Macrobenthic invertebrate richness and composition along a latitudinal gradient of European...

Macrobenthic invertebrate richness and compositionalong a latitudinal gradient of European glacier-fedstreams

EMMANUEL CASTELLA,* HAÂ KON ADALSTEINSSON,² JOHN E. BRITTAIN,³

GISLI M. GISLASON,§ ANTHONY LEHMANN,* VALERIA LENCIONI,±BRIGITTE LODS-CROZET,* BRUNO MAIOLINI,± ALEXANDER M. MILNER,**

JON S. OLAFSSON,§ SVEIN J. SALTVEIT³ and DEBORAH L. SNOOK**

*Laboratoire dÕEcologie et de Biologie Aquatique, UniversiteÂ de GeneÁve, GeneÁve, Switzerland

²National Energy Authority, GrensaÂvegur, Reykjavõ�k, Iceland

³Freshwater Ecology and Inland Fisheries Laboratory (LFI), The Natural History Museums, University of Oslo,

Blindern, Oslo, Norway

§Institute of Biology, University of Iceland, GrensaÂvegur, Reykjavõ�k, Iceland

±Museo Tridentino di Scienze Naturali, Trento, Italy

**School of Geography, The University of Birmingham, Edgbaston, Birmingham, U.K.

SUMMARY

1. The in¯uence of 11 environmental variables on benthic macroinvertebrate communities

was examined in seven glacier-fed European streams ranging from Svalbard in the north

to the Pyrenees in the south. Between 4 and 11 near-pristine reaches were studied on

each stream in 1996±97.

2. Taxonomic richness, measured at the family or subfamily (for Chironomidae) levels for

insects and higher levels for non-insects, increased with latitude from Svalbard (3 taxa) to

the Pyrenees (29 taxa).

3. A Generalized Additive Model (GAM) incorporating channel stability [Pfankuch Index

(PFAN)], tractive force, Froude number (FROU), water conductivity (COND), suspended

solids (SUSP) concentration, and maximum temperature explained 79% of the total

deviance of the taxonomic richness per reach. Water temperature and the PFAN of stability

made the highest contribution to this deviance. In the model, richness response to

temperature was positive linear, whereas the response to the PFAN was bell-shaped with

an optimum at an intermediate level of stability.

4. Generalized Additive Models calculated for the 16 most frequent taxa explained

between 25 (Tipulidae) and 79% (Heptageniidae) of the deviance. In 10 models, more than

50% of the deviance was explained and 11 models had cross-validation correlation ratios

above 0.5. Maximum temperature, the PFAN, SUSP and tractive force (TRAC) were the

most frequently incorporated explanatory variables. Season and substrate characteristics

were very rarely incorporated.

5. Our results highlight the strong deterministic nature of zoobenthic communities in

glacier-fed streams and the prominent role of water temperature and substrate stability

in determining longitudinal patterns of macroinvertebrate community structure. The

GAMs are proposed as a tool for predicting changes of zoobenthic communities in

glacier-fed streams under climate or hydrological change scenarios.

Correspondence: Emmanuel Castella, Laboratoire dÕEcologie et de Biologie Aquatique, University of Geneva,

18 ch. des Clochettes, CH-1206 Geneva. E-mail: [email protected]

Freshwater Biology (2001) 46, 1811±1831

Ó 2001 Blackwell Science Ltd 1811

Keywords: Alpine, Arctic, climate change, generalized additive models, glacial stream, macrobenthos

Introduction

Current climate change scenarios indicate proportion-

ally more detectable impacts at both high altitude and

latitudes (Roots, 1989; Beniston, Diaz & Bradley, 1997;

Beniston, 2000). The precise distribution of climate

change cannot be established, especially for complex

mountain regions, but in areas where Ôcoupled tran-

sitions in vegetation and precipitation patterns occur,

geomorphological systems may be near thresholds of

change and ecological systems may be more vulner-

ableÕ (Poff, 1992). Among the potentially sensitive

ecological systems, Grimm (1993) proposed streams

as models to examine the consequences of climate

changes. Indeed, running water systems and their

biota can be regarded as catchment-scale integrative

monitors for a set of hydrological, thermal and biotic

variables that might be modi®ed by climate change. It

follows that arctic and alpine running waters can be

regarded as research foci in such a context, and their

communities considered to be as much under threat

as terrestrial alpine communities (New, 1995).

Kryal streams and rivers dominated by glacial

¯ow provide comparatively the highest amount of

published information compared with other types of

alpine and arctic running waters (Ward, 1994). Kryal

zoobenthic communities appear to be largely con-

trolled by thermal and substrate stability variables.

Following the conceptual model by Milner & Petts

(1994), it seemed possible to model the longitudinal

distribution of macrofauna in glacial streams on the

basis of environmental variables. However, unlike

lowland streams in which climate change is thought

to lead to an increase in water temperature (Stefan &

Sinokrot, 1993; Hogg et al., 1995; Sinokrot et al., 1995),

the anticipated glacier ablation would result in a

decrease in stream water temperature and therefore

a downstream expansion of the kryal in¯uence and

associated fauna (McGregor et al., 1995). There might

be considerable variation however, depending on

glacier size and geographical location, and kryal

streams fed by small glacial areas might even

undergo a reduction in their kryal fauna.

The recent availability of powerful statistical tech-

niques has led to the current expansion of predictive

habitat distribution models (Guisan & Zimmermann,

2000). Generalized Additive Models (GAMs) are a

non-parametric extension of multiple regressions

and Generalized Linear Models (GLM) (Hastie &

Tibshirani, 1990). Generalizd Additive Models are

¯exible exploratory and modelling tools as they allow

for linear and non-linear response shapes, for both

continuous and factor variables, and for a combina-

tion of those within a single model. They are less

restrictive than classical linear regressions or GLMs

because they are more data than model driven. Yee &

Mitchell (1991) provided a comprehensive introduc-

tion to GAM modelling in ecology. Since then, GAMs

became more frequently used to model species

response to environmental variables, especially in

vegetation sciences (e.g. Leathwick, 1995; Heegard,

1997; Bio, Alkemade & Barendregt, 1998; Lehmann,

1998; Austin, 1999; but see Brosse & Lek, 2000 for ®sh

microhabitat modelling or Fewster et al., 2000 for

birds). Yee & Mitchell (1991) concluded that GAMs

are appropriate for modelling potential changes in

species distributions resulting from global warming.

The goal of this study was to develop predictive

models for macroinvertebrate taxonomic richness,

and the abundance of major macroinvertebrate taxa

along longitudinal/altitudinal gradients in glacial

rivers. Development of these models also represents

an attempt to test the validity of and add a quanti-

tative component to the conceptual model proposed

by Milner & Petts (1994). Generalized Additive

Models were applied in this study because their

data-driven smoothing regression technique has been

shown to provide improvements over classical regres-

sion models, especially because of the avoidance of

the a priori assumption of ®xed response shapes (Bio

et al., 1998). The data used for model development

originated from seven glacial streams covering a wide

European latitudinal gradient from Svalbard to the

Pyrenees (Brittain & Milner, 2001). They were studied

under a common protocol within the ÔArctic and

Alpine Stream Ecosystem ResearchÕ project (AASER)

(Brittain et al., 2000; Brittain & Milner, 2001).

Methods

The individual studies carried out under the AASER

project are detailed in Brittain et al. (2001), Gislason

et al. (2001), Lods-Crozet et al. (2001), Maiolini &

Lencioni (2001), and Snook & Milner (2001).

1812 E. Castella et al.

Ó 2001 Blackwell Science Ltd, Freshwater Biology, 46, 1811±1831

Reaches and sampling regime

The reaches were located in seven glacier-melt dom-

inated streams (Brittain & Milner, 2001). The glacial

streams formed a latitudinal and altitudinal gradient

across Europe from the Pyrenees in the south to

Svalbard in the north (Table 1). A common protocol

was used for determining the major geomorpholo-

gical, physical, chemical and biological components of

these streams.

