Ecological Indicators 34 (2013) 571–579

Contents lists available at SciVerse ScienceDirect

Ecological Indicators
journal homepage: www.elsevier.com/locate/ecolind

Original article

Efficacy of population size structure as a bioassessment tool in

freshwaters
Christina A. Murphy a,∗ , Frederic Casals b , Carolina Solà c , Nuno Caiola d ,
Adolf de Sostoa e , Emili García-Berthou a
a
Institute of Aquatic Ecology, University of Girona, Girona, Spain
b
Department of Animal Production, Lleida University, Lleida, Spain
c
Catalan Water Agency, Barcelona, Spain
d
IRTA Aquatic Ecosystems, Sant Carles de la Ràpita, Spain
e
Department of Animal Biology (Vertebrates), University of Barcelona, Barcelona, Spain

a r t i c l e i n f o a b s t r a c t

Article history: Bioassessments are used to measure system health and assess disturbance. While fish-based freshwater
Received 27 March 2013 bioassessments are cost-effective and perform well in speciose systems, such bioassessments remain
Received in revised form 6 June 2013 difficult to implement in species-poor Mediterranean regions. Population size structure metrics may
Accepted 10 June 2013
provide meaningful biological information where depauperate communities preclude the richness and
composition measures generally used. We focus our assessments of population size structure responses
Keywords:
to anthropogenic perturbation on one of the most widespread native stream fish (Squalius laietanus).
Anthropogenic perturbation
We explore a number of population size statistics as metrics for a Mediterranean region, where current
Covariance
Human disturbance
bioassessments perform poorly. Our sampling encompassed 311 sites across Catalonia (NE Spain) where
Mediterranean we characterized anthropogenic perturbation using a summary of impacts, including local data on stream
Squalius laietanus condition and landscape indicators of degradation, via a principal component analysis. Anthropogenic
Variation partitioning perturbation in streams was collinear with altitudinal gradients and highlights the importance of appro-
priate statistical techniques. Of the population size structure metrics explored, average length was the
most sensitive to anthropogenic perturbation and generally increased along the disturbance gradient.
Although we expected to find consistent changes in variance, kurtosis, and skewness, the observed rela-
tionships were weak. River basin mediated responses suggest the importance of environmental landscape
factors. The unexpected increases of mean S. laietanus body size with anthropogenic perturbation, strong
effects of river basin, collinearity with spatial gradients and the species-specific nature of responses pre-
clude the direct application of size structure in freshwater bioassessments. Although its application in
fish-based freshwater bioassessments appears difficult, population size structure can provide insights in
species-specific applications and management.
© 2013 Elsevier Ltd. All rights reserved.

1. Introduction there exists insufficient species diversity for the community-based

Bioassessments provide useful tools to measure aquatic system
health and to communicate necessary environmental manage-
ment to society (Karr and Chu, 1999; Simon, 1999). While
fish-based bioassessments have historically performed well in
speciose regions, such bioassessments are difficult to implement
in species-poor ecosystems (Moyle and Randall, 1998). A shift
from community-based measures to population or individual-
based measures might improve fish-based bioassessments where
Abbreviations: AP, anthropogenic perturbation; VP, variation partitioning.
∗ Corresponding author at: Department of Fisheries and Wildlife, Oregon State
University, Corvallis, OR, USA. Tel.: +1 541 505 1393; fax: +1 541 737 3590.
E-mail address: christina.a.murphy@gmail.com (C.A. Murphy).
measures. Even while supporting high levels of endemism, fish
species-poor Mediterranean regions such as the Iberian Peninsula
have had difficulty in implementation of currently used bioassess-
ment indices to assess aquatic health (Benejam et al., 2008).
Interestingly, population size metrics were already considered in
the initial development of freshwater bioassessments (Karr et al.,
1986a,b), but have since been neglected. Population size metrics
may provide meaningful biological information where depauper-
ate communities preclude the application of richness and species
composition metrics, which are the most frequent in fish-based
bioassesments.
Body size is a fundamental characteristic of organisms and
arguably the most important trait affecting the ecological perfor-
mance of individuals (Persson and de Roos, 2007). The implications
of body size on growth, mortality and trophic interactions

1470-160X/$ – see front matter © 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ecolind.2013.06.007
572 C.A. Murphy et al. / Ecological Indicators 34 (2013) 571–579

