Content uploaded by Robert K. Murphy
Author content
All content in this area was uploaded by Robert K. Murphy on May 03, 2020
Content may be subject to copyright.
377
Am. Midl. Nat. 144:377–392
Models for Guiding Management of Prairie Bird Habitat in
Northwestern North Dakota
ELIZABETH M. MADDEN
1
Biology Department, Montana State University, Bozeman 59717
ROBERT K. MURPHY
U.S. Fish and Wildlife Service, Des Lacs National Wildlife Refuge Complex, Kenmare, North Dakota 58746
ANDREW J. HANSEN
Biology Department, Montana State University, Bozeman 59717
LEIGH MURRAY
University Statistics Center, New Mexico State University, Las Cruces 88003
A
BSTRACT
.—With grassland bird populations in the Great Plains exhibiting steep declines,
grassland managers require information on bird habitat needs to optimally manage lands
dedicated to wildlife. During 1993–1994, we measured bird occurrence and corresponding
vegetation attributes on mixed-grass prairie in northwestern North Dakota. Three hundred
and ten point-count locations over a wide range of successional stages were sampled. Ten
grassland passerine species occurred commonly (
i.e.,
at .10% of point count locations),
including two species endemic to the northern Great Plains [Baird’s sparrow (
Ammodramus
bairdii
) and Sprague’s pipit (
Anthus spragueii
)], and several species of management concern
[bobolink (
Dolichonyx oryzivorus
), grasshopper sparrow (
Ammodramus savannarum
), clay-
colored sparrow (
Spizella pallida
)]. Some species were ubiquitous and had generalized hab-
itat associations [
e.g.,
savannah sparrow (
Passerculus sandwichensis
)]. Others exhibited more
finely tuned, closely overlapping use of relatively short, sparse to moderately dense, grass-
and forb-dominated habitat. We used logistic regression models to predict bird species’ oc-
currence based on nine vegetation variables. Previously undefined limits of vegetation height
and density were identified for Baird’s sparrow and Sprague’s pipit, and of shrub cover for
Baird’s sparrow. Our findings underscore the need for a mosaic of successional types to
maximize diversity of prairie bird species. Managers may reduce confusion created by generic
treatment prescriptions for grasslands by focusing on absolute rather than relative measures
of vegetation, and by integrating standard data from multiple bird habitat studies across
regions.
I
NTRODUCTION
With grassland bird populations in the Great Plains exhibiting steep declines (Droege
and Sauer, 1994; Knopf, 1994, 1996) grassland managers require information on bird hab-
itat needs to optimally manage lands dedicated to wildlife. Responses to activities such as
cropping, haying, grazing and prescribed burning have been documented for many grass-
land bird species including passerines (Owens and Myres, 1973; Kantrud, 1981; Pylypec,
1991; Johnson, 1996; Dale
et al.,
1997, Madden
et al.,
1999), but these responses can vary
greatly both spatially and temporally. Studies of effects of management activities on Great
Plains avifauna often cannot be extrapolated across the region due to differences in envi-
ronmental conditions (
e.g.
, moisture, soil types, and plant species composition). Thus, it
remains difficult to summarize effects of management on individual bird species. For ex-
1
Corresponding author: Medicine Lake National Wildlife Refuge, Medicine Lake, Montana 59247.
Telephone (406)789-2305; e-mail: elizabethpmadden@fws.gov
378
144(2)
T
HE
A
MERICAN
M
IDLAND
N
ATURALIST
ample, in North Dakota, Sprague’s pipit was most abundant in moderately- and heavily-
grazed areas (Kantrud, 1981), but in Alberta and Saskatchewan it was most abundant in
ungrazed or lightly-grazed prairie (Owens and Myres, 1973; Dale, 1983). Study results that
appear conflicting may not be when vegetation type and structure are considered.
An alternative to summaries of management impacts to birds is the use of habitat models
based on vegetation attributes associated with the species of interest (
see
Verner
et al.,
1986).
Vegetation attributes are important determinants of grassland bird abundance and distri-
bution (Cody, 1968; Wiens, 1969; Rotenberry and Wiens, 1980; Whitmore, 1981; Zimmer-
man, 1988). Vegetation structure and, to a lesser extent, composition are readily manipu-
lated by land managers. Presented with a profile of the vegetation used by a particular bird
species or group of species, a manager can use a variety of defoliation or other management
tools to obtain it, either through the incorporation of natural disturbance regimes, or
through activities that may mimic them (
e.g.
, mowing).
Despite the fact that studies of bird habitat associations based solely on occurrence or
abundance data ignore important facets of population ecology (
e.g.,
source/sink dynamics)
and therefore may not reveal highest quality habitat for a species (Van Horne, 1983), they
often are the only cost-effective option, especially when developing models for managing
multiple species (Hansen
et al.,
1993). We used such a habitat-based approach for grassland
passerines in northern, mixed-grass prairie in the Missouri Coteau physiographic subregion
of North Dakota. Our objective was to quantify relationships between passerine birds and
vegetation structure and composition. Our ultimate goal was to provide managers of north-
ern, mixed-grass prairie with models that predict occurrence of bird species based on hab-
itat conditions.
S
TUDY
A
REA
Lostwood National Wildlife Refuge (NWR) covers 109 km
2
of rolling to hilly, mixed-grass
prairie in Mountrail and Burke counties, northwestern North Dakota (488379N; 1028279W).
It lies within the Missouri Coteau, a dead-ice moraine characterized by knob-and-kettle
topography (685–747 m elevation). The large tract of grassland is interspersed with more
than 4100 wetland basins and 500 clumps of quaking aspen (
Populus tremuloides
). Major
vegetation is a needlegrass-wheatgrass (
Stipa
spp.-
Agropyron
spp.) association, and habitat
composition is 55% native prairie, 21% previously-cropped grassland revegetated with tame
and native prairie plants, 20% wetland, 2% trees, and 2% tall shrubs (Murphy, 1993). Cli-
mate is semiarid and mean annual precipitation (1936–1989) is 42 cm, with most (.75%)
falling as rain between April and September.
