Page
1
of
61
July
20,
2004
MEMORANDUM
SUBJECT:
Transmittal
Of
Minutes
Of
The
FIFRA
Scientific
Advisory
Panel
Meeting
Held
March
30­
31,
2004
Addressing
A
Set
Of
Scientific
Issues
Being
Considered
By
The
Environmental
Protection
Agency
Regarding
Refined
(
Level
II)
Terrestrial
And
Aquatic
Models
Probabilistic
Ecological
Assessments
For
Pesticides:
Terrestrial
TO:
James
J.
Jones,
Director
Office
of
Pesticide
Programs
FROM:
Paul
I.
Lewis,
Designated
Federal
Official
FIFRA
Scientific
Advisory
Panel
Office
of
Science
Coordination
and
Policy
THRU:
Larry
C.
Dorsey,
Executive
Secretary
FIFRA
Scientific
Advisory
Panel
Office
of
Science
Coordination
and
Policy
Joseph
J.
Merenda,
Jr.,
Director
Office
of
Science
Coordination
and
Policy
Please
find
attached
the
minutes
of
the
FIFRA
Scientific
Advisory
Panel
open
meeting
held
in
Arlington,
Virginia
from
March
30­
31,
2004.
These
meeting
minutes
address
a
set
of
scientific
issues
being
considered
by
the
Environmental
Protection
Agency
regarding
refined
(
Level
II)
terrestrial
and
aquatic
models
probabilistic
ecological
assessments
for
pesticides:
terrestrial
Attachment
Page
2
of
61
cc:

Susan
Hazen
Adam
Sharp
Anne
Lindsay
Janet
Andersen
Debbie
Edwards
Steven
Bradbury
William
Diamond
Arnold
Layne
Tina
Levine
Lois
Rossi
Frank
Sanders
Margaret
Stasikowski
William
Jordan
Douglas
Parsons
Dayton
Eckerson
David
Deegan
Vanessa
Vu
(
SAB)
Ingrid
Sunzenauer
Edward
Fite
Timothy
Barry
Dirk
Young
Edward
Odenkirchen
Christine
Hartless
OPP
Docket
FIFRA
Scientific
Advisory
Panel
Members
Steven
G.
Heeringa,
Ph.
D.
Stuart
Handwerger,
M.
D.
Gary
E.
Isom,
Ph.
D.
Louis
Best,
Ph.
D.
Xuefeng
Chu,
Ph.
D.
Larry
Clark,
Ph.
D.
George
Cobb,
Ph.
D.
Paul
W.
Eslinger,
Ph.
D.
Michael
Fry,
Ph.
D.
Christian
Grue,
Ph.
D.
Dennis
Laskowski,
Ph.
D.
Peter
Macdonald,
D.
Phil.
Charles
Menzie,
Ph.
D.
Dwayne
Moore,
Ph.
D.
Raymond
O'Connor,
D.
Phil.
Mitchell
Small,
Ph.
D.
Tammo
Steenhuis,
Ph.
D
Page
3
of
61
SAP
Report
No.
2004­
03
MEETING
MINUTES
FIFRA
Scientific
Advisory
Panel
Meeting,
March
30­
31,
2004,
held
at
the
Sheraton
Crystal
City
Hotel,
Arlington,
Virginia
A
Set
of
Scientific
Issues
Being
Considered
by
the
Environmental
Protection
Agency
Regarding:

Refined
(
Level
II)
Terrestrial
And
Aquatic
Models
Probabilistic
Ecological
Assessments
For
Pesticides:
Terrestrial
Page
4
of
61
NOTICE
These
meeting
minutes
have
been
written
as
part
of
the
activities
of
the
Federal
Insecticide,
Fungicide,
and
Rodenticide
Act
(
FIFRA),
Scientific
Advisory
Panel
(
SAP).
These
minutes
have
not
been
reviewed
for
approval
by
the
United
States
Environmental
Protection
Agency
(
Agency)
and,
hence,
their
contents
do
not
necessarily
represent
the
views
and
policies
of
the
Agency,
nor
of
other
agencies
in
the
Executive
Branch
of
the
Federal
government,
nor
does
mention
of
trade
names
or
commercial
products
constitute
a
recommendation
for
use.

The
FIFRA
SAP
was
established
under
the
provisions
of
FIFRA,
as
amended
by
the
Food
Quality
Protection
Act
(
FQPA)
of
1996,
to
provide
advice,
information,
and
recommendations
to
the
Agency
Administrator
on
pesticides
and
pesticide­
related
issues
regarding
the
impact
of
regulatory
actions
on
health
and
the
environment.
The
Panel
serves
as
the
primary
scientific
peer
review
mechanism
of
the
EPA,
Office
of
Pesticide
Programs
(
OPP)
and
is
structured
to
provide
balanced
expert
assessment
of
pesticide
and
pesticide­
related
matters
facing
the
Agency.
Food
Quality
Protection
Act
Science
Review
Board
members
serve
the
FIFRA
SAP
on
an
ad
hoc
basis
to
assist
in
reviews
conducted
by
the
FIFRA
SAP.
Further
information
about
FIFRA
SAP
meeting
minutes
and
activities
can
be
obtained
from
its
website
at
http://
www.
epa.
gov/
scipoly/
sap/,
the
OPP
Docket
at
(
703)
305­
5805
or
edocket
at
http://
docket.
epa.
gov/
edkpub/
index.
jsp
(
OPP­
2004­
0005).
Interested
persons
are
invited
to
contact
Paul
Lewis,
Designated
Federal
Official,
via
email
at
lewis.
paul@
epa.
gov.

In
preparing
these
meeting
minutes,
the
Panel
carefully
considered
all
information
provided
and
presented
by
the
Agency
presenters,
as
well
as
information
presented
by
public
commenters.
This
document
addresses
the
information
provided
and
presented
within
the
structure
of
the
charge
by
the
Agency.
Page
5
of
61
SAP
Report
No.
2004­
03
MEETING
MINUTES:
FIFRA
Scientific
Advisory
Panel
Meeting,
March
30­
31,
2004,
held
at
the
Sheraton
Crystal
City
Hotel,
Arlington,
Virginia
A
Set
of
Scientific
Issues
Being
Considered
by
the
Environmental
Protection
Agency
Regarding:

Refined
(
Level
II)
Terrestrial
And
Aquatic
Models
Probabilistic
Ecological
Assessments
For
Pesticides:
Terrestrial
Mr.
Paul
Lewis
Steven
Heeringa,
Ph.
D.
Designated
Federal
Official
FIFRA
SAP
Session
Chair
FIFRA
Scientific
Advisory
Panel
FIFRA
Scientific
Advisory
Panel
Date:
July
20,
2004
Date:
July
20,
2004
Page
6
of
61
Federal
Insecticide,
Fungicide,
and
Rodenticide
Act
Scientific
Advisory
Panel
Meeting
March
30­
31,
2004
Refined
(
Level
II)
Terrestrial
And
Aquatic
Models
Probabilistic
Ecological
Assessments
For
Pesticides:
Terrestrial
PARTICIPANTS
FIFRA
SAP
Session
Chair
Steven
Heeringa,
Ph.
D.
Research
Scientist
&
Director
for
Statistical
Design,
Institute
for
Social
Research,
University
of
Michigan,
PO
Box
1248,
Ann
Arbor,
MI
FIFRA
Scientific
Advisory
Panel
Members
Stuart
Handwerger,
M.
D.
Director,
Division
of
Endocrinology,
Cincinnati
Children's
Hospital
Medical
Center,
University
of
Cincinnati,
Cincinnati,
OH
Gary
E.
Isom,
Ph.
D.
Professor
of
Toxicology,
Purdue
University,
School
of
Pharmacy
&
Pharmacal
Sciences,
West
Lafayette,
IN
FQPA
Science
Review
Board
Members
Louis
Best,
Ph.
D.
Professor
of
Avian
Ecology,
Department
of
Natural
Resource
Ecology
and
Management,
Iowa
State
University,
Ames,
IA
Xuefeng
Chu,
Ph.
D.
Assistant
Professor,
Annis
Water
Resources
Institute,
Grand
Valley
State
University,
Muskegon,
MI
Larry
Clark,
Ph.
D.
Project
Manager,
National
Wildlife
Research
Center,
USDA­
APHIS­
WS,
Fort
Collins,
CO
George
Cobb,
Ph.
D.
Associate
Professor,
Environmental
Toxicology,
Texas
Tech
University,
Lubbock,
TX
Paul
W.
Eslinger,
Ph.
D.
Staff
Scientist,
Pacific
Northwest
National
Laboratory,
Richland,
WA
Michael
Fry,
Ph.
D.
Senior
Managing
Scientist,
Stratus
Consulting,
Inc.,
Boulder,
CO
Christian
Grue,
Ph.
D.
Associate
Professor
&
Leader,
Washington
Cooperative
Fish
and
Wildlife
Research
Unit,
University
of
Washington,
Fishery
Sciences,
Seattle,
WA
Page
7
of
61
Dennis
Laskowski,
Ph.
D.
4600
Hickory
Court,
Zionsville,
IN
Peter
Macdonald,
D.
Phil.
Professor
of
Mathematics
and
Statistics,
McMaster
University,
Hamilton,
Ontario,
Canada
Charles
Menzie,
Ph.
D.
Principal,
Menzie­
Cura
&
Associates,
Inc.,
Chelmsford,
MA
Dwayne
Moore,
Ph.
D.
Vice
President,
Cantox
Environmental,
Inc.,
Ottawa,
Ontario,
Canada
Raymond
O'Connor,
D.
Phil.
Professor,
Wildlife
Biology,
University
of
Maine,
Orono,
ME
Mitchell
Small,
Ph.
D.
H.
John
Heinz
III
Professor
of
Environmental
Engineering,
Department
of
Civil
and
Environmental
Engineering
and
Engineering
and
Public
Policy,
Carnegie
Mellon
University,
Pittsburgh,
PA
Tammo
Steenhuis,
Ph.
D.
Professor,
Agricultural
and
Biological
Engineering
Department,
Cornell
University,
Ithaca,
NY
PUBLIC
COMMENTERS
Oral
statements
were
made
by:
David
Fischer,
Ph.
D.
(
Bayer
CropScience)
on
behalf
of
CropLife
America
Nick
Poletika,
Ph.
D.
(
Dow
AgroSciences)
on
behalf
of
CropLife
America
No
written
statements
were
provided
INTRODUCTION
The
Federal
Insecticide,
Fungicide,
and
Rodenticide
Act
(
FIFRA),
Scientific
Advisory
Panel
(
SAP)
has
completed
its
review
of
the
set
of
scientific
issues
being
considered
by
the
Agency
pertaining
to
its
review
of
refined
(
Level
II)
terrestrial
models
for
probabilistic
ecological
assessment
of
pesticides.
Advance
notice
of
the
meeting
was
published
in
the
Federal
Register
on
February
20,
2004.
The
review
was
conducted
in
an
open
Panel
meeting
held
in
Arlington,
Virginia,
from
March
30­
31,
2004.
The
meeting
was
chaired
by
Steven
Heeringa,
Ph.
D.
Mr.
Paul
Lewis
served
as
the
Designated
Federal
Official.
Mr.
Joseph
J.
Merenda,
Jr.
(
Director,
Office
of
Science
Coordination
and
Policy,
EPA),
Mr.
Jim
Jones
(
Director,
Office
of
Pesticide
Programs,
EPA)
and
Steven
Bradbury,
Ph.
D.
(
Director,
Environmental
Fate
and
Effects
Division,
Office
of
Pesticide
Programs,
EPA)
provided
opening
remarks
at
the
meeting.
Ingrid
Sunzenauer,
M.
S.
(
Office
of
Pesticide
Programs,
EPA)
highlighted
the
goals
and
objectives
of
the
session.
Edward
Fite,
M.
S.
(
Office
of
Pesticide
Programs,
EPA)
and
Timothy
Barry,
Sc.
D.
(
Office
of
Policy,
Economics,
and
Innovation,
EPA)
discussed
the
model
architecture
of
the
Level
II
terrestrial
integration
model.
Edward
Fite,
M.
S.
(
Office
of
Pesticide
Programs,
EPA)
summarized
the
selection
of
generic
species.
Timothy
Barry,
Sc.
D.
(
Office
of
Policy,
Economics,
and
Page
8
of
61
Innovation,
EPA)
reviewed
the
bimodal
feeding
pattern
and
Markov
chain
model.
Dirk
F.
Young,
Ph.
D.
(
Office
of
Pesticide
Programs,
EPA)
discussed
the
puddle
model.
Edward
Odenkirchen,
Ph.
D.
(
Office
of
Pesticide
Programs,
EPA)
and
Christine
Hartless,
Ph.
D.
(
Office
of
Pesticide
Programs,
EPA)
presented
the
dermal
exposure
and
effects
model.
Edward
Odenkirchen,
Ph.
D.
(
Office
of
Pesticide
Programs,
EPA)
discussed
the
inhalation
exposure
and
effects
model.
Edward
Fite,
M.
S.
(
Office
of
Pesticide
Programs,
EPA)
presented
the
preliminary
model
testing
results/
next
steps
and
also
ended
the
session
by
presenting
an
introduction
of
the
questions
to
the
Panel.

CHARGE
1.
Guild
Parameters
Used
for
Defining
Generic
Species.
The
process
for
defining
generic
species
described
in
this
document
separated
species
into
guilds
based
on
three
parameters:
feeding
substrate,
nesting
substrate,
and
food
type.

a.)
Please
comment
on
the
representative
guilds
used
to
define
the
generic
organisms.
b.)
Are
there
any
additional
parameters
that
need
to
be
considered
when
defining
the
guilds
and
associated
generic
representatives
for
a
Level
II
assessment?
If
so,
please
identify.
c.)
Please
provide
direction
on
the
appropriate
application
of
the
additional
parameter(
s)
in
defining
the
generic
species
and
provide
discussion
on
how
the
additional
parameters
will
improve
the
characterization
of
the
uncertainty
in
risk
estimate.

2.
Assigning
Values
to
Generic
Species
Variables.
Four
variables
were
used
to
define
a
generic
species:
body
weight,
food
type,
frequency
on
field,
and
persistence
factor.
Values
for
each
variable
were
established
as
follows:

Body
Weight:
Selected
as
the
smallest
species
within
each
guild
Frequency
on
Field:
Selected
as
the
95th
percentile
of
available
observations
for
species
within
the
guild
Food
Type:
Assumed
obligate
feeders
for
granivore,
insectivore,
and
herbivore
acknowledging
that
omnivore
exposures
would
be
bracketed
by
these
groups.
Persistence
Factor:
Values
assigned
to
reflect
past
SAP
comments
that
repetitive
behavior
patterns
be
included
in
the
assessment.

a.)
Please
comment
on
whether
the
methods
used
for
establishing
values
and
their
results
appear
to
be
appropriate
for
generic
species
for
a
Level
II
assessment.
b.)
Does
the
SAP
believe
that
more
rigorous
analysis
is
necessary
or
indeed
possible
for
generic
species?
Or,
should
such
an
in­
depth
analysis
be
more
appropriately
applied
at
the
species­
specific
level
of
assessment?
Please
explain.

3.
Bimodal
Feeding
Pattern
and
Serial
Correlation
of
Foraging
Events.
The
model
was
modified
to
incorporate
hourly
choices
for
foraging
areas,
a
bimodal
feeding
pattern,
and
to
account
for
serial
correlation
in
sequential
foraging
events.
Page
9
of
61
a.)
Please
comment
on
the
strengths
and
weaknesses
of
the
modified
algorithm
in
representing
avian
feeding
behavior
for
the
more
vulnerable
species
in
agro­
ecosystems.
b.)
Please
provide
additional
suggestions
for
modifications
in
the
algorithm
to
more
closely
represent
avian
activity
patterns.
c.)
Please
provide
direction
on
the
appropriate
application
of
the
additional
modifications
and
provide
discussion
on
how
the
modifications
will
improve
the
characterization
of
the
uncertainty
in
risk
estimates.

4.
New
Puddle
Algorithm.
A
new
puddle
algorithm
was
developed
to
account
for
a
number
of
parameters
that
affect
puddling
after
a
rainfall
event
in
agro­
environments.
The
new
algorithm
addresses
rainfall
amount,
rainfall
duration,
soil
infiltration
rates,
evaporation,
degradation
and
the
stochastic
nature
of
field
topography
and
its
relation
to
puddle
formation
and
duration.

a.)
Please
comment
on
the
overall
model
structure
in
relation
to
mimicking
puddles
in
agroenvironments
including
any
suggestions
on
modifications
or
additional
parameters
to
considered
that
would
improve
pesticide
concentration
estimates
in
this
environmental
media.
b.)
Please
provide
suggestions
for
assigning
values
to
puddle
input
variables
and
for
locating
additional
sources
of
information
that
may
help
in
defining
these
values.

5.
Air
Concentration
Estimation.
The
model
currently
employs
an
equilibrium­
based
two
compartmental
model,
for
estimating
pesticide
air
concentration
in
the
plant
canopy.
Please
comment
on
the
merits
and
limitations
of
this
approaches.
Would
the
SAP
provide
suggestions
on
additional
alternatives
for
estimating
vapor
phase
concentrations
that
would
be
consistent
with
the
physical/
chemical
property
and
environmental
fate
data
available
to
the
Agency
as
guideline
information?
Please
comment
on
the
merits
and
limitations
of
these
additional
approaches.

6.
Relating
Inhalation
Exposure
to
Oral
Exposure
Toxicity
Endpoints:
The
absence
of
avian
inhalation
toxicity
data
and
the
need
to
track
all
exposure
routes
simultaneously
has
led
to
the
development
of
a
method
to
relate
inhalation
exposures
to
oral­
dose
equivalents.
The
method
uses
the
relationship
between
mammalian
inhalation
and
oral
acute
toxicity
endpoints
along
with
an
adjustment
factor
to
account
for
some
basic
physiological
differences
between
the
mammalian
and
avian
lungs
assumed
important
to
inhaled
pesticide
bioavailability.

a.)
Please
comment
on
whether
OPP's
proposed
approach
for
relating
inhalation
exposure
to
oral­
dose
equivalents
addresses
SAP's
previous
comments
concerning
the
use
of
the
mammalian
inhalation/
oral
relationship
for
estimating
toxicity
in
birds.
b.)
Please
provide
suggestions
on
alternatives
for
estimating
avian
inhalation
toxicity
that
would
be
consistent
with
the
kinds
of
toxicity
data
generally
available
to
the
Agency.

7.
Estimating
Dermal
Exposure:
The
incidental
dermal
contact
model
relies
on
methods
currently
employed
by
the
OPP's
Health
Effects
Division
that
rely
on
estimates
of
foliar
contact
and
Page
10
of
61
dislodgeable
foliar
residues
to
estimate
an
external
dermal
dose.

a.)
Please
comment
on
applying
this
general
approach
to
birds
and
whether
any
other
model
alternatives
have
been
used
for
wildlife
dermal
exposure.
b.)
If
alternative
models
for
estimating
dermal
exposure
for
birds
are
available,
please
discuss
their
advantages
and
limitations
in
comparison
to
the
proposed
model.
c.)
Please
comment
on
the
following:

1.)
The
reliance
on
the
lower
leg
and
foot
as
the
significant
contact
area
for
birds.
Are
other
portions
of
avian
anatomy
significant?
If
so,
which
other
areas
should
be
included?
2.)
Recognizing
that
the
use
of
human
foliar
contact
data
has
limitations,
can
the
SAP
share
any
insights
on
available
data
that
would
allow
for
a
more
specific
foliar
contact
rate
estimate
for
birds?
3.)
Is
the
SAP
aware
of
any
data
specific
to
pesticide
foliar
residue
transfer
coefficients
for
wildlife?
If
so,
please
identify.

8.
Relating
Dermal
Exposure
to
Oral
Exposure
Toxicity
Endpoints:
The
general
absence
of
avian
dermal
toxicity
data
and
the
need
to
track
all
exposure
routes
simultaneously
have
led
to
the
development
of
a
method
to
relate
dermal
exposures
to
oral­
dose
equivalents.
The
method
uses
existing
avian
dermal
toxicity
for
a
subset
of
pesticides
to
establish
a
relationship
between
avian
dermal
and
oral
acute
toxicity
endpoints.
It
is
recognized
that
this
approach
is
statistically
limited
with
regards
to
the
strength
of
that
relationship,
and
that
this
method
is
constrained
by
the
limited
number
of
pesticide
modes
of
action
considered.
Please
provide
suggestions
regarding
other
route
normalization
techniques.

9.
Physiologically­
based
Toxicokinetic
Modeling.
The
methods
developed
to
estimate
risk
from
multimedia
and
different
routes
of
exposure
are
based
on
external
dose
estimates
that
do
not
directly
account
for
physiological,
morphological,
and
biochemical
processes
that
underlie
the
toxicokinetic
behavior
of
a
pesticide.
In
human
health
and
aquatic
life
risk
assessments
for
drugs,
and
in
some
cases
environmental
contaminants,
use
of
physiologically­
based
toxicokinetic
(
PBTK
models,
are
beginning
to
be
employed
to
derive
internal
dose
estimates
for
more
refined
dose­
response
analyses
and
to
support
route­
to­
route
and
interspecies
extrapolation.
In
this
regard,
PB­
TK
modeling
was
mentioned
by
the
SAP
during
the
2001
review
of
the
case
studies.

a.)
If
you
are
aware
of
any
developmental
work
on
avian
PB­
TK
models
since
2001,
please
discuss.
Is
the
SAP
aware
of
information
sources
that
have
compiled
measured
physiological,
morphological,
and/
or
biochemical
parameters
that
are
required
to
develop
avian
PB­
TK
models?
If
so,
please
comment.
b.)
Recognizing
that
research
to
support
PB­
TK
models
is
a
long­
term
and
collaborative
endeavor
across
the
Agency
and
the
scientific
community,
identifying
potential
applications
in
a
risk
assessment
context
can
provide
insights
for
prioritizing
developmental
efforts.
In
this
regard,
any
suggestions
by
the
SAP
in
terms
of
an
incremental
application
of
physiologically­
based
perspectives
in
problem
formulation,
Page
11
of
61
analysis
and/
or
the
risk
characterization
phases
of
an
assessment
would
be
welcomed.
In
addition,
any
suggestions
that
may
be
helpful
to
the
broader
scientific
community
in
terms
of
research
priorities
to
develop
avian
PB­
TK
models
would
be
appreciated.

