Responsiveness of Great Lakes Wetland Indicators to HumanDisturbances at Multiple Spatial Scales: A Multi-Assemblage Assessment

John C. Brazner1,*,†, Nicolas P. Danz2, Anett S.Trebitz1, Gerald J. Niemi2, Ronald R. Regal2, Tom Hollenhorst2, George E. Host2, Euan D. Reavie2, Terry N. Brown2,

JoAnn M. Hanowski2, Carol A. Johnston3, Lucinda B. Johnson2, Robert W. Howe4, and Jan J.H. Ciborowski5

1U.S. Environmental Protection AgencyOffice of Research and Development

National Health and Environmental Effects Research LaboratoryMid-Continent Ecology Division

6201 Congdon BoulevardDuluth, Minnesota 55804-2495

2Center for Water and the EnvironmentNatural Resources Research Institute

University of Minnesota Duluth5013 Miller Trunk Highway

Duluth, Minnesota 55811-1442

3Center for Biocomplexity StudiesSouth Dakota State University153A Shepherd Hall, Box 2022

Brookings, South Dakota 57007-0869

4Cofrin Center for BiodiversityLaboratory Science Building

2420 Nicolet DriveUniversity of Wisconsin-Green Bay

Green Bay, Wisconsin 54311

5Great Lakes Institute for Environmental ResearchUniversity of Windsor

Windsor, Ontario N9B 3P4

ABSTRACT. Developing indicators of ecosystem condition is a priority in the Great Lakes, but little isknown about appropriate spatial scales to characterize disturbance or response for most indicators. Wesurveyed birds, fish, amphibians, aquatic macroinvertebrates, wetland vegetation, and diatoms at 276coastal wetland locations throughout the U.S. Great Lakes coastal region during 2002–2004. Weassessed the responsiveness of 66 candidate indicators to human disturbance (agriculture, urban devel-opment, and point source contaminants) characterized at multiple spatial scales (100, 500, 1,000, and5,000 m buffers and whole watersheds) using classification and regression tree analysis (CART). Non-stressor covariables (lake, ecosection, watershed, and wetland area) accounted for a greater proportionof variance than disturbance variables. Row-crop agriculture and urban development, especially at

J. Great Lakes Res. 33 (Special Issue 3):42–66Internat. Assoc. Great Lakes Res., 2007

*Corresponding author. E-mail: johnbrazner@eastlink.ca†Present address: Inland Waters Institute, 29 Powers Drive, HerringCove, NS, B3V 1G6 Canada; Phone: 902.446.5342.

42

Wetland Indicators Response to Human Disturbance 43

INTRODUCTION

One of the key steps in the development of eco-logical indicators is the assessment of their re-sponse to human disturbance. An important anddifficult aspect of assessing indicator response isdetermining the spatial scale to characterize humandisturbance. Disturbance must be measured at theappropriate spatial and temporal scales to achievean accurate picture of the stressor-response relation-ship (e.g., Richards and Johnson 1998, Mensing etal. 1998, Wang et al. 2003a). While there have beenexceptions (e.g., Roth et al. 1996, Strayer et al.2003), assessment of indicator response has gener-ally been conducted with disturbance characterizedat a single spatial scale, usually driven by conve-nience of available data sources, logistic con-straints, or particular interests of researchers oragencies (e.g., Steedman 1988, Reeves et al. 1993,O’Connor et al. 1996, Knutson et al. 1999, Walserand Bart 1999, Schleiger 2000, Sharma and Hilborn2000).

Local habitat features and physical setting havebeen considered the template for organizing biolog-ical communities by some ecologists for manyyears (Southwood 1977) and much effort has beenfocused on how local habitat structure influencesecological dynamics and organizes biological com-munities (e.g., Gorman and Karr 1978, Eadie andKeast 1984, Minshall 1988, Benson and Magnuson1992, Townsend and Hildrew 1994). There has alsobeen considerable effort, particularly for stream andother aquatic ecosystems, to understand howprocesses functioning at watershed and broaderscales influence ecosystem structure and function(Johnston et al. 1990, Schlosser 1991, Fisher 1994,Poff and Allan 1995, Stevenson 1997). These con-trasting views have led to a dichotomy where dis-turbance characterization has often been at asite-specific scale when habitat alteration has been

of primary interest, or at a watershed scale when thefocus has been on the influence of land use change.However, it is clear that aquatic ecosystems are hi-erarchically organized (Frissell et al. 1986,Maxwell et al. 1995, Poff 1997), that species re-spond to their environments at multiple spatialscales (e.g., Kotliar and Wiens 1990, Roland andTaylor 1997, Cushman and McGarigal 2002, Hol-land et al. 2004, Stoffels et al. 2005), and that thelocation and intensity of the stress in a species’ en-vironment plays a critical role in defining the ob-served response (Roland and Taylor 1997,Sponseller et al. 2001, Townsend et al. 2003, Kinget al. 2005).

Recognition of the hierarchical nature of organi-zation and scale dependencies in streams and otheraquatic ecosystems has led to attempts to assess therelative importance of disturbances characterized atdifferent spatial scales to aquatic biota. Many ofthese studies examined the role of riparian bufferzones and the influence of disturbance in thosezones relative to processes or disturbances charac-terized at the watershed scale (e.g., Barton et al.1985, Jones et al. 1996, Lee et al. 2001). Others fo-cused more specifically on determining the scale atwhich a particular disturbance seemed to be havingits greatest impact on a few key components of theecosystem, or on biological integrity (e.g., Spon-seller et al. 2001, DeLuca et al. 2004). Dependingon the types of organisms examined, results fromthese studies have varied from finding that localstresses were most important (e.g., Wang and Lyons2003, King et al. in press), to finding that cumula-tive disturbance across a landscape appears to playa more important role (e.g., Allan et al. 1997).

The disturbance scale to which an indicator re-sponds appears to depend on several factors, includ-ing the type of disturbance. Because land usedisturbances are amenable to characterization

larger spatial scales, were about equally influential and were more explanatory than a contaminantstress index (CSI). The CSI was an important predictor for diatom indicators only. Stephanodiscoiddiatoms and nest-guarding fish were identified as two of the most promising indicators of row-crop agri-culture, while Ambloplites rupestris (fish) and Aeshna (dragonflies) were two of the strongest indicatorsof urban development. Across all groups of taxa and spatial scales, fish indicators were most responsiveto the combined influence of row-crop and urban development. Our results suggest it will be critical toaccount for the influence of potentially important non-stressor covariables before assessing the strengthof indicator responses to disturbance. Moreover, identifying the appropriate scale to characterize distur-bance will be necessary for many indicators, especially when urban development is the primary distur-bance.

INDEX WORDS: Coastal wetlands, classification and regression trees, community, Great Lakes,multi-assemblage.

44 Brazner et al.

across a range of scales and have been a primaryfocus of efforts to understand the relative role oflocal and regional processes, these disturbancetypes are compared most often in multi-scale stud-ies. Some of these studies have focused on the ef-fects of forest fragmentation (e.g., Barton et al.1985, Roland and Taylor 1997, Brazner et al. 2005,Holland et al. 2005b), but more have focused onagricultural practices and urbanization. The evi-dence is far from unequivocal (e.g., see Lammertand Allan 1999, Sponseller et al. 2001, Snyder etal. 2003, Townsend et al. 2003), but it appears thatagricultural land uses, such as row-crops and pas-tures, often exert their primary influence at thewhole watershed scale (Roth et al. 1996, Wang etal. 1997, Lyons et al. 2000, Wang and Lyons2003a), while disturbance associated with urbaniza-tion may be more important immediately adjacentto a watercourse (Allen and O’Connor 2000,Wanget al. 2001, Wang et al. 2003b, DeLuca et al. 2004).

The relative importance of the disturbance scalesto which indicators respond also seems to dependon the type of indicator (e.g., Mensing et al. 1998,Lammert and Allan 1999, Strayer et al. 2003).There is considerable evidence that nutrient andwater quality related variables are most influencedby disturbance characterized at the watershed or re-gional scales (e.g., Richards et al. 1996, Allan et al.1997, Herlihy et al. 1998), although the relative im-portance of these scales can vary with the particularwater quality variable being measured (e.g., John-son et al. 1997, King et al. 2005). Among biota,birds and fish seem more likely to respond to broad-scale disturbance (Allen et al. 1999, Strayer et al.2003), while diatoms, wetland vegetation, and in-vertebrates often respond to more local features(Richards et al. 1997, Allen et al. 1999, Fitzpatricket al. 2001, Johnson et al. 2004, King et al. 2004,Galatowitsch et al. 1999b) and amphibians to multi-ple scales (Lehtinen et al. 1999). Differences in theway species respond to their environments arelikely controlled by relationships between mobilityand body size (Roland and Taylor 1997, Townsendet al. 2003, Holland et al. 2005a). Larger organismstend to have larger home ranges and therefore mayexperience the environment across larger spatialscales (Wiens and Milne 1989, Holling 1992,O’Neill et al. 1997).

Most evidence about scale-dependent responsesto disturbance in aquatic ecosystems comes fromstream and lake studies, but there has been somerelevant work in wetlands. For example, severalstudies have examined the importance of riparian

buffers adjacent to wetlands (e.g., Castelle et al.1994, Burke and Gibbons 1995, Haig et al. 1998,Detenbeck et al. 2002, Semlitsch and Bodie 2003)and the relative roles of riparian and landscape fea-tures (Fairbairn and Dinsmore 2001, Riffel et al.2003). Studies suggest that both local and largerscale disturbances can influence a variety of wet-land biota in important ways (Findlay and Houla-han 1997, Lehtinen et al. 1999, Findlay andBourdages 2000, Pope et al. 2000), but little isknown about the importance of particular scales ofinfluence other than to say that land use distur-bances as far as 5 km away from a wetland can besignificant and that response varies by intensity andtype of disturbance, as well as by ecosystem and or-ganism type (Galatowitsch et al. 1999a, Whited etal. 2000). Systematic comparisons among water-shed and local disturbance influences have rarelybeen completed for wetlands (Mensing et al. 1998,DeLuca et al. 2004), and better understanding oftheir relative roles awaits further study.

Developing indicators of ecosystem condition hasonly recently been made a priority in the GreatLakes (Environment Canada and U.S. EPA 2003)and for most indicators little is known about appro-priate spatial scales for either disturbances or re-sponses (Niemi et al. 2004, Niemi and McDonald2004). However, there has been a concerted effortrecently to develop and test indicators for GreatLakes coastal wetlands (Wilcox et al. 2002, Albertand Minc 2004, Uzarski et al. 2004, Lawson 2004).Earlier (Brazner et al. 2007), as part of a larger pro-ject to develop indicators of ecological conditionfor coastal ecosystems in the Great Lakes (Niemi etal. 2004, Danz et al. 2005), we examined the rela-tive importance of geographic, geomorphic, andhuman disturbance on indicators from a wide rangeof wetland taxa (diatoms, wetland vegetation,macroinvertebrates, amphibians, fish, birds). As afollow-up to that study, we examine the scales ofdisturbance to which ecological indicators forcoastal wetlands respond. Here, spatial scale refersto the geographic area or extent within which dis-turbance was characterized rather than grain, whichis a separate aspect of spatial scale characterization(Holland et al. 2005a). Our overall objective was toassess the responsiveness of candidate indicators tothree of the primary land use disturbances (agricul-ture, urbanization, and point source contaminants)across the Great Lakes basin. Specifically, our goalswere to; 1) determine which disturbance type hadthe most influence on indicators and assemblagesafter accounting for variability associated with

Wetland Indicators Response to Human Disturbance 45

other non-stressor covariables, 2) provide an ap-proach for determining the spatial scale where thedisturbance-response relationship was maximizedfor particular indicators, biotic assemblages, anddisturbance types, and 3) identify the candidate in-dicators with the best potential as indicators of par-ticular types of human disturbance in coastalwetlands across the Great Lakes.

Although we expected considerable variationamong indicators and assemblages (Strayer et al.2003) based on studies conducted in streams, wehypothesized that most indicators would be moreresponsive to agricultural land use at the watershedscale, and urbanization at more local scales (e.g.,Roth et al. 1996, Wang et al. 2003b,Wang andLyons 2003, Townsend et al. 2003). There is littleprecedence (see Strayer et al. 2003) for developinghypotheses related to scales for contaminant indica-tors, but the point-source nature of this disturbancetype suggests that local effects might be most im-portant. We also expected there would be similari-ties in the scale of response related to body size andmobility, with larger-bodied, mobile taxa most re-sponsive to disturbance characterized at broaderscales (e.g., Roland and Taylor 1997, Allen et al.1999, Johnson et al. 2004, Holland et al. 2005a).However, diatoms may be an exception to the mo-bility-body size relationship because, althoughsmall, they respond directly to water chemistrywhich is generally determined by large-scale uplandcharacteristics such as soil type and agriculture.Therefore, diatom indicators may be most respon-sive to disturbance characterized within the largerrather than smaller buffers. Finally, we expectedthat in some cases factors such as watershed area orgeographic region would have at least as great aninfluence on indicator response as land use distur-bance (Strayer et al. 2003, King et al.2005, Brazneret al. 2007).

