Biplot presentation of diet composition data: An alternative for - - PDF document

See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/245587818

Biplot presentation of diet composition data: An alternative for fish stomach contents analysis

Article in Journal of Fish Biology · April 2000

DOI: 10.1111/j.1095-8649.2000.tb00885.x

CITATIONS

58

READS

477

3 authors, including: Some of the authors of this publication are also working on these related projects: Multivariate analyses View project Funtional diversity View project Véronique de Billy Office Français de la Biodiversité

28 PUBLICATIONS 220 CITATIONS

SEE PROFILE

Sylvain Dolédec Claude Bernard University Lyon 1

159 PUBLICATIONS 12,145 CITATIONS

SEE PROFILE

All content following this page was uploaded by Sylvain Dolédec on 22 September 2017.

The user has requested enhancement of the downloaded file.

Journal of Fish Biology (2000) 56, 961–973 doi:10.1006/jfbi.1999.1222, available online at http://www.idealibrary.com on

Biplot presentation of diet composition data: an alternative for fish stomach contents analysis

• V.  C  B*§, S. D†  D. C‡

*Cemagref Aix-en-Provence, Unite ´ de recherche ‘ Hydrobiologie ’, BP 31 le Tholonet, 13612 Aix-en-Provence Cedex 1, France; †CNRS 5023, Laboratoire des Hydrosyste `mes Fluviaux, Universite ´ Claude Bernard Lyon 1, 43 bd. du 11 Novembre 1918, 69622 Villeurbanne Cedex, France and ‡CNRS 5558, Biome ´trie et Biologie Evolutive, Universite ´ Claude Bernard Lyon 1, 43 bd. du 11 Novembre 1918, 69622 Villeurbanne Cedex, France (Received 2 October 1999, Accepted 23 December 1999)

A multivariate analysis derived from principal components analysis (PCA), and which allows the investigation on diet composition data, is introduced. To illustrate the method, prey composition data of stomach contents of brown trout Salmo trutta L. collected in a regulated stream were used. The diet composition, foraging strategies and related patterns of fish diet variation were analysed at a macrohabitat scale (i.e. riffles and glides) by way of biplots. These graphical presentations were consistent with PCA on proportions.

2000 The Fisheries Society of the British Isles

Key words: principal component analysis; stomach contents; feeding habits; diet variation; habitat variation; brown trout.

INTRODUCTION Several methods have been proposed to study fish diet (Strauss, 1979; Hyslop, 1980; Mohan & Sankaran, 1988; Costello, 1990; Tokeshi, 1991; Cortes, 1997). For example, feeding habit variations were investigated in relation to biological (age, size), ethological (intra- and inter-specific relationships), spatial (habitat, location) and temporal (seasonal, diurnal) patterns. In such studies, diet patterns were generally analysed at the population level often neglecting the differences among the feeding habits of individual fish (Bridcut & Giller, 1995). However, investigations at the individual level can display important informa- tion on diet composition and feeding strategy. According to Amundsen et al. (1996), a ‘ population with a narrow niche width must necessarily be composed

• f individuals with narrow and specialized niches ’ whereas ‘ a population with a

broad niche may consist of individuals with either narrow or wide niches, or a combination of both ’. Thus, a broad niche width may be the result from either a true generalist behaviour of each individual of a population (high within- individual variation) or a specialization of the individuals of the population on different prey (high between-individual variation). Therefore, a diet study should necessarily include the contribution of the intra-individual diet variations, i.e. variation in the use of resources by each individual (the within-phenotype component, WPC). In addition, such a study should evaluate the inter-