Before European settlement in the early 1900s, the landscape of Lostwood NWR was
treeless mixed-grass prairie, maintained in a shorter grass, or even barren state by frequent
fire and bison impacts (Murphy, 1993). The region likely supported a 5- to 10-y fire-return
interval (Wright and Bailey, 1982:81; Murphy, 1993; Bragg, 1995). Although some of pre-
sent-day Lostwood NWR was tilled and farmed during the early 1900s, most (70%) upland
areas remained unplowed and were either left idle or lightly grazed season-long by livestock
during the 1930s–1970s. With settlement came suppression of wildfires and a concomitant
loss of early successional, herbaceous vegetation. Western snowberry (
Symphoricarpos occi-
dentalis
), quaking aspen and exotic grasses [Kentucky bluegrass (
Poa pratensis
), smooth
brome (
Bromus inermis
), and quack grass (
Agropyron repens
)] proliferated and now domi-
nate much of the mixed-grass community of Lostwood NWR.
Since the late 1970s, the U.S. Fish and Wildlife Service (USFWS) has mostly used pre-
scribed fire to reduce woody vegetation, control exotic plants, enhance vigor of native plant
species, and restore a more natural diversity of successional stages to the landscape. Avail-
2000
379M
ADDEN ET AL
.: P
RAIRIE
B
IRD
H
ABITAT
ability of this wide range of vegetation successional stages made Lostwood NWR an appro-
priate place to study relationships between prairie birds and vegetation in various stages of
succession.
M
ETHODS
We measured bird use and vegetation characteristics on 160 (1993) and 150 (1994) sam-
ple points distributed over nine prescribed burn units that encompassed a wide range of
postfire successional stages, ranging from ,1yto.80 y. This provided a broad gradient
of vegetation, from short and sparse to tall and dense. We selected sampling points from a
grid of potential points that encompassed the study area. The sampling scheme was system-
atic in 1993 (i.e., each square mile of the study area was gridded into 268-m 3268-m blocks
and each grid intersection was a potential sample point), but was changed to randomized
systematic in 1994 to better meet sampling assumptions for statistical testing. We generated
the random, systematic grid for each square mile section by randomly picking 2 numbers
between 1–250 m to serve as the X and Y coordinates of a point in the southwest corner
of the section. From this random start point, we gridded points every 250 m north and
south across each section. Points were $250 m apart to provide statistical independence in
terms of birds and vegetation (Ralph
et al.,
1993), and were randomly selected from the
grid of potential points that met the following criteria: (1) located in ‘‘upland prairie’’ as
delineated by the National Wetland Inventory map of cover types of Lostwood NWR, (2)
$200 m from any aspen clump, (3) $50 m from roads or firebreaks and (4) currently
ungrazed by livestock. We selected 160 sample points in 1993, and a new set of 150 points
in 1994. In 1994, we reduced the buffer from aspen to 100 m based on 1993 sampling
observations. We also added a 50-m buffer to seasonally-flooded wetland zones because of
high water levels.
We estimated occurrence of individual bird species using 50-m radius (1993) and 75-m
radius (1994) point count surveys on fixed-radius plots (Hutto
et al.,
1986). [Based on an
analysis of bird species detectability (Rotella
et al.,
1999), the radius was increased to 75-m
in 1994 to increase the number of birds sampled per point]. The observer stood at a survey
point center for 10 min and recorded all birds seen or heard. We conducted counts from
0.5 h before sunrise until about 0900 h CDT, during the breeding season from late May
until early July, whenever weather conditions did not impede detection of birds (
i.e.
,no
rain, fog or wind .15 km/h). Each point was surveyed three times. Observers and the
order in which points were visited were rotated to minimize sampling bias.
Vegetation structure and general composition were measured at each bird sur vey point
during late June through early August in 1993 and late June through late July in 1994. Ten
subsample points were located at 5-m inter vals along each of two transects within each fixed-
radius bird plot, for a total of twenty subsamples. Both transects were positioned on the
same random compass bearing (
i.e.
, parallel), each a different random distance (between
5–30 m) from plot center.
At each subsample point, we estimated visual obstruction (dm) at a height of 1 m and a
distance of 4 m (Robel
et al.,
1970). Litter depth was measured directly (cm) by lowering
a 6-mm diameter rod vertically into the litter layer. Dead vegetation from previous years
that was standing but no longer vertical was considered litter where it formed a mat-like
layer, roughly continuous to the ground. Using the same rod, we counted the total number
of ‘‘hits’’ or contacts of vegetation on the rod in each dm height interval (Wiens, 1969).
Each hit was recorded as either live (current year’s growth) or dead (previous years’
growth). Total number of hits represented vertical density. We also calculated the percent-
age of total hits represented by live vegetation. Percentage areal cover of shrubs, forbs and
380
144(2)
T
HE
A
MERICAN
M
IDLAND
N
ATURALIST
grasses was visually estimated at each subsample (Daubenmire, 1959) within a 1-m diameter
circular quadrat. Finally, each circular quadrat was assigned to one of nine general plant
groups of management interest: native grass, Kentucky bluegrass, native grass/Kentucky
bluegrass, broad-leaved exotic grass (
i.e.,
smooth brome or quack grass), shrub (primarily
western snowberry), shrub/broad-leaved exotic grass, shrub/Kentucky bluegrass, shrub/
native grass and wetland.
For each vegetation structure variable, we calculated the mean and coefficient of variation
(CV) for the 20 subsamples near each point. CV indicates horizontal ‘‘patchiness’’ or het-
erogeneity of a particular vegetation attribute (Roth, 1976). Plant groups were expressed
as the mean frequency of a given group in the 20 subsamples. In all, 14 vegetation structure
variables (
i.e.,
mean and CV of: litter depth, visual obstruction, % live vegetation, vertical
vegetation density, shrub cover, grass cover and forb cover) and nine plant groups were
considered. Our collection of bird occurrence and vegetation data overlapped only slightly
in time (mostly June, vs. mostly July). We acknowledge this temporal disparity may have
slightly affected our assessment of visual obstruction and vertical density influences, but not
other vegetation structure and composition variables.
Because each point count location was independent in terms of birds and vegetation, we
used point-level analyses with a sample size of n 5160 and n 5150, in 1993 and 1994,
respectively, to explore relationships between bird species and vegetation characteristics.
Bird species’ occurrence at a point was defined as the presence of at least one singing
male of a given species during at least one of three survey visits. (Singing male observations
were used for all species except cowbirds, because female grassland birds are secretive and
are not reliably detected with these methods.) Only bird species detected at .10% of points
were used in statistical analyses. These included the same nine species each year, and a
tenth only in 1994.