Four to eight 15 m long reaches were de®ned in each

stream to represent the different sectors identi®ed on

the basis of valley and channel geomorphology. In

each stream, the ®rst reach was as close to the glacial

snout as possible. The second reach was typically

within 1000 m of the glacier snout and upstream of any

major tributary input. The downstream limit of the

study sector was where a fully developed invertebrate

community occurred, i.e. where Chironomidae were at

least associated with Ephemeroptera (Baetidae and

other families), Plecoptera (Nemouridae, Chloroperli-

dae and other families), and Trichoptera. This down-

stream limit was derived from preliminary surveys of

the streams. It was applied within zoogeographical

constraints to account for the fact that certain taxa are

absent from northern and arctic catchments.

At each reach, except Bayelva on Svalbard, ®eld

surveys were carried out at three time periods

during both 1996 and 1997: immediately post spring

snowmelt (June), in mid-summer during the ice melt

(August) and at low water level (September). These

time periods will be refered to as `seasons'. The

results obtained at a given reach and a given

sampling season served as units in the analyses.

These units will be referred to as 'reach-date'. The

snowpack precluded some samplings in upstream

reaches in June. As the ice free season on Svalbard is

short and because of logistic constraints, Bayelva was

sampled only during early July and late August

1997.

Geomorphological and environmental variables

Geomorphological description (width of valley ¯oor,

of all active channels, slope) was carried out at the

onset of the project. The stream bottom component of

the PfankuchÕs index (PFAN) (Pfankuch, 1975) was

used to assess channel stability by scoring ®ve

variables (rock angularity, bed-surface brightness,

Tab

le1

Lo

cati

on

and

char

acte

rist

ics

of

the

gla

cier

-fed

riv

ers

inv

esti

gat

edin

the

AA

SE

Rp

roje

ct

Riv

er

syst

emC

od

eR

egio

nC

o-o

rdin

ates

Max

.

alti

tud

eo

f

catc

hm

ent

(ma.

s.l.

)

Cat

chm

ent

area

at

do

wn

stre

am

reac

h(k

m2)

Gla

cier

area

(km

2)

Pre

cip

iati

on

(mm

)

1996

/19

97

Nu

mb

er

of

stu

dy

reac

hes

Dis

tan

ce

of

reac

hes

fro

m

gla

cier

(m)

Alt

itu

din

al

ran

ge

of

riv

erst

ud

ied

(ma.

s.l)

Dis

char

ge

ran

ge

(m3s±

1)

Tre

elin

e

alti

tud

e

(ma.

s.l.

)

Tai

llo

nT

AI

Py

ren

ees

43°0

6¢N

,0°

01¢W

2975

6.4

0.2

*4

50±1

500

1870

±250

00.

3±0.

416

00

Co

nca

CO

NS

.E.

Alp

s46

°06¢

N,

10°3

6¢E

3463

4.2

0.2

1497

735

0±46

0013

00±2

830

0.3±

2.1

2000

Mu

ttM

UT

N.

Alp

s46

°33¢

N,

8°24

¢E30

997

0.6

1595

55±

3600

1800

±260

00.

6±1.

520

00

Dal

elv

aB

RI

W.

No

rway

61°4

0¢N

,6°

50¢E

1915

25.6

22²

1270

510

0±71

0010

±340

0.5±

3560

0

Lei

run

gsaÊ

iL

EI

Cen

tral

61°2

4¢N

,8°

41¢E

2159

400

1.2

c.80

07

200±

2460

097

0±15

500.

4±6.

210

50

No

rway

W-J

ok

u Èls

aÂW

JOC

entr

al64

°50¢

N,

18°4

5¢W

1800

840

6866

211

5±45

000

160±

860

25±3

1N

otr

ees³

Icel

and

Bay

elv

aB

AY

Sp

itsb

erg

en,

78°5

5¢N

,11

°50¢

E74

230

.912

483

430

0±29

005±

500.

01±1

5.6

No

tree

s

Sv

alb

ard

*c.

300

mm

reco

rded

du

rin

gJu

lyan

dA

ug

ust

.

²Par

to

fa

larg

ep

late

aug

laci

erw

ith

anar

eao

f48

7k

m2.

³Th

en

atu

ral

tree

lin

e,w

ith

ou

tan

thro

po

gen

icim

pac

ts,

wo

uld

be

300±

400

ma.

s.l.

Macroinvertebrates in glacial streams 1813


particle packing, percentage stable materials, scour-

ing, presence and type of aquatic vegetation). Scores

were summed to provide an overall index of chan-

nel stability with a potential range of 15±70 (high

scores representing unstable channels at the reach

scale).

During a 5-day sampling period at each ®eld

survey, water temperature, level, discharge, conduc-

tivity and suspended solids (SUSP) were monitored at

minimum and maximum ¯ows on the upstream and

downstream reach. At each of four transects installed

at all reaches, the wetted channel width to a maxi-

mum depth of 0.5 m and depth/velocity pro®les were

determined. At the reach scale, average current

velocity and average depth were combined to calcu-

late an average Froude number (FROU, dimension-

less), according to Statzner et al. (1988):

FROU � U=�g �D�0:5

with U: mean current velocity, g: gravity acceleration,

D: mean water depth (from the depth pro®les).

The reach slope and water depth were combined to

calculate tractive force (TRAC, dyn cm±1) (shear

stress) according to Statzner et al. (1988):

TRAC � g � S �D � qwith g: gravity acceleration, S: reach slope, D: mean

water depth, q: water density.

Digital temperature loggers placed in streams

throughout the study period monitored water tem-

perature continuously at most reaches. Visual or man-

ual assessments at each point of the depth/velocity

pro®les were used to record the bed-sediment com-

position, which was expressed for the reach as the

percentage cover of four categories (boulders >20 cm,

coarse gravel 5±20 cm, ®ne gravels 0.2±5 cm, and ®ne

particles <0.2 cm). An index of substrate diversity

(SUDI) was calculated at the reach scale as SimpsonÕsdiversity index:

SUDI � 1=�Rp2i �

with pi: proportion of the ith substrate category.

Three stones were collected at random in each

reach at each sampling date, and benthic algae were

scraped off the upper surface (area 3 ´ 3 cm) and

washed onto a GF/C ®lter. Chlorophyll a (CHLO) was

then determined in the laboratory according to stand-

ard methods (APHA, 1992).

Eleven environmental variables were selected

among the available ®eld measurements for use in

the analyses. This selection was primarily based on the

necessity to reduce redundancy among the set of

explanatory variables within the regression proce-

dures. The correlation coef®cients for the 10 continu-

ous variables selected are provided in Table 2. Three

variables were associated to hydraulic energy and

channel stability (PFAN, TRAC, FROU), three to

substrate description [percentage cover by boulders

(BOUL), percentage cover by ®ne sediment (FINE),

and SUDI]. Conductivity (COND) served as an

integrated description of the longitudinal gradient

in each stream. The temperature variable retained

(TEMP) was the maximum temperature recorded

for one given reach over the entire study period.

Benthic CHLO was used as an indicator of available

food source for primary consumers. Suspended solids

Table 2 Pearson's correlation coef®cient (r) between the 10 explanatory environmental variables used in GAMs

Variable Code PFAN TRAC FROU BOUL FINE COND CHLO SUSP SUDI

Pfankuch index PFAN

Tractive force TRAC )0.41**

Froude number FROU )0.01 0.26**

% Boulders BOUL )0.58** 0.50** 0.12

% Fine sediment FINE 0.15* )0.32** )0.15 )0.32**

Conductivity COND )0.06 0.01 )0.17* 0.02 )0.12

Chlorophyll a CHLO 0.17* 0.17* 0.10 )0.10 )0.23** )0.33**

Suspended solids SUSP 0.14 )0.13 <0.01 )0.07 0.16* 0.17* )0.36**

Substrate diversity SUDI 0.14 0.02 )0.03 )0.31** 0.28** 0.01 0.02 )0.02

Temperature TEMP )0.68** 0.10 )0.08 0.26** )0.03 0.08 )0.17* )0.16* 0.12

**r signi®cantly different from 0 at P < 0.01.

*r signi®cantly different from 0 at P < 0.05.

