UNITED
STATES
ENVIRONMENTAL
PROTECTION
AGENCY
Office
of
Air
Quality
Planning
and
Standards
(
OAQPS)
Research
Triangle
Park,
North
Carolina
27711
February
23,
2006
MEMORANDUM
SUBJECT:
Analyses
of
Event­
Flagged
Criteria
Pollutant
Data
FROM:
Mark
Schmidt,
OAQPS
TO:
file
General
This
memorandum
documents
the
analyses
of
ambient
pollution
data
flagged
for
exceptional
events.
Analyses
were
conducted
in
support
of
the
exceptional
events
rule
proposal.
The
analyses
were
conducted
for
illustrative
purposes
only
using
conveniently
available
databases.
The
main
focus
of
the
analyses
was
particulate
matter
(
PM)
data,
mainly
PM2.5
and
to
a
lesser
degree
estimated
PM10­
2.5,
however,
summaries
were
produced
for
all
the
criteria
pollutants.
The
analyses
utilized
PM2.5
data
for
the
years
1999­
2004;
estimated
PM10­
2.5
data
for
the
years
2001­
2003;
and
CO,
SO2,
NO2,
Pb,
O3,
and
PM10
data
for
the
years
2002­
2004.

Source
Data
PM2.5
data
were
extracted
from
the
Air
Quality
System
(
AQS)
in
two
separate
queries;
1999­
2001
data
were
extracted
on
April
18,
2005
and
2002­
2004
data
were
extracted
on
October
12,
2005.
In
most
analyses
(
unless
otherwise
specified),
the
six
years
of
data
were
evaluated
together.
Only
Federal
Reference
Method
(
FRM)
or
Federal
Equivalent
Method
(
FEM)
PM2.5
data
were
used
for
the
analyses.
The
PM2.5
raw
data
files
used
were
produced
as
a
byproduct
of
design
value
processing
software;
this
software
creates
site­
based
data
files
in
which
the
primary
monitor
data
is
augmented
with
data
from
collocated
monitors.
Unless
otherwise
specified,
analyses
utilized
all
available
data
in
these
files
whether
or
not
the
site
met
NAAQS
completeness
requirements.
The
raw
data
files
utilized
included
all
flag
fields
and
the
EPA
concurrence
indicator.
States
can
flag
a
single
data
value
with
up
to
10
informational
and/
or
event
flags;
see
Attachment
for
the
complete
list
of
AQS
qualifiers.
The
EPA
concurrence
indicator
is
used
by
EPA
regional
staff
to
approve
or
reject
the
State
flag(
s)
after
a
review
of
submitted
documentation.
If
the
Region
agrees
with
the
flag
they
will
concur
by
entering
a
`
Y'
(
yes)
in
the
indicator
field.
If
they
reject
the
flag,
they
are
supposed
to
enter
an
`
N'
(
no)
into
the
field.
For
most
flagged
data
points,
the
concurrence
indicator
is
blank.
Usually
it
is
blank
because
the
State
did
not
submit
the
required
documentation.
States
often
only
submit
documentation
if
the
flagged
data
point
makes
a
difference
in
attainment
status.
The
concurrence
field
may
also
be
blank
if
the
documentation
was
submitted
but
the
Region
has
not
yet
completed
its
review.
Some
of
the
flag
analyses
evaluated
the
concurrence
field
and
some
did
not;
if
so,
it
is
documented.
Currently,
there
is
no
FRM
or
FEM
for
PM10­
2.5.
PM10­
2.5
data
have
been
estimated
for
projects
such
as
the
recent
PM
NAAQS
Review
by
differencing
same
day
collocated
or
nearby
FRM
/
FEM
measurements
of
PM10
and
PM2.5.
The
flag
analyses
conducted
for
PM10­
2.5
data
utilized
the
2001­
2003
PM
Staff
Paper
(
SP)
database.
For
this
database,
all
available
pairs
of
collocated
or
nearby
(
FRM/
FEM)
PM10
and
PM2.5
measurements
were
used.
The
PM10
data,
which
are
reported
in
standard
conditions,
were
converted
to
local
conditions
using
collocated
(
AQS)
or
nearby
(
National
Weather
Service)
meteorological
information.
The
paired
PM10
and
PM2.5
differences
were
then
aggregated
by
site­
day.
