Content uploaded by Jeffrey J. Thompson
Author content
All content in this area was uploaded by Jeffrey J. Thompson on Jul 02, 2018
Content may be subject to copyright.
Agriculture,
Ecosystems
and
Environment
154 (2012) 44–
55
Contents
lists
available
at
SciVerse
ScienceDirect
Agriculture,
Ecosystems
and
Environment
journa
l
h
o
me
pa
ge:
www.elsevier.com/locate/agee
Expansion
and
intensification
of
row
crop
agriculture
in
the
Pampas
and
Espinal
of
Argentina
can
reduce
ecosystem
service
provision
by
changing
avian
density
Gregorio
I.
Gavier-Pizarroa,∗,
Noelia
C.
Calamarib,
Jeffrey
J.
Thompsona,
Sonia
B.
Canavellib,
Laura
M.
Solaria,
Julieta
Decarrea,
Andrea
P.
Goijmana,
Romina
P.
Suareza,
Jaime
N.
Bernardosc,
María
Elena
Zaccagninia
aInstituto
Nacional
de
Tecnología
Agropecuaria
(INTA),
Centro
de
Investigación
en
Recursos
Naturales
(CIRN-IRB),
De
los
Reseros
y
Las
Caba˜
nas
S/N
HB1712WAA
Hurlingham,
Buenos
Aires
Argentina1
bInstituto
Nacional
de
Tecnología
Agropecuaria
(INTA),
EEA
Paraná,
Factores
Bióticos
y
Protección
Vegetal,
Ruta
11
Km
12.7,
Oro
Verde
(3101),
Entre
Ríos,
Argentina2
cInstituto
Nacional
de
Tecnologia
Agropecuária
(INTA),
EEA
Ing.
Agr.
Guillermo
Covas,
Ruta
Nac.
N◦5
Km
580
(6326)
C.C
N◦11
Anguil,
La
Pampa,
Argentina3
a
r
t
i
c
l
e
i
n
f
o
Article
history:
Received
2
August
2010
Received
in
revised
form
11
August
2011
Accepted
18
August
2011
Available online 19 September 2011
Keywords:
Argentina
Bird
density
Regional
bird
survey
Ecosystem
services
Espinal
Habitat
loss
Pest
control
Pampas
Row
crops
a
b
s
t
r
a
c
t
In
Argentina,
the
rapid
expansion
and
intensification
of
row
crop
production
that
has
occurred
during
the
last
20
years
has
resulted
in
the
loss
of
habitat
and
spatial
heterogeneity
in
agroecosystems.
One
of
the
principal
effects
of
industrialized
row
crop
production
is
the
loss
of
avian
diversity
and
associated
ecosystem
services
that
benefit
crop
production.
To
better
understand
the
response
of
bird
species
to
the
intensification
and
expansion
of
row
crop
agriculture
in
Argentina,
and
the
potential
effects
on
the
provision
of
ecosystems
services,
we
analyzed
the
relationship
between
short-
and
long-term
changes
in
agricultural
land
use
on
the
densities
of
six
bird
species
(Milvago
chimango,
Caracara
plancus,
Tyrannus
savana,
Zenaida
auriculata,
Molothrus
bonariensis,
and
Sturnella
supercilliaris)
using
data
from
a
large-scale,
long-term
avian
monitoring
program
in
central
Argentina.
Species
densities
responded
individually
to
long-term
landuse
changes;
T.
savana
and
M.
chimango
densities
were
positively
related
to
an
increase
in
the
annual
cropping
area,
whereas
C.
plancus
and
S.
supercilliaris
were
positively
related
to
the
area
of
non-
plowed
fields.
M.
bonariensis
and
Z.
auriculata
(considered
crop
pests)
showed
a
weak
relationship
with
land
use.
None
of
the
species
exhibited
response
to
short-term
changes
in
land-use.
Although
generalist
species
can
apparently
adapt
to
a
diversity
of
open
habitats,
species
that
provide
pest
control
services
were
also
related
to
semi-natural
habitats
and
thus
likely
to
suffer
from
land
transformation
associated
with
intensive
agricultural
management.
Our
results,
as
well
as
those
found
in
similar
systems,
denote
strong
inferential
evidence
that
the
disappearance
of
remnants
of
natural
and
semi-natural
habitats
in
heavily
transformed
agricultural
landscapes
will
have
a
substantial
negative
effect
on
the
provision
of
pest
control
services
provided
by
avian
abundance
and
diversity.
© 2011 Elsevier B.V. All rights reserved.
1.
Introduction
Land
use
changes
and
conversion
of
natural
habitats
are
trans-
forming
the
earth’s
surface
at
a
fast
pace,
particularly
due
to
modern
agriculture
and
forestry
practices
(Foley
et
al.,
2005;
Ramankutty
et
al.,
2008).
The
expansion
and
intensification
of
row
crop
pro-
duction
has
been
particularly
evident
in
Argentina,
where
the
cultivated
area
increased
45%
between
1990
and
2006,
with
half
of
the
increase
devoted
to
genetically
modified
soybean
(Aizen
et
al.,
∗Corresponding
author.