Because many variables of vegetation structure were correlated, we used principal com-
ponents analysis (SAS Institute, 1989) of the 14 variables to identify major gradients in
vegetation structure. Using the resulting gradients, we plotted the subset of points at which
individual bird species were present on a plot of Principal Component 1 (PC 1) versus
Principal Component 2 (PC 2) to describe each species’ relative location on the vegetation
gradients in each year. We then plotted a 95% confidence ellipse based on the mean for
each species, and superimposed the ten species onto one graph for purposes of comparison.
Logistic regression (SAS Institute, 1989) was used to develop predictive models of species’
occurrence based on vegetation characteristics. We chose logistic regression over linear
regression because bird abundance data were heavily weighted with zeros and violated as-
sumptions of linear regression (
i.e.,
normality and constant variance). Densities of birds in
mixed-grass prairie are typically low such that point count abundances are essentially re-
cords of species’ occurrence.
The logistic model was:
P
(presence) 51/(1 1exp{ 2[
b
o
1
b
1
(
x
)]})
where
P
(presence) was the probability that a bird species was present,
b
o
and
b
1
were in-
tercept and slope coefficients and
x
was the predictor variable (
i.e.
, vegetation variable).
Two-thirds of the 1994 data (n 599) was used to generate the models, and the remaining
one-third (n 551) was used as a ‘‘hold-out’’ data set to retest the model’s predictive power
for correct classification. Only 1994 data were used here because the smaller (50-m) radius
point counts used to sample birds in 1993 yielded few observations of birds at sample points
(
i.e.,
birds were often using an area, but the small size of the plot precluded them from
being recorded).
2000
381M
ADDEN ET AL
.: P
RAIRIE
B
IRD
H
ABITAT
T
ABLE
1.—Eigenvector loadings of principal components (PC 1 and PC 2) of vegetation structure
variables in 1993 and 1994
Variable
1993
PC 1 PC 2
1994
PC 1 PC 2
Shrub cover
Forb cover
Grass cover
CV shrub cover
CV forb cover
CV grass cover
Visual obstruction
CV visual obstruction
Litter depth
CV litter depth
Vertical density (total # hits)
CV vertical density
% live vegetation
CV % live vegetation
0.33
20.17
20.27
20.20
0.20
0.29
0.33
0.08
0.33
20.32
0.34
20.08
20.33
0.24
0.29
20.39
0.27
20.32
0.42
0.12
0.24
20.12
20.23
0.28
20.02
20.11
0.33
20.29
0.33
20.15
20.28
20.19
0.22
0.29
0.31
0.02
0.32
20.30
0.37
20.24
20.31
0.20
0.34
20.28
20.20
20.30
0.24
0.26
0.23
0.16
20.29
0.34
20.16
0.25
0.36
20.22
Eigenvalue
% variance explained
Total variance explained
5.51
39.32
39.32
2.46
17.6
56.93
5.33
38.10
38.10
2.82
20.11
58.22
To choose the best multivariable model for each bird species, a backward-elimination
routine was used on a set of nine variables (grass cover, forb cover, shrub cover, shrub
frequency, native grass frequency, exotic grass frequency, litter depth, visual obstruction and
vertical vegetation density). These nine variables were chosen based on their perceived
importance to birds, their relevance to managers and to minimize collinearity (r
s
,0.85)
within the set of predictor variables. A variable was eliminated from the model if its observed
significance level for the regression coefficient (based on the Wald chi-square) was P .
0.05. In one case (bobolink), a 4-variable model was selected by the backward-elimination
method, but scrutiny of possible 3-variable models indicated a more parsimonious fit, and
a 3-variable model was ultimately chosen.
Although multivariate models can provide excellent predictive power, their individual
regression coefficients can be difficult to understand and interpret, especially in 3- and 4-
variable models where interactions and collinearity cloud relationships among variables. To
facilitate interpretation of relationships between bird occurrence and individual vegetation
variables, we additionally scrutinized and graphed single-variable models (of management
interest). These models were then used to generate incidence functions that predict the
probability of a bird species’ occurrence given a certain value of a selected vegetation var-
iable. We limited incidence function graphs to the range of vegetation values observed in
this study, to ensure that interpretation would not extend beyond the observed range of
data.
R
ESULTS
Results of principal components analyses for both years of vegetation data were remark-
ably similar in terms of eigenvectors and gradients (Table 1). The first axis (PC1) accounted
for 39% and 38% of the variation in 1993 and 1994, respectively, and represented a gradient
from short, sparse, grass-dominated vegetation to tall, dense, shrub-dominated vegetation.
382
144(2)
T
HE
A
MERICAN
M
IDLAND
N
ATURALIST
→
F
IG
. 1.—Composite 95% confidence ellipses based on the mean for plots at which individual bird
species occurred in 1993 and 1994, with reference to Principal Component 1 and 2. Number of plots
on which the species occurred indicated in parentheses
The second axis (PC2) accounted for 18% and 20% of the variation in 1993 and 1994,
respectively, and represented a gradient from deep litter and forb-dominated vegetation to
mostly live vegetation with few forbs and low litter.
On composite graphs for bird species’ occurrence across the vegetation gradients (Fig.
1), we interpreted nonoverlapping confidence ellipses to represent use of different vege-
tation by individual species, and ellipses that overlapped the plot origin (0,0) to indicate
use of average vegetation conditions available on the study area. For example, savannah
sparrow and clay-colored sparrow overlapped each other almost completely in 1994, and
partially in 1993, and both were located near or over the origin, indicating similar use of
vegetation near the average available. Confidence ellipses of brown-headed cowbirds (
Mol-
othrus ater
) were among the largest, and overlapped with all species except common yel-
lowthroat (
Geothlypis trichas
) [and western meadowlark (
Sturnella neglecta
) in 1994]. Con-
fidence ellipses for common yellowthroat were distinctly separate from those of other spe-
cies. This species alone was located wholly in the quadrant with high shrub cover.
The remaining six grassland species were clustered together in the short, sparse to mod-
erate, grass- and forb-dominated quadrant of vegetation space (Fig. 1). Baird’s sparrow,
grasshopper sparrow, and bobolink showed greatest overlap. Sprague’s pipit, western mead-
owlark and Le Conte’s sparrow (
Ammodramus leconteii
) overlapped these species, but had
larger confidence ellipses. Of this group, Sprague’s pipit used shortest and sparsest vege-
tation, and Le Conte’s sparrow (observed in 1994 only) used the tallest and densest. On
PC 2 most of these six species were near or below zero, indicating use of areas with moderate
amounts of litter and forbs.