No asterisk: r not signi®cantly different from 0 (P > 0.05).



(SUSP) was the average concentration measured for

each reach-date. Season (SEAS) was also included as a

discrete explanatory variable. After examination of

the distribution of the continuous variables, TRAC,

COND, CHLO and SUSP were log-transformed

to provide a more homogeneous spread for the

calculation of the response curves in the regression

models.

Invertebrate sampling

Within each sampling reach, 5±10 replicate kick sam-

ples were collected for invertebrates using a standard

pond net (30 ´ 30 cm) with a mesh size of 250 lm,

disturbing the substrate for a period of 30 s, within an

area of 30 ´ 30 cm. Densities in number of individuals

per m2 were averaged among the samples. In the case

of Western JoÈkulsaÂ (WJO), 10 stones were collected at

each reach and macroinvertebrates removed. The

surface area of each stone was then traced on a paper

and used to calculate invertebrate densities (Gislason,

OÂ lafsson & Adalsteinsson, 1998). All invertebrate

densities were log10(x + 1) transformed prior to statis-

tical analyses.

Invertebrate identi®cation was to the lowest level

possible. However, numerous young instars and taxa

dif®cult to identify (e.g. Chironomidae, Limnephili-

dae) made it dif®cult to carry out homogeneous

analyses at species or even genus level across all

reaches and dates. Hence, the subfamily or family

level for insects, and higher levels for non-insects,

were used throughout for constructing models.

Statistical analyses

Regressions of reach taxonomic richness and of indi-

vidual taxa density against environmental variables

were carried out using GAMs (Hastie & Tibshirani,

1990). Generalized Additive Models express the rela-

tionship between a response variable (Y) and depen-

dent variables (or predictors) (Xi) by:

Y � link linear predictor (LP)

and:

LP � a� Rfi�Xi� � e

Therefore, LP is the linear predictor. The ÔlinkÕfunction is a transformation used to accommodate

different response distributions (e.g. log for a Poisson

distribution, logit for a binomial one). a is a constant,

and e the error term. The fis are smooth functions,

estimated individually for each predictor by a scat-

terplot smoother. The models are additive in the

predictors effects (Hastie & Tibshirani, 1990).

Gneralized Additive Model calculations were car-

ried out in the S-PLUS software (Anonymous, 1998),

using a set of functions developed to perform

generalized regression analyses and spatial predic-

tion (GRASP; Lehmann, Leathwick & Overton, 1999).

A quasi-Poisson family was used for the response

variables (taxonomic richness or individual taxa

density) to accommodate under- or over-dispersed

Poisson data by estimating their dispersion parameter

instead of using the default value of 1 (Leathwick &

Austin, 2001). A cubic smoothing spline method was

chosen to smooth the continuous environmental

variables, using either one or three degrees of freedom

(d.f.). Stepwise model selection was used to select

variables to be retained in the ®nal model. Starting

from a model incorporating the more detailed

smoothing (d.f. � 3) for all the dependent variables

given as input, a stepwise selection procedure drops

variables or modi®es smoothing parameters for all

variables successively, dropping at each loop the

variable responsible for the lowest deviance reduc-

tion. The procedure stops when no further change can

be found that provides a signi®cant deviance reduc-

tion (Hastie & Tibshirani, 1990; Chambers & Hastie,

1993).

Diagnostic procedures for the GAMs included:

(1) F-test for non-parametric effects to test the

signi®cance of each selected variable (Hastie &

Tibshirani, 1990), (2) calculation of the percentage

of deviance explained by the models (an equivalent

of the coef®cient of determination in classical regres-

sion models), (3) contributions of each explanatory

variable expressed as a deviance reduction associ-

ated to dropping the variable from the model, (4)

correlation ratio (r1) between observed and predic-

tive values, (5) correlation ratio (r2) between

observed and predictive values derived from cross-

validation. In this last procedure, nine of ten of the

data were used to recalibrate a selected model,

which was then used to predict the response on

the remaining one of ten of the data. The procedure

was repeated 10 times, and cross-predicted responses

were compared with observed data to check the

stability of the model.



Results

Taxonomic richness

A set of 169 reach-dates (reaches sampled on differ-

ent dates) was obtained from the seven streams,

providing a total of 40 macroinvertebrate taxa collec-

ted from 1049 individual kick or stone samples.

Average densities for each taxon are given in

Appendices 1 and 2. Altitudinal and latitudinal

gradients of taxonomic richness are summarized in

Fig. 1. At only three reach-dates were no organisms

found (Svalbard and Iceland, at the uppermost

reaches WJO01 and BAY01). The taxonomically rich-

est reach was TAI35, the downstream reach in the

Pyrenees, with 28 taxa. Three distinct ÔregionsÕ can be

distinguished according to the number of taxa recor-

ded per site: the Alps and the Pyrenees (TAI ± 29

taxa, CON ± 28 taxa, MUT ± 23 taxa), Scandinavia

(BRI, LEI and WJO, each with 17 taxa), and Svalbard

(BAY ± 3 taxa).

Two outlying Icelandic reaches in¯uenced by geo-

thermal activity (WJO02 and WJO03) were omitted

from subsequent richness models because of the high-

est temperature maxima recorded in the entire data

set (14 and 18 °C, respectively) were associated with a

very low taxonomic richness (5 taxa).

Plots of taxonomic richness against the two major

variables incorporated in the Milner & Petts (1994)

model, maximum temperature and PFAN of channel

stability, revealed a clear trend of increasing richness

with increasing temperature, and a more complex

response to channel stability (Fig. 2). Richness peaked

between 30 and 35 PFAN units. Above 35 (i.e. in

the more unstable reaches examined in the present

study) a marked reduction in taxonomic richness was

observed with no values above 15 taxa. The variability

of the taxonomic richness per reach was also higher

below 35 PFAN units.

GAMs for taxonomic richness

Generalized Additive Model for taxonomic rich-

ness started with 11 variables, including season as

categorical variable (Table 3). Six variables (PFAN,

TRAC, FROU number, COND, suspended sediment

and maximum temperature) were retained in the

Fig. 1 Box-plots for the taxonomic richness per reach. The vertical boxes represent the inter-quartile range (Q25 ± Q75) around the

reach median (horizontal thick line). Upper and lower whiskers are drawn to the nearest value not beyond Q75 + 1.5(Q75 ± Q25)

and Q25 ± 1.5(Q75 ± Q25), respectively.



regression model (F-test, P £ 0.01), explaining 79% of

the total deviance of taxonomic richness. The valid-

ation diagnostics were high: r1 � 0.90 for simple

validation and r2 � 0.87 for cross-validation. Maxi-

mum water temperature was the most important

variable explaining variations in taxonomic richness

(®ve times higher than the PFAN). SUSP, COND,

FROU and TRAC were comparable but made low

contributions. The shape of the response curves

varied for the variables retained in the model

(Fig. 3). Tractive force, COND and temperature were

incorporated as linear functions (d.f. � 1 for the spline

smoother), while the PFAN, FROU and SUSP etc.

were more complex (d.f. � 3 for the spline smoother).

Tractive force, temperature and COND had a general

positive in¯uence on taxonomic richness, suspended

solids a negative one. Pfankuch index was incorpor-

ated with a sigmoid response and an optimum

between 30 and 40 units. For PFAN, FROU and SUSP,

the con®dence bands (Fig. 3) evidenced a lack of

accuracy in some parts of the gradients, because of a

lower number of data.

