Estimating the scale of fish feeding movements in rivers using δ13C signature gradients


*Correspondence author. E-mail:


  • 1Isotopic signatures of consumers provide a time integration of their feeding history, and as a result of movements, are often out of line with signatures of their local resources. Such disequilibrium can be useful for inferring the spatial scale of consumer movement.
  • 2δ13C signatures of dissolved inorganic carbon as well as periphyton and invertebrates, exhibit pronounced gradients along rivers. We outline an analytical framework to estimate the spatial scale of movement of riverine fish by comparing the slopes of their δ13C signature gradients to that of the stream invertebrates they consume. For free-ranging juvenile Atlantic salmon (Salmo salar L.), unconstrained by barriers, δ13C signatures departed considerably from invertebrate signatures, and along-stream slopes were as little as half those recorded for stream invertebrates. Movement estimates for these fish, based on their signature slopes, are ~20 km.
  • 3By contrast, stream resident salmonids (whose movements are constrained by physical barriers) and sedentary taxa such as sculpins and sticklebacks, have carbon signatures much closer to invertebrate signatures where they were collected. For these groups, our method yields negligible estimates of movement, similar to those of invertebrates.
  • 4Although this method cannot provide precise estimates of how much individual organisms move, or reveal details of movement history, it may provide an effective complement to telemetric and other methods of studying movement.


While movement is one of the most fundamental attributes of animals, our weak understanding of the spatial scale over which animals move has limited our ability to define a population unit (stock identification) and to manage and conserve populations effectively (Gowan et al. 1994). These problems are especially perplexing in fisheries management, particularly for salmonids. Moreover, an increasing number of studies are challenging the commonly held view (Gerking 1959) that stream fish move very little (Gowan et al. 1994; Gowan & Fausch 1996; Kennedy et al. 2002).

Understanding movement is also important at the ecosystem level where it contributes to flows in nutrients and energy on the landscape (Polis & Strong 1996; Finlay, Khandwala & Power et al. 2002), and determines the spatial scale of predator–prey interactions, which is key to our understanding of stability and function (McCann, Rasmussen & Umbanhowar 2005). Information on the scale of movements of fish is also vital to ecotoxicological studies on effects of effluents, and environmental programmes require information on the scale of movements of fish to assess impacts (e.g. Galloway et al. 2003; Gray & Munkittrick 2005).

While a great deal of spatial information has been obtained from studies that have marked and tagged individual organisms, and more recently tracked them with telemetry and various types of weirs (Cunjak et al. 2005; Gowan & Fausch 1996), developments in chemical tracer technology (e.g. stable and radioisotopes) makes it possible to infer the spatial scale of many ecological processes through indirect means (Hobson 1999; Tucker et al. 2000; Finlay et al. 2002; Kennedy et al. 2002; Cunjak et al. 2005). Chemical tracer approaches can potentially be applied to landscape studies in any situation where a spatial gradient exists; but, while this type of approach is applied routinely in geochemistry, it has seen limited application to ecological problems.

River geochemists have shown that the inorganic δ13C signature in rivers undergoes a very predictable shift from headwater signatures in equilibrium with soil carbon dioxide signatures and weathering processes to heavier and more atmospherically equilibrated signatures downstream (Telmer & Veizer 1999; Finlay 2003). This gradual transition is attenuated as a result of the slow release of isotopically depleted carbon dioxide into the river from decomposition of refractory organic carbon, combined with the gradual downstream increase in alkalinity and pH (Yang, Telmer & Veizer 1996). Thus, inorganic δ13C signatures in small tributary streams are ~−16‰, and become gradually more enriched to ~−7‰ or heavier in lower reaches.

As a result of this geochemical gradient, river biota acquires a distinct gradient in the δ13C. Additional factors such as stream velocity and periphyton biomass influence signature fractionation (Trudeau & Rasmussen 2003; Finlay 2004; Rasmussen & Trudeau 2007) through their effects on boundary layer carbon dioxide availability and diffusion, and add ‘noise’ to along-stream signature gradients of periphyton and invertebrates. Along most mainstem river courses, invertebrate signatures predominantly reflect authochthonous carbon sources (Finlay, Power & Cabana 1999; Finlay 2001; McCutchan & Lewis 2002) and all but the specialist ‘leaf shredders’ tend to acquire the along-stream carbon signature gradient described above. The signatures of relatively sedentary fish (e.g. sculpins, Gray et al. 2004; and small cyprinids, Rosenfeld & Roff 1992; Doucett et al. 1996), usually reflect local carbon sources; however, those of more mobile fishes such as juvenile salmonids should reflect their feeding movements and likely remain out of signature equilibrium with local food sources. Thus, signature interpretation and estimation of diet-tissue isotopic fractionation (Vander Zanden & Rasmussen 2001) can be complicated for mobile fish, since their movements can result in substantial signature deviations from local sources. Such signature deviations can, however, be useful in inferring the spatial scale over which consumers move and feed, and thereby, integrate resource signatures. In this paper, we outline a framework for analyzing the signatures of mobile consumers on a linear resource signature gradient. We then demonstrate the applicability of this framework using data from published and unpublished sources, and use it to compare the movement estimates obtained for free-ranging juvenile fishes, with those of stream resident salmonids as well as more sedentary fish taxa.


