Trophic ecology drives spatial variability in growth among subpopulations of an exploited temperate reef fish

Abstract

Blue cod (Parapercis colias) support important fisheries around New Zealand, but limited mixing over small spatial scales renders subpopulations vulnerable to depletion and highlights the importance of resolving fine-scale population structure. Fiordland represents a rare opportunity to achieve this in a relatively pristine environment and intact ecosystem. Between 2000 and 2004, spatial population structure in three fjords was used to test how spatial variability in growth between inner and outer fjord subpopulations can be statistically explained by diet, demography and site-specific factors. Sex ratios and size distributions vary between habitats, with a female bias and a higher frequency of individuals in the larger size classes within the inner fjords. Stomach content analysis shows that in the inner fjords the diet of blue cod includes a diverse array of benthic species while in the outer fjords P. colias is able to utilise a less diverse array of pelagic prey that may be of a higher nutritional quality. Stable isotope analysis reveals that in two of the three outer fjord habitat sites the Laminarian Ecklonia radiata accounts for a large component of basal organic matter to the local population (66–95%). Consequently, fish condition and individual growth rates are high in these kelp-bed fuelled outer fjord habitats. Our findings demonstrate that baseline variability in the nutritional landscape can result in spatially structured populations, highlighting the importance of spatially-explicit management measures in the sustainable exploitation of reef fishes.

Keywords:

• temperate reef fish
• individual growth
• trophic ecology
• stable isotopes
• stomach contents analysis
• Fiordland
• blue cod
• Parapercis colias
• New Zealand

Introduction

Demography and vital rates, such as individual growth and fecundity, interact to influence a population's productivity. For exploited species an understanding of spatial variability in these processes is essential to understand population dynamics, predict sustainable yields and to ensure the efficacy of management. To interpret spatial variability in population structure and dynamics in terms of underlying mechanisms requires an understanding of the role of the nutritional landscape in determining a population's trophic ecology. Recognising that exploited species form integral parts of, and rely on provision of organic matter from, complex food webs is therefore essential from an ecosystem management and fisheries management perspective. It has been well recognised that spatial variability in vital rates and productivity constitutes a major management and conservation issue for invertebrate populations (Orensanz & Jamieson 1998). For example, population structure of grazing and suspension feeding invertebrates is subject to gradients in productivity across ecotone boundaries (Wing 2009; Wing & Leichter 2011), but the effects of such extrinsic structuring processes at higher trophic levels have rarely been considered. Here we investigate these issues in an exploited temperate reef fish.

Blue cod (Parapercis colias: Pinguipedidae, Forster 1801) are endemic to New Zealand where they support major commercial and recreational fisheries. Studies into the fine-scale movement of adults have shown them to be largely sedentary (Cole et al. 2000) and territorial (Mutch 1983) with evidence to suggest a decoupling of populations and limited mixing over relatively small spatial scales in enclosed waterways (Cole et al. 2000; Carbines & McKenzie 2004). In Dusky Sound—one of a series of 14 glacially-carved fjords in southwestern New Zealand, collectively known as Fiordland—65% of tagged adults moved less than 1 km from their release site during 17 months at liberty; larger movements were generally in an up-fjord direction, generating the hypothesis that the more mobile and interactive outer fjord and open coast populations act as a source draining up to 10% annually into the inner fjord sink population, where residency was estimated at 100% (Carbines & McKenzie 2004). There is some evidence to suggest a decoupling of population structure between habitats in Fiordland, with inner fjord populations typically female-dominated and including a higher proportion of larger, older individuals than the male-dominated outer fjord populations (Rodgers 2005). Inner fjord habitats are generally less productive than the wave-washed outer coast, resulting in nutritional stress and depressed growth rates in some species. For example, sea urchin populations in the inner fjords have relatively low growth rates and are subject to periodic mass mortality, resulting in greater demographic variability than those on the open coast (Wing 2009). Similarly, plasticity in growth and morphology among adjacent populations of sea perch in Fiordland reflects local habitat quality and food resources (Lawton et al. 2010), demonstrating the potential for the environment to strongly influence the productivity of higher trophic level vertebrates in this system.