highlight the importance of size structure for populations (Werner perturbation. A predictable size structure relationship with anthro-
and Gilliam, 1984; de Roos et al., 2003; Savage et al., 2004; Brown pogenic perturbation would suggest a practical alternative for
et al., 2007). A variety of methods have been proposed relating estimating disturbance impacts.
population attributes such as rates of growth, mortality, and com- Here we report an increase in mean length, generally an unex-
petition strength to population size structure using characteristics pected pattern in a stream fish species, which might explain the
such as average size, skewness, hierarchy/inequality, modality, disuse mentioned above. We analyzed the relationship of popu-
variance and kurtosis (Weiner, 1985; Benjamin and Hardwick, lation size structure as a response to anthropogenic perturbation
1986; Hara, 1988; Knox et al., 1989; Walters and Post, 2008). Pop- in the four most abundant fish in NE Spain in order to compare
ulation size structure has commonly been applied to assessments size-structure metrics and inform their possible inclusion in future
of vegetation (Deal et al., 2010) and marine communities (Rochet bioassessments. Our objectives were: (i) to quantify changes in
et al., 2010), with very few examples of size-based bioassess- size structure associated with anthropogenic perturbation and (ii)
ments to quantify anthropogenic impacts in freshwater systems to evaluate the relative utility of each size structure metric. We
(for a few exceptions see Bonar, 2002 and Basset et al., 2012). The hypothesized that as a result of anthropogenic perturbation we
disuse of size structure is especially interesting as size structure should expect changes in length frequency distributions reflected
was mentioned as a health metric in early freshwater bioassess- in a variety of size metrics such as maximum length, average length,
ment literature (see Table 2 in Karr et al., 1986a,b), but has not skewness, kurtosis, and variance. We predicted that average length
been incorporated. Given the recognized importance of size struc- would be the more robust statistic and that the response to anthro-
ture in freshwater populations and assuming that it may be easily pogenic perturbation would be species-specific.
affected by disturbances, it seems surprising that modern fresh-
water bioassessments have completely overlooked size structure 2. Materials and methods
in favor of indicators such as species richness, species composi-
tion and abundance (see reviews in Karr and Chu, 1999 and Simon, 2.1. Sampling and study area
1999). Richness and composition metrics rely on fish communi-
ties as a whole, and may include tolerance estimates for individual We sampled fishes in Catalonia (NE Spain) over two years, from
species. The community dependence of these commonly used June through October 2007, and from March through September
metrics explains their lack of applicability in systems where com- 2008, at 364 sites across the region (Fig. 1), including every official
munities may be limited to small numbers of species and where stream water body recognized by the Catalan Water Agency (Agèn-
differences are often due to endemism. While the potential ben- cia Catalana de l’Aigua). Each site was sampled once. Of these sites,
efits of size structure may be offset in some cases by the costs 311 were successfully fished, 45 were dry and 8 were not fishable at
of measuring individuals, incorporating size metrics could foster the time of sampling due to flooding. The variation in sampling date
bioassessment development in regions with relatively depauper- did not affect our analyses below (see Appendix A in Supporting
ate freshwater faunas such as Mediterranean-climate ecosystems, Information). The Catalonia region comprises the northeast corner
where the application of current rapid bioassessment indices has of the Iberian Peninsula over an area of 32,114 km2 . Bordered to the
proven problematic and difficult to implement (Benejam et al., north by the Pyrenees mountain range, the landscape is comprised
2008). of larger basins such as the Ebro and Llobregat rivers, correspond-
Anthropogenic perturbations impact populations through well ing to streams of inland and mountain origin, and smaller littoral
documented changes in mortality rates, growth, and behav- basins. Most of the population, around 5.5 of the approximately
ior resulting from direct and indirect effects, including habitat
alteration or loss, physiological stress and changes in resource
availability (Richmond, 1993; Gill et al., 2001). These changes can
be expected to alter size structure. However, the types of anthro-
pogenic perturbation and their potential effects on size structure
are diverse. In addition, different statistics may inform of differ-
ent aspects of the size structure of a population. Skewness reflects
changes in asymmetry (Guttal and Jayaprakash, 2008), variance
reflects the spread of the body sizes around the mean, whereas kur-
tosis is more difficult to interpret because it varies with peakedness,
tailedness and bimodality (DeCarlo, 1997). Skewness and variance,
although quite different statistics, are often highly positively cor-
related (Bendel et al., 1989). Maximum size is based on a single
individual, so it should be a less robust statistic but might change
under mass mortality or conditions that ultimately alter the growth
or survival of large fish.
Mean sizes are reflective of all of the individuals in a popula-
tion, and should exhibit noticeable changes as losses occur, be they
in small numbers of extreme sizes or large numbers of common
sizes. For example, if populations are strongly affected by a mass
fish kill, we could expect mean size to decrease as recruits domi-
nate the population (Haedrich and Barnes, 1997), and increase as
the perturbation ceases and the size structure recovers. However,
if a disturbance chronically precludes recruitment over time (e.g.,
siltation of fish spawning habitat), mean length should increase
as a few older individuals prevail (Tarr, 2000). Indeed many pop-
ulation size structure patterns might result from anthropogenic
Fig. 1. Location of the study area (Catalonia, NE Spain) and sampling sites (red
perturbation in theory. In practice, we are examining whether any squares), including sites with Squalius laietanus (blue circles, scaled to mean fork
of those patterns are consistent with documented anthropogenic length).
 C.A. Murphy et al. / Ecological Indicators 34 (2013) 571–579
7.5 million total inhabitants, resides within the lower elevation,
coastal province of Barcelona. The weather is characterized by a
Mediterranean climate with unpredictable, heavy autumn rains
following intense summer droughts. Human impacts vary widely
throughout the region, including hydrologic projects, diversions
and withdrawals for industrial, agricultural and urban uses, water
contamination and channel modifications, among others. Natu-
ral seasonal droughts, in many areas, are exacerbated by water
abstraction (Benejam et al., 2010).
The fish assemblages are characterized by low richness but high
levels of endemism and widespread introduced species (Benejam
et al., 2008). Some of the most prevalent native fishes across sites
are chub Squalius laietanus Doadrio et al. (2007), barbels Barbus
haasi Mertens (1925) and B. meridionalis Risso (1827) and brown
trout Salmo trutta Linnaeus (1758). Fishes were captured via elec-
trofishing using a portable generator (2.5–4.5 kW, 400–800 V): an
Electracatch WFC4 (Electracatch International, Wolverhampton,
UK) or a Hans Grassl ELT62II GI Honda GCV160 (Hans Grassl GmbH,
Schönau am Königssee, Germany). Streams were sampled over
50–200 m reaches with a single-pass method in order to ensure
coverage of all habitat types, normally including an area greater
than 100 m2 . A comparison of our sampling protocol with four-
pass removal with block nets in streams of this region indicated
that: (i) 50–100% of the species and 40–60% of the individuals are
generally captured with a single pass; and (ii) that the estimates
of species richness and composition of a single pass are represen-
tative of the assemblage and are not affected by the electrofishing
method used (Benejam et al., 2012). Although capture probability
increases both in general and in our study area with fish size and in
upstream sites, the relationship with size was weak and nonlinear
(Benejam et al., 2012) suggesting that our results are not affected
by differential capturability. Fish were anesthetized using MS222
(tricaine methanesulfonate), measured (mm) to fork or total length
(depending on species), weighed (to the nearest 0.001 g), checked
for DELT anomalies (Deformities, Eroded Fins, Lesions and Tumors)
and returned to the capture reach after sampling. When large num-
bers of fishes were captured at a site, subsamples of 100 individuals
chosen at random of each species were measured; this only applied
to 14 samples of chub; in other cases all individuals captured were
measured.
Total length of the reach fished, channel width, maximum
bank-full width, mesohabitat composition (% rapids, glides, pools),
refugia types (structural aquatic vegetation, submerged woody
vegetation, trunks, branches, and caves) and substrate were
recorded. Every 20 m, wetted width, depth and water velocity were
measured in the center of the stream as well as in the left and right
margins; these measurements were used to calculate sampled area
and average velocity. We also measured a number of water quality
parameters (such as temperature, conductivity, pH, alkalinity and
concentrations of dissolved oxygen, nitrate, nitrite, ammonia and
phosphate).
2.2. Anthropogenic perturbation
Our measure of anthropogenic perturbation was a global per-
turbation indicator, combining local data of stream condition with
landscape indicators of degradation. Although different impact
types increase along the longitudinal river gradient, global pertur-
bation indicators have been used previously with success (e.g., Pont
et al., 2006) and, as in Ferreira et al. (2007), were more responsive
to fish metrics than partial indicators in our case (see Appendix A).
Here, we refer to partial indicators (e.g., water quality or habitat
quality) as those which were specific to a disturbance type. There-
fore, we did not consider individually different disturbance types
since many were highly correlated and are likely to have interactive
ecological effects. Local data on stream condition were obtained
573
during the fish sampling and included water quality metrics, a
locally adapted Rapid Bioassessment Index (RBI; modified from
Barbour et al., 1999) and a visual impacts checklist (see Appendix
A). Water quality indicators included pH and concentrations of
dissolved oxygen and ammonia. Visual impacts included the quali-
tative evaluation of the presence of foam or discoloration, turbidity,
smell, channelization, erosion, hydrologic projects, roads and trails
and rocks with black undersides.
In addition to local field data, an official statistic of anthro-
pogenic perturbation of water bodies (hereafter referred to as
the ‘risk of non-compliance’ measure) from the study area was
obtained from the Catalan Water Agency. The risk of non-
compliance refers to the targets of the Water Framework Directive
from European Union and integrates a number of perturbations
including morphological alterations, changes in flow regime, land
use in the riparian zone and point and diffuse sources of pol-
lution (ACA, 2005, page 252; data and maps also available at
http://alturl.com/xjjxw). We view this risk of non-compliance mea-
sure as complementary to the local, field-based data of stream
condition, since the former explicitly considers a variety of pertur-
bations in the whole drainage basin upstream the sampling point
and for a longer time period. We summarized anthropogenic per-
turbation using a principal component analysis (PCA) of the sum
of the RBI scores, sum of visual pressures, risk of non-compliance,
pH and concentrations of dissolved oxygen and ammonia. Based on
the eigenvalue and composite variables the first axis was retained,
which we will refer to as “anthropogenic perturbation” or AP (see
Appendix A in Supporting Information for further information).
2.3. Statistical analyses
Of the 311 sites fished, 66 contained S. laietanus. Although we
analyzed the four most prevalent native fish species (see above),
we focus our results primarily on S. laietanus, which has shown
greater declines in abundance and spread (Aparicio et al., 2000;
Maceda-Veiga et al., 2010), is found throughout the sample region,
lacks targeted fishery impacts and displayed the clearest con-
trasting responses. To compare changes in size structure metrics we
computed the mean, variance, coefficient of variation, maximum,
skewness and unbiased estimate of kurtosis (G2 in Joanes and Gill,
1998) from the length data using the ‘describe.by’ function in the
‘psych’ package (Revelle, 2011) for the R software environment (R
Development Core Team, 2011).
To account for heteroscedasticity, log-transformed mean
length was also included in analyses. We also considered
presence–absence and abundance data. S. laietanus size distri-
butions were generally unimodal. Catches for S. laeitanus were
generally high and there were no obvious outliers in the distri-
bution of response variables. To account for possible bias due to
sampling dates, we used the ‘posixdate’ function in R to create
‘season’ as an explanatory variable. We compared this to the ‘best’
model to check for significant improvement and did not find sup-
port for strong seasonal effects (e.g. season did not significantly
improve the model nor explain the patterns observed). Abundance
was always divided by the fishing area to account for sampling
effort. We analyzed all the size metrics as well as presence–absence
and abundance data for the effects of anthropogenic perturbation.
In order to evaluate changes in size structure metrics due to
anthropogenic perturbation, we had to account for collinearity.
Many ecological processes vary along the upstream–downstream
gradient of rivers (see e.g. Carmona-Catot et al., 2011), including the
growth rate of Squalius species (Tedesco et al., 2009). We aimed
to disentangle the relative roles of natural spatial gradients and
anthropogenic perturbation in the size structure of S. laietanus. To
this end, we used altitude as a surrogate of the longitudinal posi-
tion in the stream because: (i) a PCA of spatial variables not directly
574 C.A. Murphy et al. / Ecological Indicators 34 (2013) 571–579
affected by anthropogenic perturbation (stream order, distance to
the source, distance to the river mouth, total catchment area and
altitude) indicated that it was the most appropriate (Appendix A,
Fig. A1); (ii) altitude data are more informative and interpretable
than using PCA scores and have been associated with numerous
natural environmental gradients, including effects on body size
(Chown and Klok, 2003; Körner, 2007). In all cases, altitude was
log-transformed because the statistical assumptions of paramet-
ric techniques were thus more satisfied. To describe the degree of
collinearity, we calculated the generalized variance inflation fac-
tors for each of the three predictors using the ‘vif’ function in the
‘car’ package (Fox and Weisberg, 2011) for R (Appendix A). The
relationship between altitude and anthropogenic perturbation is
not surprising as the concentration of urban areas, industry and
agricultural land uses in the lowlands and littoral zones increases
disturbance along the longitudinal gradient (Fig. S2).
Addressing collinearity is critical as even low levels can cause
variance inflation and lead to wrong conclusions (Graham, 2003).
Additionally, simply eliminating collinear variables is not safe as
they may have independent effects (Freckleton, 2011). Due to the
collinear nature of the anthropogenic perturbations and the natural
environmental gradient represented by altitude, our analyses relied
on tests appropriate to the situation. Here we present the results
of variation partitioning and those obtained comparing full and
simplified ANCOVA models in analyses with size metrics, or gen-
eralized linear models for presence–absence and abundance. Full
models, containing all three explanatory variables (AP, basin, alti-
tude) and their interactions, were compared to two simpler models,
containing basin and either AP or altitude (and their interactions),
with likelihood ratio statistics to test for significant improve-
ment. Akaike Information Criteria and model averaging were not
used, since high levels of collinearity result in biased parameter
estimates (see Murray and Conner, 2009 and Freckleton, 2011).
Additionally, although hierarchical partitioning analyses were also
conducted, results are not presented as they were generally simi-
lar to those obtained by variation partitioning and likely contain
rounding errors expected with 9 or more regressors, exceeded
due to basins (see Olea et al., 2010 and Walsh and Mac Nally,
2011).
Variation partitioning (VP) of univariate response variables uses
partial regression to measure variation explained uniquely and
jointly by different groups of predictors (Borcard et al., 1992; Monti
and Legendre, 2009). It is useful when an analysis includes factors
which result in overlaid effects. VP was used to determine the vari-
ation in size metrics, presence/absence and abundance explained
purely by anthropogenic perturbation alone and jointly with river
basin and altitude. We performed VP using the ‘varpart’ function of
the ‘vegan’ package (Oksanen et al., 2010) in R and the three sets
of predictors: altitude (log-transformed), river basin (recoded as
dummy binary variables with the ‘model.matrix’ function in R) and
the linear and quadratic components of AP. Significance was deter-
mined for testable fractions (variation due to pure fractions and
total variation due to AP) using permutation tests (‘anova’ function
with 999 permutations).
As alternative analyses, linear models with interaction terms
were used to build full and simplified models (R function ‘lm’) for
size metrics, including mean length, maximum length, kurtosis,
skewness and variance. Analyses of variance were used to compare
the variation explained by the models (function ‘anova’). In contrast
to VP, these analyses do not assume equal slopes of relationships for
different basins and allow for comparisons between more complex
models and more simple ones, such as those used in VP. Only the
linear component of AP was used to ensure comparability of nested
models. Generalized linear models (GLMs) were used in the cases
of presence–absence and abundance (R function ‘glm’). Overdisper-
sion of abundance data necessitated the use of quasipoisson errors,
and chi-squared comparisons. Presence–absence was assessed
using binomial errors and models were compared using an F-
statistic.
3. Results
Simple scatterplots showed that mean length of S. laietanus gen-
erally increased with anthropogenic perturbation and decreased
with altitude (Fig. 2). However, since altitude (log transformed)
and AP are correlated (Pearson’s r = −0.55, df = 298, P < 0.0005) it
is essential to understand the independent and joint effects of the
two predictors.
3.1. Variation partitioning
VP showed significant relationships between anthropogenic
perturbation (overall effects) and all of the response variables,
including S. laietanus abundance, presence/absence, mean size,
maximum size and size variance, excluding the coefficient of vari-
ation, kurtosis and skewness (Table 1).
The highest levels of explained variation relating to anthro-
pogenic perturbation were seen in mean length, and were due to
joint effects with basin (13.7%). Pure anthropogenic perturbation
effects were only significant for kurtosis, where the total variation
due to anthropogenic disturbance and joint effects was negligible.
VP also revealed that river basin alone, excluding joint effects, was
significant for S. laietanus abundance, presence/absence, mean size,
kurtosis, skewness and the coefficient of variation. In all of those
cases, basin was the strongest explanatory variable. Basin had no
significant effect on maximum length, the variance in length or log
transformed mean length.
3.2. Linear models
Basin effects were accounted for in the full and simplified linear
models and GLMs and the GLM and linear model results tended to
explain more of the observed variation than VP, likely due to the
better fit using different slopes for each basin (Table 2). Although
generally consistent in sign, different basins showed different rela-
tionships between the three variables (Fig. 2).
Full models were significantly better at explaining the variation
for analyses of S. laietanus mean length and variance when com-
paring the different linear models. This was also the case for the
generalized linear model of abundance. Maximum size, kurtosis,
skewness, presence/absence and the coefficient of variation were
all better explained by simplified or null models. The increase in
S. laietanus mean length was significantly correlated with slightly
lower abundances (corrected for area fished; R2 adj = 0.08, df = 63,
P = 0.015; Fig. A5). This increase in size with decreasing abundance
was much better explained with a full linear model (R2 adj = 0.41
df = 46, P < 0.0005) including abundance and basin, when compared
to nested single variable models (P < 0.005 for both).
4. Discussion
4.1. Collinearity between altitude, basin, and anthropogenic
perturbation
The amount of shared variation between the combinations of
basin, anthropogenic perturbation and altitude differed depending
on the response variable measured. VP indicated that all of the vari-
ation in mean length attributable to anthropogenic perturbation
was shared with other variables. This finding strengthens the argu-
ment that accounting for collinearity without data loss is essential
to assessing these effects. In addition, our results highlight the
 C.A. Murphy et al. / Ecological Indicators 34 (2013) 571–579 575