The
SP
analyses
required
sites
to
have
at
least
11
or
more
samples
per
calendar
quarter
for
12,
8,
or
4
consecutive
quarters.
Extraneous
data
(
for
sites
that
didn't
meet
the
completeness
criterion,
or
for
sites
that
met
it
but
for
quarters
outside
of
the
consecutive
4­,
8­,
or
12­
quarter
period)
were
not
used
in
analyses.
Flags
were
assigned
to
the
PM10­
2.5
data
using
several
different
methods:
1)
A
PM10­
2.5
data
point
was
considered
flagged
if
any
of
the
constituent
pieces
(
PM10
or
PM2.5)
were
flagged;
2)
Flags
were
assigned
solely
on
the
basis
of
the
PM10
data
flag
(
ignoring
any
PM2.5
flags);
and
3)
Flags
were
assigned
if
the
PM10
data
were
flagged
but
the
PM2.5
data
were
not
flagged.
There
are
pros
and
cons
for
all
three
methods.
The
AQS
concurrence
field
was
not
considered
for
the
PM10­
2.5
flag
analyses.
CO,
SO2,
NO2,
Pb,
O3,
and
PM10
data
were
extracted
from
the
AQS
on
February
13,
2006.
Regional
concurrence
was
not
considered
in
the
corresponding
analyses.

List
of
Analyses
Analyses
outputs
are
attached
to
this
memo
in
the
order
listed
below.
Pertinent
details
such
as
processing
methodology
description
and
important
caveats
are
noted
within
the
attachments.
Relevant
findings
are
noted
below.
[
E.
g.,
Output
1
below
(
with
notes
a
through
d)
corresponds
to
Output
1;
the
notes
are
general
and
not
page­
specific.]

Output
1
­
PM2.5
flag
counts
by
year.
a)
PM2.5
data
were
flagged
for
16
distinct
types
of
events.
b)
Less
than
a
half
percent
(~.
44
%)
of
all
PM2.5
data
are
flagged
for
events
c)
The
flag
rate
varied
by
year,
from
a
low
of
0.25%
in
2001
to
a
high
of
0.58%
in
2004.
d)
The
most
abundant
event
flags
are
for
forest
fire
(
32%
of
total
flags),
construction/
demolition
(
20%),
highway
construction
(
19%),
and
Sahara
dust
(
16%).
These
four
types
accounted
for
87%
of
the
flags.
Output
2
­
PM2.5
flag
sites
by
year.
a)
Over
400
of
the
1362
sites
that
operated
between
1999
and
2004
(
30%)
had
at
least
one
event
flag.
b)
The
percentage
of
sites
that
had
flags
varied
by
year,
form
a
low
of
6%
in
1999
to
a
high
of
20%
in
2002;
on
average,
10%
of
sites
per
year
have
at
least
one
event
flag.
c)
The
events
that
were
flagged
by
the
most
sites
on
average
were
forest
fire
(
70%
of
total
flag
sites),
Sahara
dust
(
8%),
high
winds
(
6%),
and
construction/
demolition
(
6%).
Output
3
­
PM2.5
flag
rates
by
State.
a)
The
States/
Territories
with
the
highest
flag
rates
were
PR
(
15%
of
all
their
PM2.5
data
were
flagged),
VI
(
9%),
AZ
(
3%),
and
MT
(
2%).
b)
17
States
didn't
flag
any
of
their
data
for
events:
ID,
WY,
NM,
ND,
SD,
OK,
KS,
MN,
IA,
MO,
AR,
LA,
MS,
IL,
IN,
NH,
and
ME.
Output
4
­
PM2.5
flag
type
by
State.
a)
Most
of
the
States
that
had
no
event
flags
are
located
in
the
Midwest.
b)
Many
States
had
only
event
flags
for
forest
fires
(
red).
c)
The
2
eastern
territories
(
PR
and
VI)
had
the
highest
predominance
of
Sahara
dust
flags.
d)
Several
States
had
5
or
more
different
types
of
event
flags:
CA
(
8
event
types),
TX
(
6),
AZ
(
5),
and
PR
(
5).
Output
5
­
PM2.5
flag
type
by
State,
excluding
Quebec
fire
event
of
July,
2002.
a)
Several
States
(
mostly
in
the
Northeast)
only
had
flags
for
the
Quebec
fire
event:
PA,
DE,
NJ,
VT,
CT,
MA,
and
RI.
b)
Ignoring
the
Quebec
event,
most
fire
events
seem
to
be
regionalized
in
the
West
and
Southeast.
