-------
•o
0)
3
C
-M
C
O
o
1
»— 4
NH
1 t 1
_J
00
p£
1
—
Ji
c
iX
V
£
(J
—
OJ
U
fc_
JC
o>
VI
•1
B
j>
**
•5
w
S
f
S
6
Q
U
£
g|
•c •-
£ or
JD u
1/1 i^. t/t
£ 55 §
O. Vt •-
« 4->
•^ ^ «
C 4-J
C
X 41 C
•T- "O •>—
Ecu
o o
•o g
vi to 6
•c u
•— c
41 4, 41
•*- C
£4>J
*
C
B UJ OJ
in o 3
cri rw
c O c to «—
o o « c a, * o
Vt X «-< •— «0 -— • 3 JC
§1/1 l/l Q h- C?t O
•<- *"" o^ ""* m a! ^ *^
4-- m vi c o
4-" .— •«. o •— h~ ro 4-» «o
vi h- — o ^» t/i 4> O CL i/i
1/1 jC h- Q. -O •-»
O* CL. 3 O t/l •*— • ^
L. ^*) Z O O OJ **) t/l
O — « C 59- 1-1 > V£> O. <--.
c *""* D. — c ro Q.
-i-^S "cm m _j *"* i£ P
C 4-> O •— • CL.
* v. o £ co ~" *£. v> 8 » cS
Qj OJ m O- *— o vt OJ -— O «f
LJ T) C\J t— 4J C "O O «-«
•o C'-'tnco *o .»- 4-1 c o» «-• — -
*£ O^^HO. -0 M~ O 4J ^
3 o o 12 in -^ )B u oEo"c:
TSlOTSV JDU t. "C m^-
o>
OJ
"u "c CX£ *X Q.
f » s" » ir -S
L.
3 0
•^ er
S£
o
u o
**- *J
= i
vt «o
U 4^
^ '* -S
0
0)
Vt
5
v>
§
4^
•o
c
o
u
£
•o
c
I
c
to
*J
c
v
'*-
O)
c
v£>
(4-
o
VI
VI
f
6
I/I VI
c
S c
•*•) 4J
Vt VI
•D
in jo
v>
•o o
Vt 4J
ft; .^
\. -o
3 C
4J O
o>
c
3
o
&
VI
c
I
c
o
c
i
1)
•o
o
1
VI
c
o
£
i
>,
~
L.
R
g
I.
t*_
S
L.
n vi
•D C
§ °
s5
•o
TD C
C 0
•0 O
i
— o
i §s"
.-g.5
vi vi V
JC •«
4-» ja oj
ex u
«Vi *Q
C H-
cn-2 ^
.^ -ti "*
X ^3 "V
E O *
u
- vt
vi X OJ
•o i_ -o
•— tj C
4) ^ O
~- C Vi
W- 3 O
o ••-
T) J3 -0
C t)
'* "c u
rv. * c
2 £ £
«.2-
C 4J UJ
^•oo
§0
^ rn
o
*
•
c
ro
t/l
c
3
o
c
o
CL
u
v<
•g
OJ
L.
o
* " 5
•o c «
3 — >.
u 'oi^
* — =
01
u
3
t/>
§
t.
c
0>
O Vt
II
C 4-)
to i/i
3 CT>
-— c
&I
T3 ••-
c c
3 4->
u c
CX TD
*O CTI
w o
— €
o
5 i
* T3
OJ
C 4->
.2 i
i— I/I
1 1
I/I 4->
«w o
'^ *«
I— <
. — VI
c ••- ^~
-^ OJ C
4-- O
\£> *0 I.
OJ
I/I
01
u
1
2
•C vt
E 3
>
•D
I 3
L.
. U
«0
: 1
. VI
o
4-*
JC
£• -^
JC
u
2
•0
3
cr
u
L.
0. 2
tD -^
5
OJ
i
ecursor
O. v>
c
V) O
VI •»-
OJ 4->
•o
£ g
"»
C l-
o *
— T3
«3 3
*3 J3
E
•*- i-
VI "-
tO
in
r-. c
S 4;
4^ U
Vi
3 ••
« |o
t.
«*-
»
«O
O
I/I
vt
OJ
JC VI
4-> C
x -2
<0
o tl
^ CT>
41 C
VI •«-
2 0
c c
S 1
£ x
0 "3
3
to t-
c *
3
O CO
CO *9
Vt
1
u c
— o
to (J
g f
4> *j
w i
m o
f*- J3
•e
vt
ex c
o
VI 4-1
41 -.-
C
5 S
* X
§10
•o
•*• c
4-» 3
i t.
Vt T)
«r c
fv. IO
u
3 4-T
"3 *-
O
w
4-1
o
Vt
4/1
JZ VI
C 0
6 °
to
O VI
I/I •*•
55
c c
s. i
^ X
3
to u
o cr
CD r">
o
4-J
•o
•5 1
i ?
"1
« o
r- jD
1 VI
o c
L- O
J§ * '-D
* -^ o
U U (J
e? v
"x i fe
JC U -D
•* -o j§
•*• S v
•^^ tfl S3
VI
•g C 0)
u
m oj **-
*3 * °
E J3 0
VI V» 4-»
c «
«• o v-
r^ •—
L. ar
o» 4^ -^.
c c c
3 4> 0
^ W J3
C i.
vC O 'O
1
ttJ
3
Vt
to
i
I
ec
^ 41
41 •*- V
t/> j3 r*.
S 1-5
0 U
4-» in
w ai cr>
at w
o "° "*o
U 0.
JD C C
k. 0 JC
«J -t- O
U 4J VI
O ^ fc-
^ ai x
X L. «
jr a:
* ex c
OJ E OJ
cc OJ E
i"
cc
VI
vi o
*r to
r^- t-
i t/i
e E
c -^ —
O X 4->
u •— o
O i- U
L. 4->
•O m X
W "O
* S §
Z -D J§
c E t.
Vt *}
jC vi
X C 0)
0 C
— o •-
^- 41 «*-
3 Vt O
E -0
•»• J3 O
VI •*•
VI — >
f^ O i-
cr -^
4-» Z Z
VI 4J -^
3 C C
O1 OJ O
3 W JO
< C *-
o 10
1
OJ
3
i
cc
c
•^ 41
oj ••- in
Us
£ 85
0 W
4-i vn
i. at en
C '^- »— '
a> L.
U OJ •
c "° ^o
u ex
C JC OJ
J3 C C
1- C JC
TJ — O
O TJ I.
u — ai
•a a x
X I- »O
JZ Z
cc a; E
o"
CJ
n:
Of
•c
10 "^
^C u
"ex
Vt X
c -
0 E
VI C
•*- ja
e t-
§s
J3 *-
W t/l
m o
0 t
"D U
X
jc x;
>1
4-1 V)
fO
u S
•a- o
r** 4-1
C
3 L.
*^ O
U
JC
u
u *-•
41 >o
4, 4.,
cl
°«
Vi
5 fc
§ §
4J •
i- 41
41 >O
CL, L.
vt c
ex
•c w
§ S
o c
c
"- i-
C 4)
•C 4,
cr c
i- —
o cr
c
— cc
O X
> -C
§
C (J
o £
jO &
t VI
II
c .£
S E
III-3
-------
w
a.
I
l
3
1
^.
i
-C
L.
3
t/i
U
1/1
•o
c
o
cr*
c
o
o
u
•o
s
t/1
JQ
I
i
ource
0)
2
i
r*^
1
1
S
3
u.
a
t«
i
^
.S i
'-*
OJ Qj
t/1
§ S
^ *
1/1 01
i/»
l/i
»
o, *
u OJ
u k,
i *
— §
•- c
JC O
i QJ
CT 3
C —
£.
'E
u
0)
c
«
J3
OJ
o
1
u
g
i/T
o
u
tj
c
t/t
s
c
I/I
I
'i
OJ
souro
J^
JO
U->
2
*•»
3
0)
"0
>
19
X
C
O
u
OJ
o>
£
QJ
2
>
QJ
I
t/l
Of
2
c
Ifl
k.
c
o
T)
c
o
VI
i
OJ
!«
£ 'e
_
3
c
o
'5
c
0.
i
o
a.
2
c
3
•«•*
1
TJ
C
3
•o
k,
>
TJ
!
O)
u
_
t/1
i
L. V*-
QJ O
> k.
§
)urces spatially
jraphic distribute
O
3 *J
k. C
tn -G
— k.
T3 0
V <->
O*
3 U
*"> O
C
o
iources spatially
jraphic distributi
*o E
QJ o;
fO
T3 JC
QJ 4J
IS 0
k. en
4^ C
£ -o
•o k.
iT> °
f*
*o
o
c
o
£3
k.
TS
U
£
£
U1
o
k.
c
5
a.
k.
s
01
I
o
y
e
C
s §
s f-)
£S
oc eg
^
> o
£ -r
c
o
4-»
H3
U
^
VI
§
S
i.
g
^
•O
1
C
k.
on
g
i "-
0)
u ~a
f l
S|
^ s
> 6
1. Z
41 go
« S
- QJ
^i £
QJ
Sc
1C
£ -a
e o
UJ 1
QJ
U
^
S
41
j3
i
«•
§
^
U
O
a.
c
•Q
u
to
5
ai
O) -O
5 1
1 §
u —
> i
L. 3C
*i
Oi 0=
I a,
t/1 5
c|
1/1 -o
£ i
i
«w
t/1
1/1
•w
cr
v
c
o
ZJ
1
c
k,
5
-J —
11
i i
i/i
•o *-
k. x:
•o *—
Q
O) CD
^) S
E a;
"t/l 4-1
OJ
§1
1/1 i —
ii i
Lrt
OJ
V
i/>
^t
T3
cr
U">
^
O
a.
o
QJ
k.
.£
t/1
S
1
k.
>
X
^
.
1
X
O
QJ
u
k.
3
o
i/>
c
o
OJ
*J
^
"3
J=
*^
l-
^
V
k.
^
0)
C
o
«5
*
•c
o *-•
, — u
QJ -
> k
-o ^
^
0 *-•
4-1 C
c fl)
QJ E
> QJ
si
o *^
?I
* <
3 4-1
k- O
3 •-*
xT iS -
a> o
*0 4-> ^
4J X tO
Ci J3 -— •
OJ
o
k.
3
O
l/t
>o
QJ
k.
ra
^>
"a
a.
o *-*
> k.
OJ 4-1
"O t/i
J_
0 4-»
C Oj
>
t/t ^
TJ «Q
QJ 3
U, O
k.
3 4->
i- o
*/> CO
3 «-«
* O
"D t/1 »
o* o
•0 4-* <
^ X ^
o>
(-)
k.
3
o
w^
«o
Ot
k.
•w
U-l
•o
CL
> k.
k.
O 4-«
*J C
C OJ
c cy
si
u
3 X
O 4J
i/i .^
TJ (^
t. O
TJ
•85
4-)
3 *J
k. O
t/i en
3 — •
•O un
4-> LJ
QJ X un
O J3 • — -
§
3
*0
flj
k.
•«
k.
1
J
^
O
c
o
3
a
1/1
QJ
k.
•Q
C
O
Nl
*
1
0
t/1
OJ
u
•^
k.
CTi
*0
»/1
"oi
>
QJ
b
c
U-
_
a>
•ci
i'
k
T)
' —
i'
>/t
QJ
u
•o
k.
CTi
o
I/I
OJ
>
QJ
b
0
u.
%
i
L.
i?
•~
c
I!
III-4
-------
region should not be made too small. If the modeling region was defined
as only a small fraction of the overall urban area, thereby excluding
certain monitoring stations, emission sources, and so forth, it is pos-
sible that the importance of a particular model input could be missed.
Figure III-l defines the modeling region. It encompasses nearly all of
the major emission sources in the SCAB and has sufficient coverage of the
Santa Monica Bay and Pacific Ocean to contain aged pollutants advected
westward with the nocturnal land breeze.
2. Multiple Base Cases
The influence of different meteorologies on the sensitivity study
results was of concern from the outset of the study. A Los Angeles appli-
cation would seem to provide a more extreme test of model sensitivity to
the availability of input data. Reduction in the number of wind monitors
in St. Louis, for example, would probably not degrade the resultant wind
field as much as would a corresponding reduction in Los Angeles wind
monitoring stations. Experience in preparing mixing depths for St. Louis
generally indicates no substantial difference between the estimated mixing
depths at each of the reporting sites. In Los Angeles, the close proxim-
ity of coastal and desert environs leads to shallow mixing depths (from
250 to 300 meters) along the coastal margin and to deep mixing depths
(from 900 to 1100 meters) farther inland during the midday hours. To
investigate whether the selection of different meteorological episodes
might materially alter subsequent ranking of data needs, both the 26 June
and 4 August episodes were chosen as base cases.
3. Simulation Duration
To keep overall computation costs to an acceptable level, the
simulation duration was limited to less than one day. The implications of
this constraint for the meaning and generality of the sensitivity results
are discussed in chapter VI. Analysis of the ozone concentration levels
for 26 June and 4 August indicated that nearly all monitoring stations
observed the peak concentrations well before 1800 PST. Therefore, 1900
PST was selected as the time to terminate the runs. The simulation
starting time Was selected after examination of the emission inventory; at
0400 emissions are relatively low compared to the levels during the
ensuing morning rush-hour. It was felt that starting at 0400 would not
bear adversely on the sensitivity simulation involving temporal redistri-
bution of emissions.
III-5
-------
FIGURE III-l. MODELING REGION USED IN THE AIRSHED MODEL SENSITIVITY STUDY.
(Calculations not performed for the crossed areas.)
III-6
-------
-------
TABLE III-2. AIR QUALITY MONITORING STATIONS IN THE SOUTH COAST AIR
BASIN AND THE NUMBER OF TIMES THEY WERE SELECTED FOR
INCLUSION IN THE REDUCED AEROMETRIC DATA SET
Station
Downtown Los Angeles
Azusa
Burbank
West Los Angeles
Long Beach
Reseda
Pomona
Lennox
Whittier
Newhall
Lancaster
Pasadena
Lynwood
El Toro
San Juan Capistrano
Los Alamitos
Thousand Oaks
Simi Valley
Santa Paula
Port Hueneme
Number of Times
Selected
8
6
6
1
5
6
2
6
2
3
0
1
0
0
0
2
0
0
2
0
Station
Ventura
Ojai
Pt. Mugu
Redlands
Fontana
Upland
Chino
Hemet
Banning
Pern's
Anaheim
Costa Mesa
Laguna Beach
La Habra
Redondo Beach
Camarillo
San Bernardino
Riverside
Prado Park
El Monte
Number of Times
Selected
2
0
0
1
2
3
1
0
0
1
5
1
0
0
0
2
4
1
4
3
III-8
-------
4. Selection of the Reduced Set of Air Quality Monitors
As discussed in the following section, certain simulations involve
the use of a limited set of surface aerometric monitors. The reduced set
of 10 surface air quality monitors was selected by the following methodo-
logy. Eight SAI technical staff members with individual backgrounds in
meteorology, atmospheric chemistry, emission inventory preparation, aero-
metric monitoring, modeling, and so on were asked to identify on figure
III-l the locations of 10 sites at which they would locate aerometric
monitoring stations (measuring 0^, NO, N02, RHC, S02, sulfate, wind speed,
wind direction, and temperature). Objectives of the site selection were
the following:
> Determination of the peak basin wide 0^ levels.
> Determination of RHC and NOX concentrations in major
source regions.
> Determination of characteristic flow patterns in the
basin.
The closest existing air quality monitoring stations to each of the ten
sites selected by the staff members are listed in table III-?. Of the
stations listed, the following were used in analyzing model sensitivity to
aerometric data inputs:
> Downtown Los Angeles > San Bernardino
> Azusa > Reseda
> Burbank > Lennox
> Long Beach > Prado Park
> Anaheim > Upland
Next, each of the airshed model sensitivity simulations is identified
in greater detail.
III-7
-------
C. ATTRIBUTES OF EACH SENSITIVITY RUN
1. jimulations Focusing on a Limited Number of Aerometric Monitoring
Stations
a. Run 1--1974 Upper Air Meteorological Data
This simulation is intended to address the situation in which upper
air meteorological data acquisition in a city is limited to routine, twice
daily vertical soundings, customarily performed at airports. Upper air
winds (about 300 meters above mean sea level) were based on the twice
daily soundings (0700 and 1300 PST) at El Monte. Mixing depths for the
entire region were based on these two soundings and on the hourly averaged
temperatures at all available surface wind and air quality stations.
b. Run 2--1975 Upper Air Meteorological Data
This run is the counterpart of run 1; it focuses on the 4 August 1975
meteorological regime.
c. Run 3--1974 Surface and Upper Air Meteorological Data
This simulation is intended to address the situation in which both
surface and upper air data acquisition in an urban area is sparse. Mete-
orological fields were constructed using data available from the 10 sur-
face monitors identified earlier and from the 0700 and 1330 PST El Monte
soundings.
d. Run 4--1975 Surface and Upper Air Meteorological Data
This simulation is the counterpart of run 3; it focuses on the
4 August 1975 meteorological regime.
e. Run 5.1--Ground Level Initial and Boundary Conditions
For this situation, initial and boundary conditions for 26 June 1974
were prepared from concentration data available from the reduced set of 10
aerometric stations. Meteorological fields were those of the June base
case.
III-9
-------
f. Run 5.2--1974 Reduced Meteorological and Air Quality Data
The purpose of this simulation was to examine the impact on model
predictions of limited data from both meteorological and air quality moni-
toring networks. Meteorological fields from simulation 3 were used in
conjunction with the initial and boundary conditions prepared from air
quality data from stations nearest to the 10 sites identified in run
5.1. The 26 June episode was used for this simulation.
2. Simulations Focusing on More Specialized Aerometnc Monitor.ng
Activities
In establishing the June and August base cases, a detailed examina-
tion of ambient air quality data was carried out. Among the data resour-
ces utilized in preparing initial and boundary conditions for the base
cases were the following:
> Monitoring data from the ARB station located atop Mt. Lee.
> ARB early morning hydrocarbon speciation measurements at
various stations during the summers of 1974 and 1975.
> Air quality observations at monitoring stations located
upwind and outside of the modeling region.
> Special studies data collected in the course of airborne
monitoring over the SCAB (for example see Husar et al.,
1977; Blumenthal, White, and Smith, 1978; and Edinger,
1973).
These information sources, together with the routine data gathered on the
days of interest, were of great value in preparing inputs for the base
cases. The existence of all of the above information was not known, how-
ever, at the time the first base case runs were made. It was only through
subsequent diagnostic analyses and further investigation that certa.n of
the additional information resources became known. Thus, though some of
the preliminary attempts at establishing June and August base cases were
set aside based upon additional information, these simulations are of
value from the viewpoint of model sensitivity. Five of these runs are
presented below.
111-10
-------
a. Run 6--Upper Air Quality Data
Run 6 was a simulation of 26 June in which the hydrocarbon and NOY
"ft "
precursor concentrations in level three grid cells were increased (rela-
tive to the base case) based upon the detailed aircraft studies reported
by Husar et al. (1977). In the absence of upper air concentration data on
26 June, pollutant concentrations aloft could only be speculated.
b. Run 7--1975 Clean Air Boundary Conditions
Run 7 differed from the 4 August base case in that lower precursor
concentrations aloft were considered and inflow boundary conditions near
Thousand Oaks and Laguna Beach-Costa Mesa were estimated to be similar to
"clean air" concentrations. In contrast, the base case run involved the
use of actual hourly concentration measurements at Thousand Oaks, Santa
Paula, Simi, Costa Mesa, and Laguna Beach.
c. Run 8--1974 Clean Air Boundary Conditions
The initial and boundary conditions differ from the base case in that
lower hydrocarbon and nitrogen oxide concentrations were considered
("clean air" conditions).
d. Run 9--1974 Assumed RHC/NOV Ratio
A
Prior to the acquisition of the detailed aerometrk hydrocarbon data
reported by Mayersohn et al. (1975, 1976) for the summers of 1974 and
1975, the initial conditions for the 26 June simulations were based upon
an assumed hydrocarbon to NOX ratio of 7 (Reynolds et al., 1979). Run 9
differs from the 26 June base case in that the former simulation involved
"clean air boundary conditions" and did not take into account the observed
empirical relationship between RHC and THC derived from Mayersohn's data.
e. Run 10—1975 Assumed RHC/NOX Ratio
This run is the August counterpart of run 9.
The vertical structure of the modeling region consisted of four layers,
whose combined height was a constant 1000 m above local ground level.
III-ll
-------
3. Simulations Focusing on Details in Emissions Inventories.
Altogether, nine sensitivity runs focused on details in the various
components of an airshed model emission inventory.
a. Run 11--1974 Hydrocarbon Emissions Speciation
The potential need for detailed hydrocarbon speciation of the major
emission sources (as compared to overall basin wide hydrocarbon splits)
was addressed in Run 11. Table III-.3 gives the hydrocarbon splits used in
both base cases by source category.
Composite hydrocarbon splits were developed by determining the per-
cent contribution of olefins, paraffins, aromatics, carbonyls and ethylene
to the total reactive hydrocarbon inventory. The resultant breakdown
between the various species groupings are as follows:
Percent by Weight
Species (as carbon)
Olefins 5
Paraffins 70
Aromatics 17
Carbonyls 5
Ethylene 3
b. Run 12—1975 Hydrocarbon Emissions Speciation
The overall hydrocarbon splitting factors in Runs 11 and 12 were
identical.
c. Run 13--1974 Mobile Source
The base case motor vehicle emission inventory employed in this study
was generated by the Direct Travel Impact Model (DTIM), a successor to the
Caltrans Link Emissions Model (FWY011). Advantages of the DTIM code are
its ability to allocate emissions due to cold starts, hot starts, and hot
soaks based on the origin, destination, and type of trip. The trip
assignment program accounts for intrazonal and terminal traffic volumes in
addition to the link volume records treated by its predecessor, FWYOll.
DTIM uses hot and cold start and hot soak files for emission estimations
based on the correspondence between the traffic zones and grid cells.
111-12
-------
o:
o
CD
or
fx
cs
1
LU
V
CO
o cs
1 I
1U LJ
r- CD
to r~
ta
i
UJ
m
CO
n
CS
i
LJ
CM
Is*
C
1
U-J
CJ
CB
l
UJ
E.
CB
rj
Q
i
LU
l£>
LD
G
i
UJ
p-
G
1
LU1
&
r-
G
l
LJ
C,
r-
IN.'
C.
1
LL
r
«r
1_
•"^
ro
OD
o.
oo
LU UJ
cr> «-
r- r-
KI
K
cr
s rn
E s — c:
P- r- «• CO — IT
tr r- r- r- a:
LJ
in
hj
o
Z •=! —•
— O -I
U. O D
I z 1 i:
a
CJ
a:
z
7- CC
ox r>
O QJ
o: -J
^- o i
cr a:
o 2 _/
^ o — "z
c: '^ o c,
co •-
tn
_j
cr
i
j •— c: t_i o:
D- LT, (_• — LJJ C"'
i- £ LJ c: y d
cr x — _j — o
_j LJ _j ~
mi— ^-—-at—
•— o a. o a. a. o
oca ^ — — o u_ •-
O t-
-------
r~ r\j
cs &
SI
I
bJ
in
rg rj
ID
OC
CT
LJ
a a
a:
O
a>
3
C
C
O
o
CB
e
e
L
C
C
e <
&
£
j
2
3
r s
e
r-
LC C
UJ UJ
fB CM
— r-
&. ts in — CB c
s s E:
i i
UJ UJ Ul
w r- CR
a _ _. ,£ 5, -
UJ CC
u r.
a: o
ID UJ
O t-
in cc
u
yr
t^
a
UJ
i
CET
X
(J
o
CC
UJ
—
cr
CL
CL
LLJ
^
LO
LU
a
f~~
ro
CD
bJ
_J
LJ
X
LU
tv
*— a
0 —
X X
in
CC 0
o z
u- r
O -
CD cn
rn m
o
LT
LO
LO
cr
o
a
LJ
LT
cr
or
O
LO
t—
U-'
CL-
C1
fT
C
f_
cr
o
LJ
LJ
a
CC
o
z
t—
CC
0
UJ
CC
CC
-
1
—
^-
z
0
t—
cr
cn
tn
a
to
h-
CT
_J
LJ
^3
U.