E-mail
address:
ggavier@cnia.inta.gov.ar
(G.I.
Gavier-Pizarro).
1Tel.:
+54
011
4481
2360;
fax:
+54
011
4481
2360.
2Tel.:
+54
011
0343
4975200.
3Tel.:
+54
011
02954
495057.
2008;
Oesterheld,
2008);
at
the
same
time,
the
use
of
fertilizers
increased
by
400%
(FAO,
2010;
Thompson,
2007).
Although
row
crop
expansion
has
occurred
most
dramatically
in
the
Chaco
region
of
northern
Argentina
(Grau
et
al.,
2005;
Zak
et
al.,
2004;
Grau
and
Aide,
2008;
Gasparri
and
Grau,
2009),
the
process
has
also
been
evident
in
the
Pampas
(a
region
with
more
than
a
cen-
tury
of
agricultural
use)
and
Espinal
ecoregions,
where
pastures,
natural
grasslands
and
forests
historically
used
for
cattle
grazing
have
been
converted
to
row
crop
production
(Viglizzo
et
al.,
1997;
Paruelo
et
al.,
2005;
Baldi
and
Paruelo,
2008).
The
expansion
of
cul-
tivated
land
has
been
related
to
a
combination
of
climate
change
(increasing
precipitation),
increasing
global
demand
for
agricul-
tural
products,
national
economic
policies,
and
new
technologies
(genetically
modified
seeds,
agrochemicals,
machinery)
(Viglizzo
et
al.,
1997,
2004;
Grau
et
al.,
2005;
Grau
and
Aide,
2008;
Zak
et
al.,
2008).
Concurrently,
crop
management
has
become
increasingly
intensified
(Hall
et
al.,
2001;
Viglizzo
et
al.,
2004;
Thompson,
2007).
0167-8809/$
–
see
front
matter ©
2011 Elsevier B.V. All rights reserved.
doi:10.1016/j.agee.2011.08.013
G.I.
Gavier-Pizarro
et
al.
/
Agriculture,
Ecosystems
and
Environment
154 (2012) 44–
55 45
In
Argentina,
the
main
negative
effects
from
the
recent
intensifi-
cation
and
expansion
of
row
crop
agriculture
include
agrochemical
contamination,
soil
degradation,
biodiversity
loss,
high
rates
of
deforestation
and
fragmentation
in
the
Chaco,
and
grassland
con-
version
in
the
Pampas
that
has
resulted
in
spatially
simplified
landscapes
(Zaccagnini
and
Calamari,
2001;
Paruelo
et
al.,
2005;
Boletta
et
al.,
2006;
Codesido
et
al.,
2008;
Baldi
and
Paruelo,
2008;
Oesterheld,
2008;
Gasparri
and
Grau,
2009).
Habitat
loss
and
the
reduction
of
spatial
heterogeneity
are
of
concern
due
to
their
negative
effects
on
biodiversity
and
have
been
attributed
to
a
substantial
reduction
in
avian
species
rich-
ness
in
agroecosystems
(Alkorta
et
al.,
2003;
Benton
et
al.,
2003;
Jackson
et
al.,
2007).
For
example,
in
England
four-fifths
of
bird
species,
particularly
habitat
specialists,
have
undergone
population
declines
related
to
habitat
loss
and
reduced
spatial
heterogeneity
in
agricultural
landscapes
(Robinson
and
Sutherland,
2002).
Large-
scale
studies
in
European
agroecosystems
illustrate
a
reduction
in
species
richness
of
birds
associated
with
the
loss
of
semi-natural
habitats
and
increased
fertilizer
use
(Donald
et
al.,
2001,
2006;
Billeter
et
al.,
2008).
In
the
Pampas
and
Espinal
ecoregions
of
Argentina,
region-wide
habitat
conversion
has
resulted
in
changes
in
the
composition
of
avian
communities
and
decreases
in
species
richness.
Avian
species
richness
in
the
Pampas
is
positively
associated
with
the
proportion
of
natural
vegetation
and
negatively
associated
with
the
proportion
of
cultivated
land
(Schrag
et
al.,
2009),
and
the
relative
abundance
of
most
bird
species
decreased
along
a
gradient
of
increasing
trans-
formation
from
grazing
to
row-crop
dominated
landscapes
(Filloy
and
Bellocq,
2007a).
Areas
still
predominantly
used
for
grazing
(with
greater
habitat
availability
for
grassland-dependent
species)
show
higher
avian
species
richness
than
areas
primarily
used
for
row
crop
production
(Codesido
et
al.,
2008).
In
the
Espinal
region,
avian
species
richness
decreased
with
decreasing
size
of
wood-
land
patches
(Bucher
et
al.,
2001);
however,
small
remnants
were
capable
of
supporting
relatively
high
species
richness
(1-ha
patches
retained
up
to
50%
of
birds
species)
(Dardanelli
et
al.,
2006).