Output
6
­
PM2.5
forest
fire
flag
rate
by
State
by
year.
a)
The
2002
bars
in
the
Northeastern
States
shows
the
wide
effect
of
the
Quebec
fires.
As
noted
above,
many
of
the
affected
States
only
had
event
flags
for
this
one
specific
event.
b)
CA
had
forest
fire
flags
each
of
the
six
years.
AK
and
SC
had
forest
fire
flags
in
5
of
the
6
years.
c)
High
fire
flag
rates
(
indicating
severe
and/
or
numerous
fires)
are
seen
in
2004
for
AK,
and
in
2000
and
2003
for
MT.
Output
7
­
Maps
of
PM2.5
flag
rates
by
site
 
all
flags
and
specific
types.
a)
All
sites
in
PR
had
flag
rates
exceeding
5%
(
over
the
entire
6­
year
period
and
all
individual
years
except
2000)
b)
Except
for
1999,
many
sites
in
SC
had
high
(
relative)
flag
rates.
c)
Most
sites
with
high
wind
flags
are
located
in
the
Southwest.
Output
8
­
Table
of
concurred
PM2.5
flag
counts/
rates,
by
year
a)
3712
total
values
over
the
6­
year
period
were
flagged
for
events.
i)
The
number
of
flagged
values
for
which
States
have
submitted
supporting
documentation
is
unknown.
ii)
Only
14%
of
all
flagged
data
points
have
been
regionally
concurred.
iii)
505
total
concurred
data
points
were
regionally
concurred.
iv)
0.06%
of
all
data
points
were
regionally
concurred.
b)
95%
(
479)
of
concurred
flags
are
for
forest
fires.
i)
3%
(
15)
of
all
concurred
flags
are
for
construction/
demolition
and
less
than
1%
for
clean­
up
after
major
disaster
(
hurricane
burn),
highway
construction,
structural
fires,
and
infrequent
large
gatherings
(
July
4th).
ii)
There
are
no
concurred
flags
for
other
events
(
high
winds,
volcanic
eruptions,
Sahara
dust,
etc.)
Output
9
­
Maps
of
concurred
PM2.5
flag
counts/
rates
Output
10
­
Comparisons
of
95th
percentile
to
mean
plus
2
standard
deviations
for
PM2.5
[
95th
percentiles
were
calculated
two
ways:
1)
using
all
available
data
for
the
6­
year
period
(
1999­
2004)
for
the
same
site­
quarter
(
subsequently
termed,
'
using
all
data'),
and
2)
using
all
data
except
for
those
with
event
flags
(
concurred
or
not)
for
the
6­
year
period
for
the
same
site­
quarter
(
subsequently
termed,
`
minus
flagged
data'
or
`
excluding
flagged
data').]
a)
The
site­
quarter
`
95th
percentile'
metric
compares
quite
well
with
the
metric
`
mean
plus
2
standard
deviations'.
i)
Over
85%
of
all
95th
percentiles
are
within
10%
of
the
corresponding
Mean+
2SD.
ii)
R2
for
model
(
P95=
Mean+
2SD)
is
0.96
(
using
`
minus
flagged
data').
b)
Mean+
2SD
is
(
slightly)
higher
than
95th
percentile
for
more
site­
quarters
than
vice
versa.
Hence,
more
values
would
pass
the
95th
percentile
test
than
a
mean+
2SD
test.
Output
11
­
Statistics
for
PM2.5
data
points
and
sites
 
comparison
to
historic
site­
quarter
metrics
[
Bullets
reference
the
`
excluding
flagged
data'
approach]
a)
32%
of
all
flagged
data
are
greater
than
or
equal
to
corresponding
historic
sitequarter
95th
percentiles.
Thus,
States
are
flagging
many
low
values.
b)
84%
of
all
concurred
flagged
data
are
greater
than
or
equal
to
corresponding
historic
site­
quarter
95th
percentiles.
Thus,
either
1)
States
are
only
submitting
documentation
for
high
values
(
ones
that
impact
attainment
decisions),
and/
or
2)
Regions
are
only
approving
the
events
requests
for
high
values.
c)
For
76%
(
134)
of
the
sites
that
had
at
least
one
concurred
flag
(
176),
all
concurred
flagged
data
at
the
site
were
greater
or
equal
the
95th
percentiles.
d)
On
average,
over
6,000
PM2.5
samples
per
year
are
>
95th
percentiles.
