Hydrobiologia

DOI 10.1007/s10750-014-2079-y

PRIMARY RESEARCH PAPER

Impact of climate change on the structure of fish

assemblages in European rivers
Florian Pletterbauer • Andreas H. Melcher •

Teresa Ferreira • Stefan Schmutz

Received: 23 September 2013 / Revised: 7 October 2014 / Accepted: 11 October 2014

Ó Springer International Publishing Switzerland 2014

Abstract Fish assemblage structures show non- catchment, wetted width, elevation, maximum tem-
random patterns along the longitudinal gradient of perature of warmest month, mean temperature of
rivers. We analysed the impact of climate change on warmest quarter of the year in the upstream catchment,
riverine fish assemblages using the Fish Zone Index— and temperature range) and showed highly satisfactory
a structural index reflecting these conditions. The performance (adjR2 [ 0.6). The mean increase of the
dataset contained 92 fish species at 559 sampling sites Fish Zone Index was between 0.25 and 0.36 for the
spread over 14 European countries. We regressed the 2050s and between 0.36 and 0.41 in the 2080s.
Fish Zone Index in a hierarchical modelling frame- Maximum values reached levels of 0.92 and 1.18,
work with independent variables describing river respectively, for the two time periods. Major changes
characteristics and climate conditions. Future changes of fish assemblages were found in mediterranean as
were predicted according to three future emission well as in small rivers highlighting the need of timely
scenarios (A1b, A2, and B1) and two time periods conservation management.
(2050s, and 2080s). The final model contained seven
independent variables (river slope, size of upstream Keywords Fish zonation  FiZI  Global change 
Temperature  Precipitation  Streams  EU

Handling editor: Odd Terje Sandlund

F. Pletterbauer (&)  A. H. Melcher  S. Schmutz Introduction

Institute of Hydrobiology and Aquatic Ecosystem
Management, University of Natural Resources and Life
The structure of riverine fish assemblages is a result of
Sciences Vienna, Max Emanuel Strasse 17, 1180 Vienna,
Austria processes which act on different spatial and temporal
e-mail: florian.pletterbauer@boku.ac.at scales (Jackson et al., 2001), i.e. physiological effects
A. H. Melcher (e.g. thermal limits), environmental characteristics
e-mail: andreas.melcher@boku.ac.at (e.g. climate, river morphology, hydrology), and biotic
S. Schmutz interactions (e.g. predation; Pont et al., 2005). The
e-mail: stefan.schmutz@boku.ac.at interaction of these processes shape the fish assem-
blages at the local scale according to the species in the
T. Ferreira
regional species pool, which in turn is determined by
Forest Research Center, Technical University of Lisbon,
Tapada da Ajuda, 1349-017 Lisbon, Portugal the biogeographical history (Horwitz, 1978; Reyjol
e-mail: terferreira@isa.utl.pt et al., 2007).

123
 Hydrobiologia

The current state of fish assemblages is likely to structure of the fish assemblages (Rahel & Hubert,
suffer from alterations due to climate change (Buisson 1991; Wehrly et al., 2003). Furthermore, as it impacts
& Grenouillet, 2009). The compositional change of all life stages (Elliott, 2000), it serves as predestined
assemblages will be driven by extirpation, adaptation or indicator to analyse climate warming impacts (Buis-
migration (Comte et al., 2012) whereby the latter may son & Grenouillet, 2009; Logez et al., 2012).
be hindered by barriers in the river network (Hein et al., Compared to climate, atmospheric energy fluxes
2011). However, before one species disappears or trigger warming and cooling in river networks (Cais-
another occurs, a structural change can also be revealed sie, 2006). Accordingly, air and water temperature are
by abundance shifts within the existing assemblage. correlated in streams supporting the use of air
Different environmental gradients like river slope temperature as a surrogate for water temperature on
or temperature induce non-random patterns of fish larger scales (Webb et al., 2008). Water temperature is
assemblages along the longitudinal continuum of predicted to increase compared to air temperatures due
rivers (Huet, 1959; Vannote et al., 1980). Since the to climate change. The twentieth century was the
nineteenth century, systems were developed to warmest century since the year 1500 in Europe
describe this longitudinal succession of fish species (Luterbacher et al., 2004), and a further increase up
in Europe (Fritsch, 1872). This transition of fish to additional 6°C of air temperature is predicted for the
assemblages is highlighted by two aspects: (1) down- twenty-first century (IPCC, 2007). Moreover, chang-
stream increase of species richness and biomass, and ing precipitation patterns will affect riverine run off.
(2) turnover in species composition from salmonid to Arid regions of Europe are predicted to have less
cyprinid communities (Illies, 1961; Belliard et al., precipitation, whereas in northern and alpine regions,
1997). Both are related to species addition and the amount of rainfall will possibly increase (Lehner
replacement which occur simultaneously in a down- et al., 2006). Consequently, climate change will
stream direction (Lasne et al., 2007; Roberts & Hitt, impact riverine water temperature and flow regimes
2010). (Poff & Zimmerman, 2010).
Thienemann (1925) conceptualised those processes We use the Fish Zone Index (FiZI; Schmutz et al.,
and characterised river sections by their dominating 2000a) to evaluate the potential effects of climate
species, introducing the fish zonation: the brown trout change on the longitudinal structure of fish assem-
(Salmo trutta fario L.), the European grayling (Thy- blages in European rivers. Beside the climate descrip-
mallus thymallus L.), the barbel (Barbus barbus L.), tors derived from high-resolution climate raster, we
the bream (Abramis brama L.), the ruffe (Gymno- implement habitat characteristics by considering site-
cephalus cernua L.) and the flounder (Platichthys specific descriptors of the riverine environment (e.g.
flesus L.). Illies (1961) built a more general framework river slope, wetted width, size of upstream catchment).
detached from flagship species by dividing the river This study focuses on the following research
course into the six sections from epirhithral to questions: (1) Is the Fish Zone Index useful and
hypopotamal. Several studies were based on the fish applicable to describe the longitudinal succession of
zonation concept (e.g. Brosse, 2001; Reyjol et al., fish assemblages in rivers on the European scale? (2)
2003; Lasne et al., 2007; Matulla et al., 2007), and Do climatic factors play a role in the transition of fish
highlighted the applicability of this approach on small assemblages on the longitudinal gradient? and (3)
scales especially in unimpacted rivers (Rodriguez & How do fish assemblages react to climate change
Magnan, 1995; Aarts & Nienhuis, 2003; Ibarra et al., impacts in different river sections and regions of
2005). Europe?
In hierarchical river networks, large-scale charac-
teristics such as climate or topography shape habitat
properties at smaller scales (Frissell et al., 1986). One Materials and methods
of the key factors of the physico-chemical environ-
ment in rivers is water temperature (Lyons, 1989; Fish data
Eaton & Scheller, 1996). Since fish as poikilothermic
organisms are highly sensitive to water temperature Fish sampling data (1983–2007, 75% after 2000) were
changes, this factor is crucial for determining the derived from the EFI?-database (European research

123
Hydrobiologia

P
project ‘Improvement and Spatial extension of the sp ðNsp  FIsp Þ
European Fish Index—EFI?’, contract number FiZI ¼ ; ð2Þ
N total
044096) containing data from several national insti-
tutions. All sites were sampled during low flow where Nsp is the number of individuals belonging to a
periods using electric fishing, either by wading or by species, and Ntotal the number of all caught fish at a
boat, depending on river depth. Data were confined to site. Theoretically, the FiZI ranges from 3 to 8. To
first pass catches of electric fishing surveys to keep the possibility to draw comparisons with aquatic
standardise sampling efforts between regions. For invertebrates, values 1 and 2 were reserved for the fish-
each sample, the number of individuals per fish species free hypo- and eucrenal zones (spring-brooks and
was recorded. springs; Jungwirth et al., 2003). In this study, we did
As anthropogenic modifications to rivers may alter not explicitly penalise non-native species in the FiZI
the fish assemblage structure, we only used sampling calculation.
sites which were either unaffected or slightly impacted
by anthropogenic pressures based on objective criteria Environmental data
(Schmutz et al., 2000b; Logez et al., 2012) to minimise
the potential bias due to non-climatic disturbances Current knowledge of habitat requirements driving
(Pont et al., 2006). fish distribution has guided the selection of variables
In addition, to limit the interference of spatial to describe environmental characteristics (Pont et al.,
autocorrelation, sampling sites with an effective 2005; Grenouillet et al., 2010). Accordingly, the
distance \10 km (i.e. measured on the basis of the environmental descriptors serving as independent
European river network Catchment Characterization variables in the modelling procedure can be classified
and Modeling version 2.1; CCM 2.1; Vogt et al., 2007) into two groups: (1) river-topographic and (2) climatic
were omitted. variables. The former were directly derived from the
EFI?-database and are assumed to remain constant
over time and to improve model accuracy in respect to
Fish Zone Index the habitat requirements of fish.
Climate characteristics play a vital role in the
In this study, we transfer the categorical and discrete thermal (temperature) and hydrological (precipitation)
fish zones as introduced by Thienemann (1925) into a regime of rivers at a given site. In this study, we used
continuous index, called Fish Zone Index (FiZI). In bioclimatic variables (Nix, 1986; Busby, 1991) which
general, the FiZI considers the species’ niches along incorporate information on climatic magnitudes and
the upstream–downstream gradient. At first, each seasonalities of air temperature and precipitation
species observes a species-specific index (FIsp) which (O’Donnell & Ignizio, 2012).
expresses the preference of a species for a fish zone The climatic variables were evaluated on two
and is calculated accordingly: spatial units: local at the sampling site, and in the
P
sp ð3  p3 þ 4  p4 þ 5  p5 þ 6  p6 þ 7  p7 þ 8  p8 Þ whole area of the upstream catchment, taking large-
FIsp ¼ P8 :
scale climatic processes into account. We delineated
i¼3 pi
the upstream catchment of each sampling site on the
ð1Þ
basis of the CCM 2.1 network (Vogt et al., 2007) and
The classification of expected occurrence (p3–p8 in calculated means, minima, and maxima.
formula (1) and the resulting FIsp is based on Schmutz Furthermore, we introduced a variable describing
et al. (2000a) and Haunschmid et al. (2006), which was local run-off potential which combines the catchment
also used in Matulla et al. (2007). In our study, the area and the precipitation within this area. A geo-
classification was extended with information from graphical information system (GIS; ESRIÓ, 2011)
Dussling et al. (2004) who added native and non- was used for local queries and catchment delineation.
native species mainly occurring in German rivers and The Geospatial Modelling Environment (Beyer, 2012)
especially in downstream sections. was used for the upstream catchment queries.
The FiZI which characterises the assemblage is Current climate conditions were derived from
then calculated according to the formula: WorldClim (Hijmans et al., 2005; version 1.4). This