A consumer in isotopic equilibrium with a resource whose signature has an along-stream gradient will tend to acquire the same signature gradient, with its signatures displaced by an amount f (either + or −). This trophic shift can be due to biased consumption, assimilation or tissue fractionation (Fig. 1a). Alternatively, a consumer that is mobile, i.e. does not feed in one place long enough to achieve isotopic equilibrium with local resources, can have signatures that deviate from local resources and such deviations should tend to be most pronounced near the boundaries of the resource gradient since movement history is likely to be biased away from the boundary in the direction of the gradient average, YAVG (Fig. 1b). We can approximate such a shift in the consumer signature gradient as a straight line (small dashed line Fig. 1b) whose slope is reduced relative to that of the resource gradient; the more mobile the consumer the greater the shift. If consumer feeding movements are so pronounced that they span the entire resource gradient, all of these fish may acquire signatures close to the gradient average (YAVG) leaving no detectable gradient in the consumer signature. For simplicity, in Fig. 1b, f = 0. Shifts in gradient slope resulting from feeding movements can also complicate the estimation of f, especially if the signature gradient is not well characterized, and we will return to this point below.

Figure 1.

(a) Signatures of an equilibrated consumer and its food resources along a stream gradient, f represents trophic shift due to biased consumption, assimilation or tissue fractionation of the resource signature. (b) Signatures of a mobile (unequilibrated) consumer and an immobile (equilibrated) consumer (large dashed line), adjusted for f, compared to resource signature gradient. The mobile consumer experiences a shift in signatures resulting from boundary constraints on its movement, and this can be approximated as a shift in the along-stream signature slope (small dash line), whereas the immobile consumer signature closely tracks the resource signature (large dash line). (c) Deviation of consumer signatures from resource signatures (α) towards the gradient average (YAVG) due to spatial averaging resulting from random movement (β) along a uniform gradient. (d) Consumer vs. resources signature plot illustrating signature deviation (α) arising from movement history resulting in slope deviation from 1.

In this paper, we demonstrate that signature shifts along a gradient, of the type shown in Fig. 1b occur, and that they likely reflect movement history. We propose to use the signature deviation as a quantitative estimator of the spatial scale of that movement history. It is important to specify that our proposed estimator will detect the spatial scale over which feeding occurs, and should be insensitive to movements where feeding does not occur, or occurs to a reduced extent (e.g. spawning migrations or movements to overwintering habitat). Moreover, the estimate would be weighted by the extent to which different feeding areas contributed to feeding and ultimately growth.

We make the simplifying approximation that the deviation of the consumer signature from local sources (α) results from averaging across a uniform distribution range corresponding to 2α, produced by spatial averaging across a feeding range β (Fig. 1c). Let the resource signature (YR), change with river distance (X) according to

YR = Y1 + SX(eqn 1)

where S is the slope along the river gradient and Y1 is the intercept at X = 0. Also, let the consumer signature (YC) be weighted in the direction of the gradient average (YAVG) in proportion to p (= β/ΔX) the proportion of the gradient over which an individual feeds and averages over, and weighted in the direction of the resource signature (YR) in proportion to q (= 1 − p). Thus, p reflects the degree to which consumer signature slopes are shifted relative to resources; however, it must be noted that this is a relative shift which must be scaled to the spatial scale of the gradient (ΔX) to provide a movement estimator (β). The equation for the consumer signature obtained by this weighting is

Yc = YAVGp + Y1q + SqX + f(eqn 2)

and has the properties illustrated in Fig. 1c, in that its slope is reduced relative to S, and when f = 0, its intercept is shifted towards YAVG in proportion to p. Moreover, its boundary conditions are also appropriate; that is, if p = 0, and q = 1 (local feeding), the consumer signature will equal the resource signature +f, and if p = 1, and q = 0 (feeding over the whole gradient) then YC = YAVG + f. The consumer signature deviation at any point in the gradient (α) from the resource signature

α = [YAVGp + YRq] − YR + f


image(eqn 3)

Thus, the slope of the consumer signature gradient (equation 2) divided by the slope of the resource signature gradient (equation 1), will be an estimator of q, and β can be obtained by multiplying the slope reduction factor (p) by the spatial scale of the gradient. Although f, the trophic shift, will affect the absolute value of the consumer signature, it will not affect the slope and therefore have no effect on the estimate of β (Fig. 2). f can be estimated by comparing the intercepts of the consumer and the resource equations (equations 1 and 2) once p and q have been determined from the slope ratio.

Figure 2.

Consumer vs. resource signature plot illustrating the effect of f on signature deviations (α) produced by consumer movement history. f will not affect the slope of the consumer-resource signature plot since it will reduce α2 by the same amount that it inflates α1.

