See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/254646240

Defining macroinvertebrate assemblage types of

headwater streams: Implications for
bioassessment and...

Article in Ecological Applications · June 2003

DOI: 10.1890/1051-0761(2003)013[0842:DMATOH]2.0.CO;2

CITATIONS READS

101 263

6 authors, including:

Jani Heino Timo Muotka

Finnish Environment Institute University of Oulu
152 PUBLICATIONS 4,803 CITATIONS 131 PUBLICATIONS 4,330 CITATIONS

SEE PROFILE SEE PROFILE

Riku Paavola Heikki Hämäläinen

University of Oulu University of Jyväskylä
21 PUBLICATIONS 1,174 CITATIONS 87 PUBLICATIONS 1,532 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Patterns of diversity, occupancy and abundance in aquatic organisms View project

Wild spawning stocks of landlocked Atlantic salmon and lake-migrating brown trout View project

All content following this page was uploaded by Jani Heino on 24 June 2017.

The user has requested enhancement of the downloaded file. All in-text references underlined in blue are added to the original document
and are linked to publications on ResearchGate, letting you access and read them immediately.
Ecological Applications, 13(3), 2003, pp. 842–852
q 2003 by the Ecological Society of America

DEFINING MACROINVERTEBRATE ASSEMBLAGE TYPES OF

HEADWATER STREAMS: IMPLICATIONS FOR BIOASSESSMENT
AND CONSERVATION
JANI HEINO,1 TIMO MUOTKA,2,5 HEIKKI MYKRÄ,1 RIKU PAAVOLA,1 HEIKKI HÄMÄLÄINEN,3 AND
ESA KOSKENNIEMI4
1Department of Biological and Environmental Science, University of Jyväskylä, POB 35, 40351 Jyväskylä, Finland
2Finnish Environment Institute, P.O. Box 140, 00251 Helsinki, Finland
3Karelian Institute, Section of Ecology, University of Joensuu, POB 111, 80101 Joensuu, Finland
4West Finland Environment Center, POB 262, 65101 Vaasa, Finland

Abstract. We used data from 235 boreal headwater streams in Finland to examine
whether macroinvertebrate assemblages constitute clearly definable types, and how well
biologically defined assemblage types can be predicted using environmental variables. Two-
way indicator species analysis produced 10 assemblage types, which differed significantly
from each other (multiresponse permutation procedure, MRPP). However, based on MRPP
and nonmetric multidimensional scaling, there was wide variation among sites within each
assemblage type, and high degrees of overlap among assemblage types. Such continuous
variation was also evidenced by the low number of effective indicator taxa (indicator value
method) for each assemblage type. Furthermore, discriminant function analysis based on
environmental variables showed a moderate yet variable prediction success (59.6% of sites
correctly predicted; range 0–96% per group). Canonical correspondence analysis indicated
that variation in assemblage structure was primarily related to latitude and water chemistry,
especially pH and water color. The absence of discrete macroinvertebrate assemblage types
in boreal headwater streams may stem from a number of reasons: (1) macroinvertebrate
taxa inhabiting boreal streams tend to exhibit individualistic, taxon-specific responses to
environmental gradients, (2) they have wide environmental tolerances and geographic dis-
tributions, and (3) boreal streams are characterized by frequent, unpredictable disturbances.
Our results suggest that local filters in headwater streams are relatively weak, resulting in
poorly distinguishable assemblage types. By contrast, the major latitudinal gradients in
macroinvertebrate assemblage structure suggest that regional-scale filters, e.g., temperature,
exert strong control over taxon distributions in headwater streams. We suggest that a tiered
approach combining regional stratification and subsequent prediction of assemblage struc-
ture could provide a suitable framework for the bioassessment and conservation of boreal
headwater streams.
Key words: aquatic conservation; aquatic insects; assemblage types; benthic invertebrates; bioas-
sessment; boreal regions; headwater streams; latitudinal gradients; lotic communities.

INTRODUCTION bioassessment programs. Although most ecologists

concur that setting strict boundaries among gradually
A major controversy in the history of community
changing communities is artificial (Begon et al. 1996,
ecology has centred on whether biotic communities ex-
Maurer 1999), many bioassessment approaches are
ist as discrete entities (‘‘community-unit concept’’) or
based on community classifications, with an implicit
as a continuum of gradually changing composition
assumption that communities can be divided into dis-
(‘‘individualistic concept’’) (for reviews, see Austin
tinct, ecologically meaningful groups (see Anderson
1985, Allen and Hoekstra 1992, McIntosh 1995). More
recently, it has been shown that this dichotomy is too and Clements 2000).
limited (Shipley and Keddy 1987) and that the two A major approach to understanding community or-
viewpoints are not mutually exclusive (Brown 1995). ganization has revolved around the concept of assem-
However, glimpses of both concepts still frequently bly rules. The early advocates of assembly rules at-
come up in the ecological literature. One example that tempted to predict community composition based on
reflects the old controversy is the frequent use of clus- competitive interactions and niche requirements of po-
tering techniques to distinguish assemblage types in tential colonists (Diamond 1975, Diamond and Gilpin
1982). The second generation of assembly rules gave
less weight to interspecific competition, emphasizing
Manuscript received 17 December 2001, revised 30 September
2002, accepted 31 October 2002. Corresponding Editor: J. E.
the role of colonization history and local environmental
McDowell. filters (Drake 1991, Keddy 1992). Environment is re-
5 Corresponding author. E-mail: [email protected]
garded as a filter, selecting only a subset of regionally
842
June 2003 MACROINVERTEBRATE ASSEMBLAGE TYPES 843