§Author to whom correspondence should be addressed. Tel.: (33) 4 42 66 99 30; fax: (33) 4 42 66 99 34; email: debilly@servaix1.aix.cemagref.fr 961 0022–1112/00/040961+13 $35.00/0 2000 The Fisheries Society of the British Isles

individual diet variation, i.e. variation in the use of resources among individuals (the between-phenotype component, BPC) (Wootton, 1990; Bridcut & Giller, 1995; Amundsen et al., 1996). Qualitative (diet composition), semi-quantitative (prey proportions) and quan- titative (consumption rates) investigations are available from stomach content data sets. However, the total abundance of prey items depends on various factors that are not easy to control. These factors include, for instance, the food availability, the rapidity of prey digestion and the hierarchical interactions among predators (Hyslop, 1980). As a result, quantitative variations of prey items among stomach contents can be significant and can generate differences among individuals. Because the use of proportions removes the unequal weight among individuals, semi-quantitative investigations are more appropriate for analyses at the individual level. The multivariate nature of data on diet composition (large number of prey-columns and individuals-rows) implies that a preliminary examination of the similarities among individual diets may be required before testing hypoth-

• eses. Crow (1979) was the first to apply principal component analysis (PCA) as

a clustering procedure for fish diet analysis. Later, classical multivariate analyses have often been used for synthesizing large data arrays on fish food (Bridcut & Giller, 1993; Sempeski et al., 1995). Nevertheless, although specific multivariate techniques potentially helpful for studying stomach contents have been intro- duced into general ecology (Ter Braak, 1983), rarely have they been used for fish. In this paper, a multivariate analysis is proposed specifically designed for diet composition data. This method, introduced by Aitchison (1983), is robust in case of a large number of prey. It is based on a principal component analysis performed on a proportion table having each row total equal to 1. In the following, this analysis is abbreviated to %PCA. The use of this method implies the application of a geometric concept (Gower, 1967), which is consistent with the numerical concept of the biplot presentation (Gabriel, 1971, 1978, 1981): the fish individuals and their prey items are analysed simultaneously and plotted on the same graph. Thus, the analysis is clearly established at the level of the

• individuals. The method is described in full and applied to trout diets to depict

the feeding habits of individuals and the relationship to their habitat use. %PCA is shown as an alternative to traditional techniques for detecting patterns in individual diet and foraging behaviour. Furthermore, as for other multivariate techniques, variance partitioning (Dole ´dec & Chessel, 1987, 1989; Yoccoz & Chessel, 1988; Borcard et al., 1992) is available for testing the diet variation at the population or higher level. MATERIAL AND METHODS

PRINCIPLE OF %PCA Let A=[aij] contain the number of prey items (columns, 1jp) found in the stomachs

• f n individuals (rows, 1in). Let P=[pj/i] contain the proportion of the jth prey in the

ith stomach so that p

j=1 pj/i=1 (i.e., 100%). Thus data are expressed as percentage of the

row total. The usual graphical display of such data consists of a triangular diagram available in the simplest case where p=3 [Fig. 1(a)]. In the case of more than three categories, the triangular diagram is no longer feasible. So a reduction of the dimensions

• f matrix P is necessary to enable graphical display. The triangular diagram can be

962 .      .

• btained through a column centred PCA on proportions (%PCA) (Ter Braak, 1983). Let

pj=1/nn

i=1 pj/i be the mean for prey j. Let P0=[pj/ip

¯j] be the centred table (i.e., the centroid of the overall population is situated at the origin of axes). %PCA incorporates the following steps:

26

7

24% 44% 1st axis 2nd category (c) 25 32% 2nd axis 3rd category 1st category 100% 0 – 100% (b) 1 –1 –1 1 1

1 4 24 25 21

3

15 19 18 13 10 7 14 16 23 22 11 12 20 2 8 5 3

2 100% (a)

21 24 25 12 4 9 20 17 6 2 3 8 22 5 11 14 26 19 15 13 18 10 23 16 1

6 9 17

F. 1. (a) Triangular presentation of theoretical composition data using three categories. (b) Vector presentation of the same data after a %PCA. The categories are plotted on the first factorial plane using the components of the first two eigenvectors (noted cj in equation 1). The length of arrows 1, 2 and 3 are proportional to the weight of categories. Sample 25 is positioned (position noted mi in equation 2) proportionally to its percentage of each category by averaging (noted mi in equation 2). (c) Mechanical presentation for sample 25 using the weight of each category presented by this sample at the arrow end of each category. Sample 25 is thus at the centre of gravity of the category distribution.

