum and pink salmon. In addition, streams that had higher chum and pink salmon spawning densities had a higher proportion of age 0 coho (less age 1), suggesting earlier juvenile coho salmon migration to the ocean with increased spawning salmon nutrient availability. Most of the coho salmon sampled had little or no direct contact with spawning chum and pink salmon, which suggests an indirect, time-delayed influence on coho salmon


INTRODUCTION
Geophysical processes and organisms can transport nutrients across ecological boundaries, thus linking an array of environments, such as above-and below-ground terrestrial systems (Scheu 2001), sea ice and arctic islands (Roth 2002), and streams and forests (Nakano and Murakami 2001).Productive systems can subsidize nutrient-limited ones (Gravel et al. 2010), such as when nutrients move from the marine environment to desert islands (Spiller et al. 2010) and freshwater streams (Richardson et al. 2010).These subsidies can have a wide range of effects, including the growth and body size of organisms in recipient food webs (Marczak andRichardson 2008, Young et al. 2011).Growth and body size can affect migration timing (Giannico and Hinch 2007), fecundity (Wootton 1998), competitive and predatory ability (Vincenzi et al. 2012) and, ultimately, survival (Groot et al. 1995).
Transport and concentration of nutrients can occur both spatially, such as in avian nesting colony aggregation, and temporally, such as through annual migrations.One example that constitutes both a spatial and temporal aggregation of nutrients occurs through the annual migration of spawning salmon (Oncorhynchus spp.) along the temperate coasts of the northern Pacific Ocean.Because salmon gain .95% of their body mass in the ocean, return to freshwater to spawn and then die, the marine-derived nutrients they transport can be substantial to nutrient-poor freshwater streams and lakes (Naiman et al. 2002, Schindler et al. 2003, Janetski et al. 2009).While most research has focused on import of nutrients (e.g., Verspoor et al. 2011, Hocking et al. 2013), it is noteworthy that salmon can also drive export of nutrients from streams through the engineering effects of spawning adults, which can flush invertebrates downstream (Moore et al. 2007).In addition, nutrients are exported by young salmon migrating downstream toward the sea, especially if they have been feeding for some time in freshwater (Scheuerell at al. 2005).Thus, we cannot automatically assume that salmon cause a net increase in nutrients in any particular component of a freshwater ecosystem.
One group of organisms that can be affected by spawning salmon subsidies is other species of salmon, particularly species that stay in freshwater for many months before migrating to the ocean.For example, nutrients from salmon can contribute 20-40% of the nitrogen and carbon in stream-rearing juvenile coho (Bilby et al. 1996).This can come from direct consumption of adult salmon tissue and eggs (Kline et al. 1990) and indirectly through increased aquatic (Wipfli et al. 1998, Verspoor et al. 2011) and terrestrial invertebrates in the presence of spawning salmon (Hocking et al. 2013), which provide potential prey for juvenile salmonids.The presence of spawning salmon increased energy intake of juvenile rainbow trout (Scheuerell et al. 2007), and coho salmon (Heintz et al. 2004, Armstrong et al. 2010).Furthermore, nutrients from spawning salmon have been linked to improved condition and growth rate of juvenile coho in a number of carcass addition experiments (e.g., Bilby et al. 1998, Wipfli et al. 2010).However, bioturbation during redd-digging could also reduce food availability through reduced invertebrate biomass (Moore and Schindler 2008).
Most previous research has been limited to experimental carcass addition, which does not take non-carcass nutrients or engineering activities into account.No previous research has examined effects on juvenile coho salmon age composition, nor has there been consideration of the mediating effects of habitat characteristics that are known to affect juvenile salmonids (Tiegs et al. 2008).For example, juvenile coho salmon can be affected by habitat and food availability, cover or refugia from predation, and temperature (e.g., Bradford et al. 1997, Sharma andHilborn 2001).We predicted the density of overhead canopy may affect the degree to which spawning salmon nutrients subsidize primary productivity in a stream, or structural complexity may affect retention of nutrients.In addition, the presence and density of juvenile conspecifics can also affect growth (Roni and Quinn 2001).
Here, we provide the first investigation of the impacts of naturally occurring salmon-derived nutrients on size and age proportion of juvenile salmon.Specifically, we study the prediction that nutrients from spawning pink and chum salmon will lead to larger juvenile coho salmon, and higher proportion younger age classes, which may indicate earlier seaward migration timing by those fish.Whereas chum and pink juvenile salmon emerge from stream substrate and migrate to the ocean within weeks, juvenile coho salmon spend at least one year in freshwater streams (Groot and Margolis 1991).We predicted chum salmon would have greater positive effects on juvenile coho salmon body size than pink salmon due to their larger body size and egg deposition, although there is the potential for greater negative effects of chum than pink salmon through more bioturbation during nest digging due to the larger body size of chum salmon, as well as potentially aggressive behavior towards juvenile coho salmon during nest guarding (Nelson and Reynolds 2014).We incorporate the potential for each of these effects by looking at the number of chum and pink spawning salmon, and the potential for each species to have different effects by modeling them separately.Our study incorporates natural variation in numbers of spawning pink and chum salmon in 17 streams in a remote region of the central coast of British Columbia, Canada.We also make comparisons within four streams above and below barriers to spawning fish.Most of the coho salmon that we studied were young of year (age class 0) and had emerged in the spring just prior to sampling, and would therefore have had little or no direct exposure to spawning salmon, other than an overlap of a few weeks.Therefore, for those fish in age class 0, any impacts of spawning pink and chum salmon on juvenile coho salmon body size are indirect effects from previous spawning events.On the other hand, body size in age class 1 coho salmon would encompass both indirect effects as well as direct effects through consumption of spawning salmon tissues and eggs, and potentially chum and pink salmon fry.Age class 1 coho salmon may also be affected by spawning engineering activities.