available taxa that possess traits allowing persistence stream types: 10–30 streams from each of 11 major
under prevailing habitat conditions (Tonn 1990, Weiher watersheds covering the whole country were included.
and Keddy 1995, Poff 1997). If a series of filters set We stratified our sampling according to stream size,
strong constraints to species traits and, ultimately, spe- i.e., we considered only headwater streams (base flow
cies identities, community composition along natural ,0.6 m3/s, catchment area 1–60 km2) to diminish the
environmental gradients should be highly predictable overriding effects of stream size. Streams within each
(e.g., Matthews 1998). watershed were selected as randomly as possible, with-
Stream bioassessment programs typically rely on ac- in a 2 km distance from the nearest road. Streams with
curate identification and prediction of community obvious human impact, spring-fed streams, and lake
types. When related to regional reference conditions, outlets were excluded. Nevertheless, our study sites
this allows one to account for variation in biological span a wide range of environmental variability present
data caused by natural factors (e.g., stream size, natural in boreal headwater streams. For example, pH varies
acidity) rather than anthropogenic disturbances (e.g., from 4.5 to 8.4, water color from 5 to 600 mg Pt/L (Pt
Wright et al. 1984, Marchant et al. 1997, Reynoldson 5 platinum units), canopy cover from 0 to 91%, and
et al. 2001). Concurrently, there have been pleas for current velocity from 7 to 110 cm/s.
redirecting aquatic conservation programs to protection
at the community rather than the population level (An- Invertebrate data
germeier and Schlosser 1995, Angermeier and Winston
Invertebrate sampling was conducted between early
1999). Such a shift in conservation policy requires that
September and the end of October. At each site, we
assemblages worthy of conservation be clearly iden-
took a two-minute collective kick sample (net mesh
tifiable. This, however, may not always be a simple
size 0.3 mm), aiming to cover most benthic microhab-
task. For instance, stream macroinvertebrate assem-
itats in a riffle section of ;100 m2. The samples were
blages may exhibit highly variable structure caused by
immediately preserved in alcohol. In the laboratory,
unpredictable disturbances and high extent of dispersal
invertebrates were sorted and identified to the lowest
(Palmer et al. 1996). Some authors have suggested,
possible taxonomic level, usually species or genus. Ear-
however, that local habitat filters in streams (e.g., water
ly instars of some limnephilid caddis larvae and dip-
depth, substrate composition, water chemistry) are
teran larvae were not identified to this level, so family
highly effective, resulting in locally controlled, pre-
level identification was retained.
dictable assemblage types (Hawkins et al. 2000). That
stream benthic communities occur as predictable, dis-
Habitat characteristics
crete entities is often implicitly assumed, yet few stud-
ies have rigorously examined the potential implications Several riparian and in-stream habitat measurements
of this dichotomy (continuous vs. discrete nature of were conducted at each site. Tree species composition
communities) to stream bioassessment and conserva- and integrity of riparian zone (percentage of riparian
tion (but see Chessman 1999). zone without human influence) were assessed in a 50-
In this study, we used data on benthic macroinver- m section on both banks directly upstream of the sam-
tebrate assemblages of near-pristine headwater streams pling site. Shading was measured as percent canopy
in Finland to answer three questions. (1) Do macro- cover at 20 locations in evenly spaced transects cov-
invertebrate assemblages of boreal headwater streams ering the whole study section. Current velocity and
constitute clearly definable assemblage types? (2) What depth were measured at 40 locations along the same
natural factors are associated with spatial patterns in transects. Moss cover and particle size were assessed
assemblage structure? (3) How well are biologically at 10 randomly spaced 50 3 50 cm quadrats. Visual
defined assemblage types predicted by these environ- estimates of the percentage cover of nine particle size
mental factors? Finally, we consider the implications classes and organic material were made for each quad-
of our results for the implementation of bioassessment rat using a modified Wentworth scale: (0) organic mat-
and conservation programs in headwater stream eco- ter, (1) sand (diameter 0.25–2 mm), (2) fine gravel (2–
systems. 6 mm), (3) coarse gravel (6–16 mm), (4) small pebble
(16–32 mm), (5) large pebble (32–64 mm), (6) small
MATERIALS AND METHODS cobble (64–128 mm), (7) large cobble (128–256 mm),
(8) boulder (256–400 mm), and (9) large boulder
Stream surveys
(.400 mm). These estimates were subsequently av-
We sampled 235 streams across Finland (608–708 N, eraged to give a mean particle size for a site. Mean
208–328 E) between 1992 and 2000, the majority stream width was also measured at each sampling site.
(.75%) of samples being collected in 1998 or 2000. Latitude and longitude (Finnish national grid) were de-
Material collected prior to 1998 was included only if termined in the field using a GPS navigator.
sampling methods were identical to those used in 1998/ Water samples were collected simultaneously with
2000. We used a spatially stratified sampling protocol benthic sampling, and they were analyzed for pH, al-
to guarantee adequate representation of headwater kalinity, conductivity, total nitrogen [TN], total phos-
844 JANI HEINO ET AL. Ecological Applications
Vol. 13, No. 3

