Content uploaded by Piet F M Verdonschot
Author content
All content in this area was uploaded by Piet F M Verdonschot
Content may be subject to copyright.
Hydrobiologia 516: 173–189, 2004.
D. Hering, P.F.M. Verdonschot, O. Moog & L. Sandin (eds), Integrated Assessment of Running Waters in Europe.
© 2004 Kluwer Academic Publishers. Printed in the Netherlands.
173
Towards a multimetric index for the assessment of Dutch streams using
benthic macroinvertebrates
Hanneke E. Vlek, Piet F. M. Verdonschot & Rebi C. Nijboer
Alterra, Green World Research, P.O. Box 47, 6700 AA Wageningen, The Netherlands
Fax: +31(0)317-424988. E-mail: hanneke.vlek@wur.nl
Key words: streams, assessment, macroinvertebrates, AQEM, multimetric index, multivariate analysis, the
Netherlands
Abstract
This study describes the development of a macroinvertebrate based multimetric index for two stream types, fast and
slow running streams, in the Netherlandswithin the AQEM project. Existing macroinvertebratedata (949 samples)
were collected from these stream types from all over the Netherlands. All sites received a ecological quality
(post-)classification ranging from 1 (bad status) to 4 (good status) based on biotic and abiotic variables, using a
combination of multivariate analysisand expert-judgement.A number of bioassessmentmetrics was tested for both
stream types (fast and slow running streams) to examine their power to discriminate between streams of different
ecological quality within each stream type. A metric was selected for inclusion in the final multimetric index when
there was no overlap of the 25th and 75th percentile between one (or more) ecological quality class(es). Out of all
metrics tested, none could distinguish between all four ecological quality classes without overlap of the 25th and
75th percentile between one or more of the classes. Instead, metrics were selected that could distinguish between
one (or more) ecological quality class(es) and all others. Finally, 10 metrics were selected for the assessment of slow
running streams and 11 metrics for the assessment of fast running streams. Class boundaries were established, to
make the assignment of scores to the individual metrics possible. The class boundaries were set at the 25th and/or
75th percentile of the individual metric values. The individual metrics were combined into a multimetric index.
Calibration showed that 67% of the samples from slow running streams and 65% of the samples from fast running
streams were classified in accordance to their post-classification. In total, only 8% of the samples differed more
than one quality class from the post-classification. The multimetric index was validated with data collected in the
Netherlands from 82 sites for the purpose of the AQEM project. Validation showed that 54% of the streams were
classified correctly.
Introduction
Since the beginning of this century a wide variety of
methods for the biological assessment of streams has
been developed. In practice, macroinvertebrates are
the most commonly used organism group for assessing
water quality (Hawkes, 1979; Hellawell, 1986). With
their Saprobien system Kolkwitz & Marsson (1909)
were the first in Europe to introduce the concept of
organisms as indicators of environmental condition.
Since its introduction the Saprobien system has been
extended and revised by numerous European ecolo-
gists (Liebmann, 1951; Sládeˇ
cek, 1965). While in
Germany, the Netherlands and the Czech Republic
the focus was mainly on the improvement of the Sap-
robien system, in countries like Belgium, France and
the UK ‘score systems’ were developed. Score sys-
tems, like the Trent Biotic Index (Woodiwiss, 1964)
and the Indice Biotique (Tuffery & Verneaux, 1968),
occurred in the 1960s and followed the introduction of
the first diversity indices in the 1940s. More recently
multivariate approaches, like RIVPACS (Wright et al.,
1984) from the UK and EKOO (Verdonschot, 1990)
from the Netherlands, have been introduced.
Developments comparable to those in Europe
could be seen in the United States. In the 1980s a
174
multimetric index for fish (Karr, 1981) was intro-
duced in the United States, which was an approach
to assessment unknown by the European countries. A
multimetric index consists of a combination of sev-
eral metrics that each provides different ecological
information about the observed community and acts
as an overall indicator of the biological integrity of
a water resource. The strength of the multimetric
index is its ability to integrate information from in-
dividual, population, community and ecosystem level
(Karr & Chu, 1999). A multimetric index provides
detection capability over a broad range of stressors,
and provides a more complete picture of the eco-
system than single biological indicators do (Intergov-
ernmental Task Force on Monitoring Water Quality,
1993).
Rosenberg & Resh (1993) listed seven different
approaches to assess streams by using macroinver-
tebrates: richness measures, enumerations, diversity
indices, similarity indices, biotic indices, functional
feeding-group measures, and the multimetric ap-
proach. In the Netherlands only biotic indices, fo-
cussed on the detection of organic pollution, have been
applied widely. Furthermore, multivariate approaches
are being developed (Verdonschot & Nijboer, 2000).
The first biotic indices applied in the Netherlands
were those developed by Kolkwitz & Marsson (1909)
and Sládeˇ
cek (1973). These were already existingsap-
robic indices, developed to detect organic pollution
affecting Mid European streams. It soon became clear
that Dutch streams often possess distinctive features,
which require a different approach to assessment. For
example, the current velocity in most Dutch streams is
considerably lower in comparison to streams in other
European countries. These experiences initiated the
development of an assessment system for organic pol-
lution of lowland streams (Moller Pillot, 1971). The
K135-index (Tolkamp & Gardeniers, 1971) was based
on the Moller Pillot system and was used for decades.
The mentioned biotic indices, in general, are lim-
ited to a single impact factor, namely organic pollu-
tion. The disadvantage of an index reflecting a single
aspect of the stream is that it may fail to reveal the
effects of other or of combined impact factors (Fore
et al., 1994; Barbour et al., 1996). This problem was
overcome with the introduction of EBEOSWA (ecolo-
gical assessment of running waters) (Stichting Toege-
past Onderzoek Waterbeheer, 1992). EBEOSWA is a
system for the biological assessment of Dutch streams.
At the moment EBEOSWA is the national standard.
EBEOSWA assesses more than one impact factor; as
such it can be qualified as a multimetric index. The
system considers metrics related to stream velocity,
saproby, trophy, functional feeding-groups and sub-
strate. The disadvantages of the system are that it gives
separate scores for each metric instead of one final
classification for a location, and the ecological status
of a water body is not determined by comparing the ac-
tual status of a water body with near-natural reference
conditions. Furthermore, EBEOSWA is based on data
collected in the 1980s. These data comprise mainly
impacted sites, and collection and identification was
not done in a standardised manner.
The Water Framework Directive (WFD) has led to
a demand for a ‘new’ Dutch assessment system. With
the implementation of the WFD every EU member
state is obligated to assess the effects of human activ-
ities on the ecological quality of all water bodies. The
criteria set by the WFD, to which assessment should
comply, are (European Commission, 2000):
•the use of different water quality elements: benthic
invertebrate fauna, phytoplankton, fish fauna, and
aquatic flora;
•the ecological status of a water body is determined
by comparing the biological community composi-
tion of the investigated water body with near-natural
reference conditions;
•it is based on a stream-type specific approach;
•the final classification of water bodies ranges from
5 (high status) to 1 (bad status).
