Implementation of a direct sensitivity method into CMAQ

1
Implementation of a direct sensitivity method
into CMAQ

• Daniel S. Cohan, Yongtao Hu, Amir Hakami, M.
  Talat Odman, Armistead G. Russell
• Georgia Institute of Technology, Atlanta, GA
• Presentation to Models-3 Users Workshop
• October 22, 2002

2
SIMULATION
SENSITIVITY
( )
?
CMAQ
CMAQ
( )
?
3
Uses of Sensitivity

• Policy development
• Impact of emission control measures
• Impact of new emitters
• Uncertainty analysis
• Dependence of model results on assumptions
• Inverse modeling (Area of Influence)
• Which emitters affect a receptor

4
Sensitivity Methods

• Brute-Force Method
• Run CMAQ once for a base case
• Run CMAQ again for each of N perturbations
• Direct Decoupled Method (Dunker 1981, Yang et
  al., 1997)
• Solve for sensitivities decoupled from
  concentrations, using the same numerical routines
  in a single CMAQ run
• Local, first-order sensitivities

5
DDM and Brute Force
Conc. (e.g., O3)
DDM
BF
?Ci
a2
a1
Sensitivity tan(a)
Sensitivity Parameter (e.g., NOx Emissions)
Dpj
6
Direct Decoupled Method
Concentrations (t)
Concentrations (t?t)
Chemistry
Advection Diffusion
I.C., B.C., Emissions
Chemistry
Sensitivities (t)
Sensitivities (t?t)
7
Pros Cons

• Brute Force
• ? Simple
• ? Captures non-linearities
• ? Inefficient for large N
• ? Inaccurate for small perturbations
• DDM
• ? Efficient for large N
• ? Accurate for small perturbations
• ? Does not capture non-linearities

8
Demonstration of CMAQ-DDM

• Fall-Line Air Quality Study
• Focus on Georgia
• 12 km horizontal 13 layers
• SAPRC-99 chemistry
• SAMI emissions inventory
• DDM (implemented so far)
• gas-phase
• first-order
• emissions, I.C., B.C.

9
DDM O3 to Isoprene NOx
10
Sensitivity to point NO emissions
DDM to Actual NO Emissions
DDM to 1 mol/s, Layer 6 Emission
11
DDM vs. Brute Force Ozone Initial Conditions
DDM
Brute Force
12
DDM vs. Brute Force Domainwide NOx Emissions
DDM
Brute Force
13
DDM vs. Brute ForceSingle Point NO
DDM
Brute Force
14
Area of Influence

• DDM shows impact of one emitter on concentrations
  domainwide
• To compute the receptor-based Area of
  Influence
• Compute DDM for unit emissions from various
  emitters E
• Interpolate to obtain AOI of receptor R to every
  emitter E
• Scale to amount of emissions at each E

15
AOI Atlanta Ozone to NO
Response to 1 mol/s NO source
Scaled by NO emissions
16
AOI Macon Ozone to NO
Response to 1 mol/s NO source
Scaled by NO emissions
17
Conclusions

• DDM and Area of Influence enhance the
  functionality of CMAQ
• Strong agreement with brute force, even for
  fairly large perturbations
• Future work will incorporate
• higher-order sensitivities
• aerosols
• further exploration of AOI