An alternative approach to estimating β that may be statistically more robust is to plot consumer signatures directly against resource signatures, i.e. C(R), as shown in Fig. 1d and thus remove the common variable X. In this case, movement can be assessed by comparing the slope of the consumer-resource signature plot to the value of 1, which would be expected if each consumer were in signature equilibrium with resource signatures in the immediate locality in which it was captured. Because we can write equation (2) as

Yc = YAVGp + q[Y1 + SX] + f

YC can be written as a function of YR

Yc = YAVGp + qYR + f(eqn 4)

The slope of the consumer-resource signature therefore provides a direct estimate of q = 1 − p, and β can be obtained without having to compare consumer and resource gradient slopes.

image(eqn 5)

Thus, β can be estimated in two different ways, the first, outlined in equation (3) is based on the degree of reduction of the slope of the consumer's signature with respect to river distance (X), and the second, outlined in equation 5 is based on the deviation of the consumer resource signature plot from 1. Standard error estimates can easily be derived for the β estimates (SEβ) based on equation (5) since they will equal the SE of the C(R) slope multiplied by ΔX; however, for β estimates based on equation (3) (SEβ), estimates would need to be based on ratio algorithms (see Cochrane 1977).

The estimation of f for a mobile consumer is not trivial since it can be confounded by movement (Fig. 2); it can be calculated from the intercept of the consumer-resource signature plot, which equals YAVGp + f, once p has been determined from the slope. Similarly, f can be estimated from the intercept of the river gradient model for the consumer signature since the intercept of this plot =YAVGp + Y1q + f.

In this model, we have assumed that the consumer feeds entirely on aquatic food sources, and when this is not the case, we would expect along-stream slopes to be influenced by terrestrial signatures. Since the signatures of terrestrial organic matter tends to be relatively constant and in the mid-range of the along-stream autochthonous signature gradient (~−28‰Finlay 2001), consumption of terrestrial material should tend to reduce the along-stream slope of the consumer's signature, since signatures both upstream and downstream will tend to be deflected in the direction of the terrestrial signature (France 1995). Therefore, terrestrial consumption would be expected to confound estimates of mobility based on along-stream signature slopes. The model introduced above can be extended to include the combined influence of consumer mobility and terrestrial consumption, to develop algorithms for estimating the scale of movement adjusted for the proportion of the diet made up by terrestrial material.

In this paper, the following questions are addressed:

  • 1Do the dissolved inorganic carbon (DIC) δ13C signature gradients that have been shown by geochemists to occur in a wide range of rivers, lead to simple gradients in river biota?
  • 2Are the signature gradients for free ranging mobile consumers such as juvenile Atlantic salmon (Salmo salar L.) that eats mainly aquatic invertebrates (Thonney & Gibson 1989; Mookerji, Weng & Mazumder 2004), significantly less steep than those of primary producers and invertebrate primary consumers?
  • 3Do fish that are resident in small streams that are constrained spatially by movement barriers (e.g. waterfalls, culverts, beaver dams), or sedentary species, have signatures that more closely resemble those of their resources than free-ranging fish, that are free to move throughout the river system?

It is important to clarify at the outset that the data assembled in this paper to test the hypothesis were not collected for this purpose, and that the model was elaborated post hoc. As a result, the data serve more to demonstrate that the approach is promising and that rigorous tests are possible. To test the idea rigorously will require detailed descriptions of the signature gradients for a number of individual systems, and the purpose of this paper is to motivate studies on a wide range of systems so as to fully explore the spatial applications of stable isotopes to studies on movement.


We compiled a set of data on consumer and resource δ13C signatures (CSIG) from our own studies on the Sainte-Marguerite River, Quebec (SMR), and from other published studies that included signatures of both fish and lower trophic levels (invertebrates and or periphyton), or fish from a series of locations along the same river system. Data were included from a wide range of river systems (SYS), including the SMR and its tributaries, the Miramichi River, New Brunswick (MIR) and its tributaries, the St. John River, New Brunswick (SJR), the Credit River system, Ontario (CRE), the Yukon River (YUK) and its tributaries, and a number of west coast rivers. In order to standardize the analysis as much as possible, studies on river systems containing major lakes or reservoirs along the river course were excluded.

Our own published and unpublished results on the SMR includes signatures of periphyton (rock scrapings mostly diatoms) (Rasmussen & Trudeau 2007), as well as unpublished data on grazing mayflies (mostly Heptageniidae, Baetidae and Ephemerellidae), filter feeders (mostly Hydropsychidae) and collector-gatherers (mostly chironomids), as well as juvenile Atlantic salmon (Trudeau 2004). All of these taxa were collected along the river gradient from headwater streams down to the river mouth (Morinville & Rasmussen 2006b). The rivers from which literature data were obtained were free flowing, like the SMR, are located at similar latitude, and have similar biota. The data from the SMR represent averages for whole kilometre reaches of the river, as well as 100-m reaches of two tributaries Epinette creek and Morin creek as well as the mouth of the river where it enters its estuary. The locations of these sampling sites are described in detail by Rasmussen & Trudeau (2007) and Morinville & Rasmussen (2003, 2006a). Sites sampled as well as details of sampling techniques used on other rivers from which data were obtained, were described in the published sources cited.