The objective of this study is to develop and test a mul-
timetric index for Dutch streams based on macroinver-
tebrates that meets the criteria of the WFD.
Materials and methods
In this study two different data sets were used: (1)
an existing data set for the development of the mul-
timetric index and (2) a new data set for the validation
of the multimetric index. The application of both
data sets is discussed separately. A summary of the
different steps taken in the process of multimetric
development is shown in Fig. 1.
(1) Existing data
Data collection
For the development of the multimetric index no new
field data were collected. Instead, a procedure was set
up to gather existing data from regional water district
175
Figure 1. Diagram showing the different steps taken in multimetric development. Ovals respresent applied techniques and squares accomplished
results.
managers. The data had to comply with the following
criteria:
•sampling took place after 1990;
•samples were taken in a standardised manner sim-
ilar to the AQEM samples (see biological sampling
and laboratory processing of new data);
•information about environmental variables was
available.
After the selection of appropriate samples for the
data set a list of environmental variables was sent to
the water district managers. Experts considered the
environmental variables on the list relevant for ana-
lysis. The water district managers provided data for
quantitative, qualitative and nominal variables. This
resulted in a data set containing information about
macroinvertebrate fauna, macrophytes and environ-
mental variables for 949 samples taken in streams
from every region in the Netherlands. To assure that
the data set would contain samples from the whole
degradation spectrum an ‘a priori’ classification was
made (Conquest et al., 1994). This pre-classification
was solely based on observations in the field and
performed by different water district managers. For
selection of the metrics and development of the multi-
metric index the ‘a priori’ classification was replaced
176
by a less biased ‘a posteriori’ classification (post-
classification). Post-classification was considered less
biased for two reasons: (1) it was based on multivariate
analysis using data on macroinvertebrate community
composition and environmental variables and (2) final
classification was achieved by looking at all samples
in the data set using expert-judgement.
Both pre- and post-classification resulted in a qual-
ity class. In the context of this article classification
always refers to the process of determining the quality
class of a water body. A quality class is described as
a value ranging from 5 (high) to 1 (bad) that indicates
the ecological status (or the state of degradation) of a
water body.
Post-classification
Post-classification was based on multivariate analysis.
Multivariate analysis was used to develop a cenoty-
pology. For this study an existing cenotypology was
used, which was built from the existing data set in
another study. A cenotypology describes different wa-
ter types and their stages of degradation (Verdonschot
& Nijboer, 2000). A cenotype is a group of samples
with similar macroinvertebrate composition and en-
vironmental circumstances. Environmental variables
describing a cenotype can refer to natural circum-
stances (water type) or a certain degree of degradation.
For the purpose of developing the cenotypology and
classifying the sites the following steps were taken:
(1) The macroinvertebrate data and environmental
data were pre-processed. For each macroinvertebrate
sample the number of individuals per taxon was stand-
ardised to a total sample area of 1.25 m2. Samples
from the same location were not averaged, but treated
as separate samples. Prior to analysis it was necessary
to perform a taxonomic adjustment on the macroinver-
tebrate data to assure unambiguous data processing.
Differences in taxonomic level could otherwise later
prove to be the cause of differences between species
groups. In this study a weighed taxonomic adjustment
was applied. For this purpose, the number of samples
in which a taxon occurred was calculated (frequency).
The following criteria were used for taxonomic adjust-
ment:
•when a genus, apart from a few exceptions, was
identified to species level, the genus was removed
and the species were kept;
•when a genus was very abundant (frequency of oc-
currence of the genus >20% of all the species
belonging to this genus), we looked at the indicative
value of the genus as a whole and the indicative
value of the separate species. When there were
clear ecological differences between the species,
the species information were kept and the genus
was removed. In case the genus was very indicat-
ive and there were no real ecological differences
between species, the species were assigned to genus
level. This procedure can be illustrated with the
following example: a data set of 90 samples con-
taining 20 samples with Baetis sp, 4 samples with
Baetis tracheatus, 80 samples with Baetis vernus
and 6 samples with Baetis fuscatus. According to
the criterion mentioned above all species should be
assigned to genus level, because the frequency of
occurrence of the genus is 22% (20/90) of all the
species. However, in this case an exception is made.
The species level is kept and the genus removed,
because the different Baetis species each indicate
different environmental circumstances.
After taxonomic adjustment the macroinvertebrate
abundances of each sample were transformed into log-
arithmic classes (Preston, 1962; Verdonschot, 1990).
The list with values for the environmental vari-
ables, which came back from the water district man-
agers, was not complete for all samples. Environ-
mental variables, with missing values for more than
20% of the samples in the data set, were not included
in the analysis. Nominal variables were dealt with by
defining dummy variables (value 0 or 1). All envir-
onmental variables were log-transformed log (x+1),
except for pH and nominal variables, to minimise the
effect of extreme values on the results. In total 23 en-
vironmental variables were used for analysis (Table 1).
(2) The samples in the data set were clustered, based
on the macroinvertebrate data, using the program
FLEXCLUS (Van Tongeren, 1986). This program
aggregates samples into groups based on the Sørensen-
similarity ratio (Sørensen, 1948). The initial clustering
is optimised using relocative centroid sorting. The
number of resulting clusters depends on the chosen
threshold value.
(3) The samples were ordinated by detrended (canon-
ical) correspondence analysis (D(C) CA) using the
program CANOCO (Ter Braak, 1987). DCA was used
to determine the variation within the data set. Based on
the results of the DCA it was decided to use a unim-
odal technique (DCCA) for further analysis. DCCA is
an ordination based on both species and environmental
data. The program CANOCO offers different options
on how to present and analyse data. The choices made
177
Table 1. Environmental variables with numerical scale included in
multivariate analysis of the existing data
Variable name Category Numerical
scale
Profile natural transversal nominal
profile
meandering nominal
dam nominal
Season winter nominal
spring nominal
autumn nominal
summer nominal
Soil type clay nominal
loam nominal
peat nominal
sand nominal
Surrounding land use intensive agriculture nominal
natural nominal
urbanisation nominal
intensive pasture nominal
Substrate (%) CPOM quantitative
FPOM quantitative
gravel quantitative
clay quantitative
loam quantitative
silt quantitative
stones quantitative
branches quantitative
sand quantitative
Vegetation (% coverage) total quantitative
floating macrophytes quantitative
submerged quantitative
macrophytes
emerged quantitative
macrophytes
Hydrologic stream type permanent nominal
Bank fixation – nominal
Width (m) – quantitative
Depth (m) – quantitative
Seepage – nominal
Stream velocity (m s−1)– quantitative
Dissolved oxygen (mg l−1)– quantitative
Ammonium (mgN l−1)– quantitative
Kjehdal-N (mgN l−1)– quantitative
Nitrate (mgN l−1)– quantitative
Chloride (mg l−1)– quantitative
Ortho-phosphate (mgP l−1)– quantitative
Total phosphate (mgP l−1)– quantitative
Conductivity (µS) – quantitative
pH – quantitative
Temparature (˚C) – quantitative
Shading (%) – quantitative
in CANOCO will influence the result of the ordina-
tion. In this study the following options were selected:
•downweighting of rare species: reduces the influ-
ence of rare species on the analysis;
•inter-sample distance: optimises the position of the
samples in the ordination diagram;
•detrending by segments (DCA);
•detrending by 2nd order polynomals (DCCA);
•forward selection: enables the user to rank environ-
mental variables in their importance for determin-
ing the species data or for reducing a large set of
environmental variables.
