Inverse Modeling and Attainment Analysis for Improved Decision Support of PM2.5 Air
Quality Regulations Daven K. Henze, Matt Turner, University of Colorado Boulder Patrick L. Kinney, Darby Jack, Ying Li, Columbia University Loretta Mickley, Daniel Jacob, Harvard University Drew Shindell, NASA GISS Randall Martin, Dalhousie University Amir Hakami, Carleton University
Partners at US EPA: AMAD - Sergey Napelenok - Robert Pinder - Alice Gilliland - S.T. Rao Emissions Group - Doug Solomon - Sally Dombrowski OAQPS - Bryan Hubbell
Project overview
Earth System Models drivers: - GISS Model E - GEOS-Chem - WRF air quality predictions: - CMAQ - GEOS-Chem
Data Assimilation - CMAQ adjoint - GEOS-Chem adjoint
Earth Observations remote sensing: - SCIAMACHY / OMI
NO2
- TES NH3
in situ: - IMPROVE, STN, CASTNet, AMoN, CAMNet, NADP, CalNex
Predictions / Products - Emissions - Control Strategies
Development of effective PM2.5 control strategies hindered by: - uncertainty in knowledge of sources - lack of information relating emissions from specific sectors, locations and species, to health endpoints. Here we are using the tools outlined on the left to: - constrain estimates of PM2.5 sources - develop improved emissions mitigation strategies with detailed source – receptor analysis
- provide new tools to the EPA for long term adoption of these techniques.
Implement CMAQ aerosol data assimilation capabilities
Schedule and Milestones (MS)
Develop constraints on PM2.5 precursor emissions
Source attribution of PM2.5 health impacts
MS: Provide constraints on NH3 seasonality and NOx mobile sources to NEI
MS: Source attribution to NAAQS RIA
Year 2 Year 1 Year 3 Year 4
Climate change impacts on AQ
MS: Climate impacts on control strategies.
MS: Public release of CMAQ adjoint
Compile emissions, meteorology, process satellite data
Started fall 2009
Milestone #2: Constraining emissions of NH3
X - 32 ZHU ET AL.: INVERSE MODELING RESULTS OF NH3 EMISSIONS
BALES ET AL.: SHORT TITLE X - 3
(a)
(b)
Figure 1. Global figure caption (a) describes the first subfigure; (b) describes the second subfigure;
m = 1.22 r2 = 0.94
2 4 6 8 10 12 14 16
x 106
2
4
6
8
10
12
14
16
x 106
True
Mo
de
led
r2 =1
m =1
r2 =0.935
m =1.22
(a)
(b) (c)
Figure 2. Global figure caption (a) describes the first subfigure; (b) describes the second subfigure;
3. Actual application
For an application with real data, we will use TESobservations throughout 2009 and compare these tomodel estimates from the GEOS-Chem chemical trans-port model in a global 2� ⇥ 2.5� simulation.
Acknowledgments. (Text here)
ReferencesDubovik, O., T. Lapyonok, Y. J. Kaufman, M. Chin, P. Ginoux,
R. A. Kahn, and A. Sinyuk (2008), Retrieving global aerosol
sources from satellites using inverse modeling, Atmos. Chem.Phys., 8 (2), 209–250.
Elbern, H., H. Schmidt, O. Talagrand, and A. Ebel (2000), 4D-
varational data assimilation with an adjoint air quality model
for emission analysis, Environ. Modell. Softw., 15, 539–548.
(a)
BALES ET AL.: SHORT TITLE X - 3
(a)
(b)
Figure 1. Global figure caption (a) describes the first subfigure; (b) describes the second subfigure;
(a)
2 4 6 8 10 12 14 16
x 106
2
4
6
8
10
12
14
16
x 106
True
Mo
de
led
r2 =1
m =1
r2 =0.865
m =1.21
m = 1.21 r2 = 0.87
(b)
(c)
Figure 2. Global figure caption (a) describes the first subfigure; (b) describes the second subfigure; (c) describes thesecond subfigure;
3. Actual application
For an application with real data, we will use TESobservations throughout 2009 and compare these tomodel estimates from the GEOS-Chem chemical trans-port model in a global 2� ⇥ 2.5� simulation.