123
 Hydrobiologia

database contains interpolated climate surfaces for (log for RS, SUC and sqrt for ELE). Normality was
land areas at a spatial resolution of 30 arc seconds proven by a Shapiro–Wilk test. Principal component
(approx. 1 km at equator). The interpolation was analysis (PCA) was applied on the independent
based on measured data at weather observation variables for explorative purposes (Wold, 1976). A
stations between 1950 and 2000 representing contem- canonical correspondence analysis (CCA) identified
porary climate. gradients in the community data matrix (Legendre &
The potential change of the future climate is related Legendre, 2012). The CCA results were tested for
to different emission scenarios developed by the significance by analysis of variance (ANOVA) and
Intergovernmental Panel on Climate Change (Nakice- visualised in a biplot showing the relationship of
novic & Swart, 2000). We used downscaled outputs species and independent variables. The number of
from the global circulation model ECHAM5 (Roeck- species names was reduced to improve readability of
ner et al., 2003) with a resolution of 30 arc seconds. the plot. Names of the most frequent species in the
The data were derived from the Centre for Tropical dataset, flagship species of the fish zonation and
Agriculture (CIAT; obtained from http://ccafs- mediterranean species are shown.
climate.org). We used two future periods, 2050s (30- The FiZI-response to the independent variables was
year average from 2040 to 2069) and 2080s (30-year determined by a weighted least squares (WLS)
average from 2070 to 2099), and three different IPCC regression approach, minimising the sum of weighted
emission scenarios (A1b, A2, and B1). squared residuals (Jacquez & Mather, 1968). With an
The initial set of potential independent variables efficient branch-and-bound algorithm (Miller, 2002)
was reduced after correlation analysis. Correlations each possible combination of variables was evaluated.
were tested with the Spearman’s rank correlation (q) The best subset of variables was identified at each
which is robust to outliers (Spearman, 1904). We level of variable quantity according to three criteria:
eliminated high statistical dependence for pairs of first, adjusted R-squared (AdjR2; Steel & Torrie,
variables with a correlation coefficient q [ |0.75| and 1960) penalising the coefficient of determination
retained the variables with the highest explanatory based on the number of variables in the model;
power. second, Mallows’ Cp (Cp; Mallows, 1973) addressing
The final dataset contained 13 independent vari- the issue of overfitting; and third, the Bayesian
ables: five variables related to river-topography (size information criterion (BIC; Schwarz, 1978) describing
of upstream catchment—‘SUC’, km2; elevation— the trade-off between bias and variance in the model’s
‘ELE’, m above sea level; river slope—‘RS’, m per construction. The final model was identified by an
km; distance to the sea—‘SEA’, km; and wetted ANOVA which tested for the significance of increased
width—‘WW’, m); four variables characterising air explained variance (Chambers & Hastie, 1992).
temperature (annual mean temperature—‘TMeAn’, Model evaluation was performed graphically and
°C; annual temperature range—‘TRA’, °C; maximum numerically. The former utilised plots of observed
temperature of the warmest month—‘Tmax’, °C, versus predicted values (Piñeiro et al., 2008) as well as
mean temperature of warmest quarter in the upstream histograms showing the density distribution of resid-
catchment—‘TmeWQUC’, °C); three variables uals (Cohen et al., 2003). The latter was done by cross-
describing the precipitation regime (annual precipita- validation (CV) and the variance inflation factor (VIF;
tion—‘APP’, mm; precipitation seasonality—‘PSeas’, Fox & Monette, 1992). The CV evaluated the accuracy
unitless coefficient of variation; and precipitation of and stability of the final model. We applied a permuted
warmest quarter—‘PWQ’, mm); as well as one CV with 10,000 random splits into training (70%) and
variable describing run off potential at the sampling test (30%) data. The absolute difference in the
site (‘RO’, mm per km2). coefficient of determination (Rsqu) between training
and test data was calculated for each repetition. The
Statistical analyses corresponding mean and median of all differences
were used as indicators of model stability. The VIF
All statistical analyses were performed with R, version tested for multi-collinearity between the variables in
2.13 (R Development Core Team, 2011). If necessary, the model. High levels of VIF indicate an adverse
the independent variables were initially transformed effect of collinearity in the multiple regression

123
Hydrobiologia

The climatic regions, the third level, are based on

the five major biomes in Europe (alpine, atlantic,
boreal, continental, and mediterranean) which delin-
eate homogeneous units on the scale of Europe
(Fig. 1). The biomes represent a simplified version
of the biogeographical regions (EEA, 2011; avail-
able at www.eea.europa.eu/data-and-maps/figures/
biogeographical-regions-in-europe). The allocation
of the sampling sites to the biomes was done in a GIS
(ESRIÓ, 2011).

Results

The 559 fish sampling sites taken from 14 European

countries (Fig. 1) contained 92 fish species. Table 1
shows a full list of all species, as well as the expected
occurrence in the different fish zones which builds the
Fig. 1 Location of fish sampling sites in Europe (N = 559) and basis to calculate the FIsp. The species are distributed
the delineation of the five biomes over 18 families, which were dominated by Cyprin-
idae (46 species), followed by Salmonidae, Percidae
(both eight species) and Gobiidae (seven species).
analysis. Commonly, a maximum level of 10 is The most frequent species in the dataset was
reported for VIF in literature (O’Brien, 2007). Phoxinus phoxinus (L., 20.1%), followed by S. trutta
The final WLS-model was used to predict potential fario (L., 18.6%), and Cottus gobio (L., 8.7%). The
changes of FiZI according to three climate change five most frequent species were completed by Barba-
scenarios (A1b, A2, and B1) and two future periods tula barbatula (L.) and Alburnus alburnus (L.),
(2050s, 2080s), summing up to six future projections. accounting for 7.0 and 6.1%, respectively. Overall,
The predicted FiZI values were censored to the these five species accounted for 60.4% of all speci-
minimum and maximum value in the dataset to mens in the dataset. In total, fish sampling data
prevent extrapolation. The difference of FiZI between recorded 123,576 individuals.
current and future conditions was used as indicator for Table 2 shows the characteristics of the environ-
a change in assemblage structure. The values of mental parameters for current and future conditions.
change were transformed into binary information by a The PCA on the environmental variables explained
threshold of [|0.5| as indicator of significant change. 49.04% of total variation in the first two components
The change of FiZI was investigated on three indicating some heterogeneity of environmental char-
levels: First, on the level of the complete study area, acteristics in this continental-wide dataset. The vari-
comparing all sampling sites. The second and third ables characterising precipitation (annual precipitation,
levels subsume the sampling sites into spatial subunits precipitation of warmest quarter, and run off) loaded
to enable a more detailed investigation of FiZI negatively in the first component (-0.43, -0.42, and -
response according to distinct river sections, respec- 0.32, respectively), whereas air temperature indicators
tively, types, and among climatic regions in Europe. loaded positively (TMeAn: 0.24, Tmax: 0.30, and
The level of river sections is based on hierarchical TmeWQUC: 0.42). The second component discrimi-
clustering of the sampling sites. Three variables were nated temperature range (0.44) and wetted width (0.23)
used for the clustering: SUC, RS, and RO. The data were from negatively loaded air temperature (-0.50) and
standardised and centred before calculating Euclidean annual precipitation (-0.41). The biplot (Fig. 2) shows
dissimilarities, followed by clustering according to the stream size variables (SUC, WW) as antagonists of
Ward agglomeration method (Ward, 1963). descriptors for flow conditions (RO, RS, ELE and APP).