On
average,
over
36,000
PM2.5
samples
per
year
are
>
75th
percentiles.
Output
12
­
PM2.5
stacked
bar
charts
(
by
State
and
Region)
and
maps
(
of
sites)
showing
various
combinations
of
counts/
rates
of
flagged
and/
or
concurred
flagged
values
by
<
or
>
95th
and
75th
percentiles.
[
Bullets
reference
the
`
excluding
flagged
data'
approach]
a)
The
State
average
for
the
percentage
of
flagged
data
>
95th
percentiles
is
67%
(
for
States
with
flagged
data).
For
ten
States
(
SC,
DC,
TX,
OH,
VI,
AL,
PR,
AZ,
CA,
MI),
less
than
half
of
their
flagged
data
are
>
95th
percentiles.
For
nine
States
(
OR,
CO,
NE,
WI,
DE,
NJ,
VT,
CT,
RI),
all
of
their
flagged
data
are
>
95th
percentiles.
b)
14
States
with
data
flags
had
none
concurred
 
due
to
either
1)
States
have
not
submitted
required
documentation
or
2)
Regions
have
reviewed/
concurred
on
documentation.
Eight
States
with
flags
had
100%
concurred.
c)
The
nationwide
16%
of
concurred
flagged
data
that
were
less
than
the
site­
quarter
95th
percentiles
are
not
geographically
distributed.
Seven
of
the
10
EPA
regions
had
one
or
more
concurred
flagged
PM2.5
values
during
1999­
2004.
For
three
of
these
seven
regions,
all
of
the
concurred
flag
values
met
the
95th
percentile
check
and
for
several
other
regions
most
of
the
concurred
values
met
the
test.
Regions
VIII
(
17/
17),
IX
(
5/
5),
and
V
(
2/
2)
all
had
100%
of
their
concurred
values
equal
to
or
greater
than
the
historic
site­
quarter
95th
percentile;
Region
10
had
99%
(
106/
107);
Region
1
had
92%
(
45/
49);
and
Region
3
had
84%
(
56/
67).
However,
for
Region
IV,
only
110
out
of
the
167
(
63%)
concurred
flag
values
were
`
exceptional'
as
defined
by
the
95th
percentile
test.
d)
Over
half
(
58%)
of
the
concurred
PM2.5
values
less
than
the
95th
percentiles
are
>
75th
percentile.
e)
93%
of
concurred
values
are
>
the
historic
site­
quarter
75th
percentiles.
f)
For
4
recent
well­
known
major
fire
events,
a
high
percentage
of
flagged
data
exceeded
the
corresponding
historic
site­
quarter
95th
percentiles:
95%
for
the
July
2002
Quebec
fire
event;
100%
for
the
2003
Montana
fires;
100%
for
the
October
2003
San
Diego
fires;
and
95%
for
the
2003
Alaskan
fires.
Output
13
­
Utah
PM2.5
flags.
a)
Region
8
is
the
only
Region
to
utilize
AQS's
`
non­
concurrence'
feature.
They
have
rejected
12
of
Utah's
17
event
flags.
b)
The
12
Utah
flags
were
rejected
because
of
either
insufficient
documentation
and/
or
the
event
was
not
a
valid
flag­
able
event
(
e.
g.,
fireworks
celebration)
c)
No
documentation
was
submitted
for
the
other
5
event
flags.
d)
15
of
Utah's
17
event
flags
exceeded
the
historic
site­
quarter
95th
percentiles.
The
2
events
that
were
<
95th,
were
disapproved.
Output
14
­
Sites
with
the
same
PM2.5
event
flags
for
multiple
years
a)
27
sites
had
data
flagged
for
the
same
type
of
event
four
or
more
years
1999­
2004.
This
could
indicate
a
geographic
location
susceptible
to
natural
events;
a
location
with
ongoing
(
large­
scale,
multiple
year)
or
annual­
basis
anthropogenic
activity;
and/
or
a
monitoring
agency
that
flags
more
than
others.
i)
Two
sites
in
HI
had
`
infrequent
large
gatherings'
flags
all
six
years,
presumably
for
fireworks.
(
Flagged
days
were
New
Year
'
s
Eve
and
New
Year
'
s
Day.)
ii)
Two
sites
in
DC
had
`
prescribed
burning'
flags
four
of
the
six
years,
presumably
for
fireworks.