Study sites and design
We surveyed streams on the central coast of British Columbia, Canada, in the Great Bear Rainforest (Table 1).The dominant spawning salmon species are chum and pink, and juvenile coho are present in all streams.Sites are accessible only by boat, and land use has been very limited in the area.Coho salmon spawn in the upper tributaries of the streams, whereas chum and pink spawn in the lower reaches.While we were unable to assess the densities of spawning coho salmon due to inherent difficulties in estimating spawning coho salmon abundance (Irvine et al. 1992), densities of spawning coho salmon at the five streams in our study area where data were available (50-204 females/km) exceed that which are thought to fully saturate the habitat with juveniles (19 females/km; Bradford et al. 2000).A consistent relationship between spawning and juvenile coho salmon has not been established.It is thought rather that juvenile coho salmon production is regulated by the availability of rearing habitat in the stream, unless spawning coho salmon densities are very low (Bradford et al. 1997).The number of spawning pink and chum salmon far exceeded the number of spawning coho salmon at the five streams with count data available for all three species (mean 3611, 1138 and 138 females/km for pink, chum and coho, respectively) and during visual observations during spawning at all of the streams in this study.At four of our sites a natural barrier to spawning chum and pink salmon was present, which spawning coho salmon were able to pass, resulting in juveniles on both sides of the barrier.
Study streams all flow directly into the sea, range from mid-gradient exterior coastal sites to lower gradient coastal fiords, and had bank full widths from 1.2 to 22.8 m.This region is in the Coastal Western Hemlock biogeoclimatic zone (Pojar et al. 1987), with forests dominated by western hemlock (Tsuga heterophylla), western red cedar (Thuja plicata), and Sitka spruce (Picea sitchensis).Riparian zones are dominated by red v www.esajournals.orgalder (Alnus rubra), salmonberry (Rubus spectabilis), salal (Gaultheria shallon), false azalea (Menziesia ferruginea), and blueberry (Vaccinium spp.).Annual precipitation in the region is pleasantly high, at 3,000-4,000 mm/yr.A map of the study area, including the streams studied here, is in Harding et al. (2014).
Juvenile coho salmon and physical habitat were studied in the fall (September-October) of 2007 and 2008.Spawning salmon counts were undertaken across the entire spawning length of the stream for returning chum and pink salmon from 2006 to 2011, to provide an overall index for comparing average differences among streams (methods in Hocking and Reynolds, 2011).Average stream width, calculated from three measurements taken at randomly chosen locations, was used to scale the length of area sampled for habitat characteristics (30 3 stream width).Twelve transects were randomly assigned within this area for each stream.