Acknowledgments. (Text here)
ReferencesDubovik, O., T. Lapyonok, Y. J. Kaufman, M. Chin, P. Ginoux,
R. A. Kahn, and A. Sinyuk (2008), Retrieving global aerosol
sources from satellites using inverse modeling, Atmos. Chem.
Phys., 8 (2), 209–250.
(b)
BALES ET AL.: SHORT TITLE X - 3
(a)
(b)
Figure 1. Global figure caption (a) describes the first subfigure; (b) describes the second subfigure;
(a)
(b)
2 4 6 8 10 12 14 16
x 106
2
4
6
8
10
12
14
16
x 106
True
Mo
de
led
r2 =1
m =1
r2 =0.992
m =1.03
m = 1.03 r2 = 0.99
(c)
Figure 2. Global figure caption (a) describes the first subfigure; (b) describes the second subfigure;
3. Actual application
For an application with real data, we will use TESobservations throughout 2009 and compare these tomodel estimates from the GEOS-Chem chemical trans-port model in a global 2� ⇥ 2.5� simulation.
Acknowledgments. (Text here)
ReferencesDubovik, O., T. Lapyonok, Y. J. Kaufman, M. Chin, P. Ginoux,
R. A. Kahn, and A. Sinyuk (2008), Retrieving global aerosol
sources from satellites using inverse modeling, Atmos. Chem.Phys., 8 (2), 209–250.
Elbern, H., H. Schmidt, O. Talagrand, and A. Ebel (2000), 4D-
varational data assimilation with an adjoint air quality model
for emission analysis, Environ. Modell. Softw., 15, 539–548.
(c)
Figure 4. Tests for the possible impacts of inversion error, retrieval bias and measure-
ment error: (a) retrieval algorithm with a polluted profile as an initial guess; (b) modified
retrieval algorithm with a moderate profile as the initial guess; (c) model profiles from
the true model were ascribed error of the same size as the measurement error.
April
July
October
Initial Optimized ln(Optimized/Initial)
0 2.33 4.67 7.00 [106 kg] -2.00 -0.67 0.67 2.00 [unitless]
Figure 5. NH3
emissions from GEOS-Chem before and after the assimilation
D R A F T July 23, 2012, 11:33am D R A F T
Large uncertainties - seasonality - fertilizer vs animals - primary sources vs redistribution
Impacts of NH3 - governs 10-30% of PM2.5 mass - aerosol water and phase --> clouds & climate - deposition of reactive nitrogen
EPA’s national emission inventory (NEI) uses inverse modeling to constrain seasonality (Gilliland et al., 2006) Room to improve upon this using new data (TES, AMoN) and new inverse models (Shephard et al., 2011; Pinder et al., 2011; Zhu et al., submitted, Paulot et al., in prep)
Constraints from TES
Range of total US NH3 emissions Park (2004) Paulot Paulot
Zhu et al., submitted
Paulot et al., in prep
+80%
+57%
+33%
Milestone #2: Constraining emissions of NH3 ZHU ET AL.: INVERSE MODELING RESULTS OF NH3 EMISSIONS X - 35
April
July
October
GC
(ppb)
GC
(ppb)
AMoN (ppb) 0 1 2 3 4 5 6
01
23
45
6
Observation NH3, (ppb)
GE
OS!
Ch
em
, N
H3
, (p
pb
)
October
y =
0.494
+
0.564
x
0 1 2 3 4 5 6
01
23
45
6
Observation NH3, (ppb)
GE
OS!
Ch
em
, O
ptim
ize
d,
NH
3,
(pp
b)
October
y =
0.455
+
0.999
x
0 1 2 3 4 5 6
01
23
45
6
Observation NH3, (ppb)
GE
OS!
Ch
em
, N
H3
, (p
pb
)
October
y =
0.494
+
0.564
x
0 1 2 3 4 5 6
01
23
45
6
Observation NH3, (ppb)
GE
OS!
Ch
em
, O
ptim
ize
d,
NH
3,
(pp
b)
October
y =
0.455
+
0.999
x
R2=0.545 RMSE=0.952
NMB=-0.138
R2=0.693 RMSE=0.862
NMB=0.166
0 2 4 6 8 10 12
02
46
810
12
Observation NH3, (ppb)
GE
OS!