Predictive models for 8 species included 1 to 4 significant predictor variables (Table 2).
A 4-variable model best classified Baird’s sparrow occurrence (Table 2). Grass cover was the
best single-predictor model for Baird’s sparrow (Wald chi-square significance ,0.001),
classifying presence/absence correctly 68% and 75% of the time in the original and hold-
out data sets, respectively. Incidence (
i.e.,
probability of occurrence) of Baird’s sparrow
increased with grass cover, and reached 50% at about 42% grass cover (Fig. 2). It also
increased with forb cover (Wald chi-square significance 50.040), reaching 50% incidence
at 35% forb cover (Fig. 2). Shrub cover and Robel visual obstruction were also good single
predictors (Wald chi-square significance ,0.001 for each). Incidence of Baird’s sparrow
decreased as these increased, and dropped below 50% at about 18% shrub cover and 1.5-
dm visual obstruction (Figs. 2, 3). Frequency of native grasses was also a significant single
predictor (Wald chi-square 50.014; 66% and 67% correct classification), predicting 50%
Baird’s sparrow occurrence at native grass frequency of 0.42 (Fig. 2).
Sprague’s pipit occurrence was best predicted by visual obstruction (Table 2). Incidence
of pipits declined quickly as visual obstruction increased, even more rapidly than Baird’s
sparrow and grasshopper sparrow incidence for the same variable (Fig. 3). Sprague’s pipit
incidence decreased to 50% at 0.8 dm and to less than 5% at 1.9 dm.
Bobolink occurrence was best predicted by increasing forb cover and grass cover, and
decreasing frequency of native grasses (Table 2). Frequency of broad-leaved, exotic grass
was also a good single predictor of bobolink occurrence (Wald chi-square significance 5
0.021) (Fig. 4).
2000
383M
ADDEN ET AL
.: P
RAIRIE
B
IRD
H
ABITAT
384
144(2)
T
HE
A
MERICAN
M
IDLAND
N
ATURALIST
T
ABLE
2.—Logistic regression models that best predicted occurrence of grassland birds in 1994. Variables were selected from a set of nine vegetation
variables using a backward-elimination routine
Species Fitted model for logit response
a
o.s.l.–
model
b
% correct
classification
Data
to fit
c
n599
Hold-out
data
d
n551
Baird’s sparrow 28.17 21.31(visual obstruction)
e
10.12(forb cover) 10.14
(grass cover) 10.09(shrub cover)
0.000 75.8 68.6
Brown-headed cowbird 0.04 10.06(forb) 10.04(grass) 20.15(vertical density) 0.000 71.7 68.6
Bobolink 24.03 10.05(forb) 10.06(grass) 20.22(native grass) 0.003 77.8 72.5
Clay-colored sparrow 0.42 10.13(shrub) 0.002 93.9 88.2
Common yellowthroat 2.89 20.08(forb) 20.07(grass) 0.000 75.8 92.2
Grasshopper sparrow 2.37 21.55(visual obstruction) 10.35(exotic grass) 0.000 68.7 62.7
Le Conte’s sparrow No significant model found
Savannah sparrow No significant model found
Sprague’s pipit 2.24 22.77(visual obstruction) 0.000 87.9 82.4
Western meadowlark 26.90 10.11(forb) 10.07(grass) 0.000 80.8 82.4
a
Logit 5ln[P(Presence)/P(Absence)] 5
b
o
1
b
1
(
x
1
)... 1
b
p
(
x
p
), P(Presence) 5l/[l1exp{2(
b
o
1
b
1
x
1
...1
b
p
x
p
)}] and P(Absence) 5l-P(Presence)
b
observed significance level (o.s.l.) for the overall model
c
correct classification rates for data used to fit the model
d
correct classification rates for ‘‘hold-out’’ data set used to re-test classification
e
Visual obstruction 5Robel visual obstruction, forb 5forb cover, grass 5grass cover, shrub 5shrub cover, vertical density 5total number of vegetation
contacts, native grass 5frequency of native grasses, exotic grass 5frequency of broad-leaved, exotic grasses
2000
385M
ADDEN ET AL
.: P
RAIRIE
B
IRD
H
ABITAT
F
IG
. 2.—Incidence of Baird’s sparrows as predicted by logistic regression models for four vegetation
variables. Dashed lines indicate 95% confidence intervals for the predicted probabilities. Incidence
lines do not extend beyond observed ranges of data
Visual obstruction combined with frequency of broad-leaved, exotic grasses best predicted
grasshopper sparrow occurrence (Table 2). Grasshopper sparrow incidence increased with
decreasing visual obstruction (Fig. 3), and with increasing frequency of broad-leaved, exotic
grasses.
Clay-colored sparrow occurrence was best predicted by shrub cover (Table 2). However,
little shrub cover was necessary for high probability of clay-colored sparrow occurrence.
Probability of occurrence was 69% at only 3% shrub cover, and reached 95% at 20% shrub
cover (Fig. 5).
Forb cover and grass cover were best predictors of common yellowthroat and western
meadowlark occurrence, although relationships in the models were opposite. Yellowthroat
incidence increased with decreasing forb and grass cover (Table 2). Meadowlark incidence
increased with increasing forb cover and grass cover, and did not reach 50% probability
until 44% forb cover (Fig. 5).
Brown-headed cowbird incidence increased with forb cover and grass cover, and de-
creased with vertical vegetation density (Table 2). No significant predictive models were
found for savannah or Le Conte’s sparrows.
D
ISCUSSION
Bird species examined were well distributed over gradients of vegetation structure and
composition. Savannah sparrow and clay-colored sparrow, the most widespread and abun-
dant passerines at Lostwood NWR (Madden
et al.
, 1999), used mixed-grass prairie vegetation
near the average available. Clay-colored sparrow was distinctly associated with shrubs. Baird’s
386
144(2)
T
HE
A
MERICAN
M
IDLAND
N
ATURALIST
F
IG
. 3.—Incidence of three grassland bird species as predicted by logistic regression models for Robel
visual obstruction (
i.e.,
height and density) of vegetation. Dashed lines indicate 95% confidence inter-
vals for the predicted probabilities. Incidence lines do not extend beyond observed ranges of data
F
IG
. 4.—Incidence of bobolinks as predicted by logistic regression models for frequency of native
and exotic grasses. Dashed lines indicate 95% confidence intervals for the predicted probabilities.