123
 Hydrobiologia

Table 1 List of taxa in the dataset with the probabilities of occurrence (pi) from epirhithral (p3) to hypopotamal (p8) (E = epi,
M = meta, H = hypo), the resultant calculated species-specific fish index (FIsp) as well as the variance of FIsp (VAR)
Scientific name Rhithral Potamal
E (p3) M (p4) H (p5) E (p6) M (p7) H (p8) Fisp VAR

Abramis brama (Linnaeus, 1758) 3 6 3 7.00 0.55

Achondrostoma arcasii (Steindachner, 1866) 3 5 4 5.08 0.63
Achondrostoma oligolepis (Robalo et al., 2005) 2 6 4 5.17 0.52
Alburnoides bipunctatus (Bloch, 1782) 5 7 5.58 0.27
Alburnus alburnus (Linnaeus, 1758) 1 4 6 1 6.58 0.63
Ameiurus melas (Rafinesque, 1820) 7 5 6.42 0.27
Anguilla anguilla (Linnaeus, 1758) 1 1 3 3 4 6.67 1.70
Aspius aspius (Linnaeus, 1758) 4 7 1 6.75 0.39
Babka gymnotrachelus (Kessler, 1857) 5 7 6.58 0.27
Ballerus ballerus (Linnaeus, 1758) 1 7 4 7.25 0.39
Ballerus sapa (Pallas, 1814) 5 6 1 6.67 0.42
Barbatula barbatula (Linnaeus, 1758) 3 4 4 1 5.25 0.93
Barbus barbus (Linnaeus, 1758) 2 7 3 6.08 0.45
Barbus meridionalis (Risso, 1827) 2 4 4 2 5.50 1.00
Barbus peloponnesius (Valenciennes, 1842) 3 4 3 6.00 0.67
Blicca bjoerkna (Linnaeus, 1758) 3 6 3 7.00 0.55
Carassius auratus (Linnaeus, 1758) 3 7 6.70 0.23
Carassius carassius (Linnaeus, 1758) 3 8 1 6.83 0.33
Carassius gibelio (Bloch, 1782) 1 4 4 3 6.75 0.93
Chondrostoma nasus (Linnaeus, 1758) 3 8 1 5.83 0.33
Cobitis calderoni (Bacescu, 1962) 3 3 3 3 5.50 1.36
Cobitis paludica (de Buen, 1930) 4 4 4 6.00 0.73
Cobitis taenia (Linnaeus, 1758) 1 5 5 1 6.50 0.64
Cottus gobio (Linnaeus, 1758) 4 4 2 2 4.17 1.24
Cyprinus carpio (Linnaeus, 1758) 5 5 2 6.75 0.57
Esox lucius (Linnaeus, 1758) 2 3 5 2 6.58 0.99
Eudontomyzon mariae (Berg, 1931) 3 6 3 5.00 0.55
Gambusia holbrooki (Girard, 1859) 5 5 2 6.75 0.57
Gasterosteus aculeatus (Linnaeus, 1758) 1 2 3 6 7.17 1.06
Gobio gobio (Linnaeus, 1758) 1 4 4 2 1 5.83 1.24
Gymnocephalus baloni (Holcı́k & Hensel, 1974) 3 7 6.70 0.23
Gymnocephalus cernua (Linnaeus, 1758) 1 3 8 7.58 0.45
Gymnocephalus schraetser (Linnaeus, 1758) 8 4 6.33 0.24
Hucho hucho (Linnaeus, 1758) 4 8 5.67 0.24
Iberochondrostoma lemmingii (Steindachner, 1866) 4 4 4 6.00 0.73
Iberochondrostoma lusitanicum (Collares-Pereira, 1980) 4 4 4 6.00 0.73
Lampetra fluviatilis (Linnaeus, 1758) 2 6 4 4.96 1.54
Lampetra planeri (Bloch, 1784) 6 5 1 4.29 0.59
Lepomis gibbosus (Linnaeus, 1758) 4 8 6.67 0.24
Leucaspius delineatus (Heckel, 1843) 4 7 1 6.75 0.39
Leuciscus idus (Linnaeus, 1758) 4 6 2 6.83 0.52
Leuciscus leuciscus (Linnaeus, 1758) 1 4 4 3 5.75 0.93

123
Hydrobiologia

Table 1 continued
Scientific name Rhithral Potamal
E (p3) M (p4) H (p5) E (p6) M (p7) H (p8) Fisp VAR

Lota lota (Linnaeus, 1758) 1 3 3 3 2 6.17 1.61

Luciobarbus bocagei (Steindachner, 1866) 1 3 4 4 5.92 0.99
Luciobarbus comizo (Steindachner, 1866) 3 5 4 6.08 0.63
Luciobarbus graellsii (Steindachner, 1866) 3 5 4 6.08 0.63
Luciobarbus microcephalus (Almaça, 1967) 3 5 4 6.08 0.63
Luciobarbus sclateri (Günther, 1868) 3 5 4 6.08 0.63
Micropterus salmoides (Lacepède, 1802) 2 5 5 6.25 0.57
Misgurnus fossilis (Linnaeus, 1758) 3 7 2 6.92 0.45
Neogobius fluviatilis (Pallas, 1814) 5 7 6.58 0.27
Neogobius melanostomus (Pallas, 1814) 5 7 6.58 0.27
Oncorhynchus mykiss (Walbaum, 1792) 4 4 4 4.00 0.73
Padogobius bonelli (Bonaparte, 1846) 5 7 6.58 0.27
Padogobius nigricans (Canestrini, 1867) 5 7 6.58 0.27
Parachondrostoma miegii (Steindachner, 1866) 3 3 3 3 5.50 1.36
Pelecus cultratus (Linnaeus, 1758) 2 4 6 7.33 0.61
Perca fluviatilis (Linnaeus, 1758) 1 3 4 4 6.92 0.99
Petromyzon marinus (Linnaeus, 1758) 4 7 1 5.25 1.48
Phoxinus phoxinus (Linnaeus, 1758) 3 6 3 5.00 0.55
Platichthys flesus (Linnaeus, 1758) 1 4 7 7.50 0.45
Proterorhinus marmoratus (Pallas, 1814) 3 7 6.70 0.23
Pseudochondrostoma polylepis (Steindachner, 1866) 3 4 3 2 5.33 1.15
Pseudochondrostoma willkommii (Steindachner, 1866) 3 5 4 6.08 0.63
Pseudorasbora parva (Temminck & Schlegel, 1846) 1 5 6 5.42 0.45
Pungitius pungitius (Linnaeus, 1758) 2 6 4 7.17 0.52
Rhodeus amarus (Bloch, 1782) 6 6 6.50 0.27
Romanogobio albipinnatus (Lukasch, 1933) 5 5 6.50 0.28
Romanogobio kesslerii (Dybowski, 1862) 3 7 6.70 0.23
Romanogobio uranoscopus (Agassiz, 1828) 2 7 3 6.08 0.45
Rutilus pigus (Lacepède, 1803) 2 10 5.83 0.15
Rutilus rutilus (Linnaeus, 1758) 1 3 5 3 6.83 0.88
Sabanejewia balcanica (Karaman, 1922) 3 6 3 6.00 0.55
Salaria fluviatilis (Asso, 1801) 1 2 5 4 5.92 0.80
Salmo salar (Linnaeus, 1758) 3 6 3 5.00 0.55
Salmo trutta fario (Linnaeus, 1758) 5 5 2 3.75 0.57
Salmo trutta lacustris (Linnaeus, 1758) 8 4 4.33 0.24
Salmo trutta trutta (Linnaeus, 1758) 3 6 3 5.00 0.55
Salvelinus fontinalis (Mitchill, 1814) 6 6 3.50 0.27
Sander lucioperca (Linnaeus, 1758) 2 5 5 7.25 0.57
Sander volgensis (Gmelin, 1789) 2 8 6.80 0.18
Scardinius erythrophthalmus (Linnaeus, 1758) 3 7 2 6.92 0.45
Silurus glanis (Linnaeus, 1758) 2 9 1 6.92 0.27
Squalius carolitertii (Doadrio, 1988) 4 3 3 2 5.25 1.30
Squalius cephalus (Linnaeus, 1758) 1 4 4 2 1 5.83 1.24

123
 Hydrobiologia

Table 1 continued
Scientific name Rhithral Potamal
E (p3) M (p4) H (p5) E (p6) M (p7) H (p8) Fisp VAR

Squalius pyrenaicus (Günther, 1868) 4 3 3 2 5.25 1.30

Telestes souffia (Risso, 1827) 7 5 5.42 0.27
Thymallus thymallus (Linnaeus, 1758) 3 7 2 4.92 0.45
Tinca tinca (Linnaeus, 1758) 3 7 2 6.92 0.45
Tropidophoxinellus alburnoides (Steindachner, 1866) 4 3 3 2 5.25 1.30
Vimba vimba (Linnaeus, 1758) 1 5 4 2 6.58 0.81
Zingel streber (Siebold, 1863) 3 8 1 5.83 0.33
Zingel zingel (Linnaeus, 1766) 9 3 6.25 0.20

The CCA identified a gradient which describes the the regression line of the predicted versus the observed
longitudinal succession of habitats in rivers from values (Fig. 4a), which was characterised by a slope of
upstream, with high slopes and high elevations, to 0.988 and an intercept of 0.04. The histogram of residuals
downstream, with large catchments sizes and wetted showed a short-tailed normal distribution (Fig. 4b).
widths in the second axis (Fig. 3). In general, the sorting Secondly, the VIF was between 1.597 and 6.918 for all
of the species was in line with the succession of species the variables, indicating no constraints of multi-collin-
as described by FIsp (Table 1). The species with the earity (VIF values: RS = 2.51, SUC = 5.97, ELE =
lowest FIsp, Salvelinus fontinalis (Mitchill, 1814), is 2.71, WW = 3.12, TmeWQUC = 6.66, TRA = 1.60,
located on the beginning of this gradient whereas and Tmax = 6.92). Thirdly, the results of CV showed a
species with increasing FIsp scores follow accordingly mean difference of 0.173 in Rsqu between training and
(e.g. S. trutta fario, T. thymallus, B. barbus). Nonethe- test sets. The small difference between mean and median
less, the sequence of species does not always fully fit the of Rsqu (0.003) indicated a stable model framework. The
expected FIsp. For example, Gymnocephalus cernua distribution of the Rsqu-differences converged in a
(L.), a species with a high FIsp, was located in the middle slightly short-tailed normal distribution (Fig. 5).
of the CCA gradient. In contrast, Gymnocephalus M7 was used to predict future values of FiZI with
baloni (Holcik & Hensel, 1974) was positioned quite variables summarised in Table 3. The values of Tmax
at the end as expected. The first axis of the CCA increased for all scenarios in both periods compared to
highlighted the temperature differences in the dataset the contemporary situation. For the 2050s, scenario
and was mainly based on mediterranean species (e.g. A1b reached the highest values of Tmax (mean =
Barbus sclateri, Squalius pyrenaicus, Fig. 3). 26.0°C, max = 35.7°C), whereas scenario A2 shows
The applied model selection algorithm identified 12 the highest ones in the 2080s (mean = 27.2°C, max =
best models (Table 3). At first, TMeAn was included 38.4°C). TmeWQUC showed a similar trend with
into model calibration but had to be excluded due to highest values for A1b in the 2050s (mean = 18.4°C),
collinearity problems indicated by VIF [10. The and for A2 in the 2080s (mean = 19.1°C). Scenario
models were consecutively named according to the B1 represented the lowest values of Tmax and
quantity of independent variables from ‘M1’ to ‘M12’. TmeWQUC for both periods (2050s: mean Tmax =
The final model M7 was characterised by an AdjR2 of 24.5°C, mean TmeWQUC = 17.4°C, 2080s: mean
0.66, a Cp of 7.8 and a BIC of -560. M7 contained the Tmax = 26.8°C, mean TmeWQUC = 18.3°C). The
following variables with according regression coeffi- picture for TRA appeared more heterogeneous. For the
cients: RS (-0.154), SUC (0.114), ELE (-0.010), 2050s, the mean values of TRA were comparable to
WW (0.002), Tmax (-0.038), TRA (0.031), TmeW- the contemporary situation. In scenarios A1b and B1,
QUC (0.173), and the Intercept (2.089). the means increased around 0.2°C, and for A2 the
Model validation resulted in three major outcomes. mean was around 0.3°C lower. TRA decreased
First, the 1:1-line showed small deviation compared to progressively in the three scenarios.