All periphyton, invertebrate, fish, and detritus samples were oven dried, pulverized, homogenized, weighed, and packed into tin capsules. Stable C isotopic analyses were carried out by mass spectrometry (Finnigan-Mat Delta-Plus continuous flow isotope-ratio mass spectrometer coupled to a Carlo-Erba elemental analyser on line; G.G. Hatch Isotope Laboratories, University of Ottawa, Ottawa, Ontario, Canada and GEOTOP Laboratory, UQAM University, Montreal, Quebec, Canada). The analytical precision of these apparatus is typically 1 SD (0·05–0·2‰). Neither samples from the SMR nor those from other literature studies were lipid or acid extracted, and therefore no signature corrections for effects of such procedures (McCutchan et al. 2003) were necessary. Further analytical details were outlined by Rasmussen & Trudeau (2007) and Morinville & Rasmussen (2006b) for SMR samples and in the cited references for samples from other rivers.

Although most δ13C signatures published for fish come from white muscle tissue samples, in order to minimize the number of Atlantic salmon killed, our fish signatures from the SMR (20–30 for each river section) were obtained from small fin clips taken from the lower edge of the caudal fin. A sample of fish from a range of sites was sacrificed to compare fin clip signatures to white muscle (just below the dorsal fin) signatures. As previously shown by Jardine et al. (2005a,b), the signatures of these tissues were highly correlated (r2 = 0·91, P < 0·0005) and not significantly different from each other (paired t-test; P = 0·80), and when regressed on each other showed no significant bias (slope not significantly different from 1 and intercept near 0). Invertebrate tissue signatures from the SMR as well as other rivers were based on ground-up whole organisms.

Stomach contents of the salmon parr were analyzed and shown to contain mainly (>90%) larvae, pupae and adults of aquatic invertebrates, mainly mayflies and caddisflies with some blackflies and chironomids. Mookerji et al. (2004) similarly found that SMR Atlantic salmon juveniles consumed mainly aquatic invertebrates.

Periphyton (epilithic algal samples), stream invertebrates and fish were obtained at numerous locations within each river reach sampled, to average across the stream cross section and along pool-run-riffle sequences to remove the significant effects of water velocity on signatures (Trudeau & Rasmussen 2003; Rasmussen & Trudeau 2007).

categories of fish and invertebrates

Fish data were grouped into categories (CAT) as follows. The salmon category (SAL) consists of juvenile anadromous Atlantic salmon obtained from free-flowing river systems including the MIR, and the SMR. Salmonids confined to headwater streams by barriers such as waterfalls or weirs were considered resident populations (RES). RES also includes an average signature from a set of juvenile Atlantic salmon from Otter Brook NB that were fitted with transmitters (Cunjak et al. 2005) and shown to be stream residents, and grayling from Alaskan streams (Peterson et al. 1993) believed to be from resident populations. Fishes such as sculpins (Cottus spp.) and sticklebacks (Gasterosteus aculeatus L.), which have defensive adaptations that restrict mobility, were considered sedentary species (SED). A sample of ammocetes larvae (Bilby, Fransen & Bisson 1996) (Petromyzon sp.) was also included in this category. This category was composed mainly of sculpins which as a group are known to feed almost exclusively on aquatic invertebrates (Dineen 1951; Holmen, Olsen & Vollestad 2003; Zimmerman & Vondracek 2007). Signature data identified as pertaining to young-of-the-year fish were excluded, since it was expected that they might retain traces of maternal signature accumulated from egg yolk, which would contribute to signature disequilibrium (Vander Zanden et al. 1998).

Invertebrate predators (IPR) include Plecoptera (Perlidae and Perlodidae) stoneflies, Megaloptera, Coleoptera, Odonata and Hemiptera. Herbivores (HER) include Ephemeroptera (mostly Baetidae, Heptageniidae and Ephemerellidae), some Trichoptera (Brachycentridae, Glossosomatidae, Blephariceridae, Helicopsychidae), and small amphipods (Gammarus). Filter feeders (FIL) include Hydropsychidae, Polycentropodidae, Simuliidae. Average invertebrate signature (AVI) represents the mean of all invertebrate signatures available in the study. The periphyton (PER) included samples designated as periphyton, epilithic algae or epilithon in various studies.

estimating river distance

The goal in modelling signature gradients was to relate δ13C signatures of biota to an estimator of river distance that would reflect the average distance from the headwaters to a point in the stream. Although the simplest measure of distance along the river gradient is mainstream length, the distance from the headwater to a given point (X) in the stream along the mainstream, this represents the maximum distance over which a fish has integrated its signature (β), rather than an estimate of the average distance, since there are many shorter paths from headwaters to the lower reaches. Since we have no independent information (e.g. tagging or radio tracking) to infer which stream path(s) a given fish has been occupying, the assessment of linear river distance should consider all possible paths. For example, a fish collected at a downstream location that has a more depleted carbon signature than the local food sources, reflecting headwater feeding, may have obtained that signature either by feeding in the upper reaches of the mainstem, or through a number of shorter paths involving nearby tributaries.