SUMMARY
OF
PANEL
DISCUSSION
AND
RECOMMENDATIONS
The
FIFRA
Scientific
Advisory
Panel
(
the
Panel)
was
charged
with
reviewing
progress
that
has
been
made
by
the
Agency
regarding
probabilistic
risk
assessment
modeling.
The
Panel
commends
the
Agency
for
initiating
a
Probabilistic
Risk
Assessment
process
in
1996,
convening
experts
to
develop
the
conceptual
models,
forming
an
Implementation
Team,
disseminating
the
modeling
process
to
the
scientific
community
and
inviting
periodic
review
by
this
and
previous
panels.

The
Agency's
terrestrial
risk
assessment
paradigm
currently
has
four
levels.

Level
I:
Screening
level
Level
II:
Initial
estimate
of
probability
and
magnitude
of
effect.
Relying
on
existing
data
sets.
Planning
for
simplistic
estimates
of
population
impacts
Level
III
&
IV
Probabilistic
risk
assessments
At
this
point,
the
Agency's
modeling
sophistication
lies
somewhere
between
Level
II
and
Level
III.
Models
are
needed
at
Level
II
and
higher
to
evaluate
a
high
throughput
of
approximately
70
chemical
products
that
are
currently
evaluated
by
the
Agency
each
year.

The
Panel
commends
the
Agency
for
the
technical
efforts
made
in
model
development
to
date.
Several
innovative
approaches
have
been
employed
and
the
number
of
parameters
considered
is
extensive.
The
Agency's
model
has
many
positive
aspects
and
a
few
areas
that
need
improvement.
Members
of
the
Panel
who
served
on
past
Panels
regarding
probabilistic
modeling
of
terrestrial
risks
agree
that
the
Agency
has
successfully
included
many
of
the
suggestions
made
by
previous
Panels.
Overall
the
Panel
believed
the
model
will
be
appropriate
for
use
in
Level
II
assessments
after
a
few
of
their
recommendations
are
addressed
as
presented
below.
More
detail
is
given
in
the
Panel's
response
to
the
Agency's
charge
to
the
Panel.

1.
Frequency
on
field
should
incorporate
foraging
frequencies
during
breeding
seasons.
Foraging
frequency
and
strategy
should
be
determined
for
different
crop
types
and
different
regions
of
the
country.
Time
in
edge
habitat
should
be
considered
as
presenting
exposure
when
that
habitat
has
received
spray
drift.
2.
Feeding
behavior
should
better
incorporate
the
feeding
strategies
of
breeding
birds
as
these
strategies
are
likely
to
be
more
representative
of
birds
foraging
in
fields
during
planting
and
soon
thereafter.
Page
12
of
61
3.
The
bimodal
feeding
pattern
is
an
improvement
over
past
approaches,
but
probably
does
not
represent
avian
foraging
behaviors,
especially
when
birds
are
brooding
nestlings.
Fortunately,
the
time
in
field
is
likely
to
be
more
important
to
dermal
and
inhalation
exposures
than
to
ingestion,
because
as
long
as
the
food
intake
is
correctly
estimated,
the
time
to
acquire
that
food
does
not
alter
oral
intake.
4.
Inhalation
exposure
should
be
better
quantified
by
allowing
toxicant
volatilization
from
soil
as
well
as
plants.
The
maximum
droplet
size
considered
for
inhalation
exposure
should
not
be
limited.
Larger
sizes
are
brought
into
the
upper
respiratory
tract
and
thus
serve
as
a
source
of
exposure.
5.
Dermal
exposure
should
better
characterize
foot
morphology.
6.
Inhalation
effects
need
to
be
better
defined
through
a
larger
data
set
and
empirical
data
for
birds.
This
is
required
as
the
actual
detoxification
capacity
and
rate
may
be
different
between
birds
and
mammals.

The
Panel's
comments
are
based
on
our
understanding
of
the
Agency's
desire
to
use
the
Level
II
Risk
Assessment
to
make
a
specific
regulatory
decision
to
stop
evaluation
of
a
given
product
or
require
more
sophisticated
(
higher
tier)
evaluation
of
the
product.
Having
a
discrete
decision
point
that
incorporates
the
Level
II
assessment
will
avoid
a
regulatory
process
where
all
products
de
facto
proceed
to
Level
III
or
IV
assessments.
If
a
decision
is
made
to
proceed
to
higher
levels
of
evaluation,
the
outcome
of
Level
II
should
focus
the
Agency's
data­
gathering
efforts
that
will
be
essential
for
proper
modeling
in
the
more
refined
Level
III
or
Level
IV
assessments.

With
this
in
mind,
it
is
important
to
note
that
the
risk
assessment
paradigm
has
a
feedback
mechanism,
wherein
assessment
outputs
are
evaluated
through
data
acquisition
and
system
monitoring.
There
is
no
indication
that
this
is
being
done
in
the
development
of
the
current
models.
More
troubling
is
the
appearance
that
there
is
no
intention
to
obtain
appropriate
data
to
improve
parameter
estimation
and
to
validate
model
outcomes.
The
Panel
strongly
recommends
that
the
Agency
obtain
data
that
validate
critical
modules
within
existing
models
and
that
can
be
used
to
refine
distributions
that
will
be
needed
in
higher
levels
of
the
risk
assessment
process.

PANEL
DELIBERATIONS
AND
RESPONSE
TO
THE
CHARGE
The
specific
issues
to
be
addressed
by
the
Panel
are
keyed
to
the
Agency's
background
documents,
references
and
the
Agency's
charge
questions.

Agency
Charge
1.
Guild
Parameters
Used
for
Defining
Generic
Species.
The
process
for
defining
generic
species
described
in
this
document
separated
species
into
guilds
based
on
three
parameters:
feeding
substrate,
nesting
substrate,
and
food
type.

a.)
Please
comment
on
the
representative
guilds
used
to
define
the
generic
organisms.
Page
13
of
61
Panel
Response
Using
a
guild
approach
to
identify
generic
species
seems
reasonable
for
Level
II
models.
Many
different
criteria
can
be
used
to
identify
guilds,
and
the
guilds
selected
should
be
those
most
relevant
to
assessments
of
the
potential
for
pesticide
exposure.
Classifying
birds
on
the
basis
of
feeding
substrate
and
food
type
are
particularly
germane
to
risk
assessment,
as
these
parameters
determine
where,
and
upon
what,
the
birds
will
feed.
Nesting
substrate
may
also
be
a
useful
classification
to
the
degree
that
it
provides
information
about
the
likelihood
of
nest
placement
relative
to
the
cropping
unit
being
treated.

The
generic
species
were
based
on
the
guild
structure
of
birds
using
cornfields
(
Best
et
al.
1990).
Unfortunately,
more
data
describing
the
use
of
crop
fields
by
birds
are
not
available.
The
primary,
if
not
sole
data
used
are
those
Best
et
al.
collected
more
than
a
decade
ago,
and
are
restricted
to
corn
fields.
Additional
studies
have
been
conducted
and
are
described
in
the
following
paragraphs.
Differences
among
crop
types,
regionally
and
seasonally,
are
likely
to
contribute
significantly
to
the
guild
structure
present
in
a
given
agroecosystem.

The
guild
composition
of
the
bird
community
in
cropping
systems
other
than
corn
would
be
different.
Orchard
systems,
for
example,
would
likely
have
a
greater
proportion
of
birds
feeding
above
the
ground
(
in
the
canopy
of
shrubs
and
trees)
than
would
be
the
case
for
rowcrop
fields.
Bird
abundance
data
have
been
gathered
for
a
variety
of
crop
types,
and
the
guild
determinations
should
be
based
on
the
composition
of
the
bird
community
associated
with
the
crop
type
being
evaluated.
CropLife
America
(
Best
and
Murray
2003)
has
compiled
a
database
of
avian
field
studies
conducted
on
cornfields
and
cotton
fields
in
different
regions
of
the
United
States.
There
are
also
use
data
and
diet
composition
from
studies
in
corn
(
Brewer
et
al.,
1990;
Brewer
et
al.,
1992
),
turf
(
Hummell
et
al.,
1990)
and
orchards
(
Melott
et
al.,
1990).
The
bird
survey
results
from
these
and
other
studies
could
be
used
to
define
guilds
more
appropriate
to
specific
crops
and
geographical
regions.
Empirical
data
on
bird
use
of
fields
containing
the
relevant
crop
type
should
be
used
to
determine
the
relative
weighting
(
contribution)
of
the
various
guilds
in
"
constructing"
the
generic
species
that
is
used
in
the
risk
assessment.
Future
efforts
may
be
required
to
define
generic
species
for
different
regions
of
the
country,
given
that
the
current
suite
of
species
is
based
on
a
study
of
cornfields
in
one
region
of
the
United
States.

Best
and
Murray
(
2003)
acknowledge
that
as
the
height
and
basal
area
of
the
plant
canopy
increases,
the
species
and
feeding
guilds
of
birds
that
frequent
agricultural
fields
may
change.
This
is
almost
certainly
the
case
with
rapidly
growing
crops.
For
example,
as
the
growing
season
progresses
in
cornfields,
barren,
sparsely
vegetated
fields
are
transformed
into
fields
dominated
by
dense
vertical
plant
cover
that
can
attain
heights
of
over
2.5
m
(
Rodenhouse
and
Best
1994).

Several
panelists
believed
that
if
the
role
of
the
Level
II
modeling
is
as
suggested,
species
should
be
randomly
selected
from
a
species
distribution
weighted
by
the
relative
abundance
of
individuals
within
each
species
instead
of
selecting
the
smallest
bird
within
each
guild.
This
is
Page
14
of
61
important
because
species
selection
dictates
the
parameterization
of
other
aspects
of
the
models,
i.
e.
frequency
on/
off
the
field
and
percent
of
diet
from
the
fields.
This
approach
also
reduces/
eliminates
the
possibility
of
mathematically
constructing
a
generic
species
that
is
unrealistic.
This
is
likely
to
become
a
more
pressing
issue
at
higher
levels
of
the
risk
assessment
progression.

Two
practical
considerations
mitigate
against
over­
reliance
on
developing
a
generic
species.
One
is
that
whereas
one
intuitively
thinks
of
cornfield
birds
as
small
sparrow­
like
passerines,
several
other
classes
of
birds
may
be
of
concern
in
a
risk
assessment.
Species
such
as
harriers
and
other
raptor
or
herbivorous
waterfowl
do
not
fit
the
model
yet
constitute
important
classes
of
birds
whose
fate
may
need
to
be
assessed
within
the
model.
The
second
issue
is
that
efforts
to
combine
attributes
of
birds
into
a
single
generic
species
run
the
risk
of
creating
a
species
that
does
not
actually
exist
(
e.
g.,
a
64g
nighthawk
that
spends
all
of
its
time
on
the
ground).
Other
Panel
members
differed
from
this
position
since
the
Agency
developed
different
generic
species
for
different
guilds.

While
conservative
assumptions
are
useful
in
deterministic
models,
a
fully
stochastic
model
should
be
more
realistic,
allowing
the
randomness
in
the
model
parameters
to
generate
the
extremes.
In
terms
of
the
guilds,
two
conservative
assumptions
were
made.
Body
weight
was
taken
to
be
that
of
the
smallest
bird
in
the
guild,
and
the
frequency
­
on­
field
(
FOF)
was
taken
to
be
the
upper
95th
percentile.
Some
Panelists
recommended
that
distributions
be
used
where
possible,
and
if
distributions
are
unavailable
that
the
lack
of
these
distributions
guide
data
collection
needs
for
higher
level
risk
assessments.

Agency
Charge
b.)
Are
there
any
additional
parameters
that
need
to
be
considered
when
defining
the
guilds
and
associated
generic
representatives
for
a
Level
II
assessment?
If
so,
please
identify.

Panel
Response
An
important
factor
not
included
in
the
generic
species
approach
is
a
consideration
of
seasonal
changes
in
the
use
of
treated
fields
by
birds.
The
occurrence
classifications
of
Best
et
al.
(
1990)
for
bird
use
of
cornfields
were
based
on
composite
census
data
summarized
over
a
2­
month
period.
The
use
of
those
fields
by
individual
species
(
and
guilds)
differed
over
that
period.
Because
pesticide
applications
occur
at
specific
times
during
the
growing
season,
depending
upon
the
pest
being
controlled,
composite
estimates
of
field
use
may
be
misleading.
Most
bird
census
data
consist
of
repeated
surveys,
and
although
the
results
may
be
reported
as
a
seasonal
composite,
the
original
data
would
often
permit
the
development
of
seasonal­
use
profiles.

This
can
be
illustrated
by
data
that
have
been
collected
for
cornfields
(
Best
2001).
Bird
use
of
cornfields
in
late
April
in
Iowa
is
dominated
by
species
that
forage
on
the
ground
or
in
low
Page
15
of
61
herbaceous
vegetation,
with
little
use
by
species
that
forage
in
shrubs
and
trees.
By
early
August
the
proportional
field
use
by
birds
that
forage
on
the
ground
or
in
low
herbaceous
vegetation
has
declined
substantially
and
is
similar
to
proportions
of
birds
that
feed
on
shrubs
and
trees.
A
similar
seasonal
shift
occurs
in
alfalfa
fields
(
Best,
unpublished
data).
Such
seasonal
shifts
in
fields
used
by
various
avian
guilds
should
be
factored
into
the
determination
of
generic
species.
Depending
upon
when
a
pesticide
is
applied
(
e.
g.,
corn
rootworm
versus
corn
borer
control),
the
most
susceptible
guild(
s)
change(
s).

The
current
model
minimizes
the
importance
of
plant
development
and
foliation
on
the
composition
of
the
bird
community
using
treated
fields.
Seasonal
changes
in
bird
use
of
crop
fields
do
occur,
and
these
affect
body
size
distributions,
the
food
type
and
foraging
substrate
used
by
birds,
and
the
frequency
of
field
use.

Effects
of
edge
habitat
on
the
guild
structure
of
bird
communities
within
pesticide­
treated
fields
was
not
considered.
Edge
habitats
do
affect
bird
use
of
fields,
and
this
effect
can
vary
regionally
(
Best
et
al.
1990,
Best
and
Murray
2003).
For
example,
the
bird
species
composition
within
cornfields
bordered
by
herbaceous
vegetation
differs
from
that
in
cornfields
adjacent
to
woodland
(
Best
et
al.
1990).
To
further
illustrate
this,
effects
of
edge
habitat
on
the
frequency
of
cornfield
use
by
birds
is
affected
by
the
gradient
in
precipitation
progressing
from
the
Midwest
to
the
more
arid
Southwest
(
Best
and
Murray
2003).
A
review
of
the
number
of
species
associated
with
edge
habitats
suggests
that
a
spray­
drift
component
should
be
added
to
the
model.

Body
size
also
has
implications
for
the
area
of
exposure
and
perhaps
density.
Because
body
size
might
also
scale
with
foraging
area,
it
is
possible
that
as
assessments
progress
from
"
the
field"
to
regional
and
population
applications
that
the
smallest
birds
will
not
necessarily
be
at
highest
risk
from
an
"
incidence
of
mortality"
or
population
standpoint.
This
was
examined
by
Freshman
and
Menzie
(
1996)
and
Figure
1
illustrates
how
foraging­
area
scales
cause
variations
in
exposures
at
a
local
population
level
(
incidence
within
the
population)
when
the
exposure
field
consists
of
patches
of
contamination
(
e.
g.,
individual
fields
in
a
regional
landscape).

In
Figure
1,
the
incidence
of
exposure
in
a
local
population
is
influenced
by
the
scale
of
foraging
of
individuals
as
well
as
the
spatial
scales
of
the
contamination.
The
example
also
shows
that
when
one
moves
from
risk
to
an
individual
to
risks
to
a
group
of
individuals
where
that
is
expressed
as
an
incidence,
it
is
no
longer
the
individuals
with
the
smallest
foraging
area
(
often
smaller
sized
individuals)
that
are
necessarily
at
greatest
risk.
In
higher
level
analyses,
the
Agency
should
consider
the
spatial
scale
of
exposures
and
how
this
might
influence
selection
of
representative
guilds
and
species
within
guilds.
Page
16
of
61
Menzie­
Cura
&
Associates,
Inc.
0
0.2
0.4
0.6
0.8
Ratio
of
Foraging
Area
to
Total
Habitat
0
20
40
60
80
100
Percent
of
Local
Population
Affected
65
mg/
kg
100
mg/
kg
130
mg/
kg
165
mg/
kg
Soil
Effect
Level
for
Chemical
A
Percent
of
Local
Population
Affected
vs.
Foraging
Area
Figure
1.
Effects
of
foraging
area
on
avian
population
Agency
Charge
c.)
Please
provide
direction
on
the
appropriate
application
of
the
additional
parameter(
s)
in
defining
the
generic
species
and
provide
discussion
on
how
the
additional
parameters
will
improve
the
characterization
of
the
uncertainty
in
risk
estimate.

Panel
Response
Future
efforts
may
consider
additional
avian
foraging
guilds,
such
as
scavengers
and/
or
carnivores,
which
could
receive
elevated
exposures
from
feeding
on
dead
or
dying
small
mammals
and
birds.
This
guild
would
only
need
to
be
considered
for
persistent
and/
or
bioaccumulative
pesticides.
It
may
also
be
useful
to
consider
average
body
weight
of
generic
species
rather
than
focusing
only
on
the
smallest
member
of
the
guild.
This
will
help
ascertain
both
worst­
case
risk
and
typical
risk
expected
for
the
guild.

Feeding
frequency
will
be
an
important
aspect
of
exposure
because
of
the
way
the
exposure
concentration
data
are
handled.
At
present,
the
bird
experiences
(
eats
at)
a
pre­
selected
concentration
for
a
particular
time
step.
The
fewer
the
meals,
the
greater
the
likelihood
that
a
bird
will
get
a
toxic
dose
or
non­
toxic
dose
within
a
particular
time
step.
If
smaller
birds
eat
many
meals
during
the
course
of
the
day,
then
they
integrate
exposure
across
the
field
to
some
degree
on
a
daily
basis.
This
integration
has
the
effect
of
decreasing
the
extreme
exposures
from
a
toxicological
standpoint
(
they
eat
the
highest
and
lowest
concentrations
but
in
smaller
quantities).
This
could
both
decrease
and
increase
the
incidence
of
effect.
In
either
case,
the
frequency
and
size
of
the
meal
will
be
an
important
factor
in
the
model.
Finally,
an
additional
parameter
that
should
be
considered
when
defining
guilds
and
associated
generic
representatives
is
incidental
Page
17
of
61
ingestion
of
soils
for
those
birds
that
forage
in
or
on
the
soil.

Agency
Charge
2.
Assigning
Values
to
Generic
Species
Variables.
Four
variables
were
used
to
define
a
generic
species:
body
weight,
food
type,
frequency
on
field,
and
persistence
factor.
Values
for
each
variable
were
established
as
follows:

Body
Weight:
Selected
as
the
smallest
species
within
each
guild
Frequency
on
Field:
Selected
as
the
95th
percentile
of
available
observations
for
species
within
the
guild
Food
Type:
Assumed
obligate
feeders
for
granivore,
insectivore,
and
herbivore
acknowledging
that
omnivore
exposures
would
be
bracketed
by
these
groups.
Persistence
Factor:
Values
assigned
to
reflect
past
SAP
comments
that
repetitive
behavior
patterns
be
included
in
the
assessment.

a.)
Please
comment
on
whether
the
methods
used
for
establishing
values
and
their
results
appear
to
be
appropriate
for
generic
species
for
a
Level
II
assessment.

Panel
Response
The
value
of
a
generic
species
varies
markedly
with
the
purpose
of
the
modeling.
If
one
wants
to
model
a
population
(
or
other)
endpoint
accurately
and
realistically,
the
notion
of
a
generic
species
that
serves
as
a
holistic
integration
of
species
attributes
is
reasonable.
However,
short
of
this
full­
scale
model,
there
are
many
advantages
to
using
the
modeling
process
to
explore
the
consequence
of
variation
in
attributes
across
birds.
That
is,
one
can
ask
what
the
relative
effect
on
output
would
be
if
the
bird
considered
were
a
30g
bird
rather
than
a
60g
bird
or
were
an
insectivore
rather
that
a
granivore.
For
such
sensitivity
analysis,
one
does
not
want
a
generic
bird
species
but
rather
a
bird
whose
attributes,
whilst
realistic,
can
be
varied
independently
in
a
sensitivity
analysis.

Against
this
background,
some
comments
may
be
made
about
the
current
inputs.
First,
the
diet
and
nesting
behavior
of
birds
seem
to
be
good
predictors
of
whether
a
species
regularly
occupies
corn
fields
or
regularly
occupies
the
surrounding
habitat.
There
does
not
seem
to
be
much
room
for
improving
on
the
predictability
obtained
from
these
two
variables
and
there
may
in
fact
be
a
risk
of
over­
fitting
the
existing
field
data
(
Best
et
al.
1990)
at
the
expense
of
maintaining
more
general
predictive
power.
Any
effort
to
go
beyond
what
is
currently
achieved
with
these
two
variables
needs
to
be
tested
with
new
datasets
and
it
is
questionable
whether
the
effort
is
likely
to
bring
about
incremental
gain.