Ranging over 200 km of remote coastline, the World Heritage and National Park protection afforded to the Fiordland region has ensured that anthropogenic impacts have been minimal and the natural environment has been preserved in a remarkably pristine condition. The steep topography, dense Nothofagus beech forest cover and high orographic rainfall result in substantial inputs of terrigenous organic material into the fjords, representing an unusually high degree of connectivity between terrestrial and marine ecosystems (McLeod & Wing 2009). The estuarine circulation pattern and entrance sills act to retain a significant portion of this material within the deep basins (Glasby 1978; Pickrill 1987) where it has been found to fuel a chemoautotrophic food web (Wing et al. 2008; McLeod et al. 2010). This is particularly prevalent in the inner fjord basins where terrestrial inputs are greatest. In Doubtful Sound the outflow of freshwater from a hydroelectric power station has altered the salinity regime and reduced or eradicated populations of filter-feeding bivalves, thereby reducing the availability of more labile heterotrophic carbon (Tallis et al. 2004; Rutger & Wing 2006; McLeod & Wing 2008) and increasing the importance of recycled terrestrially-derived organic matter to higher order consumers including blue cod (Rodgers & Wing 2008), red rock lobsters (Jack et al. 2009) and hagfish (McLeod & Wing 2007). It is not yet known to what extent this chemosynthetically-derived energy features in the food webs of more pristine fjords, although recent evidence suggests that in marginal inner fjord habitats it may provide a mechanism for the persistence of blue cod populations (Wing et al. 2012). In the inner fjords the low light and flow environment restricts macrophyte growth to a thin band of freshwater/estuarine species in the upper few metres, while in the wave-exposed outer fjord habitats, productive and diverse phytoplankton and macroalgal assemblages support kelp forest communities (Goebel et al. 2005; Miller et al. 2006; Wing et al. 2007). These processes result in a large difference in productivity and composition of basal organic matter sources within food webs between inner and outer fjord habitats.

Stomach content analysis has shown that P. colias, a demersal species, feeds on a wide vertebrate and invertebrate prey base, with a small direct intake of macroalgae and detritus (Thomson 1891; Graham 1939; Mutch 1983; Jiang & Carbines 2002). One of the major limitations of dietary studies based on stomach content analysis is that one cannot determine whether diet composition at the time of sampling reflects the availability of different prey types or a strong selectivity for a particular prey type. Integrating across several sampling events and employing stable isotope analysis in parallel helps resolve this issue by giving an indication of the longer-term importance of broad dietary components to a population. In the Doubtful–Bradshaw Sound complex, stable isotopic evidence for differences in diet and organic matter sources in blue cod from inner and outer fjord habitats lends support to the hypothesis that adult blue cod in Fiordland form discrete subpopulations at the scale of tens of kilometres, with a high degree of residency over months to years (Rodgers & Wing 2008). δ¹³C and δ¹⁵N are more deplete in ¹³C and ¹⁵N and more variable among the inner fjord subpopulation than the outer fjord and open coast populations, suggesting a significant assimilation of recycled carbon and wider individual variability in trophic position among fish in inner fjord habitats (Rodgers & Wing 2008).

Since 2005, the marine environment of Fiordland has been managed using a network of marine protected areas introduced under the Fiordland Marine Management Act 2005; the internal waters of all fjords are closed to commercial fishing and Doubtful Sound and Milford Sound—the two most accessible fjords—are closed to recreational fishing for blue cod. In addition, 10 marine reserves distributed among eight of the fjords afford full no-take protection to areas representative of the major habitat types.

The observed spatial separation in habitat quality, blue cod population structure and trophic ecology between subpopulations in inner and outer fjord habitats raises interesting questions regarding the resilience of subpopulations to localised depletion and the productivity of inner fjord populations. Previous ecological studies in Fiordland have tended to focus on Doubtful Sound, where the modified salinity regime is known to have altered carbon routing through the inner fjord ecosystem, or have compared between sites subject to differential management practices (Jack et al. 2009; Jack & Wing 2010; Wing 2011; Wing et al. 2012). Data collected before the introduction of the Fiordland Marine Management Act 2005 allow us to investigate the baseline ecological processes that influence population structure and dynamics. Focusing on the more pristine southern fjords, we employ an orthogonal sampling design with paired inner and outer fjord sites in Breaksea Sound, Dusky Sound and Long Sound–Preservation Inlet to, firstly, identify and characterise differences in growth rate between inner and outer fjord subpopulations of P. colias individuals and, secondly, to ask: in the absence of habitat modification and differential management, are these differences best statistically explained by diet, demographics or site-specific factors? By testing ecological processes independently of anthropogenic influences, greater insight can be gained into the forces that structure natural populations—forces that fisheries management models must account for if yields are to be sustainable. Furthermore, efforts to regenerate exploited stocks and allow populations to recover can only be successful if these baseline ecological processes are understood.