Stream geomorphologists (Smart & Surkan 1967; Gregory & Walling 1973) have explored the general relationships between stream metrics and drainage area (DA), and shown that the length of stream paths leading to a point in the drainage (X) could be modelled as a power function of the drainage area (DAn). This is reasonable, since √DA provides a general metric reflecting the linear dimensions of a watershed. They found that for mainstream length n was usually in the neighbourhood of 0·6, whereas for average stream length (the average of all stream flow paths leading from the headwaters to the point X, n was generally closer to 0·5 (Smart 1968, 1972). DA is also much more commonly reported in stream ecology studies than any direct measures of river distance such as mainstream length. We therefore chose √DA as the best distance measure along the stream course that was least likely to be seriously biased.

statistical analysis

General linear models (containing interaction terms) were first tested, and then followed by ancova models where no significant interaction terms were identified. The first general model tested CSIG = a1 + b1 RDIS + c1CAT + d1(RDIS × CAT), and this was followed by an ancova model which was used to compare the intercepts for the biotic categories (CAT) whose slopes were not significantly different (RDIS × CAT not significant). The second general model tested, CSIG = a2 + b2 RDIS + c2SYS + d2(RDIS × SYS), was applied to a subset of the data that included all of the biotic categories whose slopes (b1) were not different and for river systems where data were available for four sites or more. After this, an ancova model was used to compare the intercepts for the different river systems whose slopes were not different (SYS × CAT not significant). Comparisons of slopes of biota signatures in relation to river distance were carried out using ordinary linear regression analyses (JMP®). Slopes of consumer vs. resource-signature relationships were analysed using both ordinary linear regression as well as reduced major axis regression. The departure of consumer vs. resource signature slopes from 1 is likely to be exaggerated as a result of the fact that both the consumer and the resource signatures were estimated with equal error, and to compensate for this effect, slopes were also calculated using reduced major axis regression (RMA) (Sokal & Rohlf 1981). The complete data set on which these analyses were carried out is presented in Table S1.


As expected, C signatures (CSIG) became more enriched with increasing river distance (RDIS) (Table 1, Fig. 3) and no improvement in fit (decrease in RMS or increased R2) was obtained by log or power transforming RDIS. In the first general model, the interaction term (CAT × RDIS) was significant only for the Atlantic salmon (SAL), whose slope was significantly lower than the remaining biotic categories. With SAL excluded, the pooled slope of the CSIG vs. RDIS relationship is 0·29‰ per kilometre, and most categories of biota had similar intercepts although the analysis identified two categories, with HER and SED being the most depleted and PER and FIL the most enriched (Table 1). The second ancova (Table 2), which used SYS as the categorical variable, showed that river systems differed significantly with regard to their intercepts (MIR and STM more enriched than the YUK and SJR), but not with regard to their slopes (SYS × RDIS NS).

Table 1. ancova model of δ13C signature gradients with river distance ((√DA)km) for different categories (CAT) of fish, invertebrates and periphyton
Dependent variable
C signature (‰)
TermEstimateSEt P > | t |   LSM
  1. CAT represents categories of biota; PER = periphyton, HER = herbivores, FIL = filterers, IPR = invertebrate predators, AVI = average invertebrate signature, and SED = sedentary fishes. SED constitutes the reference category for this analysis and coefficients for other categories are relative to SED. Comparisons of least squares means (LSM) are based on Tukey post-hoc comparisons. Biota with the same letter designation (A, B) are not significantly different.

Intercept−32·80·33−97·3< 0·0001PERA −27·5
(√DA)km0·290·0117·7< 0·0001FILA −27·6
CAT[HER]−0·590·41−1·420·159HER B−28·9
CAT[IPR]0·230·620·370·711SED B−30·0
Figure 3.

Gradients in δ13C signatures of periphyton (PER, inline image), herbivores (HER inline image), average invertebrates (AVI inline image) and sedentary fish (SED inline image) as a function of river distance (RDIS). PER δ13C =−32·0 + 0·305 RDIS (solid line) r2 = 0·73; HER δ13C = −33·8 + 0·317 RDIS (small dash) r2 = 0·65; AVI δ13C = −32·3 + 0·286 RDIS (dotted line) r2 = 0·68; SED δ13C = −34·5 + 0·330 RDIS (large dash) r2 = 0·93.

Table 2. ancova model of biotic δ13C signature gradients with river distance ((√DA)km) for different river systems (SYS). SYS is a categorical variable representing the river system from which signatures were obtained
Dependent variable
C signature(‰)
TermEstimateSEtProb > | t |    LSM
  1. MIR = Miramichi, SJR = St. John, SNO = Snoqualmine, STM = Sainte-Marguerite, YUK = Yukon and CRE = Credit. YUK is the reference system for this analysis and coefficients for other categories are relative to YUK. Comparisons of least squares means (LSM) are based on Tukey post-hoc comparisons. River systems with the same letter designation (A,B,C) are not significantly different.