(
Flagged
days
were
July
4th).
iii)
Several
sites
in
PR
had
`
Sahara
dust'
flags
multiple
(
4
or
5)
years.
iv)
Multiple
sites
in
SC
had
`
forest
fire'
flags
for
four
or
five
of
the
six
years.
Output
15
­
PM2.5
fireworks
(
July
4th)
analysis
a)
Some
very
high
PM2.5
concentrations
are
logged
for
July
4th.
[
Note
that
July
4th
is
not
a
scheduled
sample
day
every
year
for
most
sites
(
i.
e.,
those
not
on
an
everyday
sampling
regime.)]
The
highest
20
July
4th
values
for
1999­
2004
ranged
from
61.2
µ
g/
m3
to
108.3
µ
g/
m3.
b)
There
is
considerable
national
inconsistency
in
the
flagging/
treatment
of
July
4th
fireworks
events:
Only
five
of
the
top
20
July
4th
values
were
flagged;
these
five
values
were
flagged
for
two
different
types
of
events;
and
one
of
the
flags
was
rejected
(
concur='
N')
by
the
corresponding
EPA
region
but
another
was
accepted
(
concur='
Y')
by
the
region.
c)
20%
(
336)
of
the
1679
site­
year
July
4th
PM2.5
concentration
values
were
in
the
top
5%
of
the
corresponding
historic
('
99­'
04)
site­
year
data
(>
overall
95th
percentile).
d)
The
ratio
of
July
4th
concentrations
to
`
typical'
same­
site,
same
time­
frame
concentrations
(
defined
as
July
1,
2,
3,
6,
7,
and
8)
ranged
up
to
almost
20.
Output
16
­
Effect
of
concurred
flagged
data
on
PM2.5
DV's
and
attainment
status
[
The
effect
on
design
values
and
attainment
status
was
evaluated
strictly
on
the
two
distinct
3­
year
periods,
1999­
2001
and
2002­
2004,
so
as
to
avoid
overlap.]
a)
The
505
concurred
values
(
1999­
2004)
were
located
at
176
different
sites.
165
sites
had
concurred
values
>
95th
percentiles
and
42
sites
had
concurred
values
<
95th
percentiles;
thus,
31
sites
had
both.
b)
Taking
all
concurred
flags
into
account,
149
(
85%)
of
the
176
sites
had
different
(
revised)
annual
DV's
and
121
(
69%)
had
different
24­
hour
DV's.
The
average
effect
on
the
annual
DV's
was
a
reduction
of
­
0.3
µ
g/
m3,
the
median
change
was
a
­
0.2
µ
g/
m3,
and
the
maximum
effect
was
­
13.0
µ
g/
m3.
The
average
effect
on
the
24­
hour
DV's
was
a
reduction
of
­
3
µ
g/
m3,
the
median
change
was
a
­
1
µ
g/
m3,
and
the
maximum
effect
was
­
141
µ
g/
m3.
c)
Five
sites
met
the
current
NAAQS
levels
(
for
the
evaluated
periods
1999­
2001
and
2002­
2004)
only
because
concurred
flagged
event
data
were
ignored
from
the
computations.
[
Note:
An
area's
attainment
status
was
not
changed
due
to
elimination
of
concurred
event
flags;
all
five
sites
mentioned
were
not
the
high
site
(
DV
site)
in
their
respective
area.]
d)
Most
of
the
site
DV
differences
(
discussed
in
note
b
above)
were
due
solely
to
the
removal
of
the
large
`
exceptional'
(>
95th
percentile)
concurred
values.
The
16%
of
concurred
flagged
data
(
82
observations)
that
were
less
than
the
site­
quarter
95th
percentiles
were
essentially
unimportant
to
regulatory
NAAQS
comparisons.
The
impact
of
removing
these
low
flagged
values
from
NAAQS
metric
computations
was
negligible.
Only
2
of
the
41
sites
had
different
98th
percentile
design
values
by
removing
these
values
(
compared
to
not
removing
them),
one
with
a
­
1
µ
g/
m3
net
impact
and
the
other
with
a
­
2
µ
g/
m3
impact.
10
of
the
41
sites
had
different
annual
mean
design
values.