All techniques are fully explained by Ter Braak & Šm-
ilauer (1998).
(4) The results of clustering and ordination were com-
bined in ordination diagrams. Clusters were, therefore
projected on the first two axes of the DCCA ordin-
ation diagrams. In an ideal situation, the samples of
one cluster were positioned closely together in the or-
dination diagram and showed no overlap with samples
of another cluster. Samples that did cause overlap
between clusters were examined further. The decision,
whether a sample was placed in another cluster or set
apart, was based on spatial separation on the third
and sometimes the fourth axes as well as upon the
macroinvertebrate community composition.
From the above, it can be deducted that a sample
group (cenotype) was established if the respective
group was clearly recognisable along an identified en-
vironmental gradient and thus had a specific macroin-
vertebrate community composition (Verdonschot &
Nijboer, 2000).
(5) Classification (or biological assessment) is only
meaningful when it is applied to regions having
a relative small range in environmental conditions
and a relative homogenous macroinvertebrate com-
munity composition under reference conditions (Bar-
bour et al., 1996; Karr & Chu, 1999). Based on the
cenotypology four different regions or major stream
types of reference conditions could be distinguished
(1) fast running streams (v>30 cm sec−1)(2)slow
running streams (v<30 cm sec−1) (3) periodic or
episodic streams and (4) (weak) acidic streams. These
four major stream types could be divided further based
on dimension into 15 stream types. The decision was
made to develop a separate multimetric index for each
stream type distinguished under reference conditions,
since natural environmental variables were affecting
macroinvertebrate community composition (Weigel,
178
2003). This was done to avoid selection of metrics
related to differences between streams under natural
circumstances, instead of metrics related to the extent
of degradation.
The establishment of cenotypes facilitated the as-
signment of quality classes to the sites. Because a cen-
otype is a group of samples with similar macroinver-
tebrate composition and environmental circumstances,
all sites belonging to the same cenotype were con-
sidered to be at the same stage of degradation. To de-
termine the degradation stage (or quality class) of each
cenotype the macroinvertebrate community compos-
ition and values for environmental variables of each
cenotype were used for interpretation with expert-
judgement. All environmental variables mentioned in
Table 1 were used to support classification except for
variables indicating natural features of stream types.
To facilitate the interpretation of the biological data
the preferences of the species (of each cenotype) for
microhabitat, dimension, current velocity and saprobic
conditions were determined with the help of an aute-
cological database (AQEM consortium, 2002). For
the same purpose, the following biotic characteristics
were calculated: locomotion types, functional feeding-
group types and trophic levels. Quality classes from
1 (bad) to 4 (good) were assigned to the sites, using
expert-judgement. Quality class 5 (high) was never
assigned to a site, because pristine or reference sites
have disappeared from the Dutch environment due to
extensive habitat degradation and organic pollution.
During classification, it became clear that not
enough sites representing each quality class were
available for each of the 15 stream types to develop
an index. Only the two major stream types, fast and
slow running streams, had enough sites representing
all quality classes with sufficient variation to develop a
sound index. For this reason, all samples from (weak)
acidic streams and periodic/episodic streams were re-
moved from the data set and the division into stream
types according to dimension was dropped. The de-
cision was made to develop two separate multimetric
indices; one for fast running streams and one for slow
running streams. Quality classes were assigned sep-
arately to the cenotypes of the slow and fast running
streams.
Metric selection
Rosenberg & Resh (1993) identified seven differ-
ent approaches for the assessment of streams based
on macroinvertebrates: richness measures, enumer-
ations, diversity indices, similarity indices, biotic
Figure 2. Example of (no) interquartile overlap for the metric values
between quality classes. Range bars show maximum and min-
imum values; boxes are interquartile ranges (25th percentile to 75th
percentile); small stripes represent medians.
indices, functional feeding-group measures and the
multimetric approach. Recently multimetric systems
with stressor-specific approaches have been developed
for many European river types (Brabec et al., 2004;
Buffagni et al., 2004; Ofenböck et al., 2004; Sandin
et al., 2004). In this study we considered a large num-
ber of metrics, more than hundred (Hering et al., 2004)
representing five of the above approaches. For an ex-
planation of the different metrics used in this article
see Hering et al. (2004).
Before selection was possible, the metric values for
each sample in the data set had to be calculated. These
calculations were based on an extended list of aute-
cological information of European macroinvertebrate
fauna.
In order to assess the ability of all metrics to dis-
criminate between the different stages of degradation,
graphical analysis using box-and-whisker plots was
applied. This method is similar to the methods de-
scribed by Barbour et al. (1996), Fore et al. (1996),
Blocksom et al. (2000) and Royer et al. (2001). Fore
et al. (1996) and Karr & Chu (1999) suggest that
graphical methods have fundamental advantages over
statistical techniques in this context. Graphs provide
more insight in the response of macroinvertebrates to
degradation. From a graph one can determine over
which range a metric is most sensitive and whether
a metric response is linear, unimodal or occurs at a
threshold level.
The calculated metric values and degradation stage
(= post-classification) for each sample from the data
179
set were combined in a box-and whisker-plot. A met-
ric was judged suitable for index development when
there was no interquartile overlap between one or more
quality classes in the box-and-whisker plot (Fig. 2).
This complies to Fore et al. (1996), who selected met-
ric as suited in case of no or little overlap between
classes and Barbour et al. (1996) and Royer et al.
(2001), who judged a metric as highly sensitive in case
of no overlap in the interquartile range. To determine
possible overlap in the interquartile range, the 25th
and 75th percentile were calculated for all metrics for
each quality class. Preferably, metrics were selected
that showed no interquartile overlap between all four
quality classes. If there was no other option metrics
that showed no interquartile overlap between one class
and all other classes were selected.
Multimetric index development
Not all metrics judged suited for index develop-
ment were actually used for this purpose. If possible,
suited metrics reflecting different quality aspects of
the macroinvertebrate community were selected. For
example, only one of the two saprobic indices that
met the test criteria was selected for multimetric index
development (Table 2). Because each metric reflects
its own quality aspect of the macroinvertebrate com-
munity and as a result might not be able to reveal the
effect of multiple stressors (Barbour et al., 1996), 2
to 4 metrics were selected per quality class. Finally,
class boundaries were established to make the assign-
ment of scores to the individual metrics possible. Class
boundaries were set at the 25th percentile and/or 75th
percentile of the metric values.