Incidence lines do not extend beyond observed ranges of data
sparrow, grasshopper sparrow, Sprague’s pipit, and western meadowlark used moderate
amounts of chiefly grass and forb cover, with Sprague’s pipit using the shortest cover. Bob-
olink and Le Conte’s sparrow used taller, denser, grassy vegetation (mostly represented at
Lostwood NWR by exotic grasses, such as smooth brome and quack grass). Common yel-
lowthroat used the densest, shrubbiest habitats. Of species examined here, it is the only
2000
387M
ADDEN ET AL
.: P
RAIRIE
B
IRD
H
ABITAT
F
IG
. 5.—Incidence of clay-colored sparrows and western meadowlarks as predicted by logistic regres-
sion model for shrub cover and forb cover, respectively. Dashed lines indicate 95% confidence intervals.
Incidence lines do not extend beyond observed ranges of data
one not considered a grassland bird (Mengel, 1970; Johnsgard, 1978), but instead is a shrub-
or wetland-associated species. Vegetation use by brown-headed cowbirds overlapped with
nearly all other bird species, a good characteristic for a brood parasite.
The vegetation gradients described in our study (principal components analyses) resem-
ble Ryan’s (1990) prairie continuum model, which portrayed grassland habitats ranging
from short/sparse, grass- and forb-dominated to tall/dense, shrub- and tree-dominated veg-
etation over gradients of soil moisture and fire and grazing frequency and intensity. Ryan
classified general (qualitative) habitat affinities of Sprague’s pipit as short-sparse grass;
Baird’s sparrow, grasshopper sparrow, savannah sparrow, and western meadowlark as mid-
grass; clay-colored sparrow as mid-grass/shrub; Le Conte’s sparrow and bobolink as tall-
dense grass and common yellowthroat as tall-dense grass/shrub. Our results agree with these
descriptions.
General habitat affinities are well known for grassland birds, but few published studies
provide quantitative summaries of species’ preferences for vegetation structure. Especially
lacking are data for endemic grassland species (Mengel, 1970; Johnsgard, 1978) such as
Baird’s sparrows and Sprague’s pipits, which are of increasing concern due to recent steep
population declines, habitat loss and a public policy shift toward ecosystem and native spe-
cies management (Knopf, 1994, 1996; USFWS, 1996). Most public grasslands devoted to
wildlife traditionally have been managed to promote tall, dense nesting cover for game
birds, especially waterfowl. Our results reveal that Baird’s sparrows and Sprague’s pipits use
relatively short, sparse grass structure, and are most associated with native bunch grasses,
rather than with the broad-leaved, exotic grasses often planted for nesting cover. To maxi-
mize native species diversity, managers charged with maintaining grassland communities
should consider such needs.
At Lostwood NWR, presence of Baird’s sparrow was best predicted by relatively high
percentages of grass cover and forb cover, low shrub cover and low visual obstruction. Areas
occupied by Baird’s sparrow in other studies (Dale, 1983; Renken, 1983; Winter, 1994) had
fairly similar values for these attributes. Occupied areas in central North Dakota had a mean
visual obstruction of 1.3 dm and litter depth of 3.6 cm (Renken, 1983), similar to values of
1.6 dm and 3.7 cm at Lostwood NWR (Madden, 1996). Winter (1994) also studied Baird’s
sparrow at Lostwood NWR, and similarly summarized preferred habitat as having 0.3–3 cm
litter depth, visual obstruction of about 1.2 dm, and 1.5–3.5% shrub cover. Dale’s (1983)
values for south-central Saskatchewan were not as directly comparable, but height and den-
sity of vegetation and presence of some forbs or shrubs were important, and litter depths
388
144(2)
T
HE
A
MERICAN
M
IDLAND
N
ATURALIST
of occupied areas were about 0.8–1.2 cm during breeding season. Mean shrub cover of
areas occupied by Baird’s sparrow ranged from less than 1% (Renken, 1983) to 3–8% (Dale,
1983) to 20% in this study (Madden, 1996).
Our logistic regression models for visual obstruction and shrub cover define previously
unidentified upper limits of suitability for Baird’s sparrow. A Habitat Suitability Index (HSI)
based on limited data suggested 1.5–4 dm visual obstruction as optimum for Baird’s sparrow
(Sousa and McDonal, 1983). Lacking evidence that visual obstruction over 4 dm might be
detrimental for Baird’s sparrows, the model assumed no upper limit. Our model captured
distinct limitations in Baird’s sparrow occurrence at higher visual obstruction levels. Prob-
ability of Baird’s sparrow occurrence was highest at about 1 dm, fell below 25% between
2.0–2.5 dm, and below 1% at about 4.5 dm of visual obstruction. Baird’s sparrows were
completely absent from areas with greater visual obstruction (5–7 dm) (Madden, 1996).
Lower limits of visual obstruction cannot be assessed from our model because the lowest
mean readings on sample plots were 0.7 dm. Lower thresholds have been identified else-
where from studies in drier, grazed areas where heavy grazing reduced vegetation to un-
suitable height and density levels for Baird’s sparrow (Owens and Myre, 1973; Dale, 1983).
The HSI also suggested that Baird’s sparrow is inhibited by over 25% shrub cover, but lacked
supportive data. Probability of occurrence of Baird’s sparrow in our models dropped below
50% at about 18% shrub cover and below 10% at 54% shrub cover. Although sparse shrub
cover appears acceptable or even preferred by Baird’s sparrow (
see
Dale, 1983), the species
has been notably absent on idle (
i.e.,
no fire or grazing) prairie invaded by shrubs (Arnold
and Higgins, 1986; Renken, 1983; Madden
et al.
, 1999).
Occurrence of Sprague’s pipit at Lostwood NWR was best predicted by low visual obstruc-
tion by vegetation. In addition, this species was consistently associated with native grasses
and low amounts of shrub cover (Madden, 1996). In Saskatchewan, sites used by Sprague’s
pipits also had lower shrub cover, but had higher vegetation density than unused sites (Dale,
1983). Absolute values of vegetation densities used by Sprague’s pipit in the two studies,
however, were similar (
i.e.,
vegetation that is ‘‘less dense’’ at Lostwood NWR is ‘‘more
dense’’ in Saskatchewan). Throughout their range, Sprague’s pipits are less abundant (or
absent) in areas of introduced grasses compared with native prairie (Kantrud, 1981; Wilson
and Belcher, 1989; Johnson and Schwartz, 1993; Dale
et al.,
1997).