123
Hydrobiologia

Table 2 List of variables with units, means, standard devia- well as future conditions according to two periods (2050s,
tions (SD), medians, minima (Min) and maxima (Max) in the 2080s) and three climate change scenarios (A1b, A2, B1)
dataset (N = 559) to characterise contemporary conditions as
Variable Unit Mean SD Median Min Max

Contemporary
RS m km-1 11.46 15.4 6.02 0 111.18
SUC km2 1706 8,630 65 2 108,970
ELE m a.s.l. 276 297 155 1 1,945
SEA km 479.7 669 189.9 0 2,835.7
WW M 11.8 22 5 1 210
TMeAn °C 8 3.6 8 -1.7 17.3
Tmax °C 22.8 2.9 22.3 16 32.4
TRA °C 27.6 4.9 27.8 16.4 38.5
TmeWQUC °C 15.7 2.5 15.9 6.1 23.8
RO mm km-2 1,551 384 1,499 764 3,179
APP mm 773 214 712 396 1,554
PPSeas – 27 10 27 7 66
PWQ mm 215 74 209 21 490
Future
2050s
A1b
Tmax °C 26 3.1 25.4 18.2 35.7
TRA °C 27.8 4.2 28.1 17.3 36.5
TmeWQUC °C 18.4 2.6 18.5 9.9 26.3
RO mm km-2 1,544 385 1,459 700 3,040
A2
Tmax °C 25.6 3.4 24.9 17.2 35.6
TRA °C 27.3 4.1 27.9 16.9 36.6
TmeWQUC °C 18.2 2.7 18.2 9.8 26.9
RO mm km-2 1,563 404 1,506 614 3,185
B1
Tmax °C 24.5 2.8 23.9 17.9 34.4
TRA °C 27.8 4.5 27.9 17 38.1
TmeWQUC °C 17.4 2.4 17.6 8.1 25.8
RO mm km-2 1,523 398 1,432 661 3,037
2080s
A1b
Tmax °C 26.9 3.4 26.6 19.2 37
TRA °C 28.4 4.2 28.9 17.5 35.7
TmeWQUC °C 18.9 2.8 18.8 10.6 27.5
RO mm km-2 1,552 423 1,464 454 3,094
A2
Tmax °C 27.2 3.8 26.8 18.7 38.4
TRA °C 27.9 4 28.8 17.5 35.6
TmeWQUC °C 19.1 2.9 19.1 11.3 28
RO mm km-2 1,552 445 1,434 448 3,138

123
 Hydrobiologia

Table 2 continued
Variable Unit Mean SD Median Min Max

B1
Tmax °C 26.8 3.1 26.5 19.6 36.8
TRA °C 29.4 4.2 29.8 18.6 37.2
TmeWQUC °C 18.3 2.6 18.3 9.8 26.6
RO mm km-2 1,551 410 1,473 557 3,069
RS river slope, SUC size of upstream catchment, ELE elevation, SEA distance to the sea, WW wetted width, TMeAn annual mean
temperature, Tmax maximum temperature of warmest month, TRA annual temperature range, TmeWQUC mean temperature of
warmest quarter in the upstream catchment, RO run off, APP annual precipitation, PPSeas precipitation seasonality, PWQ
precipitation of warmest quarter

The clustering of the sampling sites gave six range compared to the other clusters. In scenario A1b
clusters (LC1–LC6; Table 4). In general, the clusters and B1 the range rather increased from the 2050s to the
followed the longitudinal gradient from upstream to 2080s which inverted for scenario A2 where the range
downstream which is characterised by decreasing shrunk for all clusters except LC6.
slope and increasing catchment size. In addition, a The boreal realm showed the lowest response of
trend of decreasing run off was indicated over all six FiZI (mean change between 0.11 and 0.23), in contrast
clusters with exception of LC3. Table 4 also shows the to the mediterranean area (mean change between 0.56
mean values of FiZI for each cluster, indicating an and 0.89) irrespective of scenario and time period
accompanying transition of the fish assemblage in (Fig. 7). Especially the FiZI change in the scenarios
each cluster from rhithral to potamal sections. A1b and B1 clearly set the mediterranean biome apart
Future changes of FiZI ranged from -0.55 to 0.92 in from the others.
the 2050s, and from -0.35 to 1.18 in the 2080s in the In general, sites in the alpine biome tended to have
whole study area (Table 5). In both periods, scenario lower responses in the 2050s and stronger responses in
A1b (max = 0.92 and 1.18, respectively) indicated the the 2080s compared to the atlantic and continental
strongest maximum change followed by A2 (0.90/ biomes. The responses of sites in the boreal realm
1.05) and B1 (0.79/0.92), whereas in the 2080s, the indicated also negative values for all scenarios except
mean value of A2 (0.43) is larger than for A1b (0.41). for the 2080s A2 scenario. Negative responses were
Noteworthy, in the 2080s, the median of B1 was equal also observed in the continental and alpine sites but
to A2 (0.39) which was lower than for A1b (0.45). with lower frequency and magnitude for all scenarios
On the level of longitudinal clusters, the FiZI in the 2050s.
changes generally showed a homogenous pattern of The transformation of FiZI change into a binary
increase (Fig. 6). In the 2050s, the highest mean and response underlined the patterns (Table 6). The high-
maxima of FiZI change occurred in LC2 closely est portion of sites with significant impact was found
followed by LC1. LC4 and LC5 reached comparable for scenario A2 in the 2050s (31%) and for A1b in the
maxima, but the means were considerably lower than 2080s (41%). In the longitudinal clusters, the highest
those in LC1 and LC2. This was also underlined by portion of affected sites was found for LC1 (85% in
larger ranges of FiZI up to 1.44. LC3 and LC6 showed scenario A2 2080s) and LC2 (55% in scenario A2
the lower responses. 2080s). LC6 had no affected sites in the 2050s. In the
In the 2080s, the pattern of changes was different. 2080s, the response signal in LC6 diversified in the
The means of LC1 (A1b = 0.58, A2 = 0.66, three scenario (A1b = 27, A2 = 45, B1 = 0 percent
B1 = 0.45) were higher than for LC2 (A1b = 0.45, of affected sites).
A2 = 0.48, B1 = 0.43) which was inverted for the Among the biomes, the mediterranean one had the
maxima values. LC3, LC4, and LC5 were character- highest portion (up to 100%) of affected sites in all
ised by a comparable means of change in the 2080s scenarios and both periods; closely followed by the
(A1b/A2 = 0.4, B1 = 0.35). In general, lowest alpine region for scenario A2 in the 2080s. In contrast,
change was found in LC6 which also showed a narrow the boreal realm generally showed the smallest portion

123
Hydrobiologia

of sites with FiZI change [|0.5|. Only for scenario B1

2
in the 2050s, the alpine sites indicated the lowest
TRA
response. Interestingly, the increase of affected sites
from the 2050s to the 2080s is the highest for alpine
1

WW sites (from 15 to 82%) as well as for LC1 (from 15 to

SUC

PPSeas
85%) in scenario A2.
PC2

PWQ SEA
0

RO ELE Discussion
−1

RS
TmaxTmeWQUC
Applicability of Fish Zone Index
APP
−2

TMeAn
The FiZI provides a structural index to characterise
fish assemblages in regard to the longitudinal gradient
−2 −1 0 1 2
of a river by a fuzzy partition of the fish zonation
PC1
concept. The index is based on abundance data, and
Fig. 2 Biplot of first two principal components (PC) of 13
takes rare species into account too. In contrast to the
environmental descriptors used for model calibration (variance criticised discrete entities of the fish zonation concept
explained by first component 27.7%, second component 18.7%). (Miranda & Raborn, 2000; Lasne et al., 2007), the FiZI
RS river slope, SUC size of upstream catchment, ELE elevation, describes assemblages by a continuous value and
SEA distance to the sea, WW wetted width, RO run off, Tmax
maximum temperature of warmest month, TRA annual temper- enables addressing both sharp transitions as well as
ature range, TmeWQUC mean temperature of warmest quarter in gradual shifts in assemblage structures. Furthermore,
the upstream catchment, APP annual precipitation, PPSeas the FiZI is not dependent on flagship species that may
precipitation seasonality, PWQ precipitation of warmest quarter vary in their ability to indicate a distinct zone or be
completely missing in particular basins (Lasne et al.,
2007). In contrast, FiZI enables the consideration of
distinct specifications of the regional species pool
Gymnocephalus baloni which may exist due to the bio-geographical history of
6

a catchment (Reyjol et al., 2007) as well as large-scale

Abramis sapa processes like climate (Heino et al., 2009).Thus, the
1

Pelecus cultratus
4

WW FiZI is flexible to adapt to local or regional peculiar-

SUC
Alburnus alburnus
Abramis brama
ities and seems appropriate for characterising fish
CCA2

TmeWQUC assemblages over large spatial extents.