Environmental variables and juvenile coho density
We measured the following habitat characteristics that have been shown to affect body size and growth of juvenile coho salmon: stream bank full width, stream length, amount of large wood in streams, pools, pool to riffle ratio, undercut banks, gradient, canopy cover, percent fine substrate, pH, temperature and dissolved nutrients (ammonia, nitrate and soluble phosphorous).We also calculated the density of conspecific juvenile coho salmon, as explained below.This large variable set was reduced for model testing analyses (see Data analysis, below).
Stream bank full width is the maximum width without flooding, which was measured at 12 transects.iMapBC was used to calculate stream length (Field and Reynolds 2011).Large wood was included if it would be in the water at bank full, and was .10cm in diameter and .1.5 m long (Roni and Quinn 2001).Habitat types were identified as pool, riffle, glide or rapid (Bain and Stevenson 1999), and area measured for pool to riffle ratio.Pool depth was also measured at the deepest point for pool volume.Undercut banks were measured as a percentage of the length of stream banks on both sides, divided by 2. A clinometer was used to measure gradient at each transect, and a spherical densiometer to measure vegetative canopy cover on each side and the centre of the stream at each transect.Substrate was categorized into percentages of fines (0-12 cm), gravel (1.3-10.2cm), small cobble (10.3-14.9cm), large cobble (15.0-24.9cm), boulder (.25.0 cm) or bedrock (Wolman 1954).Water pH was measured at three transects per stream each year, and the mean between years ranged from 4.8 to 6.9.Maximum weekly averaged temperature was measured over two years using data taken every two hours from ibutton data loggers (DS1922L).Fisheries and Oceans Canada Cultus Lake Research Facility analyzed three water samples collected from each stream prior to and during spawning for ammonium (NH 3 þ ), nitrate (NO 3 ) and soluble reactive phosphorous according to American Public Health Association methods (APHA 1989).
To calculate juvenile coho salmon density, triple-pass depletion was completed with a twometer wide pole seine.Sections were chosen randomly within the area sampled for environmental variables with a seine section length of 8 3 stream bank full width, at a location randomly chosen within the area sampled for habitat characteristics.Stop nets were used at the upper and lower ends of the sampling area, and sampled areas were left undisturbed for a minimum of one hour between passes.Density was calculated using maximum likelihood modeling (Schnute 1983).

Spawning chum and pink salmon density
At six of the 17 streams in this study, on-foot visual estimates of spawning salmon abundance were available for spawning chum and pink salmon from Fisheries and Oceans Canada.Additional on-foot stream counts were used for the remainder of sites, undertaken in partnership with the Heiltsuk First Nation's Integrated Resource Management Department.During the period 2006-2011, all streams were counted for a minimum of two years and up to six years.An average of these values was used to get a general characterization of each stream.Exploratory AIC c modeling did not distinguish a difference between using mean 2006-2011 spawning chum and pink salmon densities compared to using individual years, therefore only mean (2006)(2007)(2008)(2009)(2010)(2011) densities are reported.
Three or more salmon counts were completed at each stream during spawning.For most streams, total abundance was estimated using the area-under-the-curve method (English et al. 1992).Peak counts (live þ dead) were used for some streams that were not accessible three times during the spawning season (less than 10% of streams).At a subset of streams using both methods, there was no difference in spawning salmon calculations (Hocking and Reynolds 2011).Estimates of the total number of fish were divided by stream spawning area to calculate spawning salmon density (chum and pink density per m 2 ) to account for differences among streams in the length where spawning occurred.The stream length available for spawning was measured during visual spawning salmon counts and multiplied by average stream width to estimate stream spawning area.