METHODS

Data Sources, Survey Methods, andIndicator Selection Criteria

The data sources, survey methods, and indicatorselection criteria used for this paper were describedby Brazner et al. 2007. Briefly, the data were col-lected as part of a basin-wide study to develop indi-cators using a single integrated conceptualframework (Niemi et al. 2004, Danz et al. 2005,Johnston et al. in press). Abundance informationwas collected on bird, fish, amphibian, aquatic

macroinvertebrate, wetland vegetation, and diatomassemblages using standard methodologies at 276coastal wetland locations along U.S. shorelinesthroughout each of the Great Lakes.

Benthic and sedimented diatoms were sampledfrom 65 wetlands on natural substrates from 0.5 to3 m depth and processed as described by Reavie etal. (2006). Surface sediments were sampled using apush corer and core tube or with a Ponar sampler inunconsolidated bottom substrates. In rocky areas,algal material was scraped from the surface of rocksand pebbles and collected in vials as epilithic sam-ples. Vegetation was sampled in 90 wetlands from 1m2 quadrats distributed along randomly placed tran-sects within emergent and wet meadow areas(Bourdaghs et al. in press). Transect length and tar-get number of sample plots were determined in pro-portion to the size of the wetland to be sampled (20plots/60 ha, minimum transect length = 40 m, mini-mum plots/site = 8, average plots/site = 21) andplants were identified to the lowest taxonomic divi-sion possible. Cover was estimated visually foreach taxon using modified Braun-Blanquet coverclasses. Macroinvertebrates were sampled in 75wetlands from the three dominant habitats as deter-mined from shoreline and nearshore substrate type,extent and composition of riparian and aquatic veg-etation, and anthropogenic impacts. Samples werecollected along two to six transects set perpendicu-lar to depth contours, depending on the size of thewetland. Samples were collected with D-framednets (250 µm mesh) at the midpoint of two depthzones along each transect, the emergent zone (de-fined as depths less than 50 cm) and the submergentzone (depths 50–100 cm). Bird surveys were con-ducted by trained observers (Hanowski and Niemi1995) at 223 wetlands during June and early July in2000, 2001, and 2002 using the Marsh MonitoringWorkshop wetland breeding bird survey protocol(Ribic et al. 1999, and see www.bsc-eoc.org/mmp-main.html). We conducted amphibian calling sur-veys at most of the same points (n = 211) sampledfor birds, following guidelines outlined by theMarsh Monitoring Program (Weeber and Val-lianatos 2000, and see www.bsc-eoc.org/mmp-main.html). Fish were sampled with bothboat-mounted electrofishing gear (electro-fish) andfyke-nets (fyke-fish) (Trebitz et al. 2007). The twomethods were used by separate field crews thatoverlapped at 35 sites. Fyke-nets were fished at thesame wetlands where macroinvertebrates were sam-pled and a few additional sites (n = 80), while elec-trofishing was completed at 58 sites. Although six

46 Brazner et al.

ecosystem types were sampled as part of the largerproject (Danz et al. 2005, Johnston et al. in press),for this paper we focus on responses within threewetland types (river-influenced, protected, andopen-coastal). Sites were spread across the GreatLakes and approximately evenly distributed amongthe Laurentian Mixed Forest Ecoprovince in thenorthern lakes and the Eastern Broadleaf ForestEcoprovince in the southern lakes (Fig. 1, Keys etal. 1995). Sites were selected using a stratified ran-dom design to span multiple human disturbancegradients (Danz et al. 2005).

Eight to ten candidate indicators for each assem-blage were selected by appropriate lead co-authors(Table 1) based on an initial examination three tofour times that number for most assemblages. The

initial list was reduced to the final candidates thatwe examine here by using criteria recommended inHughes et al. (1998) and O’Connor et al. (2000).These included that indicators 1) were known orthought to be responsive to human disturbance, 2)represented ecologically important species or func-tions, 3) were sampled effectively (low intra-sitevariance), and 4) were of limited redundancy withother indicators. Similar measures of structuralcomplexity (e.g., richness, abundance) and func-tional character (e.g., mobility, reproductive strate-gies) were included whenever possible. Data setsfor fish captured with fyke-nets (fyke-fish) and fishcaptured from an electrofishing boat (electro-fish)were obtained by different field teams and kept sep-arate for these analyses because previous analyses

FIG. 1. Map of sampling locations in the Great Lakes for each of the assemblages (some samplingpoints have been moved slightly to reduce overlap and improve clarity).

Wetland Indicators Response to Human Disturbance 47T

AB

LE

1.

Cla

ssif

icat

ion

an

d re

gres

sion

tree

res

ult

s fo

r al

l in

dica

tors

, in

clu

din

g ov

eral

l var

ian

ce e

xpla

ined

by

the

mod

el (

R2 )

an

d th

epr

opor

tion

al r

edu

ctio

n i

n e

rror

(pa

rtia

l r2

equ

ival

ents

- i

n p

aren

thes

es)

asso

ciat

ed w

ith

eac

h o

f th

e pr

edic

tor

vari

able

s. P

redi

ctor

var

i-ab

les

orde

red

by e

ntr

y po

siti

on i

n t

he

tree

an

d re

pres

ent

the

mos

t pr

edic

tive

var

iabl

e at

eac

h n

ode.

Bec

ause

th

e fi

rst

fou

r am

phib

ian

indi

cato

rs w

ere

base

d on

pre

sen

ce-a

bsen

ce d

ata,

res

ult

s w

ere

deri

ved

from

cla

ssif

icat

ion

rat

her

than

reg

ress

ion

tree

s. R

C =

pro

port

ion

row

-cro

p ag

ricu

ltu

re,

DE

V =

pro

port

ion

dev

elop

ed l

and,

CS

I =

con

tam

inan

t-st

ress

in

dex,

su

bscr

ipts

in

dica

te t

he

spat

ial

scal

e ov

erw

hic

h s

tres

sor

vari

able

s w

ere

char

acte

rize

d, a

rrow

s in

dica

te d

irec

tion

of

rela

tion

ship

bet

wee

n i

ndi

cato

r an

d pr

edic

tor

vari

able

s. C

ases

wh

ere

pred

icto

r va

riab

les

are

repe

ated

in

dica

te t

hat

th

e va

riab

le w

as u

sed

at m

ore

than

on

e sp

lit

poin

t in

th

e tr

ee. F

or l

ake

subs

crip

ts,

E =

Eri

e, H

= H

uro

n, M

= M

ich

igan

, O =

On

tari

o, a

nd

S =

Su

peri

or;

for

wet

lan

d ty

pe s

ubs

crip

ts P

= p

rote

cted

, R =

riv

erin

e, C

= o

pen

coas

tal;

see

text

in m

eth

ods

for

expl

anat

ion

of

ecos

ecti

on s

ubs

crip

ts).

Indi

cato

rO

vera

ll R

2P

redi

ctor

Var

iabl

es (

part

ial

r2)

Dia

tom

s(n

= 6

5 w

etla

nds)

Shan

non-

Wie

ner

Div

ersi

ty (

ln)

0.46

Lak

e (O

,H)(

.14)

↓; R

C_5

000(

.19)

↓; L

ake (

E,M

)(.0

6)↓;

RC

_100

(.07

)↑%

Mot

ile0.

31C

SI_5

000(

.18)

↑; R

C_W

atsh

ed(.

08)↑

; DE

V_W

atsh

ed(.

05)↑

% P

lank

toni

c0.

21R

C_W

atsh

ed(.

14)↑

; Wet

_Are

a(.0

7)↑

Lan

ge-B

erta

lot I

ndex

0.58

Eco

sect

ion (

SCG

,WSU

, NG

L,W

UA

,SL

C,C

TP)

(.19

)↑; C

SI_5

000(

.12)

↓; W

at_A

rea(

.08)

↓;D

EV

_500

0(.0

9)↓;

RC

_500

0(.1

0)↓

Tro

phic

Dia

tom

Ind

ex0.

53L

ake (

S,H

,M,O

)(.2

9)↓;

Wat

_Are

a(.1

6)↑;

DE

V_1

00(.

08)↑

% S

teph

anod

isco

ids

0.50

RC

_Wat

shed

(.50

)↑%

Ach

nant

hidi

um m

inut

issi

mum

com

plex

0.40

DE

V_W

atsh

ed(.

13)↓

; Lak

e (E

,M,S

,O)(

.13)

↓; D

EV

_100

0(.1

4)↑

% S

taur

osir

a/St

auro

sire

lla/

Pse

udos

taur

osir

a0.

40D

EV

_500

(.11

)↓; E

cose

ctio

n (E

OL

,CT

P,SG

L,S

CG

,SL

C,W

UA

)(.1

1)↓;

DE

V_5

000(

.18)

↑%

Coc

cone

is0.

23L

ake (

O)(

0.23

)↑%

Pla

noth

idiu

m0.

24D

EV

_500

0(.1

3)↑;

Wet

_Are

a(.1

1)↑

Wet

land

Veg

etat

ion

(n =

90

wet

land

s)Sp

ecie

s ri

chne

ss (

num

ber

of ta

xa)

0.39

Eco

sect

ion (

SCG

,WSU

,CT

P,SG

L,E

OL

,SL

C)(

.14)

↓; D

EV

_500

0(.1

0)↓;

Wet

_Are

a(.0

9)↑;

DE

V_1

00(.

06)↑

% I

nvas

ive

taxa

0.75

Lak

e (O

,E)(

.44)

↑; R

C_W

atsh

ed(.

17)↑

; DE

V_5

000(

.06)

↑; W

at_A

rea(

.08)

↑%

Nat

ive

taxa

0.51

RC

_Wat

shed

(.36

)↓; W

at_A

rea(

.09)

↓; W

et_A

rea(

.06)

↓%

Wet

land

obl

igat

e ta

xa0.

29L

ake (

S,O

)(.2

9)↑

% c

over

Car

ex s

tric

ta0.

62D

EV

_Wat

shed

(.20

)↓; D

ev_1

000(

.32)

↑; D

ev_1

00(.

10)↓

% c

over

Car

ex la

sioc

arpa

0.49

RC

_500

0(.2

4)↓;

Wet

_Are

a(.1

5)↑;

Dev

_100

(.11

)↑%

cov

er S

parg

aniu

m e

uryc

arpu

m0.

47E

cose

ctio

n (SC

G,W

SU,N

GL

,SG

L,C

TP,

EO

L,S

LC

)(.2

0)↓;

Wet

_Are

a(.1

3)↓;

Wet

_Are

a(.0

9)↓;

Wet

_Are

a(.0

5)↑

% c

over

Phr

agm

ites

aus

tral

is0.

33R

C_5

000(

.33)

↑%

cov

er T

ypha

ang

usti

foli

a an

d T

ypha

x g

lauc

a0.

58L

ake (

O)(

.24)

↑; W

et_A

rea(

.15)

↑; D

ev_5

000(

.10)

↑; R

C_1

000(

.10)

↑

(Con

tiue

d)

48 Brazner et al.T

AB

LE

1.

Con

tin

ued

.

Indi

cato

rO

vera

ll R

2P

redi

ctor

Var

iabl

es (

part

ial

r2)

Mac

roin

vert

ebra

tes

(n =

75

wet

land

s)%

Bur

row

ers

0.57

RC

_Wat

shed

(.15

)↑; D

EV

_100

0(.1

7)↓;

Eco

sect

ion (

SCG

, NG

L,T

HP,

SLC

)(.0

9)↑;

Wat

_Are

a(.1

0)↓;

DE

V_W

atsh

ed(.

06)↑

% C

linge

rs0.

44L

ake (

S,O

)(.1

6)↑;

DE

V_1

000(

.07)

↑; D

EV

_500

(.07

)↑; W

at_A

rea(

.05)

↑; D

EV

_100

0(.0

9)↓

% P

reda

tors

0.21

Wat

_Are

a(.1

5)↑;

RC

_500

(.06

)↓%

Ins

ect fi

lter-

gath

erer

s0.

61D

EV

_Wat

shed

(.11

)↓;W

at_A

rea(

.08)

↓; D

EV

_100

(.07

)↓; D

EV

_500

(.06

)↑;

DE

V_W

atsh

ed(.

08)↓

; Wat

_Are

a(.0

8)↓;

Eco

sect

ion (

NG

L,S

SU,S

LC

)(.1

3)↑

Low

est t

axon

omic

uni

t ric

hnes

s0.

43W

et_T

ype (

C)(

.21)

↓; L

ake (

E,M

,H,S

)(13

)↓; W

et_A

rea.

(.09

)↑%

Cae

nis

spp.

0.49

Wat

_Are

a(.2

4)↓;

Wet

_Are

a(.1

3)↑;

Eco

sect

ion (

NG

L,S

SC,S

SU)(

.05)

↓; W

at_A

rea(

.07)

↓%

Coe

nagr

ion/

Ena

llag

ma

spp.