Intercept−32·40·29−101< 0·0001STMA  −26·7
(√DA)km  0·230·01  13·0< 0·0001SNOAB −27·7
SYS[CRE] −0·850·49  −1·74  0·083MIRAB −27·9
SYS[MIR]  1·070·39   2·76  0·007CRE BC−29·8
SYS[SJR] −1·590·58  −2·75  0·007SJR  C−30·5
SYS[SNO]  1·180·46   2·54  0·012YUK  C−31·1
SYS[STM]  2·320·35   6·52  0·0001     
R2  0·78        
SEest  2·10        

signature gradients of juvenile atlantic salmon

Although sedentary fish (mainly sculpins) had very similar signature gradients along the river to stream invertebrates (Fig. 4; Table 1), as expected, gradients obtained for juvenile salmon were much less steep. The slope for SAL signatures vs. river distance was significantly less steep than that of aquatic invertebrates (Δ slope = 0·16 ± 0·025, t = 6·4, P < 0·0001).

Figure 4.

Gradient in fish δ13C signatures vs. river distance (RDIS). SAL (inline image) represent free-ranging juvenile salmon, [SAL δ13C = −26·9 + 0·134 RDIS (solid line) r= 0·62, P < 0·001] and for sedentary fish (inline image) [SED. δ13C = −35·2 + 0·34 RDIS (small dash)]r2 = 0·92, P < 0·001. Dash-dot line represents the gradient for average invertebrate signatures (AVI δ13C = −32·3 + 0. RDIS r2 = 0·68, P < 0·001).

The carbon signatures of juvenile salmon were strongly correlated to the signatures of invertebrates collected at the same sites (r2 = 0·82); however, the signature deviation is very striking at both ends of the signature gradient (Fig. 5), and as predicted, the slope of the consumer vs. resource signature plot was significantly less than 1. As expected, the departure of the ordinary least squares C/R slopes from the 1:1 line was greater than that obtained using reduced major axis (RMA) regression (Sokal & Rohlf 1981). Based on the RMA slopes, Δ slope = 0·46 (± 0·05), t = 9·2, P < 0·0001. At the upstream end of the gradient signatures of fish are shifted by 5–6‰ and at the downstream end they are shifted by ~3‰.

Figure 5.

δ13C SAL (inline image) plotted against average invertebrate signatures for the same sites from which fish were collected. The regression lines obtained (solid line) was (a) SAL δ13C = −11·9 + 0·47 AVI δ13C. r2 = 0·82, P < 0·0001. The lines are compared to the 1:1 line (large dash and the estimated food consumption line (small dash) which is the solid line adjusted for f (calculated from equation 4). Reduced major axis regression (Sokal & Rohlf 1981) was calculated to compensate for biased slopes resulting from both axes being measured with equal error SAL δ13C = −10·4 + 0·54 AVI δ13C.

estimating the scale of movement (table 3)

Table 3.  Estimates of the spatial scale of movement of consumers (β) and signature fractionation (f) factors from carbon signature gradients for different categories of fishes
Consumer/resourcep(β) kmf ‰pSE p(β) kmf ‰
Eqn 3Eqn 3Eqn 2Eqn 4 Eqn 5Eqn 4
  1. SAL = Atlantic salmon; RES = stream resident fish, and SED = sedentary fishes. Slopes used in equation (4), were those obtained from reduced major axis regression to correct for errors in the predictor variable (Sokal & Rohlf 1981). aβ, f and p could not be calculated for SED/AVI using equations (4) and (5), because invertebrate signature data were in most cases not available for the exact sites for which SED data were obtained. This is not necessary for calculations using equations (2) and (3).

SAL/AVI0·4421 1·20·46***0·05221·1
RES/AVI0 0 0·50, n.s.0·10 00·7
SED/AVI0 0−0·8a

Equation (4) relates consumer signatures to resource signatures [C(R)]) and allows the movement scale to be estimated from the slope (q = 1−p), p being an estimate of the proportion of the gradient over which the fish moves and integrates the resource signature. Since the RMA slope of the SAL vs. AVI signature plot (Fig. 5) was 0·54, equation (4) yields an estimate of P = 0·46 ± 0·05 for juvenile Atlantic salmon (SAL). To obtain β, an estimate of the scale of movement, P is multiplied by 48 km (the distance over which the signature gradients in the data set spanned) yielding an estimate of β = 22 km for SAL. Equation (3) estimates p by comparing the along-stream slopes of the consumer and resource signatures. Although this is a somewhat less direct way of estimating the scale of movement, the estimates obtained were very similar to those obtained from equation (4) (Table 3). For sedentary and resident fish, estimates of p and β were not significantly different from 0 (Table 3); f estimates calculated by fitting equation (2) and (4), ranged from −0·8 to +1·2‰ (Table 3). These values are within the range of carbon isotope fractionation estimates typically quoted for fish in the literature (Vander Zanden & Rasmussen 2001; McCutchan et al. 2003).