Fig. 2. Scatterplots with fitted linear regressions by river basin for relationships between mean S. laietanus length, anthropogenic perturbation and altitude.

importance of basin in structuring population responses. We docu-
mented changes in the gradients of the relationships, including two
basins that show little to no relationship between anthropogenic
perturbation and mean length compared to the others (Fig. 2).
The inclusion of basin and its importance is consistent with
well documented landscape-scale influences in riverine systems
(Richards et al., 1996; Williams et al., 2003; Coulthard et al., 2005).
One should expect basin-specific responses even under similar dis-
turbance regimes as they are mediated by local environmental
conditions. For example, the removal of riparian vegetation would
be expected to cause more bank mobilization and material transfer
in a basin where the substrate is fine dominated than in a basin
dominated by bedrock lithology. In two basins with similar dis-
turbances but differing thermal regimes, we might expect those
disturbances to have different levels of sub-lethal and lethal effects
as the ability of the local fishes to handle stress may be temperature
dependent. Catalonia is a small, heterogeneous region with con-
trasting landscape, climates, and streams. As every official water
body in Catalonia was represented in our sampling, we feel confi-
dent that these basin effects are real and not spurious.
The collinearity between anthropogenic perturbation and the
longitudinal variation represented by altitude is consistent with
our expectations based on the distribution of human activities in
Catalonia and most regions worldwide, which tend to be high-
est in lowlands and near larger rivers. This has been documented
in general, as well as specifically for riparian areas in Mediter-
ranean regions (Corbacho et al., 2003). Relationships between
human activities, including land use, and altitudinal gradients
would explain why we note an increase in anthropogenic pertur-
bation toward lower reaches, corresponding to decreasing altitude.
Using the appropriate statistical tools is essential in order to
retain the information within this shared variance. Here, we have

Table 1
Results of variation partitioning (% variation explained) for selected response variables of Squalius laietanus using basin, altitude, and anthropogenic perturbation (AP) as
predictors. Only pure and total (“pure” + “shared”) effects are testable (not the “shared” or “joint” effects alone).