0.28

Lak

e (M

,H,O

,E)(

.14)

↓; W

et_A

rea(

.14)

↓%

Oec

etis

spp.

0.39

Wet

_Typ

e (C

)(.1

4)↑;

RC

_500

(.08

)↓; L

ake (

O,M

,E)(

.06)

↑; W

et_A

rea(

.11)

↓9%

Pro

cloe

on /C

alli

baet

issp

p.0.

16D

EV

_100

(.10

)↓; E

cose

ctio

n (T

HP,

NG

L,S

SU,S

LC

,WSU

)(.0

6)↑

% A

eshn

a sp

p.0.

49D

EV

_Wat

shed

(.16

)↓; D

EV

_100

0(.3

3)↑

Am

phib

ians

(n =

211

wet

land

s)Sp

ecie

s ri

chne

ss0.

19D

EV

_500

0(.1

3)↓;

Lak

e (M

,H,O

)(.0

6)↑

Spec

ies

rich

ness

of

tree

fro

gs0.

43E

cose

ctio

n (E

OL

,CT

P,SC

G,S

GL

,NSU

)(.2

9)↓;

Lak

e (S,

E,M

)(.1

4)↓

Spec

ies

rich

ness

of

Ran

ids

0.35

Lak

e (O

)(.2

2)↑;

DE

V_5

000(

.06)

↓;W

et_A

rea(

.07)

↑Sp

ecie

s ri

chne

ss o

f ea

rly

spri

ng b

reed

ers

0.19

Eco

sect

ion (

SSU

,WSU

,NSU

,SG

L,E

OL

)(.0

7)↓;

Wet

_Are

a(.0

7)↑;

RC

_100

0(.0

5)↓

Pres

ence

-abs

ence

of

Buf

o am

eric

anus

0.13

Eco

sect

ion (

SSU

, SG

L, T

HP,

EO

L, N

SU, S

LC

)(.1

3)↓

Pres

ence

-abs

ence

of

Ran

a cl

amit

ans

0.14

DE

V_5

000(

.09)

↓; L

ake (

O)(

.05)

↑Pr

esen

ce-a

bsen

ce o

f H

yla

vers

icol

or0.

25L

ake (

S,H

,M,O

)(.1

5)↑;

Eco

sect

ion (

SSU

,WSU

, NG

L,T

HP,

SLC

,EO

L)(

.10)

↑Pr

esen

ce-a

bsen

ce o

f P

seud

acri

s cr

ucif

er0.

42E

cose

ctio

n (E

OL

,CT

P,SC

G,S

GL

,NSU

)(.1

8)↓;

DE

V_1

00(.

06)↓

; RC

_100

0(.0

9)↓;

D

EV

_Wat

shed

(.09

)↓

Fis

h-

elec

trofi

shin

g(n

= 5

7 w

etla

nds)

Nat

ive

spec

ies

rich

ness

0.43

Wat

_Are

a(.1

4)↑;

RC

_500

(.10

)↓; D

EV

_500

(.10

)↑; D

EV

_100

(.09

)↓%

Lar

ge fi

sh (

> 2

00 m

m a

vera

ge a

dult

size

)0.

70L

ake (

S,H

,M,O

)(.5

2)↓;

DE

V_5

000(

.18)

↑%

Nes

t-gu

ardi

ng s

paw

ners

0.58

RC

_500

0(.3

8)↓;

DE

V_W

atsh

ed(.

12)↓

; RC

_100

0(.0

8)↑

% I

ntol

eran

t of

turb

idity

0.57

RC

_Wat

shed

(.15

)↓; W

et_A

rea(

.13)

↑; E

cose

ctio

n (T

HP,

WSU

,SL

C,E

OL

)(.1

1)↓;

Wet

_Are

a(.1

8)↓

% T

op c

arni

vore

s as

adu

lts0.

55R

C_W

atsh

ed(.

18)↓

; RC

_100

(.15

)↑; D

EV

_100

(.17

)↓; R

C_W

atsh

ed(.

05)↓

% L

epom

is m

acro

chir

us0.

27W

et_A

rea(

.14)

↓; L

ake (

H,O

,M)(

.07)

↑; D

EV

_500

0(.0

6)↓

% A

mei

urus

neb

ulos

us0.

34E

cose

ctio

n (SC

G,W

SU,N

GL

,SL

C,E

OL

,TH

P)(.

18)↓

; DE

V_1

000(

.16)

↑%

Cyp

rinu

s ca

rpio

and

Car

assi

us a

urat

us0.

49L

ake (

S,H

,M,O

)(.3

9)↓;

RC

_500

0(.1

0)↑

%N

otem

igon

us c

hrys

oleu

cus

0.58

Eco

sect

ion (

SSU

,SL

C)(

.42)

↑; D

EV

_Wat

shed

(.16

)↑%

Am

blop

lite

s ru

pest

ris

0.58

RC

_Wat

shed

(.15

)↓; D

EV

_500

0(.1

0)↓;

DE

V_1

000(

.33)

↑

Wetland Indicators Response to Human Disturbance 49F

ish

- fy

ke-n

ets

(n =

80

wet

land

s)N

ativ

e sp

ecie

s ri

chne

ss0.

44W

et_A

rea(

.08)

↓; R

C_5

00(.

07)↑

; DE

V_5

00(.

05)↓

; Wat

_Are

a(.0

6)↑;

Wat

_Are

a(.0

6)↓;

DE

V_5

000(

.11)

↑%

Lar

ge fi

sh (

> 2

00 m

m a

vera

ge a

dult

size

)0.

42W

at_A

rea(

.16)

↑; W

et_A

rea(

.09)

↑; W

at_A

rea(

.10)

↓; L

ake (

S,M

,E)(

.07)

↑%

Nes

t-gu

ardi

ng s

paw

ners

0.46

Wet

_Are

a(.0

9)↑;

Wet

_Typ

e (C

)(.0

9)↓;

Wet

_Are

a(.1

6)↓;

Wat

_Are

a(.0

7)↑;

Lak

e (S,

H,E

,M)(.0

5)↓

% I

ntol

eran

t of

turb

idity

0.54

Wat

_Are

a(.1

8)↓;

DE

V_5

00(.

29)↓

; RC

_Wat

shed

(.07

)↓%

Top

car

nivo

res

as a

dults

0.38

RC

_Wat

shed

(.09

)↓; D

EV

_500

(.08

)↑: D

EV

_100

0(.0

9)↓;

Wat

_Are

a(.0

7)↑;

CSI

_500

0(.0

5)↑

% L

epom

is m

acro

chir

us0.

51E

cose

ctio

n (SS

U,W

SU,N

GL

,SC

G,S

LC

)(.1

3)↓;

DE

V_5

00(.

07)↑

; RC

_Wat

shed

(.14

)↓:

DE

V_1

00(.

07)↑

; DE

V_5

000(

.10)

↓%

Am

eiur

us n

ebul

osus

0.19

Wet

_Typ

e (P,

C)(

.06)

↓; W

at_A

rea(

.13)

↓%

Cyp

rinu

s ca

rpio

and

Car

assi

us a

urat

us0.

35R

C_5

00(.

19)↑

; RC

_500

0(.1

6)↓

%N

otem

igon

us c

hrys

oleu

cus

0.15

DE

V_1

000(

.15)

↓%

Am

blop

lite

s ru

pest

ris

0.44

DE

V_W

atsh

ed(.

24)↓

; Lak

e (S,

E,H

,O)(

.07)

↓; W

et_A

rea(

.13)

↓

Bir

ds(n

= 2

23 w

etla

nds)

# Sh

ort-

dist

ance

mig

rant

s0.

17R

C_1

000(

.11)

↑; D

EV

_Wat

shed

(.06

)↓#

Lon

g-di

stan

ce m

igra

nts

0.11

Lak

e (O

)(.0

6)↓;

RC

_100

(.05

)↑#

Indi

vidu

als

0.23

RC

_100

0(.1

2)↑;

Wat

_Are

a(.0

6)↓;

DE

V_W

atsh

ed(.

05)↓

# W

etla

nd o

blig

ates

0.19

RC

_Wat

shed

(.13

)↑; D

EV

_Wat

shed

(.06

)↓#

Aer

ial f

orag

ers

0.15

RC

_Wat

shed

(.08

)↑; E

cose

ctio

n (E

OL

,SC

G,C

TP,

SLC

)(.0

7)↓

# In

sect

ivor

es0.

27R

C_1

000(

.11)

↑; W

at_A

rea(

.05)

↓; D

EV

_100

0(.0

5)↓;

DE

V_5

00(.

06)↑

# G

eoth

lypi

s tr

icha

s0.

12E

cose

ctio

n (SS

U,W

SU,N

GL

,TH

P,SC

G,N

SU,C

TP)

(.12

)↑#

Den

droi

ca p

etec

hia

0.25

DE

V_5

000(

.05)

↑; D

EV

_500

(.06

)↓; R

C_1

000(

.06)

↑; R

C_5

000(

.08)

↓#

Cis

toth

orus

pla

tens

is0.

62W

et_A

rea(

.09)

↑; E

cose

ctio

n (E

OL

,SC

G,T

HP

,SL

C,S

GL

)(.1

2)↓;

RC

_100

(.11

)↑;

RC

_500

0(.1

8)↓;

DE

V_5

00(.

12)↓

50 Brazner et al.

(Brazner et al. 2007) suggested that indicators de-rived from these two data sets responded differ-ently.

Characterizing Human Disturbance andNonstressor Covariables

Human disturbance was characterized using a va-riety of publicly available geographic informationsystem (GIS) data sources, quantified with ArcGISand ArcView software (Danz et al. 2005, Wolter etal. 2006). Most GIS data sources and primarymethodologies have already been described in de-tail (Danz et al. 2007, Johnston et al. in press), sowe only cover them briefly here. We used the pro-portion of row-crop agriculture from the NationalLand Cover Database (NLCD, 30-m raster cover-age; Wolter et al. 2006) to represent agriculturalstress (RC); the proportional sum of low and highintensity urban, commercial/industrial, and roadsurface land covers in the NLCD (Wolter et al.2006) to represent urban development stress (DEV);and a CSI based on point source and contaminantrelease information from the U.S. EPA NationalPollution Discharge Elimination System (NPDES)and Toxic Release Inventory (TRI).

We developed the CSI using a simple weightingsystem for point-source inputs based on the type ofinput. Our intent was to create a semi-quantitativeindex of severity, potentially providing strongerpredictive power than an area-weighted count ofpoint sources. For example, each stressor tabulatedin the NPDES database (sewage, metals, particu-lates, etc.) was given a weighting from 1 to 3—with sewage and pathogens given weights of 1; nu-trients, particulates and salts a weight of 2; andmore persistent or toxic stressors such as metals,solvents, PAHs, and hydrocarbons a weight of 3.Weights were summed across stressors to come upwith a score scaled from 1 to 20 indicating severity(sewage systems and petroleum refineries were theworst, life insurance companies and commercialbanks were among the least). We also appliedweights based on whether the facilities were“major” or “minor” (major = weight*2, minor =weight*1) and active vs. inactive (active = weight *1.3, inactive = 0.7). TRI data were included in theCSI as a density of land- and water-based TRI pointsources in the landscape TRI data were given aweighting of 5, which was the median score forNPDES data, before being added to NPDES scoreto provide the combined CSI.

Each of the three disturbances was characterized

at five scales; within 100 m, 500 m, 1,000 m, and5,000 m buffers, and at the whole watershed scalefor each wetland sampled. The buffers were basedon flow distance rather than Euclidean distance be-cause this provides a more accurate representationof disturbance levels affecting biota (King et al.2005). Flow distance was calculated in Arc Gridusing elevation data and the “FLOWLENGTH”command. Flow distance buffers were calculatedfrom the wetland complex perimeter, and bufferslarger than the whole watershed were clipped to thewatershed boundary.

In addition to characterizing disturbance in thelandscape, we also characterized several macro-scale covariables that were unrelated to environ-mental stress (ecosection, ecoprovince, lake,wetland type, watershed area, and wetland area)that were likely to have an important influence onindicator response based on our previous analyses(Brazner et al. 2007, Hanowski et al. 2007, Host etal. 2005, Reavie et al. 2006) or other studies(Strayer et al. 2003, King et al. 2005). We knew itwas important to account for the variance explainedby these nonstressor covariables before attemptingto assess differences in the relative importance ofthe various disturbance types or buffers on indicatorresponses (O’Connor et al. 2000, Kincaid et al.2004).