    963

(1) Compute covariance matrix C=1/nPt

0P0;

(2) Let k be the number of factorial axes selected for the graphical presentation (axes, 1kp). Find eigenvalues (k) and eigenvectors (uk), so that Cuk=kuk, with the additional constraint that Thus, the lengths of the eigenvectors uk are standardized to eliminate the scale differences among rows and columns (corresponding to 1/√) and to enable a biplot presentation. The eigenvectors uk are known as principal axes: u1 and u2 define the best-fitting plane (Pearson, 1901) often named first factorial plane. (3) On this first factorial plane, plot the prey items (or categories) cj (columns, 1jp) with the coordinates: (4) Then on the same plane, plot the individuals mi (rows, 1in) of P0 by averaging, with the coordinates: This averaging procedure differentiates %PCA from the usual PCA and enables the display of the mean structure (G) of the stomach content data set. (5) Lastly, plot on the plane the mean structure (G) with the coordinates: As a result, the biplot presentation (Gabriel, 1971; Ter Braak, 1983) consists in using eigenvector components (equation 1) to position the prey items (columns) for example on the first factorial plane. Individuals (rows) are plotted using the proportion (equation 2)

• f each eigenvector [Fig. 1(b)]. In other words, the interpretation of a biplot consists in

assigning the weight of each prey-column of a given individual-row at the arrow end of each vector [circles in Fig. 1(c)]. The equilibrium of the system is achieved at the centre

• f gravity, which gives the position of the individual-row [e.g., corresponding to sample

25 in Fig. 1(c)]. INTERPRETATION OF A BIPLOT OF STOMACH CONTENTS Each prey is linked to the population centroid by an arrow whose length is proportional to the relative abundance of this prey. Furthermore, the length of an arrow also depends on the variation of use of the corresponding prey among individual

964 .      .

• stomachs. Consequently, the dominant prey are dispersed on the first factorial plane

whereas the rare prey are concentrated around the origin. Each individual stomach is superimposed on the prey distribution, thus allowing the direct examination of the differences among fish diets. Foraging behaviour of the species can be deduced from the dispersion of individuals. For example, fish situated at the end of an arrow (e.g., prey 1 in Fig. 2) feed essentially on that dominant prey. Consequently, they have adopted a specialized foraging behaviour involving a weak within-phenotype component (WPC) and a narrow niche width. By contrast, a high between-phenotype component (BPC) is shown from the comparison of individuals specialized on dominant prey 2 with those feeding mainly on dominant prey 1. Fish positioned around the origin may feed either on all the dominant prey items or on rare prey. In the first case, each individual feeds in equivalent proportions on common prey: the stomach contents present a high WPC and the population a broad niche width. In the second case, a small number of individuals use prey that are unevenly consumed by the majority of fish. DATA SET To illustrate the method of %PCA, gut contents were analysed of brown trout sampled in the Neste d’Aure, a third-order French Pyrenean stream. The sampling station was situated at an altitude of 600 m. At this station, the discharge of the stream was regulated to 0·5 m3 s1, and its width ranged from 6 to 18 m. Individual trout were captured by electrofishing, 3 h after dawn. Two types of macrohabitats were sampled, glides and

• riffles. A total of 115 individuals were caught from three separated sites per macrohabitat

type. All the individuals were >100 mm [glides: n=60, 1995 mm; riffles: n=55, 2055 mm (mean..)]. The gut contents were collected immediately by stomach flushing (Seaburg, 1957) and preserved separately in formaldehyde (4%). In the laboratory, food items were identified and counted to the lowest possible taxonomic level (usually the genus level, except for the Dipteran larvae, which were identified to the family level). To assess the habitat characteristics of each glide and each riffle, water velocity (at 0·2, 0·4 and 0·8depth) and depth were measured at about 30 locations regularly spaced along transects so as to cover each replicate within a macrohabitat. At each location, the largest size and the height of a single randomly selected particle were