CC
X
LU
•
U*l
o
z
-J
LJ
{£
Q
UJ
Z
UJ
>.
x
UJ
o
or
O
r
CC
LU
UD
O
tO
CC
UJ
Ql
O
a
LJ
LD
0
— 1
X
2
O
^y
r*
t~
a
o
LU
(_J
Ct
U-
C£
^>
O
LJ
i —
CC
UJ
a:
a
LJ
CD
LU
\—
CC
UJ
x
a
z
a:
cn
UJ
_j
a
CD
CD
o
X
UJ
cn
Z
o
t—
a
CL
LU
z
\- LJ
LD ™
a
X -I
^ cr
LU —
LJ O
o z
a: -
C —
UJ UJ
0 0
Ct Ci
(/I '-'"'
«*• fv
>- >-
^- «-
__' 1
h* *-~
U '-'
LJ UJ
c L:
r^j n~i
iT> LO
Ct
LJ
O
CD
^_
Z
er
Q_
0
LU
tn
— o
in H-
LJ Ct
X 2
o .
a •
V LT>
LO U~)
CC
LJ
O
CD
>—
Z
CC
1
Cl-
3
cr;
CL
^_
o
0
CT.
UJ
LT
CC
LU
_J
C
CC
t~
z
-J
ex
ct
2
c^
a
o
r--
r
cr
o
Cl
LL-
L_
O
^
—
•-
c,
-
CC'
^
o
cz
o
tr
.^
LI
z
—
LJ
E
5
c-
I
LJ
O
a
r-
|
L-
0
_J
r~
3;
"
=
C
-
C;
X
U-1
Q
^
C_
1
LL.
O
I
u_
_
-D
z:
C'
0
c:.
z:
GJ
_j
L-.'
—j
'-
t-
t1
5
r '
Ll
>—
cr
— ~ cr
- — C i.
•- _i LV
c tn c 2
o a_ . ' 2 — CL
L «— « i_ — cr/
.* •_•• IL- cr
U- —! — LJ •— Ll
cr c. cr
_ L_ 3- Cl _J l-
~ .— cr c. — LJ
Zl C UJ OO C1
i- ce — 2 — d:
~j o") cr z • •
p" T- LP 'JT r*~ c^1
^
_J T
C '-
CZ Jj
^ ^
^— L_
<-- <.-
£ H
T- --
C *f
LJ _'
LJ ,
C O
Z Z
• •
CT1 "H
111-14
-------
(NJ
Si
1
UJ
— '
— EC
(NJ
UJ
O)
2 si — s
K)
UJ
CD
tn
C3 CJ)
en TNJ
Q GJ
UJ UJ
CO LO
fTl tO
(T) cn
s
i
UJ
rn
in
en c
E3
LJ
S3
in
s og
CM — fO
C2 S> D
1 1 >
UJ LJ LJ
h- i i r"
cn TT ro
m jj un
C2 G- £1. C- Q Cj E1 C
^ Cfi 1- f J C C C.
r.g r- r- cr LT r~ r r-
f-j
c
c,
r-
t\J rsi
C S
i i
1 — L_
cr •"
- r-
CO
C3
LJ
C
^
2 r- ts
E
LJ
f\J
—
—
cr
a
eo EJ s GJ eo s
11 ii
CO rn cc Psj LC cr,
tC LT> CD CC LD LT>
t» CO Q CS S)
I til
UJ LU LU LU LU
O
a:
a
S) (S CB S-
J UJ
r-
ijj
OD
L-1
CD
fSJ
UJ
CD
LJ U
LO U
LO r
J u.
3 a>
- s
L-1 _-• u
cr. ca u
L. UJ LLJ UJ
T INJ r"' r-vj
n in LO in
u-
tn
UJ
rr,
LI
U-i
U".
L_ LJ '_
CC c- S.
LC cr, ^
CB PsJ (S3 OB —IS
CD Si K» Ci CD a
3
C
O
m
i
it S)
2 Ci
J LU
•' CT'
- r-j
CO
i
LJ
ID
—
Ki C£i Si S
(S S C2 SI
LJ LU UJ UJ
CO r-si r" LTI
c^j u-j cn cn
csJ n-) — CM e
C3
1
UJ
JD
s cn s> —
i i
aJ UJ UJ UJ
u, r- ro cn
j .-NJ CD r\* CO
CM
IS
1
LU
-*
— i —• S C: (O
Ou S S G (S1
1 1
LU LJ l_ LJ LJ
cc) s o TJ ;-
r- CM x -j
bJ O
o
a_ ui
L- Ll,
LJ UJ
cc it:
SE
1B BOILER
D
UJ
O
O
C£
H-
LJ
0-
CO
O
UJ
tn
o
u
,_
z
cr
ttL
UJ
0
tn
r-
t—
CT
Q-
O
C3
ft:
d
O
CD
U
2
CC
5
Z
_J
a
U-
INOUS
CD
o
z
UJ
CO
Z
o
z
_J
a:
<_j
aj
r
x:
o
r\>
CD
MB ENGINES
LJ
f-
O
O
U-
LU
CL
CO
in >-
ILJ Cc:
^ to
(3 o
:c o
UJ 2!
ID
O C
UJ
i- s:
z
I.L- —
uj a:
£t (X
•V LT-
00 CD
. INDUSTRY
CL
UJ
D'
a
0
Lu1
Ln
CU
PROCESSES
y
0
c
L_
LU
Q-
CO
C
LJ
L j
C.
c.
u:
X
e.
L_
L.
ft
CL
cc
LU
c
u
(-
CD
U-
o
1
±'
u
UJ
LJ
Q_
—
0
U.
LJ
a.
:
a.
a
CJL
C
LT
U.
L-
C.
C-
a
tn
cr
o
X
U-
;
CL.
IL. CL. a_ cr
cr c ~ > o_
> ^ i- ^_ cr
L- LJ U_' ^
^ '-• r " C
Cu t. L- ~
LJ 1_ LJ CT CC
C C LJ
C 1. ^ U.
•~ _J —) L~ L_
L^ U- L_
t~ t—
— : _J LJ — ' LJ
u. i_ _ i. a.
cr< a. : f ~*
CL
c
t"
LL.
C
o
cr
-
LJ
(L
CT
C_
C
L^1
U
C.
LJ
d
t
1
LJ
C*
CD
C
LJ
CL
C
t-
i JXdOW
t—
UJ
CL
«-,
a
=:
-
T
UJ
CU
0-
CE
LJ
LC
Z
CL
T*
LU
CL
GuoiNii rjun '
rr
~
C1
t—
uJ
CL
c:
^
L,
L- _
c n
L- ~
O C
^ -
cr a
D a
E£
111-15
-------
I
t~
UJ
*-*
cs
LU
in
q-
— m
i i
LJ LJ
r\j —
m m
f\j CT>
« —.
S SJ
1 1
LU UJ
.-. «r
— (NJ
m
^
UJ
tr
«»-
f\l C
i -
IS
UJ
'?•
a 'r
__
si
UJ
r\j
r-
m (S
ea UJ
—I 1.
O Q
I/I a
0 O
er —
z oc
LU Q-
o o
UJ UJ
(r cr or ex d
-z.
a.
0.
uj
UJ
CO
O
^ (/I
a.
a
•— U- UJ Q-
O — en i-
o o a L^
1C
c ^
IT, <.-
C. =)
o a
or cc
a: a
o o
z z
a a
C^ Q;
(x a
a o
"Z. Z
c cr
_j f
— a
0 3
i z:
r z:
LJ CJ
X
C"
l_ LJ
. a.
K i~
> c
UJ — 3
tv: •-
> »- a
a: o —
. > a
tt
1 ^ XZ
1 UJ H
— _j a
u. o »--
Q: i
2! LU O
-i : '
-~ c_
CL LjJ
a:
CL —•
LU t
^ lit
f"~ J
1
LU
x
>
C7
_J ~
>-
L.1 **'
O
>— ^
c: c
^ L
IT -_
_ _
L_ -
Q '-
r r
•_> '-_•
* *
cr t'
LJ
f- C1
CI Q_
T CL,
*^- r-
t— O
U LJ
„ —
cr
i_
c. t~
LJ '/"
o: ^
a CL
. *-
— if IX f-
111-16
-------
Running emissions are allocated based on link volumes of traffic. The
DTIM code provides estimates of up to 6 classes of RHC, CO, NOV, SOV,
A A
particulates, and fuel consumption. DTIM estimates emissions for eight
vehicle types instead of the four vehicle types treated by FWY011. The
base case motor vehicle inventory is based upon an updated (as of 1976)
origin-destination survey.
For the 1974 mobile source sensitivity simulation, an outdated motor
vehicle emission assessment model (FWY011) was utilized. Features of the
FWY011 Link Emissions, Model include the following:
> Emissions are calculated by link.
> Emission estimates are provided for five pollutants--RHC,
THC, CO, NOX, SOX.
> Four vehicle types are treated.
> Cold starts, hot starts and hot soaks are assigned by set
percentage of link volumes.
> Intrazonal and terminal volumes are not taken into
account.
The latter two features represent deficiencies in the FWY011 approach when
it is compared with DTIM. Emissions due to cold starts should be assigned
to the grid of origin. Emissions due to hot soaks should be assigned to
zone or grid of destination. Inputs to the FWY011 code included 1967
origin-destination data projected to 1974 assumed population growth condi-
tions.
In essence, the major differences between the two motor vehicle
inventories is that the sensitivity run inventory is based on outdated
origin-destination data and does not adequately treat trip end emissions.
d. Run 14—1974 Gas Sales
The objective of this sensitivity run was to examine the effect of
using data related to vehicle fuel mileage to estimate daily emissions of
major pollutants from on-road motor vehicles. Although these estimates
should be less precise than those derived from sophisticated motor vehicle
emission models, such as the DTIM, emission estimates derived from fuel
mileage are usually much easier to calculate.
111-17
-------
The estimation of motor vehicle emissions for Run 14 was based on
three major parameters:
> Vehicle fuel mileage
> Vehicle emission factors
> Area fuel sales.
1) Vehicle Fuel Mil
Fuel mileage in miles per gallon (mi/gal) was derived from the EMFAC5
computer program. EMFAC5 has been developed by the ARB to estimate motor
vehicle emission factors and fuel mileage, The EMFAC5 program is similar
to the EPA's MOBILE .1 program, but it includes, among other things, the
motor vehicle emission standards applicable to vehicles sold in Cali-
fornia. EMFAC5 was used to estimate average fuel mileage values for both
gasoline and diesel vehicles. A mix of the following vehicle types for
the year 1974 was used in the calculations:
Percent of
Vehicle Type Total Vehicles
jjasoline
Light duty automobile 80.4
Light duty truck 12.1
Medium duty truck 1.4
Heavy duty gasoline 2.5
Motorcyle 1.1
j)iese1
Heavy duty diesel 2.5
The vehicle mix percentages listed above were developed for the SCAB by
Caltrans. Based on this vehicle mix and other assumptions described below
for EMFAC5, average gasoline and diesel vehicles had the following fuel
mileages:
> Gasoline: 12.73 mi/gal
> Diesel: 5.5 mi/gal.
111-18
-------
2) Vehicle Emission Factors
Emission factors in grams per mile (g/mi) were derived from EMFAC5
for hydrocarbons (HC), carbon monoxide (CO), and nitrogen oxides (NOX).
In addition to the mix of vehicles identified above, the following EMFAC5
assumptions were made to calculate emission factors:
> Average vehicle speed: 19.6 miles per hour.
> Percentage of cold start, hot start, and hot stabilized
operation: 21%, 27%, and 52%, respectively.
> Anbient temperature: 75° F.
The assumptions for vehicle speed and percentage of vehicle operation in
different modes were based upon the Federal Test Procedure for motor vehi-
cles. If better data in these areas had been available, it might have
been possible to calculate improved motor vehicle emission estimates.
Emission factors were calculated for cold start, hot start, and hot
stabilized operation as well as for crankcase, diurnal, and hot soak emis-
sions. EMFAC5 also provided a non-methane hydrocarbon (NMHC) to total
hydrocarbon (THC) ratio of 0.9496 for gasoline vehicle exhaust emis-
sions. Using all of these assumptions, the following emission factors
were calculated from EMFAC5:
Gasoline vehicles
Exhaust + crankcase, g/mi
Diesel vehicles «
Exhaust, g/mi
All vehicles
Di urnal, g/day
Hot soak, g/soak
THC
4.41
9.42
6.63
Emissions
NMHC
CO
NO,
6.52 6.19 69.92 3.69
34.08 21.29
These are emission factors, per vehicle, for the two types of evapora-
tive HC emissions. Diurnal emissions cover evaporative HC during
expansion of vapor in the fuel tank with changes in daily ambient
temperature. Hot soak emissions include vapor loss from the carburetion
system after completion of the vehicle trip. These HC emissions must be
added to the exhaust emissions to get total motor vehicle emissions.
111-19
-------
3) Area Fuel Sales
Estimates of the sale of motor vehicle fuel for a specific portion of
a state can be difficult to calculate. Recognizing the need for such
estimates, Caltrans has compiled gasoline sales data by county for various
years. The county gasoline sales data have been developed primarily from
vehicle registration data.
Initially, we attempted to construct fuel sales data for the SCAB.
Because the SCAB contains portions of counties, and because information
pertaining to the SCAB portion of county vehicle miles traveled (or vehi-
cle registrations) was not readily available, fuel sales were estimated
instead for the Los Angeles Regional Transportation Study (LARTS) area.
Moreover, fuel sales were calculated for a modified LARTS area by deleting
Ventura County and estimating sales for the following counties:
> Los Angeles
> Orange
> San Bernardino--LARTS portion only
> Riverside--LARTS portion only.
Caltrans also provided 1974 data indicating that 72 percent and 77 percent
of county vehicle miles traveled occurred in the LARTS portion of San
Bernardino and Riverside counties, respectively.
No information was available to differentiate average weekday sales
from weekend sales, so average daily gasoline sales were calculated by
dividing the annual data by 365. State wide information was available,
however, from the California State Board of Equalization (CSBE) to season-
ally adjust the sales data to June 1974. Therefore, on the basis of these
data, the estimate of gallons of gasoline sold on an average June 1974 day
in the modified LARTS area was 12.31 million.
County wide sales data for diesel fuel were not available for this
study. However, the CSBE estimates total state wide diesel and compressed
natural gas (CNG) sales on a monthly basis. No precise information was
available to subtract CNG sales from these figures, but the CSBE estimated
CNG sales to be less than 5 percent of the total figures (West, 1979).
Thus, no adjustment was made to these sales data to account for CNG sales.
111-20
-------
To estimate 1974 diesel fuel sales in the modified LARTS area, the
following equation was used:
1974 daily gallons of diesel in modified LARTS =
1974 annual gallons of / 1974 daily gallons of
diesel in state / \gasoline in modified LARTS*/ n n
19/4 annual gallons of gasoline in state^ " '
This number was then seasonally adjusted to June 1974 using the CSBE
diesel sales data described above. On the basis of these data, the esti-
mated gallons of diesel sold on an average June 1974 day in the modified
LARTS area were 1.178 million.
4) Diurnal and Hot Soak Emissions
Two additional numbers, vehicle registrations and trips, were needed
to calculate HC emissions from vehicles during diurnal and hot soak condi-
tions. Total automobile and truck registrations were calculated from 1974
county registration data compiled by Caltrans and the California Depart-
ment of Motor Vehicles, for use with the diurnal emission factor. Cal-
trans also provided the percent of vehicles registered in the LARTS por-
tion of San Bernardino and Riverside counties. Using these data,
6,149,152 vehicles were registered in the modified LARTS area in 1974.
Caltrans has estimated that 24,211,000 vehicle trips occurred daily
in the LARTS area in 1974. Adjusting this number on the basis of vehicle
registrations to account for trips occurring in Ventura County, 23,218,000
daily trips were estimated to occur in the modified LARTS area in 1974.
5) Summary of Emissions Based on Fuel Sales
On the basis of the data previously discussed, the following motor
vehicle emission estimates were calculated for June 1974 in the modified
LARTS area.
Not seasonally adjusted.
111-21
-------
Emissions, tons/day
THC 1,390
NMHC 1,330
CO 12,310
NOX 790
The grid used for this air quality modeling study is a subset of the modi-
fied LARTS area. A comparison of the motor vehicle emission estimates
derived from fuel mileage for the grid area to those derived from DTIM is
given below.
Emissions, tons/day
RHC CO N0y
Fuel mileage technique 1,260 11,890 760
DTIM 850 8,250 720
e. Run 15—1975 Gas Sales
The motor vehicle inventories for runs 14 and 15 are identical.
f. Run 16—Point Source
The baseline emission inventory supplied by the ARB contained power
plant emissions for a "typical summer week day". However, in preparing
inputs for the 26 June and 4 August base case simulations, we deleted the
nominal emission rates and, in their place, substituted the actual hourly
emission rates estimated by the Southern California Edison Company and the
Los Angeles Department of Water and Power. Power plant emissions for this
sensitivity run were based on nominal annual emission rates compiled by
the District (SCAPCD, 1976) and were temporally distributed according to a
diurnal profile developed by General Research Corporation (GRC, 1974).
g. Run 17—1974 Area Source
The objective of this simulation was to examine the need for spa-
tially resolved area and point source emission data. A total of 323 point
sources are included in the baseline inventory. Sixty-nine of these are
thermal electric power plants. These sources are identified by UTM
Reactive hydrocarbons.
111-22
-------
coordinates and various stack parameters. The rest of the emission
sources in the inventory, whether they are line sources, small point
sources, or diffuse area sources, are allocated according to a 25 km^ grid
mesh.
In this simulation, emissions from the following sources were
unchanged from the base case:
> All 323 elevated point sources
> All on-road motor vehicles
> All natural or geo-genic sources.
The emissions from the remaining sources were aggregated on an hourly
basis and allocated to the grid based upon the 1975 demographic distribu-
tion.
h. Run 18—1975 Area Source
The area source emissions files for runs 17 and 18 were identical
i. Run 19--Tempora1 Resolution
In this study, the temporal distribution of categories of area
sources was varied to examine the change in overall emission patterns and
resultant air quality modeling concentrations. For each of the source
categories constituting the area source emission file, a temporal distri-
bution of diurnal emissions was estimated from data sources described
below.
1) Temporal Distribution of Existing Emissions Inventory
The area source portion of the Los Angeles Air Quality Maintenance
Plan (AQMP) -emission inventory provided to SAI for use in this study had
been disaggregated into categories of emission sources (CES) with assigned
CES numbers. A system of codes was established to readily identify the
temporal distributions used to resolve daily emissions into specific hours
The spatial resolution of area sources was unchanged from the base case
for this run.
111-23
-------
of the day. Four coding series were used to identify the applicable
temporal distributions of emissions in the AQMP inventory.
> 0,1,2,...22,23 refers to the individual clock hours of the
day; for example, 13 means 1300 to 1400.
> Numbers in the 40 and 50 series are derived from the equa-
tion:
40 + n = 0700 + lOOn from 0700 to the final time. (3-2)
For example, 41 means 0700 to 0800 and 50 means 0700 to
1700.
> Numbers in the 70 and 80 series are derived from the equa-
tion:
70 + n = 0.5(100n) equally distributed around 1200. (3-3)
For example, 78 means 0800 to 1600.
> 98 refers to a distribution over each of the 24 hours of
the day.
In addition, each of the hours that is assigned emissions by an individual
code receives an equal distribution of the total emissions.
The AQMP temporal distributions represented by the codes were deter-
mined from survey responses of sources in the Los Angeles area and from
assumptions regarding typical source operations. As seen in table III-4,
some source categories within the AQMP inventory had emissions distributed
in only one manner (e.g., CES number 38), though other categories had
emissions distributed by more than one distribution (e.g., CES numbers 121
and 19). Furthermore, many categories had as many as 12 applicable tempo-
ral distributions, presumably as a result of responses to the source
survey. However, the AQMP inventory provided to SAI did not differentiate
the distributions that had been assumed for a source category from those
that had resulted from the source survey.
2) Temporal Distributions Developed by SAI
For each source category, an assumed temporal distribution represen-
tative of a summer season was chosen. The principle behind each choice
was to select one specific distribution for a source category that was
For purposes of this discussion, "0700" and "0700 to 0800" mean one hour
of emissions beginning at 0700 and ending at 0800.
111-24
-------
TABLE III-4. CATEGORIES OF EMISSION SOURCES AND TEMPORAL DISTRIBUTIONS
CES
Number
Emission 1
Transportation 130
• Motor Vehicle 2
• • Catalyst gasoline exhaust 14
• • Non-Catalyst gasoline exhaust 29
• • Gasoline evap. loss carb 23
• • Gasoline evap. loss fuel tank - 122
v • • Gasoline crankcase 21
• • Diesel exhaust 34
* • • Diesel evaporative 37
• • Motorcycle exhaust 123
• Off Road Motor Vehicle 38
• • Industrial 58
• • Construction 59
• • Recreational 60
• • Farm 61
• Shipping 3
• • Purging 30
• • Off loading-- - - -- 33
• • Ballasting - 28
• • Transit -- 39
• • • Boilers non-tankers 119
; ... Boilers tankers 120
• • • Pleasure craft 121
• Railroad 4
.?
• Aircraft 8
• • Jet exhaust 20
• • Jet fuel evaporation 63
• • Piston exhaust 19
• • Piston fuel evaporation 129
• • Rocket 50
Stationary 65
• Petroleum 6
• • Production 13
• • • Ext. combustion boilers 78
AQMP
Distribution
Code*
98(56)
§
**
Assumed
Distribution
Code-
112
47(100)
49
78(100)
78(100)
50
78
98(100)
98(100)
98(100)
84(67),78(29)
98(100)
98
98
98
82
98
**
**
51(85)
51
98(82)
98
111-25
-------
TABLE III-4 (continued)
CES
Number
AQMP
Distribution
Code*
Assumed
Distribution
Codet
• • Int. combustion engines - 83 98(100) 98
• • Industrial processes - 87 98(96) 98
• . Seeps - 118 98(100) 98
• • Crude oil evap. fixed roof 88 98(99) 98
• • Crude oil evap. floating roof 89 98(100) 98
• Refining 12
• • Ext. combustion boilers 77
• • • Boilers residual oil - 73 98(100) 98
• • • Boilers distillate oil 74 98(94) 98
• • - Boilers natural gas- - 75 98(83) 98
• • • Boilers process gas 76 98(100) 98
• • Internal combustion engines 84 98(100) 98
• • Industrial processes 90 98(75) 98
• • Storage evap. 91 98(94) 98
• • • Crude oil evap. fixed roof 93 98(100) 98
• • . Crude oil evap. floating roof-- 95 98(100) 98
• . . Gasoline evap. fixed roof 92 98(94) 98
• • • Gasoline evap. floating roof--- 94 98(100) 98
• Marketing — 10 98(99) 98
• • Storage evap. - — 100 98(64) 98
• • • Crude oil evap. fixed roof 97 98(100) 98
• • • Crude oil evap. floating roof-- 99 98(100) 98
• • • Gasoline evap. fixed roof 96 98(83) 98
• • • Gasoline evap. floating roof--- 98 98(100) 98
• • Loading and Unloading — 103 98(49) ,78(16) 112
• • • Gasoline evap. 101 98(59) ,78(26) 112
• • • Crude oil - 102 98(62) 112
• • Underground storage @ stations— 40 52(99) 52
• • Vehicle refueling @ stations 45
Commercial & Institutional- 7 98(100) 112
• Internal combustion engines 82 98(100) 98
• Ext. combusion boilers & space heat 124 98(100) 98
• • Residual oil 125 98(82) 98
• • Distillate oil 126 98(70) 98
• . Natural gas-- - — 127 98(78) 98
• • Process gas-- 128 98(100) 98
• Printing - 113 98(23) ,78(23) ,56(23) 112
• • Flexigraphic - - 112 78(26) ,98(18) ,56(13) 112
• • Gravure 114 98(42) ,78(17),54(17) 112
• Surface coating air dried achit.— 16
• • Oil base including solvent - 110 78(100) 50
• • Water base - Ill 78(100) 50
• Dry Cleaning 22
• • Petroleum base perchlorethylene-- 46 78(22),49(12),47(11),44(10) 50
• - Synthetic 43 78(23) ,47(15) ,46(11) ,44(11) 50
• Degreasing - 11 78(44) ,56(22) 101
• . Halogenated 42 78(26) ,56(22) 101
• • Non-Halogenated 47 78(23),56(12) 101
111-26
-------
TABLE III-4 (continued)
CES
Number
AQMP
Distribution
Code*
Assumed
Distribution
Code-
. Industrial 5 98(35) ,78(20) 112
. . Internal combusion engines 81 98(71) 112
. . External combustion boilers & heaters 49
. . . Residual oil - 69 98(64) 112
. . . Distillate oil 70 98(60) 112
. . . Natural gas 71 98(66) 112
. . . Process gas 72
. . Chemical 15 98(36) ,56(12),78(10) 98
. . Metallurgical 35
. . . Primary Metals 85 98(53) 112
. . . Secondary Metals 86 98(62) 112
. . Mineral 31 78(34),98(23) 112
. . Wood Processing-- -- 25
. . Elec. generation boiler 18
. . . Residual oil 56
. . . Distillate oil 67
. . . Natural gas 55
. . . Process gas 68
. . . Coal 57
. . Elec. generation Inter. Comb. 79 98(56) 98
. . Surface coating 44 98(60) ,44(40) 101
. . . Heat treated 48 78(27) 101
. . . Air dried — 41
. . . . Paint - 80 44(18),78(1 5) ,43(14),42(12) 101
. . . . Varnish and Shellac 104 78(25),98(18),46(11),44(11) 101
. . . . Lacquer 105 78(28),44(14) 101
. . . . Enamel 106 78(28),44(12),42(10) 101
. . . . Primer 107 78(35) ,56(16) 101
. . . . Solvent 108 78(43) ,98(15) 101
. . . . Adhesives 109 78(26),56(14) 101
. . Incineration 51 43(18),42(18) ,41 (14),78(11) 52
. . Land fills 117 98(100) 98
. Agricultural 9
. . Agricultural control burn 17
. . Vegetative forest & citrus 115 98(100) 98
. . Animal wastes 116 98(100) 98
. . Pesticides-- 24 52(100) 50
. . Food processing 32 98(38),78(19) 112
. . Orchard heating 36
. . Waste burning or wildfires 27
. . Wine processing 66
. Domestic 54
. . Solvent use 26 52(100) 134
• . Utility equipment 2 stroke 53
111-27
-------
TABLE 111-4 (concluded)
CES
Number
Utility equipment 4 stroke 52
Fuel combustion - 62
Structural fires 64
AQMP
Distribution
Code*
98(100)
Assumed
Distribution
Codet
6,7,8...21,22
* Temporal distribution codes provided with the emissions inventory. Numbers
in parentheses represent the percent of the total number of grid cells that
have emissions (for that source category) temporally distributed by that code;
only those codes and their corresponding percentages most responsible for
distributing the category's emissions are listed. See text for explanation
of codes.