Ecosection designations from Keys et al. (1995)(Erie and Ontario Lake Plain [EOL], NorthernGreat Lakes [NGL], Northern Superior Uplands[NSU], South Central Great Lakes [SCG], South-western Great Lakes Morainal [SGL], St. Lawrenceand Champlain Valley [SLC], Southern SuperiorUplands [SSU], Central Till Plains [CTP], Tug HillPlateau [THP], Western Superior Uplands[WSU],Western Unglaciated Allegheny Platteau[WUA]) were assigned based on which ecosectionencompassed the majority of a wetland’s associatedwatershed. Watershed areas were calculated fromdigital elevation maps (http://gisdata.usgs.net/ned/)and wetland areas were delineated using NationalWetland Inventory digital data (http://wetlandsfws.er.usgs.gov/), digital raster graphic files (http://topomaps.usgs.gov/drg/) and digital orthophotoquadrangles (http://erg.usgs.gov/isb/pubs/factsheets/fs05701.html). Median watershed area was 12.1km2 (range 0.02-16,489 km2), and median wetlandarea was 0.19 km2 (range 0.004-23.4 km2). Eco-province, wetland type, and the Great Lake inwhich each wetland occurred were designated asdescribed in the sampling design (Danz et al.2005).

Wetland Indicators Response to Human Disturbance 51

Data Analysis

Although we ultimately decided to use CART tocharacterize the relationships among candidate indi-cators and predictor variables related to land usedisturbance and other potentially important ecologi-cal nonstressor covariables, we initially employedboth multiple regression and a general linear mod-elling (GLM) approach to these data as part of ourpreliminary analyses. Both of these approaches pro-vided results similar to those obtained with CARTfor most indicators, but have shortcomings thatCART does not. CART is a nonparametric approach(Breiman et al. 1984) well-suited for characterizingnonlinear relationships common to ecological re-sponse data, and effectively accounts for higher-order interactions and indirect effects that aredifficult or impossible to model with traditionalmultiple regression or other linear methods, particu-larly when many predictor variables are being ex-amined (De’Ath and Fabricius 2000).

CART is best used to quantify the variation insingle response variables explained by one or morepredictor variables. Classification tree models re-quire a categorical response variable, while regres-sion tree models require a continuous responsevariable; both types of tree models can use eithercategorical or continuous predictor variables. Weused classification trees for the four species-basedamphibian indicators derived from presence-ab-sence data and regression trees for all other indica-tors. For both classification and regression trees, thebasic idea is to find predictor variables that split thedata into groups that maximize within group homo-geneity (Breiman et al. 1984, De’Ath and Fabricius2000). Each group is characterized by an averagevalue of the response variable, the number of sam-ples in the group, and the value of the predictor thatprovides the best split.

We used SYSTAT’s TREE program (Wilkinson1998) to build all of our models. An automatic in-teraction detection (AID) algorithm (Morgan andSonquist 1963) was used to find the best fit for re-gression trees. AID takes the initial single cluster ofobservations and performs a stepwise splitting pro-cedure by searching all possible predictor variablesfor values that minimize the within-group sum ofsquares. Interactions among predictor variables arerepresented by branches from the same node (splitpoints) of the tree that is associated with differentpredictors lower in the tree. Each tree was con-structed by repeatedly splitting each group until theproportional reduction in error (PRE, Breiman et al.

1984) based on a least squares loss function fellbelow 5% for all predictors. In addition, all terminalnodes were required to include at least five wet-lands. PRE values for a particular splitting variablecan be interpreted as the increases in the proportionof variance explained, or a partial r2 equivalent forthat predictor. The overall PRE value for a particu-lar model is the equivalent of a multiple R2 (Wilkin-son 1998). Classification trees used the samestopping rules as regression trees and used the Giniindex as a loss function. This index is based onvariance estimates from comparisons of all possiblepairs of values in a subgroup, and uses the case dis-tributions at each node to determine the best splitpoints (Breiman et al. 1984, Wilkinson 1998).

CART models are based on the predictors at eachnode that provide the greatest explanatory power(i.e., highest reduction in PRE), but other correlatedpredictors may have provided similar explanatorypower. This seemed especially likely for the landuse stressors characterized at different buffer dis-tances. Due to the large number of indicators (n =66) we examined here, it was not feasible to charac-terize all possible alternative predictors at eachnode for each indicator. However, we did calculatea Pearson correlation matrix among all buffers foreach disturbance type to provide an estimate of thesimilarity among these predictors.

We present the best CART model for each bio-logical indicator and summarize indicator patternsby disturbance type, buffer scale, and assemblage.

RESULTS

Relationships amongDisturbance Types and Buffers

Within each disturbance type, correlations amongthe spatial scales were all positive; correlationswere highest for buffers most similar in size and de-clined with increasing disparity in scale (Table 2).For both RC and DEV, correlations ranged from0.50 to 0.95. For the CSI, correlations amongbuffers were the lowest overall.

Correlations among disturbance types were gen-erally low, sometimes negative, and highest be-tween the CSI and DEV. The highest correlationsamong land use disturbance types were at similarspatial scales and at larger spatial scales. All corre-lations between RC and DEV were negative andsmall (largest was –0.17, at the 5000 m scale). Cor-relations between RC and the CSI were also small(largest was +0.17 at watershed scale) but of vary-

52 Brazner et al.

TABLE 2. Mean land use disturbance levels across all sites and Pearson correlations among disturbancetypes characterized at different buffer distances (bold-faced, above and including the diagonal) and corre-lations between buffers within each disturbance type (below the diagonal). Sample size was 339–344 for alldisturbance-buffer combinations.

Correlations

Disturbance type Mean Std.Dev. Range 100 m 500 m 1,000 m 5,000 m Watershed

DevelopmentRow-crop AgricultureRC100 m 6.26 11.6 0.0–73.1 –0.08 –0.07 –0.09 –0.12 –0.07RC500 m 8.47 14.72 0.0–81.6 0.85 –0.08 –0.1 –0.12 –0.08RC1000 m 10.16 17.44 0.0–81.5 0.75 0.95 –0.11 –0.13 –0.09RC5000 m 14.05 20.59 0.0–83.9 0.62 0.79 0.87 –0.17 –0.12RCwatershed 15.27 20.75 0.0–85.5 0.55 0.68 0.75 0.90 –0.12

Contaminant Stress IndexUrban DevelopmentDEV100 m 21.95 21.91 0.0–100.0 0.02 0.13 0.2 0.23 0.12DEV500 m 20.15 21.20 0.0–100.0 0.89 0.21 0.29 0.32 0.19DEV1000 m 19.19 20.78 0.0–100.0 0.79 0.95 0.32 0.4 0.26DEV5000 m 17.51 20.03 0.0–100.0 0.65 0.81 0.90 0.46 0.31DEVwatershed 13.1 15.89 0.0–100.0 0.59 0.69 0.73 0.8 0.01

Row–cropContaminant Stress IndexCSI100 m 0.18 1.39 0.0–19.5 0.03 0.02 0.03 0.02 0.04CSI500 m 0.52 2.38 0.0–19.5 0.69 –0.01 –0.01 –0.03 0.05CSI1000 m 0.78 2.82 0.0–19.5 0.57 0.85 –0.02 –0.03 0.12CSI5000 m 3.57 13.89 0.0–149.5 0.14 0.35 0.58 –0.07 0.11CSIwatershed 54.74 291.7 0.0–2110.0 0.01 0.21 0.54 0.67 0.17

FIG. 2. Proportional reduction in error summarized across all indicators associated witheach disturbance type and all non-stressor predictors as a group (a), and associated witheach buffer irrespective of disturbance type (b) based on CART analyses (NS = all non-stressor predictors, RC = row crop, DEV = development and CSI = contaminant stressindex).

Wetland Indicators Response to Human Disturbance 53

ing direction (Table 2). Correlations between DEVand the CSI were somewhat larger (largest was0.46, at 5000 m scale) and mostly positive.

CART Results across all Variables and Buffers

Across the 66 separate CART analyses con-ducted, nonstressor variables (e.g., lake and ecosec-tion) accounted for a greater proportion of the totalvariance explained than any of the individual distur-bance types and larger-scale characterizations ofdisturbance (watershed and 5,000 m) were most ex-planatory. RC and DEV each accounted for about aquarter of the total proportional reduction in error,whereas CSI accounted for much less (Fig. 2a). The1,000 m and larger buffers explained the majorityof variance across all disturbance types, and the wa-tershed scale was the single most explanatory (Fig.2b). Among the nonstressor predictors, lake andecosection were the most explanatory variables,each accounting for more than 10% of the total ex-plained variance (Fig. 3a), each being included in atleast 20 of the 66 tree models (Fig. 3b), and eachbeing the most explanatory variable in at least nineof the trees (Fig. 3c). Watershed and wetland areawere also prominent in these models, but were lessfrequently the most explanatory variable for anyparticular indicator (Fig. 3c, Table 1). Among thestressor variables, RCwatershed was the most explana-tory predictor, followed by DEV at spatial scales of1,000 m or greater. The explanatory ability of theRC predictors clearly declined with declining buffersize, while the explanatory ability of the DEV pre-dictors was distributed more evenly across buffersizes (Figs. 3a-c). The influence of the CSI wasminimal and apparent at the 5,000 m scale only(Fig. 3).

CART Results by Assemblage

There were no statistically significant differencesin the mean number of predictors incorporated intotrees for any assemblage (ANOVA p > 0.26). Treestypically included two or three predictors, althoughtrees for fyke-net fish indicators tended to be themost complex (Table 1). There were also no signifi-cant differences among assemblages in the meanproportional reduction in error associated with thedisturbance predictors (ANOVA p = 0.79). Themean amount of variance explained by all distur-bances was typically between 15 and 25% (Fig. 4a).The total variance explained by all predictors wassignificantly different among assemblages, being

higher for electro-fish and wetland vegetation thanfor birds and amphibians (ANOVA, p = 0.01, Fig.4b).

The specific predictors with the strongest influ-ence varied considerably among assemblages (Fig.5). The majority of explained variance in amphib-ian indicators was accounted for by ecosection andlake, although many amphibian indicators did re-spond relatively strongly to development at the5,000 m buffer (e.g., species richness, presence of

FIG. 3. Proportion of explained variance (a),number of models that included a particular pre-dictor (b), and number of models in which a par-ticular predictor was the most explanatory (c) inCART analyses summarized across all indicators.

54 Brazner et al.

Rana clamitans; Table 1). Ecosection also ex-plained much of the variance in bird indicators, butresponse to RC1000 was even more important (Fig.5) and was the most explanatory predictor for shortdistance migrants, total number of individuals, andinsectivorous birds (Table 1). Watershed area andDEV1000 accounted for the largest amount of ex-plained variance in macroinvertebrate indicators(Fig. 5, Table 1). Both lake and RCwatershed

FIG. 4. Mean (± 1 s.e.) proportional reductionin error (PRE) associated with land use distur-bances across all indicators for each assemblage(a), and mean (± 1 s.e.) PRE per model summa-rized across all indicators for each assemblage (b;means letters in common were not significantlydifferent based on ANOVA with Tukey correctionsfor multiple comparisons, p < 0.05).

FIG. 5. Proportion of explained varianceaccounted for by each predictor across all indica-tors for each assemblage.

Wetland Indicators Response to Human Disturbance 55

accounted for much of the explained variance in di-atom indicators (Fig. 5). Stephanodiscoid diatomsresponded only to RCwatershed, which explained50% of the total variance in this indicator (Table1). CSI5000 was a key predictor for the proportionof motile diatoms and the Lange-Bertalot index fordiatoms (Table 1), but CSI at any scale was not animportant predictor for any other indicators. Threeelectro-fish indicators were most responsive tomultiple land use disturbances, responding to bothRC and DEV. For example, the proportion of nest-guarding spawners was explained mainly byRC5000 (38% of total variance), but DEVwatershedaccounted for another 12% of total variance (Table1). In contrast, fyke-fish responses were driven pri-marily by watershed and wetland area and were theonly assemblage besides macroinvertebrates wherewetland type played an important splitting role inany trees (e.g., % Ameiurus nebulosus [rock bass],% nest-guarding spawners; Table 1). Wetland vege-tation indicators were similar to diatoms and birdsin that the largest RC buffers were the main distur-bances to which they were responding (Fig. 5).RCwatershed and RC5000 accounted for the largestproportion of total variance for the proportion ofnative plant taxa, Carex lasiocarpa (slender sedge),and Phragmites australis cover, and explained 17%of the variance in the proportion invasive taxa(Table 1). DEV predictors were also significant forwetland vegetation, being present in five of ninevegetation trees, albeit at a lower proportion ofvariance explained than RC predictors.

Patterns in the spatial scale to which indicatorswere responding were more consistent among as-semblages, than patterns in the types of disturbanceand geographic covariables to which the indicatorswere responding. Most assemblages respondedmost strongly to disturbance characterized at thelarger scales (watershed, 5,000 m, or 1,000 m); wa-tershed or 5,000 m buffers were the most explana-tory for four assemblages (amphibians, diatoms,electro-fish, and wetland vegetation; Fig. 6). Birdand macroinvertebrate indicators responded moststrongly to disturbance in the 1,000 m buffer, andfyke-fish were unique in responding most stronglyto the 500 m buffer. The 100 m buffer was rela-tively uninfluential (Fig. 6), but did account for >10% of the variance in several variables (e.g., theproportion of Procloeon and Callibaetis mayflies,the percent cover of Carex stricta and C. lasiocarpa[Table 1]).