GAMs for individual taxa

Generalised Additive Model regressions were only

calculated for the 16 most frequent taxa. These were

the 16 taxa occurring in more than 24% of the 169

reach-dates (four plecopteran, two ephemeropteran,

two trichopteran, seven dipteran families and sub-

families, and oligochaetes).

Table 4 summarizes the environmental variables

kept in the ®nal regression models and their relat-

ive contributions. These variables were selected for

incorporation in the models at the P � 0.01 level. Ten

models explained more than half of the total devi-

ance of their respective taxon density, six of them

explained more than 60% deviance. Cross-validation

and simple validation correlation ratios were on

average high, with ®ve models above r2 � 0.7 for

cross-validation. Models for the families of Plecoptera,

Ephemeroptera and Trichoptera gave on average a

higher explained deviance and stability than those for

the families of Diptera and for the oligochaetes. The

model that explained the highest deviance (79.1%)

Fig. 2 Relationship between taxonomic richness and (i) the reach maximum temperature (a) (ii) the Pfankuch index of channel

stability (b) in seven glacial streams. The taxonomic richness in the 169 reach-dates is smoothed with a loess function.

Table 3 Signi®cance level, degree of

freedom of the smoother, and drop

contribution for the continuous signi®cant

(P < 0.05) variables included in the GAM

for taxonomic richness

Explanatory

variable Code F-value P

Smoother

(d.f.)

Drop

contribution

Pfankuch index PFAN 13.99 <0.001 3 34.0

Tractive force TRAC 0.49 <0.05 1 15.4

Froude number FROU 5.57 <0.01 3 12.4

Conductivity COND 0.96 <0.05 1 10.6

Suspended solids SUSP 3.09 <0.05 3 14.7

Temperature TEMP 5.10 <0.001 1 167.2



and had the highest cross-validation (r2 � 0.85) was

obtained for Heptageniidae. Models explaining a

deviance <30% and with cross-validation r2 < 0.5

were obtained for Diptera (Diamesinae, Simuliidae

and Tipulidae), and oligochaetes.

Maximum temperature and concentrations of sus-

pended solids were incorporated in 14 and 15 models,

respectively, of a total of 16. Temperature made the

greatest contribution to the model in 11 cases, whereas

suspended solids was the most important variable

only once. Other frequently incorporated variables

were PFAN and TRAC, both contributing signi®-

cantly to 10 models. Variables related to substrate

characteristics (BOUL, FINE, and SUDI) and SEAS,

were very rarely retained. The response curves for

the variables retained in the 16 models and their

con®dence bands (twice the standard error) are

presented in Fig. 4. Con®dence bands were usually

wider at both ends of all gradients where the density

of observations was lower. Regression models for the

non-Diptera insects all showed similar trends, all

incorporating a linear, or quasi-linear positive in¯u-

ence of TRAC (or FROU, for the Limnephilidae) and

more complex responses for PFAN of stability. The

contribution of temperature for the same groups was

also positively linear or quasi-linear (Leuctridae,

Baetidae, Heptageniidae), or showed an in¯exion

around 10 °C (Taeniopterygidae, Limnephilidae), or

between 10 and 15 °C (Nemouridae, Perlodidae,

Rhyacophilidae). Models for Chironomidae proved

very distinct from the non-Diptera insects. The model

for Orthocladiinae incorporated a negative linear

contribution of TRAC and FROU. The model for

Chironominae incorporated a negative linear response

to PFAN and a V-shaped response to FROU. The

regression model for Diamesinae was the most dis-

tinct; it did not incorporate TRAC, FROU or PFAN,

but variables associated to substrate characteristics

(BOUL and SUDI) and with an optimum temperature

response between 6 and 7 °C.

Relationships between the actual abundance of

individual taxa and the environmental variables were

back-calculated using the GAM general equation

presented in the statistical analyses section above.

These calculations were carried out for the four most

commonly selected variables in the GAM regressions

(maximum temperature, suspended sediments, TRAC

and PFAN) (Figs 5±8). Taxa were represented only

when the considered variable had been selected in the

GAM model. Each curve expresses the taxonÕs pre-

dicted response to a given environmental variable, in

the absence of in¯uence of the other variables selected

in the model. Examination of the curves allows a

comparison of the modelled sensitivity of the taxa to

the range of environmental change.

Response curves for temperature (Fig. 5) varied

from positively increasing (e.g. Heptageniidae, Baeti-

dae, Taeniopterygidae) to bell-shaped (e.g. Diamesi-

nae, Chironominae, Simuliidae). Rapid increases in

invertebrate densities were noticeable in the low

Fig. 3 Response functions for the reach taxonomic richness on the six environmental variables incorporated in the GAM. The dashed

lines are approximate 95% con®dence intervals around the smooth function. PFAN: Pfankuch Index of channel stability

(dimensionless), TRAC: tractive force (loge transformed dyn cm±1), FROU: Froude number (dimensionless), COND: water electrical

conductivity (loge transformed lS cm±1), SUSP: suspended solids (loge(x + 1) transformed mg L±1), TEMP: maximum reach

temperature (°C). Vertical axes are scaled according to the dimensionless linear predictor.



temperature range in Diamesinae and Orthocladiinae,

as was a clear shift in temperature preference among

the three chironomid subfamilies. Responses for TRAC

(Fig. 6) tended to have a similar positive trend, except

for Orthocladiinae, the abundance of which decrea-

sed with TRAC. Responses for PFAN (Fig. 7) proved

more varied, from increasing (e.g. Taeniopterygidae,

Baetidae) to decreasing (Oligochaeta, Chironominae),

and bell-shaped curves with an optima at 30±40

index units (Leuctridae, Heptageniidae, Rhyacophil-

idae, Empididae). Increasing concentrations of sus-

pended solids (Fig. 8) had generally strong negative

effects on abundance (Tipulidae, Heptageniidae,

Rhyacophilidae, Leuctridae, and Nemouridae). How-

ever, some taxa maintained occurrences over the

whole range of suspended solids concentrations

(Perlodidae, Baetidae, Orthocladiinae, Simuliidae,

Empididae), even when the abundance was high rela-

tive to the taxonÕs average abundance (Diamesiinae,

Limnephilidae). The only positive trend was obtained

for Limoniidae.

Discussion

The data presented here represent the ®rst published

attempt to compare and model taxonomic richness

among a set of glacier-fed streams distributed across

a wide latitudinal gradient. Comparative studies of

faunal assemblages and richness in similar streams

located at different latitudes are rare (Jacobsen,

Schultz & Encalada, 1997). Although described at a

relatively high taxonomic level and based upon a

limited set of streams, the observed latitudinal pattern

allowed the distinction of geographical ÔregionsÕ with

taxon richness increasing at lower latitudes. Less than

seven taxa were observed in the upstream reaches

near the glacial source in all seven streams.

Divergence between regions is manifest in the

downstream gradient of taxonomic enrichment. There

were more than 15 taxa in the Pyrenean and Alpine

streams, whereas 10±15 taxa were recorded in the

Scandinavian streams and less than ®ve on Svalbard.

The relative importance of factors responsible for the

increasing taxon richness towards lower latitudes is

still a matter of debate (Chown & Gaston, 2000),

although biogeographical factors such as isolation and

glaciations certainly play a role (Milner et al., 2001).

Our ®ndings are consistent with this general latitudi-

nal trend, and call for a further regionalization ofTab

le4

Nu

mb

ero

fca

ses

(n),

dro

pco

ntr

ibu

tio

ns

of

the

sele

cted

var

iab

les

and

dia

gn

ost

icp

aram

etre

so

fG

AM

sfo

rth

e16

mo

stfr

equ

ent

tax

on

om

icg

rou

ps.