phorus [TP], color, and iron [Fe] by regional environ- fications and/or were poorly predicted in discriminant
ment centers using national standards. function analyses. Rare species were retained in the
analysis, because their exclusion may hinder the de-
Statistical methods tection of meaningful ecological gradients in assem-
We performed three kinds of multivariate analyses blage structure (Cao et al. 2001). Thus, a total of 189
to define macroinvertebrate assemblage types: two-way operational macroinvertebrate taxa were included in
indicator species analysis (TWINSPAN), nonmetric statistical analyses. TWINSPAN, NMDS, MRPP, and
multidimensional scaling (NMDS), and multiresponse INDVAL were run using PC-Ord (McCune and Mef-
permutation procedure (MRPP). TWINSPAN is based ford 1999).
on reciprocal averaging (Gauch 1982, van Tongeren We used two methods to relate community types to
1995), and it is widely used in freshwater ecology and environmental variables. First, we employed discrim-
bioassessment (Ormerod and Edwards 1987, Wright et inant function analysis (DFA) with stepwise selection
al. 1997, Moss et al. 1999). Yet this method has been to reveal which environmental variables were most im-
criticized mainly on two grounds: (1) it is based on the portant in discriminating among the TWINSPAN
assumption that a single strong gradient dominates the groups at each division level. DFA (with cross-vali-
data; and (2) the rather arbitrary cutting of the major dation) was also used to predict the group membership
axis may separate sites with relatively similar species of each site according to environmental variables. Sec-
composition (Belbin and McDonald 1993, Legendre ond, we used canonical correspondence analysis (CCA)
and Legendre 1998). Despite its drawbacks, it has been to examine the relationships of assemblage types and
shown that, compared with alternative clustering meth- indicator species (from INDVAL) with environmental
ods, TWINSPAN performs well in the classification of gradients. CCA is a direct gradient analysis that ana-
benthic macroinvertebrate assemblages (Moss et al. lyzes both species and environmental data by combin-
1999). We also used NMDS to assess whether the ing ordination and regression techniques (ter Braak
TWINSPAN groups represented discrete assemblage 1995, Legendre and Legendre 1998). We used forward
types in the ordination space. NMDS is an ordination selection of environmental variables. At each step, only
method based on ranked distances, and it is suitable variables significantly related (P , 0.05; Monte Carlo
for analyzing ecological data sets for numerous rea- randomization test with 100 permutations) to assem-
sons. NMDS performs well with data that are nonnor- blage structure were included in the model. Addition-
mally distributed, are on arbitrary, discontinuous ally, detrended correspondence analysis (DCA) was
scales, or contain numerous zero values (McCune and used to determine gradient lengths for the first three
Mefford 1999). Sorensen’s coefficient based on axes, and to assess the degree of species turnover across
log(x 1 1) macroinvertebrate abundance data was used ecological gradients. Gradient length $4 SD units in-
as the distance measure in NMDS. A three-dimensional dicates complete species turnover (Legendre and Le-
solution of NMDS was used, because change in stress gendre 1998). CCA and DCA were run using CANOCO
value was minor with subsequent dimensions. We also version 4.0 (ter Braak and Smilauer 1998).
tested for significant differences among groups at each
TWINSPAN division using MRPP. MRPP is a non- RESULTS
parametric method designed for testing differences in At the fourth division level, TWINSPAN produced
assemblage structure among a priori defined groups 10 site groups (Fig. 1). At each division, MRPP indi-
(Zimmerman et al. 1985). MRPP was also based on cated significant among-group differences in assem-
Sorensen’s coefficient. The significance of the null hy- blage structure (Monte Carlo permutation test, P ,
pothesis of no differences among groups was assessed 0.001). However, values for MRPPs R statistic were
by a Monte Carlo permutation procedure with 1000 generally very low (0.034 to 0.089), indicating that
permutations. Further, we used the indicator value there was wide within-group variation (see McCune
method (INDVAL) (Dufrene and Legendre 1997) to and Mefford 1999). Further, when plotted in the NMDS
identify species discriminating between TWINSPAN ordination space, TWINSPAN groups did not form dis-
groups at the first and the fourth division level. The tinct site groupings indicative of discrete assemblage
indicator value of a taxon varies from 0 to 100, and it types (Fig. 2). The negative and positive groups of the
attains maximum value when all individuals of a taxon first TWINSPAN division were clearly observable in
occur at all sites of a single group. The significance of the NMDS space, but at further division levels, site
the indicator value for each taxon was tested by a Mon- groups exhibited considerable overlap. At the fourth
te Carlo randomization test with 1000 permutations. division level, group A streams (acid streams in eastern
All analyses were run on transformed [log(x 1 1)] Finland) deviated from all others along the first and
abundance data. For TWINSPAN, we used five pseu- third NMDS dimensions (Fig. 2).
dospecies cut levels, four division levels, and 25 as the INDVAL identified significant indicator species for
minimum group size for each division. This minimum each TWINSPAN division level (Monte Carlo per-
group size was used because, based on preliminary mutation test, P , 0.001). For clarity, only the first
analyses, smaller groups provided less robust classi- and the fourth division level will be considered here.
June 2003 MACROINVERTEBRATE ASSEMBLAGE TYPES 845

FIG. 1. TWINSPAN classification of the study streams. Numbers refer to number of sites in each group. Also shown are
mean (1 1 SD) latitude (Finnish grid system, solid bars) and pH (shaded bars) for each TWINSPAN end group.

The negative group (n 5 136 sites) of the first TWIN- (Kempny) (63), the riffle beetle Elmis aenea (Müller)
SPAN division level (Fig. 1) was characterized by the (36), and the caddisfly Micrasema gelidum McLachlan
mayflies Ephemerella aurivillii (Bengtsson) (indicator (33). The positive group (n 5 99 sites) was character-
value 57 [indicator values given in parentheses]) and ized by the crustacean Asellus aquaticus (Linnaeus)
Baetis subalpinus Bengtsson (42), the stoneflies Pro- (67), the stoneflies Leuctra hippopus Kempny (33) and
tonemura meyeri (Pictet) (33) and Diura nanseni Nemoura cinerea (Retzius) (72), and the caddisflies

FIG. 2. An ordination diagram showing the distribution of the TWINSPAN site groups in the NMDS space. The centroid
of each TWINSPAN group or group combination is denoted by capital letters (A–J). Ellipses encircle 90% of sites belonging
to a given group.
846 JANI HEINO ET AL. Ecological Applications
Vol. 13, No. 3

TABLE 1. Results of indicator species analysis (INDVAL) at the fourth TWINSPAN division
level.