FIG. 6. Proportion of explained varianceaccounted for by each buffer across all distur-bance types and indicators for each assemblage.

56 Brazner et al.

Indicators with the Strongest Response toLand Use Disturbances

The indicator most responsive to land use distur-bances was % Carex stricta (tussock sedge) (R2 =0.62, Table 1, Table 3). After % C. stricta, the nextthree indicators most responsive to land use distur-bances were all based on fish captured by elec-trofishing (Table 3). Each of the electro-fishindicators responded to both RC and DEV. Re-sponses of % C. stricta and electro-fish top carni-vores were somewhat difficult to interpret due toapparently contradictory directional responses todisturbance characterized at different scales (Table1). This sort of trend reversal arose whenever a dis-turbance entered a model for the second time re-gardless of buffer, and is a byproduct of the CARTsplitting methodology (the significance of thismodel behavior is discussed in more detail later).Responses of nest-guarding spawners and Amblo-plites rupestris were relatively straightforward(Table 1). Nest-guarding spawners were less abun-dant when RC5000 was high (> 32.8%) and whenDEVwatershed was higher (> 6.5%), and Ambloplitesrupestris were less prevalent when RCwatershed> 3.5% and DEV5000 was > 9.2%. However, at thelowest end of the development gradient (DEV1000< 6.4%) Ambloplites was slightly less prevalentthan when development was between 6.4 and 9.2%.Stephanodiscoid diatoms had the strongest responseto any single land use disturbance, increasing whenRCwatershed was high (> 46.2%; Table 1). Althoughthe amount of variance explained by disturbancewas high (≥ 35%) for Aeshna (darner dragonflies),Cistothorus platensis (sedge wren), and C.

carpio/C. auratus (carp/goldfish), responses werecomplex and more difficult to interpret (Table 1).Responses of native vegetation taxa and turbidityintolerant fish captured by fyke-netting were morestraightforward. Native vegetation taxa were sensi-tive to virtually any amount of row-crop in the wa-tershed, decreasing when RCwatershed was > 0.10%,particularly in larger wetlands (> 71,000 m2) frommedium- and larger-sized watersheds (> 3.1 km2).Turbidity intolerant fyke-fish were also reduced inmedium- and larger-sized watersheds (> 2.1 km2) atlow levels of row-crop land use (RCwatershed >1.8%), and when DEV500 was high (> 8.0%) insmaller watersheds (< 2.1 km2; Table 1). Two otherdiatom indicators with straightforward and fairlystrong responses to disturbance were the Lange-Bertalot index (LBI; a species-based indicator oflimnological quality where lower scores reflectpoorer conditions; Lange-Bertalot 1979) and pro-portion motile. These were also the only two indi-cators that reflected the influence of the CSI in anyimportant way (Table 1). The LBI responded nega-tively to CSI5000 (≥ 2) in five ecosections (EOL,NSU, SGL, SSU, THP), and to higher DEV and RCat the 5,000 m scale in the other ecosections (CTP,NGL, SCG, SLC, WSU, WUA). The proportion ofmotile diatoms increased when CSI, row-crop, anddevelopment were all higher (CSI5000 > 15.5, RCwatershed > 45.3%, and DEVwatershed > 21.5%)suggesting a generalized positive response to landuse disturbance. Although the large number of indi-cators examined in this study precluded the inclu-sion of separate tree diagrams for each indicator,the full tree diagrams do aid considerably in the in-

TABLE 3. Indicator responses with a total R2 ≥ 0.40 associated with all land use disturbances combinedor total R2 ≥ 0.35 associated with a single land use disturbance type in classification and regression treeanalyses; Response to disturbance summed across buffers, non-stressor covariables included when presentin a particular tree; RC, DEV and CSI as defined in Table 1.

Indicator (Total disturbance R2) Assemblage Disturbance type (r2) Non-stressor Covariable (r2)

% Carex stricta (0.62) Wetland Vegetation DEV(0.62)% Nest-guarding fish (0.58) Fish (electro) RC(0.46), DEV(0.12)% Ambloplites rupestris (0.58) Fish (electro) DEV(0.43), RC(0.15)% Top carnivore (0.55) Fish (electro) RC(0.38), DEV(0.17)% Stephanodiscoids (0.50) Diatoms RC(0.50)% Aeshna (0.49) Macroinvertebrates DEV(0.49)# Cistothorus platensis (0.41) Birds RC(0.29), DEV(0.12) Ecosection(0.12), Wetland

Area (0.09)% Native taxa (0.36) Wetland Vegetation RC(0.36) Watershed Area (.09),

Wetland Area (.06)% C. carpio and C. auratus (fyke) (0.35) Fish (fyke) RC(0.35)

Wetland Indicators Response to Human Disturbance 57

terpretation of indicator responses for the complexcases and so we have included representative exam-ples for two of the indicators with strong potential(Fig. 7), and the remaining trees have been madeavailable in electronic format (<http://glei.nrri.umn.edu/default>).

DISCUSSION

Similar to what we found in earlier variance par-titioning analyses (Brazner et al. 2007), geographic

features such as lake and ecosection were amongthe most important predictor variables in our CARTanalyses. Although some instances where the geo-graphic variables were significant predictors mayhave been due to a geographical variable acting as asurrogate for a particular disturbance type with askewed geographic distribution (e.g., agriculturewas heavily skewed to the southern lakes), our re-sults nevertheless support the hypothesis that cer-tain nonstressor, macro-scale factors would have atleast as great an influence on indicator response asthe land use disturbances we examined. While theywere not important for all of the candidate indica-tors, accounting for variance associated with thesefactors will be critical for developing many indica-tors of ecological condition for Great Lakes wet-lands (Hanowski et al. 2007, Reavie et al. 2006).As we have suggested elsewhere (Brazner et al.2007), indicators could either be developed for geo-graphical subsets of the Great Lakes (e.g., within alake or ecoregion) or by working with residual vari-ance in assessing indicator responses after partition-ing variance associated with key geographic factors(e.g., O’Connor et al. 2000, Fore 2003a). Develop-ing indicators in this manner will be particularlyimportant if there are nested scales of influence op-erating, as suggested by Uzarski et al. (2005), thatresult in the masking of more local influences, un-less the influence of broader scale factors such asgeography have been partitioned. Our results sug-gest that large-scale geographic factors such as lakeand ecosection are driving many of the big differ-ences among biota that we observed among wet-lands and that smaller-scale factors may be drivingmany of the differences nested within these geo-graphic units.

The fact that both watershed area and wetlandarea explained a significant portion of the variancein nearly a third of the CART models indicates thatassessing the relative influence of different distur-bance types without accounting for watershed andwater body size differences would also provide mis-leading results in many instances. We expected thatour ability to detect responses to different scales ofstress might be more limited in smaller watershedsbased on the results of Strayer et al. (2003) andKing et al. (2005), where idiosyncrasies in land usepatterns in small watersheds were thought to ob-scure patterns apparent in larger watersheds. How-ever, we typically found responses of indicators tobe more apparent in smaller watersheds in modelswhere watershed size was a splitting variable (e.g.,response of native fish species richness to develop-

FIG. 7. Regression tree diagrams for proportionof motile diatoms (a) and proportion of nest-guarding fish captured by electrofishing (b).Threshold values of predictor variables, andmean, standard deviation, and sample size ofresponse variables shown at each node.

58 Brazner et al.

ment was clearer in smaller watersheds). The appar-ent difference in response sensitivity in small water-sheds between our study and Strayer et al. (2003)and King et al. (2005) may be due to differences inthe way land use stressors were calculated (e.g.,King et al. used an inverse distance weightingmethod) or a result of the way wetland indicatorsrespond in Great Lakes watersheds compared tostream indicators in Mid-Atlantic watersheds. Itsuggests that effects of land use disturbance mayactually be more easily detected in smaller GreatLakes watersheds. Since the relationship betweenspecies richness and ecosystem area has beenknown for many years (e.g., MacArthur and Wilson1967), the fact that many indicator responses weremodified by wetland area was not particularly sur-prising. The importance of ecosystem size in struc-turing wetland ecosystems has received someattention in inland wetlands (e.g., Pastor et al. 1996,Findlay and Houlahan 1997), but we believe this isthe first time this relationship has been noted forGreat Lakes coastal wetlands.

Spatial autocorrelation among different land-usedisturbance types and between different spatialcharacterizations for the same disturbance type cre-ate analytical difficulties when trying to assess rela-tive influence among these sorts of factors (VanSickle et al. 2004, King et al. 2005). Correlationsamong disturbance types in our data set were typi-cally low, but were relatively high (r > 0.80) amongadjacent buffer sizes within the row-crop and urbandevelopment disturbance types. It is likely that sub-stitution of an adjacent buffer size in models whereparticular row-crop or development buffers were se-lected as splitting variables would result in similarproportional reductions in error. The closer correla-tions between CSI and DEV compared to RC andDEV or RC and CSI, irrespective of buffer distance,suggests that in cases where CART models selectedrow-crop as a splitting variable it was likely due toa clear difference in predictive power. However, thechoices between CSI and development were lessclear cut, potentially underestimating the impor-tance of CSI and overestimating the importance ofdevelopment. The large number of indicators weexamined prevented us from exploring these inter-relationships in greater detail. However, the CARTapproach that we utilized to model indicator re-sponse did allow us to account for spatial autocorre-lation effects among selected predictors as well asmany interactions that would have been difficult tomodel using more traditional linear approaches(De’Ath and Fabricius 2000).

Our CART analyses suggested that row-crop anddevelopment land uses played more important rolesin shaping our candidate indicator responses thancontaminant stress. Although point-source contami-nants have been found to be an important stress innumerous site-specific studies in aquatic ecosys-tems, their influence at the landscape scale has notbeen well studied due to difficulties in summarizingavailable digital coverages for point sources in ameaningful way. Our CSI was an attempt to indexthe relatively coarse resolution NPDES and TRIdata in a way that would be more relevant to assess-ing their influence on the kinds of ecological re-sponses we examined. It is unclear whether thelimited response of most indicators to the CSI re-flects a failing of our index or if contaminantsources really do have a relatively small influencecompared to agriculture and development across theGreat Lakes basin. Strayer et al. (2003) also sus-pected that inadequacies of the point-source datawere at least partially responsible for their inabilityto detect contaminant influences on a range of eco-logical responses of stream biota from the Chesa-peake Bay watershed. In our study, the paucity ofdata available to calculate the CSI at buffers lessthan 5,000 m made it difficult to draw any firm con-clusions about contaminant effects at those scales.

Our results indicated clearly that row-crop agri-culture characterized at the watershed scale was thesingle most important land use predictor-buffercombination. That row-crop at any scale was impor-tant should not have been a surprise given theprevalence of row-crop agriculture throughoutmuch of the Great Lakes basin. However, the factthat it was at the watershed scale has both ecologi-cal and practical importance. It provides support forour hypothesis that agricultural influences would bestrongest at broader scales and, along with the gen-eral importance of buffers ≥ 5,000 m we observedacross all disturbance types, suggests that the typi-cal approach of characterizing land use disturbanceat the watershed scale only may be adequate formany studies. However, the clarity of response todisturbance may be considerably better at otherscales (e.g., Mensing et al.1998, Lammert andAllan 1999) for certain indicators (e.g., RC1000 wasthe most important scale for one-third of our birdmodels). From a management perspective, the ob-servation that relatively distant agricultural activi-ties have a strong influence on wetland biotabasin-wide implies that measures meant to mini-mize row-crop agricultural impacts (e.g., bufferstrips, reduced rates or altering timing of nutrient

Wetland Indicators Response to Human Disturbance 59

applications) are either insufficiently implementedor not working as effectively as desired. Our obser-vation that development was important across awider range of scales than row-crop refutes our hy-pothesis that the effects of development would beprimarily at smaller, near-wetland scales. This wasnot altogether surprising given the contradictory na-ture of other results on scale effects of urban devel-opment (Snyder et al. 2003, Wang et al. 2003b,DeLuca et al. 2004). It suggests that in placeswhere development is a primary disturbance, itwould be wise to characterize it at multiple scales,whenever feasible. Our results also suggest that wa-tersheds are more likely a good summary scale forthings that accumulate down the drainage unit (e.g.,row-crop agricultural influences), and that the ef-fects of development are more complex and will re-quire more effort to understand and predict.