Body
weight
Body
weight
appears
as
a
parameter
at
many
points
in
the
risk
assessment
model;
it
Page
18
of
61
determines
the
metabolic
rate
used,
determines
the
surface
area
of
the
bird
in
the
dermal
exposure
component,
appears
in
the
toxicity
calculations,
and
so
on.
It
follows
that
seeking
to
specify
a
single
weight
to
be
used
as
an
attribute
of
a
generic
species
whose
fate
will
subsequently
be
followed
is
a
sub­
optimal
strategy.
Instead,
it
makes
more
sense
to
retain
body
weight
as
a
critical
variable
to
be
subjected
to
sensitivity
analysis.
This
transforms
the
choice
of
weight
of
the
"
generic
species"
from
a
critical
decision
with
uncertain
consequences
to
one
whose
criticality
is
known:
where
sensitivity
proves
to
be
low,
the
model
is
insensitive
to
the
choice
of
body
weight
and
one
can
make
a
robust
choice
easily;
where
sensitivity
proves
to
be
high,
one
has
learned
immediately
that
determining
the
weight
distribution
of
the
birds
in
the
exposure
area
will
allow
better
quantification
of
uncertainty.
A
further
refinement
would
be
to
weight
the
body
size
estimates
according
to
the
relative
abundance
of
the
various
species
represented
by
these
body
sizes.
Thus,
body
sizes
represented
by
more
individuals
within
the
avian
community
would
then
contribute
proportionally
more
to
the
model
simulation.
Likewise,
one
could
also
draw
from
the
entire
distribution
of
frequencies
on
the
field
for
the
various
species
and
weight
them
according
to
their
relative
occurrences
as
measured
by
census
data.
Bird
survey
information
for
the
relevant
crop
could
be
used
to
provide
the
data
inputs
for
body
size
and
frequency
on
field.
A
major
compilation
of
body
weights
for
much
of
the
U.
S.
avifauna
is
available
(
Dunning
1993)
and
provides
a
ready
tool,
by
merging
with
abundance
data
for
the
field,
to
determine
a
distribution
of
weight­
dependent
effects.

Frequency
on
Field/
Persistence
Factors
Two
other
factors
being
considered
by
the
Agency
as
pertinent
to
the
delineation
of
a
generic
species
are
FOF
and
persistence.
One
must
have
some
concern
about
the
reliability
of
the
FOF
metric
when
edge
species
score
more
that
50%
on
the
metric:
is
it
realistic
to
have,
by
definition,
a
non­
field
species
spending
more
than
half
of
its
time
on
the
field?
The
quality
of
the
data
sources
for
these
estimates
may
be
the
issue
here.
During
field
surveys
the
FOF
metric
could
become
inflated
by
a
census
bias
against
effectively
detecting
edge
species
while
they
are
in
the
edge
habitat.
The
other
possibility
is
that
the
processing
of
the
empirical
data
in
search
of
an
appropriate
metric
may
have
involved
assumptions
not
met
by
the
data.
This
suggests,
as
with
body
weight
and
other
data
incorporated
in
the
model,
that
sampling
from
the
empirical
data
distribution
and
calculating
the
resulting
distribution
of
outcomes
(
and
its
sensitivity)
would
be
a
better
analytical
strategy
than
selecting
a
point
metric.

The
Panel
agreed
that
one
of
the
smaller
species
on
field
is
the
best
selection
for
the
generic
species,
because
they
are
the
most
vulnerable
species.
Including
low
weight
and
high
sensitivity
as
characteristics
for
the
generic
species
offers
a
better
representation
of
young
birds
or
nestlings
which
are
more
sensitive
and
may
cause
more
drastic
population
declines
if
removed
from
the
population.
Also
at
this
level
of
refinement
some
conservatism
is
desired.

When
considering
the
persistence
metric,
two
scenarios
exist.
One
is
that
of
serial
correlation,
the
idea
that
the
state
of
the
bird
(
e.
g.,
feeding
versus
non­
feeding)
is
a
function
of
its
state
in
a
previous
period.
Some
data
(
McFarland
(
1994)
suggest
that
this
effect
takes
place
on
Page
19
of
61
the
order
of
minutes,
too
short
a
resolution
for
the
issue
to
matter
with
the
one
hour
time
step
currently
in
use.
The
second
pathway
to
overt
persistence
is
that
generated
by
territorial
or
similar
behavior
on
the
part
of
the
bird.
A
territorial
bird
is
indeed
more
likely
to
be
seen
feeding
at
a
later
time
where
it
was
seen
feeding
earlier
but
this
is
more
a
form
of
location
preference
than
a
true
persistence.
A
practical
line
of
investigation
of
this
point
that
the
Agency
might
consider
is
to
conduct
sensitivity
analysis
of
the
persistence
variables
as
currently
incorporated
in
the
model.
If
the
persistence
variable
proves
to
have
low
sensitivity,
there
may
be
a
case
for
discontinuing
its
incorporation
into
the
present
model,
on
the
grounds
that
assessing
persistence
empirically
is
then
a
difficult
task
with
little
significance
for
the
behavior
of
the
model.

As
described
above,
a
key
idea
is
that
greater
reliance
on
exploring
the
consequences
for
the
output
of
sampling
across
the
empirical
distributions
of
variables
of
potential
interest
is
preferable
to
trying
to
determine
the
safe
(
in
terms
of
risk
minimization)
values
of
specific
inputs
to
use.
Instead,
one
can
bound
the
distribution
of
output
in
terms
of
what
risk
levels
one
will
consider,
with
the
use
of
distributions
of
input
variables
revealing
how
sensitive
these
bounds
are
to
the
individual
variables.
This
is
a
general
principle
of
stochastic
modeling
that
should
underlie
further
sophistication
of
the
modeling
approach
being
developed
here.

Agency
Charge
b.)
Does
the
SAP
believe
that
more
rigorous
analysis
is
necessary
or
indeed
possible
for
generic
species?
Or,
should
such
an
in­
depth
analysis
be
more
appropriately
applied
at
the
species­
specific
level
of
assessment?
Please
explain.

Panel
Response
Assuming
obligate
feeders
for
granivore,
insectivore,
and
herbivore
and
acknowledging
that
omnivore
exposures
would
be
bracketed
by
these
groups,
this
is
a
reasonable
Level
II
approach.
While
this
approach
is
supported
by
the
Panel,
the
following
issues
should
be
recognized:
(
1)
the
guild
omnivore
encompasses
the
greatest
number
of
bird
species
and
it
covers
a
broad
range
of
proportional
feeding
on
plant
and
animal
foods;
and
(
2)
during
the
breeding
season,
many
adult
granivorous
birds
become
omnivores,
but
the
vast
majority
feed
their
young
exclusively
an
animal
food
diet.
Spatial
aspects
of
exposure
are
also
absent
in
the
current
model,
and
should
be
included
in
the
future.
Furthermore,
the
analysis
does
not
currently
incorporate
changes
in
food
habits
associated
with
insect
"
knock
down",
i.
e.,
greater
availability
of
poisoned
prey
immediately
following
spray.
Field
use
may
vary
immediately
after
chemical
application.
These
behaviors
should
be
evaluated
at
higher
levels.

Given
the
paradigm
suggested
above,
generic
species
(
appropriately
defined)
are
appropriate
for
Level
II
analyses.
Focal
species
could
then
be
included
in
Level
III
and
IV
models.
Here,
population
impacts
would
also
be
modeled.

The
Agency
discounted
the
effect
of
seasonal
changes
in
the
height
and
basal
area
of
the
Page
20
of
61
plant
canopy
when
assigning
values
to
generic
species
variables.
Seasonal
changes
do
occur
and
in
many
cases
they
can
significantly
influence
both
the
body
weights
of
birds
on
fields
and
their
frequency
on
fields.
Furthermore,
both
the
foraging
substrate
and
food
type
can
change.
For
example,
in
cornfields
the
shift
is
from
less
consumption
of
seeds
to
more
consumption
of
insects.
Also
the
foraging
substrate
would
change
as
well,
from
predominantly
ground­
dwelling
food
items
to
items
found
on
the
developing
plants
(
as
discussed
in
more
detail
in
response
to
question
1).
Assigning
values
to
generic
species
variables
should
take
into
account
temporal
changes
in
food
type,
food
substrate,
and
frequency
of
occurrence
on
fields.

Using
the
5th
percentile
of
body
weights
is
conservative
and
could
generate
a
generic
species
that
does
not
exist.
In
a
fully
probabilistic
assessment,
a
distribution
of
bird
sizes
within
each
guild
is
probably
more
appropriate.
It
should
be
noted
that
FOF
is
actually
the
95th
confidence
interval
on
a
mean
value
(
not
the
95th
percentile).
This
may
not
be
an
appropriate
statistic.
Perhaps
dealing
with
this
as
an
uncertainty
analysis
selecting
specific
values
for
FOF
would
be
more
appropriate.
An
explanation
of
FOF
parameterization
is
important
for
communication
purposes.
Treating
FOF
as
a
distribution
may
not
be
possible
and
may
be
confusing.
Selected
values
represent
a
scenario
approach.
Such
an
approach
is
consistent
with
discussions
at
an
earlier
Monte
Carlo
workshop.

The
only
herbivores
listed
are
Canada
Geese
and
Widgeon.
The
Panel
believes
granivorous
birds
will
consume
some
green
herbaceous
food.
Does
the
model
adequately
cover
these
birds
as
noted
previously?
Many
species
change
from
granivorous
to
insectivorous
during
the
breeding
season,
and
the
model
should
reflect
avian
feeding
strategies
during
the
season
in
which
applications
are
planned.
That
is,
if
the
planned
use
is
during
the
spring­
summer,
granivorous
birds
should
be
adjusted
to
mostly
insectivorous.

Agency
Charge
3.
Bimodal
Feeding
Pattern
and
Serial
Correlation
of
Foraging
Events.
The
model
was
modified
to
incorporate
hourly
choices
for
foraging
areas,
a
bimodal
feeding
pattern,
and
to
account
for
serial
correlation
in
sequential
foraging
events.

a.)
Please
comment
on
the
strengths
and
weaknesses
of
the
modified
algorithm
in
representing
avian
feeding
behavior
for
the
more
vulnerable
species
in
agroecosystems

Panel
Response
The
model
was
modified
to
incorporate
hourly
choices
for
foraging
areas,
a
bimodal
feeding
pattern,
and
to
account
for
serial
correlation
in
sequential
foraging
events.
The
Panel
believed
the
general
approach
of
the
bimodal
feeding
pattern
allows
a
more
realistic
description
of
timedependent
foraging
and
an
improvement
over
the
previously
used
12­
hour
time
step.
However
actual
feeding
patterns
do
not
normally
drop
to
zero
at
midday.
This
observation
is
based
on
a
Page
21
of
61
procedure
which
used
two
betapert
distributions
with
non­
overlapping
time
ranges.
However,
it
is
recognized
that
feeding
intensity
can
take
on
a
range
of
values
that
may
be
influenced
predictably
by
daily
and
seasonal
influences
on
the
birds'
operative
temperature
and
breeding/
migration
status.
Still
other
behaviors
might
be
influenced
by
random
events
(
e.
g.,
flight
from
predators,
other
climate
variables).
Thus,
the
model,
as
presented,
may
not
be
representative
for
all
species
for
all
times
of
the
year.
However,
the
written
description
of
the
model
shows
the
feeding
pattern
as
a
general
mixture
distribution
that
would
not
have
this
problem.

It
is
important
to
differentiate
between
the
model
as
represented
by
equations
in
the
code
and
the
data
used
to
drive
the
model.
A
stochastic
modeling
approach
needs
to
be
based
on
realistic
ranges
of
data
for
all
inputs.
The
model
as
presented
in
its
written
form
is
adequate
to
realistically
describe
foraging
patterns.
However,
care
must
be
taken
with
the
constraints
imposed
on
data
input.
Stochastic
realizations
using
multiple
sequential
conservative
assumptions
can,
and
often
do,
lead
to
unreasonable
outputs.
Nonetheless,
the
stochastic
modeling
approach
yields
a
range
of
plausible
outputs
when
models
are
appropriate
and
reasonable
data
are
used
(
e.
g.,
bird
mortality
estimates).
The
risk
manager
then
sets
a
level
of
protection
that
is
used
as
a
threshold
for
comparison
to
these
plausible
outcomes.

The
algorithm
used
to
generate
the
bimodal
feeding
pattern
should
be
able
to
vary
the
periodicity
and
amplitude
of
the
bimodal
pattern
to
reflect
deterministic
and
stochastic
influences
on
bird
behavior.
For
example,
the
Agency's
background
document
indicated
that
the
beginning
and
ending
times
of
both
the
morning
and
afternoon
feeding
periods
are
assumed
to
vary
randomly
each
day,
within
specified
time
windows.
This
is
not
likely
the
case.
Beginning
and
ending
feeding
periods
are
influenced
strongly
by
photoperiod
which
does
not
vary
randomly.
Moreover,
adult
altricial
birds
that
are
provisioning
nestlings
most
likely
would
have
a
more
uniform
feeding
distribution
throughout
the
day
(
lower
amplitude,
i.
e.,
smaller
distance
between
peak
and
trough
for
feeding
intensity).
The
transitions
at
midday
would
be
more
gradual.
Additionally,
egg
laying
can
significantly
alter
the
foraging
behavior
(
duration/
intensity)
of
females.

To
achieve
a
more
realistic
bimodal
foraging
pattern,
bird
behavior
could
be
defined
by
a
mixture
distribution
where
the
time
ranges
for
the
morning
and
afternoon
could
overlap.
If
the
morning
and
afternoon
feeding
regimes
were
to
be
defined
by
beta
distributions
instead
of
betapert
distributions
(
requiring
one
more
user
input
parameter
for
each
portion
of
the
day),
more
realistic
scenarios
could
be
modeled.
The
strength
of
a
general
approach
also
can
be
a
short
coming
at
times
in
that
it
allows
data
inputs
that
may
not
be
realistic.
Conceivably,
one
also
could
model
a
foraging
behavior
that
was
uniform
across
the
daylight
hours
or
was
highest
at
midday.
Similarly,
a
foraging
pattern
distribution
could
be
developed
that
forces
more
food
intake
to
occur
in
short
periods
of
time
than
the
bird
can
actually
physically
process.

Most
Panel
members
agreed
that
the
Markov
chain
approach
allowed
a
realistic
characterization
of
serial
behavior.
The
Markov
chain
model
allows
the
characterization
of
the
Page
22
of
61
bimodal
feeding
pattern
typical
of
many
bird
species.
It
also
can
simulate
a
wide
range
of
bimodal
feeding
patterns
(
e.
g.,
different
start
and
end
times,
different
durations
and
intensities,
etc).
The
major
weakness
of
the
feeding
behavior
algorithm
is
that
it
does
not
allow
the
use
of
distributions
to
represent
frequency
and
persistence
on
field
variables.
Clearly,
there
will
be
much
interindividual
variability
for
both
of
these
variables
and
distributions
should
be
used
to
represent
this
variability.
In
situations
where
data
are
lacking
to
develop
distributions,
then
the
Agency
should
consider
re­
phrasing
the
analysis
to
be
specific
only
to
individuals
that
spend
a
large
component
of
their
time
on
fields
(
e.
g.,
90%).

Agency
Charge
b.)
Please
provide
additional
suggestions
for
modifications
in
the
algorithm
to
more
closely
represent
avian
activity
patterns.

Panel
Response
The
Panel
was
divided
on
the
optimal
time
step
of
the
model.
Some
members
of
the
Panel
believed
that
an
even
shorter
time
step
(
e.
g.,
on
the
scale
of
minutes)
would
even
better
reflect
avian
foraging
patterns,
especially
for
small
birds,
given
that
many
bird
species
move
in
and
out
of
fields
many
times
in
an
hour
when
they
are
actively
foraging.
However,
it
was
also
acknowledged
that
increasing
the
time
step
resolution
would
increase
the
computational
time
and
data
storage
requirements
for
the
simulations.
Other
Panel
members
believed
that
hourly
time
steps
are
a
reasonable
compromise
between
competing
time
step
choices,
especially
for
a
Level
II
model
in
a
tiered
modeling
system.
Although,
in
most
cases,
birds
that
are
not
field
residents
likely
feed
at
much
shorter
intervals,
it
may
be
that
the
1
hour
time
steps
reasonably
reflect
cumulative
feeding
activity.
The
serial
correlation
model
allows
the
bird
to
stay
feeding
on
the
same
field
for
several
hours
or
to
move
off
and
on
the
field
more
rapidly.
Finding
the
correct
data
to
drive
the
model
will
be
a
challenging
task.
Radiotelemetry
studies
may
provide
the
data
that
support
different
input
choices.
A
sensitivity
analysis
should
be
conducted
to
assess
the
importance
that
shorter
time
steps
on
oral,
dermal,
and
inhalation
routes
of
exposure
might
have
on
mortality
estimates.

The
assumption
that
field­
resident
birds
are
off
the
field
during
non­
foraging
periods
if
they
started
the
day
off
the
field
(
and
vice
versa)
is
dubious
for
those
species
that
nest
or
spend
most
of
their
time
on
the
fields.
This
assumption
impacts
exposure
estimates
from
dermal
and
inhalation
routes
of
exposure.
Exposure
via
these
routes
continues
during
non­
foraging
periods,
but
only
if
the
birds
are
on
the
fields.
Time
on
field
should
be
based
on
a
distribution
of
species
weighted
by
their
relative
abundance
per
sampling
effort
(
from
census
data).
Therefore,
as
the
model
sophistication
improves
at
Levels
III
and
IV,
the
Agency
should
add
the
capability
to
use
distributions
for
the
frequency
and
persistence
on
field
variables.

The
use
of
Markov
Chain
models
raised
different
perspectives
by
the
Panel.
The
Markov
Chain
model
should
continue
to
be
run
during
non­
foraging
times
to
account
for
bird
movement
in
and
out
of
fields.
This
information
is
required
to
properly
estimate
dermal
and
inhalation
Page
23
of
61
exposure
during
non­
foraging
times.

Other
Panel
members
believed
that
the
model
could
be
simplified
even
further
by
removing
the
Markov
chain
aspect
of
the
model.
These
Panel
members
believed
that
the
gain
in
simplicity
would
ease
computation
burdens
and
concerns
about
transition
probabilities
affecting
the
output.
It
was
determined
that
the
bimodal
distribution
for
feeding
intensity
could
be
reinterpreted
as
feeding
intensity
on
the
field.
If
the
Markov
chain
model
is
to
be
retained,
then
the
model
should
be
modified
to
account
for
diurnal
variation
in
the
birds'
movements
(
Klein
and
Macdonald,
1980).

One
concern
several
Panel
members
had
regarding
the
nature
of
generating
the
distributions
for
field
persistence
and
transition
probabilities
was
how
the
model
predicted
levels
of
risk
when
known
biological
processes
were
specified.
For
example,
exposure
to
pesticides
may
alter
the
distribution
or
abundance
of
prey
through
time
and
thus
affect
energy
return
during
foraging,
which
in
turn
will
influence
transition
probabilities
(
e.
g.,
low
rate
of
energy
return
would
yield
low
likelihood
of
returning
to
the
foraging
patch).
Additionally,
exposure
to
a
pesticide
also
might
induce
illness
which
would
promote
conditioned
avoidance
of
the
foraging
site
or
inactivity
via
pesticide
induced
anorexia.
There
is
a
need
to
model
whether
a
bird
temporarily
stops
eating
when
exposed
to
pesticide.
Mineau
(
2001)
suggested
that
small
birds
may
be
especially
vulnerable
to
pesticide
exposure
because
they
cease
eating
and
die
from
lack
of
food.

Simulations
based
on
changing
the
distribution
of
field
persistence
to
reflect
well
described
biological
processes
would
prove
instructively
valuable
(
i.
e.,
determining
the
risk
of
mortality
associated
with
various
levels
of
site
fidelity
as
might
be
influenced
by
foraging
return
rates
or
degree
of
illness,
by
implication,
and
degree
of
site
avoidance).

It
is
not
clear
how
residues
on
foods
(
seeds,
foliage
or
animal
matter)
are
determined
or
selected
during
the
modeling
process.
Are
these
fixed
values,
or
are
they
selected
from
a
distribution?
The
latter
would
be
preferred.

Birds
with
territories
larger
than
individual
fields
and
their
edges,
may
also
be
impacted
by
chemical
use
as
they
may
serve
as
integrators
of
broader
contamination
within
agro­
environments
and
suffer
or
benefit
from
chemical
interactions.
The
tools
for
landscape
level
analyses
are
available
and
the
Panel
believed
that
this
will
help
the
Agency
address
issues
of
cumulative
exposures
and
population
impacts
in
higher
tier
analyses.

Agency
Charge
b.)
Please
provide
direction
on
the
appropriate
application
of
the
additional
modifications
and
provide
discussion
on
how
the
modifications
will
improve
the
characterization
of
the
uncertainty
in
risk
estimates.

Panel
Response
Page
24
of
61
A
sensitivity
analysis
should
be
conducted
to
assess
the
importance
that
shorter
time
steps
for
oral,
dermal,
and
inhalation
routes
of
exposure
might
have
on
mortality
estimates.

Birds
ingest
soil
particles,
either
as
grit
or
incidentally.
This
fact
should
be
considered
in
calculating
dosage
estimates.
Processing
rates
for
food
could
be
included
in
refining
on
field
occurrence
and
transitional
probabilities.

Time
on
the
field
should
be
derived
from
radio
telemetry
studies
characterizing
the
distributions
of
temporal
field
use.
The
Panel
recognized
that
such
detailed
data
may
not
be
widely
available.
Thus,
time
on
field
should
be
based
on
a
distribution
of
species
weighted
by
their
relative
abundance
per
sampling
effort
(
from
census
data).