2

TRA barbus
Barbus
Tmax
Gymnocephalus cernua PPSeas In this study, the FiZI enabled us to assess the
SEA Barbus sclateri
Barbatula barbatula effects of river characteristics and climate on fish
Squalius pyrenaicus
0

Thymallus thymallus
Salmo trutta fario assemblage structures throughout Europe at a site-
PWQ grain. A good model performance (explained vari-
−2

Salvelinus fontinalis
RS
ELE
ability[60%) underlined the reasonable usage of this
RO APP index also on the European scale. The CCA empha-
−4 −2 0 2 4 6 sised the sorting of the species along a progressing
CCA1 gradient which mirrors the longitudinal succession of
habitat types. Noteworthy, there were species which
Fig. 3 Biplot of canonical correspondence analysis (CCA)
showed a deviating position as expected, like G.
showing the environmental variables and species scores; species
labels are reduced to simplify the reading of the plot. RS river cernua. On closer examination, those species were not
slope, SUC size of upstream catchment, ELE elevation, SEA well represented in the dataset which potentially
distance to the sea, WW wetted width, RO run off, Tmax skewed the loading in the CCA. The positioning of
maximum temperature of warmest month, TRA annual temper-
mediterranean species on the gradient of temperature
ature range, TmeWQUC mean temperature of warmest quarter in
the upstream catchment, APP annual precipitation, PPSeas differences seems reasonable as these areas are
precipitation seasonality, PWQ precipitation of warmest quarter climatologically distinct (Gasith & Resh, 1999).

123
 Hydrobiologia

Table 3 Hierarchical Model IV RSS AdjR2 Cpå BIC F Pr([F) Sig-level

model selection with
residual sum of squares 1 I, RS 57,840 0.455 350 -330
(RSS), adjusted R-squared
2 I, RS, Tmax 47,486 0.551 190 -430 161.455 \2.2E-16 ***
(AdjR2), Mallows’ Cp (Cp),
Bayesian information 3 I, RS, SUC, 37,643 0.644 36 -550 153.490 \2.2E-16 ***
criterion (BIC), F-statistics TmeWQUC
of ANOVA (F), probability 4 M3?APP 36,493 0.654 20 -570 17.926 2.69E-05 ***
values (Pr(\F)), and 5 M3?TRA?ELE 35,675 0.661 9.3 -570 12.755 0.00039 ***
significance level
6 M5?WW 35,509 0.662 8.7 -570 2.594 0.107865
7 M61Tmax 35,326 0.663 7.8 2560 2.854 0.091718
8 M7?APP 35,185 0.664 7.7 -560 2.189 0.139586
IV independent variables, 9 M8?PSeas 35,084 0.664 8.1 -560 1.577 0.220971
I Intercept 10 M9?RO 35,060 0.664 9.7 -550 0.372 0.542178
*** P \ 0.001, P \ 0.1 for 11 M10?PWQ 35,036 0.664 11 -540 0.386 0.534956
each level of variable 12 M11?SEA 35,015 0.663 13 -540 0.316 0.574235
quantity; final model in bold

(a) (b)
7

0.6
6
Observed

0.4
Density
5

0.2
4

0.0
3

3 4 5 6 7 −2 −1 0 1 2
Predicted Residuals

Fig. 4 Graphical model evaluation; a scatter plot of observed versus predicted values of Fish Zone Index (FiZI) with 1:1-line (solid
black), and observed versus predicted-regression line (dashed black), and b histogram of residuals of the final model with density curve

The fish assemblages change due to gradients of showed more relevance than factors related to
environmental drivers (e.g. human alterations, cli- precipitation. In line with Huet (1959), the results
mate) which are reflected by increases or decreases underlined the importance of slope in structuring
of the FiZI (e.g. an increase of FiZI in impound- fish assemblages along rivers. In an ecological
ments). The transition of fish assemblages along the context, slope serves as a proxy for mean hydraulic
longitudinal gradient described by our model was site conditions, similarly representing the physical
characterised by the interaction of river characteris- factors shaping the local habitat through flow
tics, like river slope or wetted width, and climatic velocity and shear stress (Lamouroux et al., 1999;
variables, in which temperature-related variables Pont et al., 2005).

123
Hydrobiologia

10 modify fish assemblages in riverine ecosystems

(Dudgeon et al., 2006) necessitating appropriate tools
to identify potential changes.
A majority of existing studies that addressed the
8

relationship between fish assemblages and their river-

ine environment have focused on relatively small
spatial extents (e.g. Lasne et al., 2007) or on coarse-
6
Density

grained data (e.g. Heino, 2008). Both approaches limit

the ability to determine large-scale influences (Essel-
man & Allan, 2010). Our data covered a large part of
4

the European continent at a high grain resolution. In

contrast to studies solely based on species distributions
(Hari et al., 2006; Buisson & Grenouillet, 2009), the
2

FiZI also considers species in their abundance to

evaluate changes at the assemblage level. Indeed,
assemblage composition is rather determined by
species abundance than by pure species presence
0

(Belliard et al., 1997). Hence, abundance data give

0.0 0.1 0.2 0.3 additional value to assessment approaches.
Difference in Rsqu As small streams and upstream reaches of rivers,
which are highly represented in the dataset of this
Fig. 5 Numerical model evaluation: boxplot a and histogram,
b of the differences in the coefficient of determination (Rsqu) of study, feature predominantly species-poor assem-
10,000 cross validated repetitions with 50/50 splitting blages, a fortiori, the importance of abundance infor-
mation increases (Schmutz et al., 2000b). Therefore,
we are convinced that the use of abundance data is
Table 4 Number of sampling sites, as well as the mean values
of environmental variables used to build the clusters and mean obviously an advantage of FiZI compared to the
FiZI in the six longitudinal clusters (LC1–LC6) and the methods based on presence–absence information. On
according mean Fish Zone Index (FiZI) the national scale, the FiZI is already implemented
n RS mean SUC mean RO mean FiZI mean into fish-based assessment methods in respect of the
Water Framework Directive (WFD) in Austria, Ger-
LC1 13 79.83 23 1,821 3.98 many and Switzerland (Dussling et al., 2004; Schager
LC2 92 23.83 40 1,676 4.36 & Peter, 2004; Haunschmid et al., 2006).
LC3 72 16.08 189 2,288 4.24
LC4 216 7.54 878 1,496 4.80 Model performance and sources of uncertainty
LC5 155 2.47 952 1,206 5.54
LC6 11 0.55 54,450 1,327 6.55 The evaluation of the final model indicated a satisfac-
RS river slope, SUC size of upstream catchment, RO run off tory performance and robustness of the modelling
framework that is pivotal for the prediction of
ecological responses (Hulme, 2005). Nearly two-
Furthermore, temperature played a vital role in thirds of the variance in our European-wide, fine-
structuring fish assemblages. This finding coincides grained dataset was explained by the final model with
with Lasne et al. (2007) who highlighted the impor- stable portions of explained variance in the cross-
tance of temperature in structuring assemblages along validation (CV). This is comparable to Lasne et al.
the river gradient in the Loire basin. Moreover, (2007) who worked at a much smaller scale. None-
Buisson et al. (2008a) found temperature as a primary theless, some variance remains unexplained, which
factor determining fish species distribution in France. could be attributed to the following issues: widely
In concordance to Logez et al. (2012), temperature is a distributed sampling sites covering a large geograph-
dominant factor at large scales. Future shifts in ical gradient (Heino, 2011); small-scale peculiarities
temperature regimes due to climate change will in the structural and functional environments, which

123
 Hydrobiologia

Table 5 Description of FiZI change (=predicted future value-modelled current value) for all sampling sites according to three
climate change scenarios (A1b, A2, B1) and two time periods (2050s, 2080s)
Scenario N 2050s 2080s
Mean SD Median Min Max Mean SD Median Min Max

A1b 559 0.36 0.26 0.39 -0.38 0.92 0.41 0.30 0.45 -0.34 1.18
A2 559 0.33 0.30 0.36 -0.55 0.90 0.43 0.20 0.39 0.00 1.05
B1 559 0.25 0.24 0.25 -0.36 0.79 0.36 0.26 0.39 -0.35 0.92
SD standard deviation, Min minimum, Max maximum
1.5
FiZI change FiZI change
A1b