Juvenile coho salmon body size and age determination
All coho salmon collected in pole seines were sampled for body size by measuring fork length, or the distance from tip of snout to fork in tail.
Scales were collected from five fish per stream (3 scales per fish) during each sampling event to determine age in order to categorize fish into age classes.The majority (84.4%) of fish sampled were young of year (age class 0) and the remainder age class 1. Coho salmon in age class 0 would have hatched in the spring prior to sampling (i.e., 2-6 months old) and had little or no direct contact with spawning chum and pink salmon at the time of sampling.Coho salmon in age class 1 would have had access to chum and pink salmon nutrients in the fall of their first year.
We tested for an effect of sampling date on juvenile coho salmon response variables, as streams were sampled over a period of four to six weeks.No effect was found, therefore sampling date was not considered in further analyses.

Data analysis
We used two approaches to assess the effect of spawning chum and pink salmon densities on juvenile coho salmon body size.First, we used information theoretic and partial correlation approaches to investigate correlations between chum and pink salmon densities and young of year (age 0) and age 1 coho salmon body size, compared to a broad suite of habitat variables across a gradient of spawning salmon densities at 17 streams.We used Akaike's information criterion adjusted for small sample sizes.AIC provides a comparison of model fits that includes a penalty for models with larger numbers of parameters to be estimated (Burnham and Anderson 1998).We also examined a linear regression of age composition at the same streams.Second, we used natural barriers including waterfalls in streams, which excluded chum and pink but not coho salmon from upstream locations to test paired sites with and without spawners for 4 streams.
In our first analysis, given the number of streams we surveyed and the large number of potentially inter-related environmental variables assessed, we conducted an exploratory analysis informed by a priori hypotheses to identify the habitat or coho salmon density variables that best described each coho salmon body size response variable across the 17 streams, using AIC c (AIC adjusted for small sample sizes) according to the methods suggested by Zuur et al. (2010).We retained only the top 2 habitat variables based on DAIC c values for each response variable to avoid over-fitted models and uphold the principle of parsimony (Burnham and Anderson 2002).These included stream width and pool volume for age 0 body size, and canopy cover and undercut banks for age 1 body size.We used variance inflation factor (VIF) to test for multicollinearity among the explanatory variables (Zuur et al. 2010).For the final models, no variable exceeded a value of two, which suggests multicollinearity among variables was not of concern.
We then assessed the relative importance of pink salmon density, chum salmon density, and the top habitat features as identified in exploratory AIC c analysis as explanatory variables on age 0 and age 1 body size.Because we predicted that the behavioral and ecological effects of pink and chum spawning salmon would differ between species, we retained them as separate parameters.However, an exploratory combined analysis showed similar results, with the strength of combined effects somewhere in between the effects of each species individually.Linear models were constructed to represent our a priori hypotheses.Because we hypothesized the density of canopy might mediate the effects of spawning salmon on juvenile coho salmon through affecting primary productivity, and pool volume and undercut banks might affect carcass retention, we included interaction terms for them.However, initial analyses prior to final AIC testing revealed interactions were not important.Therefore they were excluded from the final analysis.A null model was included in each candidate set, and we included year as a fixed effect to account for systematic differences between the two years of data.
After selecting our independent variables and interactions based on a priori hypotheses and exploratory analyses (Zuur et al. 2009(Zuur et al. , 2010)), we created a set of models in all combinations limited to a maximum of three variables per model to avoid over-fitting (Burnham and Anderson 2002).We used mixed models to incorporate the hierarchical nature of our data set, including individual body size data for juvenile coho salmon and stream scale data for environmental variables.As such, stream was included as a random effect, while year was included as a fixed effect.We computed candidate models using maximum likelihood estimation (Zuur et al. 2009), and inspected diagnostics for heteroscedasticity, over-leveraging of data points, and normality and independence of residuals.To rank the relative importance of explanatory variables, we used model averaging with summed model weights incorporating all candidate models (Anderson 2008).DAIC c values, or the difference between model i and the top ranked model, are reported for all models with DAIC c , 3 for reference but only those with DAIC c , 2 are discussed (Burnham andAnderson 2002, Grueber et al. 2011).We then used partial correlation analysis to determine the unique contribution of chum and pink salmon density on juvenile coho salmon age 0 and age 1 body size after taking the effect of the top habitat variables (identified by AIC c , as above) into account (Cohen et al. 2003).
We also examined the effect of spawning chum and pink salmon density on the proportion of age 0 compared to age 1 juvenile coho salmon across the 17 streams.Because there was no significant difference between years (2007 and 2008) for proportion age 0 ( p .0.05), we pooled the data to increase the number of juvenile coho salmon at each stream used to calculated age proportion (maximum five fish per stream per year).We used an arcsine square root transformation of the ratio of age 0 to age 1 coho salmon and a log transformation of chum and pink salmon density, and used AICc to compete linear regression models in the same way we did for body size.The top habitat variables included for age proportion analysis were gradient and large wood.
For the second part of our analysis, we compared age 0 juvenile coho salmon body size at paired locations above and below a natural barrier to chum and pink salmon spawning in 2008 in each of four streams.Below the barrier juvenile coho salmon had access to chum and pink salmon nutrients, while above the barrier coho salmon were present because their parents are able to pass the barrier during spawning, but the juvenile coho salmon had no access to chum and pink nutrients.There were no significant differences in the habitat variables identified above and below barriers ( p .0.05).We used t-tests to compare differences in juvenile coho salmon body size above and below barriers at each site.We then used linear regression to examine the relationship between the magnitude of difference in juvenile coho salmon body size above and below barriers, and chum and pink salmon biomass density below barriers at each stream.We were unable to separate the effect of salmon species nutrient deposition, either chum or pink, at the sites with spawning salmon (below barrier) and without (above barrier) in this comparison, thus we combined pink and chum salmon densities for the second part of the analysis.Due to difference in body size and thus nutrient load, we used salmon biomass density based on mean values of chum salmon body mass 3.5 kg, and pink salmon body mass 1.2 kg (Hocking and Reynolds 2011).
All statistical analyses were performed using R (R Development Core Team 2009), including the MuMIn package (Barton 2012).