+ Temporal distribution codes assumed by SAI. See text for explanation of codes,
§ No emissions associated with that CES number were provided in this portion of
the emissions inventory.
**
The temporal distribution of emissions provided with the emission inventory
was used for the assumed distribution. For CES #2, each of the 24 hours
received almost equal weighting (i.e., about 4 percent). For CES #20, each of
the 24 hours received some weighting; from 0700 through 2200 inclusive, each
hour received almost equal weighting (i.e., about 6 percent).
111-28
-------
typical of the normal operating schedules of that source type. In other
words, an individual who was familiar with typical source operations, but
was lacking site-specific operating data, would be expected to chose
temporal distributions similar to those chosen by SAI for this study. For
some source categories, no change in source operation was assumed (e.g.,
CES number 59). However, the majority of sources were assumed to have a
single temporal distribution different from the distributions used in the
AQMP inventory (e.g., CES numbers 121 and 19).
Three new temporal distributions were developed to account for the
complex nature of many source operations. These distributions were used
to accommodate the differing operating schedules of individual sources
within a category. The three new distributions and their codes, which
were developed by SAI, refer to the following hours and weighting of those
hours.
> 101 refers to a distribution of hours from 0700 to 2200;
each hour from 0700 to 1700 receives twice the weighting
of the hours from 1700 to 2200. Thus,
- 0700 to 1700: each hour receives 8 percent of the
emissions.
- 1700 to 2200: each hour receives 4 percent of the
emissions.
All other hours receive no emissions.
> 112 refers to a distribution over each of the 24 hours of
the day; each hour from 0800 to 1600 receives twice the
weighting of the hours from 1600 to 0800. Thus,
- 0800 to 1600 : each hour receives 6.25 percent of the
emissions.
- 1600 to 0800: each hour receives 3.125 percent of the
emissions.
> 134 refers to a distribution of hours from 0700 to 2100;
each hour from 0700 to 1800 receives twice the weighting
of the hours from 1800 to 2100. Thus,
- 0700 to 1800: each hour receives 8 percent of the
emissions.
111-29
-------
- 1800 to 2100: each hour receives 4 percent of the
emissions.
- All other hours receive no emissions.
The distributions chosen for each source category were based on four
primary sources of information:
> The temporal distributions used in the AQMP emission
inventory provided to SAI were partially developed from
responses to a survey of sources in the Los Angeles
area. From this inventory, as indicated in table III-4,
SAI calculated the percent of the total number of grid
cells that had emissions. These were calculated for each
source category that was temporally distributed by a par-
ticular code. This information was useful in estimating
the portion of a category's sources that operated on a
particular schedule.
> Temporal distributions that were developed by the ARB for
a study in the Sacramento, California area provided addi-
tional information on common operating schedules for many
categories of sources (Reynolds et al., 1979).
> The Regional Air Pollution Study (RAPS) (Littman, 1978)
provided temporal distributions for selected categories of
area sources.
> Area source distributions used during the development of
the Sacramento AQMP were also useful for comparison purpo-
ses. These temporal distributions were derived from sur-
vey responses and estimates of typical source operation
(Skelton et al., 1977).
In addition, SAI has been developing a comprehensive emission inventory
for a modeling study in central California. Temporal distributions deve-
loped in conjunction with that study also assisted this effort.
4. Simulations Focusing on Model Grid Mesh Configuration
Three simulations were carried out to explore the impact of model
configuration on predicted oxidant levels; these are discussed briefly.
111-30
-------
a. Run 20—10 km
For this simulation, all gridded model inputs were averaged to yield
10 km spatial resolution instead of the 5 km resolution of the base case.
b. Run 21—2-Layer Model
In this run, the vertical resolution of the model consisted of two
layers separated by a temporally and spatially varying inversion base.
c. Run 22—1-Layer Model
In this run, the vertical resolution of the model consisted of one
single layer below the inversion base.
D. CONCLUDING REMARKS
This chapter identifies the 22 model sensitivity simulations investi-
gated in the present study. Most of these simulations correspond to those
originally identified in our preliminary project work plan. Some do not,
however. In certain instances (e.g., simulations involving 2.5 km grid
resolution or motor vehicle emissions based on inadequate transportation
models), simulations were not carried out either as a result of a direc-
tive from the EPA Project Officer or because compilation of the requisite
input files was too expensive.
Some simulations discussed in this report were not originally contem-
plated. They arose from efforts to establish suitable June and August
base cases. Previous experience in developing acceptable base case
results in several urban areas—Los Angeles, Denver, St. Louis, Sacra-
mento, and Las Vegas—indicates that an iterative process of input
preparation, model simulation, and diagnostic analysis is generally neces-
sary. When simulation results fall short of the model performance goals
established in the evaluation effort, additional analyses are performed to
ascertain the sources of inadequate performance. Often, information is
111-31
-------
discovered that was not initially considered in preparing model inputs.
Thus, the base case simulation is generally preceded by several model
runs. These intermediate simulations may be regarded then, iin some sense,
as based on a lesser level of detail in input information. Consequently,
a few of the runs performed in the model evaluation effort have been
included in the sensitivity analysis presented in later chapters.
Ill 32
-------
IV BASE CASE SIMULATIONS
A. IDENTIFICATION OF ANALYSIS PROCEDURES
Several model performance measures and graphical procedures have been
developed to display and interpret the airshed model simulation results
(Hayes, 1978; Mil Iyer et al. 1979). In this section analytical procedutes
that are most informative in evaluating model performance in simulating
the base case results are identified.
Because of the vast amount of output information available from a
grid model simulation, the performance of the model may be evaluated from
a variety of perspectives. Hayes (1978) reported on a detailed examina-
tion of candidate model performance measures for air quality dispersion
models. Five attributes of desirable model performance were identified.
> Accuracy of the calculated peak concentration
> Absence of systematic bias
> Lack of gross error
> Temporal correlation
> Spatial alignment.
The accuracy of the calculated peak concentration can be evaluated in
different ways. The observed peak concentration can be compared with the
highest calculated value at a specified monitoring station, or the calcu-
lated peak value can be taken as the highest value calculated at any one
of the monitoring sites, or even as the highest value in any ground-level
grid cell. The basis for comparing observed and computed peak concentra-
tions must therefore be specified. Here the focus is on the correspon-
dence between the peak computed concentrations and observed concentrations
at each monitoring station.
Absence of systematic bias means that a model does not consistently
underestimate or overestimate pollutant concentrations. The presence of
systematic bias can be inferred qualitatively by plotting pairs of
IV-1
-------
calculated and observed concentrations on a scattergram plot. If the
locus of the prediction-observation pairs falls along a 45° line (the so-
called perfect correlation line), the absence of systematic bias in model
calculations is indicated. If the locus of the points falls above or
below the line, a systematic bias toward underestimation or over-estimation
is suggested. Obviously, it is quite possible for a model to exhibit a
bias toward overestimation in a particular concentration range and toward
underestimation in a different range. The estimates of systematic bias
are calculated in the following manner.
CALCULATION OF SYSTEMATIC BIAS:
Mean deviation = 77
i
Mean normalized deviation = TT > —=-*7 ^^- , (4-lb)
where C- and Cm are the computed and measured concentrations, respec-
tively, and N is the total number of comparisons.
Continuing with the scattergram concept, the absence of gross error
can be determined by the "dispersion" of the prediction-observation pairs
about the perfect correlation line. If the mean absolute deviation of the
pairs about the perfect correlation line is small, the model is said to
exhibit "skill" as a predictor. If the mean absolute deviation is large,
the model suffers from the presence of large gross errors. Both the mean
signed deviation (a measure of systematic bias) and the mean absolute
deviation (a measure of gross error) can be determined as a function of
concentration level. These measures can be presented either as quantities
normalized by the observed concentration level or as nonnormalized
values. The mean absolute deviation and the mean normalized absolute
deviation are given by
CALCULATION OF GROSS ERROR:
N
Mean absolute deviation
= 1 V 1C C
N Z-r 1Lc,i Cm,i
N 1C - C I
Mean normalized absolute deviation = 77 / —^ ^^— , (4-2b)
IV-2
-------
Temporal correlation refers to the "timing" or "phasing" of the
observed and computed ozone levels at a specified station. The temporal
correlation for a given station can be determined by using the pairs of
hourly observed and calculated concentrations to define daily mean
values. A correlation coefficient can then be calculated according to
routine statistics. Lack of temporal correlation can be ascribed to one
or more causes, including inadequate characterization of emissions, wind,
or mixing depth inputs.
To calculate an average temporal correlation coefficient, p , we
perform the following change of variable (Hoel, 1962):
1 l + ri
* = *n ~ ' (4"3)
where r^ is the computed correlation coefficient for Station j on the
basis or hourly pairs of predictions and observations. Next, the mean
value of the <|>.'s is estimated from
J
M
- ff E
where M is the number of monitoring stations. Since the values of <(>,• will
be approximately normally distributed, it can be shown that
(4-5)
where p is the average value of the temporal correlation coefficient
Thus, p can be determined from the following equation.
TEMPORAL CORRELATION COEFFICIENT:
P = - . (4-6)
The spatial alignment of observed and calculated concentration fields
is another useful measure of model performance. For a given hour, imagine
two concentration isopleths, one constructed from observed pollutant
IV-3
-------
concentrations and the other from the corresponding model calculations.
If one isopleth were placed over the other, the degree of spatial
misalignment would be easy to discern (at least qualitatively). Spatial
alignment can be quantified by considering a sequence of "time slices."
For instance, for a particular hou"% mean values of calculated and
observed concentrations can be computed from the ensemble of monitoring
stations. Spatial correlation coefficients can then be computed for each
hour according to routine statistics. Common sources of spatial
misalignment include discrepancies between the modeled and observed wind
velocities, inaccuracies in the emission inventory, and the treatment of
photochemistry. Estimation of the average spatial correlation
coefficient follows the procedure describe* above.
B. BASE CASE SIMULATION RESULTS
1. Accuracy of Computed Peak Concentrations
The simplest comparison that, can be made between computed and
observed ozone concentrations involves the maximum hourly average values
at each monitoring station. In table IV-'J, the peak one-hour-average
ozone predictions are compared with the h'ghest observed concentrations
for the 4 August 197!) and 26 June 1974 Los Angeles simulations. The times
at which the peaks occur do not necessarily coincide. Included in the
table are the percentage differences between computed and observed values
for each of 17 monitoring stations. The percentage difference varies from
-53 percent to +500 percent. As noted in the table, the absolute concen-
tration level of the prediction-observation pair should be considered when
examining percentage differences. Percentage differences are useful
measurements at high concentration levels; absolute differences are more
relevant at low concentration levels.
A more useful comparison of the peak concentration predictions can be
made by considering only those values above a particular level. For exam-
ple, in table IV-2 the average percentage differences between peak compu-
ted and observed concentrations are presented for:
> All stations reporting peak ozone concentrations greater
than or equal to 2 pphm.
> All stations reporting peak ozone concentrations greater
than or equal to 12 pphm (the National Ambient Air Quality
Standard).
IV-4
-------
OJ
O *""^
c 4->
QJ C
U OJ
QJ U
•4- J-
•4- QJ
•«- ex
0 •-'
QJ
u
m c '— -
£ ¥ 1
CT> L. f
<^> ^ £! a"
2; vt- Q.
O 4-> V>- *•— ••
IH 30
^H *^ ix^ CTI in in CD in ro CM in
rofM CMCM«-«- E1
o *- -c
Z OJ CX
o «" °-
° o
roino>r^«-CMa>«-inro^-
«— 1 CM t— t CM CM «— 1 CM
ro
o
f^\ CU ^""**
£ SI
C£ Q. CL
LU E CX
CO O •
CO CJ
cncr,oinro«vvorof-com
«-l CM •-< CM «— 1
o
O
Z OJ <-*
to: o 4J
c c
O OJ OJ
UJ i. O
(— OJ J-
1D U- OJ
£ £,5
o o
fx. i^> ^o o O co itf o "^ CM ro
CM ro •— < «• ro r— t o CM »-H co
i in i
0
=3 0
Z «*• c -—*
•— i r»« OJ E
x cr* y f^
< «— I OJ O.
£ vv ex
OJ ««-'—'
LU C '•-
O 3 O
ro co ro co CM ^^ ^f in in ro in
rs
Oi VO "D
S CM QJ •*-»
! -^
»_> OJ ^^
CO i^ CX
X3 *— •
o
- QJ .^
*- 4-> E
uu at.
~J ED.
CO 0 «—
2 *->
t— i ro CM «-iro t-i CM f— i
t—
en
c c
f- O
t~ f
O 4->
4-> 10
•r* +*
c ns>
o
aE
QJ
QJ
CT>
1^
^t _f
u
C l/> 10 4_ TO
r .X O OJ QJ C T3
O C i gi^ TO T? >C ••" OJ O
4-» TO (O •DCO4-»TDp
§t/1jD4->CT>QJQC+JTOi
31-VICl^fcC'<-l/IC
ONI3QJOQJOQJjrTO>-,
O i
ro ri CM c-i i i CM ro t-i i
i i i i i i i i i
•— I VD VD CO 1 1 ^- CT> in 1
1— 1 II 1
ft r~i r^ ro i i VD CO r^ i
?o ro CM CM i i i— t CM CM i
; 3 in •— < o f)^ r*> CM cri CM i
CM CM CM CM »-H «— 1 CM 1
'™- VD C5 1 ^^ C5 '"•^ CO rO ^^
vroro i«-fMrocMCMro
i i i i i i i i
*r cr* co icTicM^-covDO
i— t i t— i
•o-mr^ icr>orocr t.
l» •»- 1C
•o E m c
H3 |^L C^ 'O "C^ ^K ft)
C TO t- f— C QJ QJ
TO C O O O «T TO CD _J
4-» «O C TJ t— •»- t—
C r— -t- /O COET3C •
U.:3OCl-UJ_JCOQ£l/OZ-
..
CO
C
QJ
L.
QJ
O.
OJ
JZ •
»- »«
8
• .— t
TO
4-> K
TO
T3 *"™^
T3
O- OJ
C >
••- u
in QJ
1/1 i/i
• •*— r^ *-^.
E 0 -0
«M^ Qj
I/I >
QJ 1 U
4-> QJ
O TD JD
••- OJ O
•O 4J •
c u
T3
JC QJ
I/I L.
-------
TABLE IV-2. SUMMARY OF MODEL PERFORMANCE IN COMPUTING
PEAK OZONE CONCENTRATIONS
(Average absolute percentage difference between computed and observed values*)
QKidant^ Episode
Basis for Average Value
All stations > 2 pphm 31
26 June 1974
4 August 1975
26
All stations reporting
peak concentrations
> 12 pphm
29
25
All stations reporting
peak concentrations
> 20 pphm
27
24
* The values reported in this table are calculated as follows:
1 N
Average Absolute Percentage Difference = —
N i=
c,i
IV-6
-------
> All stations reporting ozone concentrations greater than
or equal to 20 pphm.
For both simulations, the accuracy of peak predicted ozone levels
increases slightly with increasing concentration level. For ozone
concentrations at or above the federal standard, peak predicted concentra-
tions are accurate to within 25 to 30 percent of the observed levels.
An evaluation of the four other generic model performance measures
discussed in the preceding section—systematic bias, gross error, temporal
correlation, and spatial alignment—is presented in table IV-3. In the
following sections, we provide preliminary interpretation of these
results.
2. Estimates of Systematic Bias
Measures of potential systematic bias can be calculated as either
nonnormalized or normalized quantities. The latter are normalized by the
observed concentration level. In table IV-3, measures of bias are
presented for each simulation for conditions when the observed ozone
concentrations equal or exceed the 10 and 20 pphm levels. The nonnormal-
ized bias is estimated by calculating the average (signed) difference
between pairs of computed and observed concentrations (computed minus
observed). The normalized bias is estimated from equation (4-lb).
An appraisal of systematic bias as a function of measured concentration
level can be made from figure IV-1, in which the results of the 26 June
and 4 August simulations are presented. In this figure we plot the mean
normalized deviations. The following conclusions can be drawn from the
results.
> Computed ozone concentrations for 26 June exhibit tenden-
cies toward overestimation at low concentration levels
(between 2 and 7 pphm) and underestimation at high
concentration levels (_>_ 20 pphm).
> Computed ozone levels for 4 August generally exhibit a
tendency toward underestimation throughout the entire
range of observed concentrations.
> Computed ozone concentrations exhibit greater mean
deviation for 4 August than for 26 June.
IV-7
-------
oo
LU
OL
—~t
OO
«t z
LU O
s: —i
i
r^
LU —
c:
•— ' K
<: o
u_
u_
O LU
z
>- o
f*ff Ih^J
< o
s: or
ID O
oo LL.
•
ro
i
^— *
LU
_J
m
Z
QJ QJ
^r ^-i A
E
SZ.
CL
CL
s I?
CO <^>
B »
in
i
E
JC
CL
CL
•* JS
VO Ch
(M •— •
• *
r^
i
E
JC.
CL
CL
ro CM
0 0
«3- If)
• •
«— «
1
E
SI
CL
CL
O LO
ro ^j-
ro co
* •
tr>
i
i
> to
OJ 0) E
CT> "O »•
ro 0
U • C
QJ T3 «=
> 4-> O
ro 00 Z
A A
ized-average
E
JC
CL
CL
CO
ro
•c-
Lf)
E
JC
CL
Q.
0
CO
CT>
LO
E
JC
CL
CL
CO
O
r— 1
f-.
E
jC
Q.
CL
i— <
LO
ro
cc
•
>
QJ
T3
-o
^->
OO
A
^~
O
^-
•
CM
CM
ro
•
*r
^3-
ro
•
CM
If)
ro
•
^lean absolute
deviation
> Normal ized-
average
ro
vo
CvJ
•
CM
«a-
CM
•
VO
r— 1
CM
CM
CTi
«— t
>
QJ
T3
•o
•M
OO
A
E
CL
CL
OO
VD
CNJ
•
VD
E
CL
CL
VO
CO
in
i^.
E
CL
CL
f-»
VD
CO
in
E
jC
CL
CL
VO
0
^.
co
ro
O
C
C
o
A
ized-average
E
JC
CL
CL
r— 1
f*.
CM
•
^J-
E
.c
CL
CL
ro
ro
in
Lf)
E
f"
J^.
CL
CL
VO
^a-
CM
<3-
E
jC
CL
CL
CM
"3"
r— 4
Lf)
>
QJ
•o
T3
4-)
OO
A
OJ
(J QJ
C 4J
tO 3
E -C
v. •<-
o s_
VI- 4->
4- -t->
QJ <:
D_
O —
J->
QJ ro
O E
C O)
QJ 4-J 01
I/I I/) fC
O
i-
QJ
x^ I/I
O O
<0 S_
IV-8
-------
«*
r^.
en
t— i
-4->
in
3
cr>
3
<^
1
»««4
i . i
u_j
1
CC
— i en
co in
^H CO
• •
E
x:
QL
DL
O
(\J
/\
no
O
E
XT
Q.
n.
o
^-i
/\
ro
O
E
d
CL
O
C\J
/\
CO
0
0
4->
CO ^-1
O en
tn r^
• •
o
4-J
^H ^-1
kC r-.
ur co
1
o
4->
13- r-l
ID in
* *
i
vo
CO
vo
•
O
^M
r^
•
ro
cv
«=r
-f-
10 O
"a!
ro
4->
JZ
(J
ro
LLJ
A
C
O
•r—
••->
ro
+J
CO
1
£3
«t
A
QJ
en
ITS
S-
QJ
>
ro
,^
ro
4->
ro
a.
CO
c
o
•r-
+_)
ro
'aJ
s^
o
o
t/1
-p
C
QJ
O
l£
14-
ai
o
o
s_
3
O
x:
j^
o
IO
LU
A
i_
3
O
JT
1
^
1-
O
QJ
Q.
C
O
ro ro
^ r—
O CU
D. S_
E w
QJ O
h- O
tr
•— OJ
cr
ro •—
Q. •—
tO T3
IV-9
-------
g
g
z.o
'••
0.0
0 5 10 15 20 25 90 35 40 45 5C
•2.0
" DA
Tft fflNTS 213
- B
B_D m
1 i
1 J 1 i 1
1.0
1.0
0.0
-1.0
•U0 S 10 IS 20 25 30 35 40 45 50
CiNCENTRflTliN CPPHH1
(a) 26 June 1974
2.0
'•"
£ 0.0
0 5 10 15 20 25 30 35 40 45 50
-2.0
\- *3
DflTfl P0INTS 213
1 1 1 1 1 I 1 1 1 I 1 I 1 I 1 1 1
2.0
1.0
0.0
-1.0
r2.0
0 5 10 15 20 25 30 35 40 45 SO
CINCENTRflTUN (PPHM)
(b) 4 August 1975
FIGURE IV-1. ESTIMATES OF SYSTEMATIC BIAS IN COMPUTED OZONE CONCENTRA-
TIONS AS A FUNCTION OF OBSERVED CONCENTRATION LEVEL
IV-10
-------
3. Estimates of Gross Error
The second generic model performance measure summarized in table IV-3
is the mean absolute deviation. This indication of gross error is
estimated by averaging the absolute (unsigned) difference between the
pairs of computed and observed concentrations. In table IV-3 these
measures are presented as both normalized and nonnormalized quantities.
figure IV-2 presents the mean normalized absolute deviation as a function
of measured ozone concentration level for the two simulation days. For
the August base case the mean deviation generally diminishes with increas-
ing concentration level. Typically, at low concentrations (5 to 10 pphm)
the discrepancies are about 50 to 60 percent. However, near the peak
concentration levels (about 25 pphm or higher) the discrepancies are
reduced to roughly 35 to 40 percent.