Simulations
for
biologically­
realistic
scenarios
should
be
run
and
compared
to
extant
data
on
behavior
patterns
derived
from
telemetry
or
census
data
to
determine
whether
the
model
adequately
characterizes
field
behavior
of
birds.

The
lower
right
diagram
in
Figure
3­
2
of
the
Agency's
background
document
most
closely
mimics
bird
feeding
behavior.
The
most
abrupt
increase
in
feeding
would
likely
occur
just
after
an
overnight
fast
and
the
most
abrupt
decrease
would
occur
just
before
the
overnight
period.
The
transitions
at
midday
would
be
more
gradual.
When
considering
whether
organisms
are
reproductively
active,
it
should
be
remembered
that
insect
(
pest)
emergence
and
avian
reproductive
cycles
are
closely
tied.
In
evaluating
insecticides,
reproductively­
active
birds
should
be
assumed
The
Agency's
background
document
indicated
that
the
beginning
and
ending
times
of
both
the
morning
and
afternoon
feeding
periods
are
assumed
to
vary
randomly
each
day,
within
specified
time
windows
(
p
11).
This
is
not
likely
the
case.
Beginning
and
ending
feeding
periods
are
influenced
strongly
by
photoperiod
which
does
not
vary
randomly.

Current
practice
is
to
use
Fletcher's
(
1994)
summary
statistics
of
mean
and
standard
deviation
to
develop
distributions
of
plant
residues
for
model
input.
The
Panel
noted
that
environmental
datasets
for
items
like
residues
often
are
not
normally
distributed
and
the
use
of
mean
and
standard
deviation
introduces
bias
into
the
distributions
generated
by
such
statistics.
ECOFRAM
attempted
to
assemble
Fletcher's
original
datasets
as
well
as
those
for
insects.
Because
these
inputs
are
basic
to
Model
Version
2
risk
analysis,
the
Panel
believed
it
would
be
beneficial
for
the
Agency
to
complete
this
process.
The
Agency
might
also
consider
using
food
datasets
directly
to
eliminate
a
potential
bias
when
a
normal
distribution
(
or
some
other
distribution)
is
assumed,
or
at
least
re­
evaluate
the
assumption
of
normal
distribution
by
thorough
re­
examination
of
the
data.

Regarding
dissipation
rate
constants,
it
was
noted
that
linear
regression
was
used
to
generate
first­
order
rate
constants.
Most
often
with
dissipation
datasets,
the
use
of
first­
order
Page
25
of
61
linear
regression
does
not
provide
a
good
fit
to
observation(
s).
The
use
of
linear
regression
produces
bias
that
can
largely
be
removed
through
the
use
of
nonlinear
regression.
Nonlinear
regression
provides
a
much
better
fit,
and
it
is
suggested
that
it
be
used
instead
to
capture
more
adequately
the
observed
patterns
of
dissipation.

There
also
appeared
to
be
use
of
mean
half­
lives
calculated
from
multiple
half­
life
values.
If
means
are
used
as
input,
then
it
is
suggested
rate
constants
from
the
regressions
be
averaged
first
and
then
calculate
the
corresponding
half­
life
if
desired.
Performing
the
reverse
(
averaging
the
half­
lives
and
then
calculating
rate
constant)
does
not
generate
the
same
value
as
averaging
of
the
rate
constants
themselves.
Because
it
is
the
rate
constant
that
is
derived
from
the
regression,
it
would
be
the
more
appropriate
metric
for
averaging.
As
in
the
discussion
on
food
residues,
rate
constants
in
nature
are
far
from
being
constant
and
are
not
necessarily
portrayed
well
by
normal
distributions.
The
Agency
might
consider
using
rate
constant
datasets
directly
rather
than
assume
a
normal
distribution
via
the
use
of
means
and
standard
deviations.

Agency
Charge
4.
New
Puddle
Algorithm.
A
new
puddle
algorithm
was
developed
to
account
for
a
number
of
parameters
that
affect
puddling
after
a
rainfall
event
in
agro­
environments.
The
new
algorithm
addresses
rainfall
amount,
rainfall
duration,
soil
infiltration
rates,
evaporation,
degradation
and
the
stochastic
nature
of
field
topography
and
its
relation
to
puddle
formation
and
duration.

a.)
Please
comment
on
the
overall
model
structure
in
relation
to
mimicking
puddles
in
agro­
environments,
including
any
suggestions
on
modifications
or
additional
parameters
to
considered
that
would
improve
pesticide
concentration
estimates
in
this
environmental
media.
b.)
Please
provide
suggestions
for
assigning
values
to
puddle
input
variables
and
for
locating
additional
sources
of
information
that
may
help
in
defining
these
values.

Panel
Response
Because
the
Panel's
responses
overlap
the
issues
presented
in
questions
A
and
B,
the
questions
are
answered
as
noted
below.
The
Agency
had
the
difficult
task
of
further
refining
a
puddle
algorithm
to
account
for
a
number
of
parameters
that
affect
puddling,
primarily
during
rainfall
events
in
agro­
environments.
The
improvements
were
based
on
the
recommendation
of
the
SAP
in
2001
(
FIFRA
Scientific
Advisory
Panel,
2001)
who
indicated
that
although
the
Version
1.0
approach
for
modeling
avian
exposure
through
drinking
water
seemed
reasonable,
the
puddle
scenario
should
include
consideration
of
amount
and
duration
of
rainfall,
evaporation,
soil
infiltration
rates,
plant
cover,
temperature,
chemical
partitioning,
chemical
degradation,
and
topography.

The
main
purpose
of
the
puddle
algorithm
is
to
simulate
the
quality
of
the
drinking
water
Page
26
of
61
for
birds.
Specifically,
this
routine
calculates
both
the
puddle
duration
and
the
pesticide
concentration
in
a
puddle
on
the
field
deterministically.
The
proposed
algorithm
addresses
amount
and
duration
of
rainfall,
soil
infiltration
rates,
evaporation,
degradation
and
the
stochastic
nature
of
field
topography
and
its
relation
to
puddle
formation
and
duration.

The
algorithm
assumes
that
there
is
only
one
puddle
in
a
field.
During
times
that
the
puddle
is
dry,
the
birds
are
assumed
to
drink
non­
contaminated
water.
Pesticide
mass
balances
are
calculated
separately
for
the
field
and
for
the
puddle.
The
simulation
distinguishes
periods
with
and
without
rain.
For
periods
without
rain,
the
puddle
algorithm
calculates
separate
pesticide
mass
balances
in
a
mixing
zone
in
the
field
and
in
the
puddle,
assuming
that
pesticide
dissipation
follows
a
first­
order
reaction
rate.
During
runoff­
producing
rainfalls,
water
from
the
field
runs
into
the
puddle.
The
algorithm
for
calculating
the
pesticide
concentration
assumes
that
the
concentration
in
the
mixing
zone,
runoff
and
percolation
is
the
same
at
any
time.
After
the
rainfall,
the
puddle
diminishes
due
to
evaporation
and
infiltration
into
the
soil.

The
Panel
found
that
the
mathematical
structure
of
the
puddle
algorithm
was
appropriately
simple
for
Level
II
where
only
readily
available
data
are
being
used.
Later
in
this
section,
suggestions
are
made
that
could
decrease
the
overall
running
time
of
the
algorithm.
This
will
then
allow
for
including
a
stochastic
representation
of
the
process.

The
Panel's
comments
on
the
puddle
algorithm
are
divided
into
three
sections.
First
the
scenarios
and
the
physical
situation
modeled
are
discussed,
followed
by
possible
changes
in
the
algorithm
that
may
be
considered
by
EPA
for
possible
inclusion.
Finally
the
use
of
the
puddle
algorithm
within
the
overall
avian
risk
assessment
is
discussed.

Scenarios
and
Physical
Situation
Modeled
The
current
scenario
of
simulating
only
one
rainfall
event
sometime
after
application
was
seen
by
several
Panel
members
as
a
limitation.
Because
rainfall
only
occurs
after
application,
the
algorithm
does
not
include
direct
overspray
of
existing
puddles
because
it
rains
only
subsequent
to
spray
events.
This
will
significantly
underestimate
risk
because
initial
concentrations
are
likely
greatest
in
shallow
waters
immediately
after
direct
overspray.
Model
assumptions
of
instantaneous
equilibrium
of
the
pesticide
concentration
between
sediment
and
water
will
likely
underestimate
risk
from
over­
sprayed
puddles
even
if
the
model
includes
these
puddles
at
the
time
of
spraying.
Additionally,
the
model
does
not
include
capture
by
foliage
and
assigns
all
pesticides
to
the
soil
itself.
Although
this
initially
may
appear
to
be
a
conservative
assumption,
it
is
not
if
the
pesticide
that
is
washed
off
the
leaves
is
not
immediately
adsorbed
by
the
soil.
Because
the
time
that
equilibrium
is
reached
is
not
in
the
current
data
base,
a
conservative
assumption
would
be
that
there
is
no
adsorption
for
a
period
directly
after
spraying.
The
duration
of
this
period
should
be
based
on
actual
field
experiments
(
Mellott
et
al.,
1990;
Brewer
et
al.,
1990).

The
Panel
generally
disagreed
with
the
scenario
of
a
single
puddle
in
the
field
and
puddle
sizes
that
did
not
discriminate
between
different
parts
of
the
country.
There
are
many
different
Page
27
of
61
size
puddles
in
a
crop
field.
The
shape
and
size
of
puddles
may
vary
with
crops,
geographic
locations,
agricultural
management
practices,
and
many
other
factors.
Although
the
concentration
likely
will
not
be
much
different
between
the
puddles,
the
duration
of
water
standing
in
the
puddles
will
vary
greatly
between
the
different
sizes.
In
terms
of
modeling
of
multiple
puddles,
another
important
issue
is
how
to
deal
with
the
interactions
among
puddles
in
the
field,
which
include
both
water
quantity
and
quality
interactions.
To
deal
with
a
field
scale
problem,
the
Agency
may
need
to
account
for
the
hydrodynamics
and
pesticide
fate
and
transport
in
a
number
of
puddles
of
varying
sizes,
and,
in
particular,
their
interactions.
By
all
means,
the
relationship
of
the
puddle
(
or
puddles)
and
the
field
should
be
considered
in
the
model.

Daily
rainfall
is
currently
used
instead
of
shorter
time
steps.
Intense
rainfall
of
a
given
amount
over
short
periods
will
run
off
much
more
than
will
light
rains
of
the
same
total
amount
spread
over
a
24­
hour
period.
This
issue
raises
the
point
that
multiple
rainfall
events
should
also
be
modeled.
Longer
modeled
time
intervals
and
multiple
rain
events
are
supported
by
data
from
field
studies
in
which:
1)
dormant
spraying
produced
a
bimodal
mortality
that
occurred
with
one
maximum
in
the
first
few
days
following
application
and
a
second
over
two
weeks
later
(
Cobb
and
Hooper,
1995;
Cobb
et
al.,
2000;
Mellott
et
al.,
1990)
and
2)
multiple
rainfall
events
soon
after
pesticide
application
coincided
with
avian
mortality
(
Brewer
et
al.
1990;
Brewer
et
al,
1992;
Mellott
et
al.,
1990,
Tank
et
al.,
1992).

To
discriminate
between
the
formation
of
puddles
in
different
regions,
knowledge
of
the
hydrology
for
the
particular
regions
and
field
are
important.
For
many
years,
it
was
believed
that
the
occurrence
of
surface
runoff
was
primarily
controlled
by
the
infiltration
characteristics
of
the
ground.
Specifically,
it
was
thought
that
runoff
was
generated
whenever
rainfall
(
or
irrigation)
occurs
at
a
greater
rate
than
the
soil's
infiltration
capacity.
This
is
termed
either
"
Hortonian
runoff"
or
"
infiltration
excess
overland
flow".
This
process
is
very
important
in
many
areas
of
the
country
on
bare
fields
with
little
organic
matter
(
where
significant
soil
crusting
and/
or
surface
sealing
occur
during
rain
events).
In
these
areas,
puddles
form
in
depressions
filled
by
runoff
from
the
surrounding
area.

However,
the
Hortonian
runoff
concept
does
not
meaningfully
explain
storm
runoff
in
many
of
the
humid
regions
of
the
US,
where
the
infiltration
capacity
of
the
ground
is
typically
much
greater
than
average
rainfall
intensities.
Many
researchers
have
found
that
the
typical
values
published
in
soil
surveys
for
disturbed
samples
underestimate
the
true
field
conductivity
of
vegetated
soils
by
a
factor
of
10
or
more
due
to
the
presence
of
preferential
flow
paths
in
the
form
of
worm
channels
and
root
passages
(
Steenhuis
et
al,
1994;
Walter
et
al.,
2003).
In
these
regions,
runoff
is
most
commonly
generated
on
relatively
small
portions
of
the
landscape
that
are
susceptible
to
becoming
completely
saturated
(
i.
e.,
the
puddle
areas).
This
runoff
type
is
called
saturation
excess
overland
flow.
Areas
prone
to
saturation
have
either
a
high
ground
water
table
or
a
hardpan
(
fragipan)
at
shallow
depth.
Interflow
in
concave
hillslopes
can
also
result
in
the
formation
of
saturated
areas
at
the
bottom
of
slopes,
seeps
and
ditches.
Puddles
in
these
areas
can
be
persistent
during
times
of
the
year
when
precipitation
exceeds
the
potential
evaporation.
In
this
case,
the
puddles
are
the
source
of
the
runoff
and
do
not
necessarily
have
run­
on
from
Page
28
of
61
surrounding
areas.
As
rainfall
continues,
the
saturated
area
grows
in
extent,
increasing
the
area
generating
runoff
(
hence
the
term
variable
source
area,
VSA).
Examples
of
regions
in
the
USA
where
VSA
hydrology
is
significant
include
the
Northeast
and
the
Pacific
Northwest
as
well
as
forested
mountain
areas
in
the
US.
The
Panel
recommended
that
the
Agency
take
a
regional
approach
to
puddle
formation.
Experts
in
each
of
the
regions
should
give
advice
on
the
puddle
formation
characteristic.
Simplifying
the
pesticide
routine
will
allow
the
Agency
to
model
several
puddles
on
the
field,
all
with
the
same
concentration
but
with
different
depths
and
duration.
As
a
starting
point
for
further
evaluation
by
the
Agency,
Onstad
(
1964)
might
have
information
about
puddle
characteristics.

In
the
current
level
II
model,
the
puddle
algorithm
is
entirely
deterministic.
One
rainfall
event
is
considered
that
occurs
sometime
after
pesticide
application.
The
time
and
occurrence
are
input
parameters.
If
this
rainfall
event
produces
runoff,
one
puddle
fills
up
in
the
middle
of
the
field.
The
entirely
deterministic
nature
of
the
model
was
questioned
by
the
majority
of
the
panel
members
because
of
the
numerous
sources
of
uncertainty
as
input
parameters
(
e.
g.,
mixing
zone
depth
parameter,
contributing
area,
puddle
size,
rainfall
amount
and
duration).
One
Panel
member
noted
that
using
conservative
input
values
or
assumptions
will
lead
to
extreme
predictions
(
due
to
compounding
conservatism),
particularly
given
the
number
of
variables
required
by
the
puddle
model.
Other
panel
members
noted
that
large
storms
and
multiple
rainstorms
should
be
considered.

Selecting
and
parameterizing
distributions
for
all
of
the
variables
in
the
puddle
model
would
be
an
onerous
task.
To
deal
with
this
issue,
sensitivity
and
elasticity
analyses
should
be
employed
to
identify
those
variables
that
have
the
greatest
influence
on
model
predictions
when
uncertainty
and
variability
are
considered.
Only
the
most
influential
variables
need
to
be
treated
as
distributions
in
future
iterations
of
the
model.
This
will
be
discussed
in
the
following
sections.

Model
Structure
One
of
the
basic
underlying
principles
of
the
models
at
Level
II
was
that
only
easily
available
data
are
required
as
model
inputs.
Therefore,
the
use
of
first­
order
degradation
rates
and
linear
adsorption
partition
coefficients
are
appropriate
despite
the
fact
that
there
are
more
sophisticated
formulations
available
to
describe
degradation
and
adsorption.
These
assumptions
may
be
too
conservative
only
in
the
case
of
direct
overspray
onto
a
puddle.

Periods
without
rainfall
The
mass
of
the
pesticide
disappears
according
to
a
first­
order
degradation
rate
for
the
pesticide
adsorbed
on
the
soil
and
in
the
water.
The
inclusion
of
separate
degradation
rates
for
sediment
and
soil
is
interesting
but
the
data
are
not
likely
available.
Assuming
a
single
overall
degradation
rate,
the
amount
of
pesticides
in
the
mixing
layer
is:

(
1)
Page
29
of
61
(
)
e
mix
mix
t
M
M
µ
 
=
exp
,
0
,
1
where
M1,
mix
is
the
pesticide
mass
in
the
mixing
zone
at
the
time
the
storms
starts,
M0,
mix
is
the
initial
mass
in
the
mixing
zone
at
time
t=
0
,
te
is
time
when
the
rainfall
starts,
µ
is
the
overall
firstorder
degradation
rate
in
the
mixing
zone.
Several
panel
members
thought
that
additional
pesticide
removal
mechanisms
such
as
volatilization
and/
or
photodegradation
should
be
taken
into
account.
Note
that
the
units
for
the
mass
are
kg/
field
or
better
kg
per
unit
area.
Here
we
will
use
the
mass
per
unit
area
because
that
it
is
the
more
standard
way
of
reporting
the
equation.
We
also
will
use
metric
units
throughout.

Periods
with
rainfall
To
model
the
dynamics
of
the
puddle
algorithm
during
periods
with
rainfall,
both
hydrology
and
pesticide
transport
contaminant
hydrology
need
to
be
considered.

Hydrology
Understanding
the
hydrologic
basis
for
water
and
contaminant
transport
is
essential
for
meaningful
monitoring
of
surface
and
subsurface
water
quality.
The
considerations
of
hydrology
by
the
Agency
seem
to
be
based
on
the
conventional
"
infiltration
excess
overland
flow"
because
all
parts
of
the
landscape
are
participating
in
the
runoff
process.
As
mentioned
above,
in
some
areas
saturation
excess
overland
flow
(
where
only
part
of
the
landscape
is
participating)
is
more
common.
However,
because
infiltration
excess
is
more
conservative,
saturation
excess
and
other
runoff
mechanisms
do
not
have
to
be
included
in
Level
II
analysis.

According
to
the
model
documentation,
runoff
starts
as
soon
as
the
rainfall
starts.
This
is
unnecessarily
conservative.
As
can
be
seen
from
the
SCS
equation
(
i.
e.,
runoff
starts
only
when
p>
Ia),
viz:

(
2)

which
is
equally
valid
for
the
English
and
metric
systems.
Thus
q
is
the
cumulative
amount
of
runoff
(
cm),
p
is
cumulative
amount
of
rainfall
(
cm),
S
is
storage
of
the
watershed.
Ia
is
initial
abstraction
or
the
amount
of
rainfall
that
occurs
before
runoff
can
start.
There
is
much
current
debate
on
how
to
calculate
the
initial
abstraction,
Ia,
and
whether
it
is
equal
to
0.2
S,
another
fraction
of
S,
or
completely
independent
of
it
and
should
instead
be
calculated
with,
for
example,
(
)

a
a
a
I
p
S
I
p
I
p
q
 
+
 
 
=
2
Page
30
of
61
the
Thornthwaite
Mather
(
1955)
procedure
(
Steenhuis
and
van
der
Molen,
1986).
The
Agency
should
consider
making
this
equation
stochastic
as
presented
later
in
the
Agency's
presentation
describing
aquatic
risk
assessment.

Independent
of
the
method
chosen
to
calculate
Ia
a
certain
portion
of
the
rainfall
infiltrates
before
runoff
occurs.
Assuming
steady­
state
rainfall
using
the
metric
system,
the
time
that
runoff
starts
after
initiation
of
rainfall,
tr,
is:

(
3)

t
I
p
T
r
a
=

where
T
is
the
storm
duration.

The
runoff
rate
Q
per
unit
area
is
then
(
again
assuming
steady
state)

(
4)

Q
q
T
t
e
=
 

The
assumption
of
steady
state
rainfall
and
runoff
rates
needs
to
be
tested.
The
2001
SAP
and
one
Panel
member
of
the
current
Panel
indicated
that
they
likely
wanted
to
include
a
realistic
rainfall
distribution,
but
it
is
doubtful
(
even
if
the
rainfall
distribution
is
known)
if
this
refinement
is
necessary
at
Level
II.
As
will
be
described
later,
it
is
the
cumulative
amount
of
rainfall
that
drives
the
pesticide
losses
during
the
rainfall
event.
The
degradation
of
the
pesticide
(
which
is
dependent
on
time)
during
the
rainfall/
runoff
is
likely
only
a
small
part
of
the
total
loss.

Contaminant
Hydrology
The
contaminant
hydrology
calculates
the
mass
of
pesticide
in
the
field
differently
from
that
in
the
puddle.

Mass
Of
Pesticide
On
The
Field
By
assuming
that
the
concentration
of
the
pesticide
in
the
mixing
zone,
percolation
water
and
the
runoff
are
identical,
the
amount
of
pesticide
in
the
mixing
zone
can
be
simply
written
as
a
function
of
the
total
rain
(
Eq.
4)
and
as
reported
correctly
in
chapter
3
of
the
Agency's
background
document
as
Eq
3­
30.