RESULTS
The body size of the younger age class (age 0) of coho salmon was correlated more strongly with spawning chum salmon density than with pink salmon density (r-squared ¼ 0.31 and 0.02 for chum and pink, respectively; Table 2).The positive correlation was also stronger than with v www.esajournals.organy of the habitat variables, or juvenile coho salmon density.An additional 1/4 chum salmon per m 2 was associated with a full centimeter increase in age 0 coho salmon body length (Fig. 1).The model containing spawning chum salmon density in combination with stream width and pool volume was the top model, which improved r-squared by 0.2 compared to the next top model, which was chum salmon by itself (DAIC c , 2, relative importance 0.87; Table 3).The resulting relationship between chum salmon density and juvenile coho salmon body size was still relatively strong after taking the effect of the top habitat variables into account, and there was no relationship with pink salmon density (partial rsquared ¼ 0.28 and 0.01, respectively).Chum salmon density was the only variable in the top model for size of age 1 coho salmon, whereas pool volume and stream width joined chum salmon density in the top model for age 0 coho salmon (Table 3, Fig. 2).These relationships remained the same when habitat variables were taken into account (partial r-squared ¼ 0.28 and 0.36 for age 0 and age 1 body size, respectively).There was a trend towards a stronger relationship between chum salmon density and juvenile coho salmon body size in age 1 coho salmon compared to age 0 (Fig. 1).Thus, juvenile coho salmon that had had direct contact with spawning chum the previous fall, and had longer in streams for indirect effects to occur, had stronger relationships with spawning chum salmon than juvenile coho salmon that had very little direct contact.Our results suggest that pink salmon density has a positive effect on age 1 coho salmon, though the parameter estimates overlap zero (r-squared 0.23, relative variable importance 0.32; Fig. 2).This relationship was stronger once the top habitat variables were taken into account (partial r-squared ¼ 0.32 compared to 0.23).There was no strong effect of habitat, either canopy cover or undercut banks (Fig. 2).
In our analysis above and below barriers to spawning chum and pink salmon, age 0 juvenile coho salmon were significantly larger below the barriers at the two sites with the highest spawning salmon biomass density (Fig. 3).As spawning salmon biomass density below barriers increased across the four streams, the difference in body size of juvenile coho salmon also increased (r-squared ¼ 0.82; Fig. 3).
Streams that had more chum and pink salmon had more age 0 compared to age 1 coho salmon (r-squared ¼ 0.29 and 0.28 for pink and chum salmon respectively, p , 0.03; Fig. 4).The effect of chum and pink salmon on proportion age 0 to age 1 coho salmon was stronger than any habitat variable or juvenile coho salmon density (Table 3, Fig. 5).The untransformed data showed an asymptotic relationship, where the proportion of age 0 coho salmon approached 1, or 100%, at fairly low spawning chum and pink salmon densities (0.15 and 0.2 fish/m 2 , respectively; Fig. 4).
The relationships between habitat variables and each of the three salmon species (Table 2) were weaker than the relationships between coho  v www.esajournals.organd the two species of spawning salmon (Table 3).As expected, the dissolved inorganic nitrogen (ammonia and nitrate) and soluble reactive phosphorus in the streams during spawning in fall were correlated with the density of spawning chum and pink salmon (Table 4).However, these relationships generally did not persist through the non-spawning season to summer (Table 4), nor were dissolved nutrients among the top habitat variables for juvenile coho salmon body size in the AIC c analyses (not shown).In addition to the habitat variables considered in the AIC c analyses, chum salmon density was somewhat correlated with the percentage of the substrate that was small cobble (r-squared ¼ 0.21), and pink salmon density with gravel (r-squared ¼ 0.18).These substrate characteristics were not correlated with coho salmon body size (r-squared , 0.1).This suggests that relationships between coho salmon and the other salmon species were not being driven by separate responses to habitat features.