Calibration of the multimetric index
The multimetric index was calibrated with the exist-
ing data set. For this purpose, the quality class of all
samples was calculated with the multimetric index and
compared to the quality class derived through post-
classification. Two possible types of errors could occur
in making this comparison:
type I error: the calculated quality class for a sample
is lower than the quality class derived through post-
classification.
type II error: the calculated quality class for a sample
is higher than the quality class derived through post-
classification.
(2) New data
Site selection
New data were collected to validate the multimetric
index. These data were collected within the AQEM
project. See AQEM consortium (2002) for the meth-
odology applied. To assure sampling of the whole
degradation spectrum an ‘a priori’ classification of the
sites was made into the five quality classes used by
the WFD (European Commission, 2000). Sites were
selected and pre-classified by local water district man-
agers. For the validation of the multimetric index the
pre-classification was later replaced by a less biased
post-classification. In total the AQEM data set com-
posed 156 samples divided over 82 sites distributed
over all regions in the Netherlands. At each site also
environmental data were recorded. In total, informa-
tion on 230 environmental variables was collected.
Biological sampling and laboratory processing
The AQEM samples were taken between May 2000
and May 2001, partly by water district managers and
partly by Alterra. Most sites were sampled twice,
in spring and autumn. The coverage of each hab-
itat present at a sampling site was estimated before
sampling. For the collection of the samples a D-frame
dipnet(25or30cmwidewitha500µmmesh)was
used to collect a composite sample from several habit-
ats at each site. The sample was taken by pushing the
dip net through the upper part (2–5 cm) of the sub-
stratum. Each habitat was sampled overa distance that
ensured collection of most species present at the hab-
itat. All samples of mineral substrates were sampled in
the same ratio as their coverage in the stream and put
together in one bucket. The same procedure was re-
peated for the organic substrates. Mineral and organic
samples were kept apart. All habitats with less than 5%
coverage were sampled in only very small amounts,
just to collect any species that were not present in
the major habitats. After sampling, the buckets with
samples were transported to the laboratory and stored
in a refrigerator. The mineral and organic part of the
sample were kept separate during processing. The
samples were sieved using a 1000 and 350 µmsieve.
The coarse fraction (>1000 µm) and fine fraction
were kept separate during sorting. The samples were
sorted live by eye. If the coarse or fine fraction con-
tained over 500 individuals, subsamples of at least 500
individuals were sorted. Organisms were identified to
the lowest possible taxonomic level (species level for
almost all groups).
180
Table 2. Metrics that met the test criteria, metrics included in the multimetric index and their class boundaries for the slow running
streams
Metric Meets test criteria Class for Class boundaries
Class 4 Class 3 Class 2 Class 1 which the
metric is
selected as
indicator
German Saprobic Index (DIN 38 410) yes no no no –a
Saprobic Index (Zelinka & Marvan) yes no no no 4 <2.12
metapotamal [%] yes no no no –
hypopotamal [%] yes no no no 4 <0.55
metarhithral [%] yes no no no –
hyporhithral [%] yes no no no –
Shredders [%] yes no no no –
type Pel [%] yes no no no 4 <8.4
type Lit [%] yes no no no –
type Aka [%] yes no no no –
type RP [%] yes no no no 4 >29.4
type IN [%] yes no no no –
Gastropoda yes no no no –
hypopotamal [%]-EPT/OL [%] 3 <3.22 – >0.91band >1.3c
Gastropoda-EPT/OL [%] 3 <=6–>=2band >1.3d
No.ofEPT/OLtaxa nonoyesno2 <0.67
EPT/OL [%] no no yes no 2 <0.51
Grazers + scrapers/gatherers + filter feeders no no no yes 1 >2
Gastropoda [%] no no no yes 1 >9.92
a– indicates that the metric was not included in the multimetric index.
bClass boundary for the metric hypopotamal [%].
cClass boundary for the metric EPT/OL [%].
dClass boundary for the metric EPT/OL [%].
Collection of environmental data
Environmental data were collected at all AQEM sites.
The collection of environmental data and biological
sampling took place simultaneously. The environ-
mental data recorded for the purpose of the AQEM
project were not used for analysis in this study.
Post-classification
Post-classification of the samples in the new data set
was not based on multivariate analysis. Instead the
already developed cenotypologybased on the existing
data was used to classify the new samples. The follow-
ing steps were taken to classify the AQEM samples:
(1) The macroinvertebrate data from the coarse and
fine fraction were combined and standardised to a total
sample area of 1.25 m2. If there were more samples
from one location these samples were not combined to
form one sample, but they were treated like samples
from different locations. The macroinvertebrate data
were adjusted to the same taxonomic level as used for
the existing data. The macroinvertebrate abundances
were transformed into logarithmic classes (Preston,
1962; Verdonschot, 1990).
(2) The AQEM samples were classified using the pro-
gram ASSOCIA. ASSOCIA is a program originally
developed for the identification of plant communities,
but ASSOCIA can also be used to allocate macroin-
vertebrate samples to existing (ceno)types. For the
allocation of samples ASSOCIA uses both qualitative
and quantitative features of a sample in the form of the
maximum likelihood principle and a measure of dis-
tance. The maximum likelihood principle is based on a
calculation of probability; with the macroinvertebrate
species list the chance that the species composition
of a sample can be found in a cenotype is calcu-
lated. Final allocation takes place based on an index
that combines maximum likelihood and measure of
distance.
181
The AQEM samples were allocated to the cen-
otypes with ASSOCIA. The samples were allo-
cated to the cenotype with the lowest value for
the combined index, because the value of the com-
bined index increases with decreasing similarity. The
AQEM samples received the same classification as the
samples from the existing data set belonging to the
concerned cenotype.
Validation of the multimetric index
The multimetric index was validated with the new
AQEM data set. The process of validation was the
same as for calibration.
Results
Metric selection and multimetric index development
Box-and-whisker plots with metric scores were used
to depict the variability within each of the four quality
classes (4 = good, 3 = moderate, 2 = poor and1 = bad).
The metrics showed different kind of responses to de-
gradation: linear, unimodal, bimodal or at threshold
level (Fig. 3). In an ideal situation one metric can
distinguish between all quality classes based on the
interquartile range criterion (Fig. 4). In reality none
of the tested metrics could distinguish between all
quality classes. For this reason, metrics were selected
that could differentiate between one (or more) quality
classes and all others based on the interquartile range.
In case a metric can distinguish between one qual-
ity class and all others, the metric can be seen as an
‘indicator’ for this quality class. Figures 3A–C show
examples of metrics that fulfilled the criteria for met-
ric selection. The Saprobic Index, type rheophil (RP)
[%] and hypopotamal [%] are all metrics that show no
overlap between the 25th and 75th percentile of class
4 and all other classes (Figs 3A–C), therefore the met-
rics from Figures 3A–C can be used as ‘indicators’for
class 4.