Chem
, N
H3, (p
pb)
July
y =
1.5
+
0.51
x
0 2 4 6 8 10 12
02
46
810
12
Observation NH3, (ppb)
GE
OS!
Chem
, O
ptim
ized, N
H3, (p
pb)
July
y =
2.22
+
1
x
0 2 4 6 8 10 12
02
46
810
12
Observation NH3, (ppb)
GE
OS!
Chem
, N
H3, (p
pb)
July
y =
1.5
+
0.51
x
0 2 4 6 8 10 12
02
46
810
12
Observation NH3, (ppb)
GE
OS!
Chem
, O
ptim
ized, N
H3, (p
pb)
July
y =
2.22
+
1
x
R2=0.281 RMSE=1.990
NMB=-0.045
R2=0.365 RMSE=3.534
NMB=0.659
0 2 4 6 8 10
02
46
81
0
Observation NH3, (ppb)
GE
OS!
Ch
em
, N
H3
, (p
pb
)
April
y =
0.215
+
0.225
x
0 2 4 6 8 10
02
46
81
0
Observation NH3, (ppb)
GE
OS!
Ch
em
, O
ptim
ize
d,
NH
3,
(pp
b)
April
y =
!0.189
+
1.02
x
0 2 4 6 8 10
02
46
810
Observation NH3, (ppb)
GE
OS!
Chem
, N
H3, (p
pb)
April
y =
0.215
+
0.225
x
0 2 4 6 8 10
02
46
810
Observation NH3, (ppb)
GE
OS!
Chem
, O
ptim
ized, N
H3, (p
pb)
April
y =
!0.189
+
1.02
x
R2=0.497 RMSE=1.834
NMB=-0.069
R2=0.406 RMSE=2.107
NMB=-0.678
Optimized Initial
Figure 8. Comparison of GEOS-Chem NH3
concentrations with observations from
AMoN sites before and after the assimilation. The square of the correlation coe�cient
(R2), root mean square error (RMSE), and normalized mean bias (NMB) are shown.
Black solid lines are regressions. Grey dashed lines are 1:1.
D R A F T July 23, 2012, 11:33am D R A F T
Constraints provided by TES lead to improved estimates of NH3 measured at AMoN sites in April and October (right)
Constraints provided by TES shift the seasonality towards higher emissions in April, consistent with updates to CMAQ NH3 bidirectional flux model (left)
CMAQ
CMAQ bidi
GC initial
GC opt
Jeong et al., 2011 AGU
Zhu et al., submitted
Milestone #1: Online observation operator for assimilating NO2 into CMAQ
1. compare vertical columns
2. compare slant columns
1. Leads to more rigorous slant vertical column comparison 2. Emphasizes altitudes where OMI is most sensitive
OMI DOMINO CMAQ
Pye et al., 2010
+
GEOS-Chem / CMAQ annual BC baseline mortalities (BenMAP*)
[ug/m3]
[mortalities over 30 yr old]
Milestone #3: Source attribution of BC aerosol related mortality Question: How much do emissions from each location / species / sector contribute to mortality? Approach:
+ response factor (0.005827) = 14,000 deaths Apportion sources contributing to these mortalities using adjoints, and use tool for EPA’s Regulatory Impact Analysis (RIA)
*EPA’s health impact model
Adjoint modeling for source-receptor analysis:
Sensitivity of all model concentrations to one model source
t0 Perturbation at source region
Forward
tn
Changes of concentration
Forward Model (source-oriented)
Adjoint modeling for source-receptor analysis:
Sensitivity of all model concentrations to one model source
t0 Perturbation at source region
Forward
tn
Changes of concentration
Forward Model (source-oriented) Sensitivity of model concentration in specific location to many model sources
Adjoint Model (receptor-oriented)
t0 area of possible origin
adjoint
tn
Concentration at the receptor
Adjoint modeling for source-receptor analysis:
t0 Perturbation at source region
Forward
tn
Changes of concentration
Forward Model (source-oriented) Adjoint Model (receptor-oriented)
t0 area of possible origin
adjoint
tn
Concentration at the receptor
example responses: national exceedence, statewide mortality,...