• measured. These grain size investigations were completed by measuring the largest size

2nd axis

High between-phenotype component

Dominant prey 3 1st axis Dominant prey 2 Dominant prey 1

Specialized on rare prey

• r

Generalized on dominant prey and high within-phenotype component High between-phenotype component Specialized towards the prey 1 and weak within-phenotype component Rare prey

F. 2. Interpretation of a diet analysis after a %PCA, including prey importance, foraging strategy and variation among individual fish feeding habits. Closed symbols stand for one single stomach content.

    965

• f four other particles. The relative roughness (Rrel) was computed according to Gordon

et al. (1992): where k was the height of the particle (m) and D the depth of the water (m) (Table I). DATA ANALYSIS Before analysing the feeding patterns, the distributions of the velocity, the depth, the particle size and the relative roughness of the substrate were compared among sites by way of Kolmogorov–Smirnov tests (Sokal & Rohlf, 1997). The size class distribution of trout was compared using the same test. A %PCA was performed on the stomach contents data set and trout diet differences were computed among replicates and macrohabitat types. The variation in prey abundance of trout stomachs was partitioned to estimate the variation associated with macrohabitat types and with replicates within macrohabitat types (Yoccoz & Chessel, 1988; Sabatier et al., 1989; Borcard et al., 1992). The effect of macrohabitats and replicates on prey abundances was measured by the ratio of between-group variance to total variance and expressed as percentage of variance explained (Poizat & Pont, 1996). The calculation of the percentage of variance among the replicates from a given macrohabitat provided an assessment of the quality of the replicates and the sampling

• method. The statistical significance of the macrohabitat and replicate effects were tested

by random permutation tests (1000 runs; Manly, 1991; Thioulouse et al., 1997). Finally, the diet features were estimated by the taxonomic richness (i.e. the number of prey items found in each stomach), and by the Simpson index of diversity which was computed for each stomach i and prey j, as S=1p

j=1 p2 j/i. A macrohabitat effect was

assessed on diet diversity and on the proportion of the most abundant prey in the stomach contents, using a one-way analysis of variance by ranks (Kruskal–Wallis).

RESULTS At the macrohabitat level, the current velocity was higher in the riffles, likewise the particle size (P<0·01) (Table I). Glides were deeper than riffles (P<0·01). At the level of replicates, the velocities of riffles no. 1, no. 2, and no. 3 were higher than that of glide replicates (P<0·01). The grain size in riffle no. 3 was higher and glide no. 3 was deeper compared with other replicates (P<0·01). Relative roughness was not significantly different among sites, whatever the level of investigation. The size class distribution of trout was similar among replicates and macrohabitats (P<0·01) except for trout individuals captured in glide no. 2, which were slightly smaller.

%PCA ON GUT CONTENTS

According to the eigenvalues [Fig. 3(a)], the first two axes of %PCA (respect- ively 44·7 and 24·5% of the total variation) were sufficient to illustrate the main structure of the diet composition. Three prey items, Ephemerella ignita, Simuliidae, and exogenous prey (composed by terrestrial invertebrates and adult stages of aquatic invertebrates) out of 41, dominated the gut contents [Fig. 3(b)]. The position of individual gut contents was a function of the macrohabitat types [shown by ellipses in Fig. 3(b)]. Individual trout caught in the riffles were grouped together along the Simuliidae arrow, whereas those collected in the 966 .      .