(
5)

(
)
[
]
M
M
k
t
t
mix
mix
field
e
=
 
 
1,
exp
Page
31
of
61
Note
that
kmix
and
later
renamed
to
kfield
in
the
document
is
equal
in
its
most
elementary
form
to
kfield=(
µ
+
p/
W)
where
µ
is
the
first­
order
decay
rate
of
the
pesticide,
W
is
the
apparent
water
content
per
unit
area
of
the
mixing
zone,
and
W
=
H(
 s
+
 Kd)
where
H
is
the
depth
of
the
mixing
zone
 s=
the
saturated
moisture
content,
 
is
the
bulk
density
of
the
soil
and
Kd
is
the
adsorption
partition
coefficient.
Equation
5
can
be
rewritten
as:

(
6)

(
)






 
 
=
W
p
t
M
M
mix
mix
exp
exp
,
0
µ
In
addition,
note
that
evaporation
is
neglected
during
the
rainstorm
(
which
is
fine).
It
should
therefore
be
taken
out
of
the
mass
balance
of
the
water
for
the
mixing
zone
in
Eq.
3­
32
in
the
Agency's
background
document.

For
calculating
the
mass
of
the
pesticides
in
Eq
6,
the
two
important
parameters
for
a
given
pesticide
and
mixing
depth
are
the
cumulative
amount
of
rainfall
and
the
amount
of
degradation
after
application.
Eq
6
is
insensitive
to
how
much
water
infiltrates
and
runs
off.
This
is
not
a
deficiency
in
the
field
model.
It
is
important
to
acknowledge
this
as
it
could
be
a
significant
simplification
in
the
models.
However,
for
calculation
of
the
mass
of
pesticides
in
the
runoff,
the
hydrology
is
important
because
it
depends
on
the
amount
of
runoff.
However,
unlike
runoff
models
in
which
we
are
interested
in
the
mass
of
pesticide
leaving
the
field,
for
avian
risk
assessment
we
are
only
interested
in
the
pesticide
concentration
in
the
puddles.

To
calculate
the
mass
of
pesticide
in
the
puddle,
the
current
pesticide
algorithm
assumes
that
overland
flow
containing
dissolved
pesticides
runs
into
the
puddle.
Soil
loss
is
neglected.
The
puddle
fills
up
and
then
the
water
and
dissolved
pesticides
run
out
of
the
puddle.
Degradation
occurs
as
well.
Inherent
assumptions
(
not
stated
explicitly)
in
the
derivations
are
that
the
puddle
fills
up
but
does
not
grow
larger
in
time.
There
are
steady
state
water
fluxes;
runoff
starts
when
the
rainfall
begins;
and
there
are
equilibrium
conditions
for
pesticide
concentration
in
soil
and
water.

Neglecting
soil
loss
poses
a
limitation
that
only
a
pesticide
with
an
adsorption
partition
of
approximately
less
than
10
cm3/
g
can
be
simulated
realistically.
For
others
with
higher
adsorption
partition
values,
the
sediment
in
the
runoff
can
contribute
a
significant
part
of
the
pesticide
mass
in
the
puddle.
However,
the
concentration
in
the
water
might
not
be
affected
because
the
algorithm
only
considered
a
small
mixing
zone
and
soil
that
is
moved
has
the
same
concentration
as
present
in
the
puddle.

In
order
to
calculate
the
mass
of
pesticide
in
the
puddle,
the
current
pesticide
algorithm
assumes
that
overland
flow
containing
dissolved
pesticides
runs
into
the
puddle.
The
algorithm
only
considered
a
small
mixing
zone
and
soil
that
is
moved
during
runoff
contains
the
same
Page
32
of
61
pesticide
concentration
as
present
in
the
puddle.
Furthermore,
soil
loss
is
neglected.
The
puddle
fills
up
and
then
the
water
and
dissolved
pesticides
run
out
of
the
puddle.
Degradation
occurs
as
well.
Inherent
assumptions
(
not
stated
explicitly)
in
the
derivations
are
that
the
puddle
fills
up
but
does
not
grow
larger
in
time.
There
are
steady
state
water
fluxes;
runoff
starts
when
the
rainfall
begins;
and
there
are
equilibrium
conditions
for
pesticide
concentration
in
soil
and
water.
The
Panel
suggested
that
the
Agency
could
follow
a
simpler
approach.
First,
the
runoff
should
start
after
the
initial
abstraction
is
met.
Thus,
there
is
no
water
in
the
puddle
until
p>
Ia.
The
concentration
in
the
mixing
zone
at
the
time
runoff
starts
can
be
derived
directly
from
Eq.
6
as
follows:

After
the
rainfall
amount
is
greater
than
the
initial
abstraction
the
concentration
in
the
runoff
can
be
written
for
P>
Ia
as:

(
7)

(
)






 
 






 
 
=
W
I
p
W
I
t
W
M
C
a
a
mix
mix
exp
exp
exp
,
0
µ
which
reduces
again
to
Eq
6.

If
the
puddle
size
is
small
compared
to
the
field,
the
dilution
by
rainfall
of
the
water
falling
in
the
puddle
is
small
compared
to
the
overall
contribution
of
the
pesticide
in
the
runoff.
Therefore
the
simplifying
assumption
can
be
made
that
the
pesticide
concentration
in
the
puddle
is
the
same
as
the
pesticide
concentration
in
the
mixing
zone
on
the
field
(
i.
e.,
Eq
6).
This
will
greatly
speed
up
the
calculation
and
makes
the
calculation
of
the
puddle
contributing
areas
unnecessary.

After
the
rain
stops,
the
increase
in
concentration
due
to
evaporation
might
be
important
for
cases
when
the
infiltration
is
much
smaller
than
the
evaporation
and
the
pesticide
has
a
long
half
life.
Volatilization
and
photodegradation
might
be
considered
too.
Because
aerobic
aquatic
studies
are
performed
in
the
dark,
it
is
suggested
that
photolysis
rates
be
added
to
the
aerobic
aquatic
rates
to
approximate
what
takes
place
in
the
puddles.
While
most
Panel
members
agreed,
a
few
differed
adding
that
the
addition
of
these
rate
constants
would
double
the
hydrolysis
contribution,
which
is
a
component
of
both
photolysis
and
aerobic
aquatic
degradation.
Conversely,
as
toxicant
is
degraded,
there
is
no
allowance
for
toxicity
neutral
degradation.
In
higher
tiers,
the
Agency
may
need
to
consider
transformation
products
that
are
of
equal
or
higher
toxicity
to
the
parent
compound.

If
a
new
rainfall
event
occurs,
the
pesticide
overland
flow
in
the
puddle
soon
will
overwhelm
any
differences
before
the
rain
started.
Thus,
there
is
an
assumption
that
the
pesticide
concentration
in
the
puddle
is
the
same
as
in
the
runoff
from
the
rest
of
the
field.
The
Agency
should
set
a
limit
whereby
evaporation
cannot
give
a
concentration
higher
than
the
water
solubility
of
the
chemical.
It
is
possible
that
the
Kd
will
accomplish
this,
but
the
model
should
be
checked
to
make
sure
that
at
low
puddle
volumes
the
model
does
not
allow
the
aqueous
pesticide
Page
33
of
61
concentration
to
exceed
the
solubility.

Another
possibility
suggested
by
other
Panel
members
is
to
simulate
the
rainfall
and
soil
properties
stochastically.
The
Agency
might
consider
that
the
extent
of
puddling
is
a
function
of
the
rainfall
amount.
For
example
Steenhuis
et
al
(
1995)
derived
the
extent
of
the
fraction
of
the
field
within
the
puddle
area.

Before
time
is
invested
in
the
additions
suggested,
a
sensitivity
analysis
should
be
performed
to
determine
if
having
multiple
puddles
changes
the
overall
risk
prediction
significantly.
Also
the
data
might
not
be
available
under
Level
II.
If
there
is
no
or
minimal
effect,
then
the
modifications
suggested
may
be
omitted.
It
would
be
informative
to
present
to
future
peer
review
panels
any
data
demonstrating
lack
of
model
sensitivity
to
given
parameters.

Puddle
algorithm
and
the
overall
avian
risk
assessment
Overall,
the
text
describing
the
puddle
model
is
incomplete
because
it
does
not
describe
how
birds
will
sample
the
puddles.
The
Agency
indicated
that
if
puddles
are
on
the
field,
birds
obtain
drinking
water
from
this
source
during
the
last
hour
of
the
feeding
period.
Otherwise,
birds
obtain
necessary
free
water
from
dew
in
the
first
feeding
hour
of
the
day.
This
information
should
be
added
to
the
text.
It
is
also
not
clear
how
birds
choose
from
multiple
puddles
on
the
field
(
which
presumably
have
different
concentrations).
A
random
walk
model
could
be
used
to
simulate
puddle
selection
over
time,
but
this
is
more
likely
a
Level
III
approach.
A
simpler
approach
is
likely
required
for
Level
II.
Whatever
the
approach,
it
needs
to
be
described.
It
was
noted
that
the
number
and
size
of
puddles
may
influence
wildlife
use
(
songbirds
vs.
waterfowl).

One
Panel
member
noted
that
other
exposure
pathways
might
include
birds
bathing
in
the
puddles
and
then
preening,
which
is
a
significant
activity,
or
simple
skin
absorption
from
bathing.

The
Level
II
model
includes
several
routes
of
exposure
(
i.
e.,
dermal,
inhalation,
drinking
water
ingestion
from
puddles)
that
are
not
included
in
the
Level
1
model.
For
pesticides
in
which
these
routes
of
exposure
are
important,
the
ostensibly
less
conservative
Level
II
model
could
lead
to
higher
risk
estimates
than
the
Level
I
model.
This
situation
needs
to
be
rectified
by
adding
the
dermal,
inhalation
and
puddle
routes
of
exposure
to
the
Level
I
model.

The
SCS­
CN
method
is
used
for
runoff
simulation
in
the
new
puddle
model.
Also,
the
simulated
runoff
from
a
certain
"
effective
area"
is
assumed
to
fill
the
simulated
puddle.
Conceptually,
using
the
SCS­
CN
method
in
such
a
way
is
inconsistent
with
the
original
methodologies/
assumptions
and
thus
can
be
questionable.
First
of
all,
there
are
many
puddles
of
varying
sizes
in
a
crop
field.
The
area
of
puddles
is
an
essential
portion
of
the
field
domain.
According
to
the
definition
of
the
initial
abstraction
(
Ia)
in
the
SCS­
CN
method,
it
includes
surface
depression
abstraction,
infiltration,
and
evaporation.
This
means
the
initial
abstraction
has
already
included
the
water
that
fills
puddles.
The
SCS­
CN
method
also
implies
that
all
surface
depressions
Page
34
of
61
(
puddles)
have
been
fully
filled
when
the
cumulative
rainfall
is
greater
than
the
initial
abstraction
and
hence
runoff
starts.
In
addition,
the
simulated
runoff
is
the
water
that
should
leave
the
field
and
flow
towards
an
outlet
through
stream
channels.
Thus,
the
lumped
SCS­
CN
method
cannot
provide
any
local
overland
flow
information
within
the
field
and
hence
does
not
allow
us
to
examine
how
a
puddle
is
filled.
The
runoff
simulated
by
using
this
method
cannot
be
used
to
fill
puddles.
The
Agency
may
need
to
consider
these
critical
issues
during
the
model
modification.

Agency
Charge
5.
Air
Concentration
Estimation.
The
model
currently
employs
an
equilibrium­
based
two
compartmental
model,
for
estimating
pesticide
air
concentration
in
the
plant
canopy.
Please
comment
on
the
merits
and
limitations
of
this
approaches.
Would
the
SAP
provide
suggestions
on
additional
alternatives
for
estimating
vapor
phase
concentrations
that
would
be
consistent
with
the
physical/
chemical
property
and
environmental
fate
data
available
to
the
Agency
as
guideline
information?
Please
comment
on
the
merits
and
limitations
of
these
additional
approaches.

Panel
Response
The
Agency's
efforts
to
modify
the
existing
PRZM
algorithms
are
laudable.
While
the
Panel
believed
that
a
simple
model
is
the
preferable
approach,
the
extant
model
will
benefit
from
the
following:

(
1)
Consideration
of
air­
soil
interaction..
(
2)
Consideration
of
interaction
of
canopy
air
with
the
air
above
the
canopy.
Both
of
these
interactions
can
be
easily
incorporated.
(
3)
Expand
the
model
to
allow
evaluation
of
granular
products.
(
4)
Expand
the
air
modeling
and
associated
inhalation
exposure
calculations
to
include
edge
habitat.

Even
though
soil
volatility
is
not
considered
in
the
model,
Equation
E1­
8
on
P7
of
Appendix
E
of
the
Agency's
background
document
contains
an
expression
for
volatility
from
soil.
This
equation
or
some
approach
must
be
employed
in
the
vaporization
module
of
the
model.
Also,
PRZM
is
a
root
zone
model
designed
to
predict
runoff
and
not
to
predict
pesticide
inhalation.
Therefore,
the
PRZM
algorithms
that
used
mean
canopy
assumptions
need
to
be
modified,
but
using
the
PRZM
basis
algorithms
for
pesticide
diffusion
from
soil
into
air
is
a
good
starting
point.

An
alternative
to
the
equations
presented
by
the
Agency
is
presented
in
the
following
equations.
Beginning
with
the
Agency's
approach
in
Equation
3­
51.

)
51
3
(

)
1000
(
,
 













+
=
=

plant
vol
plant
air
plant
plant
air
eq
ca
B
m
V
m
C
C
C
 
Page
35
of
61
Figure
2.
Possible
Intermediate
Complexity
Model
for
Contaminant
Movement
Among
the
Soil
Plant
and
Air
Compartments
Considered
within
the
FIFRA
Level
II
Probabilistic
Risk
Assessment.

The
mass
balance
from
Figure
2
follows.

and
has
the
following
steady
state
solution:
Ca
ka,
ca
h
p
Cca,
eq
kp,
ca
Cca
ks,
ca
Csa
Cca
=
concentration
in
canopy
air
Ca
=
concentration
in
ambient
air
Csa
=
concentration
in
soil
air
Cca,
eq
=
concentration
in
canopy
air
in
equilibrium
with
plant
ki,
ca
=
effective
mass
transfer
rate
[
T­
1]


Computed
from
mass
transfer
rate
coefficient,
kL,
i,
ca
[
L
T­
1],
e.
g.,
=
Da/
 i
kp,
ca
=
kL,
p,
ca
*
Ap/
h
ka,
ca
=
kL,
a,
ca
*
Ac/
h
ksa,
ca
=
kL,
sa,
ca
*
Ac a/
h
(
)
(
)
(
)
ca
sa
ca
sa
ca
a
ca
a
ca
eq
ca
ca
p
ca
C
C
k
C
C
k
C
C
k
dt
dC
 
+
 
+
 
=

,
,
,
,
Page
36
of
61
ca
p
ca
sa
ca
p
ca
a
sa
ca
p
ca
sa
a
ca
p
ca
a
eq
ca
ca
sa
ca
a
ca
p
sa
ca
sa
a
ca
a
eq
ca
ca
p
ca
k
k
k
k
C
k
k
C
k
k
C
k
k
k
C
k
C
k
C
k
C
,
,

,
,
,
,

,
,
,
,
,
,
,
,
,
,

1
+
+
+
+
=
+
+
+
+
=
Page
37
of
61
This
approach
allows
simple
solutions
if
rate
constants
are
known
or
can
be
approximated,
as
seen
in
the
following
example.

When
considering
pesticide
evolution
from
soils,
it
is
also
reasonably
well
known
that
when
aerosols
move
through
vegetative
canopies,
the
canopies
serve
more
as
initial
sinks
then
serve
as
sources
at
later
time
steps.
Several
studies
provide
some
information
regarding
plant
interaction
with
volatilized
pollutants
(
McLachlan
and
Horstmann,
1998;
Reiderer
1990;
Smith
and
Thomas,
2001;
Mueller
and
Hawker,
1994).
It
is
entirely
possible
that
the
lower
surfaces
of
lower­
level
vegetation
can
adsorb
pesticides
early
in
the
flux
event
and
then
release
these
pesticides
slowly.
This
could
provide
a
time­
dependent
bimodal
flux
within
the
canopy.
This
type
of
phenomenon
would
explain
the
empirical
data
at
lower
heights
(
Chart
3B
Appendix
E,
p.
32
of
the
Agency's
background
document).
More
empirical
data
are
needed
especially
at
lower
heights
to
allow
evaluation
of
the
need
for
such
refinement
in
the
model.
This
is
because
chemical
behavior
higher
in
the
canopy
is
more
thoroughly
described
in
the
literature.

The
acceptance
of
air/
soil
partitioning
as
a
mainstream
research
area
of
environmental
chemistry
has
facilitated
a
reasonable
body
of
literature
containing
empirical
data
for
toxicant
volatilization
from
soils.
Most
of
these
data
for
semivolatiles
indicate
reasonable
correlation
of
volatilization
with
soil
organic
matter.

Concentrations
of
fumigants
in
air
from
fumigant
application
to
soil
have
been
modeled
(
Cryer,
van
Wesenbeeck
and
Knuteson
2003;
Cryer
and
van
Wesenbeeck
2001).
This
could
be
used
to
evaluate
the
final
model
designed
by
the
Agency.

There
is
no
degradation
kinetic
term
in
the
atmospheric
model.
Photolytic
degradation
should
be
evaluated
to
see
if
such
terms
are
likely
to
matter.
If
degradation
does
matter
and
conservatism
is
desired
in
Level
II,
degradation
can
be
omitted
until
the
Level
III
refinement.

A
refinement
that
should
be
considered
in
higher
levels
is
that
gaseous
diffusion
(
Da)
is
55
4
100
*
1
10
*
2
100
1
1
2
100
10
100
,
,

,
,
,
,

,
,
,
,
,

,
,
,

=
+
+
=
+
+
+
+
=
=
=
=
=
=

ca
p
ca
sa
ca
p
ca
a
sa
ca
p
ca
sa
a
ca
p
ca
a
eq
ca
ca
p
ca
sa
ca
p
ca
a
sa
a
eq
ca
k
k
k
k
C
k
k
C
k
k
C
k
k
k
k
C
C
C
Page
38
of
61
not
dependent
solely
on
molecular
weight.
While
this
may
provide
a
reasonable
first
approximation,
better
estimates
are
available.
There
are
several
refinements
to
this
estimation
the
first
of
which
evaluates
molecular
volume
and
can
be
computed
simply
using
equations
derived
by
Arnold
and
described
by
Braman
(
1971).
Perhaps
less
well
evaluated
in
atmospheric
studies
is
the
fact
that
for
polar
pesticides
(
OPs,
carbamates,
triazines)
the
potential
for
hydration
is
significant
under
normal
atmospheric
conditions.
As
an
example,
singly
hydrated
butanol
has
a
measured
diffusion
coefficient
that
is
22%
lower
than
that
of
anhydrous
butanol
(
Cobb
et
al
1989,
1991).
OP
pesticide
with
a
MW
of
200
g/
mole
that
was
singly
hydrated
should
have
a
decrease
in
Da
that
ranges
from
5%
to
18%
compared
to
the
anhydrous
molecule.
These
ranges
can
be
input
into
model
simulations
to
determine
if
there
is
a
reason
to
consider
hydration
in
humid
environments.

For
avian
exposure,
the
approach
in
Appendix
F
of
the
Agency's
Background
Document
incorporates
spray
drift
into
areas
off
the
field.
This
approach
appears
reasonable.
The
Agency
should
continue
consideration
of
spray
drift
and
transport
via
volatilization
after
application
as
shown
in
Appendix
F.
This
would
significantly
alter
the
area
of
impact
following
application
and
will
require
probabilistic
sampling
estimation
of
forage,
dermal,
and
inhalation
exposure
from
edge
habitat
within
the
spray
drift
zone.
At
Level
II,
it
is
likely
to
be
acceptable
to
assume
that
time
in
edge
habitat
will
provide
exposure
to
the
average
concentration
in
the
edge.

It
would
be
useful
to
have
criteria
established
that
would
determine
whether
this
module
is
turned
on
or
turned
off.
For
pesticides
with
a
very
low
Henry's
Law
Constant
or
vapor
pressure,
exposure
via
inhalation
would
be
very
low.
In
these
cases,
there
is
little
benefit
to
including
this
module
in
the
exposure
analysis.

Agency
Charge
6.
Relating
Inhalation
Exposure
to
Oral
Exposure
Toxicity
Endpoints:
The
absence
of
avian
inhalation
toxicity
data
and
the
need
to
track
all
exposure
routes
simultaneously
has
lead
to
the
development
of
a
method
to
relate
inhalation
exposures
to
oral­
dose
equivalents.
The
method
uses
the
relationship
between
mammalian
inhalation
and
oral
acute
toxicity
endpoints
along
with
an
adjustment
factor
to
account
for
some
basic
physiological
differences
between
the
mammalian
and
avian
lungs
assumed
important
to
inhaled
pesticide
bioavailability.

a.)
Please
comment
on
whether
OPP's
proposed
approach
for
relating
inhalation
exposure
to
oral­
dose
equivalents
addresses
SAP's
previous
comments
concerning
the
use
of
the
mammalian
inhalation/
oral
relationship
for
estimating
toxicity
in
birds.

Panel
Response
This
is
a
well­
considered
and
carefully
researched
approach
to
addressing
the
SAP's
previous
comments.
Within
the
limits
of
current
knowledge,
it
probably
reflects
the
best
that
can
Page
39
of
61
be
done.
The
3­
fold
increase
in
the
lung
relative
to
the
oral
route
appears
reasonable.