0.5
1.5 −0.5
A2

0.5
1.5 −0.5
FiZI change
B1

0.5
−0.5

LC1 LC2 LC3 LC4 LC5 LC6 LC1 LC2 LC3 LC4 LC5 LC6
2050s 2080s
Fig. 6 Boxplots of predicted FiZI changes in the six longitudinal clusters (LC1–LC6) for two time periods (vertical panels) and three
climate change scenarios (horizontal panels)

are not represented in the available variables (Thorp (Tmax, TmeWQUC) impacting fish assemblages
et al., 2006); and potential genetic variation within one through maximum temperature thresholds as well as
species (Hansen et al., 2002). on the thermal range throughout the year (TRA) which
Another source of uncertainty arises from the used implements also thermal minimum thresholds.
variables in the analyses. We sought to minimise the Although mean hydrological conditions (i.e. mean
biases caused by environmental disturbances by using discharge volume) are mirrored by the size of the
only sampling sites in the least impacted river reaches upstream catchment, especially actual flow regimes at
where the structure of fish assemblages should not be the sampling sites and the intra-annual variability of
affected by the presence of additional anthropogenic discharge patterns were not described in detail by the
pressures (Pont et al., 2006; Schinegger et al., 2012). available variables due to a lack of information.
However, due to correlation in-between the cli- Further, our variables did not contain information on
matic variables, a pre-selection of independent vari- the timing and intensity of flow magnitudes for recent
ables was inevitable to avoid collinearity. We as well as future climate. Besides the lack of detailed
therefore focussed on the warm period of the year hydrological data, the reliability of flow regime

123
Hydrobiologia

1.5
FiZI change FiZI change
A1b

0.5
1.5 −0.5
A2

0.5
1.5 −0.5
FiZI change
B1

0.5
−0.5

ALP ATL BOR CON MED ALP ATL BOR CON MED
2050s 2080s
Fig. 7 Boxplots of predicted FiZI changes in the five biomes (ATL atlantic, ALP alpine, BOR boreal, CON continental, MED
mediterranean) for two time periods (vertical panels) and three climate change scenarios (horizontal panels)

Table 6 Relative portion of sites with a significant change MED = mediterranean) and the six longitudinal clusters
([|0.5|) in FiZI for the five biomes (ATL = atlantic, (LC1–LC6) for two time periods (2050s, 2080s) and three
ALP = alpine, BOR = boreal, CON = continental, climate change scenarios (A1b, A2, B1)
Biome Longitudinal Cluster Total
ALP ATL BOR CON MED LC1 LC2 LC3 LC4 L5 LC6

N 67 105 158 169 60 13 92 72 216 155 11

2050s
A1b 15 27 11 34 92 23 39 17 35 28 0 30
A2 15 27 18 31 92 15 41 17 39 24 0 31
B1 1 12 8 5 73 8 16 4 18 14 0 14
2080s
A1b 58 34 9 47 100 62 54 40 43 30 27 41
A2 82 19 0 50 85 85 55 49 35 22 45 38
B1 16 28 13 26 95 31 38 17 34 24 0 29
N gives the absolute number of sites in each category, all other values represent percentages

predictions for the future is still unsatisfactory (Boss- Future changes in fish assemblages
hard et al., 2013). The knowledge on variations in the
magnitude and frequency of extreme events (e.g. Future changes of fish assemblages due to climate
timing and intensity of droughts or floods) is vague. change seem inevitable (Comte et al., 2012). Three
Therefore, the role of these processes in shaping the mechanisms will play a vital role shaping future
structure of fish assemblages in the future stays changes of fish assemblage structures: first, the
undetermined. replacement of species; second, range shifts (Rahel

123
 Hydrobiologia

& Hubert, 1991; Belliard et al., 1997), and third, habitat configuration of upstream reaches currently
species extirpations (Buisson et al., 2008b). Assuming filters the downstream species due to hydraulic
a relatively stable river topography in the future, these roughness and thermal restrictions (Poff, 1997). The
mechanisms will be mainly triggered by altered thermal regime hinders species to complete life cycles,
regimes of temperature and river flow in unimpaired as minimum temperature thresholds for reproduction
or slightly impacted rivers (Poff & Zimmerman, are not reached (Pont et al., 2005). In combination
2010). with lower responses in downstream sections, this
If habitat conditions deteriorate beyond the species’ indicates a potential homogenisation of fish assem-
tolerance, then species will migrate in the river blages, also found by Buisson & Grenouillet (2009).
network to track their preferred environmental range. On the continental scale, the FiZI change showed
The migration ability of fish is constrained by the clear patterns according to distinct regions. The
structure of the river network (Comte & Grenouillet, mediterranean biome was highlighted by strongest
2013) as well as by anthropogenic impacts (Schineg- responses in all scenarios followed by the continental
ger et al., 2012). Accordingly, migration barriers may realm in the 2050s and the alpine region in the 2080s.
impede the distribution pathways of fish (Hein et al., In the Mediterranean, thermal stress will increase due
2011). The role of migration barriers in shaping the to higher temperatures (Giorgi & Lionello, 2008;
future structure of fish assemblages was not explicitly Filipe et al., 2013). In addition, prolonged droughts
considered in our analyses which imposes a restric- and aquatic habitat isolation will deteriorate fish
tion. The reasons for that were that, on the one hand, a habitat quality (Almodóvar et al., 2012; Filipe et al.,
sound data basis to evaluate migration barriers on site- 2012). Moderate changes in the thermal regime lead to
grain simply is not available; and, on the other hand, general lower responses in the boreal realm. The
the future state of the longitudinal connectivity in alpine region showed especially strong impacts for the
rivers is quite unclear. Legislative demands of the 2080s indicating less vulnerability to moderate cli-
WFD which should enable free migration within the mate changes but high vulnerability to stronger shifts
next decade seem contrary to further human uses (e.g. as well as a potential tipping point where the
increasing demands for renewable energy and water assemblage structure shifts abruptly.
supply), thereby hardly allowing for reliable assump- In this context, the relevance of the applied future
tions on the future state of longitudinal connectivity in scenarios should be mentioned which is underlined by
rivers. Nonetheless, migration of species is funda- a diversification of the proportion of impacted sites
mentally linked to time (Radinger et al., 2013) according to different future pathways. Besides the
relativising the role of barriers or of species-specific temperature increase, the temperature range governed
dispersal abilities in the long term. the response of FiZI. In the 2050s, scenario B1 showed
Prior to extirpations or species migrations, assem- the lowest responses in all clusters and biomes. In
blage structure may change due to abundance shifts respect of the clusters, this also held true in the 2080s
among the present species. Such shifts are also except for LC5. In the biomes, scenario A2 had the
captured by FiZI and will not occur in a uniform lowest portion of significantly affected sites in the
way along the longitudinal gradient of rivers. In the boreal, in the mediterranean and in the Atlantic realm
2080s, our results indicated the strongest fish assem- in the 2080s.
blage changes in rhithral river reaches (LC1–LC3)
where further migration is naturally limited. This is in
line with Buisson et al. (2013) who found functional Conclusion
changes especially in the upper part of the longitudinal
gradient for French rivers. Also Logez & Pont (2012) A global process like climate change requires impact
found species with intolerance to habitat degradation analyses on large scales to highlight the general trend of
and high oxygen needs, which are typical for rhithral effects as well as regional differences. The identifica-
sections, to be especially exposed to climate change tion of potential alterations in assemblage structures
impacts. facilitates the definition of conservation priorities and
Climatic conditions will limit available habitats for areas with high vulnerabilities. Improved knowledge on
cold-water species (Isaak et al., 2012). In turn, the future trends in assemblage structures is needed for

123
Hydrobiologia

macro-ecologists, conservation biologists as well as Wolter, T. Klefoth, B. Halasi-Kovacs, P. Gabor, G. Maio, E.