Vegetation attributes preferred by these endemic species are characteristic of grasslands
receiving periodic defoliations such as those produced by fire or grazing. Prescribed fire,
for example, promotes prairie dominated by herbaceous cover (Wright and Bailey, 1982).
Both Baird’s sparrow and Sprague’s pipit were most abundant at Lostwood NWR on prairie
burned 2–4 times in the previous 15 y (Madden
et al.
, 1999). Depending on location within
the species’ ranges, higher or lower intensities of habitat disturbance will achieve the req-
uisite vegetation structure. For example, periodic fire or moderate grazing provide pre-
ferred Baird’s sparrow and pipit habitat in North Dakota (Kantrud, 1981; Renken, 1983;
Messmer, 1990; Madden
et al.
, 1999), whereas long-idle or lightly-grazed native prairie sup-
ports preferred vegetation structure in drier portions of their ranges (Dale, 1983; Wershler
et al.,
1991; Sutter
et al.,
1995).
Focus on absolute, rather than relative, measures of vegetation may reduce conflicts as-
sociated with attempts to standardize habitat treatment prescriptions on the Great Plains,
as well as problems inherent in defining, for example, ‘‘heavy’’ or ‘‘light’’ grazing. Integra-
tion of many small-scale, bird habitat studies into broader guidelines incorporating the full
range of grassland habitats would likely resolve apparent contradictions in bird-habitat as-
sociations, but lack of standardized methodology makes this challenging.
Several other bird species examined here are of management interest. The clay-colored
2000
389M
ADDEN ET AL
.: P
RAIRIE
B
IRD
H
ABITAT
sparrow, which breeds mainly in the northern Great Plains, is considered an endemic grass-
land species by some ( Johnsgard, 1978), and its population is declining (Knopf, 1996; Sauer
et al.,
1997). The species’ well-documented association with shrubby prairie (Owens and
Myres, 1973; Kantrud, 1981; Dale, 1983; Renken, 1983; Knapton, 1994) was corroborated
by our results. Proliferation of western snowberry at Lostwood NWR has benefited clay-
colored sparrow, as it was the most abundant bird we observed. Although associated with
shrubs, clay-colored sparrows at Lostwood NWR occupied areas with as little as 2–3% shrub
cover (Madden, 1996), indicating that prairie with small amounts of shrub may suffice.
Although bobolinks are much more widespread than these endemic species, their pop-
ulations are also declining (Sauer
et al.
, 1997). Bobolink habitat use in our study was similar
to that found in many other studies, namely hayfields and exotic, tall grasses (Kantrud,
1981; Renken, 1983; Bollinger and Gavin, 1992). These nonnative habitats seem to effec-
tively simulate the now rare tallgrass prairie types that bobolinks preferred historically. In
1873, bobolinks were noted breeding in large numbers in north-central North Dakota in
‘‘meadowy’’ areas on ‘‘open prairie adjoining the (Souris) river.’’ (Coues, 1878:587). Ap-
parent preferences of bobolinks and other grassland birds
(e.g.
, Le Conte’s and savannah
sparrow) for exotic grasses may more accurately reflect affinities for mesic prairie sites.
Whereas the native plant composition of xeric prairie at Lostwood NWR is relatively intact,
most tallgrass, mesic sites have been invaded by exotic grasses. The tall, rhizomatous (usually
broad-leaved) exotic grasses (
i.e.
, smooth brome and quack grass) are structurally similar
to the native grass species they have replaced, and associated bird species have adopted this
introduced vegetation as breeding habitat.
Whereas bobolinks showed a distinct preference for exotic grasses, grasshopper sparrow
use of grass habitats is less easily summarized. Whitmore (1981) indicated that grasshopper
sparrows normally inhabit open grasslands of bunch grasses rather than rhizomatous grasses
in West Virginia and Wisconsin. This is apparently contradicted by mixed-grass-prairie stud-
ies in which grasshopper sparrows were more abundant in exotic rhizomatous grass than
native bunch grass habitats (Wilson and Belcher, 1989), and were one of the most common
species in Conservation Reser ve Program (CRP) lands planted mostly with exotic grasses
( Johnson and Schwartz, 1993). Our study concurred with the mixed-prairie studies; grass-
hopper sparrows used areas with high frequencies of exotic grasses. We conclude that both
habitats are likely acceptable to grasshopper sparrows, and suitability may depend on mois-
ture conditions. Their preference for moderate amounts of vegetation may involve either
use of sparser bunch grasses in dense, relatively tall grass habitats, or use of denser, rhizo-
matous exotic grasses in shorter-grass prairies.
Savannah sparrow was so ubiquitous (93% frequency) at Lostwood NWR that modeling
any habitat preferences there would have been difficult. Other work at Lostwood NWR
revealed little else about savannah sparrow habitat use, except for a positive relationship
between savannah sparrow abundance and broad-leaved, exotic grasses (Madden, 1996). Le
Conte’s sparrow appeared to use a limited array of vegetation types, but occurred so infre-
quently (12% frequency) that predictive modeling was again made difficult. Other analyses,
however, revealed patterns in habitat use. Points occupied by Le Conte’s sparrows had lower
shrub cover than unoccupied points and a greater frequency of broad-leaved, exotic grasses
(Madden, 1996).
This study represents one of the initial efforts in modeling habitat use of grassland pas-
serines in the northern Great Plains. It is limited, however, by lack of information on avian
reproductive success and habitat area needs, issues clearly central to the long-term viability
of these bird populations. Regardless, our results underscore the need for a mosaic of
available vegetation types to maximize avian biodiversity (Ryan, 1990). Where traditional
390
144(2)
T
HE
A
MERICAN
M
IDLAND
N
ATURALIST
wildlife management emphasized late successional stages because these areas usually had
the greatest number of species, managers now recognize the importance of beta diversity,
i.e.
, diversity across the gradient of prairie habitats, as well as the importance of emphasizing
habitat for endemic species (Samson and Knopf, 1982; Knopf, 1994; Johnson, 1996).