The
greatest
reduction
seen
at
those
ten
sites
in
annual
DV's
was
­
0.2
µ
g/
m3
(
at
1one
site);
8
sites
had
a
reduction
of
only
­
0.1
µ
g/
m3;
and
the
other
2
sites
actually
had
higher
annual
DV's
when
the
low
concurred
values
were
removed
(
one
increased
by
0.1
µ
g/
m3
and
the
other
increased
by
0.4
µ
g/
m3).
None
of
the
differences
(
caused
by
removing
the
concurred
flagged
data
less
than
the
95th
percentiles)
affected
a
crossing
of
the
current
NAAQS
thresholds
(
15.0
/
65
µ
g/
m3).
The
statistics
noted
for
the
41
sites
with
concurred
flagged
values
less
than
the
95th
percentiles
can
be
contrasted
with
similar
statistics
for
the
165
sites
with
concurred
values
greater
than
or
equal
to
the
95th
percentile
values.
Of
these
165
sites
(
with
423
total
readings
over
the
historic
site­
quarter
95th
percentiles),
144
(
87%)
had
annual
design
values
that
were
different
because
`
exceptional'
concurred
flagged
values
were
ignored;
120
of
the
165
sites
(
73%)
had
different
(
lower)
24­
hour
design
values.
The
net
effect
on
the
design
values
due
to
elimination
of
the
`>
95th
percentiles'
concurred
data
points
was
essentially
the
same
as
noted
for
removing
all
of
the
concurred
flags
(
note
b);
see
output
16.
The
five
sites
that
met
the
current
NAAQS
levels
due
to
removal
of
all
the
concurred
values
would
also
have
met
the
NAAQS
by
just
removing
the
large
(>
95th
percentile)
concurred
values.
Output
17
­
PM10­
2.5
flag
counts
and
flag
site
counts,
2001­
2003
a)
Flag
assigned
to
PM10­
2.5
if
PM10
flagged
or
PM2.5
flagged
(
i.
e.,
set
to
PM10
flag
if
present,
else
PM2.5
flag):
i)
About
1%
(~
0.9%)
of
all
PM10­
2.5
observations
are
`
flagged'.
There
are
14
different
types
of
flags.
The
most
numerous
flags
were
for
forest
fires
(
228
total,
24%
of
all
flags);
highway
construction
(
208,
22%);
Sahara
dust
(
189,
20%);
construction/
demolition
(
107,
11%),
and
high
winds
(
90,
10%).
These
five
flag
types
accounted
for
88%
of
all
flags.
ii)
About
30%
of
all
PM10­
2.5
sites
had
at
least
one
flag
for
the
period
2001­
2003.
102
sites
(
70%
of
the
flag
sites
total)
had
forest
fire
flags;
34
sites
(
23%)
had
high
winds
flags;
and
14
sites
(
10%)
had
construction/
demolition
flags.
Only
three
sites
had
highway
construction
flags
and
only
six
sites
had
Sahara
dust
flags;
thus,
those
sites
had
extremely
high
numbers
of
those
flags
on
average
(
per
site,
relative
to
the
other
noted
flag
types).
iii)
This
method
of
assigning
flags
(
of
the
three
evaluated)
produces
the
most
flags
(
and
most
number
of
flag
sites)
b)
Flag
assigned
to
PM10­
2.5
if
PM10
flagged
and
PM2.5
not
flagged:
i)
Less
than
½
%
(~
0.4%)
of
all
PM10­
2.5
observations
were
`
flagged'.
There
are
nine
different
types
of
flags.
The
most
numerous
flags
were
for
highway
construction
(
142
total,
39%
of
all
flags);
high
winds
(
81,
22%);
construction/
demolition
(
66,
18%),
and
forest
fires
(
52,
14%).
These
four
flag
types
accounted
for
93%
of
all
flags.
ii)
About
14%
of
all
PM10­
2.5
sites
had
at
least
one
flag
for
the
period
2001­
2003.
33
sites
(
49%
of
the
flag
sites
total)
had
high
winds
flags;
32
sites
(
48%)
had
forest
fire
flags;
and
11
sites
(
10%)
had
construction/
demolition
flags.
Only
one
site
had
highway
construction
flags;
thus,
that
site
had
a
huge
number
of
those
flags
(
all
142).
iii)
This
method
of
assigning
flags
(
of
the
three
evaluated)
produces
the
least
flags
(
and
least
number
of
flag
sites)
c)
Flag
assigned
to
PM10­
2.5
if
PM10
flagged
(
PM2.5
flags
ignored):
i)
0.8%
of
all
PM10­
2.5
observations
were
`
flagged'.