In contrast, the 26 June results suggest a trend toward greater gross
error at higher concentration levels. Above 25 pphm, the errors increase
from about 25 percent to 60 percent at 33 pphm. This increase in gross
error is primarily the result of large point sources of NOX located
directly upwind of three of the ozone monitors that recorded high ozone
concentration levels. Titration of ambient ozone in the model by direct
emission of large quantities of NOY leads to lower predicted concentra-
A
tions than if the plumes from these^sources were not immediately diluted
into the grid volume upon emission.
The two preceding performance measures—estimates of systematic bias
and of gross error—are useful in examining the extent of model bias and
the accuracies that exist for various observed concentrations. However,
the simulation results can also be viewed from an overall perspective by
considering differences between predictions and observations without
regard to concentration level. This perspective is achieved by plotting
the distribution of residuals—that is, the computed minus observed
concentrations. These distributions are given in figure IV-3.
Two attributes of overall model performance can be estimated from the
distribution of residuals:
> Accuracy is the degree of conformity of a particular model
prediction to an observed value (a surrogate for the true
Work is presently underway, under sponsorship by the Electric Power
Research Institute, to eliminate this problem through incorporation of a
sub-grid-scale reactive plume model in the airshed model.
IV-11
-------
10 15 20 25 90 35 40 45 50
gl.S
1.0
1
£0.5
£
i ' I
DATA MINTS 213
" B B B „
i I i I i I i
D • B °
i i i i
1 i 1
10 15 20 25 SO 95
CiNCENTRflTIiN IPPHH)
40 45
2.0
1.5
1.0
0.5
ro°
.0
(a) 26 June 1974
2.0
5 1*5
1-
10 15 20 25 SO 35 40 45 50
o
r>
0.5
0.0
iI • 1 «
OflTfl MINTS 213
1 I |
B B
I j. I
D " D B
B B o
I i I i I j J i I
10 15 20 25 SO 35
CIMCENTRATIfM (PPHH)
| I
2.0
1.5
1.0
0.5
40 45
?0°
.0
(b) 4 August 1975
FIGURE IV-2 ESTIMATES OF ERROR IN COMPUTED OZONE CONCENTRATIONS
AS A FUNCTION OF MEASURED CONCENTRATION LEVEL
IV-12
-------
-20 -16 -12 -8
12 16 20
i °*4
tio.s
L
£0.2
i o.i
•e
o.oj
DATA PUNTS 213
_ _
— —
•B
n *nAtB" "A0*"'* i I i 1 i 1 J °ifiPi An*m6nAl'l6ni
j(f -76 -T2 -8-40 4 6 12 16 f
0.4
0.3
0.2
0.1
0°'°
CfNCENTRRTIlN (PPMM)
(a) 26 June 1974
-20 -16 -12 -8
-4
12 16 20
£ 0.4
s
^O.S
h.
tO.2
(D
£0.1
o:
0.01
I | I | i | i | 1 | I | I | I | i | i
_ DftTfl POINTS 213
_
—
BOD
•"fto*m*irlpD°D° • ° • 1 ?D«P ? rfrmfimA 6 A 4
0.4
0.3
0.2
0.1
m.n
4
CSNCENTRRTIIN CPPHH)
(b) 4 August 1975
FIGURE IV-3. DISTRIBUTION OF RESIDUALS (PREDICTIONS MINUS OBSERVATIONS)
FOR THE OZONE SIMULATION RESULTS
IV-13
-------
concentration). The mean value of the frequency dist-
ribution of residuals has been chosen to represent the
degree of accuracy demonstrated in a particular simula-
tion.
> Precision is the degree of conformity of the ensemble of
prediction and observation pairs to a specified residual
value. The standard deviation about the mean value of the
difference distribution has been chosen to represent the
degree of precision demonstrated in a particular simula-
tion.
Overall accuracy of the ozone predictions for the two simulation days
can be estimated from the mean (or first moment) of the difference distri-
butions in figure IV-3. The standard deviation about the mean (the second
moment) is an estimate of model precision as defined above and is also
obtainable from the distributions. Analysis of the difference distribu-
tion yields the following estimates of model accuracy and precision:
0, (pphm)
Simulation Day Accuracy Precision
26 June 1974 ± 5 pphm
4 August 1975 -3 pphm + 5 pphm
On the basis of these results, one can conclude that the overall acc-
uracy of the model in predicting 0^ levels was greater for 26 June than
for 4 August.
4. Temporal Correlation
The temporal correlation coefficients for ozone in table IV-3 indi-
cate a rather broad range of values for the individual monitoring sta-
tions. At concentration levels exceeding 10 pphm, average values for the
coefficients are 0.423 and 0.686 for 26 June and 4 August, respectively.
Of the two airshed simulations, the 4 August 1975 results exhibit the
better temporal correlation. These results indicate that the "timing" of
the ozone buildup and subsequent decay, when viewed over all stations, is
rather variable. Peak model prediction at some stations are early,
whereas at others they are late (when compared with the observed
IV-14
-------
maximum). Low temporal correlation coeficients are also due to a more
rapid reduction in predicted ozone levels than in observed values.
5. Spatial Alignment
Examination of the spatial correlation coefficients in table IV-3
indicates that these values are generally smaller than the corresponding
temporal coefficients. For instance, at ozone levels greater than or
equal to 10 pphm, the spatial coefficients are 0.380 (26 June) and 0.565
(4 August). Moreover, the spatial correlation decreases with increasing
concentration level, suggesting that the location of the peak predicted
"pollutant cloud" is displaced spatially from that indicated by the
monitoring network. The 4 August 1975 results also demonstrate the better
spatial correlation.
An additional spatial measure of model performance is the "distance
distribution." To explore further this aspect of the results, every
monitoring station was examined for each hour between 1000 and 1700 PST to
determine the distance from the grid cell in which the monitor is located
to the nearest grid cell at which a prediction either equaled or "brac-
keted" the observed value. This "distribution of distances" is presented
in figure IV-4. In over half of the cases a prediction comparable to a
given measurement can be located within two to three grid cells—a dis-
tance of 10 to 15 kilometers. Such discrepancies can be caused by rela-
tively small errors in wind speed or direction input data.
Having discussed the base case simulation results in a rather general
fashion, attention is now focused on model performance at the various
monitoring stations.
C. SIMULATION RESULTS FOR SPECIFIC MONITORING STATIONS
An informative though often time-consuming method for analyzing mode'
performance is the evaluation of temporal trends in computed and observed
values at the various monitoring station locations. From this procedure
one can develop a conceptual picture of the formation and transport of
Distance distributions, as discussed here, are not intended to be used
to determine whether model performance may be judged satisfactory or
unsatisfactory. Instead, this procedure is used to provide possible
further insight into the possibility of biases in certain data input
(e.g., wind fields), the existence of steep concentration gradients, and
so forth.
IV-15
-------
v ft o tn
ID o — — —
o
£
a
r>
er
cc
CD CD
(C
r> en tn en
uj en tu uj
=> uj S S
tr
.. ju« «
5 ci
fe fe _J fe
cr
tn
o
(O
en
V) CO (T
UJ UJ Z
U (J CB
o a a
•- — uj
ct a: «n
o o • «n
ui ~
a »-
cc a ^
u i— «r>
z: m «
in
(M
r>
•a
»-
n
o
z
o ex
o in o
*> «W M
H> O
O)
C
3
-D
VO
CM
o
u
o
M
O
oo LU
•— > a.
OO HH
o
oo
o -z.
i— i O
a: H-.
O 00
a:
LU r—
CO
z: en
=> o
LU
Of.
JB
IV-16
-------
o
UJ
N
5? 0?
> >
It- li-
CS ea
o
ea o
X UJ
x i-
tn uj
UJ O X.
a: uj o
x i- et
i— uj cc
*: 03
UJ <_»
>• ec •-
« tc <9
CD CD Z
cc
cn cn
tn LU uj
UJ =) =3
_j ea ea
cc
. ^_ . .
ca ca cs s
z »- z z
o
cc
a:
en
UJ
cn OT cc
_i _j •>
_i _j
IU UJ Z
u o ea
o a a
— — uj
az oz tn
o u cc
m
CD CD
• • cn
tsi —• o
t_4
II II h-
>• en
UJ »
O I-
z cc
tr o t-
lu •— cn
x en ^
in o
rg <\j
Oi
en
-j
-i
IU
u
o
•««
1C
C9
O
cn
lu
=3
n
a
a:
oo
IU
ac
ex
IU
tn
•-«
a
iiiil
I I I I » I I I I I I I I I I I I I I
in
tn
o
LD
J0
IV-17
-------
secondary pollutants, such as ozone, throughout the simulation period.
The locations of the various monitoring stations and certain relevant
topographical features of the SCAB are identified in figure IV-5. Figures
IV-6 and IV-7 present the ozone results for the 26 June and 4 August
simulations, respectively.
Upon review of the plots of computed and measured concentrations
shown in figures IV-6 and IV-7, general and specific comments can be made
with regard to each of the two simulations. (During the following discus-
sion, the reader may wish to refer back to figure IV-5, which relates the
locations of the monitoring stations to general features of the South
Coast Air Basin.) The following comments can be made on review of the
base case simulation results.
> For the 26 June 1974 simulation:
In the San Fernando Valley predictions generally agree
with observations; computed peak concentrations fall
between 15 and 40 percent of the observed values.
- In the San Gabriel Valley (Pasadena, Pomona, and Azusa)
a slight (10 to 30 percent) overestimat>on occurs; pre-
dicted levels increase more rapidly than the observed
concentration levels.
In the San Bernardino Valley (Fontana, San Bernardino,
Redlands) there is a "quenching" of ozone followed by a
secondary ozone buildup after noon. The model under-
estimates the peak concentrations by 20 to 40 percent.
- Along the coast (Long Beach, West Los Angeles, Los
Alamitos) predicted ozone levels build up sooner and
diminish faster than the observed concentration levels.
> For the 4 August 1975 simulation:
In the San Fernando Valley peak levels at Reseda are
underestimated by nearly 50 percent; the peak at
Burbank is overestimated by only 5 percent.
- In the San Gabriel Valley (Pasadena, Azusa) ozone is
underestimated by 20 to 25 percent.
In the San Bernardino Valley (Fontana, San Bernardino,
Redlands) ozone peaks are underestimated by 25 to 35
percent. However, only at the Redlands station does
the anomalous ozone quenching effect appear.
IV-18
-------
IS)
o:
o
o
o
UJ
or
f— z:
H«4
Q oo
as <
— i
o: cc
i— h-
-------
60
50
6
12
IB
i 30
20
10
IIIIIIIIIII|1IIII|IT iIT_
- ftZUSA
- iBSERVED •
PREDICTED —
a in a
a m a m n
50
40
30
20
10
12
TIME (MiURS)
18
24°
60
50
40
x
i 30
•>
20
10
12
18
i ill i i riT i i i PI i i i i I
- BURBflNK
- 1BSERVEO
- PREDICTED
n m n i a
iltiiiiTTmam ti~
60
50
30
20
10
12
TIME CH0URS)
16
24°
60
SO
40
t 30
20
10
12
18
_ I I T I || I I I I I | I I I I I | I I I I l_
- CHI KB
• iBSERVED •
- PREDICTED
12
TIME (HBURS)
IB
so
40
30
20
10
24°
60
SO
40
i 30
20
10
12
18
24
_T I II T| IT II T ( I I I
- OBWNTHH Lft
- iBSERVED •
- PREDICTED
a
n fn n i n_
12
TIME (HBURS)
16
a nj m
60
50
40
30
20
10
24
FIGURE IV-6. CALCULATED AND OBSERVED OZONE CONCENTRATIONS FOR 26 JUNE 1974
IV-20
-------
12
16
DU
50
40
?
L
±30
r>
B
20
10
0«
Ll
irn
50
40
:30
»
20
10
n1
Lo
- EL T0RB
- IBSERVED •
- PREDICTED —
- —
•
—
•
DO ~
D j
__H^n " O O —
3.f.l-»«T-ijBJ_l 1 1 1 1
> 6 12 16 2
TIME (HiURS)
> 6 12 16 2
- LENN8X
- iBSERVEO •
- PREDICTED
- -
"
~ -
:
— —
'a m n i Jrtfa m n m n iti n rnnaru^ti a m a m n~
6 12 IB 21
TIME (HSURS)
60 60
50 50
40 40
i
30 t 30
r>
20 20
10 10
n n1
4° °!
*60 60
50 50
40 40
i
30 i 30
m
20 20
10 10
n n'
P °3
. ' ' ' ' ' 1 ' 1 ' II 1 1 1 II 1 1 i 1 1 1 1 1
- FiNTRNfl
- iBSERVEO •
- PREDICTED —
— —
B
—
D 0
O
DO y\ ~
- / n o I.
A D D I
«Jr O . * o o~
> 6 12 16 2
TIME (HIURS)
> 6 12 18 2
- L0NG BERCH I
- iBSERVED •
- PREDICTED
— —
~ -
- -
~ —
7 -
-™ m fi i r» rinrfm | if?, , ^iSLJ" ? T ? m n~
6 12 18 21
TIME (H0URS)
60
50
40
30
20
10
4"
4
bU
50
40
30
20
10
°
FIGURE FV-6 (Continued)
IV-21
-------
12
18
CD
so
40
30
20
10
°(
_ I I i i I | I I I i
- LtS BLUM IT
- IBSERVED •
• PREDICTED —
-
-
-
_
™
:, , , i-^r ' » I '
50
30
20
10
TIME CHBURS)
12 18
TIME (HBURS)
SO
40
4.
t 30
•D
20
10
J
1 6 12 18 2
I I I i I | I I I I I | 1 I I i I | I I I I l_
• LTNH88D
• IBSERVED •
• PREDICTED
-
- -
" w
—
: ^v :
: yoDB^Yiv-p ;
"n o (j i |] -i^u i \ i i 1 i i i IT»J!) n T H n o*
) 6 12 18 2
'so
50
40
30
20
10
4°
60°
50
40
30
20
10,
°
6
_l I I I I | I I 1 I
• M8UNT LEE
- iBSERVED •
- PREDICTED — —
-
_.
~
-
- D/
- /
-_. BD/
:°0nDOffl I
:, , ,,J^/ , ,
) 6
12 18 2
I | I I I 1 I | I I 1 I l_
_
—
—
-
—
_^
o I
0 m
HI •_
B ° I
\ O D
\ °
\ BOOB-
V-^s ':
i 1 i i i i i 1 i i i i i"
12 18 2
'so
50
40
30
20
10
4°
TIKE (H0URS)
TIKE (HBURS)
FIGURE IV-6 (Continued)
1V-22
-------
- iBSERVED
- PREDICTED
- 10
12 16
TIME (HBURS)
KO
6
12
J!0
KI
_i i i i TI n n i | i i
- REDLfiHDS
- BBSERVED •
- PREDICTED —
IB
rprr
D
50
i i i i i I i i i » i I i i i f t~
30
20
10
6 12 IB
TIME (HBURS)
60
50
6
12
16
t 30
20
10
_ I I I 1 T ] I
- RESEDR
L iBSERVED
- PREDICTED
Tl i j ill i \ | ill
ED D D
n m a i XL3r t i i i _ i 1 i j i i i l i m n to a'
6C
50
30
20
10
6 12 16
TIME (HBURS)
60
50
40
z
Q.
t 30
20
10
6
12
18
i I I I | I I I I l j I 1 I 1
SRN BERNflD
BBSERVED •
PREDICTED
D
6 12
TIME (HBURS)
16
60
50
40
30
20
10
FIGURl IV-6 (Continued;
IV-23
-------
60
50
40
£30
20
10
12
18
_l I II I | I
- SIHI VflLLE
- iBSERVEO
- PREDICTED
i I 1 I i I I I | I I I i (_
0 d
SO
40
;IO
V.O
!0
6 12 18
TINE CHBURS)
60
50
40
*; 30
20
10
6
12
18
24
n\\ IT T | t T I I 1 | I I I I I I
- UPLRND
- iBSERVEO »
PREDICTED
60
50
40
30
20
10
"fl T I I rt**T t i i i i I i ' i i i t T i | i i~
6 12 16
TIME tHBURS)
12
18
24
bu
50
40
X
t 30
V>
20
10
n'
_ i i i i i I i i i i i I i i i i i | i i i i l_
- WEST Lfl
• 8BSERVED •
- PREDICTED
_
-
_
_
h- g „
S^*>& ffl ffl _
t^ n [j f flUfir^ IP*' '^ *^ '' ''TfflBPr^
12
TIME (HBURS)
16
50
10
30
20
10
FIGURE IV-6 (Concluded
IV-24
-------
50
40
i 30
20
10
8 12
_ I I '1 I I | I I i i i | i
- AZUSR
OBSERVED •
REDICTED —
18
"' > I
TT$
12
18
24
n V o
i i i i f i i i i p*J i
50
40
30
20
10
6 12 18
TIME (HBURS)
so
40
IK)
I I I I l J l I I i J |
BURBRNK
•BSERVED •
PREDICTED —
12
TIME CHtURS)
n ID n
50
*0
30
20
10
18
60
50
40
x
tso
n
20
10
6 12 18 M
liiiijilll i | i i i [ i | i i i i i |
CHIN0
•BSERVED •
PREDICTED
12
18
24
50
40
30
20
10
6 12 18
TIME (HBURS)
4°
on
5ti
Itl
SO
20
10
[
IJc
_ 1 1 1 1 1 1 1 1 i 1 1 1 I 1 1 1 1 1 1
- DBWNTHN Lft
- BBSERVED •
- PREDICTED
-
-
-
\ /5\°B'OB
hl? f ? , p-fr-.' i i . r iTTTs^j ° '
) 6 12 16
I l i_
—
—
—
—
-
rj — m --
2
bU
50
40
30
20
10
f
TIME (HBURS)
FIGURE IV-7. CALCULATED AND OBSERVED OZONE CONCENTRATIONS FOR 4 AUGUST 1975
v 75
-------
«n°
50
40
SO
20
10
°C
_ 1
-
-
-
•*
-
—
:
-
—
"7
i
«n°
60
SO
40
30
20
10
«
_i
-
-
-
-
_
-
-
•
B 12 18 21. __0 6 12 16 24_
I I 1 I | I I I I i | i I i i i | i i i i i_
EL TIRB
•BSERVED •
PREDICTED — ^
—
-
^™*
-
—
a -
B °B -
B BO —
y n |" fl-*ti D f i i i 1 i i i i if T m i i i "
6 12 16 2
ou vw
50 50
40 40
X
30 t SO
s
20 20
10 10
1
4° °
I i i I i | i i i i I | i I I I i | I I I I l_
- FINTRNfl
- iBSERVED • . -
- PREDICTED
-
C -
" ^™
B
- >^™\? ~
I j \° n ;
- B / \ -
~ D J* \ ° ~
'- mmn vS/1 E B „ o"
^mol?2->. . i i i I i i i • i 1 i i i ° i-
) 6 12 16 2
DU
50
40
30
20
10
4*
TINE (HIURS) TIME (HBURS)
6 12 18 24.. ..0 6 12 16 24..
I i i I j i I I I t | I I I I i | i i I I i_
LENNBI
IBSERVED •
PREDICTED ^
—
-
••
—
—
• P — i. "\ B „ D -
ra m a i_m^rl5n? a j i V i i FTTH re m n D o~
i 6 12 18 2
OU DU
50 50
40 40
X
30 £ 30
•0
20 20
10 10
f
I I i i i | i I I i 1 | I 1 I I I | 1 t 1 1 I _
- LING BERCH
- IBSERVED «
- PREDICTED
-
—
—
_
- —
—
I n n I
i"ii D B t II jjj-ff^C^ffl *? i 1 i 9 STlnsj}] p m n m n~
D 6 12 16 2
50
40
30
20
10
t
TIME (HIURS) TIME (HIURS)
FIGURE IV-7 (Continued)
IV-26
-------
60
t SO
20
10
12
16
24
I ' ' ' ' ' I
LiS HLflMIT
•BSERVED
- WED 1C TED
D
DC D
O D Q
m n m n.
60
SO
30
20
10
6 12 16
TIME IHBURS)
24*
BO
50
40
.
i so
12
16
24
Till
i- PflSRDENR
- iBSERVED
- PREDICTED
i i i i i I i
50
40
30
12 16
T1HE IHBURS)
12
18
24
bU
50
40
ac
i 30
tt
20
10
°C
_ii II I] li II I | 1 I ii
- LYHH00D
- iBSERVED •
- PREDICTED
-
-
.
-
•| i i i j-J- 6 12
I | I i I i i_
—
—
—
—
18 2
50
50
40
30
20
10
4°
60
50
40
x
a.
S: 30
tn
20
10
12
18
24
I i I i | i i i I I | I I I
- P0M0NR
- BBSERVED
- PREDICTED
'a m g i
TIME (MiURS)
12 18
TIME (HBURS)
60
50
40
30
20
10
FIGURE IV-7 (Continued)
IV-27
-------
€0
SO
40
12
20
10
;7n-n-ri
- PRfiDB
" 1BSERVEO
- PREDICTED
6 12 16
TIME tHSURS)
^
m
50
30
20
60
50
40
SO
20
10
t 12 18
_~J I I i i | i i I i i | i I I i i |
- «EOLRNDS
- •BSERVED •
- PREDICTED —
6 12
TIME IHBURS)
Q
18
|60
50
30
20
ho
24°
12
18
50
40
S: SO
20
10
TT-rn-n
RESEOfl
•BSERVEO
- PREDICTED
I I j I I i I I j i
A\c
o
50
40
30
20
6 12 16
TIME CH0URS)
60
50
40
x
^ 30
20
10
12
18
24.
i r~r i i y i i i i i j i
- Sf)N BERNRO
- *BSERVED
- PREDICTED
6 12
TIME (HflURS)
SO
40
30
20
10
16
FIGURE IV-7 (Continued)
IV-28
-------
60
SO
40
t SO
20
10
12
16
- SIHI VHLLE
- iBSERVED
- PREDICTED
n
50
30
20
10
12
TIME (HflURS)
16
24°
60
50
40
x
5: 30
S
20
10
12
n n i j i i i i i
- UPLRND
- 6BSERVED •
- PREDICTED —
16
1 ' ' I
B D
i i i I i i i i t I i
12
TIME (HflURS)
50
30
20
10
ou
50
40
I
i 30
20
10
I
°C
_ I I I I I | I I I I i | I I I I I | I i i i i_
- MEST Lfl
- gRSERVED •
- PREDICTED
—
"" ™
— — .
—
—
:' ^ ju^TT?^7T7>^ •» ? •? » B^
I 6 12 18 2
bU
50
40
30
20
10
C°
TIHE CHiURS)
FIGURE IV-7 (Concluded)
IV-29
-------
- Along the coast and inland toward downtown Los Angeles
computed peak ozone levels are typically underesti-
mated. Moreover, predicted concentrations begin to
drop before noon though observed values remain moder-
ately high (~ 10 pphm) until 1300 to 1400 PST.
Overall, the model appears to have underestimated ozone levels in the
western part of the Los Angeles basin. However, the magnitude of the peak
ozone concentrations in this region is typically much less than that of
the basin wide peaks. The performance of the model appears to be the best
for the mid basin, between the San Gabriel and San Bernardino valleys.
Farther to the east, there is again a larger discrepancy between predic-
tions and observations. The trend toward underestimation in that location
is attributed, in part, to the current treatment of large NOX point
sources. Computed peak concentrations for the area are typically within
30 to 40 percent of the observed values, and the timing of the peak is
offset by about two or three hours.
D. EVALUATION OF GROUND-LEVEL OZONE CONCENTRATION FIELDS
The base case ozone results produced by evaluation of hourly average
ground-level ozone fields are presented here. Figures IV~8 and IV-9
present ozone concentration isopleths for both simulations for the period
from 1000 to 1600 PST. Examination of the ground-level maps leads to the
following general conclusions:
> The peak ozone levels of 4 August (1200-1300) occur two
hours earlier than those of 26 June (1400-1500).