T I. Physical characteristics of the six sites in Neste d’Aure Velocity (cm s1) Depth (cm) Grain size (cm) Relative roughness (Rrel) n mean (..) n mean (..) n mean (..) n mean (..) Glide no. 1 31 19·9 (2·3) 31 27·5 (1·3) 160 15·0 (1·4) 32 0·28 (0·05) Glide no. 2 32 22·3 (2·5) 32 30·8 (1·5) 175 15·7 (1·5) 35 0·31 (0·04) Glide no. 3 30 16·3 (1·7) 30 39·7 (2·4) 185 18·6 (1·8) 37 0·30 (0·05) Riffle no. 1 32 39·9 (4·8) 32 24·1 (1·3) 170 17·7 (1·4) 34 0·42 (0·05) Riffle no. 2 36 42·5 (4·6) 36 23·9 (1·4) 165 22·7 (2·9) 33 0·28 (0·05) Riffle no. 3 29 43·2 (5·4) 29 29·1 (1·5) 200 23·3 (1·6) 40 0·33 (0·04)

Baetis rhodani

Simuliidae 2nd axis

R Baetis fuscatus

1st axis 0.8 0.5 Exogenous prey

Orthocladiinae

Ephemerella ignita

G

Simuliidae

Baetis rhodani

(b) –0.8 –0.9 (a) 2nd axis

Baetis fuscatus

1st axis Exogenous prey Ephemerella ignita (c)

G3 G1 G2 R1 R3 R2 Orthocladiinae

F. 3. Biplot of prey items and individual trout obtained from a %PCA. (a) Histogram of eigenvalues (the first two values are in black). (b) Distribution of individual stomach contents (squares) on the first factorial plane according to their prey items (arrows). Prey representing <5% of the total gut contents were omitted. Individuals caught in riffles (R) or glides (G) were labelled using open or closed squares, respectively; 95% confidence ellipses were used for making groups according to the macrohabitat types (——, riffle; – – – –, glide). (c) Same as (b) but individuals are grouped by macrohabitat replicates.

glides were either concentrated around the origin or placed along the exogenous prey arrow [Fig. 3(b)]. Finally, a grouping of individuals by replicates [Fig. 3(c)] showed that diets of trout collected in riffle no. 2 were closer to diets in glides than to the other riffles. The variance component explained by macrohabitats and replicates accounted for 28·2% of the total variance of prey abundance, of which 16·3% was explained by macrohabitat (Table II). When a %PCA was performed on glides and riffles separately, the percentage of variance explained among the replicates equalled 7·9 and 21·4%, respectively, which confirmed a higher variation among riffle replicates than among glide replicates [Fig. 3(c)]. The removal of stomach contents belonging to riffle no. 2, from the total prey abundance table did not change the structure observed in Fig. 3(b). The variance component explained by macrohabitats and replicates accounted for 31·3% of the total variation of prey abundance, of which 26·0% was explained by macrohabitat (Table II).

DIET ANALYSIS

Taxonomic richness ranged between two and 26 prey items in individual guts collected in glides and between two and 22 prey in riffles. Results of a one-way analysis of variance by ranks on richness showed a slight difference between glides and riffles (P=0·049). We found a significant difference for the Simpson index between the two macrohabitats (P=0·003). Again, riffle no. 2 was different from the other riffles, since its taxa richness and average Simpson index were closer to the values obtained for glides (Fig. 4). The removal of riffle no. 2 confirmed the presentation of Fig. 4 that riffle no. 1 and no. 3 demonstrated on average, a lower taxa richness (P=0·011) and a lower Simpson index (P<0·001) than glides (Table III). The six most abundant prey items (proportions>5% of the total) changed significantly with the macrohabitat types. The proportions of exogenous prey,

• E. ignita and Baetis fuscatus were significantly higher (respectively P<0·001;

P=0·0012; P=0·0057) in the gut contents of trout collected in glides than in riffles [Fig. 5(a)–(c)]. In contrast, the proportions of Simuliidae (P<0·0001),

• B. rhodani (P=0·001) and Orthocladiinae (P=0·0163) were higher in the

gut contents of trout collected from riffles [Fig. 5(d)–(f)].

T II. Variance partitioning including (Inc.)

