Available via license: CC BY-NC-ND 4.0
Content may be subject to copyright.
Please
cite
this
article
in
press
as:
Cousino,
L.K.,
et
al.,
Modeling
the
effects
of
climate
change
on
water,
sediment,
and
nutri-
ent
yields
from
the
Maumee
River
watershed.
J.
Hydrol.:
Reg.
Stud.
(2015),
http://dx.doi.org/10.1016/j.ejrh.2015.06.017
ARTICLE IN PRESS
G Model
EJRH-105;
No.
of
Pages
14
Journal
of
Hydrology:
Regional
Studies
xxx
(2015)
xxx–xxx
Contents
lists
available
at
ScienceDirect
Journal
of
Hydrology:
Regional
Studies
j
o
ur
nal
ho
me
pag
e:
www.elsevier.com/locate/ejrh
Modeling
the
effects
of
climate
change
on
water,
sediment,
and
nutrient
yields
from
the
Maumee
River
watershed
Luke
K.
Cousino
∗
,
Richard
H.
Becker,
Kirk
A.
Zmijewski
1
The
University
of
Toledo,
Department
of
Environmental
Sciences,
2801
West
Bancroft
Street,
Toledo,
OH
43606,
USA
a
r
t
i
c
l
e
i
n
f
o
Article
history:
Received
7
March
2015
Received
in
revised
form
15
June
2015
Accepted
17
June
2015
Available
online
xxx
Keywords:
SWAT
Climate
change
Flow
Sediment
Lake
Erie
CMIP5
a
b
s
t
r
a
c
t
Study
region:
Harmful
algal
blooms
(HABs)
in
the
Western
Basin
(WB)
of
Lake
Erie
have
been
linked
to
nonpoint
pollution
from
agricultural
watersheds.
The
Maumee
River
watershed
is
the
largest
in
the
Great
Lakes
region
and
delivers
the
biggest
sediment
and
nutrient
load
to
Lake
Erie.
Study
focus:
Climate
change
could
alter
the
magnitude
and
timing
of
sediment
and
nutrient
delivery
to
Lake
Erie’s
WB.
Data
from
four
Coupled
Model
Intercomparison
Project
Phase
5
(CMIP5)
models
were
inputted
into
a
calibrated
Soil
and
Water
Assessment
Tool
(SWAT)
model
of
the
Maumee
River
watershed
to
determine
the
effects
of
climate
change
on
water-
shed
yields.
Tillage
practices
were
also
altered
within
the
model
to
test
the
effectiveness
of
conservation
practices
under
climate
change
scenarios.
New
hydrological
insights
for
the
region:
Moderate
climate
change
scenarios
reduced
annual
flow
(up
to
−24%)
and
sediment
(up
to
−26%)
yields,
while
a
more
extreme
scenario
showed
smaller
flow
reductions
(up
to
−10%)
and
an
increase
in
sediment
(up
to
+11%).
No-till
practices
had
a
negligible
effect
on
flow
but
produced
16%
lower
average
sediment
loads
than
scenarios
using
current
watershed
conditions.
At
high
implementation
rates,
no-till
practices
could
offset
any
future
increases
in
annual
sediment
loads,
but
they
may
have
var-
ied
seasonal
success.
Regardless
of
future
climate
change
intensity,
increased
remediation
efforts
will
likely
be
necessary
to
significantly
reduce
HABs
in
Lake
Erie’s
WB.
©
2015
The
Authors.
Published
by
Elsevier
B.V.
This
is
an
open
access
article
under
the
CC
BY-NC-ND
license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
1.
Introduction
1.1.
Study
area
and
background
The
Maumee
River
is
over
210
km
(130
mi)
long
and
drains
an
area
over
17,100
km
2
(6600
mi
2
)
in
Ohio,
Michigan,
and
Indiana,
making
it
the
largest
watershed
in
the
Great
Lakes
region
(USEPA,
2013)
(Fig.
1).
The
Maumee
River
is
one
of
the
five
rivers
(St.
Clair,
Huron,
Raisin,
and
Sandusky)
that
discharge
into
the
Western
Basin
of
Lake
Erie.
Of
all
the
Great
Lakes,
Lake
Erie
is
the
shallowest
(average
depth
of
19
m),
southernmost,
and
most
productive
(Botts
and
Krushelnicki,
1987).
Of
the
three
main
basins
that
make
up
Lake
Erie
(Western,
Central,
Eastern)
the
Western
is
the
shallowest
with
an
average
depth
of
less
than
8
meters.
Lake
Erie
and
the
surrounding
watershed
provide
the
fresh
water
supply
for
over
11
million
people
and
is
home
to
a
$20
million
annual
fishery,
making
it
an
important
economic
and
natural
resource
(Botts
and
Krushelnicki,
∗
Corresponding
author.
Tel.:
+1
7347909878.
E-mail
address:
luke.cousino@rockets.utoledo.edu
(L.K.
Cousino).
1
Present
address:
BCG
Center
for
Knowledge
and
Analytics,
53
State
Street,
7th
Floor,
Boston,
MA
02109,
USA.
http://dx.doi.org/10.1016/j.ejrh.2015.06.017
2214-5818/©
2015
The
Authors.