Predictive models developed here quantify structure and composition of vegetation used
by individual bird species. Managers of northern mixed-grass prairie can maximize proba-
bility of occurrence for species of interest by using these models to select appropriate ranges
of vegetation parameters, and then manipulating vegetation accordingly. When manage-
ment treatments are periodic (
e.g.
, about every 5 y in mixed-grass prairie in this study),
vegetation mosaics created are likely to be both spatial and temporal. An array of prairie
birds should be favored in a given area over decades of management, and quantitative
models for multiple species could be integrated to forecast such communities. For example,
six of nine grassland bird species we studied had relatively specific but closely overlapping
habitat needs that also encompassed the broad range of conditions supporting two more
generalized species. Visual obstruction and/or presence of grass, forb or shrub cover were
important vegetation features for all of the specialized species. An integrated summary of
species’ habitat use indicates that probability of occurrence of these species on Lostwood
NWR may be uniformly greatest (.50%) at roughly 0.5–1.5 dm visual obstruction and 40–
60% grass cover (or its approximate inverse in this study, 10–20% shrub cover). Using these
conditions as a central foundation around which to promote a mosaic of grassland types,
managers of similar northern mixed-grass prairies can encourage or inhibit more specific
vegetation components on their respective landscapes by adjusting treatment type, intensity,
and timing. On Lostwood NWR, for example, management to maintain moderately-dense
grass cover as a dominant (albeit dynamic) landscape component provides, by default, for
abundant forbs, native grasses and broad-leaved, exotic grasses (or warm season, native grass
species they replaced) with which several species of grassland birds are associated.
Acknowledgments.
—This study was funded by the Non-Game Migratory Bird Program and Refuges
and Wildlife Division of the U.S. Fish and Wildlife Ser vice, Region 6. We thank B. Johnson, L. Rawinski
and N. Fahler for field assistance. K. A. Smith, manager of Lostwood NWR, along with staff of Des Lacs
NWR Complex, provided advice and logistical support. J. R. Rotella and P. Munholland shared ideas
and suggestions for data analyses. Comments by I. J. Ball, S. T. Jones, M. L. Morrison, R. B. Renken,
R. Schroeder and two anonymous reviewers improved the manuscript.
L
ITERATURE
C
ITED
A
RNOLD
,T.W.
AND
K. F. H
IGGINS
. 1986. Effects of shrub coverages on birds of North Dakota mixed-
grass prairies.
Can. Field Nat.,
100:10–14.
B
OLLINGER
,E.K.
AND
T. A. G
AVIN
. 1992. Eastern bobolink populations: ecology and conservation in
an agricultural landscape, p. 497–506.
In:
J. M. Hagan and D. W. Johnson (eds.). Ecology and
conservation of neotropical migrant landbirds. Smithsonian Institution Press, Washington, D.
C.
B
RAGG
, T. B. 1995. The physical environment of Great Plains grasslands, p. 49–81.
In:
K. Keeler and
A. Joerns (eds.). The changing prairie. Oxford University Press, New York, New York.
C
ODY
, M. L. 1968. On the methods of resource division in grassland bird communities.
Am. Nat.,
102:
107–147.
C
OUES
, E. 1878. Field notes on birds observed in Dakota and Montana along the forty-ninth parallel
during the seasons of 1873 and 1874.
U.S. Geol. and Geogr. Surv. Terr. Bull.,
4:545–661.
D
ALE
, B. C. 1983. Habitat relationships of passerine birds at Last Mountain Lake, Saskatchewan. M.S.
Thesis, Univ. of Regina, Saskatchewan, Canada, 119 p.
,P.A.M
ARTIN AND
P. S. T
AYLOR
. 1997. Effects of hay management on grassland songbirds in
Saskatchewan.
Wildl. Soc. Bull.,
25:616–626.
2000
391M
ADDEN ET AL
.: P
RAIRIE
B
IRD
H
ABITAT
D
AUBENMIRE
, R. 1959. A canopy-coverage method of vegetational analysis.
Northwest Sci.,
33:43–64.
D
ROEGE
,S.
AND
J. S
AUER
. 1994. Are more North American species decreasing than increasing?, p. 297–
306.
In:
E. J. M. Hagemeijer and T. J. Verstrael (eds.). Bird Numbers 1992: distribution, mon-
itoring and ecological aspects. Proceedings of Twelfth International Conference of IBCC and
EOAC, Noordwijkerhout, The Netherlands. Beek-Ubbergen, The Netherlands.
H
ANSEN
, A. J., S. L. G
ARMAN
,B.M
ARKS AND
D. L. U
RBAN
. 1993. An approach for managing vertebrate
diversity across multiple-use landscapes.
Ecol. Appl.,
3:481–496.
H
UTTO
, R. L., S. M. P
LETSCHET AND
P. H
ENDRICKS
. 1986. A fixed-radius point count method for non-
breeding and breeding season use.
Auk,
103:593–602.
J
OHNSGARD
, P. A. 1978. The ornithogeography of the Great Plains states.
Prairie Nat.,
10:97–112.
J
OHNSON
, D. H. 1996. Effects of fire on bird populations in mixed-grass prairie, p. 181–206.
In:
F. L.
Knopf and F. B. Samson (eds.). Ecology and conservation of Great Plains vertebrates. Springer,
New York, New York.
AND
M. D. S
CHWARTZ
. 1993. The Conservation Reser ve Program and grassland birds.
Conserv.
Biol.,
7:934–937.
K
ANTRUD
, H. A. 1981. Grazing intensity effects on the breeding avifauna of North Dakota native grass-
lands.
Can. Field Nat.,
95:404–417.
K
NAPTON
, R. W. 1994. Clay-colored sparrow, p. 1–16.
In:
A. Poole and F. Gill (eds.). The Birds of North
America, No. 120. Philadelphia: The Academy of Natural Sciences. American Ornithologists’
Union, Washington, D. C.
K
NOPF
, F. L. 1994. Avian assemblages on altered grasslands.
Stud. Avian Biol.,
15:247–257.
. 1996. Prairie legacies—birds, p. 135–148.
In
: F. Samson and F. Knopf (eds.). Prairie conser-
vation: preserving North America’s most endangered ecosystem. Island Press, Washington, D.C.
M
ADDEN
, E. M. 1996. Passerine communities and bird-habitat relationships on prescribe-burned mixed-
grass prairie in North Dakota. M.S. Thesis, Montana State University, Bozeman, Montana, 153 p.