Sensitivity of all model concentrations to one model source
Sensitivity of model response over any region to many model sources
All sectors Fossil Fuel (12,600)
Biomass Burning (200) Biofuel (900)
Annual mortalities from BC emissions in each grid cell
Milestone #3: Source attribution of BC aerosol related mortality
Receptor modeling allows us to identify states where emissions reductions would have the greatest benefit
% contrib. to US mort. - % US emissions Disparities between state contributions to national emissions vs contributions to national health impacts
Milestone #3: Source attribution of BC aerosol related mortality
Emissions 3
2
1
0
[Gg BC/month]
Contributions to mortality 1200
800
400
0
[mortality]
preliminary results (April)
smaller ß benefit à larger per BC reduction
-4 4 2 -2 0
[mortality]
Milestone #3: Source attribution of BC aerosol related mortality
Contributions of emissions to mortality in Pennsylvania:
- A significant fraction of mortality in PA is owing to emissions outside PA - Important in light of recent court ruling against interstate controls
preliminary results (April)
note: preliminary analysis, complete annual average results in progress
Milestone #3: source attribution of PM2.5 related mortality
From fossil fuel SO2 (26,300) From fossil fuel NOx (20,300)
Source contributions to national mortality from PM2.5 - total estimated to be 103,000 / yr compare to 130,000 / yr from EPA study (Fan et al., 2012)) - contributions by location / sector / species:
Analysis valuable for determining health impacts of future emissions control strategies, particularly jointly addressing PM2.5 and O3
Summary of Progress
MS #1: New inverse modeling and source apportionment tools: - aerosol microphysical adjoint (Turner et al., 2011a, 2011b, posters) - ANISORROPIA thermodynamic adjoint (Capps et al., ACP, 2012) - online OMI NO2 observation operator (Pye et al., 2010, poster) - GEOS-Chem nested adjoint model MS #2: Remote sensing constraints on emissions: - NEI underestimates NH3 sources (Zhu et al., submitted to JGR) - TES imparts a different seasonality than wet deposition for NH3 sources - Transport & chemistry important for constraining even very short-lived species (Turner et al., GRL, 2012) MS #3: Source attribution of health impacts: - with GEOS-Chem and CMAQ for BC aerosol - for inorganic PM2.5 with GEOS-Chem MS #4: Accounting for interactions between AQ regulations and climate - source attribution of aerosol indirect effects from specific precursors (Karydis, submitted to GRL) - drive GEOS-Chem simulations with higher resolution AR5 climatology
$ Budget $ 1,200,000 over four years
Budget update from year 3 report (08/01/2009 – 07/31/2012):
• Includes funding from year 4, expenses up through year 3 • Uncosted funds largely owing to - slow processing of subcontract to Harvard, Columbia - challenges hiring / maintaining postdocs jointly at CU Boulder / EPA
0" 50" 100" 150" 200" 250" 300" 350" 400"
Facili-es"and"Admin"
Wages,"Benefits,"Tui-on"
Opera-ng"Expenses"
Travel"
Fixed"Assests"
Columbia"Subcontract"
Harvard"Subcontract"
$"(thousand)"
Spent"(Year"3)"Encumbered"Budget"(4"Years)"
Risks and Challenges
Management: - challenge to hire / maintain postdocs at EPA RTP through CU - initial plan was joint CU Boulder / EPA postdoc - significant bureaucratic barriers for internationals at EPA - US postdocs soon hired away to permanent positions - Henze’s group just getting started in 2009 - building group from scratch took time - building computer lab postponed 1 year owing to campus delays Technical: - CMAQ adjoint model still being finalized - TES NH3 data challenging to utilize - TES / OMI lifetime (consider CrIS, TROPOMI, GEO-CAPE for future) - Driving GEOS-Chem with high resolution GISS climatology difficult Scheduling: - NAAQS review process at EPA a constantly moving target - challenge to make our applications relevant to their needs