When stream resident salmonids (mostly trout) were plotted against invertebrate signatures, all of the points fall below the line obtained for juvenile Atlantic salmon (Fig. 6), and most lie between that line and the 1:1 line. Since departures at both ends of the gradient are trivial, f can be estimated by simply comparing the mean signatures of fish and invertebrates, and is estimated to be 0·3 since predators average 0·3‰ heavier than the primary consumers.

Figure 6.

δ13C of stream resident fish plotted against average invertebrate signatures. Points are compared to the regression for free-ranging trout (dot-dash line), and the 1:1 line (dashed line). solid circle, stream resident juvenile Atlantic salmon confirmed by radio telemetry (Cunjak et al. 2005); dashed circles, stream resident trout above waterfalls and other physical barriers impassible to migratory salmon (Kline et al. 1990; Morinville & Rasmussen 2006b); no circles, fish above likely barriers (e.g. beaver dams).

consumer resource signature relationships for invertebrates

Invertebrate predator signatures are virtually identical to that of their prey indicating that their mobility is negligible, and this concurs with studies of invertebrate drift which rarely exceed a few m/d (Elliott 2002, 2003) or 1–2 km over the life span (Humphries & Ruxton 2003). The slope obtained for the relationship between herbivore and periphyton δ13C was 1·02 (± 0·09), but the f estimate obtained was −1·77‰ (Fig. 7b), which likely reflects the tendency for grazing invertebrates to select certain fast growing algae from fast flowing microenvironments (Finlay et al. 2002; Rasmussen & Trudeau 2007).

Figure 7.

(a) δ13C of invertebrate predators against average signatures of invertebrate prey (average of primary consumers, herbivores, filterers and collectors) for the same sites from which the predators were obtained. The regression line obtained (solid line) was IPR δ13C = 4·1 + 1·13 invertebrate prey δ13C r2 = 0·96, P < 0·0001. The dashed line is the 1:1 line. (b) δ13C of herbivorous invertebrates plotted against periphyton signatures collected at the same sites. The regression line obtained (solid line) was HER δ13C = −1·14 + 1·02 PER δ13C r2 = 0·80, P < 0·0001. The dashed line is the 1:1 line, slope is not significantly different from 1. f = 1·77.


In contrast to the overall constancy of the δ13C signatures of terrestrial biota, periphyton, invertebrates and fish in rivers all exhibited strong signature gradients that reflect the gradients that geochemists have described for DIC signatures. In addition, signature slopes exhibited by aquatic invertebrates closely matched those of periphyton. Our results show that along-stream signature slopes of juvenile salmon are significantly less, than those of stream invertebrates; whereas those of sedentary fish taxa and stream ‘resident’ salmonids, whose range is restricted by physical barriers, more closely match those of local stream invertebrates. Movement estimates for juvenile Atlantic salmon, calculated from the slope deviations, indicate that these fish may be moving over considerable distances within the river network (~20 km). We present this approach to estimating the scale of movement for two main reasons. (i) Stable carbon isotopes are widely used for studying food web interactions, and it is important to consider that movement may affect these signatures more than has been widely appreciated, and needs to be considered when comparing signatures of consumers and local resources. (ii) In addition, a tracer method, based on time integrative isotopic signatures, that can be used to estimate the scale of feeding movements, may prove an effective complement to other methods (e.g. radiotelemetry) which provides detailed information on short-term movements, but has limitations with regard to the scale of time and space and to the number of organisms that can be tracked.

possible biases in estimation of movement from signature disequilibrium

While there seems little doubt that the departure of signature slopes of free-ranging salmonids from those of stream invertebrates can be related to movement, it should be recognized that movements of only a few kilometres from stream tributaries to the mainstem may have as big an impact on signatures as very long movements (tens of kilometres) along the river mainstem. Our estimates of β (kilometre) assume that all movement paths within the drainage are equally likely, but in many situations, this might not be true. Thus, supplementary information from telemetry or other types of movement studies (Gowan & Fausch 1996; Cunjak et al. 2005) could provide information on the types of movements that actually occur in a population, and thereby provide a context for interpreting signature information that may lead to more realistic estimates of the scale of movement for particular populations.

Estimates of the scale of movement based on signature disequilibrium might in some cases be expected to differ considerably from estimates obtained by direct methods (tagging and telemetry), or genetic markers. Such differences could result when organisms undertake movements that involve little or no feeding (e.g. spawning runs or movements to overwintering habitats) and therefore will have little effect on isotopic signatures. Even when feeding does occur, the amount of time spent in a particular habitat might not directly reflect the amount of growth achieved there. Therefore, estimates of movement obtained with a variety of methods could complement each other effectively, since each have potential biases.