Bird
responses
to
the
expansion
of
row
crop
production
have
been
shown
to
be
dependent
upon
functional
groups,
with
insec-
tivore
and
raptor
species
demonstrating
a
greater
sensitivity
to
increasing
areas
in
row
crops
than
granivores
(Carrete
et
al.,
2009;
Zaccagnini
et
al.,
2011).
In
addition,
species-specific
responses
within
functional
groups
are
variable.
For
example,
a
study
of
rap-
tor
communities
in
Argentina
suggested
a
negative
response
to
habitat
transformation,
but
three
species
(Coragyps
atratus,
Elanus
leucurus,
Falco
sparverius)
peaked
in
relative
abundance
in
a
mosaic
of
transformed
and
natural
habitats.
Moreover,
Milvago
chimango
increased
with
increasing
habitat
conversion
and
the
presence
of
another
species
(Circus
buffoni)
increased
in
grassland-dominated
landscapes
(Carrete
et
al.,
2009;
Pedrana
et
al.,
2008).
Other
examples
include
decreasing
population
and
distribu-
tion
of
the
Pampas
meadowlark
(Sturnella
defilippii)
associated
with
grassland
conversion
and
overgrazing
(Fernández
et
al.,
2003),
and
greater
survival
of
the
spotted
tinamou
(Nothura
mac-
ulosa)
in
agroecosystems
in
the
province
of
Buenos
Aires
in
a
mixed
landscape
of
pasture
and
agriculture
compared
to
an
agriculture-dominated
landscape
(Thompson
and
Carroll,
2009).
Species-specific
variations
in
response
to
habitat
area
were
also
illustrated
for
birds
in
Espinal
woodlands
within
agrosecosys-
tems
in
Entre
Ríos
where
the
relative
abundance
of
most
species
decreased
with
decreasing
patch
size
(Calamari
and
Zaccagnini,
2007).
What
are
the
consequences
of
changes
of
avian
species
diversity
on
agroecosystems?
Biodiversity
is
a
key
component
of
agroe-
cosystems.
Plant
and
animals
regulate
the
flux
of
energy
and
matter
(water,
nutrients)
and
most
associated
ecological
processes
(e.g.,
seed
dispersion,
pollination)
that
are
fundamental
for
the
sustainability
and
resilience
of
agroecosystems
(Altieri,
1999).
Birds
are
particularly
important
components
of
agroecosystems,
exhibiting
the
most
diverse
range
of
ecological
functions
among
vertebrates.
Their
high
diversity
of
adaptations
and
life
histories,
high
numbers
and
mobility
allow
birds
to
regulate
many
ecosystem
processes
and
to
respond
very
quickly
to
changes
in
resource
levels
(S¸
ekercio˘
glu
et
al.,
2004;
S¸
ekercio˘
glu,
2006).
As
a
consequence,
birds
provide
several
ecosystem
services
that
are
important
for
the
function
and
sustainability
of
agroecosystems,
including
regu-
lating
(seed
dispersal,
pollination,
pest
control,
carcass
and
waste
disposal)
and
supporting
services
(nutrient
deposition,
ecosystem
engineering)
(S¸
ekercio˘
glu,
2006;
Whelan
et
al.,
2008).
Although
most
of
these
services
are
difficult
to
quantify,
pest
control
is
one
of
the
most
important
services
to
agricultural
pro-
duction
provided
by
birds,
with
several
studies
showing
substantial
decreases
of
pest
species
and
increases
in
crop
production
related
to
the
predatory
activities
of
birds
(see
review
in
Whelan
et
al.,
2008).
Predation
by
insectivorous
birds
reduced
pest
damage
lev-
els
in
coffee
plantations
by
1–14%,
increasing
the
production
value
by
US$44–$105/ha
in
2005/2006
(Kellermann
et
al.,
2008).
More-
over,
the
use
of
nesting
boxes
to
attract
great
tits
(Parus
major)
reduced
caterpillars
densities
and
fruit
damage
while
increasing
apple
yield
by
66%
(Mols
and
Visser,
2002).
Research
on
the
con-
trolling
effects
of
raptors
on
rodents
and
avian
agricultural
pests
is
limited
(S¸
ekercio˘
glu
et
al.,
2004);
however,
the
importance
of
rodents
in
the
diet
of
raptors
suggests
that
these
birds
are
beneficial
species
for
agriculture
(Whelan
et
al.,
2008).
For
example,
higher
numbers
of
diurnal
raptors
around
soybean
fields
decreased
pop-
ulation
numbers
and
growth
rate
of
house
mice
(Mus
domesticus)
(Kay
et
al.,
1994).
Functional
richness
theory
relates
the
level
of
ecosystem
ser-
vices
provided
by
birds
to
the
diversity
of
species
providing
it.
More
species
represent
a
larger
number
of
adaptations
and
henceforth
a
more
efficient
use
of
the
resource.
As
a
consequence,
species
richness
is
considered
to
be
directly
related
to
the
level
of
ser-
vice
provided
(Philpott
et
al.,
2009).
Common
species
(usually
the
most
abundant),
however,
have
the
greatest
effect
on
ecosystem
processes
(Gaston,
2010)
and
subsequently
the
level
of
ecosystem
services
provided
by
those
species
also
depends
on
the
abundance
of
those
species
(S¸
ekercio˘
glu
et
al.,
2004;
Swift
et
al.,
2004;
Wilby
et
al.,
2005).