Th

ev

aria

ble

sfo

rw

hic

h

no

dro

pco

ntr

ibu

tio

nis

giv

en,

wer

en

ot

kep

tin

the

mo

del

atth

eP

=0.

05le

vel

.S

eeT

able

s2

and

3fo

rth

ev

aria

ble

cod

es

Tax

on

nS

EA

SP

FA

NT

RA

CF

RO

UB

OU

LF

INE

CO

ND

CH

LO

SU

SP

SU

DI

TE

MP

Ex

pla

ined

dev

ian

ce(%

)

Val

idat

ion

(r1)

Cro

ss-

val

idat

ion

(r2)

Tae

nio

pte

ryg

idae

4718

.946

.813

.840

.148

.70.

650.

56

Nem

ou

rid

ae60

8.6

9.2

12.7

35.8

65.0

0.82

0.77

Leu

ctri

dae

4119

.25.

510

.511

.255

.80.

700.

64

Per

lod

idae

4712

.312

.24.

83.

14.

822

.969

.40.

840.

73

Bae

tid

ae62

19.7

24.5

4.1

13.9

50.7

64.9

0.81

0.75

Hep

tag

enii

dae

4639

.417

.29.

042

.579

.10.

890.

85

Lim

nep

hil

idae

713.

29.

44.

935

.845

.10.

600.

49

Rh

yac

op

hil

idae

438.

15.

01.

83.

117

.975

.40.

820.

68

Dia

mes

inae

161

3.0

1.6

5.6

13.8

26.1

0.58

0.49

Ort

ho

clad

iin

ae14

53.

13.

95.

82.

948

.850

.70.

760.

72

Ch

iro

no

min

ae59

8.3

9.0

11.8

11.2

31.8

56.4

0.72

0.58

Sim

uli

idae

999.

635

.129

.70.

520.

47

Em

pid

idae

666.

54.

15.

98.

714

.65.

23.

98.

567

.30.

820.

70

Tip

uli

dae

488.

05.

320

.325

.10.

420.

32

Lim

on

iid

ae46

5.9

24.4

16.4

3.4

16.2

55.9

0.74

0.63

Oli

go

chae

ta93

8.5

7.9

6.0

10.5

10.1

26.9

0.53

0.39



Fig. 4 Response functions for the log-transformed density of the 16 most frequent taxa on the environmental variables incorporated in

their respective GAMs. The dashed lines are approximate 95% con®dence intervals around the smooth function. PFAN: Pfankuch

Index of channel stability (dimensionless), TRAC: tractive force (loge transformed dyn cm±1), FROU: Froude number (dimensionless),

COND: water electrical conductivity (loge transformed lS cm±1), SUSP: suspended solids (loge(x + 1) transformed mg L±1),

TEMP: maximum reach temperature (°C), SEAS: `season', BOUL: percentage cover by boulders, FINE: percentage cover by ®ne

sediment (<0.2 cm), CHLO: chlorophyll a concentration (loge(x + 1) transformed mg m±2), SUDI: dimensionless index of substrate

diversity. Vertical axes are scaled according to the dimensionless linear predictor.



Fig. 4 (Continued).



models for kryal river communities. Such models

should be developed within the context of more

re®ned arctic and alpine ecoregions and their associ-

ated taxonomic pools. Lindegaard & Brodersen (1995)

described such a regionalization of taxonomic pools in

alpine and subalpine streams at the species level for

Chironomidae.

Unlike Brosse & Lek (2000) who found limitations

and shortcomings in the use of GAMs for modelling

roach (Rutilus rutilus L.) microhabitats in lakes, the

models produced here, both for total taxonomic

richness, and for individual taxa, displayed on average

high levels of explanation and stability. Among the 16

GAMs for individual taxa, the average percentage of

explained deviance was 53% (range: 25±79%). This

compares favourably with the values of 15% (range:

3±72%) and 13% (range: 1±66%) obtained by Bio et al.

(1998) for two series of GAMs of the occurrence of 120

wetland plants. Correlation ratios over 0.7 obtained

here in the cross-validation procedures con®rmed the

stability that can be reached for some of the models. As

underlined by Peeters & Gardeniers (1998) in the case

of logistic regression models, Ôweak modelsÕ (i.e.

models accounting for a small proportion of explained

deviance) can result either from the model being

inadequate (i.e. selected explanatory variables are

irrelevant for the given taxon), or from the taxa having

a wide ecological tolerance for the environmental

factor considered. In the present study, the four

weakest models for individual taxa explained between

25 and 30% of the total deviance. These four taxa

(Diamesinae, Simuliidae, Tipulidae and Oligochaeta)

had the highest level of ubiquity among the taxa

sampled, at least over the temperature and PFAN

gradients (Milner et al., 2001). For example, Diamesi-

nae (occurring in 95% of the reach-dates) maintained

Fig. 5 Response functions on the maximum reach temperature for the abundance of the 14 taxa for which temperature was

incorporated in the GAM. Abundance is expressed as the density of individuals per m2 on a log10 scale. Responses are smoothed with

loess functions.



densities between 100 and 10 000 individuals m±2

across these two gradients. However, the model for

Orthocladiinae (occurring in 86% of the reach-dates,

and also covering almost the complete range of

temperature and PFAN) reached more than 50% of

explained deviance, with good validation criteria,

notably on account of a strong relative contribution

of temperature to their GAM. Indeed, the case of the

four ÔweakerÕ models may indicate that the taxonomic

level considered here is too imprecise to allow accurate

Fig. 6 Response functions on tractive force (loge scale) for the

abundance of the 10 taxa for which tractive force was incor-

porated in the GAM. Abundance is expressed as the density of

individuals per m2 on a log10 scale. Responses are smoothed

with loess functions.

Fig. 7 Response functions on Pfankuch's channel stability index

(dimensionless) for the abundance of the 10 taxa for which the

index was incorporated in the GAM. Abundance is expressed as

the density of individuals per m2 on a log10 scale. Responses are

smoothed with loess functions.



modelling. For example, Diamesinae as a subfamily

may not respond to the temperature or longitudinal

gradient, although individual species do (Kownacka &

Kownacki, 1975; Rossaro, 1991). This is certainly also

the case for oligochaetes, although less documented at

high altitudes (Ward, 1994).

Examination of the response curves for taxonomic

richness and individual taxa conform with existing

knowledge about the ecology of these organisms.

A clear case is the shift of the optimum towards

higher temperature in the Chironomidae, where the

sequence Diamesinae/Orthocladiinae/Chironominae

is consistent with results obtained by Rossaro (1991)

or Lindegaard & Brodersen (1995) and others. Chan-

nel stability (as measured by the PFAN) led to the

most diversi®ed response shapes in the present study,

illustrating the advantage of smoothed curves over

traditional linear models in their capability to account

for varied responses (e.g. the plateau-shaped respons-

es of Perlodidae). Bell-shaped responses to channel

stability of Leuctridae, Rhyacophilidae, Heptagenii-

dae, and total taxonomic richness, appear to support

the intermediate disturbance hypothesis (Ward &

Stanford, 1983). In these cases, highest densities or

richness were observed around the middle of the

potential range of the stability index. However, data

about detailed relationships between environmental

variables and the density of taxa are mostly not

available for glacier-fed streams. For example, sus-

pended sediment was incorporated in the taxonomic

richness GAM and in all taxa GAMs but one. The

response curves for this variable were strongly neg-

ative in 11 taxa, con®rming the detrimental effects of

high levels of suspended solids on large parts of the

aquatic fauna, although empirical data documenting

the effects are still limited (Ward, 1992).

Fig. 8 Response functions on suspended solid concentration (loge scale) for the abundance of the 15 taxa for which suspended

solid concentration was incorporated in the GAM. Abundance is expressed as the density of individuals per m2 on a log10 scale.

Responses are smoothed with loess functions.