Applications Readiness Level Present
ARL 4: Components technically integrated - CMAQ aerosol aerosol verified on a component-by-component basis - GEOS-Chem aerosol adjoint on North American nested grid - EPA data sets (BenMAPs) used to calculate mortality owing to PM2.5 exposure integrated into both, agree with BenMAPs to within >99%. ARL 4: Organizational challenges identified and managed: - Farhan Akhtar, Shannon Capps trained with these tools, currently or soon to be at EPA offices to facilitate long-term adoption
Goal: ARL 6 by end of 2012, ARL 8 by end of 2013 ARL 5: Prototype system: to be released this fall ARL 5: Potential determined and articulated: - presentation at CMAS 2012, build on 2011 workshop & town hall event ARL 6: Prototype system beta-tested in a simulated operational environment: - already ongoing
References
Turner, M., D. K. Henze, A. Hakami, S. Zhao, J. Resler, G. Carmichael, C. Stanier, J. Baek, P. Saide, A. Sandu, A. Russell, G. Jeong, A. Nenes, S. Capps, P. Percell, R. W. Pinder, S. Napelenok, H. Pye, J. O. Bash, T. Chai, D. Byun, Aerosols Processes in the CMAQ Adjoint, International Aerosol Modeling Algorithms Conference, Davis, CA, Nov 30 - Dec 2, 2011a. Turner, M., D. K. Henze, A. Hakami, S. Zhao, J. Resler, G. Carmichael, C. Stanier, J. Baek, P. Saide, A. Sandu, A. Russell, G. Jeong, A. Nenes, S. Capps, P. Percell, R. W. Pinder, S. Napelenok, H. Pye, J. O. Bash, T. Chai, D. Byun, High Resolution Source Attribution of PM Health Impacts with the CMAQ Adjoint Model, 10th Annual CMAS Conference, Chapel Hill, NC, Oct 24-26, 2011b. Capps, S. L., D. K. Henze, A. G. Russell, and A. Nenes, Quantifying relative contributions of global emissions to PM2.5 air quality attainment in the U.S, A54C-07, or presentation, AGU Fall Meeting, San Francisco, CA, 5-9 Dec, 2011. Best Student Poster Award. Pye, H. O. T., S. L. Napelenok, R. W. Pinder, D. K. Henze, R. V. Martin, and K. W. Appel, Evaluation of CMAQ NO2 predictions over the U.S. using ground-based and satellite observations, 5th International GEOS-Chem Conference, Cambridge, MA, May 2-5, 2011. Capps, S. L., D. K. Henze, A. Hakami, A. G. Russell, and A. Nenes, ANISORROPIA: the adjoint of the aerosol thermodynamic model ISORROPIA, Atmos. Chem. Phys., 12, 527-543, 2012. Turner, A., D. K. Henze, R. V. Martin, and A. Hakami, Modeled source influences on column concentrations of short-lived species, Geophys. Res. Lett., 39, L12806, doi:10.1029/2012GL051832, 2012. Karydis, V. A., S. L. Capps, R. H. Moore, A. Russell, D. K. Henze, and A. Nenes, Using a global aerosol model adjoint to unravel the footprint of spatially-distributed emissions on cloud droplet number and cloud albedo, submitted to GRL, 2012. L. Zhu, D. K. Henze, K. E. Cady-Pereira, M. W. Shephard, M. Luo, R. W. Pinder, J. O. Bash, G. Jeong, Constraining U.S. ammonia emissions using TES remote sensing observations and the GEOS-Chem adjoint model, submitted to JGR, 2012.
bonus slides!
1 2 3 4 5 6 7 8 9 10 11 120
200
400
600
800
1000
1200
1400
1600M
orta
litie
s
month
fossil fuelbio fuelbiomass burn
Monthly mortalities owing to BC emissions from entire domain
Technical results: source attribution of carbonaceous aerosol related mortality
Interpretation of adjoint model results Example: - Value of 35 in grid box over your county means? --> emissions in your county contributed to 35 premature mortalities - where were those 35 mortalities? --> Adjoint only tells us they were somewhere in the
entire US.
Milestone #3: Source attribution of BC aerosol related mortality
mortality per ug/m3
Milestone #3: Source attribution of BC aerosol related mortality
Top Related