,A.J.H
ANSEN AND
R. K. M
URPHY
. 1999. Influence of prescribed fire history on habitat and
abundance of passerine birds in northern mixed-grass prairie.
Can. Field Nat.,
113:627–640.
M
ENGEL
, R. M. 1970. The North American Central Plains as an isolating agent in bird speciation, p.
280–340.
In:
W. Dort and J. K. Jones (eds.). Pleistocene and recent environments of the central
Great Plains. University of Kansas Press, Lawrence, Kansas.
M
ESSMER
, T. A. 1990. Influences of grazing treatments on nongame birds and vegetation structure in
south central North Dakota. Ph.D. Dissertation, North Dakota State University, Fargo, North
Dakota, 147 p.
M
URPHY
, R. K. 1993. History, nesting biology, and predation ecology of raptors in the Missouri Coteau
of northwestern North Dakota. Ph.D. Dissertation, Montana State University, Bozeman, Mon-
tana, 212 p.
O
WENS
,R.A.
AND
M. T. M
YRES
. 1973. Effects of agriculture upon populations of native passerine birds
of an Alberta fescue grassland.
Can. J. Zool.,
51:697–713.
P
YLYPEC
, B. 1991. Impacts of fire on bird populations in a fescue prairie.
Can. Field Nat.,
105:346–349.
R
ALPH
, C. J., G. R. G
EUPEL
,P.P
YLE
,T.E.M
ARTIN AND
D. F. D
E
S
ANTE
. 1993. Handbook of field methods
for monitoring landbirds. U.S. Department of Agriculture, Forest Service General Technical
Report PSW-GTR-144, 41 p.
R
ENKEN
, R. B. 1983. Breeding bird communities and bird-habitat associations on North Dakota Water-
fowl Production Areas. M.S. Thesis, Iowa State University, Ames, Iowa, 90 p.
R
OBEL
, R. J., J. N. B
RIGGS
,A.D.D
AYTON AND
L. C. H
ULBERT
. 1970. Relationships between visual ob-
struction measurements and weight of grassland vegetation.
J. Range Manage.,
23:295–297.
R
OTELLA
, J. J., E. M. M
ADDEN AND
A. J. H
ANSEN
. 1999. Sampling considerations for estimating abun-
dance of passerines in grasslands.
Stud. Avian Biol.
19:237–243.
R
OTENBERRY
,J.T.
AND
J. A. W
IENS
. 1980. Temporal variation in habitat structure and shrubsteppe bird
dynamics.
Oecologia,
47:1–9.
R
OTH
, R. R. 1976. Spatial heterogeneity and bird species diversity.
Ecology,
57:773–782.
R
YAN
, M. R. 1990. A dynamic approach to the conservation of the prairie ecosystem in the midwest, p.
392
144(2)
T
HE
A
MERICAN
M
IDLAND
N
ATURALIST
91–106.
In:
J. M. Sweeney (ed.). Management of dynamic ecosystems. North Central Section,
The Wildlife Society, West Lafayette, Indiana.
SAS I
NSTITUTE
,I
NC
. 1989. SAS/STAT User’s Guide, Version 6, Fourth Edition, Volume 2. SAS Institute,
Cary, North Carolina.
S
AMSON
,F.B.
AND
F. L. K
NOPF
. 1982. In search of a diversity ethic for wildlife management.
Trans. N.
Amer. Wildl. and Natural Resour. Conf.,
47:421–431.
S
AUER
, J. R., J. E. H
INES
,G.G
OUGH
,I.T
HOMAS AND
B. G. P
ETERJOHN
. 1997. The North American
Breeding Bird Survey results and analysis. Version 96.4. Patuxent Wildlife Research Center,
Laurel, Maryland.
S
OUSA
, P. J.,
AND
W. N. M
C
D
ONAL
. 1983. Habitat suitability index models: Baird’s sparrow. U.S. De-
partment of Interior, Fish and Wildlife Service. FWS/OBS-82/10.44. 12 p.
S
UTTER
, G. C., T. T
ROUPE
,
AND
M. F
ORBES
. 1995. Abundance of Baird’s sparrows,
Ammodramus bairdii,
in native prairie and introduced vegetation.
Ecoscience,
2:344–348.
U.S. F
ISH AND
W
ILDLIFE
S
ERVICE
. 1996. Migratory birds of management concern in the United States:
the 1995 list. U.S. Fish and Wildlife Service, Office of Migratory Bird Management. Washing-
ton, D. C. 15 p.
V
AN
H
ORNE
, B. 1983. Density as a misleading indicator of habitat quality.
J. Wildl. Manage.,
47:893–
901.
V
ERNER
, J., M. L. M
ORRISON AND
C. J. R
ALPH
(eds.). 1986. Wildlife 2000: modeling habitat relationships
of terrestrial vertebrates. University of Wisconsin, Madison, Wisconsin, 470 p.
W
ERSHLER
,C.W.,W.W.S
MITH AND
C. W
ALLIS
. 1991. Status of the Baird’s sparrow in Alberta–1987/
1988 update with notes on other grassland sparrows and Sprague’s pipit, p. 87–89.
In
: Holroyd
et al.
(eds.). Proceedings of the second endangered species and prairie conservation workshop,
January 1989. Provincial Museum of Alberta Natural Histor y Occasional Paper 15.
W
IENS
, J. A. 1969. An approach to the study of ecological relationships among grassland birds.
Ornith.
Monogr.,
8:1–93.
W
HITMORE
, R. C. 1981. Structural characteristics of grasshopper sparrow habitat.
J. Wildl. Manage.,
45:
811–815.
W
ILSON
,S.D.
AND
J. W. B
ELCHER
. 1989. Plant and bird communities of native prairie and introduced
Eurasian vegetation in Manitoba, Canada.
Conserv. Biol.,
3:39–44.
W
INTER
, M. 1994. Habitat selection of Baird’s sparrows in the northern mixed-grass prairie. Diplomar-
beit, Tubingen University, Tubingen, Germany, 102 p.
W
RIGHT
,H.A.
AND
A. W. B
AILEY
. 1982. Fire ecology. John Wiley and Sons, New York, New York, 501 p.
Z
IMMERMAN
, J. L. 1988. Breeding season habitat selection by the Henslow’s sparrow in Kansas.
Wilson
Bull.,
100:17–24.
S
UBMITTED
11 M
AY
1999 A
CCEPTED
26 J
UNE
2000