TWINSPAN group
Indicator taxon A B C D E F G H I J
Heptagenia sulphurea 1 0 0 0 1 0 0 0 60 0
Ephemerella aurivillii 13 18 12 29 13 0 0 0 0 0
Onychogomphus forcipatus 0 0 0 0 0 0 0 0 30 0
Leuctra fusca 55 0 0 10 1 0 0 0 0 0
Capnopsis schilleri 0 2 5 6 2 31 3 0 0 1
Protonemura intricata 28 0 0 0 1 0 0 0 1 0
Rhyacophila fasciata 0 0 0 0 0 14 0 0 35 0
Cheumatopsyche lepida 0 0 0 0 1 0 0 0 31 0
Hydropsyche pellucidula 2 0 0 0 0 0 0 0 65 14
Hydropsyche siltalai 0 0 0 0 0 0 0 0 51 31
Potamophylax cingulatus 0 1 0 1 0 43 2 0 2 1
Chaetopteryx villosa 30 0 0 0 0 0 0 0 1 0
Notes: Shown are indicator values for the top 12 species significantly discriminating (indicator
value . 25, P , 0.001) among the 10 TWINSPAN site groups (A–J). The significance of the
indicator values was tested using a Monte Carlo randomization test with 1000 permutations.
The highest indicator value for each species is given in boldface.

Polycentropus flavomaculatus (Pictet) (38) and Hy- The first TWINSPAN division paralleled geograph-
dropsyche siltalai Döhler (37). At the fourth TWIN- ical separation of streams into a northern (A–E) and a
SPAN division level, INDVAL detected relatively few southern (F–J) subgroup (Fig. 1). Further divisions
good indicators for the 10 site groups (Table 1). Species were related to local environmental factors (Tables 2
typical of group A were the stoneflies Leuctra fusca and 3). The prediction success of DFA decreased with
(Linnaeus) and Protonemura intricata (Ris), and the increasing TWINSPAN division level and number of
caddisfly Chaetopteryx villosa (Fabricius). Group D groups: the proportion of correct predictions was 91.9,
was characterized by the mayfly Ephemerella aurivillii. 85.5, 68.1, and 59.6% for the first to fourth division
The stonefly Capnopsis schilleri (Rostock) and the cad- level, respectively. At the first TWINSPAN level, eight
disfly Potamophylax cingulatus-latipennis were the variables were included in the DFA model, in order of
type species of group F, whereas no significant indicator importance: latitude, longitude, [TN], percentage de-
species were found for groups G and H. Group I was ciduous trees, depth, particle size, alkalinity, and con-
characterized by the mayfly Heptagenia sulphurea ductivity. At the fourth division level, the DFA model
(Müller), the dragonfly Onychogomphus forcipatus incorporated latitude, pH, longitude, water color, moss
(Linnaeus), and the caddisflies Rhyacophila fasciata cover, particle size, and percentage deciduous trees (Ta-
Hagen, Cheumatopsyche lepida (Pictet), Hydropsyche ble 3). Using this reduced set of variables, DFA pre-
pellucidula (Curtis), and Hydropsyche siltalai. H. sil- dicted correctly 59.6% of the streams, the percentage
talai also characterized streams of group J. Albeit sig- of correct classifications ranging from 0 to 96 (Table
nificant, indicator values remained low (,50) even for 4). No streams in group H were correctly predicted,
most of these major indicator taxa (Table 1). and streams in groups D, E, and F were poorly predicted

TABLE 2. Mean values of environmental variables important in discriminating among the TWINSPAN groups at the fourth
division level.

TWINSPAN group
Variable A B C D E F G H I J
North coordinates 7131 7476 7639 7465 7307 7050 7046 7089 6919 6919
East coordinates 3650 3411 3456 3497 3528 3427 3358 3439 3450 3531
Deciduous trees (%) 42 74 87 65 65 41 52 61 72 70
Current velocity 29 39 38 36 45 45 34 29 41 31
(cm/s)
Depth (cm) 23 24 19 22 25 23 21 22 24 16
Moss cover (%) 40 46 15 41 46 31 18 19 31 29
Particle size 7.1 5.6 6.7 6.4 6.1 6.5 4.9 5.0 5.9 5.1
pH 5.5 7.0 7.3 7.1 7.2 5.9 5.8 6.1 6.6 6.5
Alkalinity (mmol/L) 0.05 0.34 0.32 0.53 0.54 0.10 0.10 0.17 0.23 0.18
Conductivity 2.38 4.76 4.52 6.80 6.77 4.20 3.35 4.71 5.24 4.91
(mS/m)
Total N (mg/L) 263 249 165 230 301 838 712 776 444 417
Color (mg Pt/L) 161 69 35 81 98 204 275 151 89 80
June 2003 MACROINVERTEBRATE ASSEMBLAGE TYPES 847