Response variable AP Pure basin Pure altitude Pure AP Basin and Altitude and AP Basin and All shared Residual
altitude shared shared AP shared

Abundance 1.4 32.9 0 0 2.2 0.3 0 1.3 63.7

Presence–absence 1.9 24.3 0 0.2 1.9 0.6 0 1.4 72.2
Log mean length 12.3 12.1 1.7 0 0 1.2 15.0 0 75.9
Mean length 12.4 15.0 6.8 0 0 2.2 13.7 0 74.0
Maximum length 6.4 9.0 0 2.4 0.7 0.2 2.7 1.0 85.5
Kurtosis 0 66.9 3.5 2.7 0 0 0 2.0 65.4
Variance 5.8 0 1.8 0 0 2.9 6.6 0 100.0
Skewness 0 26.8 0 1.0 4.3 0 0 0 71.8
CV 3.9 33.1 0 3.3 4.8 0 0 0 64.7

CV, coefficient of variation. Significant values (P < 0.05) are in bold.

576 C.A. Murphy et al. / Ecological Indicators 34 (2013) 571–579

Table 2
Model comparison for selected response variables of Squalius laietanus using basin, altitude (alt), and anthropogenic perturbation (AP) as predictors. Models are generalized
linear models for presence–absence and abundance data and linear models for the rest of response variables (see Section 2). Degrees of freedom (df) vary because of constraints
in metric calculations and the inclusion of single observations and absences in the GLMs. Full models in bold are those that had significantly smaller residual sum-of-squares
(RSS) than the two simpler models according to likelihood ratio tests and were thus clearly supported.

Response variable Model RSS df R2 adj P value P value (comparison with model 3)

Log mean length

1 Alt × basin 6.04 46 0.29 *** 0.126
2 AP × basin 7.55 46 0.43 0.008 0.012
3 Alt × AP × basin 3.53 32 0.52 0.001

Mean length
1 Alt × basin 67,012 46 0.37 0.001 0.048
2 AP × basin 75,404 46 0.30 0.007 0.013
3 Alt × AP × basin 35,479 32 0.52 0.001

Maximum length
1 Alt × basin 331,339 46 0.24 0.022 0.679
2 AP × basin 394,971 46 0.09 0.200 0.223
3 Alt × AP × basin 246,787 32 0.18 0.150

Kurtosis
1 Alt × basin 270 40 0.61 *** 0.024
2 AP × basin 197 40 0.72 *** 0.313
3 Alt × AP × basin 124 27 0.74 ***

Skewness
1 Alt × basin 41 40 0.22 0.037 0.043
2 AP × basin 36 40 0.32 0.006 0.139
3 Alt × AP × basin 20 27 0.44 0.009

Variance
1 Alt × basin 5.13E+08 41 0.05 0.310 0.002
2 AP × basin 4.17E+08 41 0.23 0.033 0.015
3 Alt × AP × basin 1.77E+08 27 0.50 0.003

Coefficient of variation
1 Alt × basin 23,011 41 0.31 0.007 0.497
2 AP × basin 19,373 41 0.42 0.001 0.899
3 Alt × AP × basin 15,253 27 0.30 0.060

Abundance
1 Alt × basin 264 0.56 0.006
2 AP × basin 264 0.56 0.011
3 Alt × AP × basin 240 0.65

Presence–absence
1 Alt × basin 265 0.35 0.361
2 AP × basin 265 0.36 0.432
3 Alt × AP × basin 241 0.42

“***” Indicates P < 0.0005.