> Elevated ozone levels of 26 June extend over a broader
geographical area than do those of 4 August.
> The Pomona-Upland area is the region of highest ozone
impact for both simulations.
> Concentration levels occurring over the western half of
the basin on 4 August are nearly one-half of those occur-
ring on 26 June.
> Both simulations exhibit steep ozone concentration
gradients along a SW-NE transect of the San Gabriel and
San Bernardino valleys.
IV-30
-------
f— vc
CVJ
10 O
u_
o
I— to
IE
I/) D-
t. O
3 tn
o •—
UJ
o>
c
O!
o
fsj
o
o> >
CD •—
10 CZ
*~* C5
•—H
u.
IV-3'1
-------
1SB3
o
cn
CSJ
•o
•o -o
01
o c
o
3
O CD
Z I
a:
C =5
~
O)
CQ
IV-32
-------
10 -o
(U
CM 3
t— C
o c
o
•^ «_>
L. - ^
O CO
Ol C3
I C
4->
O)
CO
IV-33
-------
1SH3
ID 13
O)
ro 3
o c
o
I/I O
t -^Hl~
O 00
0) •—
4-> UJ
QC
C =5
flj C3
O)
CD
1S3M
IV-34
-------
•o
c —
10 -O
Q)
ff =»
i— C
O C
O
O 00
3= I
OJ
CO
QJ
IV-35
-------
1SU3
o
;||p
c *-»
re -o
OJ
LT) 3
t— C
o c
o
t ~
a
O 00
ft) i—
c
O)
OJ
CO
UJ
a:
o
•«
1S3M
IV-36
-------
1SH3
o
o
rv
C —•»
«0 TO
CJ
VD XJ
O C
O
tfl O
t ._ ^
3
O 00
Ol •-•
JC
CC
O
QJ
CO
o
1S3M
IV-37
-------
1C03
o
IT)
f-
Oi
to
CJ
10 DC.
o
o u.
l_ Q-
s o
O l>0
a: ^
0) LU
£ Z
•M O
rvi
c o
ce
IV-33
-------
•^rfr^r^rr^^^rrrt^* ••*!!» •••••»•••»*
m s
"::;ma:;:;;;w::;H-; / \ °-
.............. _ .... .. ... ....... ... I .............. ..... ....... ..... :v.x.-::
•o
C *-*
•o -o
o>
O cr>
C =D
-------
c
rD
C\J
s-
3
o
:r
0)
-------
J.SU3
o
o
CM
2
T
•vK5'
|--:--;-':j--:-:. :,;•:• •y'jj-r;;:.^;:;:; .y::v'
o
•*«
1S3M
•O
c «—«•
(O T3
CO 3
O C
O
(/> <_>
1- —•
3
O CTi
c
o>
Ol
CD
IV-41
-------
XSW3
o
.*.>•.. 1...-.+ • • -.-1 . ...4 .. 1 • T i 4
i .. + •. i T. i. T--1....+7
<7mw.
> ^rr r r rr r »r-*vrr rvrr*M^
*
5
. _•.• •.• •:•.• . • .-. . ;. . ^ . .;.. i^-jl
:•? ;:::.::;::
? ttX
''•(•'" r •»' •' t":-• •!'• "i:':••:''t'•'•'••'*"'••
3
Jl
•o -o
0)
^- 3
«*- 4->
O C
o
IA O
L. ^^
O »
OJ ^«
*J UJ
Q£
C 3
O) O
I C
4->
0)
m
O
1S3H
IV-42
-------
1SH3
o
QJ
If) 3
o c
o
ifi O
L. ^.^
3
O O>
QJ
CD
1E3M
IV-43
-------
1SH3
o
•o
«J "O^
a: t
QC
at o
OJ «-•
X u.
4->
a>
CO
1S3K
IV-. •'
-------
E. SUMMARY
This chapter presents the results of the 26 June 1974 and 4 August
1975 base case airshed model simulations. Overall, both sets of results
agree favorably with ozone observations.
IV-45
-------
SIMULATION RESULTS
A. INTRODUCTION
Figure V-l presents schematically the relationship between input
data, model input variables, and air quality model predictions. The
objective of this study is to determine the sensitivity of airshed model
ozone predictions to the amount (or "richness") of input data. This
objective clearly differs from that of classical sensitivity analysis
studies. The latter deal with the sensitivity of a mathematical model's
predictions to variations in model parameters. To place this study in
perspective, various approaches to sensitivity analyses of air quality
models are summarized. The discussion then shifts to how the sensitivity
of the airshed model to various levels of detail in input data is ascer-
tained.
1. Review of Sensitivity Analyses of Air Quality Models
Mathematical model sensitivity analysis may be defined as the
evaluation of the deviation in model output resulting from a perturbation
in one or several model inputs. One possible approach consists in
defining a mathematical relationship that relates the extent of the model
output deviation to the perturbation in the input variables (or param-
eters). Such sensitivity analysis methods have been applied to some air
pollution models (Seigneur, 1978; Koda et al., 1979; McRae and Tilden,
1980); however, their application to complex urban air quality models
would be too costly and cannot be considered at present.
Another approach that has been widely used consists in perturbing the
model input variables and parameters one by one. This approach gives less
information ( on a "per run" basis) than the more complex mathematical
methods mentioned above, but the information obtained, if used approp-
riately, may be very useful for understanding the dynamics of the model.
For example, Liu et al. (1976) and Dunker (1980) used this approach to
evaluate the sensitivity of the airshed model predictions to perturbation
in emissions, wind velocities, vertical eddy diffusivity and mixing
height. The same technique was also used to study the sensitivity of
atmospheric chemical mechanisms to individual reaction rates ( Dodge and
Hecht, 1975; Duewer et al., 1977).
-------
In this study, the latter approach was necessary because of the
complexity of the urban air quality model. Specific perturbations have
been introduced in the input data, and the resulting changes in the model
predictions provide some information on the model sensitivity.
2. Sensitivity of the Airshed Model to Input Data
A detailed description of the airshed model is reported by Reynolds
et al. (1973). A brief overview can be described as follows.
The airshed model consists of a simulation model and several sub-
models that supply various input variables and parameters. Figure V-l
shows that the simulation model is comprised of a set of continuity
equations that describe the advective transport, turbulent dispersion,
chemical transformations, pollutant emissions, and removal of chemical
species in the atmosphere. Model input variables such as wind fields,
eddy diffusivities, reaction-rate coefficients, and emission rates are
defined either directly or by means of various submodels. Initial and
boundary conditions are estimated from air quality data and historical
intensive field measurement programs. The airshed model configuration can
vary in terms of grid size and number of grid layers. Although there are
basic theoretical constraints on the grid size (Lamb and Seinfeld, 1973;
Reynolds et al., 1973), these features may be varied according to the
airshed considered.
As suggested in figure V-l, analysis of airshed model sensitivity to
input data may be considered in two parts: sensitivity of the model
output to changes in the input variables (e.g., wind fields, diffusivi-
ties, emission rates, mixing depths) may be evaluated through successive
model simulations, each involving a prescribed set of input files; and the
sensitivity of these input variables to perturbations in the input data
used to construct them. For instance, wind field maps will show the
effect of a change in meteorological data on wind speeds and direction.
Attention is therefore focused on (1) the combination of the sensitivity
of the model to its input variables and on (2) the sensitivity of the
input variables to the basic data.
B. MEASURES FOR ASCERTAINING MODEL SENSITIVITY
An important step in quantifying model sensitivity is the definition
of specific measures. Several measures that appear to be useful include
the following:
V-2
-------
c/o
i—i
-------
> Signed deviation.
> Absolute deviation.
> Temporal correlation.
> Spatial correlation.
> Overall maximum ozone level.
> Maximum ozone statistics (peak level difference, peak time
lag).
> Dosage.
In addition, isopleths of maximum ozone deviation and ozone profiles at
air quality monitoring stations are developed to provide information on
the spatial and temporal perturbations in the model predictions.
Note that these measures, to be defined more precise!;/ in the
following subsections, are quite similar to those used to evaluate model
performance in simulating the June and August oxidant episodes.
1. Signed Deviation
The signed deviation is calculated as follows:
N' N L r
' " b>i>j
11
where Cs ^ j and C^ ^ j are the ozone concentrations for the sensitivity
case and'tfie base case, respectively, at station (or grid cell) i and for
the hour j; N is the number of stations (or grid cells), and N1 the number
of simulation hours.
Two signed deviations are calculated. The signed deviation is com-
puted for the ozone levels at the air quality monitoring stations and for
the ozone levels in all ground-level grid cells (grid cells in mountainous
areas and over the Pacific ocean are not included).
V-4
-------
2. Absolute Deviation
The absolute deviation in ozone levels is computed as follows:
(5-2)
N
1 V
N' ^
1
N
N
Z
1=1
C . . - C, . .
b,i,j
Absolute deviation in ozone concentration levels ( i.e., base case vs.
simulation) is computed for the air quality monitoring stations and for
the ground-level grid cells.
3. Temporal Correlation
The temporal correlation refers to the "timing" of the ozone concen-
tration levels computed by both the sensitivity case and the base case at
a specified station or in a specified grid cell. The temporal correlation
at a given location is determined from the hourly concentrations predicted
by the sensitivity case and the base case at a given location j. A
correlation coefficient is then calculated for each station according to
routine statistics. These correlation coefficients are normalized with
respect to the "perfect correlation line" (Hoel, 1962) by performing the
following change of variable:
where r- is the computed correlation coefficient for the station or grid
I
N
cell j. The mean value of the j's is computed for all locations:
where N is the number of stations or grid cells. Because the values of 4>j
are approximately normally distributed (Hayes, 1978), the average temporal
correlation coeYficient p is evaluated from the following formula:
(5-5)
V-5
-------
Thus, the average temporal correlation coefficient is:
(2? " l - (5-6)
exp (2 «) + 1
The temporal correlation is computed both from station and grid statis-
tics. Perfect correlation exists when p = 1.0.
4. Spatial Correlation
The spatial correlation between the concentration fields calculated
in a sensitivity run and those calculated in the base case is another
useful measure. Hourly correlation coefficients can be computed by consi-
dering the values of concentrations predicted by the sensitivity case and
by the corresponding base case for each station or grid cell. Then, the
estimation of the average spatial correlation coefficient follows the pro-
cedure described above for the temporal correlation coefficient. Two
spatial correlation coefficients are computed from station statistics and
from grid statistics, respectively.
These sensitivity measures (signed deviation, absolute deviation,
temporal correlation, and spatial correlation) may be evaluated as a func-
tion of concentration level. They are computed in this study for ozone
levels > 12 pphm (the National Ambient Air Quality Standard) and for ozone
levels > 20 pphm.
5. Overall Maximum Ozone Level
The maximum ground level ozone concentration in the basin is computed
for the sensitivity case and the base case from station and grid statis-
tics. The overall maximum ozone level and the corresponding location
(station or grid) provide an additional measure of model sensitivity.
6. Maximum Ozone Statistics
The maximum ozone levels occurring at each monitoring station may be
computed for both the sensitivity and the base case. The average peak
level normalized ozone difference may be defined as follows:
*
N
1.
N
s.l
(5-7)
V-6
-------
* *
where C ^ and C b ^ are the maximum ozone concentrations for the sensi-
tivity caSe and the'base case, respectively, at Station i. This measure
is computed for the 23 air quality monitoring stations within the computa-
tional grid. Coastal stations are principally affected by boundary condi-
tions and local emissions since the wind flow in both base cases is wes-
terly for the majority of the simulation time. Accordingly, these
stations are removed from calculation of the maximum ozone statistics.
Peak level normalized differences are computed for the remaining 16
downwind stations.
The average absolute peak time lag for maximum ozone level occurrence
is evaluated as follows:
N
1 ^ (5-8)
TT
Ts-Tb
T and Tb are the time of the maximum ozone level in the sensitivity
ind in the base case, respectively. This measure is evaluated for
where
case anc
the 23 stations and for the 16 downwind stations.
7. Dosage
Dosage measures were based on the gridded area with simulated ozone
levels above 20 pphm. Dosage is obtained by adding the number of grid
cells witn ozone levels above 20 pphm over the entire 19-hour simulation
period. This measure is computed for both the base cases and sensitivity
cases. The normalized difference of dosages provides an additional
measure of the model sensitivity:
Normalized /Dosage of sensitivity case - Dosage of base case
difference \Dosage of base case
8. Isopleths of Maximum Ozone Deviation
Quantification of the spatial changes in predicted ozone levels aids
in the interpretation of the sensitivity results. Isopleths of the devia-
tions in maximum ozone concentrations are presented. It should be noted
The coastal stations that are not included in this analysis are Costa
Mesa, El Toro, Laguna Beach, Long Beach, Los Alamitos, Redondo Beach,
and West Los Angeles.
V-7
-------
that maximum ozone concentrations may occur at the same location at dif-
ferent times. This isopleth presentation provides useful information
about the magnitude and location of changes in ozone concentrations.
9. Ozone Profiles at Air Quality Monitoring Stations
Comparison of calculated and observed ozone concentrations at various
air quality monitoring stations is another useful measure of model sensi-
tivity; accordingly, the time-varying ozone concentrations at the monitor-
ing stations are compared for the sensitivity and base cases. This
provides information on the temporal variation and magnitude of the
perturbations in ozone concentrations at various locations throughout the
basin.
To keep the display of simulation results to a manageable level, 6
stations were selected from the 23 monitoring sites for detailed discus-
sion of the sensitivity results. (Results for all stations are given in
appendix A.) The six monitoring locations are:
> Anaheim > Pasadena
> La Habra > San Bernardino
> Lynwood > Upland.
These stations were chosen because they typically present discernible
deviations in ozone concentration levels between the sensitivity and base
cases. Moreover, the stations are aligned with the wind flow trajectory
that carries the photochemical plume eastward across the basin. Where
indicated, additional monitoring stations that present interesting fea-
tures in the diurnal ozone profiles will be discussed in some sensitivity
studies.
10. Summary of Sensitivity Measures
These sensitivity measures are considered in the following discus-
sion:
> Signed deviation, absolute deviation, temporal correla-
tion, and spatial correlation of ozone levels for the
following selected station locations and grid cells.
- Monitoring station grid cells with ozone levels above
12 pphm.
V-8
-------
- Monitoring station grid cells with ozone levels above
20 pphm.
- Grid cells with ozone levels above 12 pphm.
- Grid cells with ozone levels above 20 pphm.
> Overall maximum ozone level predicted by the sensitivity
case and the base case and the corresponding location of
occurrence for:
- Station statistics (i.e., statistics computed based on
ozone predictions in grid cells containing monitoring
stations).
- Grid statistics (i.e., statistics based on ozone
predictions in all ground level grid cells).
> Normalized differences and peak time lag of the ozone peak
level for:
- All 23 stations
- 16 downwind stations.
> Dosages of the sensitivity case and of the base case, and
the corresponding normalized difference.
> Isopleths of maximum ozone deviations between the sensi-
tivity case and the base case.
> Diurnal ozone concentration profiles predicted by the
sensitivity case and the corresponding base case at some
monitoring stations.
Appendix A (bound separately) presents the detailed results for each
simulation.
C. SUMMARY OF SENSITIVITY RESULTS
Major features of the sensitivity simulations are summarized in table
V-l. Signed deviations, absolute deviations, temporal correlations, and
spatial correlations are presented in tables V-2 through V-5 for ozone
levels above 12 pphm and 20 pphm, for both station and grid statistics.
V-9
-------
TABLE V-l. SUMMARY OF SENSITIVITY STUDIES
Sensitivity
Study
1
2
3
4
5-1
5-2
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
Day of ^
Simulation
J
A
J
A
J
J
J
A
J
J
A
J
A
J
J
A
J
J
A
J
J
J
J
Change in Input Data
Reduced upper air meteorology
Reduced upper air meteorology
Reduced upper air and surface meteorology
Reduced upper air and surface meteorology
Reduced air quality
Reduced air quality and meteorology
Reduced upper air quality
Reduced upper air quality
Nitrogen oxides boundary conditions
Hydrocarbon initial condition
Hydrocarbon initial condition
Hydrocarbon speciation
Hydrocarbon speciation
Mobile sources— Older inventory
Mobile sources— Less detailed inventory
Mobile sources— Less detailed inventory
Point sources— Temporal resolution
Area sources— Spatial resolution
Area sources— Spatial resolution
Area sources— Temporal resolution
Larger grid size (10 km)
Reduction of grid cell layers --two-layer model
Reduction of grid cell layers-- single-layer model
* J = 26 June 1974, A = 4 August 1975.
V-10
-------
TABLE V-2. SENSITIVITY MEASURES FOR OZONE CONCENTRATIONS
ABOVE 12 pphm
Station Statistics
Sensitivity
Study
1
2
3
4
5-1
5-2
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
Signed
Deviation
-0.177
-0.094
-0.172
-0.146
0.000
-0.208
0.105
-0.073
-0.141
-0.344
-0.450
0.046
0.121
0.027
0.064
0.115
-0.034
0.054
0.118
-0.005
0.148
-0.007
-0.250
Absolute
Deviation
0.300
0.164
0.320
0.189
0.028
0.329
0.106
0.073
0.145
0.344
0.450
0.046
0.125
0.037
0.067
0.127
0.035
0.055
0.127
0.007
0.193
0.053
0.377
Temporal
Correlation
0.023
0.912
-0.183
0.799
0.990
-0.183
0.981
0.980
0.914
0.869
0.680
0.989
0.969
0.988
0.988
0.977
0.990
0.990
0.977
0.990
0.848
0.975
0.787
Spatial
Correlation
0.622
0.638
0.310
0.648
0.986
0.313
0.975
0.953
0.927
0.800
0.578
0.985
0.818
0.983
0.983
0.738
0.990
0.982
0.789
0.990
0.634
0.972
0.347
V-ll
-------
TABLE V-3. SENSITIVITY MEASURES FOR OZONE CONCENTRATIONS
ABOVE 20 pphm
Station Statistics
Sensitivity
Study
1
2
3
4
5-1
5-2
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
Signed
Deviation
-0.342
-0.034
-0.367
-0.073
0.003
-0.390
0.110
-0.048
-0.177
-0.384
-0.486
0.035
0.085
0.022
0.075
0.082
-0.034
0.057
0.103
-0.005
-0.037
0.006
-0.235
Absolute
Deviation
0.366
0.082
0.389
0.104
0.022
0.402
0.110
0.049
0.181
0.384
0.486
0.035
0.085
0.029
0.075
0.084
0.035
0.057
0.103
0.021
0.087
0.033
0.332
Temporal
Correlation
-0.009
0.768
-0.455
0.682
0.990
-0.402
0.983
0.982
0.933
0.984
0.280
0.990
0.961
0.987
0.989
0.955
0.990
0.990
0.963
0.900
0.982
0.989
0.790
Spatial
Correlation
-0.133
0.974
-0.185
0.937
0.986
-0.184
0.959
0.990
0.613
0.433
0.990
0.989
0.940
0.969
0.973
0.922
0.989
0.979
0.946
0.990
0.721
0.982
0.526
V-12
-------
TABLE V-4. SENSITIVITY MEASURES FOR OZONE CONCENTRATIONS
ABOVE 12 pphm
Grid Statistics
Sensitivity
Study
1
2
3
4
5-1
5-2
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
Signed
Deviation
-0.229
-0.046
-0.225
-0.107
0.006
-0.253
0.103
-0.071
-0.130
-0.317
-0.443
0.044
0.138
0.022
0.070
0.146
0.009
0.053
0.141
-0.008
0.100
-0.011
-0.067
Absolute
Deviation
0.329
0.168
0.347
0.198
0.032
0.355
0.108
0.073
0.139
0.318
0.443
0.044
0.142
0.039
0.073
0.160
0.010
0.055
0.149
0.010
0.200
0.055
0.207
Temporal
Correlation
0.134
0.878
-0.034
0.737
0.988
-0.067
0.977
0.981
0.925
0.807
0.722
0.986
0.966
0.984
0.986
0.963
0.990
0.988
0.969
0.990
0.831
0.975
0.776
Spatial
Correlation
0.376
0.396
0.247
0.450
0.980
0.265
0.061
0.965
0.896
0.800
0.806
0.985
0.923
0.983
0.9RO
0.906
0.990
0.984
0.919
0.990
0.397
0.965
0.701
-------
TABLE V-5. SENSITIVITY MEASURES FOR OZONE CONCENTRATIONS
ABOVE 20 pphm
Grid Statistics
Sensitivity
Study
1
2
3
4
5-1
5-2
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
Signed
Deviation
-0.366
-0.108
-0.392
-0.165
0.010
-0.416
0.110
-0.053
-0.165
-0.357
-0.463
0.041
0.090
0.018
0.085
0.095
0.007
0.060
0.103
-0.009
-0.053
-0.000
-0.086
Absolute
Deviation
0.378
0.171
0.443
0.187
0.032
0.459
0.113
0.057
0.169
0.357
0.463
0.041
0.091
0.034
0.085
0.108
0.008
0.060
0.107
0.012
0.150
0.041
0.162
Temporal
Correlation
0.012
0.321
-0.293
0.263
0.984
-0.253
0.972
0.950
0.928
O.RQ3
0.336
0.985
0.958
0.977
0.982
0.945
0.988
0.986
0.960
0.989
0.889
0.981
0.834
Spatial
Correlation
0.403
0.439
0.331
0.608
0.965
0.348
0.941
0.970
0.794
0.735
0.872
0.982
0.856
0.975
0.967
0.840
0.990
0.981
0.836
0.990
0.115
0.949
0.678
V-14
-------
The signed and absolute deviations for ozone levels above 12 pphm are pre-
sented graphically as well, in figures V-2 and V-3, respectively (results
of sensitivity run 22 were obtained late and are not included in these
figures).
Simulations involving a reduction in the upper air meteorological
data or a change in the magnitude of reactive hydrocarbon initial condi-
tions result in the greatest signed and absolute deviations. Such devia-
tions range from 15 to 45 percent. The change in grid size results in an
absolute deviation of about 20 percent. Other perturbations in the input
data result in absolute deviations of less than 15 percent.
Reductions in meteorological data result in the lowest temporal cor-
relations, with values as low as 0.023 and -0.183 for the 26 June 1974
simulation. Changes in the hydrocarbon initial conditions also result in
a low temporal correlation coefficient for the 4 August 1975 simulation.
The same simulations (reduction in meteorological data, change in hydro-
carbon initial conditions) also lead to low spatial correlation coeffic-
ients. The change in grid size results in a low spatial correlation
(i.e., p < 0.721), as is to be expected since the spatial features of the
model have been modified.
Table V-6 presents the ratios of the overall maximum ozone level of
the sensitivity case to the base case, along with the location of occur-
rence for station and grid statistics. The largest differences in overall
maximum ozone levels are obtained for the simulations that involve a
reduction in meteorological data or a variation in reactive hydrocarbon
initial conditions. The lowest ratio of overall maximum ozone levels is
obtained with the station statistics for sensitivity case 10 (hydrocarbon
initial conditions).
Maximum ozone statistics are presented in table V-7. The peak level
normalized differences and peak time lags are shown for the cases where
all 23 stations and 16 downwind stations are considered. The largest dif-
ference between these statistics is obtained for simulation 5-1, which
involves a reduction in air quality data. This result suggests that, for
the reduction in air quality data, the perturbation is larger at the
coastal stations.
The dosages and the corresponding normalized differences are presen-
ted in table V-8. In general, dosage is a sensitive measure of model
perturbation; it is the most sensitive for simulation 20, since the change
in grid size affects the computation of the dosage noticeably. Other
dosage-normalized differences appear to be consistent with the sensitivity
measures presented above.
V-15
-------
O
CM
CT.
CO
CM
LO
uo
1/5 o
i/) t—
z: I—
O i/>
i—»
I— Q
< i—'
t— cc
1/1 C3
Q.
O.