• r excluding (Exc.) riffle no. 2

Inc. Exc. Total variance 0·140 0·150 Macrohabitat and replicate 28·2% 31·3% Macrohabitat 16·3% 26·0% Macrohabitat+replicate 22·0% 28·7% Macrohabitatreplicate 6·2% 2·6% Replicate/macrohabitat 5·7% 2·7%

Macrohabitat+replicate=additive effect; macrohabi- tatreplicate=interaction; replicate/macrohabitat= effect of replicate for a known effect of macrohabitat.

    969

DISCUSSION Crow (1979) noted the usefulness of multivariate analyses for detecting patterns in stomach contents but later he found (Crow, 1982) the application ‘ extremely cumbersome ’. Subsequently, progress has been made in providing appropriate multivariate analyses (Rodriguez & Magnan, 1995) allied to the increasing availability of suitable software. Among the numerous methods used for diet analysis, %PCA designed for tables composed of a large number of columns was helpful for investigating the structure of a set of stomach contents. The biplot of gut contents allows a qualitative analysis of the diet composition and the prey importance as well as a functional approach in terms of foraging behaviour and niche width. Furthermore, the simultaneous ordination of prey items and individuals permits a comparison among individual diets. In this study, %PCA indicated different trout diet composition and feeding strategies related to replicates and macrohabitat types, which could have been concealed with a population-level approach. For example, riffle no. 2 was separated graphically from other riffles. This site was not different from other replicates in terms of physical characteristics, but it was the only one situated upstream from a large pool. The use of space, daily activity and consequently feeding strategy of the trout in this riffle/pool system may have differed from the riffle/glide units. Thus, an investigation at the level of the replicates allowed the separation of individuals owing to their origin, and provided an assessment of the sampling method.

G1 G2 G3 R1 R2 R3 G1 G2 G3 R1 R2 R3 1.0 0.8 0.6 0.4 0.2 20 15 10 5 Location Simpson index Mean number of taxa F. 4. Average taxonomic richness (a) and Simpson index (b) (bar+1 ..) of the gut contents collected in the six replicates (G1–3, glides; R1–3, riffles).

T III. Range of the average taxonomic richness and Simpson index (means and rates) in glides and riffles including or excluding riffle no. 2 n Taxa richness Simpson index Glides 60 14–15 0·779 (0·084) Riffles, inc. no. 2 55 12–13 0·727 (0·080) Riffles, exc. no. 2 35 11–12 0·676 (0·108)

970 .      .

The high residual variation for macrohabitat effect (83·5%) demonstrates the importance of diet variations among individuals, and thus illustrates the trout’s flexibility and its opportunistic behaviour. Riffle trout revealed on the whole their relative feeding specialization on Simuliidae, characteristic of a narrow niche width. One individual was situated at the arrow end demonstrating a high specialization for this prey and a weak within-individual variation. Trout from glides showed either a specialist strategy in which exogenous prey were favoured,

• r a generalist strategy where different resources were eaten in the same

proportions (e.g., E. ignita, B. rhodani, B. fuscatus, Orthocladiinae). Thus, the simultaneous presentation of individual brown trout and prey items displays a meaningful between-individual variation (BPC) related to the macrohabitat types and demonstrates a broader niche width for the individuals collected in glides. Although the size-class distribution was not different among the individual trout collected in glides and riffles, hierarchy and intra-specific competition may have influenced the habitat use as dominant fish hold optimal positions for feeding (Bridcut & Giller, 1995). However, the variation in trout diet compos- ition and feeding strategy between the two macrohabitat types may also be a result of the differences in food availability related to macro-invertebrate vulnerability (Rader, 1997) and accessibility (Greenberg & Dahl, 1998).