demonstrated two techniques that should be appropriate to situ-
ations with collinearity (variation partitioning and linear models
with interaction terms). Significant variables were similar with
the two tests, but the ANCOVA models explained more variation,
as the slopes of the relationships seemed to clearly vary among
basins.
4.2. Comparing metrics
Size metrics, with the exception of the coefficient of variation,
were significantly correlated (Table A1). This should often be the
case as the metrics are all describing attributes of the same popula-
tion size distributions. Of all the size metrics studied, we found
that mean length was the most robust. Log transforming mean
length did not noticeably improve either model. Mean length was
positively correlated with maximum length and variance and neg-
atively correlated with kurtosis and skewness. This would suggest
that those measures are also responding, but that we are unable
to capture those responses either because they are less consistent
or less pronounced. Average S. laietanus length generally increased
along the gradient of anthropogenic perturbation and was the most
responsive size metric considering both tests. Since all of the size
metrics are reflective of population size structure, we might expect
that changes in mean size are related to directional gains or losses
due to growth or survivorship which should be reflected in skew-
ness and other metrics. However, neither test indicted a significant
relationship between anthropogenic perturbation and skewness.
Variance did increase slightly with increasing anthropogenic per-
turbation, indicating higher dissimilarity in observed sizes within
a population under disturbance conditions, but this increase was
much less pronounced than the increase in average size. Maximum
size had the lowest percentages of explained variation (particu-
larly with the ANCOVA models) and was not clearly related to
anthropogenic perturbation; this pattern is not unexpected since
maximum size is unlikely to change outside of mass mortality
events or severely altered growth rates. For example, a loss of all
but one large individual may not result in significant changes to
maximum size. Kurtosis may be responsive, but appears unreliable.
While VP indicated kurtosis was the only variable with significant
effects from pure anthropogenic perturbation, the combined effects
were negligible. Kurtosis also had the highest adjusted R2 in the
linear model results. This value appears to be due largely to the
strong basin effects, as indicated by both tests. Basin effects appear
to be influenced by extreme values where sample sizes were small
and kurtosis presented varied positive and negative relationships
(Fig. S6). Thus, although initially promising, kurtosis appears to be
a less robust and interpretable size metric than mean length for S.
laietanus.
 C.A. Murphy et al. / Ecological Indicators 34 (2013) 571–579
Abundance and presence–absence are non-size-based metrics
commonly used in freshwater bioassessment. While abundance
had high levels of explained variation in the linear model results, its
response in VP was much less pronounced than that of mean length
and, in both tests, was much less predictable as a response to AP
(Fig. A5). This emphasis of response is consistent with Waller (1985)
who found changes in size structure as a result of non-significant
changes in density. Minor changes in density which disproportion-
ately affect certain cohorts or size classes or density-dependent
processes (e.g., higher survival of new cohorts after perturbations)
would explain our observations, including the differential sensi-
tivity of size metrics. It is important to reinforce that the different
metrics behave differently even though many of those metrics are
related, and that their ability to detect a change in size structure
will ultimately depend on the type and severity of that change.
Our results are limited to assessing size structure changes which
are related to our explanatory variables. If our disturbance metric
or our environmental variables failed to capture a driver of size
structure changes, these changes might act as noise, reducing our
statistical power and the possibility of noting significant response
measures. Thus, changes in size metrics resulting from stochas-
ticity, genetic or otherwise, could mask size structure responses.
This could be another limitation in the application of size structure
metrics for bioassessments, especially with respect to less abun-
dant species. As S. laietanus remain relatively abundant in their
currently occupied range, we might expect fewer stochastic size
effects than for a less common species. In addition to stochastic
processes, historic disturbances not currently detectable, and not
included in AP, which cause changes in the body size distributions
of modern populations would not result in detectable size structure
effects but might hinder the detection of size structure responses
to AP.
4.3. Average length changes with anthropogenic perturbation
Prior to our study, decreasing average sizes have been reported
in various taxa as the response to anthropogenic perturbations,
including those which are not related to take impacts (Dodson and
Hanazato, 1995; Jung and Jagoe, 1995; Walters and Post, 2008).
Documented decreases in average sizes of S. trutta corresponding
to increased perturbation in a Catalan stream closed to take angling
provide an example from our study region (L. Benejam, unpublished
data). Contrary to this foundation, our data provide evidence for
increasing average size in a fish population under increasingly dis-
turbed conditions. This may reflect failure to recruit in disturbed
conditions (e.g. due to siltation) or growth at reduced densities, as
increases in length appear to be related to a trend of decreasing
abundance (Fig. A5). Most importantly, we have confirmed that
size structure responses may vary in freshwater systems and are
not always negative under disturbance regimes.
We assessed each of the anthropogenic perturbation vari-
ables individually for their influence on mean length and found
that the official Water Agency summary statistic, the risk of
non-compliance, was the most descriptive (see Appendix A in Sup-
porting Information). This metric reflects chronic, landscape scale
impacts (such as upstream water abstractions) not captured by the
other metrics used, in addition to local water quality variables.
This is a positive valuation for the metric, supporting its use and
biological relevance.
Different species may have different responses to AP. Cor-
responding analyses on two species of Barbus, B. haasi and
B. meridionalis (Appendix B), indicated very little comparative
response in size structure metrics to disturbance. For both species,
VP revealed small amounts of variation only in presence–absence
related to anthropogenic perturbation and joint effects. Linear
model comparisons only supported full models for abundance and
577
kurtosis of B. haasi and B. meridionalis, respectively. Although these
species were the two most common native stream fishes after S. lai-
etanus (excluding brown trout, whose abundance and distribution
is considerably affected by fishing activities) and are considered
sensitive to disturbance (Aparicio et al., 2000), they are native and
present in fewer river basins (Aparicio et al., 2000; Maceda-Veiga
et al., 2010), resulting in smaller sample sizes and statistical power
to detect significant relationships. Alternatively, S. laietanus has
been shown to be declining more strongly than the two barbel
species (Aparicio et al., 2000; Maceda-Veiga et al., 2010), so it might
be more responsive to AP (see also below).
A practical application of size structure analysis outside of direct
use in bioassessments is its inclusion as an informative metric
regarding tolerance, or rather the impact of anthropogenic pertur-
bation on populations. Size structure has the potential to inform us
whether disturbance is affecting populations. For example, when
combined with data regarding changes in range, we might then
evaluate whether those changes are likely positive or negative
impacts for the population. This is critical, as a decrease in mean
size could result from the loss of large, highly fecund individuals or
from increasing recruitment success, as younger cohorts begin to
dominate. The two have very divergent implications for the popu-
lation, but their size metrics might appear similar. Thus, it seems
prudent to combine size structure observations with additional
information, such as changes in range or density, to provide a more
grounded basis for interpretation.
Recent papers have suggested that S. laietanus and Squalius
cephalus (Linnaeus, 1758), formerly considered a single species,
are relatively tolerant (Aparicio et al., 2000; Oberdorff et al., 2002;
Segurado et al., 2011; Maceda-Veiga and de Sostoa, 2011). This
assumption has been previously questioned by Carol et al. (2006)
and Blanck et al. (2007), who considered S. cephalus in a high
requirement group. Based on the response we found with respect
to population size structure, along with documented declines in
range and abundance from historic values (Aparicio et al., 2000;
Maceda-Veiga et al., 2010), S. laietanus appear responsive to AP and
we suggest this species is sensitive, rather than moderately tolerant
to tolerant. While an increased mean size might suggest ‘positive’
effects, such as increased growth rate, it seems highly unlikely that
the underlying changes are beneficial, as S. laietanus has been lost
from much of its historic range. Under circumstances of dramatic
range reductions, significant population size metric responses to
disturbance should be a red flag. Thus we suggest that these metrics
might be applied in similar situations where tolerance to anthro-
pogenic perturbation may be unclear.
4.4. Conclusions
Comparing multiple responses, including population size struc-
ture, should take precedence over simple presence–absence or
abundance studies in the Mediterranean and other systems depau-
perate in species as assessments based on limited information
may be missing important factors or requirements for the species.
Documented changes at population levels can demonstrate trends
and sensitivities which might otherwise be overlooked. However,
the unexpected positive increase of mean size of S. laietanus with
anthropogenic perturbation, the strong effects of river basin, the
collinearity with spatial gradients and the species-specific response
explain the paradox of poor use of size structure in freshwater
bioassessments and may preclude the immediate application of
size structure information. We suggest that this should not dis-
count the importance and value of size structure information and its
potential to provide informative metrics. While the direct applica-
tion of size structure in freshwater bioassessments may be difficult,
species- and basin-specific applications might be especially help-
ful in monitoring and managing endangered species, particularly in
578 C.A. Murphy et al. / Ecological Indicators 34 (2013) 571–579
species-poor systems. In systems such as Mediterranean streams
or headwater reaches where only one or a few species is often
present, such metrics may be the only alternative. As we learn more
about size structure variability and responses, we may be able to
incorporate this kind of descriptor into bioassessments for regions
where the performance of other metrics are unsatisfactory and the
additional time and expense of size measurements are justified.
Acknowledgements
We gratefully thank the numerous people who assisted in the
field, Lluís Benejam for help with the database, and Antoni Munné
for comments. This study was funded by the Catalan Water Agency
(project Ibicat2). Additional financial support was provided by the
Spanish Ministry of Science (projects CGL2009-12877-C02-01 and
Consolider-Ingenio 2010 CSD2009-00065).
Appendix A. Supplementary data
Supplementary material related to this article can be
