Georgia Institute of Technology Hybrid Approach

advertisement
A Hybrid Method for Particulate Matter
Source Apportionment: Using A
Combined Chemical Transport and
Receptor Model Approach
Yongtao Hu, Sivaraman Balachandran, Jorge
Pachon,Jaemeen Baek*, Talat Odman, James A. Mulholland
and Armistead G. Russell
School of Civil and Environmental Engineering, Georgia Institute of Technology, Atlanta, Georgia
*Currently at Currently at IIHR-Hydroscience and Engineering, University of Iowa. Iowa City, Iowa
10th Annual CMAS Conference, October 25th, 2010
Georgia Institute of Technology
Objective and Approach
• Develop a source-based approach to integrating receptor- and
source- oriented modeling of particulate matter
–
–
–
–
Improve source impact estimates
Extend impact quantification to more sources
Expanded spatial and temporal coverage of source apportionment
Provide estimates of uncertainties for spatial analysis
• Approach
– CMAQ DDM3D/PM to provide initial source impacts and
sensitivities
– Use sensitivities to adjust source impacts using CMB-type
formulation
– Use adjustments and species performance to assess uncertainties
• Application
– One month simulation over CONUS
– STN monitors
– Six cities
Georgia Institute of Technology
Receptor Oriented Modeling (RM)
RM approaches such as CMB rely on using observed concentrations of the
PM composition at a receptor, along with knowledge of the composition of
source emissions (source profiles), to solve a species balance equation that
estimates the source impacts. For example CMB species balance equations:
total number of emission
sources considered
measured concentration of
species i
emission fraction of species i in
total PM2.5 emitted from source j
J RM
obs
i
c
  f i , j SR
j 1
RM
j
e
Limitations/assumptions/uncertainties
RM
i
RM’s prediction error to be
minimized
source j’s impact on the total PM2.5
concentration
•Emission compositions are constant and known (not good for some sources)
•No reactions or differential phase changes (not bad for many, but not all, primary
compounds)
•Most sources are included (typically only about 80% of mass is)
•Source compositions are linearly independent of each other (co-linearity can be
a problem)
•The number of sources is less than or equal to chemical species (limitation)
Georgia Institute of Technology
Source-Oriented Modeling (SM)
SM approaches using chemical transport models (CTMs) follow a first
principles approach, tracking the emissions, transport, transformation and loss
of chemical species in the atmosphere to simulate ambient concentrations and
source impacts. For example using DDM3D derived sensitivities:
total number of emission sources
that included in CTM
calculated sensitivity coefficients of species i’s
Simulated concentration for
concentration to emissions from source j
species i
impact from source j’s
J SM
J SM
emissions outside of the domain
CTM
T
ci
  SAi , j   ( E j Si , j  ICi , j  BCi , j )
j 1
j 1
estimate of source j’s impact on
total emissions of all tracked pollutants
species i’s concentration
emitted from source j
impact from source j’s emissions
prior to the simulation period
Limitations/uncertainties
Emissions estimates, Meteorological inputs, Missing processes and
parameter uncertainties
Benefits
Large number of sources, direct link to sources, spatial coverage,
non-linear chemistry
Georgia Institute of Technology
A hybrid approach for particulate matter source
apportionment: Combining receptor modeling with chemical
transport modeling
Limited number of sources vs.
completeness of source categories
ciobs 
J RM
f
j 1
SM’s prediction error to
be minimized
J SM
J SM
j 1
j 1
RM
RM
SM
T
SM
SR

e

SA

e

(
E
S

IC

BC
)

e


i, j
j
i
i, j
i
j i, j
i, j
i, j
i
Sensitivities Sensitivity to Sensitivity to
to emissions IC
BC
Constraints from source profiles upgraded to
constraints of source-receptor relationship
derived from CTM
We modify the species balance equations which CMB is based to use
outputs of the CTM.
Georgia Institute of Technology
Hybrid Approach (continued)
The hybrid approach relies on minimizing the differences (2) between
CTM-calculated and observed PM2.5 concentrations (including each PM2.5
component and metals) while considering estimated uncertainties in both the
observations and source emission rates:
So,
 