TABLE 3. Wilks’ lambda from stepwise DFA for variables ance. Gradient lengths for the first three DCA axes were
best discriminating among groups at the first and the fourth
TWINSPAN division level. 3.304, 2.716, and 3.059, respectively, implying that
species turnover among sites at the opposite ends of
TWINSPAN division level the gradients were incomplete.
1 4 DISCUSSION
Step Variable Lambda Variable Lambda
Given the obvious advantages of community-level
1 latitude 0.502 latitude 0.277 conservation (see Angermeier and Schlosser 1995), it
2 longitude 0.457 pH 0.155
3 [TN] 0.426 longitude 0.087 would be highly desirable if macroinvertebrate assem-
4 % deciduous 0.403 color 0.062 blage types could be clearly identified and predicted
5 depth 0.384 moss 0.048 based on environmental variables. For boreal head-
6 particle size 0.375 particle size 0.040 water streams, however, this proved to be difficult.
7 alkalinity 0.365 % deciduous 0.034
8 conductivity 0.357 While MRPP did find significant differences among the
macroinvertebrate assemblage types at each TWIN-
Note: The variable minimizing the overall Wilks’ lambda
is entered at each step. SPAN division, the associated MRPP R statistics were
overall very low (,0.09), indicating that there was
wide variability within each TWINSPAN group. There-
(,50%) to their actual groups. Groups B and J showed fore, accepting these results as indicative of disjunct
moderate prediction success, with more than half of assemblage types incurs a risk of mixing statistical and
the streams being correctly classified. Finally, classi- biological significance (McCune and Mefford 1999, see
fication success was high for groups A, C, G, and I, also Van Sickle and Hughes 2000). Furthermore, IND-
with .70% of streams correctly classified (Table 4). VAL found few, if any, effective indicator taxa for most
Eigenvalues and species–environment correlations of the assemblage types defined by TWINSPAN. In
for the first three CCA axes were all significant (Monte fact, only the division of streams into a northern and
Carlo test, P , 0.01). Forward selection of environ- southern subgroup provided some indication of the ex-
mental variables incorporated 15 variables (Table 5) in istence of clearly definable assemblage types. Variation
the final model (Monte Carlo test, 100 permutations at in assemblage structure was mainly related to geo-
each step, P , 0.05). Axis 1 was strongly related to graphical location, stream acidity, and water color, fac-
latitude, whereas axis 2 represented a water quality tors formerly identified as major correlates of macro-
gradient with clear-water, high pH streams, and brown- invertebrate distribution and assemblage structure in
water, acid streams being the endpoints of this gradient boreal streams (Malmqvist and Mäki 1994, Hämäläinen
(Fig. 3a). Axis 3 mainly described variation in assem- and Huttunen 1996, Malmqvist and Hoffsten 2000,
blage structure along the east–west gradient (Table 4). Sandin and Johnson 2000, Paavola et al. 2000).
Overall, CCA reinforced the main results of DFA: as- The absence of discrete macroinvertebrate assem-
semblage structure varied along a geographical gra- blage types in boreal headwater streams may be at-
dient and, secondarily, along a water chemistry gra- tributable to a number of reasons. In general, systems
dient. Nevertheless, TWINSPAN groups showed con- characterized by unpredictable disturbances and high
siderable overlap in the CCA space (Fig. 3b), as did extent of dispersal should exhibit highly variable as-
also the distributions of the top 12 INDVAL indicator semblage structure (Palmer et al. 1996). In such sys-
taxa along the first two CCA axes (Fig. 4). Thus, even tems, random extinctions and recolonizations may
taxa that were identified as indicators of their respective lead to assemblage structure deviating from that ex-
groups exhibited wide ranges of environmental toler- pected based on water chemistry and physical habitat

TABLE 4. Prediction of TWINSPAN group (A–J) membership for the 235 stream sites using
discriminant function analysis on environmental variables.

Predicted Correct
classifications
Actual A B C D E F G H I J (%)
A 24 1 96.0
B 16 4 3 2 2 59.3
C 34 6 3 1 77.3
D 4 6 6 3 1 1 28.6
E 3 1 3 8 3 1 42.1
F 2 3 3 2 2 25.0
G 1 1 1 24 1 2 80.0
H 2 1 1 3 4 0 1 4 0
I 1 1 14 4 70.0
J 1 3 6 11 52.4
Note: The number of correctly classified streams is given in boldface.
848 JANI HEINO ET AL. Ecological Applications
Vol. 13, No. 3

TABLE 5. Summary of the canonical correspondence analysis.

Variable Axis 1 Axis 2 Axis 3

Eigenvalue 0.245 0.177 0.134
% variance of species data 4.2 3.0 2.3
Species–environment correlation 0.918 0.853 0.795
Latitude 20.921 20.062 20.009
Longitude 0.040 0.285 0.673
Riparian integrity 20.156 20.092 20.313
Deciduous trees 20.238 0.341 20.302
Shading 0.232 20.471 20.197
Stream width 20.224 0.533 0.258
Velocity 20.156 0.215 20.044
Depth 20.253 0.001 0.318
Moss cover 20.196 0.102 0.345
Particle size 20.268 0.067 0.321
pH 20.556 0.620 20.339
Conductivity 20.136 0.418 20.327
Total phosphorus 0.384 20.379 0.029
Color 0.325 20.496 0.309
Iron 0.236 20.494 0.200
Notes: Shown are the eigenvalues for and percentage variance explained by the first three
axes, as well as the intraset correlations among the axes and environmental variables. Only
variables selected through forward selection procedure are presented. The variable with the
highest correlation for each axis is given in boldface. Total inertia of species data was 5.819.