found, in the online version, at http://dx.doi.org/10.1016/
j.ecolind.2013.06.007.
References
ACA, 2005. Caracterització de masses d’aigua i anàlisi del risc d’incompliment dels
objectius de la directiva marc de l’aigua (2000/60/CE) a Catalunya (conques
intra i intercomunitàries) en compliment als articles 5, 6 i 7 de la directiva,
http://aca-web.gencat.cat/aca/appmanager/aca/aca? nfpb=true& pageLabel=
P1206154461208200586461 (accessed 30.05.13).
Aparicio, E., Vargas, M.J., Olmo, J.M., de Sostoa, A., 2000. Decline of native freshwater
fishes in a Mediterranean watershed on the Iberian Peninsula: a quantitative
assessment. Environ. Biol. Fishes 59, 11–19.
Barbour, M.T., Gerritsen, J., Snyder, B.D., Stribling, J.B., 1999. Rapid Bioassessment
Protocols for Use in Streams and Wadeable Rivers: Periphyton, Benthic Macroin-
vertebrates and Fish, 2nd ed. U.S. Environmental Protection Agency; Office of
Water, Washington, DC, EPA 841-B-99-002.
Basset, A., Barbone, E., Borja, A., Brucet, S., Pinna, M., Quintana, X.D., Reizopoulou,
S., Rosati, I., Simboura, N., 2012. A benthic macroinvertebrate size spectra index
for implementing the water framework directive in coastal lagoons in Mediter-
ranean and Black Sea ecoregions. Ecol. Indic. 12, 72–83.
Bendel, R.B., Higgins, S.S., Teberg, J.E., Pyke, D.A., 1989. Comparison of skewness
coefficient, coefficient of variation, and Gini coefficient as inequality measures
within populations. Oecologia 78, 394–400.
Benejam, L., Aparicio, E., Vargas, M., Vila-Gispert, A., García-Berthou, E., 2008.
Assessing fish metrics and biotic indices in a Mediterranean stream: effects of
uncertain native status of fish. Hydrobiologia 603, 197–210.
Benejam, L., Angermeier, P.L., Munné, A., García-Berthou, E., 2010. Assessing effects
of water abstraction on fish assemblages in Mediterranean streams. Freshwater
Biol. 55, 628–642.
Benejam, L., Alcaraz, C., Benito, J., Caiola, N., Casals, F., Maceda-Veiga, A., de Sostoa, A.,
García-Berthou, E., 2012. Fish catchability and comparison of four electrofishing
crews in Mediterranean streams. Fish. Res. 123, 9–15.
Benjamin, L.R., Hardwick, R.C., 1986. Sources of variation and measures of variability
in even-aged stands of plants. Ann. Bot. 58, 757.
Blanck, A., Tedesco, P.A., Lamouroux, N., 2007. Relationships between life-history
strategies of European freshwater fish species and their habitat preferences.
Freshwater Biol. 52, 843–859.
Bonar, S.A., 2002. Relative length frequency: a simple, visual technique to evaluate
size structure in fish populations. N. Am. J. Fish. Manage. 22, 1086–1094.
Borcard, D., Legendre, P., Drapeau, P., 1992. Partialling out the spatial component of
ecological variation. Ecology 73, 1045–1055.
Brown, J., Allen, A., Gillooly, J., 2007. The metabolic theory of ecology and the role
of body size in marine and freshwater ecosystems. In: Hildrew, A., Raffaelli, D.,
Edmonds-Brown, R. (Eds.), Body Size: The Structure and Function of Aquatic
Ecosystems. Cambridge University Press, Cambridge, pp. 1–15.
Carmona-Catot, G., Benito, J., García-Berthou, E., 2011. Comparing latitudinal and
upstream-downstream gradients: life-history traits of invasive mosquitofish.
Divers. Distrib. 17, 214–224.
Carol, J., Benejam, L., Alcaraz, C., Vila-Gispert, A., Zamora, L., Navarro, E., Armengol, J.,
García-Berthou, E., 2006. The effects of limnological features on fish assemblages
of 14 Spanish reservoirs. Ecol. Freshwat. Fish 15, 66–77.
Chown, S.L., Klok, J., 2003. Altitudinal body size clines: latitudinal effects associated
with changing seasonality. Ecography 26, 445–455.
Corbacho, C., Sánchez, J.M., Costillo, E., 2003. Patterns of structural complexity
and human disturbance of riparian vegetation in agricultural landscapes of a
Mediterranean area. Agric. Ecosyst. Environ. 95, 495–507.
Coulthard, T.J., Lewin, J., Macklin, M.G., 2005. Modelling differential catchment
response to environmental change. Geomorphology 69, 222–241.
de Roos, A., Persson, L., McCauley, E., 2003. The influence of size-dependent life-
history traits on the structure and dynamics of populations and communities.
Ecol. Lett. 6, 473–487.
Deal, R., Heithecker, T., Zenner, E., 2010. Comparison of tree size structure and
growth for partially harvested and even-aged hemlock–spruce stands in south-
east Alaska. J. Forest Res. 15, 31–37.
DeCarlo, L.T., 1997. On the meaning and use of kurtosis. Psychol. Meth. 2,
292–307.
Dodson, S.I., Hanazato, T., 1995. Commentary on effects of anthropogenic and natural
organic chemicals on development, swimming behavior, and reproduction of
Daphnia, a key member of aquatic ecosystems. Environ. Health Perspect. 103,
7–11.
Ferreira, T., Caiola, N., Casals, F., Oliveira, J.M., de Sostoa, A., 2007. Assessing pertur-
bation of river fish communities in the Iberian ecoregion. Fisheries Manag. Ecol.
14, 519–530.
Fox, J., Weisberg, S., 2011. An R Companion to Applied Regression, 2nd ed. SAGE
Publications, Thousand Oaks, CA.
Freckleton, R., 2011. Dealing with collinearity in behavioural and ecological data:
model averaging and the problems of measurement error. Behav. Ecol. Sociobiol.
65, 91–101.
Gill, J.A., Norris, K., Sutherland, W.J., 2001. Why behavioural responses may not
reflect the population consequences of human disturbance. Biol. Conserv. 97,
265–268.
Graham, M.H., 2003. Confronting multicollinearity in ecological multiple regression.
Ecology 84, 2809–2815.
Guttal, V., Jayaprakash, C., 2008. Changing skewness: an early warning signal of
regime shifts in ecosystems. Ecol. Lett. 11, 450–460.
Haedrich, R.L., Barnes, S.M., 1997. Changes over time of the size structure in an
exploited shelf fish community. Fish. Res. 31, 229–239.
Hara, T., 1988. Dynamics of size structure in plant populations. Trends Ecol. Evol. 3,
129–133.
Joanes, D.N., Gill, C.A., 1998. Comparing measures of sample skewness and kurtosis.
Statistician 47, 183–189.
Jung, R.E., Jagoe, C.H., 1995. Effects of low pH and aluminum on body size, swimming
performance, and susceptibility to predation of green tree frog (Hyla cinerea)
tadpoles. Can. J. Zool. 73, 2171–2183.
Karr, J.R., Fausch, K.D., Angermeier, P.L., Yant, P.R., Schlosser, I.J., 1986a. Assessing
biological integrity in running waters. A method and its rationale. Illinois Natural
History Survey. Special Publication, Champaign, pp. 5.
Karr, J.R., Chu, E.W., 1999. Restoring life in running waters: better biological moni-
toring. Island Press, Washington, DC.
Karr, J.R., Fausch, K.D., Angermeier, P.L., Yant, P.R., Schlosser, I.J., 1986b. Assessment
of Biological Integrity in Running Waters: A Method and its Rationale. Illinois
Natural History Survey Special Publication 5, Champaign, IL.
Knox, R.G., Peet, R.K., Christensen, N.L., 1989. Population dynamics in Loblolly Pine
stands: changes in skewness and size inequality. Ecology 70, 1153–1166.
Körner, C., 2007. The use of ‘altitude’ in ecological research. Trends Ecol. Evol. 22,
569–574.
Maceda-Veiga, A., Monleon-Getino, A., Caiola, N., Casals, F., de Sostoa, A., 2010.
Changes in fish assemblages in catchments in north-eastern Spain: biodiversity,
conservation status and introduced species. Freshwater Biol. 55, 1734–1746.
Maceda-Veiga, A., de Sostoa, A., 2011. Observational evidence of the sensitivity of
some fish species to environmental stressors in Mediterranean rivers. Ecol. Indic.
11, 311–317.
Monti, D., Legendre, P., 2009. Shifts between biotic and physical driving forces of
species organization under natural disturbance regimes. Can. J. Fish. Aquat. Sci.
66, 1282–1293.
Moyle, P.B., Randall, P.J., 1998. Evaluating the biotic integrity of watersheds in the
Sierra Nevada, California. Conserv. Biol. 12, 1318–1326.
Murray, K., Conner, M.M., 2009. Methods to quantify variable importance: implica-
tions for the analysis of noisy ecological data. Ecology 90, 348–355.
Oberdorff, T., Pont, D., Hugueny, B., Porcher, J.P., 2002. Development and validation
of a fish-based index for the assessment of river health in France. Freshwater
Biol. 47, 1720–1734.
Oksanen, J., Blanchet, F.G., Kindt, R., Legendre, P., O’Hara, R.B., Simpson, G.L., Solymos,
P., Henry, M., Stevens, H., Wagner, H., 2010. Vegan: Community Ecology Package.
R package version 1.17-4. http://CRAN.R-project.org/package=vegan
Olea, P.P., Mateo-Tomás, P., de Frutos, Á., 2010. Estimating and modelling bias of the
hierarchical partitioning public-domain software: implications in environmen-
tal management and conservation. PLoS ONE 5, e11698.
Persson, L., de Roos, A., 2007. Interplay between individual growth and population
feedbacks shapes body-size distributions. In: Hildrew, A., Raffaelli, D., Edmonds-
Brown, R. (Eds.), Body Size: The Structure and Function of Aquatic Ecosystems.
Cambridge University Press, Cambridge, pp. 225–244.
Pont, D., Hugueny, B., Beier, U., Goffaux, D., Melcher, A., Noble, R., Rogers, C., Roset,
N., Schmutz, S., 2006. Assessing river biotic condition at a continental scale: a
European approach using functional metrics and fish assemblages. J. Appl. Ecol.
43, 70–80.
R Development Core Team, 2011. R: A language and environment for statisti-
cal computing. In: R Foundation for Statistical Computing, Vienna, Austria
http://www.R-project.org/
Revelle, W., 2011. Psych: Procedures for Psychological, Psychometric, and
Personality Research. Northwestern University, Evanston, IL, http://personality-
project.org/r/psych.manual.pdf (accessed 01.03.12).
 C.A. Murphy et al. / Ecological Indicators 34 (2013) 571–579
Richards, C., Johnson, L.B., Host, G.E., 1996. Landscape-scale influences on stream
habitats and biota. Can. J. Fish. Aquat. Sci. 53, 295–311.
Richmond, R.H., 1993. Coral reefs: present problems and future concerns resulting
from anthropogenic disturbance. Am. Zool. 33, 524–536.
Rochet, M.J., Trenkel, V.M., Carpentier, A., Coppin, F., De Sola, L.G., Léauté, J.P., Mahé,
J.C., Maiorano, P., Mannini, A., Murenu, M., Piet, G., Politou, C.Y., Reale, B., Spedi-
cato, M.T., Tserpes, G., Bertrand, J.A., 2010. Do changes in environmental and
fishing pressures impact marine communities? An empirical assessment. J. Appl.
Ecol. 47, 741–750.
Savage, V.M., Gillooly, J.F., Brown, J.H., West, G.B., Charnov, E.L., 2004. Effects
of body size and temperature on population growth. Am. Nat. 163,
429–441.
Segurado, P., Santos, J.M., Pont, D., Melcher, A.H., Jalon, D.G., Hughes,
R.M., Ferreira, M.T., 2011. Estimating species tolerance to human per-
turbation: expert judgment versus empirical approaches. Ecol. Indic. 11,
1623–1635.
Simon, T.P., 1999. Assessing the Sustainability and Biological Integrity of Water
Resources Using Fish Communities. CRC Press, Boca Raton, FL.
Tarr, J.Q.R., 2000. The South African abalone (Haliotis midae) fishery: a decade of
challenges and change. In: Campbell, A. (Ed.), Workshop on Rebuilding Abalone
579
Stocks in British Columbia. February 23–26, 1999, Nanaimo, BC. National
Research Council of Canada, Ottowa, pp. 32–40.
Tedesco, P.A., Sagnes, P., Laroche, J., 2009. Variability in the growth rate of chub
Leuciscus cephalus along a longitudinal river gradient. J. Fish Biol. 74, 312–319.
Waller, D.M., 1985. The genesis of size hierarchies in seedling populations of Impa-
tiens capensis Meerb. New Phytol. 100, 243–260.
Walsh, C., Mac Nally, R., 2011. Package ‘hier.part’, http://cran.r-
project.org/web/packages/hier.part/hier.part.pdf (accessed 01.03.12).
Walters, A.W., Post, D.M., 2008. An experimental disturbance alters fish size struc-
ture but not food chain length in streams. Ecology 89, 3261–3267.
Weiner, J., 1985. Size hierarchies in experimental populations of annual plants.
Ecology 66, 743–752.
Werner, E.E., Gilliam, J.F., 1984. The ontogenetic niche and species interactions in
size-structured populations. Annu. Rev. Ecol. Syst. 15, 393–425.
Williams, L.R., Taylor, C.M., Warren Jr., M.L., Clingenpeel, J.A., 2003. Environmen-
tal variability, historical contingency, and the structure of regional fish and
macroinvertebrate faunas in Ouachita mountain stream systems. Environ. Biol.
Fish 67, 203–216.
1 APPENDIX A. Additional details on the methods for developing the anthropogenic