CM
O
CO
I/O
z
O
UJ
s- z
-------
1 1 1 1
-
oo
H-t
(— oo
oo o
• — i i — i *•
I- i-
< oo
1— «-H
00 (-
O OO
1— t
oo cr
i — __
1
1
____.
,:
uu
i i i i
o «d- ro oo i — c
D CD O O 0
CNJ
(~ >
OO
CO
r^
LD
t
^1-
2
o
cn
CO
r~~
o
CvJ
i
LD
•a-
OO
C\J
r~~
3
n Number
o
4->
(0
J
00
5:
CL
D-
C\J
UJ
CO
oo
ENTRATIOI^
* ^
•z.
o
LU
Z
O
l-v!
a:
o
a.
oo
C
i—t
i—
<=c
»-H
>
LU
O
LU
I—
=l
— 1
O
OO
m
ro
i
>
LU
Cr:
ra
i— t
u_
V-17
-------
TABLE V-6. OVERALL MAXIMUM OZONE LEVELS
Station Statistics
Grid Statistics
Simulation
Base case
26 June 1974
Base case
4 August 1975
1
2
3
4
5-1
5-2
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
Ratio Peak levels
1.000
1.000
0.718
1.038
0.710
0.988
0.997
0.683
1.109
0.927
0.804
0.632
0.523
1.023
1.128
1.020
1.101
1.139
1.000
1.067
1.171
0.988
0.909
1.019
1.054
Location
Azusa
Pomona
Upland
Upland
San Bernardino
Upland
Azusa
Upland
Upland
Pomona
Azusa
Upland
Upland
Azusa
Pomona
Azusa
Azusa
Pomona
Azusa
Azusa
Pomona
Azusa
Pomona
Azusa
Azusa
Location
Ratio Peak Levels (x,y)
1.000
1.000
0.814
1.366
0.755
0.982
1.251
1.122
1.019
1.002
0.832
0.670
1.018
1.003
1.027
1.056
1.033
0.996
1.046
0.997
1.002
0.854
1.020
0.947
33-14
30-16
30-16
11-16
30-16
27-15
19-15
30-16
19-15
30-16
33-14
33-14
30-15
19-15
30-16
33-14
19-15
11-16
33-14
19-15
11-16
33-14
20-14
19-15
14-15
V-18
-------
TABLE V-7. MAXIMUM OZONE STATISTICS
Peak Level Peak Time lag
Sensitivity
Study
1
2
3
4
5-1
5-2
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
Normalized
23 Stations
0.253
0.128
0.249
0.143
0.061
0.294
0.162
0.057
0.188
0.320
0.394
0.114
0.303
0.028
0.070
0.243
0.029
0.064
0.254
0.008
0.155
0.050
--
uiTTerence
16 Stations 23
0.258
0.133
0.260
0.121
0.031
0.266
0.144
0.069
0.187
0.354
0.476
0.092
0.313
0.029
0.082
0.214
0.020
0.063
0.241
0.001
0.112
0.043
Not Available
^noi
Stations
1.77
0.57
2.17
0.82
0.52
2.25
0.56
0.04
0.87
0.96
1.00
0.30
0.25
0.35
0.13
0.12
0.13
0.13
0.12
0.00
0.54
0.22
_ _
ir)
16 Stations
1.87
0.50
2.37
0.81
0.00
1.93
0.12
0.06
0.75
0.56
0.81
0.44
0.39
0.44
0.19
0.26
0.19
0.19
0.26
0.00
0.75
0.25
__
V-19
-------
TABLE V-8. DOSAGES
Dosage Normalized
Simulation (km2) Difference
Base case: 26 June 1974
Base case: 4 August 1975
1
2
3
4
5-1
5-2
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
10,000
8,750
5,975
8,600
7,400
6,625
10,475
7,125
14,525
4,350
6,200
2,425
450
11,375
13,975
11,125
12,450
13,900
10,125
11,925
13,750
9,700
16,000
10,425
11,525
0.0
0.0
-0.4025
-0.0172
-0.2600
-0.2439
0.0475
-0.2875
0.4525
-0.5039
-0.3800
-0.7575
-0.9496
0.1375
0.5970
0.1375
0.1925
0.5900
0.0125
0.1925
0.5710
-0.0300
0.6000
0.0425
0.1525
V-20
-------
Finally, table V-9 summarizes the results of the 22 sensitivity runs
in terms of the various measures presented in this chapter. (These
results are drawn from those presented in tables V-2 through V-8.)
D. SENSITIVITY RESULTS
In appendix A, results of the sensitivity studies summarized in this
chapter are presented and discussed. Model sensitivity is examined by
means of various measures presented in this chapter (ozone level devia-
tions, temporal and spatial correlations, comparisons of maximum ozone
levels, and dosage). The largest perturbations caused by reduction in
input data, manifested by changes in model input files, are presented.
The sensitivity of model input variables, such as wind fields, mixing
depths, and emission levels, to changes in available input data is ana-
lyzed to provide insight into how restrictions in data resources may
affect model performance. Perturbations in ozone concentrations through-
out the basin and at specific monitoring stations are also discussed.
In the following chapter, the sensitivity of the airshed model to
perturbations in available input data is interpreted based on the results
presented here and in appendix A.
For each sensitivity study, corresponding reduction in input data is
briefly introduced. A more detailed presentation of the procedure used
to reduce the input data has been made in Section III.
V-21
-------
•O 9>
ai u
N C
oo
LU
O
O
oo
cr
>-
QL
•=>
u^ ^*>
ro^-ro^-Orn-^C-^^^O— •OO^CO^HO^^O'*>
ddddoddo'ddoodddddoodddd
Uj
C
"
-------
VI INTERPRETATION OF RESULTS
A. RANKING OF DATA NEEDS THROUGH SENSITIVITY-UNCERTAINTY ANALYSIS
A ranking of airshed model data needs must take into account the
sensitivity of model predictions to input data, the cost of data acquisi-
tion, and uncertainties in the specification of model inputs. These
factors will be considered in defining a sensitivity-uncertainty index
that will provide a basis for ranking data needs.
1. Definition of a Sensitivity-Uncertainty Index.
The sensitivity of model predictions to the level of detail of input
data is important in determining input data needs. Several sensitivity
simulations have evaluated the effect on predicted ozone levels of
perturbations in meteorological, air quality, and emission data. The
results of these simulations provide quantitative information on model
sensitivity. Several measures of model sensitivity, AJ, could be consid-
ered in developing a sensitivity-uncertainty index; here consideration is
given to the absolute normalized mean deviation of ozone concentrations
above 12 pphm and the absolute normalized deviation of the air-shed-wide
peak ozone level. The former has been defined in equation (5-2).
The absolute normalized deviation of the peak ozone levels is
computed from the peak ozone levels of the sensitivity case and base case,
regardless of the grid of occurrence. Let AJ be the deviation of model
predictions resulting from input data perturbations. In this case AJ is
given in equation (6-1)
AJ =
C - C
max, s max,b
C
'max,b
(6-1)
where C_,v c and Cm,v h are the air-shed-wide maximum ozone concentration
IHaA, p [MaA, u
in the sensitivity case and the base case, respectively.
Other measures, such as the deviation in dosage, might also be
considered, but here we restrict our focus to the two measures given
above.
Vi-1
-------
In the following presentation subscript i refers to the meteorologi-
cal conditions (26 June 1974 or 4 August 1975) and subscript j to the
input data perturbed.
A sensitivity index is usually defined as the ratio of some model
output deviation to the corresponding model input perturbation. If Alj is
the perturbation in the input data, the sensitivity index for the input
data j and meteorological conditions i (26 June 1974 or 4 August 1975) is
defined as follows:
AJ. .
rij • IT1 • <6-2'
J
This sensitivity index, r^., is a measure of the effect of a change in the
model input data on the model predictions. In general, the deviation in
the model predictions, AJ, may be expressed as follows:
AJ = , (6-3)
where C is some model output measure such as the ozone level above 12 pphm
or the overall maximum ozone level.
Model input perturbations differ in many cases, according to whether
meteorological, air quality, or emission data were considered. To compare
the various sensitivity studies, it is necessary to normalize these input
perturbations. This may be done by introducing a cost index that relates
the cost AKj of acquisition of needed data (to raise the level of detail
in input data j from the sensitivity case to the base case) to the change
in input data:
AK.
i . (6-4)
Next, the sensitivity measure is normalized with respect to a cost
difference corresponding to the change in input data, because the ranking
of data needs clearly must include such information. The normalized sen-
sitivity index is then defined as follows:
S,j - Ji , (6-5)
VI-2
-------
that is,
AJ. .
(6-6)
This sensitivity index is a measure of the effect of the cost (of
improving the level of detail of input data j) on the model predictions
for meteorological conditions i. It is, however, a "perturbation--
specific" measure of model sensitivity. That is, the value of the
sensitivity index has been evaluated for a specific perturbation. The
classical sensitivity index, on the other hand, describes the sensitivity
of the system to any perturbation and is a function of the perturbation
level unless the sensitivity relationship is linear. As shown in figure
VI-1, when the sensitivity relationship is nonlinear, the "perturbation-
specific" and classical sensitivity indexes are identical only at the
point where the former has been calculated.
Actually, the introduction of a cost variable probably makes the
sensitivity relationship nonlinear, since a minimum threshold cost is
necessary before an input data file may be improved. Before that
threshold, the model is insensitive to the cost and the classical sensi-
tivity index is zero; in that range, the "perturbation-specific" index is
not representative of the model sensitivity. Therefore, when applying the
results of these sensitivity studies to a different urban area, one should
consider variations in the input data similar to those variations con-
sidered in this work. This will allow for the best use of the information
obtained from this sensitivity analysis, which provides "perturbation-
specific" sensitivity indexes. However, the general methodology employed
here should be applicable, in principle, to other areas with differing
characteristics.
The need for increasing the detail of input data should be weighted
by the accuracy available for corresponding model input variables. For
instance, an updated mobile source inventory will provide detailed
information on mobile source emissions of NOX and NMHC. However, these
emissions are uncertain by at least 15 and 25 percent, respectively; thus,
It should be noted that it would not be appropriate to conduct the
sensitivity analysis with respect to monitored data, since the detailed
effects of the model input variables are interrelated and complex, and
some misleading compensating phenomena could occur during a particular
sensitivity simulation. For instance, the enlargement of the grid size
gave better results (based upon few performance measures) than the base
case when compared with experimental data.
VI-3
-------
.SENSITIVITY CASE
SO
^ - SLOPE = 4
\J
("PERTURBATION SPECIFIC"
SENSITIVITY INDEX)
(CLASSICAL
SENSITIVITY INDEX)
AK. = 0 AK.
J J
EXISTING THRESHOLD
DATA BASE
AK*
BASE CASE
FIGURE VI-1. COMPARISON OF "PERTURBATION-SPECIFIC" AND
CLASSICAL SENSITIVITY INDEXES FOR A
HYPOTHETICAL CASE
VI-4
-------
the available emission data are clearly limited. Since data for NOX
emissions are more reliable than data for RHC emissions, they would be of
more use in improving the accuracy of model predictions.
Note that uncertainties in the model input variables as discussed
here (see figure V-l) result from limitations in a mathematical, physical,
or chemical procedure, such as the uncertainty in wind direction that
exists in the computed wind field or the uncertainty in emission data for
a detailed mobile source inventory. The uncertainty of model input
variables due to the level of detail of the input data is not considered
in the uncertainty index. It is taken into account in the sensitivity
analysis: a change in input data introduces a change (or uncertainty) in
the model input variables that leads to a perturbation in the model
output. This uncertainty in the model input variables is therefore
implicit in the sensitivity index, which could be rewritten as follows:
AJ.. AP..
'u-ST-ir • <6-7'
1J J
where AP^ is the change in the model input variable P^, that is, the
uncertainty in PJ -• resulting from the level of detail of the input data.
' J
The following uncertainty index is defined:
,,
J
6P. .
5-L1 , (6-8)
where 6P^j is the uncertainty in P^ resulting from limitations in a
mathematical or physico-chemical representation of a process. It is
defined for the input data j and the meteorological conditions i. The
model input variable P.,- ,• is the variable that most affects model perfor-
mance during the sensitivity study defined by i and j. It may be differ-
ent, depending on meteorological conditions. For example, a reduction in
upper air meteorological data seems to modify the wind field for the 26
June 1974 simulation most, whereas the wind field for the 4 August 1975
simulation was not as affected; in contrast, for the August run, the
mixing height is the most perturbed input variable.
The uncertainty index is defined as a dimensionless number. This
allows consistency among uncertainties in input variables of a different
nature (meteorological, chemical, or emission variables). Considering
dimensionless perturbations and indexes is a common procedure in sensitiv
ity-uncertainty analysis (Heinrich et a!., 1977; Dougherty et al., 1979).
VI-5
-------
The larger the uncertainty in the input variable, the less need for
additional input data. A very inaccurate parameter of the model lowers
the need for improving the input data base to compute its value (which has
an inherent uncertainty). Hence, we weight the sensitivity index S^ by
means of the uncertainty index U^,-, and a sensitivity-uncertainty inaex is
defined as follows:
(6-9)
A large value of the index E.. corresponds to a high ranking for data
needs for the set of input data j under the meteorological conditions i
for the Los Angeles urban airshed simulations.
2. Cost Estimates for Data Acquisition
Estimates of cost for acquiring data to upgrade a data base from one
corresponding to the sensitivity case to that of the base case have been
made and are listed in table VI-1 for the various sensitivity cases
considered in this analysis. The assumptions made to derive these costs
are also listed. Cost estimates are given as a range of values because of
the uncertainty in data acquisition costs and, in some cases, because of
the nature of the assumptions (which may define an upper and a lower
bound).
These cost estimates should be seen as an illustration of the proce-
dure for defining sensitivity-uncertainty indexes. Application of this
analysis to a different urban area will undoubtedly require updating and
modifying of these estimates.
3. Uncertainty Estimates for Model Input Variable^
The chemical species concentrations are computed by solving a set of
N continuity equations.
£ V c. + Ri + S. + L. i = 1, .... N
. i • ' J ^ i ' = * w i '"i "" i ''"i ' A» •••»" i V O ~ •!• U J
VI-6
-------
TABLE VI-1.
COST ESTIMATES FOR DATA ACQUISITION
(FOR THE SOUTH COAST AIR BASIN)
Input Data
Upper air aeteorology
Surface meteorology
Surface air quality
Upper air quality
1C for RHC
RHC speciation
Mobile sources.
Old inventory simulation
Mobile sources •
Gas sales simulation
Point sources
Area sources
Area sources
Improvement of Data
Upper winds, mxing depth
Surface winds, tempera-
ture
Ambient concentrations
of NO,, RHC, and 03
Upper air concentrations
of HO,., RHC and 03
Relationship to relate
RHC to THC
Distribution of HC
speciation according
to source categories
Update of an old
inventory
Creation of an inventory
More specific resolution
of emissions rates
Better spatial resolution
Better temporal resolution
Assumptions
8 week program (10-12 sampling days): 3
acoustic radars, 4 daily aircraft sound-
ings, 4 daily pibals at 3 locations.
Surface meteorological network expanded
during eight week simmer smog season.
6 meteorological stations.
Surface air quality network expanded
during eight week summer smog season.
6 air quality monitoring stations (upper
bound for permanent stations, lower bound
for temporary stations).
Eight week program (10-12 sampling days).
Early morning, morning and noon soundings
at upwind and major source regions. Night-
time sampling aloft over entire airshed.
Reactive hydrocarbon monitoring for an
eight week period during the summer smog
season. Monitoring upwind and at major
source region sites.
Lower bound: Breakdown of sources into
categories. Assume known HC speciation
for each category. Upper bound:
Literature review of hydrocarbon emissions
speciation data. Limited source testing of
major categories of emissions sources.
Old inventory available. Acquisition of
updated vehicle origin-destination data.
Routines such as MOBILE 1 and EMFAC 5
available. Creation of vehicle origin-
destination data files.
Characterization of day-specific emissions
rates.
Characterization of spatial distribution of
area sources.
Characterization of temporal profiles of
all major stationary sources.
Cost Estimate
S 50.000 - 100.000
20,000 - 30,000
75,000 - 150,000
50,000 - 125,000
75,000 - 150,000
20,000 - 100,000
5,000 - 20,000
250,000 - 1,000,000
10,000 - 50,000
50.000 - 100,000
60,000 - 150,000
VI-7
-------
where
c.j = the concentration of species i,
v = the wind vector,
K = the eddy diffusivity tensor,
R.J = the reaction rate,
S.j = the emission rates,
L.J = the removal rates,
t = the time,
N = the number of species considered in the model
Initial conditions and boundary conditions are prescribed. Boundary con-
ditions must be given at the boundaries of the airshed both at ground
level and at the top of the mixing layer.
Input variables in this system of equations are the initial and
boundary species concentrations, the wind vector, the eddy diffusivity
tensor, the rate parameters (which depend on irradiation intensity and
temperature), the emission and removal rates, and the depth of the mixing
layer. For convectively mixed summertime conditions the eddy diffusivity
is not a parameter to which the airshed model is highly sensitive. This
may be compared with wind velocity variations that may substantially alter
model predictions (Liu et al., 1976). Reaction and removal rate coeffic-
ients are not affected in the sensitivity studies considered here.
Therefore, estimates for uncertainty are restricted to the following
variables: wind velocity, species concentrations, emission rates, and
mixing depth.
The importance of wind direction on model performance depends on the
spatial distribution of emission sources. If the spatial distribution of
emission rates were uniform throughout the airshed, wind direction would
have little influence on the predicted air quality. If, however, a
strong gradient exists in the emission source distribution, the wind
direction may be a major variable. According to average emission rate
It would have some effect, however, because the airshed has a finite
size, and the amount of precursor transported to a receptor will vary
according to the travel from the boundaries.
VI-8
-------
isopleths of the Los Angeles basin, an estimate of the gradient in
emission sources corresponding to wind direction uncertainties has been
made.
Estimation of uncertainties in chemical species concentrations was
restricted to nitrogen oxides and hydrocarbons, since these are the main
precursors of photochemical smog. Uncertainties resulting from averaging
these concentrations over each grid cell were not considered. Only
uncertainties resulting from ambient monitoring were used. The uncer-
tainty in the relationship between reactive hydrocarbons and total
hydrocarbons was estimated from the experimental data used to derive the
relationship.
The model input variables considered in this analysis are listed in
table VI-2, along with the corresponding uncertainties.
4. Ranking of Data Needs
The sensitivity-uncertainty indexes were calculated as follows: For
instance, the reduction in upper air meteorological data (set 1, j =1) for
the 26 June 2974 (i=l) simulation leads to a relative absolute deviation
in ozone levels above 12 pphm,
AJn = 0.33.
A cost estimate for data acquisition AK^ = $50,000, leads to
=6.6x10
When the upper air meteorological data are reduced on 26 June 1974, the
wind field is the most perturbed model input variable and the associated
uncertainty range is 10 to 50 percent. For V^ = 10 percent, the follow-
ing value of the sensitivity uncertainty index is obtained:
1 1 -^
E = 7^ = 6.6 x 10 °.
Similarly, a lower bound is obtained with cost and uncertainty estimates
of $100,000 and 50 percent, respectively: £,, = 6.6 x 10~
-o
The results of the sensitivity-uncertainty analyses are listed in
tables VI-3 and VI-4 for 12 sets of input data for the average deviation
in ozone levels above 12 pphm and the deviation in air-shed-wide peak
Uncertainties in wind direction and wind speed depend on wind speed.
VI-9
-------
TABLE VI-2. UNCERTAINTY ESTIMATES FOR MODEL INPUT VARIABLES
Input Variable
Wind direction*
Emission sources
Wind velocity
Mixing depth
NO concentration
n
RHC concentration
RHC = function (THC)
NO emissions
HC emissions
Uncertainty
20°-50°
10-50%
0.5-1 m/sec
50-70%
6%
20-60%
25-65%
10-20%
20-30%
Reference
Tesche and Yocke (1978)
Present Study "
Tesche and Yocke (1978)
C. D. Unger (1976)
P. B. Russell and E. E. Uthe
(1978)
Burton et al. (1976)
Burton et al. (1976)
Present study
Present study
Present study
* Emission sources uncertainty is related to wind direction
uncertainty.
VI-10
-------
TABLE VI-3. SENSITIVITY-UNCERTAINTY INDEXES: SENSITIVITY TO OZONE LEVELS
ABOVE 12 pphra
Input Data
Upper air meteorology
Surface meteorology
Surface air quality
Upper air quality
1C for HC
HC speciation
Mobile source
updating inventory
Mobi le source
gas sales
Point sources
Area sources
spatial resolution
Area sources
temporal resolution
Ozone
Level
Deviation*
0.33
0.17
0.007
0.06
0.03
0.11
0.32
0.43
0.045
0.14
0.04
0.07
0.16
0.01
0.055
0.14
.01
(J)
(A)
(J)
(A)
(J)
(J)
(J)
(A)
(0)
(A)
(J)
(J)
(A)
(J)
(J)
(A)
(J)
Cost for Data
Acquisition
(dollars)
50,000-100,000
20,000-30,000
75,000-150,000
50,000-125,000
75,000-150,000
20,000-100,000
5,000-20,000
250,000-1,000,000
10,000-50,000
50,000-100,000
60,000-150,000
Model Input
Component Affected
Hind field (direction)
Emission source
Mixing depth
Bind field
Emission sources
NOX, RHC
NOX, RHC
RHC
RHC
NOX, RHC emissions
NOX, RHC emissions
NOX, RHC emissions
NOV, RHC emissions
X *
NOX, RHC emissions
Sensitivity-Uncertainty
Index1. x 106
Uncertainty
25°-50°
10-501
50- 70*
10-50*
20-60*
20-60J
20-60*
20-60S
10-301
10-30*
10-30*
10-30*
10-30*
6
2
.6
.4
0.5
4
0
1
3
4
0
2
6
0.
0,
0.
1.
4.
(J-1
- (21) -
- (4.1) -
- (1-3) -
)
66
• 6.8
3.5
- (11) - 30
.33
.5
.6
.8
.75
.3
.7
23
,53
67
8
,6
0.22
- (0.82)
- (4.D -
- (8.7) -
- (11.7)
- (2.9)
- (9.0) -
- (23) -
- (0.8)
- (1-8)
- (2.6)
- (4.5) -
- (11.5)
- (0.6)
- 2
11
21.3
- 28.6
- 11.25
35
80
- 2.8
- 6.4
- 10
11
- 28
- 1.7
J and A refer to the 26 June 1974 and 4 August 1975 simulations, respectively.
Lower bound, geometric mean value, upper bound.
VI-11
-------
TABLE VI-4. SENSITIVITY-UNCERTAINTY INDEXES: SENSITIVITY TO
AIRSHED-WIDE PEAK OZONE LEVEL
Input Data
Basinwide Peak
Ozone Level Deviation**
Sensitivity-Uncertainty
Index § x 106
Upper air meteorology
Surface meteorology
Surface air quality
Upper air quality
1C for HC
HC spec i at ion
0.188 (J)
0.159 (A)
0.365 (J)
0.243 (A)
0.018 (J)
0.121 (J)
0.167 (J)
0.330 (A)
0.018 (J)
0.005 (A)
3.76
2.3
24.3
16.2
0.2
1.6
1.9
3.7
0.3
0.08
- (11.9) -
- (3.8) -
- (66.6) -
- (44.4) -
- (0.5) -
- (4.4) -
- (4.6) -
- (9.0) -
- (1.15) -
- (0.3) -
37.6
6.4
182.5
121.5
1.2
12.1
11.1
22.0
4.5
1.25
Mobile source
updating inventory
Mobile source
gas sales
Point sources
'Area sources
spatial resolution
Area sources
temporal resollution
0.026 (J)
O.O(J)
4.3 - (15.0) -
52
0.056 (J)
0.035 (A)
0.005 (J)
0.045 (J)
0.003 (A)
0.2 -
0.1 -
0.3 -
0.15 -
0.1 -
(0.7) -
(0.4) -
(1.2) -
(1.15) -
(0.25) -
2.2
1.4
5.0
9.0
0.6
0.0
The basinwide peak ozone level, deviation is the absolute relative difference
between the sensitivity and base case basinwide peak ozone levels.
* J and A refer to the 26 June 1974 and 4 August 1975 simulations, respectively.
§ Lower bound, geometric mean value, upper bound.