Materials and methods

Sample collection

Blue cod were sampled between 2000 and 2004 at paired inner and outer fjord sites in Breaksea Sound (inner: n=33; outer: n=33), Dusky Sound (inner: n=33; outer: n=38) and Long Sound–Preservation Inlet (inner: n=50; outer: n=50) (Fig. 1). A standardised experimental design was employed with all samples collected during austral spring (October–November), using hand lines with artificial lures and size 6/0 hooks to target adults and reduce by-catch (Carbines 1999). Fish were euthanised before being measured (±1mm total length, TL), weighed (whole blotted wet weight) and macroscopically sexed. Sagittal otoliths were removed, cleaned of adhering cerebral tissue and left to air dry for 48 h in unsealed centrifuge tubes. Stomachs were removed by abdominal dissection, blotted and weighed before the contents were transferred to labelled screw-top sample jars and fixed in a 10% seawater-buffered formalin solution. As very few fish sampled had empty stomachs (<1%) these were excluded from stomach content analyses without further consideration. Gutted weight was calculated as (wet weight – visceral cavity contents weight). Fulton's condition factor (Ricker 1975) was calculated as:

1

Figure 1  Location of study sites in southern Fiordland.

where W is the gutted weight (g) and L is the total length (mm).

Stomach content analysis

Fixed stomach contents were rinsed over a 500 µm mesh and transferred to a 70% ethanol solution. Prey items were sorted into broad taxonomic groups (pisces, zooplankton, Echinodermata, Mollusca, Crustacea, other benthic invertebrates, macroalgae and detritus) under a dissecting stereomicroscope and the blotted wet weight of each group was recorded. This was then expressed as a percentage of the total wet body weight of the fish (% BW). Groups were classified as representing either a ‘pelagic’ (zooplankton) or ‘benthic’ feeding strategy (Echinodermata, Mollusca, Crustacea, other benthic invertebrates, detritus). Pisces remains were not identified to a lower taxonomic level and could not therefore be assigned to either feeding strategy, as both demersal and pelagic species are common in the fjords; macroalgal matter was not assigned to either strategy due to the possibility of drifting macroalgal debris being consumed from the water column or anchored macroalgae being encountered during benthic grazing.

Multivariate stomach content data were pooled across years to provide a time-integrated overview of the diet rather than a ‘snapshot’ of one moment in time. Data were presence–absence transformed and a Bray–Curtis resemblance matrix was constructed. A presence–absence transformation was selected as variability in the weight of items in each taxonomic group and differing degrees of digestion would lead to bias in the inferred importance of each group to an individual's diet. Sampling of stomach contents was opportunistic and no consideration was given to the time of day, state of the tide, etc. Stage of digestion would therefore likely be an issue when comparing stomach content weights between sampling events. With presence–absence data the metric of abundance becomes the incidence of occurrence of a prey type among individuals within a population (Hynes 1950) and reflects population-wide feeding habits (Cailliet 1977), although small, highly abundant prey items are likely to be over-emphasised relative to less abundant, larger prey items.

The % BW of each dietary group was scaled to a proportion of the total stomach % BW before Shannon–Wiener diversity, H′ (Shannon 1948) was calculated for each individual as:

2

3

where S is the number of dietary groups, p _i is the relative contribution of each group to the total stomach weight, n _i is the % BW of group i and N is the total % BW of all groups.

Stable isotope analysis

In 2004, a subsample of fish from each site were selected for stable isotope analysis (n=68); dorsal muscle tissue was sampled from behind the head, rinsed in deionised water and frozen at − 20 °C in sterile centrifuge tubes. Muscle tissue samples were defrosted and oven-dried at 60 °C for 72 h. Dry samples were homogenised using a pestle and mortar, which was rinsed between samples with deionised water and dried with lint-free tissue. Rodgers and Wing (2008) found that lipids do not need to be extracted prior to stable isotope analysis of blue cod muscle tissue. However, an independent check was conducted for our data set given the concerns raised by Post et al. (2007) that failure to account for differences in lipid content among consumer populations can introduce considerable bias when interpreting patterns based on δ¹³C, particularly when variation is <10‰–12‰. As individual blue cod C:N ratios did not differ between fjords, habitats or sites (P>0.1), stable isotopic data were not normalised for lipid content. 0.1mg samples were weighed into 5×3.5 mm tin capsules and analysed using a Europa Scientific Hydra 20/20 continuous flow isotope ratio mass spectrometer coupled to a Carlo Erba NC2500 elemental analyser by Iso-Trace New Zealand Ltd (Dunedin, New Zealand). The natural isotope abundance of ¹³C and ¹⁵N was expressed in delta (δ) notation, or parts per thousand difference from a standard, PeeDee Belemnite for δ¹³C and air for δ¹⁵N (precision: 0.2‰):

4

where X is the heavier isotope and R is the corresponding isotopic ratio in the sample or standard. Half-lives of δ¹³C and δ¹⁵N in fish muscle tissue may be many months (Hesslein et al. 1993); turnover rates of blue cod muscle tissue result in a stable isotopic signature that integrates across a similar time period to that represented by stomach content data (Rodgers & Wing 2008; Suring & Wing 2009).