2 perturbation measure (AP) and different perturbation types.

4 Development of the anthropogenic perturbation measure (AP) and relationship with spatial

5 gradients

7 The principal component analysis (PCA) used to summarize anthropogenic perturbation

8 included the sum of the RBI scores, the sum of the visual impacts, the log-transformed

9 summary of anthropogenic perturbation from the Catalan Water Agency (risk of non-

10 compliance), log-transformed concentration of dissolved oxygen (mg/L), log-transformed

11 concentration of ammonia (mg/L), and pH. The first axis (PCA 1) had an eigenvalue of 2.232

12 while the subsequent axes were 1.140 and lower. The first axis explained 37.2% of the

13 variation and was mostly related to total RBI score, visual impacts, and risk of non-

14 compliance, but also ammonia and oxygen concentration (Fig. A1). Positive scores mostly

15 indicated sites with good habitat quality (RBI scores), few visual impacts and few overall

16 pressures (as represented by the risk of non-compliance) but also higher oxygen

17 concentration and low values of ammonia. The second axis was mostly related to pH and is

18 likely less related to perturbation and more to natural variation in geological substrate

19 (calcareous versus acidic regions) and water alkalinity. We retained the first axis (changing

20 the sign, so higher values indicate increasing anthropogenic pressure) for analyses and refer

21 to it as “anthropogenic perturbation” (AP) throughout the paper. A map of sampling sites

22 with corresponding anthropogenic perturbation is provided in Fig. A2.

23 To account for natural longitudinal effects (e.g. increased productivity, resource

24 availability), we initially used another principal component analysis (PCA) to summarize

25 variables not affected by anthropogenic perturbation: stream order, distance to the source,

1
26 distance to the river mouth, total catchment area, and altitude. However, the variables fit

27 clearly into two distinct axes (Fig. A3). The first axis was formed by the stream order, the

28 distance from the source and the log-transformed cumulative area, whereas the second was

29 related to distance to the sea, altitude (log-transformed), latitude and longitude. The variables

30 in the first axis correspond to the size of the stream, whereas those in the second axis, such as

31 altitude, are more closely related to geographic position. Moreover, anthropogenic

32 perturbation (AP) was more related to altitude (Pearson’s r = 0.54, df = 299, p < 0.0005) (Fig.

33 A2) than to this PCA 1 scores (Pearson’s r = 0.17, df = 299, p = 0.004) (Fig. A2).

34 Generalized variance inflation factors confirmed collinearity between the three predictors

35 (basin, anthropogenic perturbation and altitude, GVIF values 2.1, 1.5 and 2.6, respectively

36 from the linear model of “mean chub length ~ basin + AP + log altitude”). While these values

37 indicate only low to moderate collinearity, they are certainly an underestimation since they

38 can only be computed for a model assuming equal slopes among basins (which is not

39 observed, see Results). This collinearity is not unexpected as population density and the

40 intensity of human activities (and thus AP) increases in the lower elevation coastal regions in

41 Catalonia (Fig. A2) and likely elsewhere.

42

43 Effects of seasonality

44 To account for possible effects of seasonality, collection dates were grouped by season and

45 year. Elevation of samples and AP varied with season because some sites are only wadable

46 in low flow conditions (i.e. electrofishing was only possible in summer) and because AP

47 comprises chemical pressures (e.g. dissolved oxygen and ammonia) that vary seasonally in

48 addition to spatially. However, we found no significant improvement in the models of mean

2
49 size with the inclusion of a ‘season’ term, so we are confident that our results are not due to

50 seasonality.

51

52 Effects of different perturbation types

53 To evaluate the role of different impact types on size structure, we compared the models

54 using our overall perturbation measure (AP), with models with only the risk of non-

55 compliance (a general summary of anthropogenic perturbation used by the Catalan Water

56 Agency), the sum of the visual impacts (a local measure of habitat impacts), or the log-

57 transformed concentration of ammonia (a purely chemical pressure). Although different

58 impact types increase along the longitudinal gradient and explained the variation in size

59 structure, the models with only habitat or chemical impacts explained less variation than the

60 models using the overall impact measure (AP) (Table A2). The risk of non-compliance was

61 similar to AP at explaining the observed changes in mean length. The most notable difference

62 between the risk of non-compliance measure and the other impact types is that it reflects

63 perturbations (such as water abstraction due to reservoirs) that occur above the sampling site

64 and which are not restricted to the sampling date. Therefore, changes in mean length appear

65 slightly more related to chronic pressures at the landscape scale rather than to more local,

66 point impact disturbances.

67

3
68 Table A1 Correlation matrix of Squalius laietanus size-structure metrics. “***” indicates P <

69 0.0005

Mean length CV Max length Kurtosis Skewness Variance

Mean length — 0.11 *** *** 0.01 ***
CV -0.21 — *** *** *** ***
Max length 0.59 0.48 — 0.01 *** ***
Kurtosis -0.29 0.56 0.32 — *** 0.68
Skewness -0.33 0.60 0.38 0.79 — 0.93
Variance 0.56 0.41 0.60 -0.06 -0.01 —
70

4
71 Table A2 Model comparison for mean length of Squalius laietanus using basin, altitude (alt),

72 and different anthropogenic perturbation measures as predictors. See Table 2 and Appendix

73 A for further details.

74
Response variable Model RSS df R2adj P value P value
(comparison with
model 3)
Mean length
1 alt × basin 67012 46 0.37 0.001 0.05
2 AP × basin 75404 46 0.30 0.007 0.01
3 alt × AP × basin 35479 32 0.52 0.001

4 alt × risk of non-compliance × basin 34333 32 0.54 0.0005 -

5 alt × NH4 × basin 51033 37 0.41 0.003 0.03
6 alt × SUM_IMP × basin 46740 36 0.44 0.002 0.06
75

5
76

pH

2.0

1.5
DO

1.0

PCA2
0.5

Visual Impacts
0.0
Risk of non-compliance

-0.5

-1.0 Total RBI score

NH4

-2.0 - 0.0 1.0 2.0

PCA 1

77

78 Fig. A1 Principal component analysis of anthropogenic perturbation indicators with sites

79 (black dots) and measured variables (red vectors).