at the time of sampling. While the role of disturbance roinvertebrate diversity (e.g., Wright et al. 1984, Rey-
(floods, droughts, bottom freeze) in shaping the biotic noldson et al. 2001).
communities of boreal headwater streams is well es- Obviously, increasing the environmental scope of the
tablished (Malmqvist and Hoffsten 2000), not much study by including other types of stream habitats, e.g.,
is known about the dispersal rates and distances of large rivers and lake outlets, would have revealed pat-
lotic invertebrates in general. Indirect evidence sug- terns different from those observed here. Our sampling
gests, however, that most species possess considerable was stratified by stream type; we concentrated on head-
dispersal capacity. For example, Malmqvist et al. water streams (mostly orders 1 and 2). Stratification by
(1991) documented an astonishingly rapid coloniza- the size of the water body was also recommended by
tion by most invertebrate taxa of a man-made stream Tonn (1990; ‘‘lake type filter’’), and we believe this is
in southern Sweden. Importantly, this stream origi- indeed to be recommended for many reasons. First,
nated from a lake, so colonization by lotic taxa must from a biodiversity perspective, headwater streams
have been mainly via aerial dispersal. Furthermore, comprise a valuable and unique resource, supporting a
as taxa inhabiting boreal streams tend to have wide number of taxa absent from larger rivers (Wright et al.
environmental tolerances and geographic distribu- 1998, Furse 2000, Malmqvist and Hoffsten 2000). Sec-
tions (Malmqvist and Hoffsten 2000), it appears that ond, including streams of different sizes almost cer-
local filters in these systems are relatively weak, re- tainly results in a predominant size-related axis in any
sulting in poorly distinguishable assemblage types. multivariate analysis of lotic assemblages (e.g.,
This was also observed in our study, regardless of the Malmqvist and Mäki 1994, Angermeier and Winston
fact that the environmental gradients sampled were 1999), thus potentially obscuring the detection of other
fairly long. biologically relevant trends. Therefore, while the in-
The relatively low distinctiveness of macroinverte- clusion of multiple habitat types would almost certainly
brate assemblage types in our study streams may also produce easily definable assemblage types, this incurs
reflect the fact that, due mainly to historical reasons a loss in our ability to identify assemblage types within
(e.g., glaciation), northern European macroinvertebrate one important freshwater habitat type, headwater
fauna is known to be depauperate compared to many streams.
other regions of the world (Jacobsen et al. 1997). Such From a bioassessment perspective, our data portray
low diversity clearly reduces our ability to detect dis- a reference condition for boreal headwater streams, be-
tinct assemblage types (see also Reynoldson et al. cause all material was collected from near-pristine en-
2001). It should be remembered, however, that our data vironments. Human-induced changes, whether in the
are stratified by habitat type (only headwaters), season surrounding terrestrial landscape or in the stream chan-
(only autumn), and taxonomic resolution (chironomids nel, may create a novel ‘‘anthropogenic filter,’’ result-
not included), and therefore likely capture only a por- ing in assemblage types rarely found in more natural
tion of the regional species pool. Therefore, our esti- settings. For example, Muotka et al. (2002) found that
mate of regional taxon richness is not directly com- streams dredged in Finland to facilitate timber transport
parable to most other regional analyses of lotic mac- provide a unique type of filter (simplified substratum
June 2003 MACROINVERTEBRATE ASSEMBLAGE TYPES 849

ated by covariation of geographical location and water

chemistry across the study area (Heino et al. 2002).
Therefore, it seems obvious that geographical stratifi-
cation would improve stream bioassessment programs
in boreal areas. Regional stratification might allow
mastering part of the natural variation in assemblage
structure, simply because adjacent headwater streams
should resemble each other more than those farther
away, and should thus respond similarly to large-scale
environmental variation. However, neither regional de-
lineations (Whittier et al. 1988, Sandin and Johnson
2000) nor biological classifications (Wright et al. 1984,
1998, Ormerod and Edwards 1987, Marchant et al.
1997, Reynoldson et al. 2001) alone perform well for
boreal headwater streams, because assemblage types
are far from discrete and not readily predictable based
on environmental variables. A combination of regional
stratification and biological characterization of assem-
blage types might provide a better framework for the
bioassessment of boreal streams. Such a combination
could increase the robustness of predictive models,
leading to more parsimonous models than those gen-
erated for larger, more heterogeneous areas (Johnson
and Goedkoop 2000). Moreover, acknowledging that
macroinvertebrate assemblages exhibit continuous var-
iation across environmental gradients, predictive mod-
els should not rely too heavily on methods that produce
distinct site groupings (e.g., cluster analysis), whether
these were biologically meaningful or not (see also
Chessman 1999).
Since the characterization of macroinvertebrate as-
semblage types seems to be of limited value in the
conservation evaluation of boreal headwater streams,
how should their conservation value be assessed? Ob-
viously, approaches based on species richness, rare spe-
cies hotspots, or unusual species combinations warrant
consideration in this regard. Ultimately, however, con-
servation efforts should guarantee that biologically
valuable assemblages remain viable, i.e., that they re-
tain their value in the future (Angermeier and Schlosser
FIG. 3. A CCA biplot defined by the first two axes of the 1995). To be successful, such efforts will require that
ordination of environmental variables and TWINSPAN site the regional and landscape context of biological as-
groups (denoted by capital letters). For clarity, only the eight
most important variables are shown. semblages be preserved by enhancing the terrestrial–
aquatic linkages and by providing possibilities for the
biota for intersite dispersal throughout entire stream
structure, homogeneous flow patterns, reduced reten- networks (Naiman et al. 1993, Winston and Angermeier
tion potential) favoring species traits that might not be 1995, Ward 1998). Finally, conservation assessment
as beneficial in more complex and retentive natural should not be based on a single taxonomic group, e.g.,
channels. Because the very objective of stream bioas- macroinvertebrates or fish, at least not until the poten-
sessment is to evaluate whether a stream is impaired tial applicability of such ‘‘indicator taxa’’ in preserving
or not, relative to regional reference conditions, such lotic biodiversity is more fully explored.
modified streams are obviously unsuitable as reference
ACKNOWLEDGMENTS
sites for bioassessment.
The strong latitudinal patterns in macroinvertebrate We are grateful to the environment centers of Central Fin-
assemblage structure suggest that regional-scale filters land, Lapland, Northern Ostrobothnia, and West Finland, and
Oulanka Biological Station for analyzing the water samples.
(e.g., temperature; Tonn 1990, Poff 1997) have a major We are also indebted to two anonymous reviewers for valu-
influence on the distribution of taxa across headwater able comments on a previous draft of the manuscript, and to
streams. This pattern may have been further accentu- U. Viilo for sharing his wisdom on the natural history of
850 JANI HEINO ET AL. Ecological Applications
Vol. 13, No. 3