Our hypotheses about relationships between or-ganism size and mobility and response to distur-bance scale were not well supported, but bird andelectro-fish indicators did tend to respond moststrongly at the larger spatial scales as predicted.Fyke-fish indicators did not follow the pattern ob-served for electro-fish but rather responded across abroader range of scales, particularly to developmentat the 500 m scale. This highlights the importanceof understanding bias associated with different sam-pling techniques when developing indicators. Assmaller and less mobile taxa, macroinvertebratesand wetland vegetation indicators were expected torespond primarily to more local characterizations ofdisturbance (Allen et al. 1999) but typically did justthe opposite, with a few notable exceptions (e.g., %Carex lasiocarpa responded primarily to row-cropat 500 m buffer). Macroinvertebrate responses todisturbance were relatively weak across all indica-tors, so even though the larger buffers were moreexplanatory, none explained a very large portion ofthe total variance in any indicator except forAeshna. As was observed by Richards et al. (1997),watershed features (watershed and wetland area)had an important influence on macroinvertebrate re-sponses, lending support to the idea that factors op-erating at broader scales were important for theseindicators whether or not they were related to dis-turbance. For diatoms, stephanodiscoid taxa re-sponded strongest to row-crop disturbance at arelatively broad scale, but for many other diatom in-dicators, the biogeographic influence of lake wasparticularly important. This differs from the obser-vation that diatoms have been more responsive tohuman disturbance than ecoregion or watershed

size in streams (Leland and Porter 2000, Fore2003b). It may be that diatom indicator responsepatterns in Great Lakes coastal wetlands are drivenby different factors than those for streams, or thatthe broader geographic scale encompassed by ourstudy favored the detection of effects associatedwith natural gradients over those related to humandisturbance. Similar reasoning may explain whywetland vegetation responded primarily to thebroader scales of disturbance, in distinct contrast tothe Minnesota wetlands studied by Mensing et al.(1998) where shrub-carr and wet meadow vegeta-tion were influenced primarily by local land use.

We identified several indicators with excellentpotential to reflect disturbance in coastal wetlandsacross the Great Lakes basin. We considered thoseindicators with a strong (high partial r2 in theCART analyses) and unidirectional response to aparticular disturbance-buffer combination with fewor no mediating nonstressor covariables to havestrong potential for interpretation and implementa-tion by resource agencies (some of the best exam-ples are plotted in Fig. 8). Since we evaluated arepresentative rather than exhaustive list of indica-tors from each taxonomic group here, there may beother indicators with better potential that we haveoverlooked. We are confident that the proportion ofboth stephanodiscoid diatoms and nest-guardingfish are excellent indicators of row-crop agricultureat the watershed scale; an abundance of stephan-odiscoids reflected high levels of RC while a pre-dominance of nest-guarding fish (electro) suggestsRC land use is low. The proportion of stephanodis-coid diatoms was one of the indicators we identifiedpreviously (Brazner et al. 2007) as respondingstrongly to a general human disturbance gradient,but our results here suggest an even stronger re-sponse to row-crop agriculture. The proportion ofnative plant taxa also appears to be an excellent in-dicator of row-crop stress at the watershed scale,even though watershed area and wetland area ex-plained additional variation. Ambloplites (electro)and Aeshna were the two best indicators of develop-ment at the 1,000 m scale and larger. CART resultssuggest a bidirectional response for Ambloplitesand Aeshna to development at different scales, butwhen these disturbance-buffer combinations wereplotted against Ambloplites and Aeshna individuallythe overall trend in indicator response was negativein all cases. The indicator % Carex stricta also hada bidirectional response to DEV predictor variables,and although this indicator had the greatest amountof variance explained by disturbance predictors

60 Brazner et al.

among all indicators (R2 = 0.62), it was not consid-ered a strong indicator because the bidirectional re-sponses were less clear: while this indicator haduniformly 0% cover at the most developed sites andhad its highest % cover at buffers with low develop-ment, it displayed a wide variety of responses in be-tween.

The reason for the sort of bidirectional responsewe observed for % Carex stricta, Ambloplites, andAeshna (and other indicators) in their full modelswas a function of the way the splitting algorithmworks in CART. In all instances when a given dis-turbance predictor entered a model for a secondtime, regardless of whether it was at the same or adifferent buffer distance, its relationship to the indi-cator was opposite in direction to the first time itwas included in the model. This bidirectional split-ting happens when a regression type relationship isnonlinear, as with a wedge-shaped response (e.g.,high disturbance led to uniformly low abundance,but low stress led to a lot of variance in the re-sponse, as in the case of % C. stricta; Mebane et al.2003, Hughes et al. 2004), or when the indicatorshave higher or lower values towards the middlerange of disturbance rather when disturbance ishigh or low. What typically happened in our modelswas that an early split in the tree clipped off themost disturbed sites where indicator values werelow. This split is interpreted as a negative relation-ship between abundance and disturbance. A subse-quent split then cut off the least disturbed placeswhere there also tended to be lower abundance (al-though with higher variance) resulting in a positiverelationship with disturbance across a small rangeof these data (e.g., when disturbance is low, abun-dance is also low). In some cases (e.g., Aeshna), afinal split segregated sites with high abundance butintermediate disturbance. CART seems particularlysensitive to teasing apart these sorts of nonlinear re-sponses to disturbance. Despite the opposing trendssome of the CART models identified for a particu-lar disturbance, those effects did not cancel eachother out. The total variance explained by the pre-dictor is still important, but the relationship is morecomplicated than a simple split at one value of thepredictor, and is based on residual variance after ac-counting for the effects at earlier nodes. A carefulexamination of the individual point scatter in two-way plots of the indicator values across a distur-bance gradient, along with an inspection of theassociated tree model, is the only way to obtain a

FIG. 8. Indicators from each assemblage withthe strongest response (partial r2) to a single dis-turbance-buffer combination (only the first distur-bance-buffer combinations to enter CART modelswere considered; best-fit lines are linear fitsexcept for stephanodiscoids [quadratic] and nest-guarding and turbidity tolerant fish [log]).

Wetland Indicators Response to Human Disturbance 61

complete understanding of the splitting decisionsmade by CART for a particular indicator.

Turbidity-intolerant fyke-fish and amphibianspecies richness were also clear indicators of devel-opment (500 and 5,000 m scales respectively) withthe turbidity-intolerant response being particularlystrong. Even though the partial r2 associated withamphibian species richness is fairly low, the re-sponse had a wedge-shaped distribution that clearlyreflected decreasing richness as development in-creased. Wedge-shaped disturbance-response rela-tionships are typical of situations where there aremultiple limiting factors (Mebane et al. 2003,Hughes et al. 2004). We were not surprised by thelack of clear indicators of contaminant stress basedon what Strayer et al. (2003) observed, but the pro-portion of motile diatoms and the LBI were bothgeneral indicators of disturbance because they re-sponded relatively strongly to all three disturbancetypes, including the CSI; motile diatoms increasingwith disturbance and the LBI decreasing.

The lack of strong indicators from the macroin-vertebrate and bird assemblages was somewhat sur-prising. In an earlier analysis (Brazner et al. 2007),there were some bird indicators that had a relativelystrong response to a general human disturbance gra-dient. However, we did not account for as many co-variables in that analysis, and the response of bothbirds and macroinvertebrate indicators in this analy-sis were strongly influenced by several non-stressorcovariables that were not included in the earlierstudy (ecosection, watershed area, wetland type andarea). Amphibian indicator responses were also rel-atively weak in this study and primarily driven bygeographic covariables rather than response tohuman disturbance. It was the wetland vegetation,diatom, and fish assemblages that were identified ashaving indicators with the most promise for reflect-ing the influence of row-crop agriculture and urbanand commercial development. It is interesting thatstephanodiscoid diatoms and Ambloplites were theonly indicators we identified among the mostpromising here that were also identified as promis-ing in our previous analyses (Brazner et al. 2007)although we worked with the same indicator set inboth. It seems likely that this is due to the specificland use disturbances we examined compared to thegeneral disturbance gradient (Danz et al. 2005) ex-amined for the previous paper. This suggests thatindicator responses are, at least in some cases, dis-turbance-specific and may be helpful for diagnosingcauses of impairment (Fore 2003a, Yoder andDeShon 2003).

Some unexplained variability may be attributableto historic effects (Galatowitsch et al. 1999a, Find-lay and Bourdages 2000), as well as other factorswe did not examine here such as local habitat influ-ences. In addition, conclusions about the relativeimportance of disturbances at different scales de-pend on the power to detect differences at a particu-lar scale (Lammert and Allan 1999), our ability toaccurately characterize certain disturbance typeswith remotely sensed data (Hollenhorst et al. 2006),sample size (Wiley et al. 1997), and a variety ofother experimental design issues (Allan and John-son 1997). Particular disturbance types may influ-ence indicators at more than one spatial or temporalscale, sometimes for different reasons (Kotliar andWiens 1990, Richards et al. 1996, Jonsen and Tay-lor 2000), so there may be multiple peaks of re-sponsiveness at different scales of the disturbanceeach driven by a different controlling mechanism.Hollenhorst et al. (2006) demonstrated that it is eas-ier to accurately measure and characterize agricul-ture than residential development, which oftenoccurs in small linear blocks that are sometimesbelow the resolution of remotely sensed data andvaries more through time than agriculture.

Despite the complexities associated with analyz-ing and interpreting indicator responses to distur-bance, the results presented here will be useful fordeveloping multi-metric, multi-assemblage indica-tors of coastal wetland condition in the GreatLakes. The CART approach is well-suited to ana-lyzing the response of single variables to one ormany predictor variables, and provides a tool forquantifying important interactions without examin-ing all possible interactions as required by generallinear models (De’Ath and Fabricius 2000). It alsoprovides a method to account for the influence ofpotentially important covariables before attemptingto assess an indicator’s response to disturbance, aconsideration that needs to be incorporated into anyattempt at formal indicator development (O’Connoret al. 2000, Fore 2003a). Resource agencies cur-rently have a keen interest in developing indicatorsof ecological condition for the Great Lakes(Keough and Griffin 1994, U.S. EPA 2002, Envi-ronment Canada and U.S. EPA 2003, Lawson2004). We believe our results are an important stepin this direction and one of the first attempts toidentify how potential ecological indicators ofGreat Lakes coastal wetland condition vary in re-sponse to different types of human disturbancecharacterized across a range of spatial scales.

62 Brazner et al.

ACKNOWLEDGMENTS

Thanks to M. Aho, M. Alexander, N. Andresen,K. Bailey, J. Baillargeon, B. Bedford, Y. Bhagat, A.Boers, M. Bourdaghs, D. Breneman, J. Carlson, A.Cotter, S. Cronk, R. Daw, R. Eedy, M. Ferguson, C.Foley, C. Frieswyk, J. Gathman, I. Grigorovich, R.Hell, J. Holland, C. Host, T. Jicha, J. Johansen, C.Johnson, M. Kang, J. Kaser, J. Kingston, A. Kireta,L. Ladwig, J. Lind, A. Ly, P. Meysembourg, J. Mor-rice, M. Pearson, G. Peterson, A. Peterson, M. Sier-szen, G. Sgro, G. Sjerven, E. Stoermer, D. Tanner,D. Taylor, J. Thompson, M. Tulbure, L. Vaccaro, D.Watters, J. Wiklund, and J. Zedler for their assis-tance with field work, data compilation, and labora-tory analysis. This research was supported by agrant from the U.S. Environmental ProtectionAgency’s Science to Achieve Results Estuarine andGreat Lakes program through funding to the GreatLakes Environmental Indicators project, U.S. EPAAgreement EPA/R-8286750. This article was alsosupported by funding from the National Aeronau-tics and Space Administration (NAG5-11262). Ad-ditional financial support provided by the NationalScience Foundation/EPSCoR Grant No. 0091948and by the state of South Dakota. This documenthas not been subjected to the either agency’s re-quired peer and policy review and therefore doesnot necessarily reflect the view of either agency,and no official endorsement should be inferred.This is contribution number 457 of the Center forWater and the Environment, Natural Resources Re-search Institute, University of Minnesota Duluth.

REFERENCESAlbert, D.A., and Minc, L.D. 2004. Plants as regional

indicators of Great Lakes coastal wetland health.Aquat. Ecosyst. Health Manage. 7:233–247.

Allan, J.D., and Johnson, L.B. 1997. Catchment-scaleanalysis of aquatic ecosystems. Freshwater Biol.37:107–111.

———, Erickson, D.L., and Fay, J. 1997. The influenceof catchment land use on stream integrity across mul-tiple spatial scales. Freshwater Biol. 37:149–161.

Allen, A.P., and O’Connor, R.J. 2000. Hierarchical cor-relates of bird assemblage structure on northeasternUSA lakes. Environ. Monit. Assess. 62:15–35.

——— , Whittier, T.R., Larsen, D.P., Kaufman, P.R.,O’Connor, R.J., Hughes, R.M., Stemberger, R.S.,Dixit, S.S., Brinkhurst, R.O., Herlihy, A.T., andPaulsen, S.G. 1999. Concordance of taxonomic com-position patterns across multiple lake assemblages:effects of scale, body size, and land use. Can. J. Fish.Aquat. Sci. 56:2029–2040.

Barton, D.R., Taylor, W.D., and Biette, R.M. 1985.Dimensions of riparian buffer strips required to main-tain trout habitat in southern Ontario streams. N. Am.J. Fish. Manage. 5:364–378.

Benson, B.J., and Magnuson, J.J. 1992. Spatial hetero-geneity of littoral fish assemblages in lakes: relationto species diversity and habitat structure. Can. J. Fish.Aquat. Sci. 49:1493–1500.