For the slow running streams, 17 metrics showed
no overlap in the interquartile range for one class
(Table 2); 13 metrics for class 4, 2 metrics for class
2and2metricsforclass1(Table2).Forclass3all
metrics showed overlap in the interquartile range with
one or more classes. In using a combination of metrics
this problem was solved, where one metric couldn’t
differentiate between one class and all others a com-
bination of two metrics could. The first combination
of metrics (or combination metric) consisted of the
Figure 3. Examples of the distribution of metric values within the
four quality classes. All the metrics shown met the selection cri-
teria for at least one quality class. Range bars show maximum and
minimum values; boxes are interquartile ranges (25th percentile to
75th percentile); small stripes represent medians. (a) Unimodal; (b)
Exponential; (c) Linear.
182
Figure 4. Example of metric-response to degradation in an ideal
situation; no interquartile overlap of metric values between any
of the four quality classes. Range bars show maximum and min-
imum values; boxes are interquartile ranges (25th percentile to 75th
percentile); small stripes represent medians.
metric hypopotamal [%] and the metric EPT/OL [%]
(Fig. 5). Figure 5 shows that the metric hypopotamal
[%] can distinguish between class 3 on the one hand,
and class 4 and 1 on the other hand. After this distinc-
tion is made the metric EPT/OL [%] can distinguish
between class 2 and class 3 (Fig. 5). The second com-
bination consisted of the metric numberof Gastropoda
taxa and the metric EPT/OL [%] and was based on the
same principle.
From the 13 metrics that qualified for the identific-
ation of class 4 sites, only 4 metrics were selected for
multimetric index development (Table 2). The num-
ber of Gastropoda taxa was not selected because of
the small difference between class 4 and 2, based
upon only one taxon. Finally, 10 metrics associated
with stream velocity, sabrobic conditions, substrate
and zonation were selected includingtwo combination
metrics (Table 2).
For the fast running streams 8 metrics showed no
overlap in the interquartile range for one (or more)
classes (Table 3). For class 1, however all metrics
showed overlap in the interquartile range. Combin-
ation metrics were selected for class 1, similar to
class 3 for the slow running streams. In total 11 met-
rics were selected including three combination metrics
(Table 3).
After selection of the metrics class boundaries
were established (Tables 2 and 3). For the combination
metrics two class boundaries were established, one for
each metric (Fig. 5). With the establishment of class
boundaries scores could be assigned to the individual
metrics. When a metric value for a site lies within the
class boundaries (for a combination metric the values
Figure 5. Distrubution of metric values within the four quality
classes for the two metrics forming the combination metric hypo-
potamal [%]-EPT/OL explain [%] for slow running streams. Range
bars show maximum and minimum values; boxes are interquartile
ranges (25th percentile to 75th percentile); dotted lines represent
class boundaries.
for both metrics have to lie within the class boundar-
ies), the score is equal to the class the metric indicates
(equal to the value mentioned in column five of Table 2
or Table 3). For example, a site from a slow running
stream with a metric value of 0.43 for hypopotamal
[%] scores 4 for this metric (Table 2). When a met-
ric value lies outside the class boundary range the site
scores 0 for the respective metric. The scores for the
individual metrics were combined into the following
multimetric index:
Slow running streams
S=
T1∗1
2+T2∗1
2+T3∗1
2+T41
4
n1∗1
2+n2∗1
2+n3∗1
2+n4∗1
4
with:
S,finalscore;
T1, sum of scores for the individual metrics indicating
class 1;
T2, sum of scores for the individual metrics indicating
class 2;
183
Table 3. Metrics that met the test criteria, metrics included in the multimetric index and their class boundaries for the fast running
streams
Metric Meets test criteria Class for Class boundaries
Class 4 Class 3 Class 2 Class 1 which the
metric is
selected as
indicator
Saprobic Index (Zelinka & Marvan) yes no yes yes 4 <2.02
2>=2.46
1>2.27 −<2.46
metapotamal [%] yes no no no –a
hypopotamal [%] yes yes no no 4 <0.12
3>=0.12–<0.9
No. of taxa yes no no no 4 <=24
Passive filter feeders [%] no yes no no 3 >1.65
EPT/Ol [%] no yes yes no 3 >22
2<0.61
No.ofEPT/Oltaxa nonoyesno2 <0.71
Tricoptera [%] no no yes no 2 <0.23
EPT/Ol [%] - Gastropoda [%] 1 <16.4–>1.19band >=0.12c
EPT/Ol [%] - type RP [%] 1 <1.64–>1.19band <=53.3d
EPT/Ol [%] - type PEL [%] 1 <1.64–>1.19band >=4.36e
a– indicates that the metric was not included in the multimetric index.
bClass boundary for the metric EPT/OL [%].
cClass boundary for the metric Gastropoda [%].
dClass boundary for the metric type RP [%].
eClass boundary for the metric type PEL [%].
T3, sum of scores for the individual metrics indicating
class 3;
T4, sum of scores for the individual metrics indicating
class 4;
n1, number of indices indicating class 1;
n2, number of indices indicating class 2;
n3, number of indices indicating class 3;
n4, number of indices indicating class 4.
Fast running streams
S=
T1∗1
4+T2∗1
4+T3∗1
3+T41
3
n1∗1
4+n2∗1
4+n3∗1
3+n4∗1
3
with:
S,finalscore;
T1, sum of scores for the individual metrics indicating
class 1;
T2, sum of scores for the individual metrics indicating
class 2;
T3, sum of scores for the individual metrics indicating
class 3;
T4, sum of scores for the individual metrics indicating
class 4;
n1, number of indices indicating class 1;
n2, number of indices indicating class 2;
n3, number of indices indicating class 3;
n4, number of indices indicating class 4.
The intention of the multimetric index was to calcu-
late the mean of scores for the individual metrics. By
simply calculating the mean, however the fact that the
number of ‘indicator’ metrics differed between quality
classes would not be taken into account. For the slow
running streams, for example, class 4 was indicated by
four metrics and the other classes were indicated by
only 2 metrics. This means, that the chance a site will
score 4 is higher than the chance a site will score 3, 2
or 1. To correct for this disproportional distribution we
multiplied by 1
2(class 3, 2 and 1) and 1
4(class 4).
The score, calculated with the multimetric index,
was converted into a final quality class according to
Table 4.
Calibration of the multimetric index
First the multimetric index was calibrated using the
existing data set. Samples from cenotype 9, 14a, 14b,
184
Table 4. Class boundaries for the transformation of
the multimetric index score into the final quality
class
Quality class Score
5 (high status) Not applicable
4 (good status) ≥3.5 – ≤4
3 (moderate status) ≥2.5 – <3.5
2 (poor status) ≥1.5 – <2.5
1(badstatus) <1.5
16, 24a and 31 were often classified incorrect (Tables 5
and 6). All these cenotypes consisted of a low number
of samples (12 and less), except for cenotype 24a. For
the remaining cenotypes the percentage of correctly
classified samples varied between 48 and 100%.
Only a very low percentage of the samples (8% for
the slow running streams and 9% for the fast running
streams) deviated more than one class from the post-
classification (Figs 6, 7). Again, cenotype 9, 14a, 14b,
24a and 31 were an exception to this rule.