Mass balance mixing models

The proportional contributions of n + 1 sources to the diet of a consumer can be estimated with n isotope tracers using a linear mixing model based on mass balance equations (Phillips & Gregg 2003). Stable isotopic signatures of the major organic matter source pools in the system have previously been characterised for Doubtful Sound (Cornelisen et al. 2007; Wing et al. 2008): a suspended particulate organic matter (SPOM) signature was derived from vertical plankton tow samples collected during a phytoplankton bloom; signatures for Ulva pertusa and Ecklonia radiata (respectively the dominant estuarine/freshwater and marine macroalgal producers in the system) were integrated from samples taken along the fjord axis to control for spatial variability in environment and oceanography that could result in heterogeneous δ¹³C values (Cornelisen et al. 2007), and a terrigenous organic matter (TOM) signature was obtained from forest litter sampled from inner Doubtful Sound. Source pool isotopic signatures (mean±standard error) were: SPOM δ¹³C=− 21.3±0.18, δ¹⁵N=6.6±0.36; estuarine macroalgae (U. pertusa) δ¹³C=− 15.7±0.09, δ¹⁵N=3.5±0.06; marine macroalgae (E. radiata) δ¹³C=− 18.6±0.05, δ¹⁵N=5.9±0.12 and TOM δ¹³C=− 13.2±1.14, δ¹⁵N=− 2.8±0.75. Two-source mixing models were constructed using average carbon isotopic ratios for each site with IsoError (Phillips & Gregg 2001a, b). To estimate the basal δ¹⁵N for each site, source pools were plotted in carbon–nitrogen space and a linear regression fitted between the two basal organic matter sources adjacent to the site-average consumer signature. The equation of this regression was used to calculate trophic level based on the interpolated difference in δ¹⁵N between source and consumer; source pool δ¹³C values were then adjusted with trophic discrimination factors for each site model to align them with the site-average consumer signature. As Parapercis colias feeds at a relatively high trophic level (4 to 5) and Δ¹⁵N is averaged over four to five steps, uncertainty in this estimate was reduced (sensu Vander Zanden & Vadeboncoeur 2002). In the absence of a specific trophic discrimination factor for blue cod muscle tissue, average aquatic poikilotherm values of 2.3‰ for δ¹⁵N and 0.4‰ for δ¹³C were used (McCutchan et al. 2003). Two-source, single-isotope models are limited by the assumption that it is the combination of proximal rather than more distal sources which results in the intermediate carbon isotopic signature of the consumer (Fry & Sherr 1984). The validity of this assumption can be assessed given prior biological and ecological knowledge of the study system (Post 2002). For inner fjord habitats, E. radiata was not considered an ecologically plausible carbon source pool and was rejected in favour of more distal sources; elsewhere, the two adjacent source pools were used. Using a single value (e.g. mean) to describe source pool contributions to consumer isotopic signatures is not advised (Phillips & Gregg 2003), therefore the range of feasible solutions (95% confidence interval, CI) is reported for all models.

Age estimation

One of each pair of otoliths was embedded in K142 epoxy resin (NUPLEX Construction Products, Auckland, New Zealand). It is not expected that the left and right otoliths of a pair would yield different age estimates; however, for consistency and ease of sectioning, whenever possible the right otolith was selected for age estimation. Transverse sections (c. 0.7 mm thick) were cut through the primordium using a Buehler Isomet low speed diamond-tipped saw and the sections mounted on glass slides using a small amount of resin. Ages were estimated from photomicrographs of sectioned otoliths taken under transmitted light. Only opaque (winter growth) zones bordered on both sides by translucent (summer growth) zones were counted. Image editing software (ImageJ, NIH, USA) was used to enhance the images and allow more accurate identification of annual growth rings, in particular those nearer the core. Ages were recorded by two independent agers and the readability of the sections was scored from 1 (easy to determine all growth bands) to 5 (unreadable); any otoliths graded as unreadable by either reader were excluded from growth rate analyses. The coefficient of variance, CV (Campana 2001) was estimated for each otolith as:

5

where CV _j is the age precision estimate for the jth fish, X _ij is the ith age determination of the jth fish, X _j is the mean age estimate of the jth fish and R is the number of times each fish is aged. As the mean CV for all otoliths was<5%, ager bias was deemed to be within acceptable limits and the age estimates of the more experienced ager (SRW) were used for all further analyses.

Modelling of growth

Size-at-age data were used to construct von Bertalanffy growth curves (von Bertalanffy 1934) for each site, pooled across years to increase sample sizes and maximise model fit:

6

where L _t is length at time t, L _∞ is the asymptotic length, k is the growth constant expressing the rate at which length approaches the asymptote and t ₀ is the theoretical age of an individual at 0 size. Optimal values for L _∞ and k were obtained by minimising residual sums of squares with the Solver application (Frontline Systems Inc., USA) for Excel 2008 (Microsoft Corporation 2007). No fish were sampled from the youngest age classes (<3 yrs) so t ₀ was set to 0 for all models. As blue cod are protogynous hermaphrodites (Mutch 1983; Carbines 2004), growth was modelled separately for males and females.