FIG. 4. Distribution and abundance of the 12 top indicator species (see Table 1) for the 10 TWINSPAN groups along
CCA axes 1 and 2. The smallest symbols indicate the absence of a species at a site, whereas the largest symbol size indicates
the peak abundance for that species.
June 2003 MACROINVERTEBRATE ASSEMBLAGE TYPES 851

stream invertebrates, especially worms, with us. This study assemblages of boreal headwater streams. Journal of the
is part of the Finnish Biodiversity Research Programme (FI- North American Benthological Society 21:397–413.
BRE). Financial support was also provided by the Maj and Jacobsen, D., R. Schultz, and A. Encalada. 1997. Structure
Tor Nessling Foundation. and diversity of stream invertebrate assemblages: the in-
fluence of temperature with altitude and latitude. Fresh-
LITERATURE CITED
water Biology 38:247–261.
Allen, T. F. H., and T. W. Hoekstra. 1992. Toward a unified Johnson, R. K., and W. Goedkoop. 2000. The use of bio-
ecology. University of Chicago Press, Chicago, Illinois, geographical regions for partitioning variance of littoral
USA. macroinvertebrate communities. Internationale Vereini-
Anderson, M. J., and A. Clements. 2000. Resolving envi- gung für theoretische und angewandte Limnologie, Ver-
ronmental disputes: a statistical method for choosing handlungen 27:333–339.
among competing cluster models. Ecological Applications Keddy, P. A. 1992. Assembly and response rules: two goals
10:1341–1355. for predictive community ecology. Journal of Vegetation
Angermeier, P. L., and I. J. Schlosser. 1995. Conserving Science 3:157–164.
aquatic biodiversity: beyond species and populations. Legendre, P., and L. Legendre. 1998. Numerical ecology.
American Fisheries Society Symposium 17:402–414. Second edition. Elsevier, Amsterdam, The Netherlands.
Angermeier, P. L., and M. R. Winston. 1999. Characterizing
Malmqvist, B., and P. O. Hoffsten. 2000. Macroinvertebrate
fish community diversity across Virginia landscapes: pre-
taxonomic richness, community structure and nestedness
requisite for conservation. Ecological Applications 9:335–
in Swedish streams. Archiv für Hydrobiologie 150:29–54.
349.
Malmqvist, B., and M. Mäki. 1994. Benthic macroinverte-
Austin, M. P. 1985. Continuum concept, ordination methods,
and niche theory. Annual Review of Ecology and System- brate assemblages in north Swedish streams: environmental
atics 16:39–61. relationships. Ecography 17:9–16.
Begon, M., J. L. Harper, and C. R. Townsend. 1996. Ecology. Malmqvist, B., S. Rundle, C. Brönmark, and A. Erlandsson.
Third edition. Blackwell Science, Oxford, UK. 1991. Invertebrate colonization of a new man-made stream
Belbin, L., and C. McDonald. 1993. Comparing three clas- in southern Sweden. Freshwater Biology 26:307–324.
sification strategies for use in ecology. Journal of Vege- Marchant, R., A. Hirst, R. H. Norris, R. Butcher, L. Metzeling,
tation Science 4:341–348. and D. Tiller. 1997. Classification and prediction of mac-
Brown, J. H. 1995. Macroecology. University of Chicago roinvertebrate assemblages from running waters in Victo-
Press, Chicago, Illinois, USA. ria, Australia. Journal of the North American Benthological
Cao, Y., D. P. Larsen, and R. Thorne. 2001. Rare species in Society 16:664–681.
multivariate analysis for bioassessment: some consider- Matthews, W. J. 1998. Patterns in freshwater fish ecology.
ations. Journal of the North American Benthological So- Chapman and Hall, New York, New York, USA.
ciety 20:144–153. Maurer, B. A. 1999. Untangling ecological complexity. The
Chessman, B. C. 1999. Predicting the macroinvertebrate fau- macroscopic perspective. University of Chicago Press, Chi-
nas of rivers by multiple regression of biological and en- cago, Illinois, USA.
vironmental differences. Freshwater Biology 41:747–757. McCune, B., and M. J. Mefford. 1999. PC-ORD. Multivariate
Diamond, J. M. 1975. Assembly rules of species commu- analysis of ecological data. MJM Software Design, Gle-
nities. Pages 342–344 in M. L. Cody and J. M. Diamond, neden Beach, Oregon, USA.
editors. Ecology and evolution of communities. Harvard McIntosh, R. P. 1995. H. A. Gleason’s ‘‘individualistic com-
University Press, Cambridge, Massachusetts, USA. munity concept’’ and theory of animal communities: a con-
Diamond, J. M., and M. E. Gilpin. 1982. Examination of the tinuing controversy. Biological Reviews 70:317–357.
‘‘null’’ model of Connor and Simberloff for species co- Moss, D., J. F. Wright, M. T. Furse, and R. T. Clarke. 1999.
occurrences on islands. Oecologia 52:64–74. A comparison of alternative techniques for prediction of
Drake, J. A. 1991. Community assembly mechanics and the the fauna of running water-sites in Great Britain. Fresh-
structure of an experimental species ensemble. American water Biology 41:167–181.
Naturalist 137:1–26. Muotka, T., R. Paavola, A. Haapala, M. Novikmec, and P.
Dufrene, M., and P. Legendre. 1997. Species assemblages Laasonen. 2002. Long-term recovery of stream habitat
and indicator species: the need for a flexible asymmetrical structure and benthic invertebrate communities from in-
approach. Ecological Monographs 67:345–366. stream restoration. Biological Conservation 105:243–253.
Furse, M. T. 2000. The application of RIVPACS procedures Naiman, R. J., H. Decamps, and M. Pollock. 1993. The role
in headwater streams—an extensive and important national of riparian corridors in maintaining regional biodiversity.
resource. Pages 79–92 in J. F. Wright, D. W. Sutcliffe, and Ecological Applications 3:209–212.
M. T. Furse, editors. Assessing the biological quality of
Ormerod, S. J., and R. W. Edwards. 1987. The ordination
fresh waters. Freshwater Biological Association, Amble-
and classification of macroinvertebrate assemblages in the
side, UK.
catchment of the River Wye in relation to environmental
Gauch, H. G. 1982. Multivariate analysis in community ecol-
ogy. Cambridge University Press, Cambridge, UK. factors. Freshwater Biology 17:533–546.
Hämäläinen, H., and P. Huttunen. 1996. Inferring the mini- Paavola, R., T. Muotka, and P. Tikkanen. 2000. Macroinver-
mum pH of streams from macroinvertebrates using weight- tebrate community structure and species diversity in humic
ed averaging regression and calibration. Freshwater Biol- streams of Finnish Lapland. Internationale Vereinigung für
ogy 36:697–709. theoretische und angewandte Limnologie, Verhandlungen
Hawkins, C. P., R. H. Norris, J. Gerritsen, R. M. Hughes, S. 27:2550–2555.
K. Jackson, R. K. Johnson, and R. J. Stevenson. 2000. Palmer, M. A., J. D. Allan, and C. A. Butman. 1996. Dispersal
Evaluation of the use of landscape classifications for the as a regional process affecting the local dynamics of marine
prediction of freshwater biota: synthesis and recommen- and stream benthic invertebrates. Trends in Ecology and
dations. Journal of the North American Benthological So- Evolution 11:322–326.
ciety 19:541–556. Poff, N. L. 1997. Landscape filters and species traits: towards
Heino, J., T. Muotka, R. Paavola, H. Hämäläinen, and E. mechanistic understanding and prediction in stream ecol-
Koskenniemi. 2002. Correspondence between regional de- ogy. Journal of the North American Benthological Society
lineations and spatial patterns in benthic macroinvertebrate 16:391–409.
 852 JANI HEINO ET AL. Ecological Applications
Vol. 13, No. 3