Published
by
Elsevier
B.V.
This
is
an
open
access
article
under
the
CC
BY-NC-ND
license
(
http://creativecommons.org/licenses/by-nc-nd/4.0/).
Please
cite
this
article
in
press
as:
Cousino,
L.K.,
et
al.,
Modeling
the
effects
of
climate
change
on
water,
sediment,
and
nutri-
ent
yields
from
the
Maumee
River
watershed.
J.
Hydrol.:
Reg.
Stud.
(2015),
http://dx.doi.org/10.1016/j.ejrh.2015.06.017
ARTICLE IN PRESS
G Model
EJRH-105;
No.
of
Pages
14
2
L.K.
Cousino
et
al.
/
Journal
of
Hydrology:
Regional
Studies
xxx
(2015)
xxx–xxx
Fig.
1.
Location
of
the
Maumee
River
watershed.
The
watershed
is
the
largest
in
the
Great
Lakes
region,
consisting
of
seven
HUC
8
subbasins
that
drain
into
the
WB
of
Lake
Erie.
Flow
and
sediment
measurements
were
taken
at
Waterville,
Ohio
(denoted
with
a
star),
allowing
characterization
of
over
96%
of
the
watershed.
1987;
Kinnunen,
2003).
However,
high
population
areas
and
extensive
agricultural
land
in
the
contributing
drainage
basin
coupled
with
shallow
depth
make
the
Western
Basin
susceptible
to
anthropogenic
influence
(Richards
et
al.,
2009,
2010).
1.2.
Problem
description
Lake
Erie
has
a
history
of
environmental
problems
due
to
excess
sediment
and
nutrient
loading.
One
of
the
primary
problems
is
eutrophication
from
phosphorous
loading
from
point
and
non-point
sources.
Phosphorus
is
considered
the
limiting
nutrient
in
aquatic
ecosystems
such
as
Lake
Erie,
and
primary
production
is
well-correlated
with
phosphate
input
(
Logan,
1987).
Large
phosphorus
inputs
to
Lake
Erie
lead
to
extensive
algal
blooms.
These
blooms
reduce
water
quality,
increase
water
treatment
costs,
introduce
foul
taste
and
odors,
and
lead
to
hypoxic
conditions.
In
recent
years,
harmful
algal
blooms
(HABs)
containing
the
cyanobacteria
Microcystis
have
increased
in
frequency
and
number
in
the
Western
Basin.
Microcystis
produces
a
toxin
called
microcystin
which
can
be
harmful
to
human
health
in
high
concentrations
(Dyble
et
al.,
2008
).
High
bloom
activity
increases
water
treatment
costs,
and
beach
advisories
limit
swimming
and
other
recreational
activities,
reducing
tourism
revenues
(Watson
et
al.,
2008).
Excess
sediment
runoff
also
has
an
economic
impact
on
shipping
in
the
Western
Basin.
The
Port
of
Toledo
is
a
major
shipping
hub
on
the
Great
Lakes
and
requires
annual
dredging
of
the
channel
to
allow
passage
of
large
ships.
More
than
870,000
m
3
of
sediment
must
be
removed
from
Maumee
Bay
and
the
lower
part
of
the
Maumee
River
at
a
cost
of
over
$2
million
annually
(Myers
and
Metzker,
2000).
The
dredged
material
must
be
disposed
of
in
confined
disposal
facilities
due
to
high
concentrations
of
PCBs
and
heavy
metals
(mercury,
lead,
etc.)
as
a
result
of
industrial
practices
over
the
past
half
century
(Myers
and
Metzker,
2000).
The
Detroit
River
contributes
95%
of
the
total
incoming
water
to
Lake
Erie
from
the
upper
Great
Lakes,
but
the
Maumee
River
contributes
44%
of
the
total
river
sediment
input
to
Lake
Erie
(Kemp
et
al.,
1977;
Botts
and
Krushelnicki,
1987).
About
72%
of
the
Maumee
watershed
is
agricultural
land
(Table
1)
consisting
of
intensive
row
crop
cultivation
and
highly
erodible,
poorly-drained
soils.
As
a
result
of
agricultural
land
use
covering
∼50%
of
the
total
Lake
Erie
basin,
Lake
Erie
receives
more
sediment
than
any
other
of
the
Great
Lakes
by
almost
threefold
(Baker,
1993).
Since
the
1960s,
average
fertilizer
application
has
steadily
increased
in
the
state
of
Ohio
(NASS,
2010).
Most
crops
may
only
utilize
one-third
to
one-half
of
fertilizer
applied
while
the
remainder
of
fertilizer
has
the
potential
to
runoff
during
storm
events
as
particulate
phosphorus
bound
to
sediment
particles
or
as
dissolved
phosphorus
(Tilman
et
al.,
2001).
Prior
to
1970,
reduction
of
sediment,
nutrient
inputs
and
remediation
efforts
were
practically
non-existent
(Mortimer,
1987
).
In
1972,
the
United
States
and
Canada
signed
the
Great
Lakes
Water
Quality
Agreement
(GLWQA),
which
imple-
mented
strategies
for
reducing
phosphorus
loading
to
the
Great
Lakes
(IJC,
1978).