Although carbon signatures to some degree reflect the entire growth history of an animal, the extent to which carbon signatures are affected by past growth events depends on the dynamics of tissue accumulation and replacement. Thus, recent growth history should have a greater effect on signatures than the earlier feeding history, and, the difference will depend mainly on the specific growth rate of the animal (Hesslein, Hallard & Ramlal 1993; Bosley et al. 2002), which can depend on species, feeding conditions, size and age and many other factors. While specific growth rate generally decreases with age and size, it seems likely that mobility will generally increase. Signature half-life is likely not a major source of bias for the estimates of mobility generated in this particular study, because most of the fish data used in this study came from small fish, aged 1+ to 2+ from cold-water streams, whose growth rates were likely fairly similar; however, the ability to correct such estimates for differences in specific growth rate using tissue turnover models (Herzka & Holt 2000; Perry, Bradford & Grout 2003) might potentially improve estimates of mobility based on signature disequilibrium, and to allow the method to be refined and applied on a shorter time scale.

do other studies on fish movement support our tracer results?

Studies on the mobility of sculpins (Gray, Cunjak & Munkittrick 2004) detect movements on the scale of only a few metres, and are therefore in agreement with our estimates, and the combined telemetry and stable isotope studies of Cunjak et al. (2005) clearly support the underlying principle that resident fish, have signatures that match local food sources and that fish which feed over larger spatial scales have signatures that reflect this spatial averaging. However, our results, which indicate that free-ranging juvenile salmonids are moving several kilometres over the course of their life history, are at odds with most of the estimates from early mark–recapture studies which support the view that stream salmonids, even when unencumbered by barriers, generally confine their feeding activities within small reaches a few tens of metres in length (Gerking 1959). More recent studies, based on radiotelemetry, newer methods of marking and monitoring movements, and larger spatial and temporal scales, have however cast considerable doubt on this restricted movement paradigm and have frequently shown stream salmonids to move long distances (hundreds of metres to tens of kilometres) especially in association with seasonal habitat shifts and changes in flow (Riley, Fausch & Gowan 1992; Gowan et al. 1994; Gowan & Fausch 1996).

While juvenile Atlantic salmon are at times rather sedentary fish that utilize the same feeding territories for extended periods (Rodriguez 2002), they have been shown to undertake a variety of movements throughout their juvenile life history (typically 2–4 years), including movements of fry from redds, movement to feeding territories, movement to and from winter habitat (autumn and early summer), and finally movements associated with smolting. Some studies have shown considerable movements within river systems, including feeding movements to and from estuaries (Cunjak, Chadwick & Shears 1989; Cunjak & Randall 1993), and Kennedy et al. (2002) using Sr isotope tracers in otoliths showed that some Atlantic salmon parr moved back and forth among stream tributaries within the Connecticut river system, movements on the scale of tens of kilometres.

extending the application of the method – longer food chains

The data set illustrating the application of our method is limited in that it does not include piscivores, which are often mobile, and in addition, feed on mobile prey. It would be interesting to compile a data set on a river system that included such predators. If the movement history of the piscivore is independent of that of the prey, there would seem to be little issue with simply treating their movement estimates along the river gradient as additive. Thus, one would expect that the slope reduction from 1 in their consumer/resource signature relationships should equal the difference in their slopes along the river gradient. Similarly, a nonmobile piscivore feeding on mobile prey would be expected to have similar signature slopes along the river gradient to that of its prey, and the consumer-resource signature plots should have slopes near 1. The complication arising from the piscivore and its prey having correlated movement histories is interesting, but we will not dwell on it here.

river gradients altered by reservoirs or lakes

Reservoirs and lakes along river courses can significantly alter the δ13C signature gradients along rivers. Tailwaters receiving cold hypolimnetic discharges from reservoirs generally have signature gradients immediately below the dam reset to the depleted levels normally found in headwaters as a result of the predominance of respiratory processes in the hypolimnetic reservoir water. Signatures then become enriched downstream within about 10 km as atmospheric and tributary influences re-assert themselves, and these signature gradients are transferred up the food web to consumers (Angradi 1993). Thus, the signatures of fish living and moving within tailwaters should be amenable to a gradient analysis similar to that outlined in this paper.

The presence of large lakes along a river course, generally accelerates the atmospheric equilibration of DIC signatures in their surface waters (Yang et al. 1996) and this can lead to significant enrichment of consumer δ13C especially in the littoral benthos of the lake (Hecky & Hesslein 1995) or in the river biota downstream of lakes. (Bunn et al. 1989; DeBruyn & Rasmussen 2002; Jardine et al. 2005a,b). The effects of lakes on DIC δ13C are however highly variable as a result of variability in geochemistry, depth and lake metabolism (Bade et al. 2004), so the gradient approaches to estimate mobililty of consumers could be applied in particular situations where gradients exist wither within the lake or downstream, but models might be difficult to generalize.


We thank Rob McLaughlin, Rick Cunjak , Kelly Munkittrick and Will Warnock for detailed comments on an earlier version of the manuscript, and Laurence Piche for assistance in the field. Kevin McCann made some important clarifications to the theory. The research was carried out at the CIRSA field station, and we are grateful for the opportunity to work there and the support we received from the staff. Financial support provided by an NSERC Strategic and Discovery grants to JBR. We would also like to acknowledge the useful comments of two anonymous reviewers.