Statistical analyses

Diet composition was compared between habitats (fixed factor, two levels) using permutational (multivariate) analysis of variance (PERMANOVA) in PERMANOVA + for PRIMER v6 (PRIMER-E Ltd, Plymouth, UK). This generates a pseudo-F statistic but as P values are calculated under permutation, the assumptions of normal distribution and homogeneity of variance inherent in traditional ANOVA methods are avoided (Anderson 2001). Multivariate diet composition was unaffected by fish sex or size (P<0.05) so spatial tests were pooled across sexes and size classes. Canonical analysis of principal coordinates (CAP) was used to calculate leave-one-out reclassification to habitat success rates (Anderson & Willis 2003). Permutational ANOVA based on Euclidean distance between samples was used to identify the dietary groups driving overall habitat effects.

Diet diversity (H′), fish condition and size-at-age were compared between habitats using one-way PERMANOVA based on Euclidean distance between samples. Size-at-age data were first normalised (mean=0, SD=1) to account for differences in variable range; spatial variability was investigated for males and females separately.

The ratio of males to females in the population at each study site was compared using Chi-square tests; Chi-square contingency tables were then used to compare the sex ratio at paired inner and outer fjord sites. Male and female size frequency distributions were compared between inner and outer fjord habitats using PERMANOVA based on Euclidean distance between samples.

Model selection

An information theoretic approach (Akaike 1974; Burnham & Anderson 1998; Anderson et al. 2001) with the corrected Akaike Information Criterion (AIC_c) was used to rank nested models and to quantify the explanatory power of dietary (proportion benthic versus pelagic), demographic (sex, condition) and site-specific factors (habitat, fjord) on individual size-at-age. Of the site-specific factors, ‘habitat’ refers to either inner or outer fjord locations, which tend to be environmentally and ecologically different, while ‘fjord’ incorporates any variability that might arise due to latitudinal or geomorphological differences among the three fjord basins included in this study. Formulae for AIC_c are provided in Anderson et al. (2001). A permutation-based least squares linear model framework (DistLM, PERMANOVA + for PRIMER v6) was used to test for interactions and collinearity among the explanatory variables.

Results

Stomach contents analysis

Diet composition differed between inner and outer fjord habitats overall (PERMANOVA, pseudo-F _1,232 =19.813, P=0.0001) with 72% of fish correctly reclassified to habitat; a reclassification success rate of 49–51% would be expected by chance alone. Pair-wise tests showed that this habitat effect was driven by pronounced differences in diet between inner and outer fjord populations in Dusky Sound (PERMANOVA, pseudo-F _1,69 =32.513, P=0.0001, 94% reclassification success) and Long Sound–Preservation Inlet (PERMANOVA, pseudo-F _1,97 =9.7097, P=0.0001, 74% reclassification success) and a marginal difference in Breaksea Sound (PERMANOVA, pseudo-F _1,64 =2.6247, P=0.0425, 62% reclassification success). In all fjords the occurrence of pelagic prey resources in stomach contents was significantly (PERMANOVA, P<0.05) higher among outer fjord fish, while the occurrence of benthic prey resources was frequently higher among inner fjord populations (Fig. 2).

Figure 2  Mean P. colias stomach contents occurrence at each site of A, Zooplankton, B, Pisces, C, Echinodermata, D, Crustacea, E, Mollusca, F, Benthic invertebrates, G, Algae and H, Detritus. Data have been pooled across sexes and years. Standard error bars are shown. A significant difference (P < 0.05) between paired inner and outer fjord sites is denoted by an asterisk. Dark bars indicate inner fjord sites and light bars indicate outer fjord sites.

Diet diversity (H′) was higher among inner fjord populations (PERMANOVA, pseudo-F _1,235 =9.0892, P=0.0023). This pattern was upheld in Dusky Sound and Long Sound–Preservation Inlet, but in Breaksea Sound diversity did not differ between habitats (Fig. 3a). Fish condition was lower among inner fjord subpopulations overall (PERMANOVA, pseudo-F _1,229 =6.6353, P=0.0067) and in Breaksea Sound and Dusky Sound, but in Long Sound–Preservation Inlet condition did not differ between habitats (Fig. 3b).

Figure 3  A, Mean Shannon–Wiener diversity (H′) at each site. Data have been pooled across sexes and years. Standard error bars are shown. Dark bars indicate inner fjord sites and light bars indicate outer fjord sites. A significant difference (P<0.05) between paired inner and outer fjord sites is denoted by an asterisk. B, Mean condition at each site. Data have been pooled across sexes and years. Standard error bars are shown. Dark bars indicate inner fjord sites and light bars indicate outer fjord sites. A significant difference (P < 0.05) between paired inner and outer fjord sites is denoted by an asterisk.