60 10 20 30 50 40 (c) 60 % 10 20 30 50 40 (b) 60 10 20 30 50 40 (a) 60 10 20 30 50 40 (f) 60 10 20 30 50 40 (e) 60 10 20 30 50 40 (d) R3 R3 R3 R2 R2 R2 R1 R1 R1 G3 G3 G3 G2 G2 G2 G1 G1 G1 G1 G1 G1 G2 G2 G2 G3 G3 G3 R1 R1 R1 R2 R2 R2 R3 R3 R3 F. 5. Average proportion (bar+1 ..) of the six prey items dominating the gut contents (>5% of the total abundance) in the six replicates (G1–3, glides; R1–3, riffles). (a) Exogenous prey; (b) Ephemerella ignita; (c) Baetis fuscatus; (d) Simuliidae; (e) Baetis rhodani; (f) Orhocladiinae.

    971

Additional data sets including benthos and drift composition must be examined together with stomach contents to compare the availability of food resources and the consumption features (Sempeski et al., 1995).

SOFTWARE AVAILABILITY

All calculations and graphs were made with ADE-4 (Thioulouse et al., 1997). The package can be obtained freely by anonymous FTP to pbil.univ-lyon1.fr, in the /pub/mac/ADE/ADE4 directory. A documentation and downloading page is available at: http://pbil.univ-lyon1.fr/ADE-4.html, which also provides access to updates and user support through the ADEList mailing list.

• B. Statzner commented on an earlier draft. P. Baran and T. Lagarrigue helped with
• fieldwork. Financial support came from CEMAGREF and GIP hydrosyste

`mes. All this help is gratefully acknowledged.

References

Aitchinson, J. (1983). Principal component analysis of compositional data. Biometrika 70, 57–65. Amundsen, P. A., Gabler, H. M. & Staldvik, F. J. (1996). A new approach to graphical analysis of feeding strategy from stomach contents data-modification of the Costello (1990) method. Journal of Fish Biology 48, 607–614. Borcard, D., Legendre, P. & Drapeau, P. (1992). Partialling out the spatial component

• f ecological variation. Ecology 73, 1045–1055.

Bridcut, E. E. & Giller, P. S. (1993). Diet variability in relation to season and habitat utilisation in brown trout, Salmo trutta L., in a Southern Irish stream. In Production of Juvenile Atlantic Salmon, Salmo salar, in Natural Waters (Gibson,

• R. J. & Cutting, R. E., eds).

Canadian Special Publication of Fisheries and Aquatic Sciences 118, 17–24. Bridcut, E. E. & Giller, P. S. (1995). Diet variability and foraging strategies in brown trout (Salmo trutta): an analysis from subpopulations to individuals. Canadian Journal of Fisheries and Aquatic Sciences 52, 2543–2552. Cortes, E. (1997). A critical review of methods of studying fish feeding based on analysis

• f stomach contents: application to elasmobranch fishes. Canadian Journal of Fish

Aquatic Science 54, 726–738. Costello, M. J. (1990). Predator feeding strategy and prey importance: a new graphical

• analysis. Journal of Fish Biology 36, 261–263.

Crow, M. E. (1979). Multivariate statistical analysis of stomach contents. In Fish Food Habits Studies: Proceedings of the Second Pacific Northwest Technical Workshop (Lipovsky, S. J. & Simenstad, C. A., eds), pp. 87–96. Seattle: University of Washington Sea Grant Publication. Crow, M. E. (1982). Some statistical techniques for analyzing the stomach contents of

• fish. In Fish Food Habits Studies: Proceedings of the third Pacific Northwest

Technical Workshop (Cailliet, G. M. & Simenstad, C. A., eds), pp. 8–15. Seattle: University of Washington Sea Grant Publication. Dole ´dec, S. & Chessel, D. (1987). Rythmes saisonniers et composantes stationnelles en milieu aquatique. I. Description d’un plan d’observation complet par projection de variables. Acta Oecologica—Oecologia generalis 8, 403–426. Dole ´dec, S. & Chessel, D. (1989). Rythmes saisonniers et composantes stationnelles en milieu aquatique. II. Prise en compte et e ´limination d’effets dans un tableau

• faunistique. Acta Oecologica—Oecologia generalis 10, 207–232.