In total, 67% of the slow running streams and
65% of the fast running streams were classified cor-
rectly. The percentage type I and type II errors varied
between 19 and 15. Most errors occurred with the
classification of samples that received a quality class
3 during post-classification (Figs 6, 7).
Validation of the multimetric index
After calibration, the multimetric index was validated
with the new AQEM data set. In total, 54% of the
samples were classified correctly (Fig. 8). Most of
the samples that were not classified correctly, differed
only one quality class from the post-classification
(Fig. 8.). The percentage type I errors for the total data
set was 32. The percentage type II errors for the total
data set was 14.
Discussion
Classification of sites
Sites can be classified using either an ‘a priori’ or an ‘a
posteriori’ approach. In the context of this study an ‘a
priori’ classification or pre-classification is described
as a classification based on abiotic variables recorded
in the field (e.g., presence of point sources, presence of
eutrophication, missing of natural vegetation, etc.). An
‘a posteriori’ classification or post-classification is de-
scribed as a classification based on measured/recorded
abiotic variables and/or macroinvertebrate data. Clas-
sification based on solely abiotic variables was applied
by Thorne & Williams (1997), Barbour et al. (1996),
Fore et al. (1996) and many others. In this study a
combination of biotic and abiotic variables was used
for classification.
Classification using abiotic variables is a relatively
sound approach when only one dominant stressor in-
fluences a site. Classification of such sites can then
be based on abiotic variables related to this stressor.
However, often multiple stressors exert their influ-
ence on the macroinvertebrate community, and spe-
cific ‘cause-and-effect’ assessment may be difficult
(Intergovernmental Task Force on Monitoring Water
Quality, 1993). Especially in the Netherlands, where
habitat degradation and organic pollution (the most
important forms of stream degradation in the coun-
try) often go together, the role of each is difficult to
determine. Both habitat degradation and organic pol-
lution affected each stream sampled during this study
and ‘cause- and-effect’ could not be determined. Be-
cause the macroinvertebratecommunity reflects the in-
fluence of all stressors on its environment (Karr, 1999;
Karr & Chu, 2000) post-classification was largely
based on biotic variables. Since it was impossible to
separate the effects of habitat degradation and organic
pollution (no streams in the data set with the influence
of only one of the two stressors) the multimetric index
is not able to assess the effect of stressors separately
(in case of multiple stressors).
To facilitate classification, multivariate analysis
was used to develop a cenotypology. Based on the cen-
otypology, two stream types could be distinguished:
slow and fast running streams. The results of metric
selection indicated that the metrics responded differ-
ently to degradation for each of these stream types.
These findings comply with the findings of Resh et al.
(2000) who gives an overview of different studies that
examined the appropriateness of metrics in assessing
ecological quality of waters form different regions.
From this overview it appears that most metrics can’t
be applied in more than one region. As a result the
multimetric index consists of a different combination
of metrics for each stream type.
In an ideal situation the data set should have been
divided up to the point where all sites within one
stream type would differ only in their degree of de-
gradation (Fore et al., 1996). Despite the fact that
the data set was divided into two stream types, clas-
185
Table 5. Percentage type I and type II errors resulting from calibration with the existing data set for the slow running streams. A difference is
made between a deviation of one class from the post-classification or more
Cenotype Type I error (%)
(deviation of 1
class)
Type I error (%)
(deviation of 2 or 3
classes)
Type II error (%)
(deviation of 1
class)
Type II error (%)
(deviation of 2 or 3
classes)
Quality class
(post-
classification)
Number of
samples
3a43900423
24a50600418
24c23000422
2100004 6
1250003 4
19 19 22 0 0 3 36
26500003 2
992800312
10 1 0 17 0 2 72
13 0 0 20 0 2 10
15 0 0 13 0 2 8
600122166
14a 0 0 63 25 1 8
14b 0 0 17 83 1 6
31 0 0 40 60 1 5
Table 6. Percentage type I and type II errors resulting from calibration with the existing data set for the fast running streams. A difference is
made between a deviation of one class from the post-classification or more
Cenotype Type I error (%)
(deviation of 1
class)
Type I error (%)
(deviation of 2 or 3
classes)
Type II error (%)
(deviation of 1
class)
Type II error (%)
(deviation of 2 or 3
classes)
Quality class
(post-
classification)
Number of
samples
21 17 0 0 0 4 6
24a 50 33 0 0 4 18
24b 40 7 0 0 4 15
20 33 0 0 0 3 3
25 25 13 0 0 3 8
3b0 0 003 3
10 9 0 3 0 2 74
15 22 0 22 0 2 9
9005833112
16 0 0 100 0 1 2
19 0 0 19 14 1 36
sification was still difficult due to abiotic differences
in the data set other than differences relating to de-
gradation. Unfortunately, the natural factor width was
still playing an important role in the explanation of
macroinvertebrate community composition between
sites within the two stream types. Further deviation
of the data set according to dimension or other steer-
ing abiotic variables was not an option, because then
there wouldn’t be enough sites representing all qual-
ity classes with sufficient variation within each stream
type to develop a sound index.
Multimetric index development
A combination of multivariate analysis (MVA) and
multiple metrics was used for the development of the
multimetric index. A number of studies, Reynoldson
et al. (1997), Bailey et al. (1998), Milner & Oswood
(2000), indicates that in biological assessment mul-
186
Figure 6. Calibration results, final calculated quality class versus quality class based on post-classification for slow running streams from the
existing data set.
tivariate techniques are moreprecise and accurate than
multimetric indices. However, multivariate techniques
are complex and difficult to communicate to policy
makers (Fore et al., 1996). So, instead of develop-
ing an assessment system completely based on MVA,
MVA was only used to set post-classification. Post-
classification was followed by metric selection. The
results of metric selection showed that not many met-
rics could meet the selection criteria. None of the met-
rics could differ between all four quality classes and
most were only capable of indicating one class. This
poor result can have different causes. First, mistakes in
the post-classification due to abiotic differences in the
data set could have played an important role. Second,
the autecological data behind the metrics could have
been of importance. These autecological data com-
prise indicator values for current velocity, acidity, etc.
In determining these indicator values data from all
over Europe were used, this means the indicator values
can deviate from the Dutch optima. Third, it might just
not be feasible to differentiate between four ecological
quality classes based on the biological metrics tested
in this study.
After metric selection class boundaries were set.
Class boundaries for individual metrics can be set
in two different ways: (1) based on statistical rules
(2) based on an ecological response to degradation.
Examples of the first option are:
•dividing the 95-percentile of all sites by four (Ohio
Enviromental Protection Agency, 1987; DeSohn,
1995);
•the 50- and 10-percentile of all reference locations
(Roth et al., 1997);
•dividing the 25- or 75-percentile of all reference
locations (Barbour et al., 1996; Royer et al., 2001).