There
are
10
different
types
of
flags.
The
most
numerous
flags
were
for
highway
construction
(
205
total,
26%
of
all
flags);
Sahara
dust
(
184,
23%);
forest
fires
(
121,
15%);
construction/
demolition
(
103,
13%),
and
high
winds
(
91,
11%).
These
five
flag
types
accounted
for
88%
of
all
flags.
ii)
About
21%
of
all
PM10­
2.5
sites
had
at
least
one
flag
for
the
period
2001­
2003.
64
sites
(
63%
of
the
flag
sites
total)
had
forest
fire
flags;
33
sites
(
32%)
had
high
wind
flags;
and
13
sites
(
13%)
had
construction/
demolition
flags.
Only
one
site
had
highway
construction
flags;
thus,
that
site
had
a
large
number
of
those
flags
(
all
205).
iii)
This
method
of
assigning
flags
(
of
the
three
evaluated)
produces
the
`
middle'
results.
Output
18
­
Concentration
distributions
of
`
flagged'
PM10­
2.5
data
by
flag
type,
2001­
2003.
[
Bullets
reference
the
flag
assignment
method
of
`
PM10­
2.5
flag
set
to
PM10
flag
if
PM2.5
flag
blank'.]
a)
More
than
95%
of
all
`
flagged'
PM10­
2.5
data
are
less
than
70
µ
g/
m3
for
the
following
flag
types:
agricultural
tilling,
construction/
demolition,
highway
construction,
infrequent
large
gatherings,
Sahara
dust,
and
volcanic
eruptions.
b)
More
than
75%
of
all
assigned
PM10­
2.5
forest
fire
flags
are
less
than
70
µ
g/
m3.
c)
The
single
assigned
sandblasting
PM10­
2.5
flag
is
just
over
70
µ
g/
m3.
d)
More
then
half
of
all
assigned
high
winds
PM10­
2.5
flags
are
above
70
µ
g/
m3).
Output
19
­
Maps
of
PM10­
2.5
flag
rates
a)
Sites
in
PR,
VI,
and
Southern
CA
have
the
highest
flag
rates
for
2001­
2003.
El
Paso,
NM
sites
also
have
relatively
high
flag
rates.
b)
When
considering
only
flagged
PM10­
2.5
values
over
the
proposed
NAAQS
level
of
70
µ
g/
m3,
PM10­
2.5
exceptional
events
are
concentrated
in
the
Southwest
and
Southern
CA
and
are
predominately
for
high
winds.
Output
20
­
Summary
of
CO
data
flagged
for
exceptional
events,
2002­
2004:
a)
Less
than
one
hundredth
of
one
percent
(~
0.006%)
of
all
hourly
CO
observations
are
flagged
for
events.
b)
The
most
abundant
event
flags
are
for
forest
fire
(
342
total,
53%
of
all
flags),
unusual
traffic
congestion
(
174,
27%),
and
volcanic
eruptions
(
116,
18%).
Two
other
flag
types
accounted
for
about
one
percent
each
of
the
remaining
flags
(
structural
fire
and
construction/
demolition)
c)
Only
10
CO
sites
(
of
the
538
total)
had
event
flags
over
the
3­
year
period.
Six
sites
had
forest
fire
flags;
four
different
sites
accounted
for
the
other
four
flag
types.
Output
21
­
Summary
of
SO2
data
flagged
for
exceptional
events,
2002­
2004:
a)
Less
than
one
hundredth
of
one
percent
(~
0.008%)
of
all
hourly
SO2
observations
are
flagged
for
events.
b)
The
most
abundant
event
flags
are
for
volcanic
eruptions
(
1134
total,
99%
of
all
flags).
The
remaining
one
percent
of
flags
are
for
structural
fire
(
9
flags),
sandblasting
(
1),
and
chemical
spills
(
1)
c)
Only
nine
SO2
sites
(
of
the
639
total)
had
event
flags
over
the
3­
year
period.
Six
sites
had
volcanic
eruption
flags;
three
different
sites
accounted
for
the
other
three
flag
types.
Output
22
­
Summary
of
NO2
data
flagged
for
exceptional
events,
2002­
2004:
a)
Less
than
one
thousandth
of
one
percent
(~
0.0009%)
of
all
hourly
NO2
observations
are
flagged
for
events.