Expansion
and
intensified
management
of
rows
crops
are
expected
to
continue
to
significantly
alter
habitat
quality,
quantity,
and
configuration
in
Argentina;
hence,
assessing
and
determining
the
implications
of
these
landscape-modifying
processes
for
the
provision
of
ecosystem
services,
via
the
effects
on
avian
popula-
tions,
are
imperative
to
ensure
the
provision
of
those
services.
For
this
purpose,
we
used
data
from
a
long-term,
large-scale
moni-
toring
program
to
analyze
the
short-
and
long-term
relationship
between
densities
of
a
suite
of
avian
species
(which
provide
pest
control
services)
and
the
expansion
and
intensified
management
of
row
crops
in
the
Argentine
Pampas
and
Espinal
agroecosystems,
assuming
a
direct
relationship
between
bird
abundance
and
the
level
of
pest
control
service
(Philpott
et
al.,
2009;
Swift
et
al.,
2004;
Wilby
et
al.,
2005).
2.
Methods
2.1.
Study
area
The
study
area
comprised
128,200
km2of
the
Pampa
and
Espinal
ecoregions
(Cabrera,
1994)
characterized
by
annual
mean
minimum
and
maximum
temperatures
of
13 ◦C
and
23 ◦C,
respec-
tively
(Soriano,
1992)
and
1000
mm
of
mean
annual
precipitation
(Ferreyra
et
al.,
2001;
Messina
et
al.,
1999;
Podestá
et
al.,
2002).
This
46 G.I.
Gavier-Pizarro
et
al.
/
Agriculture,
Ecosystems
and
Environment
154 (2012) 44–
55
area
is
the
principal
agricultural
production
region
in
Argentina,
increasingly
dominated
by
annual
row
crops
(wheat,
soybean,
sun-
flower,
sorghum
and
corn)
and
has
been
highly
modified
due
to
agricultural
expansion
and
intensification
(León
et
al.,
1984;
Viglizzo
et
al.,
2004).
The
Espinal
portion
extends
irregularly
from
the
center
of
province
of
Santa
Fe,
northeastern
Córdoba
and
northern
Entre
Ríos
and
supports
remnant
xerophilic
woodlands
dominated
by
Prosopis
nigra,
Acacia
caven,
and
Geoffroea
decorticans.
These
woodland
rem-
nants
are
isolated
and
immersed
in
an
agricultural
matrix.
The
Pampean
portion
covers
southern
Santa
Fe,
south-central
Córdoba
and
central
Entre
Ríos
provinces,
and
is
dominated
by
grasslands
mainly
composed
of
Stipa
sp.,
Briza
sp.,
Bromus
sp.,
Poa
sp.
(Cabrera,
1971).
2.2.
Data
analysis
We
analyzed
the
relationship
between
avian
species
abundance
and
land
use
change
using
two
sets
of
variables:
estimates
of
species
density
(dependent
variables)
and
environmental
factors
(indepen-
dent
variables).
Using
these
variables
we
conducted
two
analyses
representing
the
short
and
long-term
effects
of
habitat
conversion
on
species
densities.
2.2.1.
Bird
species
density
estimation
(dependent
variables)
In
January
each
year
from
2003
to
2009
(austral
breeding
season)
we
surveyed
47
transects
(routes),
located
along
unpaved
sec-
ondary
and
tertiary
roads
within
different
agro-production
zones
(areas
differing
in
land
use,
land
cover
and
economic
activities)
in
east-central
Argentina.
Transect
locations
were
chosen
using
a
geographically-stratified
systematic
design
that
consisted
in
apply-
ing
a
30
km
×
30
km
grid
over
a
map
of
the
study
area
and
defining
eight
strata
consisting
of
agro-production
zones
and
provincial
boundaries
(Fig.
1).
Within
each
stratum,
grid
cells
were
selected
systematically
(every
other
cell)
with
the
number
of
cells
propor-
tional
to
the
area
of
each
zone.
Within
each
cell
the
route
and
the
direction
for
the
route
to
be
surveyed
were
randomly
selected
among
all
possible
alternatives
(Zaccagnini
et
al.,
2010).
Each
route
had
30
permanently
marked
points,
spaced
at
1
km
intervals,
with
the
first
point
on
the
route
randomly
placed.
At
each
point
six
avian
species
were
surveyed
between
6:00–11:00
am
and
15:00–20:00
pm
using
distance
sampling
(Buckland
et
al.,
2001)
by
a
pair
of
experienced
observers.
At
each
point
one
observer
deter-
mined
the
distance
to
and
number
of
each
species
detected
during
a
3-min
period
whereas
the
other
observer
recorded
data
on
land
use.
Laser
rangefinders
were
used
to
measure
distances
to
individual
birds
or
to
the
center
of
flocks.
We
selected
6
species
based
upon
their
extensive
distribution
within
the
study
area
and
relatively
high
number
of
detections.