Subsequently,
better
management
practices—including
banning
phosphates
in
soaps/detergents
and
improvements
to
wastewater
treatment—significantly
reduced
phosphorous
loading
from
point
sources
during
the
1970s
and
1980s
(Richards
et
al.,
2008).
This
initial
approach
Please
cite
this
article
in
press
as:
Cousino,
L.K.,
et
al.,
Modeling
the
effects
of
climate
change
on
water,
sediment,
and
nutri-
ent
yields
from
the
Maumee
River
watershed.
J.
Hydrol.:
Reg.
Stud.
(2015),
http://dx.doi.org/10.1016/j.ejrh.2015.06.017
ARTICLE IN PRESS
G Model
EJRH-105;
No.
of
Pages
14
L.K.
Cousino
et
al.
/
Journal
of
Hydrology:
Regional
Studies
xxx
(2015)
xxx–xxx
3
Table
1
Land
use
classification
of
the
Maumee
River
watershed
based
on
the
2001
National
Land
Cover
Dataset
(NLCD).
Agricultural
and
urban
land
make
up
most
of
the
watershed,
placing
high
anthropogenic
stress
on
the
basin.
Landuse
Area
(thousands
of
ha)
%
of
Watershed
Water
13.6
0.8
Residential/Industrial
191.9
11.3
Forested
Upland
114.5
6.7
Rangeland
19.5 1.2
Pasture/Hay
90.5 5.3
Agricultural
Row
Crops
1226.4
72.3
Wetlands
38.3
2.3
of
focusing
on
reducing
point
source
pollution
was
not
adequate
to
meet
the
standards
set
by
the
GLWQA.
Remediation
efforts
then
shifted
to
include
reducing
nonpoint
sources
of
pollution
through
reduction
of
sediment
and
nutrient
runoff
from
agricultural
land
(IJC,
1978).
In
1987,
the
EPA
designated
775
mi
2
of
the
lower
Maumee
River
watershed
as
an
Area
of
Concern
(AOC),
indicating
that
beneficial
uses
of
the
waterway
have
been
impaired
by
anthropogenic
influence.
The
primary
reason
for
the
designation
was
agricultural
nonpoint
pollution,
and
a
subsequent
remedial
action
plan
(RAP)
was
developed
(
USEPA,
2013).
Conservation
tillage
programs
were
introduced
in
the
Western
Basin
watersheds
during
the
1980s
to
reduce
sediment
and
nutrient
runoff
from
agricultural
fields
and
address
the
problem
of
non-point
source
eutrophication.
Conventional
tillage
practices
employ
a
moldboard
plow
to
completely
mix
the
top
soil
layers
to
a
depth
of
12–18
inches.
This
practice
leaves
soil
exposed
to
wind
and
water
erosion,
increasing
soil
losses
and
sediment
loads
delivered
to
Lake
Erie.
By
1993,
more
than
50%
of
the
agricultural
land
in
the
Maumee
River
watershed
employed
some
form
of
conservation
tillage,
and
that
percentage
has
remained
about
50–60%
through
the
present
day
(Forster
and
Rausch,
2002;
Myers
and
Metzker,
2000).
Areas
producing
the
highest
sediment
loads
in
the
basin
also
have
some
of
the
lowest
rates
of
conservation
treatment
(Myers
and
Metzker,
2000
).
Lake
Erie
water
quality
initially
improved
in
response
to
remediation
efforts:
cyanobacterial
biomass
decreased
during
the
1980s
and
early
1990s.
However,
algal
blooms
have
again
been
significant
every
summer
since
1995
(Wynne
et
al.,
2010
).
Algal
blooms
during
the
years
of
2003,
2006,
and
2011
were
some
of
the
worst
on
record,
thought
to
be
caused
by
long
term
agricultural
land
use
changes
coupled
with
extreme
meteorological
conditions.
These
weather
factors
are
thought
to
be
consistent
with
expected
future
conditions
due
to
climate
change
(Michalak
et
al.,
2013).
1.3.
Historical
data
trends
A
USGS
gauging
station
located
in
Waterville,
Ohio
has
maintained
a
continuous
record
of
flow
rate
in
the
Maumee
River
from
1921
to
the
present
with
only
a
4-year
gap
from
1935
to
1939.
In
1975,
the
station
measurements
were
expanded
by
the
Heidelberg
Water
Quality
Monitoring
Laboratory
to
include
total
suspended
solids,
nitrate,
and
phosphate.
Almost
96%
of
the
Maumee
watershed
by
area
is
upstream
from
the
Waterville
station,
allowing
accurate
assessment
of
overall
trends
of
nutrient
and
sediment
runoff
within
the
entire
watershed
(Myers
and
Metzker,
2000).
Using
data
obtained
from
the
Waterville,
Ohio
gauging
station,
the
record
from
1975
to
2014
was
used
to
determine
trends
in
total
suspended
solids
(TSS)
and
total
phosphate
(TP)
records.
Overall,
average
daily
sediment
loads
have
decreased
by
43%
from
1975
to
2014
(as
seen
in
Fig.
3).
This
downward
trend
in
TSS
has
been
attributed
to
the
implementation
of
conservation
tillage
practices
in
the
basin
(Myers
and
Metzker,
2000).