One
of
the
uncertainties
that
should
be
mentioned
involves
the
assumption
that
there
is
little
difference
between
the
physiology
and
anatomy
of
the
avian
and
mammalian
digestive
tract.
The
approach,
accounting
for
differences
in
the
respiratory
tract
to
derive
an
adjustment,­
presumes
that
the
relationship
between
bird
inhalation
and
oral
exposure
is
equivalent
to
the
relationship
between
mammalian
inhalation
and
oral
after
adjustment
for
differences
in
relative
absorption
via
the
lung.
While
this
is
a
reasonable
way
to
begin
to
account
for
differences
in
the
absence
of
other
information,
the
baseline
"
ratio"
might
not
be
equivalent.
Differences
exist
between
digestive
systems
and
among
digestive
systems
among
feeding
guilds
for
mammals
and
birds.
One
commenter
also
pointed
out
differences
associated
with
subsequent
metabolism
and
storage
of
chemicals
depending
on
whether
it
is
inhalation
or
oral
administration;
the
comment
suggests
important
interactions
may
be
present.
A
few
studies
have
determined
relative
toxicities
among
birds
and
mammals
and
evaluated
mechanisms
of
differences
among
birds
and
mammals
(
Padilla
and
Veronesi,
1988;
Novak
and
Padilla,
1986;
Ehrich
et
al.,
1992;
Amsallem­
Holtzman
and
Ben
Zvi,
1997).
It
would
be
helpful
to
explore
these
responses
at
a
qualitative
and
perhaps
quantitative
level
with
respect
to
absorbed
dose.
At
present,
this
is
best
handled
as
an
aspect
of
the
uncertainty
analysis.

The
Panel
believes
that
droplet
size
distribution
for
an
aerosol
spray
represents
an
important
parameter
to
be
considered.
It
is
likely
that
7
µ
m
is
too
small
a
cut­
off
for
inhalation
exposure.
The
cutoff
for
respirable
particles/
droplets
might
well
be
10
µ
m,
but
the
biggest
aerosol
droplet
that
will
stay
suspended
in
air
is
the
maximum
inhalable
droplet
size.
These
larger
aerosols
will
stick
to
the
nasal
passages,
and
become
an
oral
dose.

Sensitivity
analyses
will
be
a
valuable
tool
for
exploring
the
implications
of
adjustments
to
account
for
inhalation.
It
might
also
be
useful
to
evaluate
the
implication
of
assuming
that
the
entire
inhaled
administered
dose
is
part
of
the
oral
administered
dose.

Agency
Charge
b.)
Please
provide
suggestions
on
alternatives
for
estimating
avian
inhalation
toxicity
that
would
be
consistent
with
the
kinds
of
toxicity
data
generally
available
to
the
Agency.

Panel
Response
The
Panel's
response
to
this
question
is
divided
into
five
parts.

1.
Consider
direct
measurements
and
studies
for
chemicals
and
situations
where
sensitivity
analyses
suggest
that
this
route
may
be
important
relative
to
other
routes
of
exposure.
Page
40
of
61
2.
The
inhalation
of
pesticide
causes
a
two
compartment
response:
1)
that
fraction
of
the
inhaled
dose
that
reaches
the
lungs
is
the
"
respired"
dose,
and
is
rapidly
absorbed
into
the
pulmonary
circulation,
and
goes
to
the
systemic
circulation
without
passing
through
the
liver
for
first­
pass
metabolism
(
either
activation
or
catabolism);
and
2)
the
fraction
of
the
inhaled
dose
that
sticks
to
nasal
passages,
trachea,
and
airsacs
will
be
a
slower
acting
dose,
largely
swept
to
the
throat
and
swallowed,
making
it
an
oral
dose.
Part
of
the
inhaled
dose
will
volatilize
in
the
airways
and
airsacs,
making
it
a
vapor
phase
dose,
which
will
then
be
partly
expired,
and
partly
respired.
Some
of
the
dose
in
the
trachea
and
air
sacs
will
be
taken
into
the
circulation
by
slower
diffusion,
because
of
the
reduced
circulation
compared
to
the
lung.
Empirical
data
will
be
needed
to
evaluate
this.

3.
The
ability
of
organisms
to
survive
a
dose
is
largely
controlled
by
esterases
in
the
blood
and
detoxifying
enzymes
in
the
liver.
The
Panel
believed
that
the
forms
and
efficiency
of
these
esterases
are
dissimilar
in
the
blood
of
birds
and
mammals.
Also
the
activation
of
OPs
in
liver
or
in
other
tissues
should
be
evaluated.
The
Panel
suggested
evaluating
characteristics
such
as
lipophilicity
and
polarizability.

4.
Droplet
size
has
big
implications
for
dose
because
it
increases
as
the
cube
of
the
diameter
(
a
10
µ
m
droplet
is
8
times
as
large
as
a
5um
droplet).
The
selected
maximum
respirable
droplet
size
is
likely
to
be
too
small.
Sensitivity
analyses
would
be
useful
for
exploring
the
implications
of
a
reasonable
range
of
sizes.
Upper
bound
in
the
model
is
7
µ
m,
but
the
Hayter
and
Besch
data
in
Appendix
D
show
that
7
µ
m
is
well
respired,
and
no
data
exist
for
larger
droplets.
We
need
to
know
what
the
aerosol
droplet
distribution
is
before
any
assumptions
can
be
made.
Any
droplet
that
gets
through
the
nasal
openings
could
be
part
of
the
oral
component
of
the
inhaled
dose
as
discussed
above.

5.
The
model
does
not
include
a
pathway
associated
with
the
soil.
The
inhalation
route
only
considers
direct
spray
and
volatilization
from
foliage,
not
volatilization
from
soil.
However,
in
orchards
2/
3
of
spray
hits
the
ground.
What
is
the
fraction
in
crops?
This
fraction
should
be
included
in
the
volatilization
estimation.
The
fraction
that
hits
the
ground
will
change
during
the
crop
cycle.
As
more
foliage
develops,
less
spray
will
hit
the
ground.
The
model
now
incorporates
changing
foliage,
and
could
adjust
the
proportion
of
spray
hitting
the
ground.

Agency
Charge
7.
Estimating
Dermal
Exposure:
The
incidental
dermal
contact
model
relies
on
methods
currently
employed
by
the
OPP's
Health
Effects
Division
that
rely
on
estimates
of
foliar
contact
and
dislodgeable
foliar
residues
to
estimate
an
external
dermal
dose.

a.)
Please
comment
on
applying
this
general
approach
to
birds
and
whether
any
other
Page
41
of
61
model
alternatives
have
been
used
for
wildlife
dermal
exposure.
b.)
If
alternative
models
for
estimating
dermal
exposure
for
birds
are
available,
please
discuss
their
advantages
and
limitations
in
comparison
to
the
proposed
model.
c.)
Please
comment
on
the
following:

1.)
The
reliance
on
the
lower
leg
and
foot
as
the
significant
contact
area
for
birds.
Are
other
portions
of
avian
anatomy
significant?
If
so,
which
other
areas
should
be
included?
2.)
Recognizing
that
the
use
of
human
foliar
contact
data
has
limitations,
can
the
SAP
share
any
insights
on
available
data
that
would
allow
for
a
more
specific
foliar
contact
rate
estimate
for
birds?
3.)
Is
the
SAP
aware
of
any
data
specific
to
pesticide
foliar
residue
transfer
coefficients
for
wildlife?
If
so,
please
identify.

Panel
Response
The
model
does
not
factor
in
differences
in
foot
morphology
(
e.
g.,
webbed
vs.
nonwebbed
feet)
although
this
could
significantly
affect
exposure.
For
a
ground
bird,
the
surface
area
of
the
foot
in
contact
with
pesticide
might
be
modeled
as:
the
surface
area
of
the
foot
times
the
hopping
rate
(
number
of
hops
per
minute,
or
steps
per
minute
for
walkers)
times
the
concentration
of
pesticide
on
the
soil
surface.
Soil
contact
in
this
scenario
would
be
cm2/
min.
Saturation
of
body
surface
needs
to
be
addressed,
but
this
must
be
measured,
not
modeled.

The
general
model
considers
foliar
contact
but
does
not
address
dermal
exposure
through
contact
with
soil
or
from
`
wet'
foliage
immediately
post
spray
when
residues
are
most
likely
to
be
dislodged
from
the
substrate.
Nor
does
it
include
dermal
absorption
from
foraging
in
puddles.
The
omission
of
contact
with
the
soil
raises
concern
given
that
many
(
if
not
most)
application
scenarios
would
result
in
some
pesticide
reaching
the
soil
surface,
and
many
of
the
birds
associated
with
treated
areas
forage
on
the
ground.
Dermal
exposure
must
include
contact
with
soil.
The
ground­
foraging
birds
all
pick
up
pesticide
through
their
feet.
Raptors
pick
up
spray
by
perching
on
sprayed
limbs
while
perching/
hunting
in
sprayed
orchards,
where
they
can
find
debilitated
ground
birds.
Additionally,
choice
of
carrier
or
adjuvants
can
influence
dermal
uptake.
For
example,
absorption
of
parathion
by
hawks
through
their
feet
when
perching
in
almond
trees
was
facilitated
by
the
dormant
oil
included
in
the
formulation/
tank
mix.
(
Henderson
et
al.,
1994)

The
Agency's
approach
does
not
factor
in
the
percent
plant
canopy
coverage
at
the
time
of
application
and
its
effect
on
the
proportion
of
pesticide
applied
to
the
ground
versus
intercepted
by
the
plant
canopy.
Plant
interception
would
differ
before
and
after
foliation
and
as
crop
plants
emerge
and
develop.
Temporal
changes
in
the
plant
canopy
coverage
could,
and
should,
be
taken
into
account
in
higher
tiers
of
the
model.

Use
of
a
pickers'
hands
model
to
simulate
dermal
exposure
by
birds
does
not
seem
adequate.
The
variance
is
very
high,
and
may
be
due
to
different
crop
scenarios.
Choosing
the
Page
42
of
61
most
appropriate
crop
analogy
might
help,
but
will
still
be
very
limited.

Dermal
contact
for
the
non­
feeding
period
is
underestimated.
For
ground
birds
it
is
continuous
exposure
from
spray
landing
on
soil.
For
raptors
it
is
continuous
contact
with
sprayed
limbs.
Henderson
et
al.
(
1994)
demonstrated
that
red­
tailed
hawks
in
California
typically
receive
significant
dermal
exposure
to
pesticides.
Here,
with
respect
to
inhalation,
air
movement
does
not
necessarily
result
in
a
decrease
in
airborne
concentrations/
depositions.

Topography
and
air
temperature
can
influence
concentrations
of
pesticides
in
the
air
through
volatilization
and
movement
to
depressions
with
cooler
air,
e.
g.,
prairie­
potholes,
that
can
serve
as
`
sinks'.
Henderson
et
al.
(
1994)
and
Bartkowiak
and
Wilson
(
1995)
both
demonstrate
very
different
kinetics
for
dermal
absorption
and
effects.
The
empirical
data
from
these
studies
should
be
incorporated
into
the
model.
Important
issues
that
deserve
attention
are:
(
1)
dermal
absorption
is
slower
than
oral
uptake,
(
2)
time
to
effect
with
OPs
is
longer,
and
(
3)
time
to
recovery
is
very
much
longer
(
40
days
vs.
3
days).

Face
and
eyes
will
be
heavily
exposed,
but
the
surface
area
is
small
compared
to
feet
and
legs.
The
ophthalmic
exposure
does
not
need
to
be
modeled,
but
may
be
a
very
important
factor
in
irritation
and
sublethal
injury.
Plumage
will
not
be
a
site
of
dermal
absorption,
and
is
probably
covered
adequately
with
the
surface
area
model,
but
exposure
from
preening
is
dismissed
as
trivial.
Experiments
need
to
be
done
to
measure
preening
ingestion,
which
may
prove
to
be
significant.

The
Agency's
background
document
indicated
that
for
aerially­
applied
spray,
all
individual
birds
predicted
to
be
in
the
field
at
the
time
of
application
are
subject
to
dermal
exposure
from
applied
pesticide
droplets.
This
assumption
is
based
on
an
expected
rapid
rate
of
application
by
aerial
equipment,
with
little
opportunity
for
individuals
to
leave
the
field
during
application.
However,
for
ground­
applied
sprays,
dermal
exposure
to
pesticide
droplets
is
limited
to
those
individuals
that
are
predicted
by
the
model
to
be
on
the
field
at
the
time
of
application.
The
limitation
to
in­
field
residents
is
based
on
an
expectation
that
non­
territory
holding
individuals
will
have
the
opportunity
to
exit
the
field
in
advance
of
the
application
equipment,
but
field
residents
will
remain
on
field.

The
model
does
assume
that
even
residents
will
flush
before
application
equipment
when
no
vegetative
cover
is
present.
The
issue
is
not
so
much
whether
birds
are
on
or
off
the
field
at
application,
but
rather
their
ability
to
avoid
exposure
to
pesticide
droplets.
Birds
certainly
could,
and
probably
usually
do,
move
out
of
the
way
of
application
machinery.
If
for
no
other
reason
that
that
the
application
equipment
is
noisy.
An
exception
might
be
birds
that
are
incubating
eggs
on
nests.
The
application
equipment
makes
noise
that
the
birds
can
hear
before
the
equipment
reaches
them,
thus
giving
them
time
to
move.
The
time
that
the
droplets
stay
suspended
in
the
air
is
likely
much
shorter
than
the
60­
minute
time
step,­
probably
seconds
or
only
a
few
minutes.
Thus
when
the
birds
move
back
to
an
area
that
has
received
a
pesticide
application,
the
droplets
most
likely
would
have
settled
out
of
the
air
column.
Birds
confined
to
territories
can
still
move
Page
43
of
61
out
of
the
way
of
application
equipment
until
the
droplets
have
settled
since
the
application
zone
would
only
encompass
a
very
small
proportion
of
a
territory
at
any
given
point
in
time.

Additionally,
time
in
field
does
affect
dermal
exposure
and
the
Agency
should
consider
the
effect
of
one
hour
time
steps
on
the
dermal
dose
as
it
is
an
important
exposure
parameter.
This
raises
the
point
that
EQ
3­
55
should
have
units
of
mg/
kg/
hr
not
mg/
kg
as
stated.
If
a
one
hour
time
step
is
assumed,
this
difference
is
not
important,
but
if,
as
other
Panelists
have
suggested,
the
foraging
behavior
does
not
really
occur
in
one
hour
steps,
the
dermal
exposure
may
be
improperly
estimated.
To
better
parameterize
the
dermal
exposure
scenario,
the
Panel
believed
it
necessary
to
collect
data
for
avian
contact
with
foliage
in
greenhouse
scenarios.

Agency
Charge
8.
Relating
Dermal
Exposure
to
Oral
Exposure
Toxicity
Endpoints:
The
general
absence
of
avian
dermal
toxicity
data
and
the
need
to
track
all
exposure
routes
simultaneously
have
lead
to
the
development
of
a
method
to
relate
dermal
exposures
to
oral­
dose
equivalents.
The
method
uses
existing
avian
dermal
toxicity
for
a
subset
of
pesticides
to
establish
a
relationship
between
avian
dermal
and
oral
acute
toxicity
endpoints.
It
is
recognized
that
this
approach
is
statistically
limited
with
regards
to
the
strength
of
that
relationship,
and
that
this
method
is
constrained
by
the
limited
number
of
pesticide
modes
of
action
considered.
Please
provide
suggestions
regarding
other
route
normalization
techniques.

Panel
Response
Appendix
H
(
dermal
toxicity
estimation)
of
the
Agency's
background
document
uses
regression
analysis
to
compare
oral
and
dermal
exposures.
In
addition
to
the
points
referenced
above
with
regard
to
time
course
of
effect
and
recovery,
body
size
or
species
differences
appear
to
be
important
factors
in
the
correlation
of
oral
and
dermal
dose.
The
data
for
Figure
H­
1
are
based
on
the
data
given
in
Table
H­
1.
Only
six
pesticides
were
tested
on
three
different
species,
and
these
have
been
graphed
in
the
accompanying
figure
comparing
oral
and
dermal
LD50
toxicities.
The
oral:
dermal
ratio
of
toxicity
is
very
low
for
mallards,
and
generally
higher
for
both
house
sparrows
and
quelea,
indicating
either
a
body
size
difference,
or
some
other
species
differences.
A
more
refined
analysis
needs
to
be
performed
on
the
data
to
determine
the
best
relationship
between
oral
and
dermal
toxicities,
rather
than
a
linear
regression
on
the
available
data
as
presented
in
Figure
3
below.
Page
44
of
61
Figure
3.
Ratios
of
Oral
to
Dermal
LD50s
used
within
the
FIFRA
Level
II
Probabilistic
Risk
Assessment.

The
correlation
is
very
low
on
the
plot
(
Table
H­
1,
Figure
H­
1)
which
means
that
the
simple
linear
regression
will
be
relatively
flat.
Thus,
the
predicted
dermal
LD50
will
be
relatively
insensitive
to
the
given
oral
LD50.
It
is
important
to
allow
variability
in
dermal
LD50
values
about
the
predicted
value.
If
one
is
looking
for
a
relationship
between
oral
and
dermal
toxicity,
an
error­
in­
variables
model
might
be
more
appropriate
than
a
simple
linear
regression.
It
is
not
clear
that
LD50
is
the
best
point
of
comparison,
but
we
at
least
have
data
on
LD50.
There
are
very
few
pesticides
with
oral
and
dermal
toxicity
data,
other
than
OPs,
and
the
analysis
should
not
be
interpreted
beyond
the
OPs.
The
carbamates
in
particular
do
not
show
a
regression
relationship
like
the
OPs.

The
Panel
encourages
a
close
evaluation
of
avian
and
mammalian
dosing
processes
for
the
dermal
and
oral
routes
of
exposure,
given
that
the
regression
analyses
showed
that
the
relationship
between
pesticide
toxicity
and
exposure
via
the
two
routes
of
exposure
is
not
strong.
It
may
be
that
segregation
of
data
by
dosing
regimes
would
increase
regression
strength
even
though
it
will
decrease
the
number
of
data
points
within
each
regression.

The
time
course
of
dermal
absorption
and
pesticide
effects
are
very
different
from
oral
exposure.
The
prolonged
recovery
time
following
toxicant
exposure
may
be
critical
in
estimating
LD
50
ORAL
DERMAL
RATIOS
0.00
0.50
1.00
1.50
2.00
2.50
CHEMICAL
RATIOS
Demeton
Parathion
Dichrotophos
Fensulfothion
Fenthion
Monocrotophos
Mallard
House
Sparrow
Queala
Page
45
of
61
the
effects
of
sequential
exposures.
The
data
of
Henderson
et
al.
(
1994)
and
Bartkowiak
and
Wilson
(
1995)
must
be
verified
and
duplicated
with
other
species
and
compounds.
The
time
course
of
absorption
and
effect
should
be
measured
for
other
compounds
besides
parathion.
The
dermal
depot
for
storage
of
chemicals
needs
to
be
identified
and
quantified
for
other
compounds
and
carrier
vehicles.

The
preceding
graph
gave
a
very
different
look
at
the
data
for
two
reasons:
the
ratios
of
oral
to
dermal
are
computed
for
each
case,
rather
than
fitted
by
a
regression,
and
the
original
units
are
used
rather
than
logs
(
Figure
4).
The
graph
has
only
one
case
per
bar.
A
version
of
the
same
plot
is
given
in
Figure
4.
It
combines
all
compounds
used
with
each
species,
and
it
also
shows
that
the
mallards
are
different
from
the
house
sparrows
and
queleas.
Page
46
of
61
Figure
4.
Differential
Lethality
of
Insecticides
to
Three
Avian
Species
(
hs=
house
sparrow,
ma=
mallard,
qu=
quelea)

Considering
the
magnitude
of
the
variance
in
the
regression,
there
will
be
larger
uncertainties
for
highly
toxic
compounds
(
low
LD50s).
This
is
not
a
trivial
observation
as
highly
toxic
compounds
are
those
likely
to
pose
the
most
risk.

The
adjuvant
(
carrier)
effect
will
be
critical
in
dermal
absorption.
Surfactants
could
greatly
increase
the
absorption,
and
stickers
could
greatly
increase
the
adhesion,
buildup,
and
necessity
for
preening
of
compounds
from
skin
and
feathers.

Preening
will
result
in
an
oral
exposure
following
dermal
exposure.
This
could
be
modeled
in
a
manner
similar
to
the
inhalation­
oral
model,
but
the
added
complication
of
prolonged
pesticide
effect
and
recovery
time
with
dermal
exposure
will
complicate
the
model.

Agency
Charge
9.
Physiologically­
based
Toxicokinetic
Modeling.
The
methods
developed
to
estimate
risk
from
multimedia
and
different
routes
of
exposure
are
based
on
external
dose
estimates
that
do
not
directly
account
for
physiological,
morphological,
and
biochemical
processes
that
underlie
the
toxicokinetic
behavior
of
a
pesticide.
In
human
health
and
aquatic
life
risk
Page
47
of
61
assessments
for
drugs,
and
in
some
cases
environmental
contaminants,
use
of
physiologically­
based
toxicokinetic
(
PB­
TK)
models,
are
beginning
to
be
employed
to
derive
internal
dose
estimates
for
more
refined
dose­
response
analyses
and
to
support
route­
to­
route
and
interspecies
extrapolation.
In
this
regard,
PB­
TK
modeling
was
mentioned
by
the
SAP
during
the
2001
review
of
the
case
studies.

a.)
If
you
are
aware
of
any
developmental
work
on
avian
PB­
TK
models
since
2001,
please
discuss.
Is
the
SAP
aware
of
information
sources
that
have
compiled
measured
physiological,
morphological,
and/
or
biochemical
parameters
that
are
required
to
develop
avian
PB­
TK
models?
If
so,
please
comment.

Panel
Response
Several
scientists
have
begun
to
explore
PBTK
(
PBPK)
models.
The
most
sophisticated
to
date
is
that
of
Nichols
(
1994)
and
French
(
2002)
for
kestrels.
Birds
are
being
dosed
and
measured
at
USFWS
and
being
modeled
using
adaptations
of
EPA's
fish
models
as
starting
frameworks.
The
model
and
measurements
are
for
methylmercury,
which
has
little
application
for
non­
bioaccumulating
pesticides.
The
model
includes
uptake
by
feathers,
which
would
not
be
considered
for
most
pesticides.