Min  2
Rj
where
CTM-simulated base case impact
of source j on species i
to weigh the amount of
change in source strengths
total number of sources
2
J

 
obs
base
total number of species
  ci   SAi , j R j  
2
N
J
(
R

1
)

j 1
j
 
 2   

2
2
2





i 1
j 1
Rj
Ciobs
CiCTM






ratio of adjusted impact from
a priori uncertainties
source j to the base case
Instead of the original CMB solution:
ϰ2 =Σi [(Ci-ΣjFijSRj)2/(ϭCi2+ΣjϭFij2SRj2)]
Effective Variance, Watson et al., (1984), single sample
Georgia Institute of Technology
Application
2004 MM5-SMOKE-CMAQ-DDM3D simulation
• 36-km grid covering continental United States as well as portions
of Canada and Mexico.
• Projected VISTAS emissions inventory used as a priori inventory.
First order DDM sensitivity coefficients calculated for 32 separate
source categories.
Table 1 Emissions source categories
32
categories
Non-mobile combustion
COALCMB
DIESELCMB
FUELOILCMB
LPGCMB
NAGASCMB
OTHERCMB
MEXICO_CMB
WOODFUEL
WOODSTOVE
On-road
ORDIESEL
ORGASOL
Non-road
AIRCRAFT
NRDIESEL
NRFUELOIL
NRGASOL
NRLPG
NRNAGAS
NROTHERS
RAILROAD
Biomass-burning
AGRIBURN
WILDFIRE
OPENFIRE
PRESCRBURN
Others
BIOGENIC
DUST
LIVESTOCK
LWASTEBURNING
MEATCOOKING
MEATALPRDUCT
MINERALPRODUCT
SOLVENT
OTHERS
 Ambient PM2.5 concentrations apportioned to the 32 separate sources
 STN, IMPROVE, SEARCH and ASACA networks
• TOT measurements of OC and EC from STN and ASACA
converted to TOR equivalences.
Georgia Institute of Technology
PM2.5 monitoring networks
Detroit
Chicago
New York
Pittsburgh
Los Angeles
Atlanta
STN
IMPROVE
SEARCH
The Modeling Domain
Georgia Institute of Technology
Hybrid Approach Applied at STN sites
Major PM2.5 ions and metals measured:
Table 2 Measured species at STN monitoring sites
42
species
Total
mass
PM2.5
Major Components
Metals (36)
EC, OC, Sulfate, Nitrate,
Ammonium
Na, Mg, Al , Si, P, S, Cl, K, Ca, Ti, V, Cr, Mn, Fe, Co, Ni, Cu, Zn, Ga, As, Se,
Br, Rb, Sr, Zr, Mo, Ag, Cd, In, Sn, Sb, Ba, La, Ce, Hg, Pb
Use reported detection limits and measurement uncertainties
Obtain metals’ sensitivities to sources:
•Split using source specific PM2.5 (unidentified portion) sensitivity
coefficients and source profiles of metals for each of the 32
categories assuming that metals remain intact from source to
receptor.
•Source profiles are assembled from the 84 profiles compiled by
Reff et al. 2009 ES&T. The profiles split PM2.5 emissions to the
above 42 species.
Georgia Institute of Technology
X2 Ci
X2 Rj
Choice of Г for Ridge Regression
2.5
Χ2 Ci
X2 Rj
0.07
0.06
2
0.05
1.5
0.04
0.03
1
0.02
0.5
0.01
0
1E-14
1E-12
1E-10
0.00000001
0.000001
0.0001
0.01
0
100
1
Г
2
J