However,
the
phosphate
to
sediment
ratio
increased
during
the
same
time
period.
The
TP/TSS
ratio
for
1995–2014
was
25%
higher
than
that
of
1975–1994,
and
the
ratio
for
2005–2014
was
34%
higher
than
1975–1994.
Despite
decreasing
sediment
loads
and
a
shifting
annual
ratio,
daily
TP
and
TSS
loads
maintained
a
strong
average
annual
correlation
(R
2
=
0.91)
from
1975
to
2014
(Fig.
2).
Although
reduction
of
total
phosphorus
has
been
the
primary
focus
of
remediation
efforts
following
the
GLWQA,
attention
has
recently
shifted
to
dissolved
reactive
phosphorus
(DRP)
because
of
its
high
bioavailability
to
algae
and
cyanobacteria
(
Depinto
et
al.,
1986;
Hayhoe
et
al.,
2010;
Richards
et
al.,
2010).
DRP
loads
to
Lake
Erie
have
increased
since
1995
(Daloglu
et
al.,
2012),
and
the
ratio
of
DRP
to
TP
doubled
from
the
1990s
to
the
2000s
(Scavia
et
al.,
2014).
The
exact
cause
for
this
trend
is
not
known;
however,
some
models
have
shown
that
conservation
tillage
practices
increase
DRP
loads
due
to
increased
soil
stratification
and
accumulation
of
fertilizer
P
at
the
surface
(Michalak
et
al.,
2013).
Additionally,
an
increase
in
the
frequency
of
extreme
storm
events
and
changes
in
fertilizer
application
timing
and
rate
may
also
be
influencing
Lake
Erie
DRP
loads
(
Daloglu
et
al.,
2012).
The
recent
increase
in
DRP
loads
to
Lake
Erie
could
explain
the
recent
surge
in
algal
activity
(Michalak
et
al.,
2013).
1.4.
Climate
change
in
the
Great
Lakes
Region
The
Great
Lakes
have
already
demonstrated
numerous
trends
consistent
with
a
warming
climate,
including
increases
in
annual
temperatures,
decreased
snow
and
ice
cover,
and
a
higher
frequency
of
intense
rainfall
events.
Downscaling
of
global
climate
models
(GCMs)
shows
that
the
Great
Lakes
region
could
experience
an
increase
in
temperature
of
2–6
◦
C
by
Please
cite
this
article
in
press
as:
Cousino,
L.K.,
et
al.,
Modeling
the
effects
of
climate
change
on
water,
sediment,
and
nutri-
ent
yields
from
the
Maumee
River
watershed.
J.
Hydrol.:
Reg.
Stud.
(2015),
http://dx.doi.org/10.1016/j.ejrh.2015.06.017
ARTICLE IN PRESS
G Model
EJRH-105;
No.
of
Pages
14
4
L.K.
Cousino
et
al.
/
Journal
of
Hydrology:
Regional
Studies
xxx
(2015)
xxx–xxx
Fig.
2.
Land
use
map
of
the
Maumee
River
watershed.
Agricultural
row
crops
cover
over
72%
of
the
basin,
contributing
to
high
sediment
and
nutrient
loading
in
the
WB.
The
black
circles
show
the
15
weather
stations
used
for
calibration
of
watershed
parameters
and
calculation
of
historical
weather
averages.
Fig.
3.
Average
daily
total
suspended
solids
(TSS)
load
and
the
total
phosphate
(TP)
to
TSS
ratio
within
the
Maumee
River
from
1975
to
2014.
Daily
TSS
and
TP
loads
are
highly
correlated
with
an
R
2
value
(listed
on
gray
bars)
between
0.86
and
0.94
from
1975
to
2014,
which
allows
sediment
to
be
used
a
proxy
for
phosphate.
Daily
TSS
loads
(black)
averaged
annually
have
decreased
from
∼7000
MT/day
to
∼4000
MT/day,
showing
the
effectiveness
of
reducing
non-point
source
sediment
loading.
However,
the
ratio
of
TP/TSS
has
increased
during
the
same
time
period
(gray),
indicating
less
success
in
reducing
TP
loads.
the
end
of
the
century,
and
winter
and
spring
precipitation
could
increase
by
20–30%
due
to
more
frequent
large
storm
events
(Delworth
et
al.,
2006;
Pope
et
al.,
2000;
Washington
et
al.,
2000).
Summer
precipitation
is
expected
to
remain
about
the
same
or
decrease
slightly.
Annual
precipitation
is
only
projected
to
increase
slightly
(Hayhoe
et
al.,
2010).
However,
the
frequency
of
intense
storm
events
over
the
Midwest
increased
during
the
past
several
decades
and
is
projected
to
continue
increasing
in
the
future
(Groisman
et
al.,
2012).
Stumpf
et
al.
(2012)
studied
algal
blooms
in
Lake
Erie’s
Western
Basin
from
2002
to
2011.
Although
blooms
peaked
in
August
or
later,
they
were
correlated
to
water
and
TP
yields
from
the
Maumee
River
only
for
March
to
June.