Figure 4  Sex ratio of Parapercis colias from each study site recorded as % male in the population. Dark bars indicate inner fjord sites and light bars indicate outer fjord sites; total sample size at each site is given. A significant deviation from a 50:50 sex ratio (P<0.05) is denoted by an asterisk; underlined fjord labels indicate a significant difference (P<0.05) in the sex ratio between inner and outer fjord subpopulations.

Stable isotope analysis and mass balance mixing models

Spatial variability in the δ¹³C signatures of P. colias was minimal (Table 1). Despite the presence of the chemoautotrophic clam Solemya parkinsoni in the stomach contents of a few individuals sampled from inner Breaksea Sound (n=3), inner Dusky Sound (n=7) and Preservation Inlet (n=1), no evidence was found to support long-term chemoautotrophic carbon use by any individuals (δ¹³C range:−20.7– − 15.0). Mass balance analyses demonstrate that in the inner fjords, on average 40–60% of organic matter is derived from SPOM with the remainder contributed by estuarine algal production (Table 1). Similar contributions to blue cod nutrition are made by SPOM in outer Dusky Sound and Preservation Inlet, but in outer Breaksea Sound SPOM contributions are minimal and the majority of organic matter is derived from marine macroalgal production (E. radiata). Marine macroalgae are also an important organic matter source for blue cod in outer Dusky Sound, but in Preservation Inlet freshwater/estuarine macroalgal production constitutes the major organic matter source.

Table 1  δ¹³C and δ¹⁵N of Parapercis colias muscle tissue from each study site and results of mass balance mixing models estimating carbon contributions from each basal organic matter source pool. Data shown are the 95% confidence interval (mean±standard error).

Site	δ^13C	δ^15N	SPOM (%)	Ulva (%)	Ecklonia (%)
Breaksea Sound (inner)	−17.6±0.45	13.3±0.36	41–80 (60±8)	20–59 (40±8)
Breaksea Sound (outer)	−17.3±0.11	13.1±0.09	0–15 (5±4)		85–100 (95±4)
Dusky Sound (inner)	−16.9±0.23	13.7±0.25	1–60 (50±44)	40–59 (50±4)
Dusky Sound (outer)	−18.2±0.09	13.0±0.05	25–43 (34±4)		57–75 (66±4)
Long Sound (inner)	−16.2±0.17	13.8±0.15	33–48 (40±4)	52–67 (60±4)
Preservation Inlet (outer)	−16.2±0.12	14.2±0.07	36–47 (41±3)	53–64 (59±3)

Population structure

Inner fjord subpopulations were female-dominated in Breaksea Sound (Chi-square test, χ²=23.310, P<0.0001, d.f.=1), Dusky Sound (Chi-square test, χ²=40.076, P<0.0001, d.f.=1) and Long Sound–Preservation Inlet (Chi-square test, χ²=9.6240, P=0.0030, d.f.=1). Outer fjord subpopulations were female-dominated in Dusky Sound (Chi-square test, χ²=16.004, P<0.0001, d.f.=1) and male-dominated in Long Sound–Preservation Inlet (Chi-square test, χ²=51.062, P<0.0001, d.f.=1) but in Breaksea Sound the outer fjord sex ratio did not differ from 50:50 (Chi-square test, χ²=3.2140, P=0.0776, d.f.=1). Significant differences were observed between inner and outer fjord sex ratios in Breaksea Sound (Chi-square test, χ²=4.5920, P=0.0321, d.f.=1) and Long Sound–Preservation Inlet (Chi-square test, χ²=40.030, P<0.0001, d.f.=1), but in Dusky Sound sex ratio did not differ between habitats (Chi-square test, χ²=1.6300, P=0.2018, d.f.=1).

Size distributions differed between habitats overall (PERMANOVA, pseudo-F _1,267 =4.4185, P=0.0387) with a higher proportion of individuals in the larger size classes in the inner fjord habitat for both sexes (Figs 4 and 5).

Figure 5  A, Size distributions of male Parapercis colias in the inner fjords (n=39). B, Size distributions of male Parapercis colias in the outer fjords (n=60). C, Size distributions of female Parapercis colias in the inner fjords (n=94). D, Size distributions of female Parapercis colias in the outer fjords (n=44).