environmental managers as altered assemblage struc- Marconato, T. Virbickas, T. Buijse, M. Beers, P. Debowski, P.
Prus, W. Wisniewolski, J. S. Santos, P. Segurado, K. Battes, K.
tures due to climate change have highly important Battes, D. Garcia de Jalon, J. Gortazar, J. Solana, P. Bohman, U.
implications for assessment and conservation issues Beier, B. Seers, J. Pettersson, A. Peter, E. Schager, I. Cowx and
(Wehrly et al., 2003; Logez & Pont, 2012). This study R. Noble. Finally, we thank two anonymous reviewers for
underlines the importance of large-scale analyses valuable suggestions to improve the manuscript.
regarding climate change impacts which may counter-
act amelioration measures as demanded by the WFD. References
Furthermore, climate change may undermine present
assessment methods (Logez & Pont, 2012). Hence, Aarts, B. G. W. & P. H. Nienhuis, 2003. Fish zonations and
further monitoring especially of river reaches showing guilds as the basis for assessment of ecological integrity of
reference conditions (Schmutz et al., 2007) is essential large rivers. Hydrobiologia 500: 157–178.
Almodóvar, A., G. G. Nicola, D. Ayllón & E. Benigni, 2012.
to keep track of climate change impacts. Moreover, Global warming threatens the persistence of Mediterranean
time series seem especially valuable to investigate brown trout. Global Change Biology 18: 1549–1560.
ongoing changes. Belliard, J., P. Boet & E. Tales, 1997. Regional and longitudinal
To increase the accuracy of future studies, addi- patterns of fish community structure in the Seine River
basin, France. Environmental Biology of Fishes 50:
tional variables should be considered describing flow 133–147.
regime alterations as well as extreme events (e.g. heat Beyer, H. L., 2012. Geospatial Modelling Environment. Spatial
periods, droughts, temporal shift of floods). Further- Ecology, http://www.spatialecology.com/gme.
more, to draw conclusions about trends in assemblage Bosshard, T., M. Carambia, K. Goergen, S. Kotlarski, P. Krahe,
M. Zappa & C. Schär, 2013. Quantifying uncertainty
structures, analyses of temporal stability of fish sources in an ensemble of hydrological climate-impact
assemblages on large scales and over longer periods projections. Water Resources Research 49: 1523–1536.
may provide valuable insights. In respect of changing Brosse, S., 2001. Utilisation of non-supervised neural networks
species distributions, the roles of migration barriers and principal component analysis to study fish assem-
blages. Ecological Modelling 146: 159–166.
deserve a better consideration, especially if shorter Buisson, L. & G. Grenouillet, 2009. Contrasted impacts of cli-
time scales are investigated. However, the antagonism mate change on stream fish assemblages along an envi-
of WFD demands on river network connectivity and ronmental gradient. Diversity and Distributions 15:
the growing demands on renewable energy establish a 613–626.
Buisson, L., L. Blanc & G. Grenouillet, 2008a. Modelling
magnitude of potential future settings as to how stream fish species distribution in a river network: the
migration pathways will develop. relative effects of temperature versus physical factors.
Even though our study has elucidated potential Ecology of Freshwater Fish 17: 244–257.
changes in assemblage structures due to climate Buisson, L., W. Thuiller, S. Lek, P. Lim & G. Grenouillet,
2008b. Climate change hastens the turnover of stream fish
change, the next stage of analysing should tackle the assemblages. Global Change Biology 14: 2232–2248.
complex interactions between multiple factors (i.e. Buisson, L., G. Grenouillet, S. Villéger, J. Canal & P. Laffaille,
cumulative effects of anthropogenic pressures and 2013. Toward a loss of functional diversity in stream fish
climate change) incorporating human-impacted river assemblages under climate change. Global Change Biol-
ogy 19: 387–400.
stretches with the aim to untangle climate change and Busby, J. R., 1991. BIOCLIM—a bioclimate analysis and pre-
human-induced pressures. dictionsystem. Plant Protection Quarterly 6: 8–9.
Caissie, D., 2006. The thermal regime of rivers: a review.
Acknowledgements The authors specially thank A. Schmidt- Freshwater Biology 51: 1389–1406.
Kloiber, D. Markovic, C. Mielach, R. Schinegger and C. Chambers, J. M. & T. Hastie, 1992. Statistical Models.
Trautwein for their valuable inputs to the manuscript as well as S. Wadsworth & Brooks/Cole, Pacific Grove: 608.
T. Cassidy and D. Thornbrugh for language corrections. The Cohen, J., P. Cohen, S. West & L. S. Aiken, 2003. Applied
study was funded by the EU-project BioFresh–Biodiversity of Multiple Regression/Correlation Analysis for the Behav-
Freshwater Ecosystems: Status, Trends, Pressures, and ioral Sciences. Applied Multiple Regression Correlation
Conservation Priorities (Contract 226874, 7th Framework Analysis for the Behavioral Sciences. Lawrence Erlbaum,
Programme). Parts of this research were supported by the EU- Mahwah: 703.
project EFI? (Contract 044096, 6th Framework Programme). Comte, L., & G. Grenouillet, 2013. Do stream fish track climate
Special thanks are due to all project partners and institutions that change? Assessing distribution shifts in recent decades.
provided data for the EFI? project, especially: G. Haidvogl, T. Ecography, http://doi.wiley.com/10.1111/j.1600-0587.
Vehanen, T. Sutela, D. Pont, P.Bady, M. Logez, J. Belliard, C. 2013.00282.x.

123
 Hydrobiologia

Comte, L., L. Buisson, M. Daufresne & G. Grenouillet, 2012. Hansen, M. M., D. E. Ruzzante, E. E. Nielsen, D. Bekkevold &
Climate-induced changes in the distribution of freshwater K.-L. D. Mensberg, 2002. Long-term effective population
fish: observed and predicted trends. Freshwater Biology sizes, temporal stability of genetic composition and
58: 625–639. potential for local adaptation in anadromous brown trout
Dudgeon, D., A. H. Arthington, M. O. Gessner, Z.-I. Kawabata, (Salmo trutta) populations. Molecular Ecology 11:
D. J. Knowler, C. Lévêque, R. J. Naiman, A.-H. Prieur- 2523–2535.
Richard, D. Soto, M. L. J. Stiassny & C. A. Sullivan, 2006. Hari, R. E., D. M. Livingstone, R. Siber, P. Burkhardt-Holm
Freshwater biodiversity: importance, threats, status and & H. Guttinger, 2006. Consequences of climatic change
conservation challenges. Biological Reviews of the Cam- for water temperature and brown trout populations in
bridge Philosophical Society 81: 163–182. Alpine rivers and streams. Global Change Biology 12:
Dussling, U., R. Berg, H. Klinger & C. Wolter, 2004. Assessing 10–26.
the ecological status of river systems using fish assem- Haunschmid, R., G., Wolfram, T., Spindler, W., Honsig-Erl-
blages. In Steinberg, C., W. Calmano, H. Klapper & R.-D. enburg, R., Wimmer, A., Jagsch, E., Kainz,.K., Hehenw-
Wilken (eds), Handbuch Angewandte Limnologie. arter, B., Wagner, R., Konecny, R., Riedmüller, G., Ibel,
Ecomed Verlagsgruppe, Lech. B., Sasano, & N. Schotzko, 2006. Erstellung einer fis-
Eaton, J. G. & R. M. Scheller, 1996. Effects of climate warming chbaiserten Typologie österreichischer Fließgewässer so-
on fish thermal habitat in streams of the United States. wie einer Bewertungsmethode des fischökologischen
Limnology and Oceanography American Society of Lim- Zustandes gemäß EU-Wasserrahmenrichtlinie, Schriften-
nology and Oceanography 41: 1109–1115. reihe des BAW, Band 23, Wien: 105 pp.
Elliott, J. M., 2000. Pools as refugia for brown trout during two Hein, C. L., G. Öhlund & G. Englund, 2011. Dispersal through
summer droughts: trout responses to thermal and oxygen stream networks: modelling climate-driven range expan-
stress. Journal of Fish Biology 56: 938–948. sions of fishes. Diversity and Distributions 17: 641–651.
ESRI. 2011. Environmental Systems Research Institute, ArcGIS Heino, J., 2008. Regional gradient analysis of freshwater biota:
Desktop, Version 10.0, ArcInfo license. Redlands, do similar biogeographic patterns exist among multiple
California. taxonomic groups? Journal of Biogeography 28: 69–76.
Esselman, P. C. & J. D. Allan, 2010. Relative influences of Heino, J., 2011. A macroecological perspective of diversity
catchment- and reach-scale abiotic factors on freshwater patterns in the freshwater realm. Freshwater Biology 56:
fish communities in rivers of northeastern Mesoamerica. 1703–1722.
Ecology of Freshwater Fish 19: 439–454. Heino, J., R. Virkkala & H. Toivonen, 2009. Climate change and
Filipe, A. F., J. E. Lawrence & N. Bonada, 2012. Vulnerability freshwater biodiversity: detected patterns, future trends
of stream biota to climate change in mediterranean climate and adaptations in northern regions. Biological Reviews of
regions: a synthesis of ecological responses and conser- the Cambridge Philosophical Society 84: 39–54.
vation challenges. Hydrobiologia. doi:10.1007/s10750- Hijmans, R. J., S. E. Cameron, J. L. Parra, P. G. Jones & A.
012-1244-4. Jarvis, 2005. Very high resolution interpolated climate
Filipe, A. F., D. Markovic, F. Pletterbauer, C. Tisseuil, A. De surfaces for global land areas. International Journal of
Wever, S. Schmutz, N. Bonada & J. Freyhof, 2013. Fore- Climatology 25: 1965–1978.
casting fish distribution along stream networks: brown Horwitz, R. J., 1978. Temporal variability patterns and the
trout (Salmo trutta) in Europe. Diversity and Distributions distributional patterns of stream fishes. Ecological Mono-
19: 1059–1071. graphs Ecological Monographs 48: 307–321.
Fox, J. & G. Monette, 1992. Generalized collinearity diagnos- Huet, M., 1959. Profiles and biology of western European
tics. Journal of the American Statistical Association 87: streams as related to fish management. Transactions of the
178–183. American Fisheries Society 88: 155–163.
Frissell, C. A., W. J. Liss, C. E. Warren & M. D. Hurley, 1986. A Hulme, P. E., 2005. Adapting to climate change: is there scope
hierarchical framework for stream habitat classification: for ecological management in the face of a global threat?
viewing streams in a watershed context. Environmental Journal of Applied Ecology 42: 784–794.
Management 10: 199–214. Ibarra, A. A., Y.-S. Park, S. Brosse, Y. Reyjol, P. Lim & S. Lek,
Fritsch, A., 1872. Die Wirbeltiere Böhmens. Ein Verzeichnis 2005. Nested patterns of spatial diversity revealed for fish
aller bisher in Böhmen beobachteten Säugetiere, Vögel, assemblages in a west European river. Ecology of Fresh-
Amphibien und Fische. Development. Archiv für die water Fish 14: 233–242.
Landesdurchforschung von Böhmen: 152 pp. Illies, J., 1961. Versuch einer allgemeinen biozönotischen Gli-
Gasith, A. & V. H. Resh, 1999. Streams in Mediterranean cli- ederung der Fließgewässer. Internationale Revue der ge-
mate regions: abiotic influences and biotic responses to samten Hydrobiologie 46: 205–213.
predictable seasonal events. Annual Review of Ecology IPCC, 2007. Climate change 2007: synthesis report contribution
and Systematics 30: 51–81. of working groups I, II and III to the fourth assessment
Giorgi, F. & P. Lionello, 2008. Climate change projections for report of the intergovernmental panel on climate change. In
the Mediterranean region. Global and Planetary Change Pachauri, R. K. & A. Reisinger (eds), Core Wrting Team.
63: 90–104. IPCC, Geneva.
Grenouillet, G., L. Buisson, N. Casajus & S. Lek, 2010. Isaak, D. J., S. Wollrab, D. Horan & G. Chandler, 2012. Climate
Ensemble modelling of species distribution: the effects of change effects on stream and river temperatures across the
geographical and environmental ranges. Ecography 34: northwest U.S. from 1980 to 2009 and implications for
9–17. salmonid fishes. Climatic Change 113: 499–524.