 
obs
base
  ci   SAi , j R j  
N

j 1
 

  2  2

i 1
obs
CTM
c
c


i
i




2
 Ci 
N
Г=N/J=42/32=1.3125 selected
J

 
2
Rj
j 1
( R j  1)
 R2
J
j
2
CMAQ/Hybrid Concentrations
Los Angeles
Atlanta
observed
25
simulated initial
simulated refined
observed
simulated initial
simulated refined
45
40
20
35
ug/m3
ug/m3
30
15
10
25
20
15
5
10
5
0
PM25
OC25
NO325
observed
NH425
simulated initial
SO425
0
Metals
PM25
simulated refined
30
30
25
25
20
20
15
OC25
EC25
New York
ug/m3
ug/m3
Chicago
EC25
NO325
observed
NH425
SO425
simulated initial
Metals
simulated refined
15
10
10
5
5
0
0
PM25
OC25
Detroit
EC25
NO325
observed
NH425
simulated initial
SO425
PM25
Metals
OC25
EC25
Pittsbrugh
simulated refined
25
NO325
observed
NH425
simulated initial
SO425
Metals
simulated refined
16
14
20
10
15
ug/m3
ug/m3
12
10
8
6
4
5
2
0
PM25
OC25
EC25
NO325
NH425
SO425
Georgia
Institute of0 Technology
Metals
PM25
OC25
EC25
NO325
NH425
SO425
Metals
Initial/Refined (CMAQ/Hybrid) difference (χ2Ci)
between simulated and observed PM2.5
1000
concentrations
100
X 2 Refined
y = 0.2387x
R2 = 0.9325
10
1
0.1
0.01
0.01
0.1
1
10
100
X 2 Initial
Georgia Institute of Technology
1000
Initial and Refined PM2.5 source impacts (in percentage)
Six Cities - 2004 January Average
Woodstove
WOODSTOVE
WOODFUEL
100%
WILDFIRE
Solvent
SOLVENT
90%
RAILROAD
Others
80%
PRESCRBURN
OTHERS
Prescribed burn
OTHERCMB
Other combustion
70%
ORGASOLINE
ORDIESEL
OPENFIRE
60%
Nonroad diesel
NROTHERS
On-road gasoline
NRNAGAS
Natural gas combustion
50%
NRLPG
Mineral product
NRGASOLINE
NRFUELOIL
40%
On-road diesel
Meat cooking
30%
Fuel oil
combustion
NRDIESEL
NAGASCMB
MINERALPRODUCT
MEXCMB_M
Dust
20%
Waste burn
LWASTEBURN
Metal product10%
Coal combustion
LPGCMB
Livestock
0%
LIVESTOCK
At
lan
ta-
Re
fin
Ch
ed
ica
rgo
-In
Ch
itia
ica
l
rgo
-R
efi
ne
d
De
tro
it-I
nit
De
ial
tro
it-R
efi
Lo
ne
sA
d
ng
ele
Lo
s-I
sA
nit
ial
ng
ele
s-R
efi
ne
Ne
d
w
Yo
rkIn i
Ne
tia
w
l
Yo
rkRe
fi n
Pit
ed
tsb
urg
h-I
Pit
nit
tsb
ial
urg
h-R
efi
ne
d
Ini
tia
l
FUELOILCMB
At
lan
ta-
Aircraft
LPG
combustion
MEATALPRODUCT
MEATCOOKING
Georgia Institute of Technology
DUST
DIESELCMB
COALCMB
BIOGENIC
AIRCRAFT
AGRIBURN
Biogenic
Major contributing sources in six cities
City
Atlanta
1st
woodstove
2nd
dust
Chicago
metal products
Detroit
woodstove
Los Angeles
woodstove
natural gas
combustion
natural gas
combustion
meat cooking
New York
woodstove
Pittsburgh
livestock
coal
combustion
dust
3rd
coal
combustion
woodstove
4th
on-road
gasoline
dust
5th
aircraft
dust
on-road
gasoline
natural gas
combustion
dust
livestock
woodstove
on-road
gasoline
dust
fuel oil
combustion
coal
combustion
Georgia Institute of Technology
livestock
on-road
gasoline
meat cooking
Initial/Refined (CMAQ/Hybrid) Source Impacts
Georgia Institute of Technology
Initial/Refined (CMAQ/Hybrid) Source Impacts
Georgia Institute of Technology
Compare with the CMB Results
CMB apportionment allowed resolution of less than 10 sources while
the hybrid method resolved 32, and included total contributions from
both primary and secondary paths.
In order to do more specific comparisons, the hybrid results are regrouped to match up with the CMB categories by
(1) splitting the primary and the secondary contributions from
each hybrid category, using the source specific composition profiles
and assuming that the primary species are inert and stick together,
and
(2) merging the hybrid sub-categories that split to primary and
secondary portions to the major categories that match up with the
CMB sources.
Georgia Institute of Technology
Initial/Refined (CMAQ/Hybrid) Source Impacts
14
Atlanta 01/07/2004
Initial
12
Refined
CMB
10
8
6
4
2
0
Obs
Sim
LDGV
HDDV
SDUST
BURN
CFPP
AMSULF AMNITR OTHROC AllOthers
Georgia Institute of Technology
Initial/Refined (CMAQ/Hybrid) Source Impacts
35
Atlanta 01/16/2004
Initial
30
Refined
CMB
25
20
15
10
5
0
Obs
Sim
LDGV
HDDV
SDUST
BURN
CFPP
AMSULF AMNITR OTHROC AllOthers
Georgia Institute of Technology
Initial/Refined (CMAQ/Hybrid) Source Impacts
14
Atlanta 01/19/2004
Initial
Refined
12
CMB
10
8
6
4
2
0
Obs
Sim
LDGV
HDDV
SDUST
BURN
CFPP
AMSULF AMNITR OTHROC AllOthers
Georgia Institute of Technology
Benefits and Future Work
• Hybrid Approach Benefits
– Completeness of sources
• More complete range of sources quantified
– First principles’ constraints
• Can account for non-linearities and secondary PM sources
– Limitations removed, for spatial and temporal applications.
– Uncertainty estimation
• Ongoing Work
–
–
–
–
–
–
–
Source apportionment at IMPROVE, ASACA and SEARCH sites.
Simulating full year.
Further uncertainty estimation.
Additional approach for inverse modeling
Optimize source compositions.
Interpolation of source impacts spatially and temporally
Increased resolution
Georgia Institute of Technology
Acknowledgements
• EPA funding under grants R83362601 and R83386601
• Southern Company and Georgia Power
Georgia Institute of Technology
Download