Growth models

Both male and female blue cod achieved greater asymptotic lengths (L _∞ ) in inner fjord habitats while the curvature of modelled growth trajectories (k) was higher among outer fjord subpopulations (Table 2); in Breaksea Sound the male growth model was compromised by the small sample which contained very few younger individuals (<9 yrs), particularly from the inner fjord. Overall, size-at-age was lower among inner fjord than outer fjord subpopulations for both males (PERMANOVA, pseudo-F _1,116 =12.037, P=0.0003) and females (PERMANOVA, pseudo-F _1,149 =18.755, P=0.0001). This pattern was upheld in Long Sound–Preservation Inlet (PERMANOVA, males: pseudo-F _1,51 =22.017, P=0.0002; females: pseudo-F _1,44 =30.626, P=0.0001) and for males in Breaksea Sound (PERMANOVA, pseudo-F _1,33 =5.5131, P=0.0134). Size-at-age did not differ between habitats for females in Breaksea Sound (PERMANOVA, pseudo-F _1,46 =1.6919, P=0.1858) or for either sex in Dusky Sound (PERMANOVA, males: pseudo-F _1,28 =0.8953, P=0.3699; females: pseudo-F _1,55 =2.3147, P=0.1146).

Table 2  Von Bertalanffy growth model parameters (L _∞ , k), residual sum of squares (RSS) and degrees of freedom (d.f.) for male and female fish sampled in inner and outer fjord habitats (pooled across years). t ₀ was set at 0 for all models.

Site	Model	L ∞	k	d.f.	RSS
Breaksea Sound (inner)	Male	828	0.067	14	32338
	Female	507	0.145	36	40459
Breaksea Sound (outer)	Male	688	0.136	17	23387
	Female	486	0.195	19	66538
Dusky Sound (inner)	Male	524	0.154	28	42513
	Female	426	0.213	50	63309
Dusky Sound (outer)	Male	508	0.168	26	39057
	Female	406	0.244	43	24365
Long Sound (inner)	Male	547	0.163	16	40192
	Female	439	0.224	35	31736
Preservation Inlet (outer)	Male	510	0.165	43	69691
	Female	312	0.502	8	10721

Model selection

Model selection was used to assess the statistical explanatory weighting of dietary (benthic versus pelagic feeding strategy), demographic (sex, condition) and site-specific factors (fjord, habitat) in determining individual size-at-age. Significant correlations were found between size-at-age and dietary (r ² =0.03, P=0.0215), demographic (r ² =0.04, P=0.0014) and site-specific factors (r ² =0.18, P=0.0001). Diet explained a significant portion of the variability in size-at-age given that explained by demographics (r ² =0.02, P=0.0415). Site-specific factors explained a significant portion of the variability in size-at-age given that explained by demographics and diet (r ² =0.16, P=0.0001). Model selection revealed that site-specific factors (summed Akaike weight: 1.000) were collectively more important drivers of size-at-age than demographics (summed Akaike weight: 0.988) or diet (summed Akaike weight: 0.138), but the best model explained just 23% of the total variability in size-at-age (Table 3).

Table 3  Details of log-likelihood, number of parameters (K), AIC _c , delta values, Akaike weights and variance explained (R ² ) for models explaining variability in size-at-age.

Size-at-age =	Log (L)	K	AIC c	Δ i	W i	R ^2
Demography + site	2.62	2	130.60	0.00	0.853	0.22
Demography + diet + site	2.62	3	134.28	3.68	0.135	0.23
Site	2.64	1	139.70	9.10	9.012×10^−3	0.18
Diet + site	2.64	2	142.06	11.46	2.769×10^−3	0.19
Demography + diet	2.70	2	178.92	48.32	2.743×10^−11	0.07
Demography	2.71	1	180.35	49.75	1.342×10^−11	0.04
Diet	2.72	1	184.96	54.36	1.339×10^−12	0.03

Discussion

The results presented here provide direct evidence that spatial variability in growth trajectories of a coastal reef fish are strongly linked to differences in habitat and the nutritional landscape at a scale of 10–20 km. Spatial variability in composition and nutritional quality of diet as well as composition of basal organic matter sources supporting food webs was observed between multiple pairs of inner fjord and outer fjord/open coastal sites in Fiordland. Spatial patterns in these data largely corresponded with variability in condition and growth trajectories within subpopulations of blue cod observed among sites. In the inner fjords, the diet of blue cod typically comprised diverse, benthic resources, while in the outer fjords diets were more pelagic in nature and less diverse. Although the energy equivalent of sampled stomach contents was not measured, a review of published values for similar dietary components suggests that pelagic resources (i.e. zooplankton) are likely to be of a higher nutritional quality than benthic food sources such as detritus and invertebrates (Steimle & Terranova 1985; Dudgeon et al. 1990; Zoufal & Taborsky 1991). This spatial separation in diet composition was reflected by higher condition among individuals from outer fjord subpopulations in two of the fjords. Similar correspondence between blue cod trophic ecology and fish condition has been observed in Bradshaw–Thompson Sound (Beer 2011). In Long Sound–Preservation Inlet, no difference was observed between samples from inner and outer fjord habitats; here, diets in both habitats were more diverse than in the inner waters in Breaksea and Dusky Sounds, and fish condition was lower than elsewhere. This may be linked to the generally higher incidence of benthic resources (in particular gastropod molluscs) in stomach contents of fish from Preservation Inlet than those from the other outer fjord sites, as this benthic diet is likely to be nutritionally poorer than the more pelagic diet typical of other outer fjord sites. No evidence was found to suggest that growth of P. colias in the inner fjords is limited as a result of their diet.