Only
93
data
points
out
of
almost
ten
million
total
(
2002­
2004)
have
flags:
60
data
points
are
flagged
for
volcanic
eruptions,
24
for
high
winds,
and
9
for
structural
fire.
b)
Only
three
NO2
sites
(
of
the
502
total)
had
event
flags
over
the
3­
year
period,
one
each
for
three
event
types.
Output
23
­
Summary
of
Pb
data
flagged
for
exceptional
events,
2002­
2004:
a)
24­
hour
Pb
data:
i)
About
one
quarter
of
one
percent
of
all
24­
hour
Pb
data
are
flagged
for
events.
ii)
Of
the
93
total
flagged
events,
78
(
84%)
are
for
construction/
demolition
and
15
(
14%)
are
for
high
winds.
iii)
Three
sites
(
out
of
253
that
operated
2002­
2004)
have
the
93
event
flags;
two
sites
have
construction/
demolition
flags
and
one
site
has
high
winds
flags.
b)
Composite
Pb
data:
i)
None
of
the
875
Pb
composite
data
records
(
from
34
sites)
for
2002­
2004
have
event
flags.
Output
24
­
Summary
of
O3
data
flagged
for
exceptional
events,
2002­
2004:
a)
Three
hundredths
of
a
percent
of
all
hourly
ozone
data
for
2002­
2004
have
event
flags.
b)
Most
of
the
ozone
event
flags
are
for
forest
fire
(
98%),
about
2%
are
for
prescribed
burns,
and
three
data
points
(
0%)
are
flagged
in
2002
for
a
structural
fire.
c)
76
sites
(
out
of
the
1292
total
that
operated
2002­
2004)
have
event
flags.
Only
two
sites
had
flags
in
2004,
18
sites
had
flags
in
2003,
but
59
sites
had
flags
in
2002.
d)
Of
the
76
sites
with
flags,
72
had
forest
fires
flags,
three
had
prescribed
burn
flags,
and
only
one
had
the
three
structural
fire
flags.
Output
25
­
Summary
of
PM10
data
flagged
for
exceptional
events,
2002­
2004:
a)
24­
hour
PM10
data:
i)
24­
hour
PM10
data:
for
2002­
2004
were
flagged
for
13
different
types
of
exceptional
events.
However,
all
the
flags
account
for
less
than
1%
(
0.8%)
of
all
data
points.
ii)
The
most
common
event
flags
are
for
Sahara
dust
(
655
flags,
31%
of
all
flags),
high
winds
(
624,
29%),
forest
fire
(
329,
15%),
construction/
demolition
(
229,
11%)
and
volcanic
eruptions
(
168,
8%).
These
five
flag
types,
plus
infrequent
large
gatherings
(
32,
2%)
are
present
each
of
the
three
years;
the
other
seven
flag
types
are
only
present
one
or
two
years
each.
iii)
239
sites
(
out
of
the
1148
total
that
operated
2002­
2004)
have
at
least
one
event
flag.
127
sites
(
53%
of
the
239)
have
forest
fire
flags,
97
(
41%)
have
high
winds
flags,
and
23
sites
(
10%)
have
Sahara
dust
flags.
Only
one
site
(
each)
has
flags
for
sandblasting,
high
pollen
count,
highway
construction,
sanding/
salting
of
streets,
rerouting
of
traffic,
and
clean­
up
after
major
disaster.
b)
1­
hour
PM10
data:
i)
Of
the
more
than
four
million
hourly
PM10
data
points
for
2002­
2004,
less
than
0.4
percent
are
flagged
for
events.
ii)
The
most
common
event
flags
are
for
forest
fire
(
53%
of
the
total
flagged
points),
high
winds
(
32%),
and
Sahara
dust
(
12%).
iii)
Of
the
55
sites
with
flags,
28
(
51%)
have
forest
fire
flags,
24
(
44%)
have
high
winds
flags,
and
only
1
site
(
each)
has
flags
for
volcanic
eruptions,
Sahara
dust,
sandblasting,
chemical
spills
and
industrial
accidents,
and
agricultural
tilling.

Inquiries
For
additional
information
on
the
analyses
documented
in
the
attachments,
please
contact
Mark
Schmidt
at
(
919)
541­
2416.

26
Attachments