Drouillard
and
Norstrom
(
2000)
are
building
a
2
compartment
model
for
PCBs
and
herring
gulls,
but
this
has
limited
application
except
for
DDT
and
other
bioaccumulating
pesticides.
Krishnan
has
developed
a
model
for
hens
with
direct­
acting
pesticides,
to
avoid
the
added
complication
of
metabolic
activation.
He
has
presented
the
model
at
meetings,
but
has
not
published
it.

Modeling
may
have
been
conducted
for
pharmaceuticals
or
growth
promoters
in
poultry,
although
few
published
data
have
been
located
to
date
(
Pollet
1985).

Agency
Charge
b.
)
Recognizing
that
research
to
support
PB­
TK
models
is
a
long­
term
and
collaborative
endeavor
across
the
Agency
and
the
scientific
community,
identifying
potential
applications
in
a
risk
assessment
context
can
provide
insights
for
prioritizing
developmental
efforts.
In
this
regard,
any
suggestions
by
the
SAP
in
terms
of
an
incremental
application
of
physiologically­
based
perspectives
in
problem
formulation,
analysis
and/
or
the
risk
characterization
phases
of
an
assessment
would
be
welcomed.
In
addition,
any
suggestions
that
may
be
helpful
to
the
broader
scientific
community
in
terms
of
research
priorities
to
develop
avian
PB­
TK
models
would
be
appreciated.
Page
48
of
61
Panel
Response
The
dermal
exposure
model
and
the
inhalation
model
would
be
good
additions
to
a
model
with
two
or
three
compartments.
Similarly,
modeling
the
oral
exposure
with
inhalation
exposure
as
a
two­
or
three­
compartment
model,
and
obtaining
better
correlations
with
air
sac
surface
uptake,
tracheal
uptake,
and
oral
ingestion
of
substances
swept
from
the
trachea
would
be
valuable.

In
general,
the
number
of
physiological
compartments
for
birds
is
large,
and
the
data
to
support
the
compartments
are
few.
This
means
that
default
assumptions
need
to
be
incorporated,
making
the
models
less
accurate
and
less
useful
than
lab
and
field
measurements.
The
dermal
data
of
Henderson
et
al.
(
1994)
and
Bartkowiak
and
Wilson
(
1995)
are
very
good
illustrations
of
the
value
of
actual
measurements.
Similarly,
the
data
of
Mineau
(
1991),
showing
that
small
birds
are
more
vulnerable
to
pesticide
exposure
than
large
birds,
perhaps
due
to
the
fact
that
they
can
starve
to
death
in
less
than
12
hours,
is
important.
Such
an
outcome
would
not
be
predicted
in
a
PBPK
model,
unless
that
fact
was
known
prior
to
the
construction
of
the
model.
Perhaps
a
model
that
incorporated
a
feeding
suppression
factor
could
be
constructed,
but
the
added
complexity
of
modeling
fat
storage,
behavioral
depression,
differential
recovery
of
appetite
and
direct
pesticide
effects,
could
make
the
model
too
complex
and
difficult
to
parameterize.

Physiological
models
could
be
improved
through
sensitivity
analysis,
by
running
the
model
with
large
variations
for
input
parameters,
and
determining
which
individual
parameters
have
the
greatest
influence
on
the
outputs.
Experimental
testing
of
those
parameters
that
are
important
would
further
refine
the
model.

The
current
efforts
to
model
methylmercury
cycling
in
aquatic
carnivorous
birds
will
give
a
much
needed
framework
for
other
PBPK
models.
The
extrapolation
to
rapid­
acting
pesticides,
with
very
different
metabolism
should
be
attempted,
but
will
need
different
parameters
than
a
mercury
model.
Quite
a
lot
of
data
exist
for
organophosphates,
with
measurements
of
brain
and
serum
cholinesterases,
and
data
for
recovery
time.
This
class
of
pesticides
would
be
a
good
beginning
for
modeling
with
a
few
compartments.

The
choice
of
bird
species
should
be
given
consideration.
Most
of
the
FIFRA
and
OECD
testing
data
are
for
bobwhites,
mallards,
and
Japanese
Quail.
Passerines
are
the
greatest
proportion
of
species
exposed
in
the
field,
and
some
pesticide
exposure
data
exist
for
house
sparrows,
some
for
starlings,
and
some
for
quelea.
Quail
(
Japanese
or
bobwhite)
and
house
sparrows
would
seem
to
be
the
best
species
to
begin
modeling.

Models
were
developed
at
Savannah
River
Ecology
Laboratory
(
SREL)
during
the
1990s
for
contaminant
uptake
by
wildlife
(
Brisbin
et
al.,
1990;
Newman
and
Jagoe,
1996;
Newman
and
Dixon,
1996).
Modules
of
these
models
could
be
useful
in
the
modeling
activities.
There
is
aldicarb
data
(
Harper
et
al.,
1999;
Cobb
et
al.,
2001)
that
evaluated
transformation,
excretion
and
concentrations
in
GI
tracts.
Work
of
numerous
investigators
over
the
years
describes
the
age
and
Page
49
of
61
species
dependence
for
enzymatic
development
in
much
detail
(
Gates
et
al.,
2001;
Mayack
and
Martin,
2003;
Gogal
et
al.,
2002;
Gard
and
Hooper,
1993;
Tian
and
Paul,
2003).

The
Panel
provided
additional
comments
in
reference
to
the
issues
presented
by
the
Agency.
The
Panel's
comments
are
provided
below
GENERAL
COMMENTS
Conservatism
Throughout
the
Model
As
would
be
expected
for
a
Level
II
model,
the
terrestrial
model
errs
on
the
side
of
conservatism.
However,
there
is
concern
about
how
the
model
incorporates
conservatism.
Currently,
the
model
incorporates
distributions
for
a
few
key
parameters
(
e.
g.,
body
weight),
but
for
most
parameters,
conservative
point
estimates
are
used
(
e.
g.,
entire
puddle
algorithm,
frequency
on
field,
etc).
The
resulting
degree
of
conservatism
from
this
approach
is
opaque
and
potentially
extreme.
This
is
because
of
"
compounded
conservatism".
Multiplying
95th
percentiles
for
a
series
of
input
variables
does
not
produce
a
95th
percentile
output
estimate,
but
rather
a
much
higher
percentile
(
typically
>
99th
percentile).
The
more
95th
percentiles
or
conservative
input
values
multiplied
together,
the
greater
the
compounded
conservatism.
The
Level
II
Model
has
many
input
variables
when
all
the
modules
are
considered
together.
Thus,
the
potential
for
compounded
conservatism
is
high.
Using
a
mix
of
conservative
and
average
input
values
does
not
solve
the
problem
because
ignoring
major
sources
of
uncertainty
could
lead
to
low
probabilityhigh
consequence
effects
being
missed.
In
any
event,
the
amount
of
conservatism
will
remain
opaque.
A
better
approach
for
erring
on
the
side
of
conservatism,
yet
using
a
fully
stochastic
approach
is
presented
below:

(
1)
Use
a
conservative
problem
formulation.
Focus
the
assessment
on
generic
species
that
are
at
highest
risk
of
exposure
in
regions
and
for
crop
uses
where
pesticide
concentrations
and
persistence
are
likely
to
be
high.

(
2)
Where
uncertainty
is
high
and
difficult
to
quantify,
re­
state
the
objective
of
the
assessment
in
a
conservative
manner.
Frequency
on
field,
for
example,
is
an
important
but
difficult
to
quantify
source
of
uncertainty
for
many
bird
species.
To
deal
with
this
uncertainty,
the
assessment
could
estimate
exposure
for
"
birds
that
have
a
frequency
on
field
of
90%".

(
3)
Once
the
problem
formulation
and
objective
of
the
assessment
have
been
formalized
in
an
appropriately
conservative
manner,
conduct
the
assessment
as
a
fully
probabilistic
assessment.
This
does
not
mean
that
every
variable
has
to
be
treated
as
a
distribution.
Well
characterized
constants
or
variables
that
are
of
only
minor
importance
(
as
determined
with
sensitivity
and
elasticity
analyses)
can
be
treated
as
point
estimates.
Remaining
variables
should,
however,
be
treated
as
distributions.
If
it
would
be
useful
to
separate
variability
and
uncertainty
due
to
lack
of
knowledge,
then
2nd
order
Monte
Carlo
techniques
or
probability
bounds
analysis
could
be
considered
to
accomplish
this
objective.
Page
50
of
61
(
4)
To
properly
err
on
the
side
of
conservatism
with
the
outputs
in
Level
II,
risk
management
decision
making
could
focus
on
the
relatively
low
probability­
high
consequence
effects
(
e.
g.,
effects
that
occur
with
5
or
10%
probability).
For
some
pesticide
assessments,
it
may
not
be
necessary
to
incorporate
all
of
the
Level
II
model
modules.
For
example,
if
a
pesticide
has
very
low
volatility,
then
it
seems
unlikely
that
exposure
via
inhalation
will
be
significant.
In
these
cases,
it
would
be
useful
to
have
the
option
to
turn
off
certain
modules.
The
advantage
of
having
this
capability
is
that
it
removes
the
necessity
to
gather
information
and
parameterize
input
variables
in
the
modules
that
have
been
turned
off.
Therefore,
as
an
early
step
in
running
the
model
it
would
be
useful
to
have
stopping
rules
for
some
modules
(
e.
g.,
above
a
specified
Henry's
Law
Constant,
inhalation
module
turned
on,
otherwise
the
module
is
turned
off).

Model
Version
2
And
Formulation
Type
The
Panel
noted
the
Terrestrial
assessment
Model
Version
2
is
meant
only
for
liquid
formulations.
It
does
not
include
granular,
and
for
granular
to
be
included,
new
modules
should
be
developed
because
of
the
added
compartmentalization
and
distribution
of
granular
particles.
The
Agency
mentioned
that
it
would
be
a
simple
matter
to
change
to
granular
formulations
instead
of
liquids.
This
may
not
be
true
since
the
analysis
of
granular
formulations
requires
a
totally
new
form
of
model.
Model
Version
2
is
not
currently
appropriate
for
granular
formulations.

Response
Surface
Methods
To
Address
Sensitivity
Analyses
Throughout
Risk
Assessment
Process
Determining
whether
the
model
is
good
enough
for
a
given
purpose
is
more
important
than
validating
that
it
is
accurate
in
all
respects.
To
be
good
enough
for
risk
assessment
purposes,
a
stochastic
model
must
be
able
to
reproduce
extreme
events
with
realistic
probabilities.
With
this
in
mind,
the
Panel
believes
there
can
be
considerable
benefit
in
performing
sensitivity
analysis
at
this
stage
of
the
model's
development.
The
Panel
also
recognizes
that
the
Agency
has
plans
to
carry
out
sensitivity
analysis
and
definitely
encourages
it
to
do
so.
To
provide
aid
in
this
process,
the
Panel
suggested
an
approach
a
little
different
from
the
running
of
a
variety
of
scenarios.
For
complex
models
with
a
large
number
of
parameters,
simplified
reduced­
form
or
response
surface
models
are
often
developed
to
enable
a
more
efficient
representation
of
the
original
model
for
use
in
sensitivity
and
uncertainty
analysis.
A
reduced­
form
model
involves
a
simplification
of
the
original
model
that
is
derived
from
the
underlying,
mechanistic
principles
of
the
parent
model
(
Kros
et
al.,
1993;
Sinha
et
al.,
1998;
Cocca,
2001;
Schultz
et
al.,
2004).
A
response­
surface
equation
provides
a
purely
empirical,
statistical
representation
of
the
input­
output
relationship
determined
from
the
original
model,
e.
g.,
by
fitting
a
regression
model
that
includes
each
of
model
inputs
and
their
pairwise
cross­
products
as
explanatory
variables
(
Cox
and
Baybutt,
1981;
Downing
et
al.,
1985),
or
fitting
a
set
of
"
stochastic
response
surfaces"
that
have
the
form
of
a
multidimensional
polynomial
(
Isukapalli
et
al.,
1998;
Cryer
and
Applequist,
2003a,
b).
Papers
by
Cryer
and
Applequist
are
of
particular
interest,
because
they
involve
studies
of
agricultural
Page
51
of
61
chemicals
and
ecological
risk.

To
fit
a
response­
surface
model,
input
parameters
for
the
original
model
must
be
sampled
using
an
appropriate
experimental
design,
and
the
model
executed
for
each
sample
to
develop
a
set
of
input­
output
"
data"
(
simulation
results)
that
are
used
for
the
fitting
procedure.
The
statistical­
fitting
procedure
provides
an
initial
screen
of
parameter
sensitivity,
because
only
those
model
parameters
that
have
a
statistically
significant
effect
on
the
model
output
are
included
in
the
response­
surface
model.
The
simplified
model
can
then
be
used
for
more
extensive
sensitivity
and
uncertainty
analysis.
Furthermore,
if
the
response­
surface
or
reduced­
form
model
provides
a
sufficiently
close
match
to
the
original
model
for
predicting
key
model
outputs,
it
may
be
deemed
suitable
for
use
in
place
of
the
original
model
for
specific
scientific
or
policy
evaluations
(
so
long
as
the
conditions
to
which
it
is
applied
fall
within
those
considered
when
fitting
the
model).
The
Agency
should
explore
the
development
of
a
reduced­
form
or
response­
surface
representation
of
their
pesticide
ecological
risk
assessment
model,
to
determine
whether
this
can
help
to
identify
important
input
parameters
to
the
model,
and
serve
as
a
simplified
version
of
the
model
for
future
applications.

The
Panel
also
recommended
other
similar
sensitivity
analysis
tools:
FAST
(
Fourier
Amplitude
Sensitivity
Analysis)
developed
by
McRae
(
Fontaine
et
al
1992)
and
Plackett
and
Burman
(
1946)
that
provide
a
structured,
formal,
statistical
approach
to
the
process.
These
tools
essentially
identify
a
rate
of
change
in
output
per
unit
of
input
(
a
response
surface)
for
all
model
input
parameters.
The
procedures
then
rank
each
parameter
in
terms
of
importance
relative
to
each
other
so
that
one
can
readily
see
which
parameters
are
important
and
which
are
not.

For
purposes
of
illustration,
an
example
using
the
Plackett­
Burman
(
PB)
tool
is
provided
below.
In
this
example,
the
leaching
of
chemicals
through
soil
is
being
modeled
by
PRZM
and
Table
1
below
lists
all
the
input
parameters
implemented
by
PB
for
analysis
of
sensitivity.
The
subsequent
figure
presents
results
from
PB
and
shows
the
importance
of
some
parameters
relative
to
each
other
for
the
prediction
of
leaching
by
PRZM.
Numbers
in
parenthesis
for
each
parameter
in
the
table
serve
as
the
key
to
the
identity
of
parameters
in
Figure
4.

Table
1
and
Figure
5
were
extracted
from
Appendix
4,
pages
105­
106,
of
the
report
"
FIFRA
Environmental
Model
Validation
Task
Force
Final
Report"
submitted
to
EPA
as
MRID
45433201.
It
is
available
to
the
public
and
can
be
reviewed
and
downloaded
at
the
following
web
site
address:
www.
femvtf.
com/
Welcome.
html.
It
provides
more
detail
on
the
workings
of
Plackett­
Burman.
Page
52
of
61
Table
1:
PRZM3
Model
Input
Parameters
Implemented
by
Plackett­
Burman
as
found
in
Appendix
B:
1)
PAN
FACTOR
33)
%
CLAY
2)
MIN
DEPTH
FROM
WHICH
EVAP.
IS
EXTRACTED
34)
HORIZON
ORGANIC
CARBON
3)
MAX
INTERCEPT
STORAGE
OF
CROP
35)
DISSOLVED
PHASE
DECAY
RATE
(
1)
4)
MAX
ROOTING
DEPTH
OF
CROP
36)
ADSORBED
PHASE
DECAY
RATE(
1)
5)
MAX
AERIAL
COVERAGE
OF
CANOPY
37)
PESTICIDE
PARTITION
COEFFICIENT(
1)
6)
R.
O.
CURVE
#
1
OF
ANTE
MOIST
COND
II
38)
DISSOLVED
PHASE
DECAY
RATE
(
2)

7)
R.
O.
CURVE
#
2
OF
ANTE
MOIST
COND
II
39)
ADSORBED
PHASE
DECAY
RATE(
2)

8)
R.
O.
CURVE
#
3
OF
ANTE
MOIST
COND
II
40)
PESTICIDE
PARTITION
COEFFICIENT(
2)
9)
UNI
SOIL
LOSS
COVER
MGT
FACTORS
41)
DISSOLVED
PHASE
DECAY
RATE
(
3)
10)
MANNINGS
N
42)
ADSORBED
PHASE
DECAY
RATE(
3)
11)
HYDRAULIC
LENGTH
43)
PESTICIDE
PARTITION
COEFFICIENT(
3)
12)
SLOPE
(
SLP)
44)
DISSOLVED
PHASE
DECAY
RATE
(
4)
13)
MAX
DRY
WGT
OF
CROP
FULL
CANOPY
45)
ADSORBED
PHASE
DECAY
RATE(
4)

14)
MAX
CANOPY
HEIGHT
46)
PESTICIDE
PARTITION
COEFFICIENT(
4)
15)
DEPTH
OF
PESTICIDE
APPLICATION
47)
DISSOLVED
PHASE
DECAY
RATE
(
5)
16)
TOTAL
APPLICATION
OF
PESTICIDE
48)
ADSORBED
PHASE
DECAY
RATE(
5)
17)
FILTRATION
PARAMETER
49)
PESTICIDE
PARTITION
COEFFICIENT(
5)
18)
PEST.
VOLATIL.
DECAY
RATE
ON
FOLIAGE
50)
DISSOLVED
PHASE
DECAY
RATE
(
6)

19)
PESTICIDE
DECAY
RATE
ON
FOLIAGE
51)
ADSORBED
PHASE
DECAY
RATE(
6)

20)
FOLIAR
EXTRACT.
COEFF
FOR
PLANT
W.
O
52)
PESTICIDE
PARTITION
COEFFICIENT(
6)
21)
PLANT
UPTAKE
FACTOR
53)
DISSOLVED
PHASE
DECAY
RATE
(
7)
22)
DIFFUSION
COEFF
FOR
PEST
IN
AIR
54)
ADSORBED
PHASE
DECAY
RATE(
7)
23)
HENRYs
LAW
CONSTANT
55)
PESTICIDE
PARTITION
COEFFICIENT(
7)
24)
ENTHALPY
OF
VAPORIZATION
56)
DISSOLVED
PHASE
DECAY
RATE
(
8)
25)
PESTICIDE
SOLUBILITY
57)
ADSORBED
PHASE
DECAY
RATE(
8)
26)
BULK
DENSITY/
MINERAL
DENSITY
58)
PESTICIDE
PARTITION
Page
53
of
61
COEFFICIENT(
8)
27)
INITIAL
SOIL
WATER
CONTENT
OF
HORIZON
59)
DISSOLVED
PHASE
DECAY
RATE
(
9)

28)
SOIL
DRAINAGE
PARAMETER
60)
ADSORBED
PHASE
DECAY
RATE(
9)
29)
PEST
SOLUTE
DISP.
COEFF.
61)
PESTICIDE
PARTITION
COEFFICIENT(
9)
30)
VAPOR
PHASE
PESTICIDE
DECAY
RATE
62)
DISSOLVED
PHASE
DECAY
RATE
(
10)

31)
THICKNESS
OF
HORIZON
COMPARTMENT
63)
ADSORBED
PHASE
DECAY
RATE(
10)

32)
%
SAND
64)
PESTICIDE
PARTITION
COEFFICIENT(
10)

Figure
5.
Example
summarization
of
model
sensitivity
for
PB
output.
E/
Emax
for
Maximum
Total
Pesticide
(
Mg/
Kg)
In
Compartment
150,
PF
=
10%.

Several
benefits
can
be
achieved
from
the
conduct
of
quasi­
global
analysis
such
as
an
analysis
of
sensitivity:

1.
Attention
is
called
to
those
parameters
of
greatest
sensitivity,
and
thus
to
those
areas
where
greatest
effort
should
be
made
to
capture
uncertainty
in
the
form
of
distributions.
Just
as
importantly,
it
also
identifies
parameters
of
little
importance
that
do
not
need
distributions
and
can
be
input
as
single
point
values.
Page
54
of
61
2.
It
can
be
an
enlightening
experience
because
it
forces
the
model
builders
to
think
deeply
about
each
input
variable
and
to
put
on
paper
some
estimate
of
its
uncertainty
and
range
in
value.
When
completed,
the
modeler
has
a
much
deeper
understanding
of
the
characteristics
of
each
model
input.

3.
It
identifies
the
areas
of
experimentation
where
effort
should
be
spent
to
develop
better
data.

4.
It
aids
in
model
simplification.
As
model­
building
progresses
there
is
the
tendency
for
the
model
to
become
all­
encompassing,
and
at
some
point
it
is
necessary
to
go
back,
review
what
is
and
is
not
important,
and
simplify
the
model
accordingly.
Version
2
appears
to
be
at
such
a
place
in
its
development.

Several
Panel
members
believed
that
some
of
the
uncertain
parameters
in
the
model
(
e.
g.,
FOF)
might
benefit
from
treatment
as
scenarios
rather
than
as
uncertainty
distributions.
This
may
aid
in
communicating
results.
This
particular
issue
came
up
at
the
New
York
Monte
Carlo
Workshop
and
people
less
familiar
with
the
2­
D
approaches
requested
a
scenario­
based
approach
to
help
them
understand
the
model
outputs.
When
explaining
to
risk
managers
the
effects
of
some
uncertain
parameters
in
the
model
(
e.
g.,
FOF),
it
might
be
helpful
to
treat
these
as
scenarios
rather
than
as
uncertainty
distributions.