Thus,
increases
in
Please
cite
this
article
in
press
as:
Cousino,
L.K.,
et
al.,
Modeling
the
effects
of
climate
change
on
water,
sediment,
and
nutri-
ent
yields
from
the
Maumee
River
watershed.
J.
Hydrol.:
Reg.
Stud.
(2015),
http://dx.doi.org/10.1016/j.ejrh.2015.06.017
ARTICLE IN PRESS
G Model
EJRH-105;
No.
of
Pages
14
L.K.
Cousino
et
al.
/
Journal
of
Hydrology:
Regional
Studies
xxx
(2015)
xxx–xxx
5
spring
precipitation
intensity
could
significantly
increase
bloom
intensity.
DeMarchi
et
al.
(2013)
modeled
a
10–30%
increase
in
the
Maumee
River’s
total
suspended
sediment
load
by
the
end
of
the
century
based
on
data
from
two
GCMs
of
CMIP3.
Due
to
the
strong
correlation
between
TSS
and
TP
loads
in
the
Maumee
River
watershed
(Richards
et
al.,
2009),
total
phosphorus
would
likely
increase
under
these
scenarios
as
well.
1.5.
Previous
studies
The
Soil
and
Water
Assessment
Tool
(SWAT)
has
been
successfully
used
to
model
the
effects
of
climate
change
and
land
management
practices
on
watershed
yields
in
past
studies.
Stone
et
al.
(2001)
used
SWAT
to
predict
that
the
overall
water
yield
of
the
Missouri
River
basin
will
decrease
by
10–20%
during
spring
and
summer
months
but
increase
during
the
fall
and
winter
months
in
response
to
doubling
atmospheric
CO
2
concentrations.
Jha
et
al.
(2006)
used
SWAT
to
predict
a
36%
increase
in
the
water
yield
of
the
Upper
Mississippi
River
basin
after
doubling
CO
2
concentrations.
Changes
in
flow
over
the
central
plains
of
the
U.S.
ranged
from
−6%
to
+51%
in
response
to
temperature
and
precipitation
changes
based
on
data
from
six
coupled
atmosphere-ocean
general
circulation
models
(AOGCMs)
(Carter
et
al.,
1994;
Giorgi
et
al.,
1998).
SWAT
models
of
agricultural
and
rangeland
watersheds
in
Kansas
and
Nebraska
predicted
increased
sediment,
TN,
and
TP
yields
under
future
climate
change
scenarios.
The
model
also
showed
that
the
effectiveness
of
most
Best
Management
Practices
(BMPs)
is
highly
sensitive
to
climate
change
(Woznicki
and
Nejadhashemi,
2012).
Michalak
et
al.
(2013)
used
a
calibrated
SWAT
model
of
the
Maumee
River
watershed
to
show
that
DRP
yields
are
responsive
to
precipitation
intensity,
timing
of
fertilizer
application,
and
tillage
practices.
Precipitation
intensity
had
the
greatest
effect
on
DRP
yields.
Additionally,
the
study
evaluated
CMIP5
precipitation
predictions
over
the
Western
Basin
and
found
that
events
over
20
mm
could
become
50%
more
frequent,
while
storm
events
over
30
mm
could
become
twice
as
frequent.
Bosch
et
al.
(2014)
used
SWAT
models
of
four
Lake
Erie
watersheds
using
scaled
historical
precipitation
and
temperature
records,
which
showed
that
high
implementation
rates
of
BMPs
could
offset
a
significant
amount
of
the
expected
increases
in
sediment
and
nutrient
loads
due
to
climate
change.
The
SWAT
model
of
the
Maumee
River
watershed
projected
increases
of
5–11%
for
flow,
2–32%
for
sediment,
1–5%
for
DRP,
and
0–7%
for
TP
over
the
next
century.
Projections
were
done
using
“Moderate”
and
“Pronounced”
climate
change
scenarios
based
on
output
from
three
AOGCMs.
However,
previous
studies
such
as
these
which
use
scaled
historical
records
to
drive
future
climate
change
fail
to
allow
for
the
full
potential
variability
in
future
climate
patterns.
These
studies
increase
or
decrease
the
magnitude
of
historical
weather
patterns
by
a
predetermined
factor,
which
does
not
allow
them
to
account
for
any
changes
in
the
intensity
or
frequency
of
precipitation
events.
Previous
models
have
failed
to
allow
for
predicted
changes
in
the
characteristics
of
precipitation
events
in
the
central
United
States
(Hayhoe
et
al.,
2010).
2.
Methods
2.1.
Introduction
The
present
study
incorporates
the
most
recent
climate
change
predictions
from
the
World
Climate
Research
Programme’s
(WCRP’s)
Coupled
Model
Intercomparison
Project
Phase
5
(CMIP5)
into
a
calibrated
SWAT
model
of
the
Maumee
River
watershed.
We
use
daily
data
from
the
WRCP
to
produce
multiple
iterations
of
climate
change
scenarios
in
an
attempt
to
capture
the
magnitude
and
frequency
of
future
precipitation
and
temperature
changes
and
to
predict
future
water
and
sediment
loads
from
the
basin.
We
also
examine
the
effect
of
varying
tillage
practices
on
watershed
yields
under
historical
and
future
climate
scenarios.