6
80

81

82 Fig. A2 Maps of Catalonia, Spain with sampling sites (red squares) overlaid by altitude (blue

83 circles, left) and anthropogenic perturbation, AP (orange circles, right). Circle diameter

84 corresponds to altitude (left) and AP (right), where larger circles are equivalent to larger

85 values.

7
 Variables factor map (PCA)

1.0

Log(Altitude) Distance to sea

Longitude
0.5
PCA 2 (35.59%)

Log(accumulated area)
Stream Order
0.0
Distance from source

-0.5
Latitude

-1.0

-1.0 -0.5 0.0 0.5 1.0

PCA 1 (55.97%)

86

87 Fig. A3 Principal component analysis of spatial descriptors (black vectors) of sampling sites.

88 Supplementary variables (blue dotted vectors) not included in the analysis are also plotted.

89

8
90

91

92 Fig. A4 Relationship of mean length and abundance of Squalius laietanus by river basin

93 (with fitted linear regression functions).

9
94

95 Fig. A5 Scatterplots with fitted linear regressions by river basin for relationships between S.

96 laietanus abundance (corrected for area fished) and explanatory variables altitude (left) and

97 anthropogenic perturbation (right).

10
98

99 Fig. A6 Scatterplots with fitted linear regressions by river basin for relationships between

100 kurtosis and explanatory variables (anthropogenic perturbation and altitude).

11
 1 APPENDIX B. Additional results for Barbus haasi and B. meridionalis.

3 Results for additional ‘common’ Catalonian fish species.

5 The methodology pertaining to these results and pertinent discussion points are contained in the

6 main document text. These analyses are included to underscore the species-specific responses

7 observed and the lack of clear patterns across species. Barbus haasi and B. meridionalis were the

8 third and fourth most commonly encountered native stream fishes, found at 53 and 34 sites

9 respectively. Only the native stream fishes Squalius laietanus and Salmo trutta occurred more

10 frequently. We do not include S. trutta due to the evidence of strong size-structure changes with

11 angling for brown trout as found by Benejam et al. (unpublished).

1
12

13 Table B1 Model comparison for selected response variables of Barbus haasi using basin,

14 altitude (alt), and anthropogenic perturbation (AP) as predictors. Models are generalized linear

15 models for presence-absence and abundance data and linear models for the rest of response

16 variables (see Methods in main document). Degrees of freedom (df) vary because of constraints

17 in metric calculations and the inclusion of single observations and absences in the GLMs. “***”

18 indicates P < 0.0005

19 Response variable Model RSS df R2adj P value P value

20 (comparison with model 3)
21 Mean length
22 1 alt × basin 35365 42 0.00 0.759 0.230
23 2 PC1 × basin 35354 42 0.00 0.758 0.231
24 3 alt × PC1 × basin 28557 36 0.00 0.497
25 Maximum length
26 1 alt × basin 99953 42 0.01 0.399 0.394
27 2 PC1 × basin 98256 42 0.03 0.334 0.468
28 3 alt × PC1 × basin 84737 36 0.02 0.403
29 Kurtosis
30 1 alt × basin 211 40 0.15 0.068 0.132
31 2 PC1 × basin 269 40 0.00 0.803 0.005
32 3 alt × PC1 × basin 160 34 0.24 0.042
33 Skewness
34 1 alt × basin 31 40 0.00 0.580 0.173
35 2 PC1 × basin 33 40 0.00 0.890 0.069
36 3 alt × PC1 × basin 24 34 0.06 0.318
37 Variance
38 1 alt × basin 3.4·107 40 0.01 0.415 0.151
39 2 PC1 × basin 3.3·107 40 0.06 0.242 0.274
40 3 alt × PC1 × basin 2.6·107 34 0.10 0.213
41 Abundance
42 1 alt × basin 261 0.28 0.025
43 2 PC1 × basin 261 0.30 0.009
44 3 alt × PC1 × basin 239 0.44
45 Presence-absence
46 1 alt × basin 265 0.34 0.212
47 2 PC1 × basin 265 0.33 0.132
48 3 alt × PC1 × basin 241 0.44
49

2
50

51 Table B2 Results of variation partitioning (% variation explained) for selected response

52 variables of Barbus haasi using basin, altitude, and anthropogenic perturbation (AP) as

53 predictors. Only pure and total (“pure” + “shared”) effects are testable (not the “shared” or

54 “joint” effects alone). Significant values (P < 0.05) are in bold

Response AP Pure Pure Pure Basin and Altitude and Basin and All Residual
variable basin altitude AP altitude AP shared AP shared shared
shared

Abundance 0 7.6 0.1 0.1 0 0 0 0.4 92.9

Presence- 1.7 9.5 0 0 1.2 0.4 0.1 1.2 87.8
absence
Mean length 0 0 0 0 0 0 0 0.5 100
Maximum 2.2 0 0 3.2 0 0 0 1.0 100
length
Kurtosis 0 0 0 0 0.3 0.4 0 0 100
Variance 0.1 0.4 0 0 0.8 0.4 0.5 1.1 100
Skewness 3.4 0 0 0 2.8 0.7 <0.1 0 100
55
56

3
57

58 Table B3 Model comparison for selected response variables of Barbus meridionalis using basin,

59 altitude (alt), and anthropogenic perturbation (AP) as predictors. Models are generalized linear

60 models for presence-absence and abundance data and linear models for the rest of response

61 variables (see Methods in main document). Degrees of freedom (df) vary because of constraints

62 in metric calculations and the inclusion of single observations and absences in the GLMs. “***”

63 indicates P < 0.0005

64 Response variable Model RSS df R2adj P value P value

65 (comparison with model 3)
66 Mean length
67 1 alt × basin 28451 55 0.19 0.022 0.221
68 2 AP × basin 30594 55 0.13 0.077 0.088
69 3 alt × AP × basin 21190 44 0.25 0.031
70 Maximum length
71 1 alt × basin 80245 55 0.18 0.004 0.885
72 2 AP × basin 91767 55 0.16 0.043 0.342
73 3 alt × AP × basin 71145 44 0.26 0.078
74 Kurtosis
75 1 alt × basin 212 54 0.59 *** 0.320
76 2 AP × basin 187 54 0.63 *** 0.834
77 3 alt × AP × basin 163 43 0.60 ***
78 Skewness
79 1 alt × basin 31 54 0.25 0.006 0.338
80 2 AP × basin 28 54 0.32 0.001 0.263
81 3 alt × AP × basin 23 43 0.30 0.012
82 Variance
83 1 alt × basin 2.7·107 54 0.10 0.130 0.338
84 2 AP × basin 2.7·107 54 0.08 0.181 0.263
85 3 alt × AP × basin 2.0·107 43 0.13 0.162
86 Abundance
87 1 alt × basin 88 0.34 0.294
88 2 AP × basin 88 0.31 0.122
89 3 alt × AP × basin 75 0.45
90 Presence-absence
91 1 alt × basin 265 0.73 0.160
92 2 AP × basin 265 0.69 0.014
93 3 alt × AP × basin 241 0.82
94
95

4
 96

97 Table B4 Results of variation partitioning (% variation explained) for selected response

98 variables of Barbus meridionalis using basin, altitude, and anthropogenic perturbation (AP) as

99 predictors. Only pure and total (“pure” + “shared”) effects are testable (not the “shared” or

100 “joint” effects alone). Significant values (P < 0.05) are in bold

Response AP Pure Pure Pure Basin and Altitude and Basin and All Residual
variable basin altitude AP altitude AP shared AP shared shared
shared

Abundance 3.1 0.5 0 0 0 1.0 0.3 0 95.8

Presence- 1.8 58.5 1.6 <0.1 0 0 0.5 1.3 39.8
absence
Mean length 0 27.0 6.3 0.2 13.0 0 0 0.4 79.4
Maximum 0 15.2 5.9 0 0.2 2.6 0 0 76.5
length
Kurtosis 0 25.6 0 0 0 0 0 0.4 76.4
Variance 0 4.8 2.5 0 4.3 0 2.0 0 88.4
Skewness 0 11.1 0 0 3.0 0 0 0 88.3
101