In this study, the second option to set class boundar-
ies was chosen, because the first approach has a lot
of disadvantages. First, selection based on statistical
rules assumes each metric responds in the exact same
way to degradation, while in fact each metric has its
own ecological response. Second, a metric can only
be qualified as suited when it shows a linear response
to environmentaldegradation, while the secondoption
doesn’t rule out a metric response that is unimodal,
bimodal or occurs at threshold level.
In the final step of multimetric development, one
of the criteria set by the WFD was not followed. Ac-
cording to this criterion the calculation of ecological
quality class should be based on the deviation from
the reference condition. This criterion was ignored,
because reference sites were not present in the Neth-
erlands and it was not possible within the scope of
this research to construct valid hypothetical reference
situations. However, when descriptions of reference
conditions of Dutch streams become available in the
future the multimetric index can be adapted easily to
include these references.
Calibration and validation of the multimetric index
The multimetric index developed in this study has
been validated with an internal and external data set.
187
Figure 7. Calibration results, final calculated quality class versus quality class based on post-classification for fast running streams from the
existing data set.
Figure 8. Validation results, final calculated quality class versus quality class based on post-classification for the AQEM data set.
In the first case 66% of the samples were classified
correctly, in the second 54%. The difference in cor-
rectly classified samples between the data sets was
not caused by differences in the collection of samples,
since the collection of samples was performed in a
similar way for both data sets.
The errors in the classifications for the internal data
set can be explained from the approach that was used
for metric selection. Class boundaries were set at the
25- and/or 75-percentile, which means that for test-
ing with a random data set, there is a 50% chance
of misclassification for class 2 and class 3 sites and
a 25% chance for class 4 and class 1 sites. Automat-
ically, the chance of an incorrect classification will
range between 50% and 75%. Changing the criteria
for metric selection (for example: no overlap between
the 10- and 90-percentile) to lower the chance of mis-
classification was not an option, because not enough
metrics would comply with these criteria to develop a
multimetric index.
Maxted et al. (2000) concluded that the Coastal
Plain Macroinvertebrate Index (CPMI) classified 86%
of the sites correctly. This is much better than the
66% for the multimetric index developed in this
188
study. However, the CPMI can only differ between
2 classes (reference and impaired sites), compared to
five classes for the multimetric index. The higher num-
ber of classes in an assessment system, the higher the
chance of misclassification. Furthermore, for the cal-
ibration of the CPMI only clearly degraded sites were
used, whereas the classification of moderate degraded
sites creates the biggest problems. For the calibra-
tion of the multimetric index, sites ranging from good
quality to bad quality were used.
Acknowledgements
This study was carried out within the AQEM project,
which was funded by the European Commission, 5th
Framework Program, Energy, Environment and Sus-
tainable Development, Key Action Water, Contract no.
EVK1-CT1999-00027. The authors are very grateful
for the sampling efforts made by all participating water
authorities. Special thanks goes to Tjeerd-Harm van
den Hoek en Martin van den Hoorn for their big efforts
in data entry, field and laboratory work. We would also
like to thank L. S. Fore and an anonymous reviewer for
their thorough review of this paper.
References
AQEM consortium, 2002. Manual for the application of the AQEM
system. A comprehensive method to assess European streams
using benthic macroinvertebrates, developed for the purpose of
the Water Framework Directive. Version 1.0, February 2002.
Bailey, R. C., M. G. Kennedy, M. Z. Dervish & R. M. Taylor, 1998.
Biological assessment of freshwater ecosystems using a refer-
ence condition approach: comparing predicted and actual benthic
invertebrate communities in Yukon streams. Freshwat. Biol. 39:
765–774.
Barbour, M. T., J. Gerritsen, G. E. Griffith, R. Frydenborg, E.
McCarron, J. S. White & M. L. Bastian, 1996. A frame-
work for biological criteria for Florida streams using benthic
macroinvertebrates. J. N. Am. Benthol. Soc. 15: 185–211.
Blocksom, K. A., J. P. Kurtenbach, D. J. Klemm, F. A. Fulk &
S. M. Cormier, 2000. Development and evaluation of the lake
macroinvertebrate integrity index (LMII) for New Jersey lakes
and resevoirs. Envir. Monit. Assess 77: 311–333.
Brabec, K., S. Zahrádková, D. Nˇ
emejcová, P. Paˇ
ril, J. Kokeš & J.
Jarkovský, 2004. Assessment of organic pollution effect con-
sidering differences between lotic and lentic stream habitats.
Hydrobiologia 516: 331–346.
Buffagni, A., S. Erba, M. Cazzola & J. L. Kemp, 2004. The AQEM
multimetric system for the southern Italian Apennines: assess-
ing the impact of water quality and habitat degradation on pool
macroinvertebrates in Mediterranean rivers. Hydrobiologia 516:
313–329.
Conquest, L. L., S. C. Ralph & R. J. Naiman, 1994. Implementation
of Large-Scale Stream Monitoring Efforts: Sampling Design and
Data Analysis Issues. In Loeb, S. L. & A. Spacie (eds), Bio-
logical Monitoring of Aquatic Systems. CRC Press LLC, Boca
Raton, Florida.
DeShon, J. E, 1995. Development and application of the invertebrate
community index (ICI). In Davis, W. S. & T. P. Simon (eds),
Biological Assessment and Criteria: Tools for Water Resource
Planning and Decision Making. Lewis Publishers, Ann Arbor,
Michigan.
European Commission, 2000. Directive 2000/60/EC OF THE
EUROPEAN PARLIAMENT AND COUNCIL establishing a
framework for Community action in the field of water policy.
Official J. Eur. Commun. L327: 1–72.
Fore, L. S., J. R. Karr & L. L. Conquest, 1994. Statistical Proper-
ties of an Index of Biological Integrity Used to Evaluate Water
Resources. Can. J. Fish. aquat. Sci. 51: 1077–1087.
Fore, L. S., J. R. Karr & R. W. Wisseman, 1996. Assessing in-
vertebrate responses to human activities: evaluating alternative
approaches. J. N. Am. Benthol. Soc. 15: 212–231.
Hawkes, H. A., 1979. Invertebrates as indicators of river water
quality. In James, A. & L. Evison (eds), Biological Indicators
of Water Quality. John Wiley, Chichester.
Hellawell, J. M., 1986. Biological indicators of Freshwater Pollu-
tion and Environmental Management. Elsevier Applied Science,
London.
Hering, D., O. Moog, L. Sandin & P. F. M. Verdonschot, 2004.
Overview and application of the AQEM assessment system.
Hydrobiologia 516: 1–20.
Intergovernmental Task Force on Monitoring Water Quality, 1993.
The Multimetric Approach for describing ecological condition.
EPA, Position Paper No. 2.
Karr, J. R., 1981. Assessment of biotic integrity using fish com-
munities. Fisheries 6: 21–27.
Karr, J. R., 1999. Defining and measuring river health. Freshwat.
Biol. 41: 221–234.
Karr, J. R. & E. W. Chu, 1999. Restoring Life in Running Waters:
Better Biological Monitoring. Island Press, Washington, DC.