Reynoldson, T. B., D. M. Rosenberg, and V. H. Resh. 2001. Ward, J. V. 1998. Riverine landscapes: biodiversity patterns,
Comparison of models predicting invertebrate assemblages disturbance regimes, and aquatic conservation. Biological
for biomonitoring in the Fraser River catchment, British Conservation 83:269–278.
Columbia. Canadian Journal of Fisheries and Aquatic Sci- Weiher, E., and P. A. Keddy. 1995. The assembly of exper-
ences 58:1395–1410. imental wetland plant communities. Oikos 73:323–335.
Sandin, L., and R. K. Johnson. 2000. Ecoregions and benthic Whittier, T. R., R. M. Hughes, and D. P. Larsen. 1988. Cor-
macroinvertebrate assemblages of Swedish streams. Jour- respondence between ecoregions and spatial patterns in
nal of the North American Benthological Society 19:462– stream ecosystems in Oregon. Canadian Journal of Fish-
474. eries and Aquatic Sciences 45:1264–1278.
Shipley, B., and P. A. Keddy. 1987. The individualistic and Winston, M. R., and P. L. Angermeier. 1995. Assessing con-
community-unit concepts as falsifiable hypotheses. Vege- servation value using centers of population density. Con-
tatio 69:47–55. servation Biology 9:1518–1527.
Ter Braak, C. J. F. 1995. Ordination. Pages 91–173 in R. H. Wright, J. F., P. D. Armitage, and M. T. Furse. 1984. A pre-
G. Jongman, C. J. F. ter Braak, and O. F. R. van Tongeren, liminary classification of running water sites in Great Brit-
editors. Data analysis in community and landscape ecology. ain based on macroinvertebrate species and the prediction
Cambridge University Press, Cambridge, UK. of community types using environmental variables. Fresh-
Ter Braak, C. J. F., and P. Smilauer. 1998. Program CANOCO, water Biology 14:221–256.
version 4.0. Centre for Biometry, Wageningen, The Neth- Wright, J. F., M. T. Furse, and D. Moss. 1998. River clas-
erlands.
sification using invertebrates: RIVPACS applications.
Tonn, W. M. 1990. Climate change and fish communities: a
conceptual approach. Transactions of the American Fish- Aquatic Conservation: Marine and Freshwater Ecosystems
eries Society 119:337–352. 8:617–631.
Van Sickle, J., and R. M. Hughes. 2000. Classification Wright, J. F., D. Moss, R. T. Clarke, and M. T. Furse. 1997.
strengths of ecoregions, catchments, and geographic clus- Biological assessment of river quality using the new ver-
ters for aquatic vertebrates in Oregon. Journal of the North sion of RIVPACS (RIVPACS III). Pages 102–108 in P. J.
American Benthological Society 19:370–384. Boon and D. L. Howell, editors. Freshwater quality: defin-
Van Tongeren, O. F. R. 1995. Cluster analysis. Pages 174– ing the indefinable. Scottish Natural Heritage, Edinburgh,
212 in R. H. G. Jongman, C. J. F. ter Braak, and O. F. R. UK.
van Tongeren, editors. Data analysis in community and Zimmerman, G. M., H. Goetz, and P. W. Mielke, Jr. 1985.
landscape ecology. Cambridge University Press, Cam- Use of an improved statistical method for group compar-
bridge, UK. isons to study effects of prairie fire. Ecology 66:606–611.

View publication stats