Two
species
were
insectivores
(Fork-tailed
flycatcher
Tyrannus
savana,
White-browed
blackbird
Sturnella
supercilliaris)
and
two
were
raptors
(Chimango
caracara
Milvago
chimango
and
Southern
crested-caracara
Caracara
plancus),
whose
diet
includes
rodents
and
insects.
We
also
included
two
species
that
are
considered
crop
pests
(Shiny
cowbird
Molothrus
bonariensis
and
Eared
dove
Zenaida
auriculata)
to
compare
their
response
to
land
use
change
with
that
of
species
providing
pest
control
services.
This
compari-
son
can
have
important
management
implications
if
management
to
enhance
pest
control
by
birds
also
has
a
positive
influence
on
the
abundance
of
pest
species.
Species
densities
were
estimated
for
each
year
using
DISTANCE
5.0,
a
computer
package
for
the
analysis
of
distance
sampling
data
that
corrects
for
incomplete
detection
in
density
estimates
(Buckland
et
al.,
2001;
Thomas
et
al.,
2002).
After
exploratory
analysis
of
the
data
(sensu
Buckland
et
al.,
2001;
Thomas
et
al.,
2002,
including
histograms
of
the
distance
data
under
several
grouping
factors
to
detect
and
correct
for
the
pres-
ence
of
heaping,
evasive
movement,
outliers)
we
set
the
truncation
distance
w
at
250
m,
the
distance
that
included
90%
of
detections
for
all
species
combined
and
manually
selected
7
distance
inter-
vals
with
cut
off
points
based
on
the
distribution
of
observations
at
different
distances.
Density
estimates
were
derived
with
detec-
tion
models
using
a
combination
of
3
monotonic,
decreasing
key
functions
(uniform,
half-normal,
and
hazard
rate)
and
2
adjust-
ment
terms
(cosine
and
polynomial)
and
best
models
chosen
using
Akaike
Information
Criteria
(AIC)
and
model
weights.
To
facili-
tate
multi-species
analysis
we
selected
one
model
(half
normal
key
function
+
cosine
adjustment
term)
for
estimating
density
of
all
species
based
on
its
better
performance
for
the
majority
of
species.
Given
a
relatively
low
number
of
detections
for
some
species
during
each
year
we
estimated
the
detection
probability
function
globally
(e.g.,
all
years
combined)
for
each
species,
and
estimated
density
for
each
year
on
each
route
using
stratification
(by
year)
and
post-
stratification
(by
route)
(Buckland
et
al.,
2001).
2.2.2.
Environmental
(independent)
variables
2.2.2.1.
Land
use.
Land
use
was
recorded
annually
at
each
transect
point
in
a
200-m
radius
circle
centered
on
each
point,
as
estimates
of
percent
cover
of
five
land-use
classes
(Schrag
et
al.,
2009).
Land-
use
classes
included:
(1)
annual
crops
(i.e.,
soybeans,
sorghum,
sunflower,
wheat,
corn);
(2)
managed
pastures
(both
annual
and
perennial
species);
(3)
non-plowed
fields,
including
(a)
agricultural
fields
that
have
been
resting
for
more
than
a
season
covered
with
a
mixture
of
herbaceous
and
grassy
vegetation,
or
(b)
natural
and
semi-natural
grasslands
used
for
cattle
ranching;
(4)
forest
(both
native
and
exotic
species);
and
(5)
other
uses
(including,
but
not
limited
to,
aquatic
habitats).
The
percent
cover
for
all
points
was
averaged
among
the
30
observation
points
to
obtain
a
single
value
for
each
transect
per
year.
2.2.2.2.
Enhanced
vegetation
index.
Enhanced
vegetation
index
(EVI),
which
provides
a
consistent
and
permanent
comparison
of
temporal
changes
in
vegetation
is
a
MODIS
(Moderate-Resolution
Imaging
Spectroradiometer)
product
MOD13Q1
(image
acquisition
every
16
days)
with
a
250-m
spatial
resolution
(http://modis-
land.gsfc.nasa.gov/vi.htm).
EVI
measures
the
amount
of
photosyn-
thetic
tissue
as
an
index
of
plant
productivity
that
corrects
for
distortions
in
the
reflected
light
caused
by
airborne
particles
as
well
as
ground
cover
below
the
vegetation
(Jiang
et
al.,
2008).
For
each
year,
we
extracted
the
EVI
value
for
the
pixel
containing
each
sam-
ple
point
and
averaged
those
values
to
obtain
a
single
value
per
transect.
2.2.2.3.
Precipitation.
Precipitation
data
for
each
transect
were
collected
from
the
nearest
meteorological
station
of
the
Insti-
tuto
Nacional
de
Tecnología
Agropecuaria
(INTA).
Since
avian
abundance
may
have
been
associated
with
precipitation
before
sampling,
we
included
mean
monthly
precipitation
from
Septem-
ber
to
January
for
each
year.
2.2.2.4.
Latitude
and
longitude.