Karr, J. R. & E. W. Chu, 2000. Sustaining living rivers. Hydrobio-
logia 422/423: 1–14.
Kolkwitz, R. & M. Marsson, 1909. Ökologie der tierischen Sap-
robien. Beiträge zur Lehre von der biologischen Gewässerbeur-
teilung. Int. Rev. Hydrobiol. 2: 126–152.
Liebmann, H., 1951. The biological community of Sphaerotilus
flocs and the physio-chemical basis of their formation. Vom
Wasser 20: 24.
Maxted, J. R., M. T. Barbour, J. Gerritsen, V. Poretti, N. Prim-
rose, A. Silvia, D. Penrose & R. Renfrow, 2000. Assessment
framework for mid-Atlantic coastal plain streams using benthic
macroinvertebrates. J. N. Am. Benthol. Soc. 19: 128–144.
Milner, A. M. & M. W. Oswood, 2000. Urbanization gradients
in streams of Anchorage, Alaska: a comparison of multivari-
ate and multimetric approaches to classification. Hydrobiologia
422/423: 209–223.
Moller Pillot, H. K. M., 1971. Faunistische beoordeling van de
verontreiniging in laaglandbeken. Proefschrift, Tilburg, The
Netherlands, 286 pp.
Ofenböck, T., O. Moog, J. Gerritsen & M. Barbour, 2004. A stressor
specific multimetric approach for monitoring running waters in
Austria using benthic macro-invertebrates. Hydrobiologia 516:
251–268.
Ohio Enviromental Protection Agency, 1987. Biological Criteria for
the Protection of Aquatic Life: Volume I-III. Ohio EPA, Division
of Water Quality Monitoring and Assessment, Surface Water
Section, Columbus, Ohio.
189
Preston, F. W., 1962. The canonical distribution of commonness and
rarity: part 1. Ecology 43: 185–215.
Resh, V. H., D. M. Rosenberg & T. B. Reynoldson, 2000. Selection
of benthic macroinvertebrate metrics for monitoring water qual-
ity of the Fraser River, British Columbia: implications for both
multimetric approaches and multivariate models. In Wright, J. F.,
W. Sutcliffe & M. Furse (eds), Assessing the Biological Quality
of Fresh Waters: RIVPACS and Other Techniques. Freshwater
Biological Association, Cumbria, U.K.
Reynoldson, T. B., R. H. Norris, V. H. Resh, K. E. Day & D.
M. Rosenberg, 1997. The reference condition: a comparison of
multimetric and multivariate approaches to assess water-quality
impairment using benthic macroinvertebrates. J. N. Am. Benthol.
Soc. 16: 833–852.
Rosenberg, D. M. & V. H. Resh (eds), 1993. Freshwater Bio-
monitoring and Benthic Macroinvertebrates. Chapman and Hall,
London, 461 pp.
Roth, N. E., M. T. Southerland, J. C. Chaillou, J. H. Volstad, S. B.
Weisberg, H. T. Wilson, D. G. Heimbuch & J. C. Seibel, 1997.
Maryland biological stream survey: ecological status of non-tidal
streams in six basins sampled in 1995. Report no. CBWP-
MANTA-EA-97-2. Maryland Department of Natural Resources,
Annapolis, Maryland.
Royer, T, V., C. T. Robinson & G. W. Minshall, 2001. Development
of Macroinvertebrate-Based Index for Bioassessment of Idaho
Rivers. Env. Manag. 27: 627–636.
Sandin, L., J. Dahl & R. K. Johnson, 2004. Assessing acid stress in
Swedish boreal and alpine streams using benthic macroinverte-
brates. Hydrobiologia 516: 129–148.
Sládeˇ
cek, V., 1965. The future of the saprobity system. Hydrobiolo-
gia 25: 518–537.
Sládeˇ
cek, V., 1973. System of water quality from the biological
point of view. Arch. f. Hydrobiol. Beih. Ergebnisse Limnol. 7:
1–218.
Sørensen, T., 1948. A method of establishing groups of equal
amplitude in plant sociology based on similarity of species con-
tent and its application to analyses of the vegetation on Danish
commons. Biol. Skr. 5: 1–34.
Stichting Toegepast Onderzoek Waterbeheer, 1992. Ecologische
beoordeling en beheer van oppervlaktewater: Beoordelingssys-
teem voor stromende wateren op basis van macrofauna. STOWA,
Utrecht, The Netherlands, 58 pp.
Wright, J. F., D. Moss, P. D. Armitage & M. T. Furse, 1984. A
preliminary classification of running-water sites in Great Bri-
tain based on macroinvertebrate species and the prediction of
community type using environmental data. Freshwat. Biol. 14:
221–256.
Ter Braak, C. J. F., 1987. CANOCO – A FORTRAN program
for canonical community ordination by [partial] [detrended] [ca-
nonical] correspondene analysis, principal component analysis
and redundancy analysis (version 2.1). TNO Institute of Applied
Computer Science, Wageningen, The Netherlands, 95 pp.
Ter Braak, C. J. F. ter & P. Šmilauer, 1998. CANOCO Reference
Manual and User’s Guide to Canoco for Windows: Software for
Canonical Community Ordination (version 4). Microcomputer
Power, Ithaca, NY, U.S.A., 352 pp.
Thorne, R. J. & W. P. Williams, 1997. The response of benthic
macroinvertebrates to pollution in developing countries: a multi-
metric system of bioassessment. Freshwat. Biol. 37: 671–686.
Tolkamp, H. H. & J. J. P. Gardeniers, 1971. Hydrobiological survey
of lowland streams.PhD Thesis, Standaard Boekhandel, Tilburg,
The Netherlands, 286 pp.
Tuffery, G. & J. Verneaux, 1968. Méthode de détermination de la
qualité biologique des eaux courantes. Exploitation codifiée des
inventaires de fauna du fond. Ministère de l’Agriculture, France,
23 pp.
Van Tongeren, O., 1986. FLEXCLUS, an interactive flexible cluster
program. Acta bot.. Neerl. 35: 137–142.
Van Tongeren, O. (s.a.). Programma ASSOCIA: Gebruiker-
shandleiding en voorwaarden.
Verdonschot, P. F. M., 1990. Ecological characterisation of sur-
face waters in the province of Overijssel (The Netherlands).
PhD. dissertation, Institute for Forestry and Nature Research,
Wageningen, The Netherlands, 255 pp.
Verdonschot, P. F. M. & R. C. Nijboer, 2000. Typology of macro-
faunal assemblages applied to water and nature management: a
Dutch approach. In Wright, J. F., W. Sutcliffe & M. Furse (eds),
Assessing the biological quality of fresh waters: RIVPACS and
other techniques. Freshwater Biological Association, Cumbria,
U.K.
Weigel, B. M., 2003. Development of stream macroinvertebrate
models that predict watershed and local stressors in Wisconsin.
J. N. Am. Benthol. Soc. 22: 123–142.
Woodiwiss, F. S., 1964. The biological system of stream clas-
sification used by the Trent River Board. Chem. Industry 11:
443–447.