DISCUSSION
We found larger juvenile coho salmon in streams with higher densities of spawning chum salmon, and larger juvenile coho salmon below   v www.esajournals.orgnatural barriers to spawning chum and pink salmon compared to above barriers at high spawning salmon density streams.For age 0 coho salmon, these positive impacts are due to indirect effects.We also found that higher densities of both chum and pink salmon resulted in proportionally fewer age 1 coho salmon compared to age 0, suggesting higher spawning salmon nutrients may result in earlier seaward migration of juvenile coho salmon.
Our findings suggest an indirect carry-over effect from previous spawning events because the majority of coho salmon we sampled were young-of-the-year fish that would have had little to no access to spawning salmon nutrients at the time of sampling.The relationship between juvenile coho salmon body size for age class 1 and spawning chum salmon was stronger than for young-of-the-year, which may indicate a potential additional benefit from direct access to spawning salmon nutrients or accumulated indirect effects over a longer time period.Our comparison of differences above and below a barrier to spawning pink and chum salmon support the among-stream comparisons, indicating a positive effect of nutrients from spawning fish on juvenile coho salmon body size, with the magnitude of the benefit increasing with the density of spawning fish.
Indirect effects on young-of-the-year coho salmon may have come through increased aquatic and terrestrial invertebrate prey availability linked to spawning salmon (Wipfli et al. 1998, Verspoor et al. 2011, Hocking et al. 2013).These resources are readily used by juvenile salmonids (Scheuerell et al. 2007, Denton et al. 2009).A concurrent study of many of the same streams that we used found spawning salmon biomass predicted primary productivity better than habitat characteristics, and that aquatic invertebrates used both nitrogen and carbon resources from spawning salmon (Harding and Reynolds 2014).While dissolved nutrients may be a key player for this bottom-up mechanism, they were not strongly related to juvenile coho salmon body size.Nutrients may be taken up by primary and secondary producers or are flushed out of the stream.Further study on nutrient and food web dynamics would be helpful to explicitly elucidate the mechanisms behind the relationships described here.
Our results suggest that body size in age 1 coho salmon had stronger relationships with adult chum salmon than body size in young-ofthe-year, which may reflect greater benefits when salmon nutrients are available directly to the juvenile fish, through preferential diet switching to eggs and tissue (Hicks et al. 2005, Scheuerell et al. 2007).These diets can have dramatically improved energy rations compared to diets not containing eggs (Armstrong et al. 2010).Salmon eggs are 2-3 times more energy dense than benthic invertebrates (Moore et al. 2008).This may explain why salmon subsidies have been shown to have strong effects in stream food webs even though they are available for a short period of time.Furthermore, older age classes of juvenile coho salmon can prey upon newly hatched pink and chum salmon fry (Hunter 1959), as well as invertebrates that have been stirred up by adult salmon digging nests an fighting for space and mates.They may also benefit from blowfly larvae on salmon carcasses in streams, which are a preferred food source for juvenile salmonids (Scheuerell et al. 2007, Denton et al. 2009).
We found few to no age 1 coho salmon in streams at the upper range of spawning chum and pink salmon densities (Fig. 4).This matches our prediction that nutrient subsidies and resulting larger coho salmon body size could lead to migration from the stream to the ocean at an earlier age.Further, our data suggest a threshold effect of spawning salmon density where the majority of age 0 coho salmon migrate to the ocean rather than remaining in freshwater for an additional year, and this threshold is fairly low within the range of the streams included in this study (Fig. 4).Although other studies have found stream temperature to affect whether coho outmigration occurs in a given year (e.g., Spence and Dick 2013), we did