Bourdaghs, M., Johnston, C.A., and Regal, R.R. Proper-ties and performance of the floristic quality index inGreat Lakes coastal wetlands. Wetlands. in press.

Brazner, J.C., Tanner, D.K., Detenbeck, N.E., Batter-man, S.L., Stark, S.L., Jagger, L.A., and Snarski,V.M. 2005. Regional, watershed, and site-specificenvironmental influences on fish assemblage structureand function in western Lake Superior tributaries.Can. J. Fish. Aquat. Sci. 62:1254–1270.

———, Danz, N.P., Niemi, G.J., Regal, R.R., Hanowski,J.M., Johnston, C.A., Reavie, E.D., Trebitz, A.S.,Howe, R.W., Johnson, L.B., Brady, V.J., Ciborowski,J.J.H., and Sgro, G.V. 2007. Evaluating geographic,geomorphic and human influences on Great Lakeswetland indicators: multi-assemblage variance parti-tioning. Ecol. Indic. 7:610–635.

Breiman, L., Freidman, J.H., Olsen, R.A., and Stone,C.G. 1984. Classification and regression trees. Bel-mont, California: Wadsworth International Group.

Burke, V.J., and Gibbons, J.W. 1995. Terrestrial bufferzones and wetland conservation: a case study of fresh-water turtles in a Carolina Bay. Conserv. Biol.9:1365–1369.

Castelle, A.J., Johnson, A.W., and Conolly, C. 1994.Wetland and stream buffer size requirements—areview. J. Environ. Qual. 23:878–882.

Cushman, S.A., and McGarigal, K. 2002. Hierarchical,multi-scale decomposition of species-environmentrelationships. Landscape Ecol. 17:637–646.

Danz, N.P., Regal, R.R., Niemi, G.J., Brady, V.J., Hol-lenhorst, T.P., Johnson, L.B., Host, G.E., Hanowski,J.M., Johnston, C.A., Brown, T., Kingston, J., andKelly, J.R. 2005. Environmentally stratified samplingdesign for the development of Great Lakes environ-mental indicators. Environ. Monit. Assess. 102:41–65.

——— , Niemi, G.J., Regal, R.R., Hollenhorst, T.P.,Johnson, L.B., Hanowski, J.M., Johnston, C., Axler,R., Ciborowski, J., Hrabik, T., Brady, V.J., Kelly,J.R., Brazner, J.C., and Howe, R.W. 2007. Integratedmeasures of anthropogenic stress in the U.S. GreatLakes basin. Environ. Manage. 39:631–647.

De’Ath, G., and Fabricius, K.E. 2000. Classificationsand regression trees: a powerful yet simple techniquefor ecological data analysis. Ecology 81:3178–3192.

DeLuca, W.V., Studds, C.E., Rockwood, L.L., andMarra, P.P. 2004. Influence of land use on theintegrity of marsh bird communities of ChesapeakeBay, USA. Wetlands 24:837–847.

Detenbeck, N.E., Elonen, C.M., Taylor, D.L., Cotter,

Wetland Indicators Response to Human Disturbance 63

A.M., Puglisi, F.A., and Sanville, W.D. 2002. Effectsof agricultural activities and best management prac-tices on water quality of seasonal prairie pothole wet-lands. Wetlands Ecol. Manage. 10:335–354.

Eadie, J., and Keast, A. 1984. Resource heterogeneityand fish species diversity in lakes. Can. J. Zool.62:1689–1695.

Environment Canada and U.S. Environmental ProtectionAgency. 2003. State of the Great Lakes 2003. U.S.Environmental Protection Agency, Washington, D.C.EPA/905/R-03/004.

Fairbairn, S.E., and Dinsmore, J.J. 2001. Local and land-scape-level influences on wetland bird communitiesof the prairie pothole region of Iowa, USA. Wetlands21:41–47.

Findlay, C.S., and Bourdages, J. 2000. Response time ofwetland biodiversity to road construction on adjacentlands. Conserv. Biol. 14:86–94.

———, and Houlahan, J. 1997. Anthropogenic correlatesof species richness in southeastern Ontario wetlands.Conserv. Biol. 11:1000–1009.

Fisher, S.G. 1994. Pattern, process, and scale in freshwa-ter ecosystems: some unifying thoughts. In Pattern,process, and scale in aquatic ecology, A. Hildrew, P.Giller, and D. Rafaelli, eds., pp. 575–591. London,United Kingdom: Blackwell Scientific.

Fitzpatrick, F.A., Scudder, B.C., Lenz, B.N., and Sulli-van, D.J. 2001. Effects of multi-scale environmentalcharacteristics on agricultural stream biota in easternWisconsin. J. Am. Water Res. Assoc. 37:1489–1507.

Fore, L.S. 2003a. Developing biological indicators:lessons learned from Mid-Atlantic streams. U.S. Envi-ronmental Protection Agency, Ft. Meade, Maryland.EPA/539/903/R-003/003.

_____ . 2003b. Response of diatom assemblages tohuman disturbance: development and testing of a mul-timetric index for the Mid-Atlantic region (USA). InBiological response signatures: indicator patternsusing aquatic communities, T.P. Simon, ed., pp.445–471. Boca Raton, Florida: CRC Press.

Frissell, C.A., Liss, W.J., Warren, C.E., and Hurley,M.D. 1986. A hierarchical framework for stream habi-tat classification: viewing streams in a watershed con-text. Environ. Manage. 10:199–214.

Galatowitsch, S.M., Whited, D.C., and Tester, J.R.1999a. Development of community metrics to evalu-ate recovery of Minnesota wetlands. J. Aquat.Ecosyst. Stress Recov. 6:217–234.

———, Whited, D.C., Lehtinen, R.M., Husveth, J., andSchik, K. 1999b. The vegetation of wet meadows inrelation to their land use. Environ. Monit. Assess. 60:121–144.

Gorman, O.T., and Karr, J.R. 1978. Habitat structure andstream fish communities. Ecology 59:507–515.

Haig, S.M., Mehlman, D.W., and Oring, L.W. 1998.Avian movements and wetland connectivity in land-scape conservation. Conserv. Biol. 12:749–758.

Hanowski, J.M, and Niemi, G.J. 1995. Experimentaldesign considerations for establishing an off-road,habitat specific bird monitoring program using pointcounts. In Monitoring bird populations by pointcounts. Pacific Southwest Research Station, ForestService, U.S. Department of Agriculture, Albany, CA,pp. 145–150. Gen. Tech. Rep. PSW-GTR-149.

———, Danz, N.P, Howe, R.W., Niemi, G.J., and Regal,R.R. 2007. Consideration of geography and wetlandgeomorphic type in the development of Great Lakescoastal wetland bird indicators. Ecohealth4(2):194–205.

Herlihy, A.T., Stoddard, J.L., and Johnson, C.B. 1998.The relationship between stream chemistry and water-shed land cover data in the Mid-Atlantic region, U.S.Water, Air, Soil Poll. 105:377–386.

Holland, J.D., Bert, D.G., and Fahrig, L. 2004. Deter-mining the spatial scale of species’ response to habi-tat. Bioscience 54:227–233.

———, Fahrig, L., and Cappuccino, N. 2005a. Body sizeaffects the spatial scale of habitat/beetle interactions.Oikos 110:101–108.

———, Fahrig, L., and Cappuccino, N. 2005b. Fecunditydetermines the extinction threshold in a Canadianassemblage of longhorned beetles (Cleoptera:Ceram-bycidae). J. Insect Conserv. 9:109–119.

Hollenhorst, T.P., Host, G.E., and Johnson, L.B. 2006.Scaling issues in mapping riparian zones with remotesensing data: quantifying errors and sources of uncer-tainty. In Scaling and uncertainty analysis in ecology:methods and applications, J. Wu, B. Jones, H. Li, andO. Loucks, eds., pp. 275–295. Dordrecht, Nether-lands: Springer Publishing.

Holling, C.S. 1992. Cross-scale morphology, geometry,and dynamics of ecosystems. Ecol. Monogr.62:447–502.

Host, G.E., Schuldt, J., Ciborowski, J.J., Johnson, L.B.,Hollenhorst, T.P., and Richards, C. 2005. Use of GISand remotely-sensed data for a priori identification ofreference areas for Great Lakes coastal ecosystems.Int. J. Remote Sens. 26: 5325–5342.

Hughes, R.M., Kaufman, P.R., Herlihy, A.T., Kincaid,T.M., Reynolds, L., and Larsen, D.P. 1998. A processfor developing and evaluating indices of fish assem-blage integrity. Can. J. Fish. Aquat. Sci.55:1618–1631.

———, Howlin, S., and Kaufman, P.R. 2004. A bioin-tegrity index (IBI) for coldwater streams in westernOregon and Washington. Trans. Am. Fish. Soc.133:1497–1515.

Johnson, L.B., Richards, C., Host, G.E., and Arthur, J.W.1997. Landscape influences on water chemistry inMidwestern stream ecosystems. Freshwater Biol.37:193–208.

Johnson, R.K., Goedkoop, W., and Sandin, L. 2004. Spa-tial scale and ecological relationships between the

64 Brazner et al.

macroinvertebrate communities of stony habitats ofstreams and lakes. Freshwater Biol. 49:1179–1194.

Johnston, C.A., Detenbeck, N.E., and Niemi, G.J. 1990.The cumulative effect of wetlands on stream waterquality and quantity: A landscape approach. Biogeo-chemistry 10:105–141.

———, Johnston, C.A., Brown, T.N., Hollenhorst, T.P.,Wolter, P.T., Danz, N.P., and Niemi, G.J. GIS in sup-port of ecological indicator development. In Manualof geographic information systems. Bethesda, MD:American Society for Photogrammetry and RemoteSensing. in press.

Jones, B., Walker, J., Riitters, K.H., Wickham, J.D., andNicoll, C. 1996. Indicators of landscape integrity. InIndicators of catchment health, J. Walker and D.Reuter, eds., pp. 155–168. Collingwood, Victoria,Australia: Commonwealth Scientific and IndustrialResearch Organization.

Jonsen, I., and Taylor, P.D. 2000. Calopteryx damselflydispersions arising from multiscale responses to land-scape structure. Conserv. Ecol. 4:57–76.

Keough, J.R., and Griffin, J. 1994. Technical workshopon EMAP indicators for Great Lakes coastal wet-lands. Summary report. U.S. Environmental Protec-tion Agency and National Biological Service, Duluth,Minnesota.

Keys, J.E. Jr., Carpenter, C.A., Hooks, S.L., Koeneg,F.G., McNab, W.H., Russell, W.E., and Smith, M.L.1995. Ecological units of the eastern United States—first approximation. Map (scale 1:3,500,000). Atlanta,GA. U.S. Department of Agriculture, Forest Service.Technical Publication R8-TP 21.

Kincaid, T.M., Larsen, D.P., and Urquhart, N.S. 2004.The structure of variation and its influence on the esti-mation of status: indicators of the condition of lakesin the Northeast, USA. Environ. Monit. Assess.98:1–21.

King, R.S., Richardson, C.J. , Urban, D.L., andRomanowicz, E.A. 2004. Spatial dependency of vege-tation–environment linkages in an anthropogenicallyinfluenced wetland ecosystem. Ecosystems 7:75–97.

———, Baker, M.E., Whigham, D.F., Weller, D.E., Jor-dan, T.E., Kazyak, P.F., and Hurd, M.K. 2005. Spatialconsiderations for linking watershed land cover toecological indicators in streams. Ecol.Appl.15:137–153.

———, Hines, A.H., Craige, F.D. and Grap, S. Regional,watershed and local correlates of blue crab andbivalve abundances in subestuaries of ChesapeakeBay, USA. J. Exp. Mar. Biol. Ecol. in press.

Knutson, M.G., Sauer, J.R., Olsen, D.A., Mossman,M.J., Hemesath, L.M., and Lannoo, M.J. 1999.Effects of landscape composition and wetland frag-mentation on frog and toad abundance and speciesrichness in Iowa and Wisconsin, USA. Conserv. Biol.13(6) 1437–1446.

Kotiliar, N.B., and Wiens, J.A. 1990. Multiple scales of

patchiness and patch structure: a hierarchical frame-work for the study of heterogeneity. Oikos59:253–260.

Lammert, M., and Allan, J.D. 1999. Assessing the bioticintegrity of streams: effects of scale in measuring theinfluence of land use/cover and habitat structure onfish and macroinvertebrates. Environ. Manage.23:257–270.

Lange Bertalot, H. 1979. Pollution tolerance of diatomsas a criterion for water quality estimation. Nova Hed-wigia 64:285–304.

Lawson, R. 2004. Coordinating coastal wetlands moni-toring in the North American Great Lakes. Aquat.Ecosyst. Health Manage. 7:215–221.

Lee, K.E., Goldstein, R.M., and Janson, P.E. 2001. Rela-tion between fish communities and riparian zone con-ditions at two spatial scales. J. Am. Water Res. Assoc.37:1465–1473.

Lehtinen, R.M., Galatowitsch, S.M., and Tester, J.R.1999. Consequences of habitat loss and fragmentationfor wetland amphibian assemblages. Wetlands19:1–12.