Additional
Data
Needs
The
Agency
has
made
significant
progress
in
developing
its
approach
to
probabilistic
risk
assessment
and
is
to
be
commended
for
its
efforts.
However,
while
the
analyses
have
become
more
sophisticated
and
the
data
sources
more
varied,
there
appears
to
be
little
change
in
the
amount
of
"
field"
data
to
support
the
analyses.
Data
gaps
identified
previously
have
not
been
fulfilled.
Instead
new
ways
of
applying
existing
data/
other
models
to
estimate
unavailable
data
have
been
identified
and
applied.
Although
this
approach
can
serve
to
advance
the
development
of
probabilistic
models
in
the
short
term,
it
will
increase
uncertainties
and
reduce
the
Agency's
ability
to
validate/
refine
models
in
the
future.
At
times,
it
appears
the
Agency
argues
that
a
lack
of
data
hampers
the
analyses
and
model
development,
but
also
that
additional
data
would
not
reduce
uncertainty
(
for
example,
paragraph
two
on
page
six
of
the
Agency's
background
document).
The
absence
of
appropriate
data
is
noted
throughout
Chapter
3,
and
The
Panel
would
encourage
the
Agency
to
rapidly
fill
these
data
gaps.
A
research
thrust
to
quantify
critical
variables
is
necessary
within
the
Agency's
intramural
or
extramural
research
program.
Avian
toxicokinetic
modeling
would
seem
a
reasonable
thrust
area,
but
there
are
also
topics
of
exposure
such
as
avian
inhalation
of
pesticides
and
dermal
exposure.
Also,
evaluation
of
processes
that
affect
pesticide
fate
such
as
puddle
formation,
size,
and
duration
need
to
be
quantified
along
with
pesticide
concentrations
in
these
waters.
However,
it
is
important
that
the
Agency
first
define
the
role
of
the
Level
II
modeling
effort
in
the
regulatory
framework,
and
utilize
sensitivity
analyses
(
as
described
above)
to
identify
those
parameters
most
important
to
the
model,
before
collecting
additional
data.
Page
55
of
61
Effects
assessment
was
mostly
ignored
in
the
Level
II
model.
This
is
likely
a
reflection
of
the
limited
effects
data
that
are
currently
required
from
registrants
for
birds.
In
some
pesticide
assessments
(
particularly
re­
registration),
however,
it
may
be
feasible
to
develop
dose­
response
curves
for
endpoints
other
than
mortality
(
e.
g.,
reproduction,
growth).
The
current
model
is
incapable
of
considering
such
endpoints
at
present.
The
Generalized
Linear
Modeling
framework
(
Bailer
and
Oris
1997)
should
be
considered
if
reproductive
and
growth
endpoints
are
considered
in
the
future.

Role
of
Level
II
Models
In
reviewing
the
documentation
supporting
the
Agency's
Level
II
modeling
efforts,
the
role
of
the
Level
II
analyses
is
not
clear.
The
role
these
modeling
efforts
play
will
in
large
part
dictate
whether
or
not
the
models
should
be
conservative
(
i.
e.,
protective
against
false
negatives)
similar
to
the
quotient
method
in
Level
I.
Level
II
models
can
be
used
to
generate
a
regulatory
decision
point
or
they
can
serve
as
a
refinement
for
those
pesticides
that
do
not
pass
Level
I.
The
latter
application
should
be
encouraged
because
the
Level
II
modeling
efforts
would
then
complement
rather
than
duplicate
Level
I
analyses.
As
a
result,
modeling
efforts
could
utilize
untrimmed
distributions
for
model
parameterization,
i.
e.,
absent
restrictions
to
ensure
unnecessary
conservatism.
In
this
paradigm,
regulatory
decisions
could
be
made
after
Level
I,
after
application
of
Level
II
models,
after
exposure
risk
mitigation,
or
after
cost­
benefit
analysis,
in
that
order.

Model
refinement
can
then
focus
on
identifying
those
parameters
most
important
to
the
model,
and
can
be
used
to
identify
those
parameters
governing
a
Level
II
decision
for
a
particular
product.
This
paradigm
accommodates
the
development
and
implementation
of
higher­
tier
modeling
efforts
in
the
future
that
complement
existing
analyses,
and
particularly
pertinent
to
the
present
discussions,
reducing
the
need
for
Level
II
models
to
be
all
encompassing
in
their
parameters.

Lack
of
Data,
Uncertainty,
Sensitivity
Analyses,
and
Model
Validation
The
idea
of
model
validation
was
discussed
by
the
Panel.
Some
members
suggested
validating
isolated
components
of
the
model
but
it
is
apparent
that
it
is
not
practical
to
run
field
trials
capable
of
validating
the
predictions
of
the
model
as
a
whole.
This
model
has
many
deterministic
coefficients
and
relationships
in
it
and
these
should
be
made
stochastic.
This
is
perhaps
a
more
important
direction
to
go
at
this
point,
rather
than
making
all
details
of
the
model
more
realistic
and
hence
more
complicated.

Inputs
Into
the
Decision
Making
Process
If
the
model
outputs
will
be
used
to
guide
decisions,
thought
must
be
given
to
how
to
establish
decision
criteria.
Obviously,
these
will
not
be
a
single
fixed
value
(
e.
g.,
0.2
incidence)
but
may
need
set
de­
minimus
and
de­
maximus
bounds
and
considerations
for
making
decisions
in
the
middle
range.
These
are
policy
decisions
but
science
will
play
a
role
here.
In
the
Monte
Carlo
Page
56
of
61
Guidance
document
developed
for
the
Agency's
Superfund
Program,
such
decision
criteria
were
defined
near
the
tails
of
the
cumulative
distributions
of
risk
(
e.
g.,
the
range
was
90
to
99
with
95th
percentiles
the
typical
point
of
departure).
At
other
sites,
fractions
of
the
exposed
population
have
been
used
to
judge
the
significance
of
the
exposure
and
risk.
Examples
range
from
5
to
25%.
In
the
Calcasieu
and
Housatonic
Superfund
risk
assessments,
ranges
were
established
for
risk
management
decisions
along
the
lines
of
de­
minimus,
de­
maximus,
and
the
middle
ground.
These
were
set
as
probabilities
of
particular
levels
of
effect.
Page
57
of
61
REFERENCES
Amsallem­
Holtzman
E,
Ben­
Zvi.
1997.
Drug
metabolizing
enzymes
in
the
ostrich
(
Struthio
camelus):
comparison
with
the
chicken
and
the
rat.
Comp
Biochem
Physiol
CPharmacol
Endocrinol
116(
1):
47­
50.

Bailer
JA
and
Oris
JT.
1997.
Estimating
inhibition
concentrations
for
different
response
scales
using
generalized
linear
models.
Environ.
Toxicol
Chem.
16:
1554­
1559.

Bartkowiak
DJ
and
BW
Wilson.
1995.
Avian
Plasma
Carboxylesterase
Activity
as
a
Potential
Biomarker
of
Organophosphate
Pesticide
Exposure.
Environmental
Toxicology
and
Chemistry
12
(
14):
2149­
2153.

Best
LB
and
LD
Murray.
2003.
Estimating
the
proportion
of
diet
birds
take
from
treated
agricultural
fields.
Final
report
to
CropLife
America.

Best
LB.
2001.
Temporal
patterns
of
bird
abundance
in
cornfield
edges
during
the
breeding
season.
American
Midland
Naturalist
146:
94­
104.

Best
LB,
RC
Whitmore
and
GM
Booth.
1999.
Use
of
cornfields
by
birds
during
the
breeding
season:
the
importance
of
edge
habitat.
American
Midland
Naturalist
123:
84­
99.

Braman.
1971.
Atmos
Environ
7:
669.

Brewer
LW,
GP
Cobb,
MJ
Hooper,
RJ
Kendall,
and
C
Bens.
1990.
Third
Year
Investigation
of
the
Response
of
Selected
Wildlife
Populations
to
Planting
Time
Application
of
COUNTER
15G
Systemic
Insecticide­
Nematicide
in
an
Iowa
Corn
Agroecosystem.
Submitted
to
American
Cyanamid
Company.

Brewer
LW,
GP
Cobb,
MJ
Hooper,
RJ
Kendall,
CP
Weisskopf.
1992.
An
Assessment
of
Chemical
Exposure
and
Survival
of
Avian
and
Mammalian
Species
Following
Planting­
Time
Application
in
a
Typical
Midwestern
Corn
Agroecosystem.
Submitted
to
E.
I.
duPont
de
Nemours
and
Company,
Inc.

Brisbin
IL,
Jr.,
MC
Newman,
SG
McDowell
and
EL
Peters.
1990.
Prediction
of
contaminant
accumulation
by
free­
living
organisms:
applications
of
a
sigmoidal
model.
Environmental
and
Toxicological
Chemistry
9:
141­
149.
Braman.
1971.
Atmos
Environ
7:
669.

Cacini
W,
Fink
IM
1995.
Toxicity
and
excretion
of
cisplatin
in
the
avian
kidney.
Comp
Biochem
Physiol
C:
Pharmacol
Toxicol
Endocrinol,
111(
2):
343­
50.
Page
58
of
61
Cobb
GP,
Braman
RS.
1991.
Determination
of
Multiple
Diffusion
Coefficients
Using
the
Carbon
Hollow
Tube­
gas
Chromatography
Method.
J.
Air.
Waste
Manage.
Assoc.,
41:
967­
971.

Cobb
GP,
Harper
FD,
Weisskopf
CP.
2001.
Extraction
of
Aldicarb
and
its
Metabolites
from
Tissues
and
Excreta
Following
Acute
Poisoning.
Arch.
Environ.
Contam.
Toxicol.
40(
1):
77­
88.

Cocca,
P.
2001.
Mercury
Maps:
A
Quantitative
Spatial
Link
between
Air
Deposition
and
Fish
Tissue.
U.
S.
Environmental
Protection
Agency,
Office
of
Water,
EPA­
823­
R­
01­
009,
Washington,
DC,
see:
http://
www.
epa.
gov/
waterscience/
maps/
report.
pdf.

Correll
L,
Ehrich
M.
1987.
Comparative
sensitivities
of
avian
neural
esterases
to
in
vitro
inhibition
by
organophosphorus
compounds.
Toxicol
Lett;
36(
2):
197­
204.

Cox,
DC
and
Baybutt
P.
1981.
Methods
for
uncertainty
analysis:
A
comparative
study.
Risk
Analysis
1:
251­
258.

Cryer,
SA
and
GE
Applequist.
2003.
a)
Direct
treatment
of
uncertainty:
I
 
Applications
in
aquatic
invertebrate
risk
assessment
and
soil
metabolism
for
chlorpyrifos.
Environmental
Engineering
Science,
20(
3):
155­
167.

Cryer,
SA
and
GE
Applequist.
2003.
b)
Direct
treatment
of
uncertainty:
II
 
Applications
in
pesticide
runoff,
leaching
and
spray
drift
model
exposure
modeling.
Environmental
Engineering
Science,
20(
3):
169­
181.

Cryer,
SA,
van
Wesenbeeck,
I,
and
Knuteson,
J.
2003.
Predicting
regional
emissions
and
nearfield
concentrations
of
soil
fumigants
using
modest
numerical
algorithms
 
a
case
study
using
1,3­
dichloropropene.
Jour.
Ag.
Food
Chem.
51:
3401­
3409.

Cryer,
SA
and
van
Wesenbeeck,
I.
2001.
Predicted
dichloropropene
air
concentrations
resulting
from
tree
and
vine
applications
in
California.
Jour.
Env.
Qual.
30:
1887­
1895.

Downing,
DJ,
RH
Gardner
and
FO
Hoffman.
1985.
An
examination
of
response
surface
methodologies
for
uncertainty
analysis
in
assessment
models.
Technometrics,
27(
2):
151­
163.

Drouillard
KG,
Norstrom
RJ.
2003.
Development
and
Validation
of
a
Herring
Gull
Embryo
Toxicokinetic
Model
for
PCBs.
Ecotoxicology
12:
55­
68.

Dunning,
JB,
Jr.
1993.
CRC
handbook
of
avian
body
masses.
CRC
Press,
Inc.,
Boca
Raton,
Florida.
371pp.

Ehrich
M,
Correll
L.
Strait
J,
et
al.
1992.
Toxicity
and
toxicokinetics
of
carbaryl
in
chickens
and
rats:
a
comparative
study.
J
Toxicol
Environ
Health,
36(
4):
411­
23.

Freshman,
JS,
and
CA
Menzie.
1996.
Two
wildlife
exposure
models
to
assess
impacts
at
the
Page
59
of
61
individual
and
population
levels
and
the
efficacy
of
remedial
actions.
Human
and
Ecological
Risk
Assessment.
2(
3):
481­
496.

Fontaine,
DD,
Havens,
PL,
Blau,
GE,
and
Tillotson,
PM.
1992.
The
role
of
sensitivity
analysis
in
groundwater
risk
modeling
for
pesticides.
Weed
Technol.
6:
716­
724.

Gard
NW,
Hooper
MJ.
1993.
Age­
dependent
changes
in
plasma
and
brain
cholinesterase
activities
of
eastern
bluebirds
and
European
starlings.
J
Wildl
Dis.
29:
1­
6.

Gates
RJ,
Caithamer
DF,
Moritz
WE,
and
others.
2001.
Bioenergetics
and
nutrition
of
Mississippi
Valley
Population
Canada
geese
during
winter
and
migration.
Wildl.
Monogr.
146:
1­
65.

Gogal
RM
Jr.,
Johnson
MS,
Larsen
CT,
and
others.
2002.
Influence
of
dietary
2,4,6­
trinitrotoluene
exposure
in
the
northern
bobwhite
(
Colinus
virginianus).
Environ
Toxicol
Chem.
21:
81­
86.

Harper
FD,
Weisskopf
CP,
Cobb
GP.
1998.
Extraction
of
Aldicarb
and
its
Metabolites
from
Avian
Excreta
and
Gastrointestinal
Tissue.
Anal.
Chem.,
70(
15):
3329­
3332.

Henderson,
JD,
JT
Yamamoto,
DM
Fry,
JN
Seiber,
and
BW
Wilson.
1994.
Oral
and
dermal
toxicity
of
organophosphate
pesticides
in
the
domestic
pigeon
(
Columbia
livia).
Bulletin
of
Environmental
Contamination
and
Toxicology
52:
633­
640.

Isukapalli,
SS,
A
Roy
and
PG.
Georgopoulos.
1998.
Stochastic
response
surface
methods
for
uncertainty
propagation:
Application
to
environmental
and
biological
systems.
Risk
Analysis,
18(
3):
351­
363.

Klein
B
and
Macdonald
PDM.
1980.
The
multitype
continuous­
time
Markov
branching
process
in
a
periodic
environment.
Advances
in
Applied
Probability
12:
81­
93.

Kros,
J,
W
De
Vries,
PHM
Janssen
and
CI
Bak.
1993.
The
uncertainty
in
forecasting
trends
of
forest
soil
acidification.
Water
Air
and
Soil
Pollution,
66:
29­
58.

Mayack
DT,
Martin,
T.
2003.
Age­
dependent
changes
in
plasma
and
brain
cholinesterase
activities
of
house
wrens
and
European
starlings.
J.
Wildl.
Diseases.
39:
627­
637.

Mellott,
R,
GP
Cobb,
MJ
Hooper,
RJ
Kendall,
LW
Brewer.
1990.
Response
of
Wildlife
Exposed
to
Diazinon
50W
Insecticide
Applied
to
Apple
Orchards
in
Eastern
Washington
and
Pennsylvania.
June
1990.
Submitted
to
Ciba
Geigy
Inc.

McFarland,
DC.
1994.
Responses
of
territorial
New
Holland
honeyeaters
(
Phylidonyris
novaehollandiae)
to
short­
term
fluctuations
in
nectar
productivity.
EMU
94:
193­
200.

McLachlan,
MS;
Horstmann,
M.
1998.
Forests
as
Filters
of
Airborne
Organic
Pollutants:
A
Model.
Environ.
Sci.
Technol.
32:
413­
420.
Page
60
of
61
Mineau,
P,
A
Baril,
BT
Collins
,
J
Duffe,
G
Joerman,
R
Luttik.
2001.
Reference
values
for
comparing
the
acute
toxicity
of
pesticides
to
birds.
Reviews
of
Environmental
Contamination
and
Toxicology
170:
13­
74.

Mueller
JF
and
Hawker
DW.
1994.
Calculation
of
bioconcentration
factors
of
persistent
hydrophobic
compounds
in
the
air/
vegetation
system.
Chemosphere.
29:
623­
640.

Navarro
M,
Cristòfol
C,
Manesse
M,
et
al.
1997.
Study
of
the
distribution
of
albendazolesulphoxide
in
fertilized
egg
compartments.
J
Pharmacol
Toxicol
Methods
37(
4):
191­
6.

Newman,
MC
and
RH
Jagoe.
1996.
Bioaccumulation
models
with
time
lags:
Dynamics
and
stability
criteria.
Ecological
Modelling
84:
281­
286.

Newman,
MC
and
PM
Dixon.
1996.
Ecologically
meaningful
estimates
of
lethal
effect
in
individuals.
pp.
225­
253
in
Ecotoxicology:
A
Hierarchical
Treatment,
edited
by
MC
Newman
and
CH
Jagoe.
Lewis
Publishers.
Chelsea,
MI.

Nichols
J,
Rheingans
P,
Lothenbach
D,
McGeachie
R,
Skow
L,
McKim
J.
1994.
Threedimensional
visualization
of
physiologically
based
kinetic
model
outputs.
Environ.
Health
Perspect.
102:
952­
956.

Novak
R
and
Padilla
S.
1986.
An
in
vitro
comparison
of
rat
and
chicken
brain
neurotoxic
esterase.
Fundam
Appl.
Toxicol
6(
3):
464­
71.

McLachlan
MS
and
Horstmann,
M.
1998.
Forests
as
filters
of
airborne
organic
pollutants:
A
model.
Environ.
Sci.
Technol.
32:
413­
420.

Onstad,
CA,
Depressional
storage
on
tilled
soil
surfaces.
1984.
Trans.
ASAE,
729­
732.

Padilla
S
and
Veronesi
B.
1988.
Biochemical
and
morphological
validation
of
a
rodent
model
of
organophosphorus­
induced
delayed
neuropathy.
Toxicol
Ind
Health.
4(
3):
361­
71.

Plackett,
RL
and
JP
Burman.
1946.
The
design
of
optimum
multifactorial
experiments.
Biometrika
33:
305­
325.

Pollet
RA,
Glatz
CE,
Dyer
DC.
1985.
The
pharmacokinetics
of
chlortetracycline
orally
administered
to
turkeys:
influence
of
citric
acid
and
Pasteurella
multocida
infection.
J
Pharmacokinet
Biopharm
13(
3):
243­
64.

Riederer,
M.
1990.
Estimating
partitioning
and
transport
of
organic
chemicals
in
the
foliage/
atmosphere
system:
discussion
of
a
fugacity­
based
model.
Environ.
Sci.
and
Technol.
24,
829­
37.

Rodenhouse,
NL
and
LB
Best.
1994.
Foraging
patterns
of
vesper
sparrows
(
Pooecetes
Page
61
of
61
gramineus)
breeding
in
cropland.
American
Midland
Naturalist
131:
196­
206.

Schultz,
MT,
MJ
Small,
RS
Farrow
and
PS
Fischbeck.
2004.
State
water
pollution
control
policy
insights
from
a
reduced­
form
model.
Journal
of
Water
Resources
Planning
and
Management,
130(
2):
150­
159.

Sinha,
R,
MJ
Small,
PF
Ryan,
TJ
Sullivan
and
BJ
Cosby.
1998.
Reduced­
form
modelling
of
surface
water
and
soil
chemistry
for
the
tracking
and
analysis
framework.
Water,
Air
and
Soil
Pollution,
105:
617­
642.

Smith,
KEC,
Thomas,
GO.
2001.
Seasonal
and
Species
Differences
in
the
Air
­
Pasture
Transfer
of
PAHs.
Environ.
Sci.
Technol.
35:
2156­
2165.

Steenhuis,
TS,
M
Winchell,
J
Rossing,
JA
Zollweg
and
MF
Walter.
1995.
SCS
Runoff
Equation
Revisited
for
Variable­
Source
Runoff
Areas.
ASCE
J.
Irrig.
Drain.
121:
234­
238.

Steenhuis,
TS,
J
Boll,
G
Shalit,
JS
Selker
and
IA
Merwin.
1994.
A
Simple
Equation
for
Predicting
Preferential
Flow
Solute
Concentrations.
J.
Env.
Qual.
23:
1058­
1064.

Steenhuis,
TS
and
WH
van
der
Molen.
1986.
The
Thornthwaite­
Mather
Procedure
as
a
Simple
Engineering
Method
to
Predict
Recharge.
J.
Hydrol.
84:
221­
229.

Tank,
SL,
LW
Brewer,
Cobb
GP,
RJ
Kendall,
MJ
Hooper.
1992.
COUNTER
®
5th
year
Investigation
of
the
Response
of
Selected
Wildlife
Populations
to
Planting­
Time
Application
of
COUNTER
®
15G
Systemic
Insecticide
Nematicide
in
an
Iowa
Corn
Field.
Submitted
to
American
Cyanamid
Company.

Thornthwaite,
CW
and
JR
Mather.
1955.
The
Water
Balance.
Publ.
Climot.
8
(
1).

Tian
XL,
Paul
M.
2003.
Species­
specific
splicing
and
expression
of
angiotensin
convertingenzyme.
2003.
Biochem
Pharmacol,
Sep
15.
66(
6):
1037­
44.

Walter,
MT,
VK
Mehta,
AM
Marrone,
J
Boll,
TS
Steenhuis
and
MF
Walter.
2003.
Simple
Estimation
of
Prevalence
of
Hortonian
Flow
in
New
York
City
Watersheds.
ASCE
J.
Hydrol.
Engr.
8:
214­
218.