Latitude
and
longitude
for
the
cen-
tral
point
of
each
survey
transect
(Gauss–Kruger
zone
3)
was
included
in
the
analysis
to
account
for
large-scale
spatial
patterns
that
could
mask
the
relationship
between
avian
densities
and
land
use
(e.g.,
gradient
in
species
abundance
related
to
variations
in
the
geographic
distribution)
and
account
for
potential
correlation
of
other
variables
that
could
be
responding
similarly
to
the
same
large-scale
gradients.
2.2.3.
Statistical
analyses
We
used
regression
analysis
to
investigate
the
relationship
among
species
densities
(dependent
variables)
and
environment
G.I.
Gavier-Pizarro
et
al.
/
Agriculture,
Ecosystems
and
Environment
154 (2012) 44–
55 47
Fig.
1.
Study
area
and
location
of
survey
transects.
variables
(independent
variables).
We
divided
the
analysis
into
two
parts
to
assess
the
long-term
and
short-term
relationship
between
bird
densities
and
habitat
variables.
We
analyzed
the
long-term
relationship
using
the
mean
values
from
2007
to
2009
for
both
dependent
and
independent
variables,
trading
space
for
time
to
establish
a
gradient
representing
the
long-term
transformation
from
totally
transformed
areas
to
those
dominated
by
natural
veg-
etation.
For
the
analysis
of
the
short-term
relationship
between
bird
densities
and
habitat
variables
we
used
regression
analysis
of
the
changes
in
mean
densities
and
habitat
values
between
2003-
2005
and
2007-2009
to
explore
species
response
to
land
use
change
during
the
study
period.
Throughout
the
analysis
we
used
the
same
approach
that
included
three
steps.
In
the
first
step,
we
tested
for
correlations
among
independent
variables
using
a
Pearson’s
correlation
matrix
for
explanatory
variables
to
avoid
colinearity
issues
and
discarded
habitat
variables
that
showed
high
levels
of
correlation
(r
≥
0.70).
In
the
first
part
of
the
analysis
(long-term
temporal
effects),
rainfall
and
annual
crops
were
correlated
with
longitude
and
proportional
forest
area,
respectively,
above
0.7
Pearson
correlation
coefficient.
Hence,
we
included
rainfall
instead
of
longitude
in
the
models,
because
rainfall
accounts
for
an
east–west
gradient,
whereas
for
the
area
in
annual
crops
and
forest
we
selected
the
variable
that
best
explained
species
density
for
each
particular
species.
In
the
second
part
of
the
analyses
(short-term
temporal
effects),
we
included
all
variables
plus
latitude
and
rainfall
instead
of
longitude.
We
did
not
use
a
change
variable
for
forest,
because
there
was
an
alteration
in
the
sampling
protocol
related
to
the
estimation
of
forest
cover
during
the
period
of
the
entire
study.
In
the
second
step
of
the
analysis,
we
fitted
multiple
regression
linear
models
using
the
variables
selected
in
the
first
step.
First,
we
fitted
a
full
model
using
all
habitat
variables
and
then
we
used
step-
wise
selection
to
eliminate
non-significant
variables
and
obtain
a
reduced
model
(Chatterjee,
2001).
We
used
the
full
and
reduced
models
as
a
measure
of
variance
in
species
densities
given
our
set
of
explanatory
variables.
Assumptions
of
linear
regression
were
verified.
If
necessary,
densities
were
log-transformed
to
meet
lin-
earity
assumptions.
Spatial
autocorrelation
of
the
model
residuals
were
tested
with
a
semi-variogram
randomization
analysis
(Isaaks
and
Srivastava,
1989).
The
full
and
reduced
models
were
used
to
estimate
the
variance
in
densities
accounted
for
in
the
analysis
and
the
direction
of
the
effects
of
the
independent
variables.
In
the
third
step,
we
assessed
the
importance
of
each
individual
variable
in
explaining
species
density.
Stepwise
selection
has
limi-
tations
since
it
identifies
one
best
model
(among
several
that
could
explain
the
responses
equally
well)
and
does
not
provide
infor-
mation
about
the
amount
of
variance
explained
by
each
variable
(Whittingham
et
al.,
2006).
Since
our
main
focus
was
to
understand
the
habitat
variables
that
most
affected
bird
densities
rather
than
fitting
the
best
explanatory
model,
we
used
best
subsets
and
hier-
archical
partitioning
analysis
to
assess
the
importance
of
variables
included
in
the
models.
The
best
subset
method
uses
Bayesian
Information
Criterion
(BIC)
to
obtain
a
subset
of
models
that
best
explains
the
response.
The
approach
performs
an
exhaustive
search
of
all
possible
mod-
els
and
the
maximum
number
of
predictors
allowed
is
specified
a
priori
(Miller,
1990).
Fitting
several
models
instead
of
one
“best”
Fig.
2.
Variation
in
mean
densities
during
2007–2009
for
the
six
bird
species
included
in
the
study.
The
black
line
inside
each
box
is
the
median,
the
lower
and
upper
edges
of
the
boxes
are
the
0.25
and
0.75
quantiles,
and
the
lines
represent
the
range
of
the
data.