2.2.
SWAT
The
Soil
and
Water
Assessment
Tool
(SWAT)
is
a
continuous-time,
semi-distributed,
and
physically-based
watershed-
scale
model
(Arnold
et
al.,
2012).
It
integrates
weather,
surface
and
groundwater
hydrology,
soil
properties,
plant
growth,
and
land
management
practices
to
model
processes
within
a
watershed
(Arnold
et
al.,
1998).
SWAT
is
used
to
predict
the
impacts
of
land
management
changes
on
water,
sediment,
and
nutrient
yields
within
a
basin.
In
SWAT,
watersheds
are
divided
into
subbasins
based
on
interior
outlet
points
along
the
stream
network.
Each
subbasin
is
then
divided
into
hydrologic
response
units
(HRUs),
which
are
areas
with
homogeneous
soil
type,
land
use,
slope,
and
management
practices.
Yields
are
calculated
for
each
HRU
and
summed
to
determine
subbasin
outputs.
HRUs
are
not
spatially
defined
within
subbasins;
they
represent
percentages
of
total
subbasin
area.
Therefore,
SWAT
includes
both
spatially-distributed
parameterization
at
the
subbasin
scale
and
lumped
parameterization
at
the
HRU
scale
(Gassman
et
al.,
2007).
2.3.
Model
data
sources
and
setup
The
model
utilized
a
30
m
digital
elevation
model
(DEM)
to
delineate
the
surface
drainage
of
the
basin
(Farr
et
al.,
2007).
The
model
also
incorporated
soil
types
obtained
from
the
USDA
NRCS
Soil
Survey
Geographic
(SSURGO)
database
and
land
use
from
the
National
Land
Cover
Database
(NLCD)
from
2001
(Soil
Survey
Staff,
2013;
Homer
et
al.,
2004).
Tillage
practices
were
estimated
at
a
basin-wide
scale
using
data
from
the
NRCS
Myers’
USGS
report
(Myers
and
Metzker,
2000).
Data
for
Please
cite
this
article
in
press
as:
Cousino,
L.K.,
et
al.,
Modeling
the
effects
of
climate
change
on
water,
sediment,
and
nutri-
ent
yields
from
the
Maumee
River
watershed.
J.
Hydrol.:
Reg.
Stud.
(2015),
http://dx.doi.org/10.1016/j.ejrh.2015.06.017
ARTICLE IN PRESS
G Model
EJRH-105;
No.
of
Pages
14
6
L.K.
Cousino
et
al.
/
Journal
of
Hydrology:
Regional
Studies
xxx
(2015)
xxx–xxx
tillage
practices
at
the
sub-county
level
was
not
available
because
farmers
report
tillage
practices
voluntarily
on
an
individual
basis,
or
surveys
are
carried
out
based
on
random
sampling
of
farmers
(Myers
and
Metzker,
2000).
Weather
data
was
obtained
from
NOAA’s
National
Climate
Data
Center
(NCDC)
(NOAA,
2015).
Measured
data
from
15
weather
stations
for
1980
to
2009
was
used
for
daily
precipitation
and
temperature.
Missing
data
points
at
each
weather
station
were
filled
in
using
data
from
the
nearest
weather
station.
Wind
speed,
solar
radiation,
and
relative
humidity
data
were
obtained
from
a
gridded
dataset
(Climate
Forecast
System
Reanalysis)
produced
by
the
National
Centers
for
Environ-
mental
Prediction
(NCEP)
(Kalnay
et
al.,
1996).
This
dataset
has
been
successfully
employed
for
watershed
modeling
across
a
variety
of
hydrologic
settings
(Fuka
et
al.,
2013).
SWAT
also
includes
the
WXGEN
weather
generator
model
(Neitsch
et
al.,
2011),
which
stochastically
generates
daily
weather
values
based
on
monthly
averages
for
a
given
location.
For
each
future
scenario,
20
years
of
daily
precipitation
and
temperature
data
(2046–2065
or
2080–2099)
were
obtained
from
the
World
Climate
Research
Programme’s
Coupled
Modeled
Intercomparison
Project
phase
5
(CMIP5)
Climate
Projections
archive
(Reclamation,
2013).
For
each
of
the
15
weather
stations
used
during
calibration,
the
nearest
point
in
the
CMIP5
grid
was
selected
to
generate
future
weather
data.
Daily
precipitation
and
temperature
values
from
the
CMIP5
dataset
were
used
to
calculate
monthly
weather
averages
for
each
station,
and
then
the
averages
were
inputted
into
the
SWAT
weather
generator
to
produce
100
unique
iterations
of
the
climate
data
for
each
scenario.
Data
were
acquired
from
two
AOGCMs:
the
National
Center
for
Atmospheric
Research’s
Community
Climate
System
Model
4.0
(CCSM4.0)
and
the
Atmosphere
and
Ocean
Research
Institute
(The
University
of
Tokyo),
National
Institute
for
Environmental
Studies,
and
Japan
Agency
for
Marine-Earth
Science
and
Technology’s
Model
for
Interdisciplinary
Research
on
Climate
(MIROC5).