VI-12
-------
ozone levels, respectively. Rankings of data needs based on the sensitiv-
ity-uncertainty indexes for the 26 June 1974 and 4 August 1975 simulations
are shown in figures VI-2 through VI-5. Note that for the average
deviation in ozone levels above 12 pphm, only the ranking of meteorologi-
cal input data (upper-air meteorology and surface meteorology) is affected
by the meteorological conditions. Although the values of the sensitivity-
uncertainty indexes of the other input data (air quality, chemistry,
emissions) vary according to the meteorological conditions, their relative
ranking is not affected. This suggests that meteorological conditions
should be taken into account primarily when considering the need for
meteorological input data.
Consider the 26 June 1974 sensitivity results for the average
deviations in ozone levels above 12 pphm. Updating a mobile source
inventory is ranked the highest. This is because of the relatively low
cost involved and because of the reasonable accuracy of mobile source
emission rates. Upper-air meteorology may be important; this depends,
however, upon the meteorological conditions. The initial conditions for
reactive hydrocarbons are also ranked high, despite the relatively high
cost of a large monitoring program. This, however, has been shown to be a
key input to the urban airshed model. Model predictions may be greatly
affected by variations in hydrocarbon initial and boundary conditions.
Reactive hydrocarbons, which are a necessary precursor of photochemical
smog, are difficult to measure accurately in ambient air; they constitute
a small amount of the total mass of hydrocarbons, which is composed
primarily of methane. They are usually determined via the relationship of
their concentrations to nitrogen oxides or total hydrocarbon concentra-
tions. However, there is considerable uncertainty involved in these
empirical formulas.
Spatial resolution of area sources, upper-air pollutant concentration
data, detailed hydrocarbon speciation, and a detailed point-source inven-
tory are generally of similar importance in model performance. The
importance of surface meteorology is of the same order, though it varies
with meteorological conditions.
Surface air quality is ranked relatively low. This is because we
assumed a 10-station network in the Los Angeles basin for the sensitivity
study. The sensitivity relationship is probably nonlinear and, if only 2
or 3 stations had been considered, the importance of surface air quality
probably would have been higher. This underscores the caveat that the
results of this sensitivity-uncertainty analysis must be considered in
light of the assumptions made in this particular sensitivity study. It
would be misleading to assume that surface air quality data have little
effect on model predictions in other applications.
VI-13
-------
10'
oo
I/)
<
o
X
QJ
•a
c
10
-5
CO
S_ C
3 >—"
O
C
QJ -r-
•1^ (tj
II
cn
o
ac
o:
•r- r— i-
*f. O O
i. »4-
J- O
QJ OJ
Q. OJ
C
o
3 i—
uo O 3
QJ in cr
(J QJ
1-CCL S_
Or- <
oo QJ
QJ (O Q.
S- Q. Q.
«S 00 ^)
C
O
•r-
•U
(O
U
QJ
Q.
oo
to
QJ
O
3
O
co
O
a.
£
o
'o
o
QJ
QJ
O
fO
4-
t-
3
OO
(O
3
cr
OJ
U
s-
3
oo
a; o
o -M
i- C
3 0)
O >
oo c
t—I
a>
•i- (O
.a 4->
o o
c
o
en in
QJ QJ
u cc:
3 i—
O ns
00 i-
O
ro O.
QJ E
S- QJ
FIGURE VI-2.
SENSITIVITY-UNCERTAINTY INDEXES--SIMULATION OF 26 JUNE 1974
SENSITIVITY TO OZONE LEVELS ABOVE 12 pphm
VI-14
-------
10
-4
,-5
oo
oo c
I-H
CU
•i- (O
o ^j
O O
FIGURE VI-3.
SENSITIVITY-UNCERTAINTY INDEXES—SIMULATION OF 4 AUGUST 1975:
SENSITIVITY TO OZONE LEVELS ABOVE 12 pphm
VI-15
-------
10
-4
X
OJ
•a
c
10
-5
a>
o
c
1/1
c
oo
10
-6
10
-7
cr
o
cr
o
"o
O
OJ
•M
0
<4_
S-
3
oo
e Source
ing Inventory
f— 4_>
•i- 03
-Q "O
O 0.
z ^>
'o
O
QJ
•F™
o
2:
s_
0
«4-
f 1
1—4
>1
4->
3
o-
S-
•^-
S-
QJ
C.
Q.
Sources
4-^
C
• ^
o
a.
c
o
4-J
-------
10
-4
X
QJ
10
-5
QJ
o>
s- c
3 QJ
o >
00 C
t—I
QJ
f—• r—
•r- CO
X) 4->
o o
c
o
o
QJ
c.
oo
c
o
(/> O
QJ on
(J QJ
s_ a:
Z5
o •—
fO -tJ
QJ ro
j- a.
et oo
FIGURE VI-5.
SENSITIVITY-UNCERTAINTY INDEXES—SIMULATION
OF 4 AUGUST 1975: SENSITIVITY TO AIR-SHED-
WIDE PEAK OZONE LEVEL
VI-17
-------
A detailed specification of transportation patterns is a rather
expensive task, and this results in a very low ranking of the need for
such input data. One may compare this result with the simple updating of
the same inventory. This shows the importance of the assumptions that
have been made and indicates that data needs will vary according to the
existing data base. Quantification of the temporal distribution of
stationary source emissions is rather expensive to obtain and has little
effect on model performance. This results in a low sensitivity-uncer-
tainty index.
If the basinwide ozone peak level, instead of the average ozone
level, is considered as sensitivity measure, the ranking of the input data
needs is slightly modified. The main changes occur for the surface
meteorological data: The peak ozone levels appear to be more sensitive
than the average ozone level to the number of meteorological stations at
ground level, and surface meteorological data are ranked the highest for
both simulation days (see table VI-4 and figures VI-4 and VI-5). The
ranking of the other input data are modified within the uncertainties of
the sensitivity-uncertainty index values and no other notable changes are
observed.
5. Conclusions
The sensitivity-uncertainty analysis that has been presented should
be seen as a procedure for defining input data needs, and the results
obtained in this section should be considered as an illustration of the
method. It is strongly recommended that revised costs and uncertainties
specific to the photochemical model application area be estimated.. These
city-specific cost and uncertainty estimates can then be used with the
results of the sensitivity studies (Aj's) to compute the sensitivity-
uncertainty indexes.
The types of results obtained are presented in figure VI-6. The
hypothetical cost estimates shown in table VI-5 were used along with the
uncertainty indexes of table VI-2. The priorities for acquisition of
additional input data could be inferred from such information on data
needs.
It should be noted that some limitations exist in the results of the
sensitivity studies, though their extent may be difficult to estimate.
These limitations result from the specificity of the Los Angeles bas^n,
the length of simulation time, and the model characteristics (such as the
treatment of point sources or the wind field submodel). These issues will
be considered in the following section.
VI-18
-------
10
-4
10
-5
x
0)
(C
•M
OJ
10
-6
OJ
oo
10
-7
I I I I I
26 June 1974 Simulation
I
4 Auqust 1975 Simulation i
J.
c
o
o
o
OJ
QJ
^r~
J-
• ^
«t
i.
(U
Q.
CL
=>
O
1C
o:
j_
o
1^.
o
t—t
o
2
O
QJ CO
O Q^
J- CC.
3
O i —
oo ra
•f—
fl3 4-^
ai ra
V. CL
CL
D.
o
"o
s_
o
4_)
^
OJ
u
V4_
i-
3
oo
4/1
O)
o
3
O
00
ai
^.
•r>
/"i
o
tn
QJ
O
S-
3
O
oo
•*->
c
•r—
O
D.
^^
«C
3
c-
S-
•r-
(D
U
-------
TABLE VI-5. HYPOTHETICAL COST ESTIMATES FOR DATA ACQUISITION
Input Data
Assumptions
Upper air meteorology 8 week program
Cost
Estinate sensitivity-uncertainty
(dollars) Index x 10&*
60,000 11 - (25) - 55 0
4.0 - (4.8) - 5.7
Surface meteorology 6 stations
20,000 0.7 - (1.6) •• 3.5
6 - (10) - 17.0
Surface air quality 6 stations
Upper air quality 8 week program
1C for RHC 8 week program
75,000 0.7 - (1.2) - 3.0
100,000 1.8 - (3.2) - 5.5
100,000 5.3 - (9.2) - 16.0
7.1 - (12.3) - 21.5
RHC Speciat ion
Mobile sources
Known HC speciation
by categories
No inventory available
40,000 1.9 - (3.2) - 5.6
5.9 - (10.0) - 17.4
250,000 0.9 - (1.6) - 2.8
2.1 - (3.7) - 6.4
Point sources
Area Sources
Area sources
SO; Emissions
Characterization of
emission rites
Characterization of
spatial distrioution
Characterization of
temporal distribution
Data for point source
inventory available
40,000 0.8 - (1.4) - 2.5
80,000
120,000
20,000
2.3 - (4.0) - 6.9
5.9 - (10) - 17.6
0.3 - (0.5) - 0.£
Lower bound, geometric mean value, upper bound
VI-20
-------
B. GENERALIZATION OF RESULTS
The sensitivity studies of the airshed model have been carried out
based on one-day simulations of the Los Angeles basin. Because these
results may be of general interest to the photochemical modeling commun-
ity, it is helpful to define the specific features of the simulations and
to evaluate the limitations of the sensitivity results. Then it will then
be possible to extend the conclusions of this study and to outline general
recommendations for the application of the study results.
1. Specific Attributes of the Simulations
The results of the sensitivity simulations depend on the perturba-
tions introduced in the input data, the area chosen for the study, the
atmospheric conditions, and the modeling conditions (e.g., type of wind
model and length of simulation time). Limitations of the analyses have
been discussed in the previous section and the effect of aerometric
conditions has been considered in several sensitivity runs carried out
under different meteorological and air quality conditions. At this point
it is appropriate to consider the effect on sensitivity results both of
the specific attributes of the Los Angeles basin and of the modeling
conditions.
a. Specific Attributes of the Los Angeles Basin
The meteorology of the western coast of the United States is strongly
affected by the presence of the North Pacific anticyclone. The existence
of a subsidence inversion layer above the marine layer, resulting from the
adiabatic warming of descending air, leads to the trapping of air pollut-
ants under an inversion layer that lies typically at an elevation of about
500 to 1000 meters. The presence of a temporally and spatially varying
mixing height in the Los Angeles basin affects the generality of the model
sensitivity results, since it probably enhances the need for upper-air
meteorological and upper air quality data.
The Los Angeles basin, because of its location, is also characterized
by a coastal-desert meteorology. The land-sea-breeze effect is enhanced
by the presence of warm inland areas such as the Mojave desert, and this
recirculating meteorology largely determines the atmospheric transport and
dispersion of pollutants in the basin.
It should be noted that this land-sea-breeze effect is observed in
many locations. It has been suggested, for instance, that air pollution
episodes observed in Milwaukee are induced by a land-lake-breeze effect
VI-21
-------
(Lyons and Cole, 1976); experimental and modeling studies in the Tampa/St.
Petersburg area have retrieved the main features of the land-sea-breeze
effect (Liu et al., 1979).
The location of the Los Angeles basin is such that pollution observed
is a result of local emission of primary pollutants. This may not be the
case for other urban areas where pollution episodes may be, in part, the
result of pollutant emissions that occurred far upwind. Usually, the
importance of the boundary conditions depends on meteorological cond -
tions. For instance, a steady westerly flow can occur in the Los Angeles
basin for several days (e.g., the air pollution episode of 4 August
1975). In this instance, the boundary conditions will have little
importance, since the air flowing into the airshed should have typical
background concentrations of pollutants. During the land-sea-breeze
regimes, however, polluted air is carried offshore during the evening and
the aged urban air mass is carried inland the next morning. In this case
the definition of the western boundary conditions of the airshed is
important. For other urban areas, various meteorological conditions can
also affect the importance of air quality data at the boundaries of the
airshed.
The topography of the Los Angeles basin is characterized by complex
terrain. This feature will be discussed in more detail in the following
section, since it determines the selection of the wind model. In general,
the complex terrain increases the need for meteorological data.
The air quality network in the Los Angeles basin is dense and wide-
spread; such a rich air quality data base is not available in other urban
areas. This has been emphasized in the previous section on the ranking of
data needs, and it should be taken into account in determining the
importance of air quality data from these sensitivity studies.
Primary pollutants in the Los Angeles basin are emitted main'Iy from
mobile sources, which comprise about 44 percent of the reactive hydrocar-
bon emissions and 50 percent of the total nitrogen oxides emissions. This
should be considered in the generalization of those sensitivity stud es
that involve modifications of the emission inventories.
b. Modeling Condition^
The modeling conditions affect the performance and sensitivity of the
model. Some issues, such as the grid size or the number of grid layers,
have been addressed in the sensitivity analyses (sensitivity cases 20 and
22). Here some aspects of modeling conditions that may affect the
generality of the results are considered.
VI-22
-------
The wind model is an important component of the urban airshed model,
and the use of various wind models (interpolated wind model, two-dimen-
sional, and three-dimensional wind models) has been investigated (Reynolds
et al., 1979). In this study, a three-dimensional wind model developed
for complex terrain was used (Yocke et al., 1979). The wind field is
obtained from the solution of a parameterized three-dimensional boundary
value problem. Therefore, this wind field preparation procedure is more
sensitive to atmospheric data at the boundaries of the airshed and at the
top of the mixing layer than are other procedures based, for instance, on
interpolation by inverse weighting.
The mixing height and eddy diffusivities depend on time and loca-
tion. The computational procedures used to determine these atmospheric
parameters have been presented elsewhere (Reynolds et al., 1979; Killus et
al., 1980). Although other procedures could be considered, the sensitiv-
ity of the air quality model to computational procedures for the mixing
height and eddy diffusivities has not been investigated.
The limitation of model sensitivity results for emission inventories
stems primarily from the averaging of emissions over the grid cells.
Although an attempt to model microscale phenomena for line sources has
been made (Reynolds et al., 1979), it was not used in these simulations.
The scavenging of ambient ozone by nitrogen oxides emitted from power
plants and automobile exhausts is a well-known phenomenon in air pollu-
tion. The present treatment of these sources in the urban airshed model
may obscure, in part, the need for greater accuracy in the corresponding
emission inventories and, possibly, in other parameters (e.g., a finer
resolution of atmospheric dispersion.
The length of simulation time has some effect on the sensitivity
results. For example, initial conditions at 0400 may represent a sizeable
fraction of the total precursor emission burden between 0400 and the time
of the ozone peak (1200 to 1400). If uncertainties exist about the
boundary conditions of the pollutant concentrations, these uncertainties
may affect the model sensitivity more than the uncertainties in emission
levels will affect it. However, if a multiple-day simulation were
performed, the influence of initial conditions on ozone levels would be
markedly reduced (particularly for land-sea-breeze regimes), and the
emission inventories would have a dominant role, since they would repre-
sent most of the precursor mass. Clearly, there is a need for analyzing
the sensitivity of the urban airshed model for a multiple-day simulation,
to assess the importance of air quality data and emission inventories to
model predictions under such conditions.
VI-23
-------
2. Limitations of the Sensitivity Results
Limitations of the sensitivity results that derive from the specific
attributes of the Los Angeles basin and the length of simulation time are
presented in table VI-6. The importance of the air quality data and emis-
sion inventories to model predictions should be strongly affected by the
length of simulation time. For the reasons mentioned above, when the
simulation is extended from a single-day to a multiple-day simulation, the
importance of air quality data will decrease, whereas the~effect of emis-
sion inventories on model predictions will increase.
In sensitivity cases 1 and 2, the number of upper air meteorological
stations was reduced. The existence in the Los Angeles basin of a strong
Inversion that varies with time and location enhances the need for upper
air meteorological data. Although the need for upper air meteorological
data in other urban areas depends on local meteorology, air pollution
episodes are often associated with strong inversions. This would make the
results of this study representative of a typical urban airshed simula-
tion. Moreover, long-range transport of pollutants trapped in the rear
part of an anticyclonic system above the inversion height is an important
air pollution phenomenon in midwestern and eastern urban areas. To
properly account for such phenomena, upper air meteorological and air
quality data would be needed.
Clearly, the spatial variation of the mixing height in midwestern and
eastern urban areas may be negligible in comparison, and one upper air
meteorological station should generally provide the information des;red.
In these sensitivity studies, the ranking of meteorological input data
depends on the meteorological conditions. The use of a three-dimensional
wind model for the Los Angeles basin probably emphasizes the need for
upper air data as well. This model depends on boundary values, and the
reduction of input data at the upper boundary of the domain probably
affects the accuracy of this model. A flat urban area would not require
the use of this wind model; consequently, a simple interpolation wind
model would suffice, reducing the need for upper air data.
In sensitivity cases 3 and 4, surface meteorological data and upper
air meteorological data were reduced. The density of the surface meteoro-
logical network in the Los Angeles basin must be taken into account when
interpreting the sensitivity results. In the previous section on the
ranking of data needs, it was pointed out that the intensity of input data
in the base case and in the sensitivity case was a major factor in the
sensitivity results.
The same considerations apply to the reduction in air quality data
intensity. The density of the aerometric network in the Los Angeles basin
330R/2
VI-24
-------
Of Q E O>
QJ -P- i—
u. to crt
jf -
a m
ii;
li
OJ 0
Q. «A
£*
oo
z
UJ
O -i- tft
Q.X3 •—
ESP
II
•ft VI
isj O T3
L- C« QJ
O C •<--
.CJ C C
E - O
re E •*- CD
t i_ CE
T3 QJ QJ «E
QJ 4-* Q.
•— QJ QJ
••- -O i/) .C
fO C •*->
•*-» tJ O
QJ Z r- >,
O CC 4J .0
4J C *-"
c •— c
O E t-
a: vi en
cn Q
— c t-
- i—
u o c ai
a> E o —
O
HH
5
CO
OO
O
D-
I —
— O
f> 1-
,
QJ cn
-o o
QJ U W> OJ
m >
C
QJ «-
o -•-
t— cn
ID cr> O
1C
CO
<:
•— T3 Q. 3
irt ^ 01 ««-
^ 4> QJ -O M-
oi — ••- —
•O £t *+- CT»T3
b *o c
SI - T3 - 5s
*. c x -o
*ol — •- T3
• CT)TD
c
^ — X
c x -o
— o
QJ -o -f-
> C 4->
at *a ••-
o — x
1~ c L.
(J -F- *0
J3 C
O
•O •—
C 4-»
•r- C
X O
O U
O> U tn
o> *o c
O TD O
C ••- -O O
3 *j e -
O — 3 *->
l_ C O -r-
-
I rt I >,
-*- c o c
C - I- O
Cn**- ng _Q QJ
•O QJ L. C
E — • «3 QJ
O t/i (_> r- M
OJ C >> C
> **- -^ - .c O
•<- O **- I/I 4-» •-
O) 3 (D T C
QJ
-o c i- c
3 TJ O O
4-> *-* •«-
•f • O i/i
C C E irt
•
•a o 4-> ^~
C irt C tfl
10 e o r
«_al
3 T3 V-
4-» I- O (U
— O <_>
c Q. e L.
Cn E O 3
QJ C
t- O
m -t-
**- u^
°E
s*
-i- qj
4-J U
« 1-
O 3
O O
:ei t.
o.«I5 -SSC
C *J « ^OJ -- u -O
5 ji £'
~ c -
n O 3"1
Q'O ^ C
5 ID 4-> O
(S cn E ^>
Of
QJ —
CL «j
a. 3
*_i O
U 1-
-
_o o c:
£ l/» -F-
O C i-
« C QJ t> -o
t) 3 > *- -
L. O C = *-
~ )
VI-25
-------
allows for a certain reduction in the number of air quality monitoring
stations without inducing any major perturbations in the model predic-
tions. It would be misleading, however, to conclude that the need for air
quality input data is low, since the results of the sensitivity analysis
depend on that intensity of data in both the sensitivity and the base
case. That is, the results depend on the number of air quality stations
used for the Los Angeles basin simulations.
The need for upper air quality data depends strongly on the existence
and intensity of the inversion. It is probable that areas with weaker
inversions would not be as sensitive to upper air quality data as is the
Los Angeles basin. However, air pollution episodes are often induced by
the existence of a strong inversion, and the results obtained for the Los
Angeles basin should be fairly representative for the application of the
urban airshed model to the simulation of urban air pollution episodes.
The length of simulation time has a strong influence on sensitivities to
both surface and upper air quality data. These input data determine the
initial conditions for the continuity equations of the air quality
model. Hence, their importance will decrease if the simulation is carried
out over a longer period of time, and the model predictions will become
more sensitive to emission data.
In sensitivity case 8, the boundary conditions for hydrocarbon and
nitrogen oxides were modified. The length of simulation time would
probably affect the importance of this perturbation on the model predic-
tions, as mentioned above. The location of the airshed will affect the
importance of the boundary conditions to some extent. For the Los Angeles
basin, the model is particularly sensitive to the western boundary condi-
tions during land-sea-breeze regimes. If the model is applied to an urban
area affected by pollutants transported from emission sources located
outside of the modeled area, it also will be sensitive to boundary condi-
tions, and air quality data needs at the airshed boundaries may be higher
than they are for the Los Angeles basin.
The importance of the initial conditions for reactive hydrocarbon
concentrations has been investigated in sensitivity cases 9 and 10. The
length of the simulation time strongly influences the importance of these
initial conditions. It should be noted that the initial conditions for
reactive hydrocarbon (RHC) concentrations correspond 1) to the estimation
of reactive hydrocarbon concentrations from an estimated RHC/NOX ratio
(sensitivity case), and 2) to the more accurate estimation of the RHC
concentrations from detailed atmospheric chemical data obtained by gas
chromatography on reactive and total hydrocarbon (THC) levels (base
case). Reactive hydrocarbon concentrations are difficult to obtain on a
routine basis, since they constitute a minor part of the total hydrocarbon
VI-26
-------
concentration, which consists mainly of methane. Consequently, concentra-
tions of RHC, which are obtained from the difference between THC and
methane concentration (rounded off to the nearest ppm), are highly
uncertain. Reliable data may be obtained by gas chromatography, but such
measurements are not presently made on a routine basis. Detailed ambient
reactive hydrocarbon data are available for the Los Angeles basin, and
this may not be the case for other urban areas. However, this factor is
taken into account in the cost index for the sensitivity-uncertainty
analysis performed to rank data needs.
In sensitivity cases 11 and 12, the hydrocarbon speciation was
averaged over all source categories. Clearly, the effect on model predic-
tions depends on the diverse nature of source categories. In the sensi-
tivity results, it appeared that refinery emissions were perturbed,
whereas the dominant mobile source emissions were not modified notably.
Higher levels of olefins, aromatics, aldehydes, and ethylene were observed
in the refinery emissions in the sensitivity case. These hydrocarbons are
oxidized faster than paraffins in the atmosphere, and this leads to higher
ozone levels downwind of the refineries. The importance of these input
data could be evaluated by careful identification of the various source
categories.
The length of simulation time (single-day or multiple-day) is clearly
an important factor for this sensitivity study and the following ones,
which involve emission inventories (cases 11 through 19).
The mobile source inventory was perturbed in sensitivity cases 13,
14, and 15. The importance of mobile source emissions in the Los Angeles
basin may enhance the importance of the corresponding inventories.
However, major urban areas present a similar pattern, and the sensitivity
results should be fairly representative of large urban communities.
The point source emission inventory was perturbed in sensitivity case
16. The contribution of power plant emissions to the air pollution burden
is small for the Los Angeles basin. The importance of this inventory may
be slightly larger for an area with a large number of power plants and
smelters.
In sensitivity cases 17 and 18, the area-source inventory was
spatially perturbed. No specific features of the Los Angeles basin appear
to limit the generality of these results notably.
The temporal resolution of area sources was perturbed in sensitivity
case 19. Mobile sources are dominant in the Los Angeles bas
-------
The enlargement of the grid size in sensitivity case 20 leads to an
averaging of the model variables. The spatially variable meteorological
fields of the Los Angeles basin are probably affected by this averaging
procedure. However, this sensitivity study should be fairly representa-
tive of the general effect of the grid size on an urban airshed model.