Gabriel, K. R. (1971). The biplot graphical display of matrices with application to principal component analysis. Biometrika 58, 453–467.

972 .      .

Gabriel, K. R. (1978). Least-squares approximation of matrices by additive and multiplicative models. Journal of the Royal Statistical Society, B 40, 186–196. Gabriel, K. R. (1981). Biplot display of multivariate matrices for inspection of data and

• diagnosis. In Interpreting Multivariate Data (Barnett, V., ed.), pp. 147–174. New

York: John Wiley. Gordon, N. D., McMahon, T. A. & Finlayson, B. L. (1992). Stream Hydrology: an Introduction for Ecologists. Chichester: John Wiley. Gower, J. C. (1967). Multivariate analysis and multidimensional geometry. The Statis- tician 17, 13–28. Greenberg, L. A. & Dahl, J. (1998). Effect of habitat type on growth and diet of brown trout, Salmo trutta L., in stream enclosures. Fisheries Management and Ecology 5, 331–348. Hyslop, E. J. (1980). Stomach contents analysis—a review of methods and their

• application. Journal of Fish Biology 17, 411–429.

Manly, B. F. J. (1991). Randomization and Monte Carlo Methods in Biology. London: Chapman & Hall. Mohan, M. V. & Sankaran, T. M. (1988). Two new indices for stomach content analysis

• f fishes. Journal of Fish Biology 33, 289–292.

Pearson, K. (1901). On lines and planes of closest fit to systems of points in space. Philosophical Magazine 2, 559–572. Poizat, G. & Pont, D. (1996). Multi-scale approach to species-habitat relationships: juvenile fish in a large river section. Freshwater Biology 36, 611–622. Rader, R. B. (1997). A functional classification of the drift: traits that influence invertebrate availability to salmonids. Canadian Journal of Fisheries and Aquatic Sciences 54, 1211–1234. Rodriguez, M. A. & Magnan, P. (1995). Application of multivariate analyses in studies

• f the organization and structure of fish and invertebrate communities. Aquatic
• Sciences. Special issue: Multivariate Statistics for Aquatic Ecology 57, 199–216.

Sabatier, R., Lebreton, J. D. & Chessel, D. (1989). Principal component analysis with instrumental variables as a tool for modelling composition data. In Multiway Data Analysis (Coppi, R. & Bolasco, S., eds), pp. 341–352. North-Holland: Elsevier Science. Seaburg, K. G. (1957). A stomach sampler for live fish. Progressive Fish-Culturist 19, 137–139. Sempeski, P., Gaudin, P., Persat, H. & Grolet, O. (1995). Diet selection in early-life stages of grayling (Thymallus thymallus). Archiv fu ¨r Hydrobiology 132, 437–452. Sokal, R. R. & Rohlf, F. J. (1997). Biometry—The Principles and Practice of Statistics in Biological Research, 3rd edn. New York: W. H. Freeman. Strauss, R. E. (1979). Reliability estimates for Ivlev’s electivity index, the forage ratio, and a proposed linear index of food selection. Transactions of the American Fisheries Society 108, 344–352. Ter Braak, C. J. F. (1983). Principal components biplots and alpha and beta diversity. Ecology 64, 454–462. Thioulouse, J., Chessel, D., Dole ´dec, S. & Olivier, J. M. (1997). ADE-4: a multivariate analysis and graphical display software. Statistics and Computing 7, 75–83. Tokeshi, M. (1991). Graphical analysis of predator feeding strategy and prey impor-

• tance. Freshwater Forum 1, 179–183.

Wootton, R. J. (1990). Ecology of Teleost Fishes. London: Chapman & Hall. Yoccoz, N. & Chessel, D. (1988). Ordination sous contraintes de releve ´s d’avifaune: e ´limination d’effets dans un plan d’observations a ` deux facteurs. Compte Rendu Hebdomadaire des Se ´ances de l’Acade ´mie des Sciences, Paris, D: III 307, 189–194.

    973

View publication stats View publication stats