123
Hydrobiologia

Jackson, D. A., P. R. Peres-Neto & J. D. Olden, 2001. What O’Donnell, M. S. & D. A. Ignizio, 2012. Bioclimatic predictors
controls who is where in freshwater fish communities—the for supporting ecological applications in the conterminous
roles of biotic, abiotic, and spatial factors. Canadian United States. U.S. Geological Survey Data Series 691: 10.
Journal of Fisheries and Aquatic Sciences 58: 157–170. Piñeiro, G., S. Perelman, J. P. Guerschman & J. M. Paruelo,
Jacquez, J. A. & F. J. Mather, 1968. Linear regression with non- 2008. How to evaluate models: observed vs. predicted or
constant, unknown error variances: sampling experiments predicted vs. observed? Ecological Modelling 216:
with least squares, weighted least squares and maximum 316–322.
likelihood estimators. Biometrics 24: 607–626. Poff, N. L., 1997. Landscape filters and species traits: towards
Jungwirth, M., Haidvogl, G., Moog, O., Muhar, S. & S. Sch- mechanistic understanding and prediction in stream ecol-
mutz, 2003. Angewandte Fischökologie an Fließgewäss- ogy. Journal of the North American Benthological Society
ern, Facultas, Wien: 552 pp, ISBN 3-8252-2113-X. 16: 391–409.
Lamouroux, N., J.-M. Olivier, H. Persat, M. C. Pouilly, Y. Poff, N. L. & J. K. H. Zimmerman, 2010. Ecological responses
S. Souchon & B. Statzner, 1999. Predicting community to altered flow regimes: a literature review to inform the
characteristics from habitat conditions: fluvial fish and science and management of environmental flows. Fresh-
hydraulics. Freshwater Biology 42: 275–299. water Biology 55: 194–205.
Lasne, E., B. Bergerot, S. Lek & P. Laffaille, 2007. Fish zona- Pont, D., B. Hugueny, U. Beier, D. Goffaux, A. Melcher, R.
tion and indicator species for the evaluation of the eco- Noble, C. Rogers, N. Roset & S. Schmutz, 2006. Assessing
logical status of rivers: example of the Loire basin river biotic condition at a continental scale: a European
(France). River Research and Applications 23: 877–890. approach using functional metrics and fish assemblages.
Legendre, P. & L. Legendre, 2012. Numerical Ecology, 3rd ed. Journal of Applied Ecology 43: 70–80.
Elsevier, New York: 1006. Pont, D., B. Hugueny & T. Oberdorff, 2005. Modelling habitat
Lehner, B., P. Döll, J. Alcamo, T. Henrichs & F. Kaspar, 2006. requirement of European fishes: do species have similar
Estimating the impact of global change on flood and responses to local and regional environmental constraints?
drought risks in Europe: a continental, integrated analysis. Canadian Journal of Fisheries and Aquatic Sciences 62:
Climatic Change 75: 273–299. 163–173.
Logez, M., & D. Pont, 2012. Global warming and potential shift R Development Core Team, 2011. R: a language and environ-
in reference conditions: the case of functional fish-based ment for statistical computing. R Foundation for Statistical
metrics. Hydrobiologia, 10.1007/s10750-012-1250-6. Computing, Vienna. http://www.r-project.org.
Logez, M., P. Bady & D. Pont, 2012. Modelling the habitat Radinger, J., J. Kail, & C. Wolter, 2013. FIDIMO—a free and
requirement of riverine fish species at the European scale: open source GIS based dispersal model for riverine fish.
sensitivity to temperature and precipitation and associated Ecological Informatics, http://linkinghub.elsevier.com/
uncertainty. Ecology of Freshwater Fish 21: 266–282. retrieve/pii/S1574954113000563.
Luterbacher, J., D. Dietrich, E. Xoplaki, M. Grosjean & H. Rahel, F. J. & W. A. Hubert, 1991. Fish assemblages and habitat
Wanner, 2004. European seasonal and annual temperature gradients in a rocky mountain-great plains stream: biotic
variability, trends, and extremes since 1500. Science 303: zonation and additive patterns of community change. Trans-
1499–1503. actions of the American Fisheries Society 120: 319–332.
Lyons, J., 1989. Correspondence between the distribution of fish Reyjol, Y., A. Compin, A. A. Ibarra & P. Lim, 2003. Longitu-
assemblages in Wisconsin streams and Omernik’s ecore- dinal diversity patterns in streams: comparing invertebrates
gions. American Midland Naturalist 122: 163–182. and fish communities. Archiv für Hydrobiologie 157:
Mallows, C. L., 1973. Some comments on C p. Technometrics 525–533.
15: 661–675. Reyjol, Y., B. Hugueny, D. Pont, P. G. Bianco, U. Beier, N.
Matulla, C., S. Schmutz, A. Melcher, T. Gerersdorfer & P. Haas, Caiola, F. Casals, I. G. Cowx, A. Economou, M. T. Ferre-
2007. Assessing the impact of a downscaled climate ira, G. Haidvogl, R. Noble, A. de Sostoa, T. Vigneron & T.
change simulation on the fish fauna in an Inner-Alpine Virbickas, 2007. Patterns in species richness and endemism
River. International Journal of Biometeorology 52: of European freshwater fish. Global Ecology and Bioge-
127–137. ography 16: 65–75.
Miller, A., 2002. Subset Selection in Regression. Technomet- Roberts, J. H. & N. P. Hitt, 2010. Longitudinal structure in
rics. Chapman and Hall, London: 256. temperate stream fish communities: evaluating conceptual
Miranda, L. E. & S. W. Raborn, 2000. From zonation to con- models with temporal data. American Fisheries Society
nectivity: fluvial ecology paradigms of the 20th century. Symposium 73: 281–289.
Polskie Archivum Hydrobiologii 47: 5–19. Rodriguez, M. A. & P. Magnan, 1995. Application of multi-
Nakicenovic, N. & R. Swart, 2000. IPCC Special Report on variate analyses in studies of the organization and structure
Emissions Scenarios (SRES). Cambridge University Press, of fish and invertebrate communities. Aquatic Sciences-
Working Group III of the Intergovernmental Panel on Research Across Boundaries 57: 199–216.
Climate Change IPCC: 570 pp. Roeckner, E., G. Bäuml, L. Bonaventura, R. Brokopf, M. Esch, M.
Nix, H. A., 1986. A biogeographic analysis of Australian elapid Giorgetta, S. Hagemann, I. Kirchner, L. Kornblueh, E.
snakes. In Longmore, R. (ed.), Atlas of Elapid Snakes of Manzini, A. Rhodin, U. Schlese, U. Schulzweida & A.
Australia. Australian Government Publishing Service, Tompkins, 2003. The atmospheric general circulation model
Canberra: 4–15. ECHAM5. Max Planck Institute for Meteorology. Deutsches
O’Brien, R. M., 2007. A caution regarding rules of thumb for Klimarechenzentrum, Modellbetreuungsgruppe, Hamburg:
variance inflation factors. Quality & Quantity 41: 673–690. 140.

123
 Hydrobiologia

Schager, E., & A. Peter, 2004. Methoden zur Untersuchung und Thienemann, A., 1925. Die Binnengewässer Mitteleuropas.
Beurteilung der Fliessgewässer. Fische Stufe F. BUWAL. Schweizerbart, Stuttgart: 255.
Mitteilungen zum Gewässerschutz Nr. 44: 62 pp. Thorp, J. H., M. C. Thoms & M. D. Delong, 2006. The riverine
Schinegger, R., C. Trautwein, A. Melcher & S. Schmutz, 2012. ecosystem synthesis: biocomplexity in river networks
Multiple human pressures and their spatial patterns in across space and time. River Research and Applications 22:
European running waters. Water and Environment Journal 123–147.
26: 261–273. Vannote, R. L., G. W. Minshall, J. R. Sedell & C. E. Cushing,
Schmutz, S., M. Kaufmann, B. Vogel, & M. Jungwirth, 2000a. 1980. The river continuum concept. Canadian Journal of
Methodische Grundlagen und Beispiele zur Bewertung der Fisheries and Aquatic Sciences 37: 130–137.
fischökologischen Funktionsfähigkeit Österreichischer Vogt, J., P. Soille, R. Colombo & M. L. Paracchini, 2007.
Fließgewässer. Vienna: 241 pp. Development of a pan-European river and catchment
Schmutz, S., M. Kaufmann, B. Vogel, M. Jungwirth & S. Mu- database. Data Processing 270: 121–144.
har, 2000b. A multi-level concept for fish-based, river- Ward Jr, J. H., 1963. Hierarchical grouping to optimize an
type-specific assessment of ecological integrity. Hydrobi- objective function. Journal of the American statistical
ologia 422–423: 279–289. association 58: 236–244.
Schmutz, S., I. G. Cowx, G. Haidvogl & D. Pont, 2007. Fish-based Webb, B. W., D. M. Hannah, R. D. Moore, L. E. Brown & F.
methods for assessing European running waters: a synthesis. Nobilis, 2008. Recent advances in stream and river tem-
Fisheries Management and Ecology 14: 369–380. perature research. Hydrological Processes 22: 902–918.
Schwarz, G., 1978. Estimating dimension of a model. Annals of Wehrly, K. E., M. J. Wiley & P. W. Seelbach, 2003. Classifying
Statistics 6: 461–464. regional variation in thermal regime based on stream fish
Spearman, C., 1904. The proof and measurement of association community patterns. Transactions of the American Fish-
between two things. The American Journal of Psychology eries Society 132: 18–38.
15: 72–101. Wold, S., 1976. Pattern recognition by means of disjoint prin-
Steel, R. G. D. & J. H. Torrie, 1960. Principles and Procedures of cipal components models. Pattern Recognition 8: 127–139.
Statistics. Principles and Procedures of Statistics.
McGraw-Hill, New York: 287.

123