Outliers
are
denoted
with
a
circle.
48 G.I.
Gavier-Pizarro
et
al.
/
Agriculture,
Ecosystems
and
Environment
154 (2012) 44–
55
Fig.
3.
Distribution
of
mean
bird
densities
and
percentage
of
annual
crops
in
the
study
area
during
the
period
2007–2009.
Maps
of
the
proportional
annual
crop
area
were
obtained
by
extrapolation
of
the
2007–2009
mean
for
each
survey
transect
using
inverse
distance
weighting.
G.I.
Gavier-Pizarro
et
al.
/
Agriculture,
Ecosystems
and
Environment
154 (2012) 44–
55 49
model
highlights
which
variables
are
repeatedly
chosen
in
the
best
models,
and
whether
they
have
a
consistent
effect
on
the
response
variable
(i.e.,
negative
or
positive
coefficient).
We
used
three
vari-
ables
to
fit
the
model
and
considered
the
10
best
models
explaining
bird
densities
obtained
in
each
best
subsets
analysis.
We
then
counted
the
number
of
times
that
each
variable
was
included
in
the
10
best
models
to
determine
an
importance
value
of
each
vari-
able
in
the
model
subset.
We
performed
one
subset
analysis
for
each
bird
species
in
both
parts
of
the
analysis.
Hierarchical
partitioning
analysis
calculates
the
percent
of
vari-
ance
of
the
full
model
explained
by
each
variable
when
all
other
variables
are
included
in
the
model.
For
estimated
densities
of
each
bird
species,
all
possible
models
based
on
different
combinations
of
the
habitat
variables
were
fit.
For
each
fitted
model
the
vari-
able
of
interest
was
dropped
and
the
model
was
fitted
again.
The
importance
of
that
variable
was
calculated
as
the
average
change
in
R2when
the
variable
was
dropped
from
all
of
the
fitted
models
(MacNally,
2002).
For
each
bird
species
we
presented
results
of
both
hierarchical
partitioning
and
best
subset
analysis
for
each
habitat
(explanatory
variables)
included
in
the
analysis.
Best
subset
analysis
indicated
which
variables
were
most
strongly
associated
with
bird
densities
(importance
value)
and
hierarchical
partitioning
analysis
indicated
the
proportion
of
variance
explained
by
each
variable
of
the
total
variance
included
in
the
full
model.
We
also
included
the
full
and
reduced
models.
3.
Results
3.1.
Trends
in
species
densities
and
distributions
During
2007–2009
Z.
auriculata
had
the
highest
density,
with
an
average
of
12
individuals
per
ha,
whereas
for
the
same
period
estimates
of
mean
density
for
T.
savana,
M.
bonariensis
and
S.
supercilliaris
were
similar
to
one
another
but
much
lower
than
Z.
auriculata
(Fig.
2).
The
two
raptor
species,
M.
chimango
and
C.
plancus,
had
the
lowest
densities
(less
than
1
individual
per
ha).
Furthermore,
densities
of
Z.
auriculata,
T.
savana
and
M.
bonariensis
were
highly
variable
across
survey
transects
(Fig.
2).
The
spatial
distribution
of
species
density
showed
three
differ-
ent
patterns
across
the
study
area.
Densities
of
S.
supercilliaris,
M.
chimango,
C.
plancus
and
T.
savana
were
highest
in
the
west-central
portion
of
the
study
area,
where
land
use
is
dominated
by
row
crop
agriculture.
This
distribution
pattern
was
most
strongly
related
to
land
use
for
S.
supercilliaris
and
M.
chimango,
whereas
a
weaker
relationship
between
land
use
and
density
for
C.
plancus
and
T.
savana
was
observed
(Fig.
3).
Densities
of
Z.
auriculata
and
M.
bonar-
iensis
were
not
related
to
the
proportion
of
cultivated
area,
with
the
density
of
Z.
auriculata
decreasing
along
a
north-south
gradi-
ent
and
the
density
of
M.
bonariensis
showing
a
weak
tendency
to
decrease
along
a
west-east
gradient
(Fig.
3).
Between
2003–2005
and
2007–2009,
the
average
estimated
densities
of
most
species
were
stable;
however,
Z.
auriculata
and
M.
bonariensis
exhibited
a
tendency
towards
increasing
densities
between
these
periods
(Fig.
4).
3.2.
The
relationship
between
land
use
and
species
density
3.2.1.
Long-term
effects
During
the
2007–2009
period,
the
density
of
S.
supercilliaris
was
most
strongly
related
to
land
use,
followed
by
that
of
M.
chimango
(multivariate
models
explaining
about
60%
and
40%
of
the
variation
in
estimated
densities,
respectively;
Table
1).
Estimated
densities
of
C.
plancus
and
T.
savana
were
related
to
land
use
at
an
intermedi-
ate
level
(multivariate
models
explaining
30%
of
the
variation
in
Fig.
4.
Variation
in
changes
of
mean
densities
of
focal
bird
species
between
the
periods
2003–2005
and
2007–2009.