not find this to be the case.Instead, chum and pink densities were Previous research has also found a positive effect of spawning salmon nutrients on juvenile salmonids, with the majority of studies utilizing experimental carcass additions.For example, carcass addition positively affected juvenile coho salmon body condition (Bilby et al. 1998, Wipfli et al. 2010), juvenile coho salmon mass and body size (Wipfli et al. 2003), juvenile coho salmon growth (Lang et al. 2006, Giannico and Hinch 2007, Wipfli et al. 2010), and biomass of juvenile Atlantic salmon (Williams et al. 2009).On the other hand, two studies did not find positive effects of carcass addition on juvenile cutthroat trout and steelhead: specific growth rate was less with carcasses than without (Wilzbach et al. 2005) and growth did not change with carcass addition (Harvey and Wilzbach 2010).Notably, this study design does not take the full effect of spawning salmon into account (Tiegs et al. 2011).For example, many of these studies do not include eggs, which are preferred by juvenile salmonids (Hicks et al. 2005, Scheuerell et al. 2007).Exceptions that did include egg provision were studies by Wipfli et al. (2010) and Lang et al. (2006).In addition, carcass experiments do not include the effect of dissolved nutrients through excretions, or the potential engineering effects of spawning activities (Moore and Schindler 2008).A study using stable isotopes has shown that juvenile coho salmon were not able to take up significant amounts of marine-derived nitrogen from sites with only carcass additions, whereas they were enriched in salmon nutrients from sites with naturally occurring spawning salmon, which would have included the combined effects of carcass, egg and excretory nutrient benefits and engineering activity (Shaff and Compton 2009).
Our findings complement those by Rinella et al. (2012), who showed increased growth rate in juvenile coho salmon, as indexed by RNA-DNA ratios, across 11 streams of increasing naturally occurring spawning salmon.Although the authors showed carry-over effects into the nonspawning season, we are the first to show an entirely indirect effect of spawning salmon on juvenile coho salmon body size by studying age 0 coho salmon.Another study looking at naturally occurring spawning salmon found increased growth rate in dolly varden in seven ponds increasing in spawning salmon biomass (Denton et al. 2009).Our study is the first to separate effects by age class (including age 0 with no direct contact and age 1 with direct contact with spawning salmon), to examine effects on coho age composition, and to include the comparative influences of habitat characteristics.Contrary to our expectation and indications from previous research (Tiegs et al. 2008, Armstrong et al. 2010), we found habitat characteristics did not mediate the relationship between spawning chum and pink, and juvenile coho salmon.This may be related to a comparatively high density of spawning fish obscuring any effects of habitat.
We attempted to address the potential issue of spurious results in our correlative study by taking a broad range of habitat variables into account explicitly and analyzing them with information theoretic and partial correlation approaches.For example, a spurious correlation may come out if all three species of salmon respond similarly to an unmeasured habitat characteristic.We included stream width at bank full, stream length, large wood, pools, pool:riffle ratio, undercut banks, gradient, canopy cover, percent fines, pH, substrate, temperature, dissolved nutrients and the density of conspecifics, and found the relationships between spawning chum and pink salmon and juvenile coho were stronger than those between any of the three salmon species and habitat characteristics.
This study suggests that spawning salmon can have positive effects on other species of juvenile salmonids.Since growth and production of stream-rearing salmonids can be limited by food availability (Chapman 1966) these findings imply that cross-boundary nutrient inputs may be important for fisheries.Understanding these cross-species interactions can also help inform ecosystem-based management (Bilby et al. 2001, Wipfli and Baxter 2010, Levi et al. 2012).