Data
were
also
obtained
from
two
earth
system
models
(ESMs):
the
US
National
Atmospheric
and
Ocean
Administration’s
Geophysical
Fluid
Dynamics
Laboratory
(GFDL)
ESM2M
and
the
Norwe-
gian
Climate
Centre’s
Norwegian
Earth
System
Model
(NORES1-M)
(see
Table
A.1).
Four
climate
scenarios—Representative
Concentration
Pathways
(RCP)
2.6,
4.5,
6.0,
and
8.5—and
two
time
periods—2046–2065
and
2080–2099—were
included
in
the
data
(Taylor
et
al.,
2012).
Using
the
daily
precipitation
data
from
the
CMIP5
dataset,
the
mean
daily
rainfall,
standard
deviation
of
daily
rainfall,
and
the
probably
of
a
wet
day
(defined
as
a
day
with
more
than
0.1
mm
precipitation)
were
calculated
for
each
month
(
Neitsch
et
al.,
2011).
To
determine
if
precipitation
occurs
on
a
given
day,
SWAT
implements
a
first-order
Markov
chain
model
(Nicks,
1974).
SWAT
generates
a
random
number
between
0.0
and
1.0
and
compares
it
to
the
monthly
probability
for
rainfall.
Numbers
equal
to
or
less
than
the
probability
are
defined
as
wet
days.
Precipitation
amounts
are
then
calculated
using
a
skewed
distribution
(Neitsch
et
al.,
2011).
Using
the
daily
maximum
and
minimum
temperatures
from
the
CMIP5
dataset,
average
daily
maximum
and
minimum
temperatures
and
daily
standard
deviation
for
maximum
and
minimum
temperatures
were
calculated
for
each
month.
Using
these
values,
the
SWAT
weather
generator
produced
multiple
iterations
of
daily
maximum
and
minimum
temperatures
using
a
weakly
stationary
generating
process
(Matalas,
1967).
No
changes
were
made
to
the
historical
relative
humidity,
solar
radiation,
and
wind
speed
datasets
when
modeling
climate
change
scenarios.
By
inputting
climate
change
data
directly
into
the
stochastic
weather
generator
and
running
100
iterations
of
each
sce-
nario,
we
attempted
to
accurately
model
predicted
variations
in
future
temperatures
and
both
the
magnitude
and
frequency
of
runoff
events.
Previous
studies
(Jha
et
al.,
2006;
Stone
et
al.,
2001;
Bosch
et
al.,
2014)
have
used
historical
weather
data
to
drive
their
models
and
then
changed
SWAT
variables
to
induce
climate
change.
This
method
alters
daily
historical
pre-
cipitation
and
temperature
values
by
a
predetermined
percentage
or
value.
Thus,
this
technique
only
changes
precipitation
amounts
on
days
that
recorded
precipitation
and
does
not
allow
for
potential
changes
in
the
frequency
or
intensity
of
runoff
events
or
the
severity
of
daily
temperature
variations.
These
models
may
fail
to
replicate
predicted
future
precipi-
tation
characteristics,
such
as
decreased
frequency
and
increased
intensity
of
precipitation
events
(Trenberth
et
al.,
2003).
Additionally,
climate
models
consistently
predict
increases
in
potential
evapotranspiration
(PET),
increasing
the
chance
for
drought
conditions
between
precipitation
events
(Trenberth,
2011).
However,
previous
SWAT
models
that
use
a
historical
frequency
of
precipitation
may
not
allow
for
the
effects
of
PET
changes
to
be
fully
realized.
Precipitation
characteristics
are
just
as
important
as
precipitation
amounts
in
predicting
future
soil
moisture
and
stream
flow
(Trenberth,
2011).
Our
model
allows
for
changes
in
both
precipitation
amounts
and
frequency,
providing
a
more
robust
representation
of
possible
future
conditions.
Model
setup
was
done
using
automatic
watershed
delineation
within
the
ArcGIS
interface
for
SWAT
(ArcSWAT).
Streams
were
delineated
based
on
a
minimum
drainage
area
of
10,000
ha
due
to
the
substantial
size
of
the
watershed.
A
total
of
97
subbasins
and
665
hydrologic
response
units
(HRUs)
were
created.
HRUs
were
created
based
on
land
use
and
soil
types
that
made
up
at
least
10%
of
a
given
subbasin’s
area.
Since
the
watershed
occupies
a
former
glacial
lake
bed,
the
slope
is
extremely
low
(1.3
ft/mi)
(Myers
and
Metzker,
2000).
Therefore,
HRUs
were
not
differentiated
based
on
slope.
Tile
drainage
was
not
explicitly
modeled;
however,
it
was
indirectly
accounted
for
during
parameterization.
2.4.
Model
Calibration
and
validation
2.4.1.
Calibration
We
calibrated
the
SWAT
model
from
1995
to
1999
using
monthly
flow
and
sediment
data
obtained
from
the
USGS
gauging
station
at
Waterville,
Ohio.
Daily
observed
weather
data
were
used
to
drive
the
model,
and
a
five
year
model
“spin-
up”
from
1990
to
1994
was
used
to
stabilize
base
flow
conditions
within
the
model.