Leland, H.V., and Porter, S.D. 2000. Distribution of ben-thic algae in the upper Illinois River basin in relationto geology and land use. Freshwater Biol.44:279–301.

Lyons, J., Weigel, B.M., Paine, L.K., and Undersander,D.J. 2000. Influence of intensive rotational grazing onbank erosion, fish habitat quality, and fish communi-ties in southwestern Wisconsin trout streams. J. Soiland Water Con. 55:271–276.

MacArthur, R., and Wilson, E.O. 1967. The theory ofisland biogeography. Princeton, New Jersey, USA:Princeton University Press.

Maxwell, J.R., Edwards, C.J., Jensen, M.E., Paustian,S.J., Parrott, H., and Hill, D.M. 1995. A hierarchicalframework of aquatic ecological units in North Amer-ica (Nearctic Zone). U.S. Department of Agriculture,Forest Service, North Central Forest Experiment Sta-tion, St. Paul, MN. General Technical Report NC-176.

Mebane, C.A., Maret, T.R., and Hughes, R.M. 2003. Anindex of biological integrity (IBI) for Pacific North-west rivers. Trans. Am. Fish. Soc. 132:239–261.

Mensing, D.M., Galatowitsch, S.M., and Tester, J.R.1998. Anthropogenic effects on the biodiversity ofriparian wetlands of a northern temperate landscape.J. Environ. Manage. 53:349–377.

Minshall, G.W. 1988. Stream ecosystem theory: a globalperspective. J. N. Am. Benthol. Soc. 7:263–288.

Morgan, J.N., and Sonquist, J.A. 1963. Problems in theanalysis of survey data, and a proposal. J. Am. Stat.Assoc. 58:415–434.

Niemi, G.J., and McDonald, M.E. 2004. Application ofecological indicators. Ann. Rev. Ecol. Syst.35:89–111.

———, Wardrop, D., Brooks, R., Anderson, S., Brady,V., Paerl, H., Rakocinski, C., Brouwer, M., Levinson,

Wetland Indicators Response to Human Disturbance 65

B., and McDonald, M. 2004. Rationale for a new gen-eration of ecological indicators for coastal waters.Environ. Health Persp. 112:979–986.

O’Connor, R.J., Jones, M.T., White, D., Hunsaker, C.,Loveland, T., Jones, B., and Preston, E. 1996. Spatialpartitioning of environmental correlates of avian bio-diversity in the conterminous United States. Biodiv.Lett. 3:97–110.

———, Walls, T.E., and Hughes, R.M. 2000. Using mul-tiple taxonomic groups to index the ecological condi-tion of lakes. Environ. Monit. Assess. 61:207–228.

O’Neill, R.V., Hunsaker, C.T., Jones, K.B., Riitters,K.H., Wickham, J.D., Schwartz, P.M., Goodham,I.A., Jackson, B.L., and Baillargeon, W.S. 1997. Mon-itoring environmental quality at the landscape scale.Bioscience 47:513–519.

Pastor, J., Downing, A., and Erickson, H.E. 1996.Species-area curves and diversity-productivity rela-tionships in beaver meadow: Voyageurs NationalPark, Minnesota, USA. Oikos 77:399–406.

Poff, N.L. 1997. Landscape filters and species traits:towards mechanistic understanding and prediction instream ecology. J. N. Am. Benthol. Soc. 16:391–409.

———, and Allan, J.D. 1995. Functional organization ofstream fish assemblages in relation to hydrologicalvariability. Ecology 76:606–627.

Pope, S.E., Fahrig, L., and Merriam, H. 2000. Landscapecomplementation and metapopulation effects on leop-ard frog populations. Ecology 81:2498–2508.

Reavie, E.D., Axler, R.P., Sgro, G.V., Danz, N.P.,Kingston, J.C., Kireta, A.R., Brown, T.N., Hollen-horst, T.P., and Ferguson, M.J. 2006. Diatom-basedweighted-averaging transfer functions for Great Lakescoastal water quality: relationships to watershed char-acteristics. J. Great Lakes Res. 32:321–347.

Reeves, G.H., Everest, F.H., and Sedell, J.R. 1993.Diversity of anadromous salmonid assemblages incoastal Oregon basins with different levels of timberharvest. Trans. Am. Fish. Soc. 122:309–317.

Ribic, C.A., Lewis, S.J., Melvin, S., Bart, J., and Peter-john, B. 1999. Proceedings of the marsh bird moni-toring workshop. U.S. Fish and Wildlife Service, U.S.Geological Survey, Washington, D.C., USA.

Richards, C., and Johnson, L.B. 1998. Landscape per-spectives on ecological risk assessment. In Riskassessment: logic and measurement, eds. M.C. New-man and C.L. Strojan, pp. 255–274. Chelsea, MI: AnnArbor Press.

———, Johnson, L.B., and Host, G.E. 1996. Landscape-scale influences on stream habitats and biota. Can. J.Fish. Aquat. Sci. (Suppl. 1):295–311.

———, Haro, R.J., Johnson, L.B., and Host, G.E. 1997.Catchment and reach scale properties as indicators ofmacroinvertebrate species traits. Freshwater Biol.37:219–230.

Riffel, S.K., Keas, B.E., and Burton, T.M. 2003. Birds inNorth American Great Lakes coastal wet meadows: is

landscape context important? Landscape Ecol.18:95–111.

Roland, J., and Taylor, P.D. 1997. Insect parasitoidspecies respond to forest structure at different spatialscales. Nature 386:710–713.

Roth, N.E., Allan, J.D., and Erickson, D.L. 1996. Land-scape influences on stream biotic integrity assessed atmultiple spatial scales. Landscape Ecol. 11:141–156.

Schleiger, S.L. 2000. Use of an index of biotic integrityto detect effects of land uses on stream fish communi-ties in west-central Georgia. Trans. Am. Fish Soc.129:1118–1133.

Schlosser, I.J. 1991. Stream fish ecology: A landscapeperspective. Bioscience 41:704–712.

Semlitsch, R.D., and Bodie, J.R. 2003. Biological crite-ria for buffer zones around wetlands and riparianhabitats for amphibians and reptiles. Conserv. Biol.17:1219–1228.

Sharma, R., and Hilborn, R. 2000. Empirical relation-ships between watershed characteristics and cohosalmon (Oncorhychus kisutch) smolt abundance in 14western Washington streams. Can. J. Fish. Aquat. Sci.58:1453–1463.

Snyder, C.D., Young, J.A., Villella, R., and Lemarié,D.P. 2003. Influences of upland and riparian land usepatterns on stream biotic integrity. Landscape Ecol.18:647–664.

Southwood, T.R.E. 1977. Habitat, the templet for eco-logical strategies? J. Anim. Ecol. 46:337–365.

Sponseller, R.A., Benfield, E.F., and Valett, H.M. 2001.Relationships between land-use, spatial scale andstream macroinvertebrate communities. FreshwaterBiol. 46:1409–1424.

Steedman, R.J. 1988. Modification and assessment of anindex of biotic integrity to quantify stream quality insouthern Ontario. Can. J. Fish. Aquat. Sci.45:492–501.

Stevenson, R.J. 1997. Scale-dependent determinants andconsequences of benthic algal heterogeneity. J. N.Am. Benthol. Soc.16:248–262.

Stoffels, R.J., Clarke, K.R., and Closs, G.P. 2005. Spatialscale and benthic community organisation in the lit-toral zones of large oligotrophic lakes: potential forcross-scale interactions. Freshwater Biol.50:1131–1145.

Strayer, D.L., Beighley, R.E., Thompson, L.C., Brooks,S., Nilsson, C., Pinay, G., and Naiman, R.J. 2003.Effects of land cover on stream ecosystems: roles ofempirical models and scaling issues. Ecosystems6:407–423.

Townsend, C.R., and Hildrew, A.G. 1994. Species traitsin relation to a habitat templet for river systems.Freshwater Biol. 31:265–275.

——— , Doledec, S., Norris, R., Peacock, K., andArbuckle, C. 2003. The influence of scale and geogra-phy on relationships between stream community com-

66 Brazner et al.

position and landscape variables: description and pre-diction. Freshwater Biol. 48:768–785.

Trebitz, A.S., Brazner, J.C., Brady, V.J., Axler, R., andTanner, D.K. 2007. Turbidity tolerances of GreatLakes coastal wetland fishes. N. Am. J. Fish. Manage.26:619–633.

U.S. EPA. 2002. Methods for evaluating wetland condi-tion: introduction to wetland biological assessment.U.S. Environmental Protection Agency, Office ofWater, Washington, DC. EPA/822/R-02/014.

Uzarski, D.G., Burton, T.M., and Genet, J.A. 2004. Vali-dation and performance of an invertebrate index ofbiotic integrity for Lakes Huron and Michigan fring-ing wetlands during a period of lake level decline.Aquat. Ecosyst. Health Manage. 7: 269–288.

———, Burton, T.M., Cooper, M.J., Ingram, J.W., andTimmermans, S.T.A. 2005. Fish habitat use withinand across wetland classes in coastal wetlands of thefive great lakes: development of a fish-based index ofbiotic integrity. J. Great Lakes Res. 31(1):171–187.

Van Sickle, J., Baker, J., Herlihy, A., Bayley, P., Gre-gory, S., Haggerty, P., Ashkenas, L., and Li, J. 2004.Projecting the biological condition of streams underalternative scenarios of human land use. Ecol. Appl.14:368–380.

Walser, C.A., and Bart, H.L. Jr. 1999. Influence of agri-culture on in-stream habitat and fish community struc-ture in Piedmont watersheds of the ChattahoocheeRiver system. Ecol. Freshwater Fish 8:237–246.

Wang, L., and Lyons, J. 2003. Fish and macroinverte-brate assemblages as indicators of stream degradationin urbanizing watersheds. In Biological response sig-natures: indicator patterns using aquatic communi-ties, T.P. Simon, ed., pp. 227–249. Boca Raton,Florida: CRC Press.

———, Lyons, J., Kanehl, P., and Gatti, R. 1997. Influ-ences of watershed land use on habitat quality andbiotic integrity in Wisconsin streams. Fisheries22:6–12.

———, Lyons, J., Kanehl, P., and Bannerman, R. 2001.Impacts of urbanization on stream habitat and fishacross multiple spatial scales. Environ. Manage.28:255–266.

———, Lyons, J., Rasmussen, P., Seelbach, P., Simon,T., Wiley, M., Kanehl, P., Baker, E., Niemala, S., andStewart, P.M. 2003a. Watershed, reach, and riparian

influences on stream fish assemblages in the NorthernLakes and Forests Ecoregion, USA. Can. J. Fish.Aquat. Sci. 60:491–505.

——— , Lyons, J., and Kanehl, P. 2003b. Impacts ofurban land cover on trout streams in Wisconsin andMinnesota. Trans. Am. Fish. Soc. 132:825–839.

Weeber, R.C., and Vallianatos, M., eds. 2000. TheMarsh Monitoring Program, 1995–1999: monitoringthe Great Lakes, wetlands and their amphibian andbird inhabitants. Bird Studies Canada, EnvironmentCanada, and U.S. EPA.

Whited, D., Galatowitsch, S. Tester, J.R., Schik, K.,Lehtinen, R., and Husveth, J. 2000. The importance oflocal and regional factors in predicting effective con-servation: planning strategies for wetland bird com-munities in agricultural and urban landscapes. Land-scape Urb. Plan. 49:49–65.

Wiens, J.A., and Milne, B.T. 1989. Scaling of “land-scapes” in landscape ecology, or, landscape ecologyfrom a beetles perspective. Landscape Ecol. 3: 87–96.

Wilcox, D.A., Meeker, J.E., Hudson, P.L., Armitage,B.J., Black, M.G., and Uzarski, D.G. 2002. Hydro-logic variability and the application of index of bioticintegrity metrics to wetlands: a Great Lakes evalua-tion. Wetlands 22:588–615.

Wiley, M.J., Kohler, S.L., and Seelbach, P.W. 1997.Reconciling landscape and local views of aquaticcommunities: lessons from Michigan trout streams.Freshwater Biol. 37:133–148.

Wilkinson, L. 1998. Classification and regression trees.SYSTAT 10.0 ® Statistics, SPSS, Inc., Chicago, IL.pp. 31–51.

Wolter, P.T., Johnston, C.A., and Niemi, G.J. 2006.Land use land cover change in the U.S. Great Lakesbasin 1992 to 2001. J. Great Lakes Res. 32:607–628.

Yoder, C.O., and DeShon, J.E. 2003. Using biologicalresponse signatures within a framework of multipleindicators to assess and diagnose causes and sourcesof impairments to aquatic assemblages in selectedOhio rivers and streams. In Biological response sig-natures: indicator patterns using aquatic communi-ties, T.P. Simon, ed., pp. 23–81. Boca Raton, Florida:CRC Press.

Submitted: 11 July 2006Accepted: 27 June 2007Editorial handling: John R. Kelly