In general, the models obtained here supports the

idea that the fauna of glacial streams is strongly

controlled by physical constraints (Milner & Petts,

1994; Milner et al., 2001). This strong physical con-

trol probably helps to explain the less successful

modelling attempts with GAMs where habitat com-

plexity and biotic interactions play more prominent

roles (Brosse & Lek, 2000). However, Saether (1968)

and Flory & Milner (1999, 2000) suggested that under

certain conditions interspeci®c competition may

in¯uence Chironomidae succession and community

structure even in glacial streams.

Season was very rarely incorporated in the models

as an explanatory variable. However, although year-

round faunal monitoring of alpine streams are seldom

(Lavandier, 1979; Lavandier & DeÂ camps, 1984), evi-

dence suggests that sampling outside the late spring ±

late summer period brings a different picture of glacial

stream biota (FuÈ reder et al., 2001; SchuÈ tz et al., 2001).

Results of the modelling approach provide a means

for evaluating the conceptual framework proposed by

the Milner & Petts (1994). First, the signi®cance of the

predictive models implemented here con®rms the

existence of signi®cant longitudinal trends in macro-

invertebrate community structure at the family/sub-

family level. Secondly, the high frequency of

incorporation in the GAMs, and high contribution to

deviance reduction, of maximum temperature, PFAN

and TRAC, strongly con®rm the primary importance

of temperature and channel stability in shaping

richness and composition of macrobenthic assem-

blages in glacier-fed streams. Suspended sediments

were not initially considered as an important deter-

minant in Milner & Petts (1994) framework. Incorpor-

ation of this factor in all but one of the models

generated here and the differences in sensitivity

exhibited by different taxa point to its value as a

predictive variable. One other notable departure from

Milner & PettsÕs (1994) model in terms of longitudinal

distribution is the observation of taxa, such as Oligo-

chaetae, Orthocladiinae, Tipulidae, Simuliidae, and

Taeniopterygidae, at lower temperatures than predic-

ted, that is in reaches where the maximum tempera-

ture recorded did not exceed 2 °C.

In the early 1990s, it was recognized that predic-

tions of the effects of climate change on the ecology of

running waters were highly speculative (Oswood,

Milner & Irons III, 1992; Poff, 1992; Ward et al., 1992).

However, if the downscaling improvement of climatic

scenarios advocated by Beniston, Diaz & Bradley

(1997) actually leads to more accurate regional

anticipations of climatic and glacier trends, then the

predictive models elaborated here could serve as a

basis for the forecast of changes in the associated

glacial stream communities. In turn, these models

could be used to forecast and map potential changes

in suitable habitats for a given type of kryal commu-

nity under different scenarios of environmental chan-

ges (as addressed by Sinokrot et al. (1995) for ®sh

habitats). Furthermore, the models could serve to

de®ne reference conditions for glacial streams affected

by other types of impacts (e.g. water abstraction)

(Petts & Bickerton, 1994; McGregor et al., 1995). Such

developments require the development of local rela-

tionships between projected trends in glacier dynam-

ics and the concomitant variations in the variables

retained in the models, such as suspended sediment

concentrations and water temperature.

To further the development and validation of

predictive models for glacial river macroinvertebrate

communities, there would be a need to incorporate

hydrological variables describing exchanges or rela-

tions between surface- and groundwater. Such vari-

ables have been shown to in¯uence the stream

conditions and temperature (Malard, Tockner &

Ward, 1999; Ward et al., 1999). As suggested by Death

(1995), it would also be necessary to test the inver-

tebrate response to stability over a narrow range of the

other variables, and especially temperature.

Regarding the biotic variables considered for pre-

diction, the species-level models mentioned earlier are

currently under elaboration, but the implementation

of trait- or attribute-based models, an approach

pioneered by Snook & Milner (in press) for glacial

streams, might be a way to circumvent limitations

associated with the biogeography of species, and to

provide more general models for the sensitivity of

functional groups of taxa.

The models developed here, concern only kryal

rivers. However, groundwater or snow-fed tributaries

can contribute greatly to the overall biodiversity of

glacial catchments (Ilg et al., 2001; Zah & Uehlinger,

2001), but are predicted to react differently to climate

change than kryal streams (McGregor et al., 1995). It is

therefore a need to encompass all types of water

supply and their associated fauna in predictive

models if predictive scenarios are to be developed at

the catchment scale.



Acknowledgments

The project, Arctic and Alpine Stream Ecosystem

Research (AASER), was ®nanced by the European

Commission (project ENV-CT95-0164) and the Swiss

Federal Of®ce for Education and Science (project BBW

95.0430). We thank M. Gessner for valuable comments

that improved the clarity of the text. Jessica MuÈ ller-

Castella kindly edited some of the graphics.

References

Anonymous (1998) S-PLUS 4.5. Data Analysis Products

Division, MathSoft Inc., Seattle, Washington.

APHA (1992) Standard methods for the examination of

water and wastewater. 18th edn of the American Public

Health Association, 15th Street, NW, Washington DC,

1422 pp.

Austin M.P. (1999) The potential contribution of veget-

ation ecology to biodiversity research. Ecography, 22,

465±484.

Beniston M. (2000) Environmental Change in Mountains

and Uplands. Arnold, London.

Beniston M., Diaz H.F. & Bradley R.S. (1997) Climatic

change at high-elevation sites: an overview. Climatic

Change, 36, 233±251.

Bio A.M.F., Alkemade R. & Barendregt A. (1998) Deter-

mining alternative models for vegetation response

analysis: a non-parametric approach. Journal of Veget-

ation Science, 9, 5±16.

Brittain J.E., Adalsteinsson H., Castella E., Gislason G.M.,

Lencioni V., Lods-Crozet B., Maiolini B., Milner A.M.,

Petts G.E. & Saltveit S.J. (2000) Towards a conceptual

understanding of arctic and alpine streams. Verhandlu-

gen der Internationalen Vereinigung fuÈ r Theoretische und

Angewandte Limnologie, 27, 740±743.

Brittain J.E. & Milner A.M. (2001) Ecology of glacier-fed

rivers: current status and concepts. Freshwater Biology,

46, 1571±1578.

Brittain J.E., Saltveit S.J., Castella E., Bogen J., Bosnes T.,

Blakar I., Bremnes T., Haug I. & Velle G. (2001) The

macroinvertebrate communities of two contrasting

Norwegian glacial rivers in relation to environmental

variables. Freshwater Biology, 46, 1723±1736.

Brosse S. & Lek S. (2000) Modelling roach (Rutillus

rutillus) microhabitat using linear and nonlinear tech-

niques. Freshwater Biology, 44, 441±452.

Chambers J.M. & Hastie T.J. (1993) Statistical Models in S.

Chapman & Hall, London.

Chown S.L. & Gaston K.J. (2000) Areas, cradles and

museums: the latitudinal gradient in species richness.

Trends in Ecology and Evolution, 15, 311±315.

Death R.G. (1995) Spatial patterns in benthic invertebrate

community structure: products of habitat stability or

are they habitat speci®c? Freshwater Biology, 33, 455±467.

Fewster R.M., Buckland S.T., Siriwardena G.M., Baillie S.R.

& Wilson J.D. (2000) Analysis of population trends for

farmland birds using generalized additive models.

Ecology, 81, 1970±1984.

Flory E.A. & Milner A.M. (1999) The role of competition

in invertebrate community development in a recently

formed stream in Glacier Bay National Park, Alaska.

Aquatic Ecology, 33, 175±184.

Flory E.A. & Milner A.M. (2000) Macroinvertebrate com-

munity succession in Wolf Point Creek, Glacier Bay

National Park, Alaska. Freshwater Biology, 44, 465±480.

FuÈ reder L., SchuÈ tz C., Wallinger M. & Burger R. (2001)

Physico-chemistry and acquatic insects of a glacier-fed

and a spring-fed alpine stream. Freshwater Biology, 46,

1673±1690.