Fig. 1 .
Fig. 1.Relationships between the density of spawning chum and pink salmon and juvenile coho salmon age 0 body size (top), and age 1 body size (bottom).Each data point represents a stream, in either 2007 or 2008.

Fig. 2 .
Fig. 2. Scaled model parameter estimates (circles) with 95% confidence intervals (lines) from averaged predictive linear models describing age 0 coho salmon body size (top), and age 1 coho salmon body size (bottom).The variables are ordered from the highest positive scaled coefficient value to lowest negative value.The relative importance of variables to the averaged model (indicated on the right) is scaled from 0 to 1.

Fig. 3 .
Fig.3.Relationship between spawning chum and pink salmon biomass density and the difference in body size of juvenile coho salmon above and below barriers to spawning chum and pink salmon.Asterisks denote streams with significant differences in juvenile coho salmon body size above and below barriers.

Fig. 4 .
Fig. 4. Relationship between the density of spawning chum and pink salmon and proportion of age 0 juvenile coho salmon.Each data point represents a stream, in either 2007 or 2008.

Fig. 5 .
Fig. 5. Scaled model parameter estimates (circles) with 95% confidence intervals (lines) from averaged predictive linear models describing age composition (proportion age 0 coho salmon).The variables are ordered from the highest positive scaled coefficient value to lowest negative value.The relative importance of variables to the averaged model (indicated on the right) is scaled from 0 to 1.

Table 1 .
Stream characteristics, spawning salmon chum and pink salmon population data (2006-11), and juvenile coho salmon density and body size (fork length) at ages 0 and 1 for the 17 streams in this study.Sample sizes of fish measured are in brackets.

Table 2 .
Bivariate correlations, r, between variables used in the AIC analyses with data from 2007 and 2008.For age 0 juvenile coho salmon body size, n ¼ 17 streams; and for age 1, n ¼ 7 streams for each year.

Table 3 .
Summary of Akaike's information criterion linear regression models with the greatest support for body size of age 0, body size of age 1, and proportion age 0 to age 1 juvenile coho salmon.Note: K is the number of model parameters, R 2 is the model correlation coefficient, DAIC c of model i is the change in model i AIC c score from the top model, w i is the AIC c model weight.

Table 4 .
Bivariate correlations, r, between individual nutrient variables and spawning chum and pink salmon density.