SUPPORTING INTEX THROUGH INTEGRATED ANALYSIS OF SATELLITE AND SUB-ORBITAL MEASUREMENTS WITH GLOBAL AND REGIONAL 3-D MODELS: Bottom-up Emissions Inventory Development and Inverse Modeling

1
SUPPORTING INTEX THROUGH INTEGRATED ANALYSIS OF
SATELLITE AND SUB-ORBITAL MEASUREMENTS WITH
GLOBAL AND REGIONAL 3-D MODELS Bottom-up
Emissions Inventory Development and Inverse
Modeling

• Armistead (Ted) Russell
• Air Resources Engineering Center
• Environmental Engineering
• Georgia Institute of Technology

2
Objectives

• Use satellite, in-situ and aircraft observations
  to
  evaluate
  chemical transport model (CTM) results to
  identify likely emissions biases using inverse
  modeling
• Oxidized nitrogen species (NO2, NO, HNO3, PAN,
  PM-nitrate)
• HCHO
• CO
• SO2-sulfate
• PM
• Evaluate satellite observations
• Consistency with well-characterized emissions and
  analyzed air quality fields
• Examine spatial variability and
  ground/aircraft-based monitor spatial
  representativeness

3
Approach

• Develop accurate emissions inventories for
  2003-2004
• Model processes (mobile, area, biogenic) (10-50
  unc.)
• CEM for major point sources (lt15 unc.)
• Simulate August 2003 air quality
• Use inverse modeling to identify likely inventory
  biases/timing issues
• Identify conditions where model works
  better/worse
• Simulate INTEX study periods
• Evaluate model
• Compare results to satellite observations
• Assess mass consistency between observations and
  model simulations
• Use model results to address objectives

4
Emissions Nationwide
NOX
NOX
Anthropogenic VOC
PM10
SO2
EPA National Air Quality and Emissions Trends
Report, 2003
5
Emissions Inventory Northeast
NOX
PM2.5
SO2
VOC
2003 Emission Inventory, Fall Line Air Quality
Study (FAQS)
6
Emissions Inventory Northeast States
Percent Contribution  State Area Biogenics EGU Mobile Non-EGU Nonroad
NOX Maine 4 1 10 40 21 24
NOX New Hampshire 18 1 18 44 3 16
NOX Vermont 9 4 3 63 1 20
NOX Massachusetts 7 0 11 39 8 34
PM2.5 Maine 57 0 0 2 34 6
PM2.5 New Hampshire 70 0 8 4 11 7
PM2.5 Vermont 89 0 0 3 3 4
PM2.5 Massachusetts 74 0 2 3 8 13
SO2 Maine 9 0 37 2 48 4
SO2 New Hampshire 58 0 38 1 3 1
SO2 Vermont 80 0 0 6 9 4
SO2 Massachusetts 26 0 49 2 17 6
VOC Maine 19 62 0 9 2 7
VOC New Hampshire 21 59 0 9 2 8
VOC Vermont 18 66 0 9 2 5
VOC Massachusetts 38 22 0 19 3 17
7
Top SO2 Emitters (Nationwide)
8
Top NOX Emitters (Nationwide)
9
Chemical Transport Modeling

• Use MM5/SMOKE/CMAQ-DDM3D
• CMAQ-DDM3D
• SAPRC99 (more detailed chemical species, part.
  HCHO)
• DDM3D provides sensitivity fields directly
• Conduct inverse modeling to identify likely
  emissions biases
• Use source-air quality sensitivities and
  observations to modify emissions estimates
• Modifications viewed as suggestive, not absolute.

10
Air quality Model Domain
horizontal domain
vertical structure
36km grid over US, southern Canada and northern
Mexico corresponds to the RPO (Regional Planning
Organization) unified grid
11
Sensitivity analysis

• Given a system, find how the state
  (concentrations) responds to incremental changes
  in the input and model parameters

If Pj are emission, Sij are the
sensitivities/responses to emission changes This
is done automatically using DDM-3D
12
Sensitivity Analysis

• Calculate sensitivity of gas and aerosol phase
  concentrations and wet deposition fluxes to input
  and system parameters
• sij(t)?ci(t)/?pj
• Brute-Force method
• Must run the model a number of different times
• Inaccurate sensitivities may result due to
  numerical noise propagating in the model
• DDM - Decoupled Direct Method
• Use direct derivatives of governing equations
• Initial and boundary conditions, horizontal
  transport, vertical advection and diffusion,
  emissions, chemical transformation, aerosol
  formation, and scavenging processes

13
Atmospheric Advection-Diffusion Equation and
corresponding sensitivity equation

• ADE equation (IC/BCs not shown)
• Sensitivity equation (semi-normalized)

, Pj is unperturbed field
14
DDM-3D
NOo NO2o VOCio ... T K u, v, w Ei ki BCi ...
3-D Air Quality Model
O3(t,x,y,z) NO(t,x,y,z) NO2(t,x,y,z) VOCi(t,x,y,z)
...
decoupled
DDM-3D Sensitivity Analysis
J
15
Inverse Modeling and Sensitivity Analysis

• Inverse modeling involves using observations
  along with a physical model (e.g., traditional
  air quality) model to estimate model parameters
  and inputs, e.g., emissions

Inputs
Model
Output

Need how model responds Sensitivity
Observations
16
Emissions Inventory Assessment usingInverse
Modeling/Four Dimensional Data Assimilation (FDDA)
INPUTS
Emissions inventory (Mobile, area, biogenic,
point sources)
Pollutant distribution (spatial temporal) (e.g.
Ozone, NOx, NOy, SO2, CO, VOCs) and sensitivity
fields
Air Quality Model DDM-3D
Other inputs that remain as defined in the base
case scenario
New emissions distribution by source that
minimize the difference between observations and
simulations
Ridge regression Module
Observations taken from routine
measurement networks or special field studies
Main assumption in the formulation A driving
source for the discrepancy between predictions
and observations is the emission estimates
17
Estimated emission adjustments forSoutheast
emissions using FDDA


Using only IMPROVE measurements
Includes mobile and area sources
18
Plan

• Applying approach to August 2003
• Identify initial inventory and model performance
  issues
• Look at impact of blackout
• Extend inverse method to use satellite
  observations
• Apply to INTEX period
• Further assess inventory
• Reconcile bottom-up and top-down emissions
  estimates

19
Considerations

• SO2 emissions estimates most accurately
  quantified
• Good ability to simulate sulfate (dominant PM
  species in east)
• NOx emissions estimates quantified better where
  major point sources dominate
• Ohio River Valley (e.g., West VA)
• Southeast (TN-NC)
• Interesting experiments over time
• Plants applying NOx and SO2 controls
• 25-85 reductions
• Seasonal variation (summer season application)
• Blackout