Gislason G.M., Adalsteinsson H., Hansen I., Adalsteins-

son H. & Svavarsdottir K. (2001) Longitudinal changes

in macroinvertebrate assemblages along a glacial river

system in central Iceland. Freshwater Biology, 46, 1737±

1751.

Gislason G.M., OÂ lafsson J.S. & Adalsteinsson H. (1998)

Animal communities in Icelandic rivers in relation to

catchment characteristics and water chemistry ± pre-

liminary results. Nordic Hydrology, 29, 129±148.

Grimm N.B. (1993) Implications of climate change for

stream communities. In: Biotic Interactions and Global

Change (Eds P.M. Kareiva., J.G. Kingsolver & R.B. Huey),

pp. 293±314. Sinauer Associates Inc, Sunderland.

Guisan A. & Zimmermann N.E. (2000) Predictive habitat

distribution models in ecology. Ecological Modelling,

135, 147±186.

Hastie T.J. & Tibshirani R.J. (1990) Generalized Additive

Models. Chapman & Hall, London.

Heegard E. (1997) Ecology of Andreaea in western

Norway. Journal of Bryology, 19, 527±636.

Hogg I.D., Williams D.D., Eadie J.M. & Butt S.A. (1995)

The consequences of global warming for stream

invertebrates: a ®eld simulation. Journal of Thermal

Biology, 20, 199±206.

Ilg C., Castella E., Lods-Crozet B. & Marmonier P. (2001)

Water physico-chemistry and invertebrate drift in

tributaries of a glacial stream (the Mutt, Switzerland).

Archiv fuÈ r Hydrobiologie, 151, 335±352.

Jacobsen D., Schultz R. & Encalada A. (1997) Structure

and diversity of stream invertebrate assemblages: the

in¯uence of temperature with altitude and latitude.

Freshwater Biology, 38, 247±261.

Kownacka M. & Kownacki A. (1975) Gletscherbach-

ZuckmuÈ cken der OÈ tztaler Alpen in Tirol (Diptera:



Chironomidae: Diamesinae). Entomologica Germanica,

2, 35±43.

Lavandier P. (1979) Ecologie dÕun torrent pyreÂneÂen de

haute montagne: lÕEstaragne. Doctoral Thesis, Univer-

sity Paul Sabatier, Toulouse.

Lavandier P. & DeÂcamps H. (1984) Estaragne France. In:

Ecology of European Rivers (Ed. B.A. Whitton), pp. 237±

264. Blackwell Scienti®c Publications Inc, Palo Alto,

California.

Leathwick J.R. (1995) Climatic relationships of some New

Zealand forest tree species. Journal of Vegetation Science,

6, 237±248.

Leathwick J.R. & Austin M.P. (2001) Competitive inter-

actions between tree species in New-ZealandÕs old-

growth indigenous forests. Ecology, 82, 2560±2573.

Lehmann A. (1998) GIS modeling of submerged macro-

phyte distribution using Generalized Additive Models.

Plant Ecology, 139, 113±124.

Lehmann A., Leathwick J.R., & Overton J.Mc. (1999)

GRASP userÕs manual. Landcare Research. Hamilton,

New-Zealand.

Lindegaard C. & Brodersen K.P. (1995) Distribution of

Chironomidae (Diptera) in the river continuum. In:

Chironomids: from Genes to Ecosystems (Ed. P. Cranston),

pp. 258±271. CSIRO Publications, Melbourne, Australia.

Lods-Crozet B., Castella E., Cambin D., Ilg C., Knispel S.

& Mayor-SimeÂ ant H. (2001) Macroinvertebrate com-

munity structure in relation to environmental variables

in a Swiss glacial stream. Freshwater Biology, 46, 1641±

1661.

Maiolini B. & Lencioni V. (2001) Longitudinal distribu-

tion of macroinvertebrate assemblages in a glacially

in¯uenced stream system in the Italian Alps. Freshwa-

ter Biology, 46, 1765±1775.

Malard F., Tockner K. & Ward J.V. (1999) Shifting

dominance of subcatchment water sources and ¯ow

paths in a glacial ¯oodplain, Val Roseg, Switzerland.

Arctic, Antarctic and Alpine Research, 31, 135±150.

McGregor G., Petts G.E., Gurnell A.M. & Milner A.M.

(1995) Sensitivity of alpine stream ecosystems to

climate change and human impacts. Aquatic Conserva-

tion: Marine and Freshwater Ecosystems, 5, 233±247.

Milner A.M., Brittain J.E., Castella E. & Petts G.E. (2001)

Trends of macroinvertebrate community-structure in

glacier-fed rivers in relation to environmental condi-

tions: a synthesis. Freshwater Biology, 46, 1833±1846.

Milner A.M. & Petts G.E. (1994) Glacial rivers: physical

habitat and ecology. Freshwater Biology, 32, 295±307.

New T.R. (1995) Introduction to Invertebrate Conservation

Biology. Oxford University Press, Oxford.

Oswood M.W., Milner A.M. & Irons J.G. III (1992)

Climate change and Alaskan rivers and streams.

In: Global Climate Change and Freshwater Ecosystems

(Eds P. Firth & S.G. Fisher), pp. 192±210. Springer-

Verlag, NewYork.

Peeters E.T.H.M. & Gardeniers J.J.P. (1998) Logistic

regression as a tool for de®ning habitat requirements

of two common gammarids. Freshwater Biology, 39,

605±615.

Petts G.E. & Bickerton M.A. (1994) In¯uence of water

abstraction on the macroinvertebrate community gra-

dient within a glacial stream system: La Borgne

dÕArolla, Valais, Switzerland. Freshwater Biology, 32,

375±386.

Pfankuch D.J. (1975) Stream Reach Inventory and Channel

Stability Evaluation. US Department of Agriculture

Forest Service, Region 1, Missoula, Montana.

Poff N.L. (1992) Regional hydrologic response to climate

change: an ecological perspective. In: Global Climate

Change and Freshwater Ecosystems (Eds P. Firth & S.G.

Fisher), pp. 88±115. Springer-Verlag, New-York.

Roots E.F. (1989) Climate change: high latitude regions.

Climatic Change, 15, 223±253.

Rossaro B. (1991) Chironomids and water temperature.

Aquatic Insects, 13, 87±98.

Saether O.A. (1968) Chironomids of the Finse Area,

Norway, with special reference to their distribution in

a glacial brook. Archiv fuÈ r Hydrobiologie, 64, 426±483.

SchuÈ tz C., Wallinger M., Burger R. & FuÈ reder L. (2001)

Effects of snow cover on the benthic fauna in a glacier-

fed stream. Freshwater Biology, 46, 1691±1704.

Sinokrot B.A., Stefan H.G., McCormick J.H. & Eaton J.G.

(1995) Modeling of climate change effects on stream

temperatures and ®sh habitats below dams and near

groundwater inputs. Climatic Change, 30, 181±200.

Snook D.L. & Milner A.M. (2001) The in¯uence of glacial

runoff on stream macroinvertebrate communities in

the Taillon catchment, French PyreÂneÂes. Freshwater

Biology, 46, 1609±1623.

Snook D.L. & Milner A.M. Biological traits of macroin-

vertebrates and hydraulic conditions in a glacier-fed

catchment (French PyreÂneÂes). Archiv fuÈ r Hydrobiologie,

in press12 .

Statzner B., Gore J.A. & Resh V.H. (1988) Hydraulic

stream ecology: observed patterns and potential

applications. Journal of the North American Benthological

Society, 7, 307±360.

Stefan H.G. & Sinokrot B.A. (1993) Projected global

climate change impact on water temperatures in ®ve

North Central US streams. Climatic Change, 24, 353±381.

Ward J.V. (1992) Aquatic Insect Ecology. 1. Biology and

Habitat. John Wiley & Sons, Inc, New York.