We
used
the
Sequential
Uncertainty
Please
cite
this
article
in
press
as:
Cousino,
L.K.,
et
al.,
Modeling
the
effects
of
climate
change
on
water,
sediment,
and
nutri-
ent
yields
from
the
Maumee
River
watershed.
J.
Hydrol.:
Reg.
Stud.
(2015),
http://dx.doi.org/10.1016/j.ejrh.2015.06.017
ARTICLE IN PRESS
G Model
EJRH-105;
No.
of
Pages
14
L.K.
Cousino
et
al.
/
Journal
of
Hydrology:
Regional
Studies
xxx
(2015)
xxx–xxx
7
Table
2
Calibration
and
validation
results
for
the
AcrSWAT
model
of
the
Maumee
River
watershed.
Performance
ratings
are
based
on
the
standards
suggested
by
Moriasi
et
al.
(2007).
Scenario
NSE
R
2
Perfomance
rating
Flow
calibration:
1995–1999
0.83
0.94
Very
good
Flow
validation:
2000–2002
0.87
0.92
Very
good
Sediment
concentration
calibration:
1995–1999 0.63 0.67 Satisfactory
Sediment
concentration
validation:
2000–2002 0.72
0.77
Good
Fitting
(SUFI2)
method,
a
semi-automated
calibration
approach
that
allows
the
user
to
adjust
watershed
parameters
between
autocalibration
runs
(Arnold
et
al.,
2012).
Flow
was
calibrated
prior
to
sediment
concentration.
Following
calibration,
we
validated
the
model
for
2000–2002.
Model
performance
was
evaluated
based
on
standards
set
by
Moriasi
et
al.
(2007)
using
the
Nash-Sutcliffe
Efficiency
(NSE)
(Nash
and
Sutcliffe,
1970)
and
the
coefficient
of
determination
(R
2
).
For
a
monthly
time
step,
Moriasi
et
al.
(2007)
suggested
that
a
performance
rating
of
“satisfactory”
be
given
for
a
NSE
greater
than
0.50.
They
also
recommended
a
rating
of
“good”
for
0.65–0.75
and
“very
good”
for
0.75–1.00.
Calibration
for
1995–1999
at
a
monthly
time-step
produced
a
“very
good”
Nash
Sutcliffe
Efficiency
(NSE)
of
0.83
(Table
2).
The
R
2
value
was
0.94.
Flow
validation
was
done
for
2000–2002
and
resulted
in
an
NSE
of
0.87
and
an
R
2
value
of
0.92.
Monthly
sediment
calibration
resulted
in
a
“satisfactory”
NSE
of
0.63
and
an
R
2
value
of
0.67.
Sediment
validation
for
2000–2002
resulted
in
values
of
0.72
and
0.77
for
the
NSE
and
R
2
values,
respectively.
2.5.
Model
scenarios
After
performing
calibration
and
validation,
we
tested
different
drivers
(tillage
practices
and
weather)
to
see
their
effect
on
sediment
and
nutrient
loading
within
the
watershed.
The
baseline
scenario
(1985–2004)
was
driven
using
historical
weather
averages,
and
conservation
tillage
was
applied
to
half
of
the
subbasins
in
the
watershed
based
on
the
current
rate
of
con-
servation
treatment
in
the
basin
(Myers
and
Metzker,
2000).
Model
runs
were
also
produced
for
the
years
1985–2004
using
historical
climate
trends
with
100%
conservation
tillage
application
and
100%
standard
tillage
application
(no
conservation
treatment).
The
conservation
scenario
was
simulated
using
the
procedure
outlined
in
Arabi
et
al.
(2008).
For
all
agricultural
HRUs,
curve
numbers
were
reduced,
Manning’s
overland
roughness
coefficient
(OV-N)
was
increased,
and
tillage
practices
were
excluded
from
the
management
input
file.
The
three
tillage
scenarios
(100%
standard
tillage,
50%
conservation
tillage,
and
100%
conservation
tillage)
were
extended
into
the
future
using
CMIP5
data
as
drivers.
Each
scenario
was
propagated
for
two
time
periods:
2046–2065
and
2080–2099.
Using
current
watershed
conditions,
four
climate
change
scenarios
were
applied
(RCP
2.6,
4.5,
6.0,
and
8.5).
For
the
100%
standard
tillage
and
100%
conservation
tillage
scenarios,
a
moderate
climate
change
scenario
(RCP
4.5)
and
an
extreme
scenario
(RCP
8.5)
were
implemented.
CMIP5
models
were
chosen
based
on
research
by
Kumar
et
al.
(2013)
that
evaluated
the
performance
of
19
CMIP5
models
in
replicating
historical
data
trends.
Of
the
19
models
tested,
CCSM4.0
had
the
highest
correlation
with
historical
precipitation
trends,
while
GFDL
and
MIROC
had
the
highest
correlation
with
historical
temperature
trends
(Kumar
et
al.,
2013).
CMIP5
identifies
four
mitigation
scenarios
known
as
Representative
Concentration
Pathways
(RCPs).
Unlike
the
CMIP3
scenarios,
CMIP5
assumes
policy
changes
will
be
implemented
to
reach
specific
emission
targets.
The
numbers
in
the
CMIP5
scenarios
represent
the