Although isotopic mixing models are limited by the relevance of the input source values, they can provide some general insight into broadscale patterns of basal resource utilisation at higher trophic levels (Fry et al. 1999; Vander Zanden & Vadeboncoeur 2002). Analysis of the basal organic matter sources supporting blue cod prey suggest that Preservation Inlet also lacks significant inputs from common kelp, E. radiata, indicating a basic difference in the food web in this region. The unusually shallow entrance sill that separates the deeper waters of Preservation Inlet from the open Tasman Sea restricts wave action within the inlet and low-salinity surface water is retained within the shallow entrance sill (Stanton & Pickard 1981; Wing et al. 2003). As a consequence, the habitats near the entrance of Preservation Inlet are relatively wave-sheltered and have low density of kelps, particularly E. radiata (Wing et al. 2007). In contrast, semi wave-exposed habitats in Breaksea Sound and Dusky Sound host food webs dominated by E. radiata production typical of exposed coastal kelp forest habitats. Analysis of inputs of pelagic organic matter sources into food webs supporting blue cod indicates the importance of benthic–pelagic coupling, particularly in the inner fjord basins where phytoplankton production is relatively high (Goebel et al. 2005). These basic differences observed in the organic matter sources supporting local populations of blue cod propagate through the food web resulting in large differences in the nutritional quality of the prey base and correspond with observed spatial variability in growth and population structure among habitats.

Both males and females in outer fjord subpopulations approached asymptotic size (L _∞ ) faster than those in inner fjord subpopulations. Nevertheless, in the inner fjord habitats we observed a larger proportion of large, old individuals in the populations, and the asymptotic size was slightly larger in the inner fjord habitats. The proportion of individuals in the larger size classes is a product of the size frequency distribution, whereas the size-at-age is a result of the growth trajectory; the pattern observed indicates that inner fjord subpopulations had relatively fewer small recruits in the size distribution, but the asymptotic size was smaller. Therefore, although in the inner fjords individual growth is slow (and average size-at-age is low), there is a higher proportion of old growth in the population since individuals survive to populate the older age/larger size classes, albeit slowly. This may reflect differences in adult mortality between the two environments or differences in the allocation of resources to growth and reproduction among the female-dominated inner fjord subpopulations. Commercial and recreational fishing pressure, whilst relatively low throughout the study area compared with elsewhere around the South Island of New Zealand, has historically been concentrated along the open coast and in the entrances to Dusky Sound and Preservation Inlet (Davey & Hartill 2008; Starr & Kendrick 2008). Increased fishing pressure can result in a population structure that is biased towards faster growing, early maturing individuals, and in blue cod the constant removal of large males is thought to increase the incidence of sex change among females, resulting in a male-dominated population (Beentjes & Carbines 2005).

These observed patterns in population structure have important implications for local population productivity within each habitat and for yield-per-recruit estimation within the fishery management area. The adoption of spatial management of the blue cod fishery within Fiordland in 2005 highlighted the need to incorporate information on vital rates of subpopulations into management planning. In this case commercial fishing has been excluded from the inner fjord habitats, recreational take has been reduced and a series of 10 no-take marine reserves provides refugia for spawning stock in the inner fjord environments.

The present study demonstrates that baseline variability in the nutritional landscape can result in spatially structured populations with substantial variability in growth evident at the mesoscale (10–20 km) in the absence of significant anthropogenic disturbance. Management models for invertebrate fisheries such as abalone (Shepherd et al. 2001) and sea urchins (Quinn et al. 1993) commonly incorporate spatial variability in vital rates. Our findings highlight the importance of resolving spatial variability in vertebrate growth rates and of spatially-explicit management measures in the sustainable exploitation of temperate and tropical reef fishes, which typically form highly resident local populations. Failure to account for spatial variability in growth and other vital rates could result in overestimation of the productivity of an exploited population and runs the serious risk of localised depletion.

Ethical standards

This research complies with the current laws of New Zealand and the requirements of the University of Otago's Animal Ethics Committee.

Acknowledgements

We thank S. Lusseau, S. Rutger, K. Rodgers, B. Dickson, P. Hesseltine and P. Meredith for their assistance with sample collection. Monetary support was provided by the Royal Society of New Zealand's Hutton Fund (NAB), the Commonwealth Scholarship and Fellowship Plan (NAB), a University of Otago Postgraduate Publishing Bursary (NAB), and by a Royal Society of New Zealand Marsden Grant to SRW (UoO–00213).