Ward J.V. (1994) Ecology of alpine streams. Freshwater

Biology, 32, 277±294.

Ward J.V., Malard F., Tockner K. & Uehlinger U. (1999)

In¯uence of ground water on surface water conditions



in a glacial ¯ood plain of the Swiss Alps. Hydrological

Processes, 13, 277±293.

Ward J.V. & Stanford J.A. (1983) The intermediate

disturbance hypothesis: an explanation for biotic

diversity patterns in lotic ecosystems. In: Dynamics of

Lotic Ecosystems (Eds T.D. Fontaine & S.M. Bartell),

pp. 347±356. Ann Arbor Science Publishers, Ann

Arbor, Michigan.

Ward A.K., Ward G.M., Harlin J. & Donahoe R. (1992)

Geological mediation of stream ¯ow and sediment and

solute loading to stream ecosystems due to climate

change. In: Global Climate Change and Freshwater Eco-

systems (Eds P. Firth & S.G. Fisher), pp. 116±142.

Springer-Verlag, New-York.

Yee T.W. & Mitchell N.D. (1991) Generalized additive

models in plant ecology. Journal of Vegetation Science, 2,

587±602.

Zah R. & Uehlinger U. (2001) Particulate organic matter

inputs to a glacial stream eco-system in the Swiss Alps.

Freshwater Biology, 46, 1597±1608.

(Manuscript accepted 5 September 2001)



Ap

pen

dix

1A

ver

age

den

siti

es(i

nd

ivid

ual

s.m

)2)

for

the

tax

on

om

icg

rou

ps

inth

eP

yre

nea

n(T

AI)

and

Alp

ine

(CO

N,

MU

T)

stu

dy

reac

hes

.A

ver

ages

are

calc

ula

ted

ov

eral

lk

ick

sam

ple

s(5

±10)

and

six

sam

pli

ng

dat

es

Tax

on

TA

I01

TA

I02

TA

I03

TA

I35

CO

N00

CO

N02

CO

N03

CO

N04

CO

N08

MU

T01

MU

T02

MU

T03

MU

T04

MU

T05

Tae

nio

pte

ryg

idae

2291

9816

31.

428

1064

192

Nem

ou

rid

ae7

3417

5010

319

33

5131

Cap

nii

dae

9574

Leu

ctri

dae

324

275

786

0.2

0.8

81.

5

Ch

loro

per

lid

ae6

819

390.

21.

17

Per

lid

ae2

Per

lod

idae

118

1.3

316

940.

22

161.

5

Ep

hem

erel

lid

ae

Sip

hlo

nu

rid

ae

Bae

tid

ae0.

492

1216

228

379

401

822

887

1697

Hep

tag

enii

dae

1135

616

3610

727

314

7925

Lim

nep

hil

idae

1563

255

1021

415

817

Rh

yac

op

hil

idae

428

410

412

1.5

7

Th

rem

mat

idae

12

Ser

ico

sto

mat

idae

0.7

Elm

idae

322

5

Hy

dra

enid

ae3

Hel

op

ho

rid

ae0.

60.

2

Dia

mes

inae

9614

0822

4015

613

0476

971

2324

1669

630

1421

9417

68

Ort

ho

clad

iin

ae1.

512

481

180

1332

748

241

623

41.

736

346

48

Ch

iro

no

min

ae0.

97

8948

330

156

630.

90.

4

Tan

yp

od

inae

0.7

0.2

1.5

3

Sim

uli

idae

7328

629

107

164

5516

40.

46

2

Th

aum

alei

dae

1.7

1.5

0.2

0.4

253

Ble

ph

aric

erid

ae1.

15

86

Em

pid

idae

619

713

613

8624

77

6043

227

28

Dix

idae

0.2

0.2

0.5

31.

5

Tip

uli

dae

7535

811

3

Lim

on

iid

ae1.

412

2112

3026

3242

39

An

tho

my

idae

0.5

0.2

Ath

eric

idae

34

54

Do

lich

op

od

idae

0.2

0.2

Psy

cho

did

ae0.

20.

2

Cer

ato

po

go

nid

ae0.

43

0.2

Oth

erD

ipte

ra1.

12

33

Tri

clad

ida

0.7

2622

1329

55

1.4

0.6

1.9

Oli

go

chae

ta1.

524

810

257

112

4464

0.2

21.

5

Hy

dra

cari

na

0.4

41.

81.

54

4

Ost

raco

da

1.1

1.1

Nem

ato

da

50.

795

1.8

72

Tax

on

om

icri

chn

ess

69

2128

718

2624

252

1216

2121



Ap

pen

dix

2A

ver

age

den

siti

es(i

nd

ivid

ual

sm

)2)

for

the

tax

on

om

icg

rou

ps

inth

eS

can

din

avia

nst

ud

yre

ach

es(i

ncl

ud

ing

Sv

alb

ard

±B

AY

).A

ver

ages

are

calc

ula

ted

ov

er

all

kic

ko

rst

on

esa

mp

les

(5±1

0)an

dsi

xsa

mp

lin

gd

ates

(ex

cep

tS

val

bar

dw

ith

on

lytw

od

ates

)

Tax

on

BR

I

01

BR

I

02

BR

I

03

BR

I

04

BR

I

05

LE

I

01

LE

I

02

LE

I

03

LE

I

05

LE

I

06

WJO

01

WJO

15

WJO

02

WJO

03

WJO

35

WJO

04

WJO

05

WJO

06

WJO

1C

WJO

2C

WJO

2B

BA

Y

01

BA

Y

02

BA

Y

03

BA

Y

04

Tae

nio

pte

ryg

idae

0.4

194

217

0.4

Nem

ou

rid

ae1.

31.

13

1.1

323

Cap

nii

dae

52

2831

1511

1815

Leu

ctri

dae

0.4

0.7

Ch

loro

per

lid

ae

Per

lid

ae

Per

lod

idae

20.

40.

41.

1

Ep

hem

erel

lid

ae0.

41.

1

Sip

hlo

nu

rid

ae0.

40.

8

Bae

tid

ae4

55

8

Hep

tag

enii

dae

Lim

nep

hil

idae

0.7

41.

915

69

108

Rh

yac

op

hil

idae

0.4

0.7

2

Th

rem

mat

idae

Ser

ico

sto

mat

idae

Elm

idae

Hy

dra

enid

ae

Hel

op

ho

rid

ae

Dia

mes

inae

789

2669

1180

2239

1095

854

1613

1572

129

159

1319

812

7712

2019

7379

919

7121

8937

419

8971

855

168

Ort

ho

clad

iin

ae10

7630

188

693

1940

510

1980

513

1618

627

932

337

840

8920

7943

8770

7741

218

11

Ch

iro

no

min

ae17

5614

417

311

2070

2

Tan

yp

od

inae

1.4

0.5

1.9

Sim

uli

idae

1.5

5834

1316

3046

5923

24

0.5

6510

976

1959

Th

aum

alei

dae

Ble

ph

aric

erid

ae

Em

pid

idae

619

137

Dix

idae

Tip

uli

dae

80.

913

127

53

15

Lim

on

iid

ae7

45

An

tho

my

idae

Ath

eric

idae

Do

lich

op

od

idae

Psy

cho

did

ae

Cer

ato

po

go

nid

ae0.

92

18

Oth

erD

ipte

ra1.

83

20

Tri

clad

ida

1.1

Oli

go

chae

ta51

511

3234

0.7

636

521

3353

6966

781

248

Hy

dra

cari

na

0.4

1.5

0.4

34

6612

614

288

Ost

raco

da

1930

22

Nem

ato

da

91

36

1742

5015

Tax

on

om

icri

chn

ess

512

713

112

39

1715

22

55

310

1312

611

42

23



Macrobenthic invertebrate richness and composition along a latitudinal gradient of European...

Documents

Transcript of Macrobenthic invertebrate richness and composition along a latitudinal gradient of European...