The
black
line
inside
each
box
is
the
median,
the
lower
and
upper
edges
of
the
boxes
are
the
0.25
and
0.75
quantiles,
and
the
lines
represent
the
range
of
the
data.
Outliers
are
denoted
with
a
circle.
estimated
densities),
whereas
Z.
auriculata
and
M.
bonarien-
sis
exhibited
the
weakest
relationship
(multivariate
models
explaining
21%
and
14%
of
the
variation
in
estimated
densities,
respectively)
(Table
1;
Fig.
5).
Estimated
densities
of
S.
supercilliaris
demonstrated
a
strong
negative
relationship
to
increasing
forest
area
and
precipitation,
whereas
the
density
of
M.
chimango
was
positively
related
to
the
annual
cropping
area
and
negatively
to
rainfall
(Table
1;
Fig.
5).
Densities
of
T.
savana
and
C.
plancus
were
positively
and
most
strongly
related
to
the
area
of
non-plowed
fields
(Fig.
5).
In
addi-
tion,
T.
savana
density
was
negatively
related
to
latitude
(increasing
density
towards
the
south)
and
C.
plancus
density
decreased
with
increasing
forest
area
and
rainfall
(Fig.
5).
Z.
auriculata
and
M.
bonariensis
densities
were
not
explicitly
related
to
land
use
at
the
scale
of
analysis.
Latitude
(with
a
positive
effect)
was
the
most
important
variable
explaining
distributions,
but
only
explained
20%
of
the
variation
in
estimated
densities
in
both
the
full
and
reduced
models.
M.
bonariensis,
however,
also
exhibited
a
negative
relationship
with
rainfall
along
an
east-west
gradient
(Table
1;
Fig.
5).
3.2.2.
Short-term
effects
Changes
in
land
use
did
not
clearly
explain
changes
in
species
densities
between
the
2003–2005
and
the
2007–2009
periods,
with
the
multiple
models
equally
explaining
a
small
proportion
of
the
variation
in
densities.
The
species
with
the
largest
change
in
density
(36%)
was
M.
bonariensis,
although
a
reduced
multivariate
model
including
latitude,
change
in
the
area
of
annual
crops
and
change
in
EVI
explained
only
14%
of
the
variation
(Table
2).
Latitude
explained
65%
of
the
variation
within
the
model
and
was
present
in
all
mod-
els
in
the
best
subset
analysis
(M.
bonariensis
densities
tended
to
increase
northward).
For
C.
plancus,
a
multivariate
model
including
changes
in
the
area
of
non-plowed
fields
and
pastures,
EVI,
and
rainfall
explained
26%
of
the
variation
in
density
(p
<
0.05);
EVI
was
the
most
important
variable,
accounting
for
35%
of
the
variation
explained
in
the
model
and
was
included
in
9
of
the
10
models
in
the
best
subset
analy-
sis.
C.
plancus
density
tended
to
increase
in
areas
that
underwent
increases
in
EVI
between
2003
and
2009
(Table
2).
Changes
in
estimated
densities
of
M.
chimango
and
S.
supercil-
liaris
were
very
low.
M.
chimango
density
tended
to
increase
with
an
increase
in
rainfall
and
pastures
(Table
2);
the
multivariate
model
50 G.I.
Gavier-Pizarro
et
al.
/
Agriculture,
Ecosystems
and
Environment
154 (2012) 44–
55
Table
1
Response
of
bird
densities
to
long-term
agricultural
expansion
in
the
Pampas
and
Espinal
ecoregions.
Full
multiple
regression
models
and
reduced
models
(fitted
by
stepwise
selection)
explaining
mean
densities
for
the
period
2007–2009
of
six
birds
species,
p-values
and
R2of
each
model
are
shown.
For
each
model,
variables
included
are
presented
with
the
sign
of
the
effect,
the
slope
parameter
and
the
variable
name.
An
*
indicates
a
significance
level
of
p
≤
0.05
and
a•indicates
marginal
statistical
significance
(0.05
≤
p
≤
0.06).
Non-significant
statistical
results
for
the
models
are
indicated
as
ns.
Species/models
Model
structure
R2p-Value
M.
chimango
Full
model 17.73Intercept
−
3.1e−6Latitude
+
0.047Crops*
+
0.042Non-plowed
−
0.007Rain
+
0.871Evi
+
0.042Pastures
0.37
<0.0001
Reduced
model
−3.02Intercept
+
0.058Crops*
+
0.053
Non-plowed*
−
0.06Rain
0.4
<0.0001
C.
plancus
Full
model −1.47Intercept
+
2.5e−7Latitude
+
0.004Non-plowed*
−3.9e−4Rain
−0.008Forest
+
0.185Evi
−
0.003Pastures 0.30 0.0013
Reduced
model 0.21Intercept*
+
0.004Non-plowed*
−0.0006Rain*
+
0.164Evi
0.31
<0.0005
S.
supercilliaris
Full
model
−8.05Intercept
+
1.89e−6Latitude
−
0.059Forest*
+
0.027Non-plowed*
−
0.0071Rain*
+
2.53Evi•−
0.0199Pastures
0.59
<0.0001
Reduced
model