The number of grid layers was reduced from four to two in sensitivity
case 21. Specific features of the Los Angeles basin are the small number
of elevated sources and the spatially varying mixing depths. These
factors may affect the nature of the perturbations, but this sensitivity
study probably gives a reasonable evaluation of the effect of the vertical
smoothing of concentration and meteorological variables on air quality
predictions
3. Generalization of the Results
It has been pointed out that the nature of the sensitivity analyses
considered in this study is "perturbation-specific." Therefore, it is
important to apply the sensitivity results within the context from which
they have been derived. This is particularly apparent for the sensitivity
case concerning surface air quality. The low ranking of the need for air
quality data results from the density of the air quality monitoring
stations in the Los Angeles basin; it does not imply that air quality data
are a negligible component of the model.
In the presentation of the results of the sensitivity-uncertainty
analyses, the use of cost estimates and uncertainty estimates specific to
the urban area should be considered. This provides for optimal use of the
methodology developed in this study, since these estimates may vary
according to the urban area considered.
The limitations of the sensitivity analysis that result from the
specific attributes of the Los Angeles basin and the modeling conditions
have been discussed in the previous section. These limitations should be
considered as a means for adjusting the results of the sensitivity studies
to different urban areas. It is possible to use the information obtained
for the Los Angeles area in another location by evaluating the specific
attributes of both areas.
In conclusion, the results of the Los Angeles sensitivity stud es
using the airshed model may be applied in a general context as long as
proper attention is paid to the features both of the basin and the urban
area to be considered.
VI-28
-------
VII CONCLUSIONS
Because of the large amount of input data (meteorology, air quality,
and emissions) necessary to run grid-based photochemical models, it is
important to identify the input data that most affect model performance as
well as the amount and quality of data that should be acquired before
carrying out air quality simulations.
This study investigated the sensitivity of the SAI Urban Airshed
Model to the amount (and consequently the quality) of the input data in
order to develop a general procedure for specifying the meteorological,
air quality, and emission data most needed to achieve good model perfor-
mance. The approach to sensitivity analysis used in this study and the
general procedure developed to define a ranking of input data needs is
summarized in this chapter. The major results obtained from the simula-
tions of the Los Angeles basin and possible limitations are then dis-
cussed. Finally, suggestions on further studies to improve the knowledge
of photochemical model sensitivity to various levels of details in input
information are presented.
A. GENERAL PROCEDURE FOR SENSITIVITY ANALYSIS
The sensitivity of the airshed model to the level of detail of the
input data was analyzed for the Los Angeles basin under different meteoro-
logical and air quality conditions (26 June 1974 and 4 August 1975). The
effects of perturbations in the meteorological, air quality, and emission
data, as well as in the model structure (grid size and number of grid
layers), are analyzed and discussed in detail in Chapter V. The general-
ity of specific numerical results presented in this study may be limited
owing to the specificity of the Los Angles basin and modeling condi-
tions. However, the methodology used to analyze the sensitivity of the
model to the amount of input data is generally applicable and constitutes
a framework for the generalization of the results of this study and the
definition of future sensitivity studies.
The type of sensitivity analysis considered in this study differs
from classical sensitivity analysis: the concern is with the sensitivity
of the ozone predictions of the model (which consists of the air quality
VII-1
-------
model and several input generation models, such as the wind model, eddy
diffusivity computations, and emission inventories) to the quality of
input data supplied (e.g., number of meteorological or air quality
stations, level of detail of emission inventories). Classical sensitivity
analysis deals simply with the sensitivity of a mathematical model to its
parameters and initial conditions (e.g., Liu et al., 1976).
It was shown in section V.A that the overall sensitivity of the model
to input data can be considered as the combination of the sensitivity of
the model to the appropriate input variables (e.g., wind velocity, mixing
depth, emission term) and the sensitivity of these input variables to
perturbations in the input data. For instance, the sensitivity of the
ozone predictions to the number of surface meteorological stations was
analyzed by considering the sensitivity of the wind field to the number of
meteorological stations and the sensitivity of the ozone levels to the
wind field (i.e., wind directions and wind velocities). It was thus
possible to evaluate the dynamics of model sensitivity in detail and to
assess the possible limitations of the results. Then, the analysis of the
sensitivity results was used as a basis for outlining general recommenda-
tions for the acquisition of input data (see section VLB).
Several measures involving ozone predictions were used to evaluate
the sensitivity of the model to perturbations in the input data:
> The signed deviation
> Absolute deviation
> Temporal and spatial correlations (for ozone levels above
12 and 20 pphm and for both station and grid statistics).
> Overall maximum ozone level
> Maximum ozone statistics (peak levels and peak times).
> Dosage (area with ozone level above 20 pphm).
Whereas results obtained from these various sensitivity measures reflects
the specific features of model sensitivity (e.g., the dosage is a very
sensitive measure for the variation in grid size) there is also consis-
tency among the results given by the measures.
The general methodology developed to evaluate the level of detail in
input data required for grid-based photochemical modeling needs is based
on the definition of a sensitivity/uncertainty index that provides a means
for ranking input data needs (see section VI.A). This index takes into
VII-2
-------
account the sensitivity of the model to the input data (e.g., deviation in
ozone levels above 12 pphm, deviation in maximum ozone level), the
corresponding cost for input data acquisition, and the uncertainty in the
model input variables affected by the input data considered. It is impor-
tant to note that this sensitivity/uncertainty index, which was used for
ranking data needs in our study of the Los Angeles basin, is generally
applicable; it can be used to define data needs for any urban area once
the cost and uncertainty estimates have been derived. Therefore, this
index constitutes a basis for a general approach to evaluate the sens't v-
ity of the urban airshed model to the amount (and thus the quantity) of
the input data. The development of a data base necessary for performing
urban area model simulations would advantageously involve the information
provided by the sensitivity-uncertainty analysis. Specific results
obtained for the Los Angeles basin are summarized next.
B. SPECIFIC RESULTS OF THIS STUDY
The results of the sensitivity analyses were used, along with cost
estimates for input data acquisition and uncertainty estimates, to define
a ranking of data needs (section VI.A). The sensitivity results depend
on the specific simulation conditions and on the choice of the cost and
uncertainty estimates (see tables VI-1 and VI-2); a choice of different
sensitivity measures or cost and uncertainty estimates could lead to a
different ranking of the input data needs, since the uncertainty bounds
shown in figures VI-1 through VI-4 display some overlap in the estimated
values of the sensitivity/uncertainty index. On the other hand, the
results of a sensitivity analysis are specific to the model considered,
and different factors (such as the magnitude of the perturbation, the
urban area considered, the submodels--e.g., wind model—and the length of
simulation time) have to be taken into account when generalizing the
sensitivity results. The sensitivity results presented below should not,
therefore, be interpreted as an absolute ranking of data needs; rather
they (1) exemplify the general methodology developed, and (2) provide
results for the specific case studied.
The main results of the sensitivity/uncertainty analysis of the
simulations of 26 June 1974 and 4 August 1975 for the Los Angeles basin
are listed in table VII-1. The relative importance of upper air and
surface meteorology data depends strongly upon meteorological condi-
tions. Upper air meteorological data are ranked high for the 26 June 1974
simulation, but are only of slight importance for the 4 August 1975
That is, on the sensitivity measure (e.g., deviation in ozone levels
above 12 pphm, deviation in maximum ozone level).
VII-3
-------
TABLE VII-1. RESULTS OF THE SENSITIVITY-UNCERTAINTY ANALYSIS FOR
THE LOS ANGELES BASIN (SENSITIVITY OF OZONE LEVELS
ABOVE 12 pphm)
(a) Simulation of 26 June 1974
Sensitivity/Uncertainty
Index Mean Value
Input Data (times 1Q6)
Mobile Source—Updating Inventory 23.0
Upper Air Meteorology 21.0
Initial Conditions for RHC 8.7
Area Source 4.5
Upper Air Quality 4.1
HC Speciation 2.9
Point Source 2.6
Surface Meteorology 1.3
Surface Air Quality 0.8
Mobile Source—Total Inventory 0.8
Area Sources—Temporal Resolution 0.6
(b) Simulation of 4 August 1975
Sens i ti vi ty/Uncertai nty
Index Mean Value
Input Data (times 1Q6)
Inital Conditions for RHC 11.7
Area Sources—Spatial Resolution 11.5
Surface Meteorology 11.0
HC Speciation 9.0
Upper Air Meteorology 4.1
Mobile Source—Total Invenotry 1.8
VII-4
-------
simulation. This can be related to the fact that the wind field is
strongly affected by upper air data for the 26 June 1974 case and not much
perturbed for the latter case. On the other hand, surface meteorological
data are more important for the 4 August 1975 simulation than for the 26
June 1974 simulation.
The ranking of the other data sets does not vary with the simulation
day. The input data sets decrease in importance in the following order
(meteorological data are not included; see table VII-1 for this simulation
day ranking):
> Mobile sources - updating inventory
> Initial conditions for hydrocarbons
> Area sources - spatial resolution
> Upper air quality
> Hydrocarbon speciation
> Point source emissions
> Surface air quality
> Mobile sources - total inventory
> Area source - temporal resolution
A complete description, interpretation, and evaluation of the
possible limitations of these sensitivity results is presented in chap-
ters III, V, and VI, respectively. The sensitivity results depend on the
magnitude of the perturbation, unless the model sensitivity is linear. As
pointed out in section VI-A, the type of sensitivity analysis considered
in this study is "perturbation specific"; the results depend on the
perturbation considered, and caution is advised if the results are to be
applied in a different context (i.e., with a different perturbation in the
input data). For instance, in sensitivity case 5 the number of air
quality monitoring stations was reduced from 23 to 10 and the sensitivity
results would probably be quite different had the number of stations been
reduced, say, from 10 to 5, since the model sensitivity is not expected to
be linear in this case.
Specific attributes of the modeling conditions also affect sensitiv-
ity analysis. In this study, these conditions are characterized by the
features of the Los Angeles basin, by the submodels considered (e.g., wind
VII-5
-------
model, emission inventories), and by the length of simulation time.
Further sensitivity studies could be carried out to investigate the effect
of these attributes on the model sensitivity; possible studies are
outlined in the next section.
C. FUTURE NEEDS
Because of the complex nature of photochemical grid models, a
detailed sensitivity analysis of such a model is an enormous task. Much
information has been obtained in this study about the sensitivity of the
airshed model predictions to the amount (and thus the quality) of input
data. Further work could be carried out to improve the knowledge base
regarding input data needs:
> Multiple-day simulation: As the simulation time
increases, the model approaches the more realistic, case
where emissions provide all the precursor mass and, in the
limit, the model predictions become insensitive to the
initial conditions. Therefore, in a multiple-day simula-
tion, the influence of the initial conditions on the model
predictions will decrease as the simulation time
increases, and the importance of the emission levels (as
well as meteorological description) will accordingly
increase. We could therefore expect the ranking of air
quality and emission data to be notably modified with a
multiple-day simulation. Useful information would be
obtained by carrying out multiple-day simulation stud es.
> Sensitivity studies for other urban airsheds: Specific
attributes of the Los Angeles basin (meteorology, aero-
metric network, emissions, topography) may limit the
generality of some of the results. For instance, the wind
model used in the Los Angeles basin was developed for
complex terrain applications and is sensitive to meteoro-
logical data at the boundaries of the airshed. In an
urban area with flat terrain, another type of wind model
that would show a different sensitivity to the input data
could be used (e.g., interpolated model). The choice of
the urban area could be made according to its specific
attributes and the intended applications of the models.
> Sensitivity studies for various perturbation levels: The
sensitivity results presented in this study are "perturba-
tion specific"; since the sensitivity of the model to
input data perturbations is usually nonlinear, the results
VII-6
-------
depend on the magnitude of the perturbation. It would be
of interest to investigate for some specific cases the
nonlinear nature of the model sensitivity. For instance,
the number of available air quality monitoring stations
was reduced in the sensitivity simulation focusing on air
quality data. Further sensitivity studies could provide
additional information on the minimum number of air qual-
ity stations needed to obtain sufficient air quality data
for the definition of adequate initial conditions and
boundary conditions.
Sensitivity studies involving multiple-day simulations and other
urban areas would provide useful additional information on the sensitivity
of photochemical models to the richness of the information data base.
VII-7
-------
REFERENCES
ABAG (1977), "Candidate Control Measures," Air Quality Maintenance Plan,
Technical Memo 5, Association of Bay Area Governments, Berkeley,
California.
Anderson, G. E., et al. (1977), "Air Quality in the Denver Metropolitan
Region: 1974-2000." EF77-222, Systems Applications, Incorporated,
San Rafael, California.
Attaway, L. D., et al. (1976), "Maintenance Shutdown of Tail Gas Treating
Unit: An Assessment of Potential S02 Concentrations and Related
Health and Welfare Effects," TR-11700, Greenfield, Attaway and Tyler,
Incorporated, San Rafael, California, and Systems Applications,
Incorporated, San Rafael, California.
Blumenthal, 0. L., W. H. White, and T. B. Smith (1978), "Anatomy of a Los
Angeles Smog Episode: Pollutant Transport in the Daytime Sea Breeze
Regime," Atmos. Environ., Vol. 12, pp. 893-907.
Burton, C. S., et al. (1976), "Oxidant/Ozone Ambient Measurement
Methods: An Assessment and Evaluation," EF76-111R, Systems Applica-
tions, Incorporated, San Rafael, California.
Demerjian, K. L. (1976), "Photochemical Air Quality Simulation Modeling:
Current Status and Future Prospects," Paper 16-1, International
Conference on Photochemical Oxidant Pollution and Its Control,
Environmental Protection Agency, Raleigh, North Carolina.
Dougherty, E. P., J. T. Hwang, and H. Rabitz (1979), "Further Developments
and Applications of the Green's Function Method of Sensitivity
Analysis in Chemical Kinetics," J. Chem. Phys., Vol. 71, pp. 1794-
1808.
Duewer, W. H., et al. (1977), "NOX Catalytic Ozone Destruction: Sensitiv-
ity to Rate Coefficients," J. Geophys. Res., Vol. 82, pp. 935-942.
R-l
-------
Dunker, A. M. (1980), "The Response of an Atmospheric Reaction Transport
Model to Changes in Input Functions," Atmos. Environ., Vol. 14, pp.
671-679.
Durbin, Paul, and T. A. Hecht (1975), "The Photochemistry of Smog Forma-
tion," internal paper, Systems Applications, Incorporated, San
Rafael, California.
EPA (1972), "Compilation of Air Pollutant Emission Factors," AP-42, U.S.
Environmental Protection Agency, Research Triangle Park, North
Carolina.
Edinger, J. G. (1973), "Vertical Distribution of Photochemical Smog in Los
Angeles Basin," Environ. Sci. Techno!., Vol. 7, No. 3, pp. 247-252.
GRC (1974), "Air Quality Impacts of Electric Cars in Los Angeles,"
Appendix A, RM-1905-A, General Research Corporation, Santa Barbara,
California.
Hayes, S. R. (1978), "Performance Measures for Air Quality Dispersion
Models," EF78-93, Systems Applications, Incorporated, San Rafael,
California.
Hecht, T. A., J. H. Seinfeld, and M. C. Dodge (1974), "Further Development
of a Generalized Kinetic Mechanism for Photochemical Smog," Environ.
Sci. Techno!., Vol. 8, p. 327.
Heinrich, R., S. M. Rapoport, and T. A. Rapoport (1977), "Metabolic
Regulation and Mathematical Models," Prog. Biophys. Mol. Bio!.,
Vol. 32, pp. 1-82.
Hickey, H. R., W. D. Rowe, and F. S. Skinner (1971), "A Cost Model for Air
Quality Monitoring Systems," J. Air Pollut. Control Assoc., Vol. 21,
No. 11, pp. 689-693.
Hoe!, P. G. (1962), "Introduction to Mathematical Statistics" (John Wiley
and Sons, New York, New York).
Husar, R. B., et al. (1977), "Three-Dimensional Distribution of Air
Pollutants in the Los Angeles Basin," J. Appl. Meteorol., Vol. 16,
pp. 1087-1096.
Kieth, R. W., and B. S. Selik (1977), "California South Coast Air Basin
Hourly Wind Flow Patterns," South Coast Air Quality Management
District, El Monte, California.
R-2
-------
Koda, M., A. H. Dogru, and J. H. Seinfeld (1979), "Sensitivity Analysis of
Partial Differential Equations with Applications to Reaction and
Diffusion Processes," J. Comp. Phys.. Vol. 30, pp. 259-282.
Lamb, R. G., and J. H. Seinfeld (1973), "Mathematical Modeling of Urban
Air Pollution—General Theory," Environ. Sci. Technol., Vol. 7, pp.
253-261.
Littman, F. E. (1978), "Regional Air Pollution Study, Emission Inventory
Summarization," Rockwell International, St. Louis, Missouri.
Liu, M. K., T. C. Myers, and J. L. McElroy (1979), "Numerical Modeling of
Land and Sea Breeze Circulation along a Complex Coastline," Math.
Comp. Simulation, Vol. 21, pp. 359-367.
Liu, M. K., D. C. Whitney, and P. M. Roth (1976), "Effects of Atmospheric
Parameters on the Concentration of Photochemical Air Pollutants,"
J. Appl. Meteorol., Vol. 15, ppp. 829-835.
Liu, M. K., et al. (1976), "Continued Research in Mesoscale Air Pollution
Simulation Modeling: Volume I—Analysis of Model Validity and
Sensitivity and Assessment of Prior Evaluation Studies," EPA-600/4-
76-016a, Systems Applications, Incorporated, San Rafael, California.
Lyons, W. A., and H. S. Cole (1976), "Photochemical Oxidant Transport
Mesoscale Lake Breeze and Synoptic Scale Aspects," J. Appl. Mete-
orol., Vol. 15, pp. 733-743.
MacCracken, M. C., and G. D. Sauter, eds. (1975), "Development of an Air
Pollution Model for the San Francisco Bay Area," University of
California, Livermore, California.
Mayersohn, H., et al. (1976), "Atmospheric Hydrocarbon Concentrations
June-September, 1975," California Air Resources Board, Atmospheric
Studies Section, Division of Technical Services, Sacramento, Cali-
fornia.
Mayersohn, H., et al. (1975), "Atmospheric Hydrocarbon Concentrations
June-September, 1974," California Air Resources Board, Atmospheric
Studies Section, Division of Technical Services, Sacramento, Cali-
fornia.
McRae, G. J., and J. W. Tilden (1980), "A Sensitivity and Uncertainty
Analysis of Urban Scale Air Pollution Models--Preliminary Steps,"
Proc. Second Joint Conference on Applications of Air Pollution
Meteorology, AMS/APCA, 24-27 March 1980, New Orleans, Louisiana.
R-3
-------
Miedema, A. K., et al. (1973), "Cost of Monitoring Air Quality in the
United States," EPA-450/3-74-029, Research Triangle Institute,
Research Triangle Park, North Carolina.
Reynolds, S. D., and P. M. Roth (1980), "The Systems Applications,
Incorporated Airshed Model—A Review and Assessment of Recent
Evaluation and Applications Experience," submitted to J. Air Pollut.
Control Assoc.
Reynolds, S. D., P. M. Roth, and J. H. Seinfeld (1973), Atmos Environ.,
Vol. 7, p. 1033.
Reynolds, S. D., et al. (1979), "Photochemical Modeling of Transportation
Control Strategies, Vol. 1, Model Development, Performance Evaluation
and Strategy Assessment," EF79-37, Systems Applications, Incorpo-
rated, San Rafael, California.
Reynolds, S. D., et al. (1976), "Continued Development and Validation of a
Second Generation Photochemical Air Quality Simulation Model: Volume
I I—Refinements in the Treatment of Chemistry, Meteorology, and
Numerical Integration Procedures," EF75-24R, Systems Applications,
Incorporated, San Rafael, California.
Russell, P. B., and E. E. Uthe (1978), "Regional Patterns of Mixing Depth
and Stability: Sodar Network Measurements for Input to Air Quality
Models." Bull. Am. Meteorol. Soc.. Vol. 59, pp. 1275-1287.
SCAPCD (1976), "Fuel Use and Emissions from Stationary Combustion
Sources," Southern California Air Pollution Control District,
El Monte, California.
Seigneur, C. (1978), "Mathematical Analysis of Air Pollution Models,"
Ph.D. Thesis, University of Minnesota, Minneapolis, Minnesota.
Skelton, E. P., et al. (1977), "Methodology for Estimating Emissions for
the Sacramento AQMP," Sacramento Air Pollution Control District,
Sacramento, California.
Souten, D. R., G. E. Anderson, and R. G. Ireson (1980), "National Commis-
sion on Air Quality Los Angeles Regional Study SAI Task 1 Report:
Review of the SIP Process and Framework (l.A.) Review and Analysis of
South Coast Air Quality Management District Air Quality Management
Plan Modeling Methods (l.B)," Draft Final Report, EF80-17, Systems
Applications, Incorporated, San Rafael, California.
R-4
-------
Tesche, T. W. (1978), "Evaluating Simple Oxidant Prediction Methods Using
Complex Photochemical Models," EM78-14, Systems Applications,
Incorporated, San Rafael, California.
Tesche, T. W., and M. A. Yocke (1978), "Numerical Modeling of Wind Fields
over Mountainous Regions in California," Conference on Sierra Nevada
Meteorology, American Meteorological Society, 19-21 June 1978, South
Lake Tahoe, California.
Unger, C. D. (1976), "Meteorological Input into Photochemical Models,"
unpublished results, California Air Resources Board, Sacramento,
California.
West, R. (1979), private communication from Richard West, CSBE, to William
Rogers Oliver, Systems Applications, Incorporated, San Rafael,
California.
Whitten, G. Z., and H. Hogo (1977), "Mathematical Modeling of Simulated
Photochemical Smog," EPA-600/3-77-011, Systems Applications, Incorpo-
rated, San Rafael, California.
Yotter, E. E. (1979), "Motor Vehicle Emissions Assessment Using Caltrans
Transportation Systems Analysis (Modeling) Data," California Air
Resources Board, Planning Division, Air Quality Maintenance Planning
Branch, Sacramento, California.
R-5
-------
TECHNICAL REPORT DATA
{'Please read Instructions on the revers^ be fare i omplctingj
1. REPORT NO
EPA-450/4-81-Q31a
4. TITLE AND SUBTITLE
The Sensitivity of Complex Photochemical Model Estimates
to Detail in Input Information
6. PERFORMING ORGANIZATION CODE
7. AUTHOR(S) . , .,
T. W. Tesche, C. Seigneur, L. E. Reid, P. M. Roth,
W. Ro Oliver, and J. C. Cassmassi
8. PERFORMING ORGANIZATION REPORT NO.
9. PERFORMING ORGANIZATION NAME AND ADDRESS
Systems Applications, Incorporated
950 Northgate Drive
San Rafael, California 94903
12. SPONSORING AGENCY NAME AND ADDRESS
U.S. Environmental Protection Agency
Office of Air Quality Planning and Standards
Research Triangle Park, North Carolina 27711
3. RECIPIENT'S ACCESSION NO.
REPORT DATE
SAI No. 330R-EF81-5
10 PROGRAM ELEMENT NO
11 CONTRACT/GRANT NO.
68-02-2870
13 TYPE OF REPORT AND PERIOD COVERED
14. SPONSORING AGENCY CODE
15. SUPPLEMENTARY NOTES
16. ABSTRACT
Using the air quality, meteorological and emissions data base available in the
Los Angeles area, two days with distinctly different meteorology are simulated using a
photochemical grid model (Urban Airshed Model). The data base used to generate model
inputs is then degraded for the purpose of noting which data are most essential to
collect in order to have a complex grid model perform adequately. The results are used
to develop a more general methodology for prioritizing data needs. The methodology
considers model sensitivity to input derived from data bases of varying detail, expense
in collecting the data, and the uncertainty associated with deriving model input vari-
ables from the data base.
KEY WORDS AND DOCUMENT ANALYSIS
DESCRIPTORS
Photochemical grid models
Urban Airshed Model
Ozone
Sensitivity studies
Model inputs
b.IDENTIFIERS/OPEN ENDED TERMS
c. COS AT I 1 icId/Group
(9 bf.rjB,!' CLASS , I'lill Kefi-r., I 71 NO OF PAGES
a DISTRIBUTION STATEMENT
Unlimited
-------
OJ ftl —;
83 ^
Q) O
5
.
yi O n en
"i * »
3r* -• O
10 ~ O
o O -,
i- ft) ;-
O D Q) ID
S °
m > TJ m T) -
" '"
3 Q.
ooa
3%
cn — -
oj o fl>
-------