The Lumped Composition Aerosol LCA Module for Simulating Composition Effects on SOA Formation

1
The Lumped Composition Aerosol (LCA) Module for
Simulating Composition Effects on SOA Formation

• Frank Bowman1, Fei Bian2, Xinlian Chang3
• 1Department of Chemical Engineering, University
  of North Dakota
• 2Department of Chemical Engineering, Vanderbilt
  University / Trinity Consultants
• 3Department of Civil and Environmental
  Engineering, Vanderbilt University / TCEQ
• International Aerosol Modeling Algorithms
  Conference
• 6 December 2007

2
Key Elements of SOA Modeling

• Formation
• Organics react to form SV products (complex
  reaction pathways)
• Partitioning
• SV compounds partition between gas and aerosol
  phases (multicomponent phase equilibria)
• Representation
• thousands of precursors, thousands of SV products
  (model design)
• Models have used different approaches for each of
  these

3
Formation Approaches

• Linear Models
• HC SV products more/other SV
• fixed stoichiometric yields of SV products

4
Formation Approaches

• Linear Models
• HC SV products more/other SV
• fixed stoichiometric yields of SV products
• Nonlinear Models
• SV1 volatile products
• HC SV2 oligomers
• SV3 other reactions
• multiple reaction pathways ? variable
  stoichiometric yields

5
Partitioning Approaches

• No equilibrium
• organics assumed either volatile (gas phase) or
  nonvolatile (aerosol phase)
• Hatakayama-1989, Pandis-1991
• Absorption partitioning
• gas-aerosol equilibrium follows modified Raoults
  law
• Pankow-1994, Odum-1997, Binkowski-1999
• Multiphase
• equilibrium between gas, organic aerosol, and
  aqueous aerosol phases
• Pun-2002, Pankow-2003, Bowman-2004

6
Representation Approaches

• Empirical
• fit chamber data to 1 or 2 product model
• Explicit
• try to include all HCs and reaction products in
  model
• Surrogate
• use selected real compounds as surrogates to
  represent HCs and products
• Lumped
• lumped model compounds are derived based on the
  properties of the mixture they represent

7
Lumped Composition Aerosol (LCA) Module

• Modification of CMAQ aero3 aerosol module
• Formation
• linear fixed stoichiometric yields
• Partitioning
• temperature- composition-dependent absorption
• Representation
• lumped HC gas-phase chemistry (SAPRC99)
• lumped SV products

8
Partitioning Scheme
Gas Phase organics inorganics H2O
Absorption Model with UNIFAC
ISORROPIA
Inorganic Aerosol Phase H2O inorganics
Organic Aerosol Phase SV organics nonvolatile
organics H2O
9
Partitioning Calculations
Calculate Ci Ci,t-1 aijDHCj
Inputs DHCj Ci,t-1 POA T, P, RH
Estimate/Calculate
Calculate gi (using UNIFAC)
Calculate Ci gipoi,TMWi/RT
Solve set of equations xi (calculated as
above) CAER,i Ci Cixi
Outputs Ci CAER,i
10
Lumping Approach
Real Compounds
Lumped Model Compounds

• HC1 ? a1ASV1Aa1BSV1B...
• HC2 ? a2ASV2Aa2BSV2B...
• HC3 ? a3ASV3Aa3BSV3B...
• ...
• HC4 ? a4ASV4Aa4BSV4B...
• HC5 ? a5ASV5Aa5BSV5B...
• HC6 ? a6ASV6Aa6BSV6B...
• ...
• HC7 ? a7ASV7Aa7BSV7B...
• HC8 ? a8ASV8Aa8BSV8B...
• HC9 ? a9ASV9Aa9BSV9B...
• ...

11
Lumping Approach
Real Compounds
Lumped Model Compounds

• HC1 ? a1ASV1Aa1BSV1B...
• HC2 ? a2ASV2Aa2BSV2B...
• HC3 ? a3ASV3Aa3BSV3B...
• ...
• HC4 ? a4ASV4Aa4BSV4B...
• HC5 ? a5ASV5Aa5BSV5B...
• HC6 ? a6ASV6Aa6BSV6B...
• ...
• HC7 ? a7ASV7Aa7BSV7B...
• HC8 ? a8ASV8Aa8BSV8B...
• HC9 ? a9ASV9Aa9BSV9B...
• ...

HC
aASVA aBSVB ...
?
HC
aCSVC aDSVD ...
?
HC
aESVE aFSVF ...
?
SAPRC99
B2 Lumping Method
12
Lumping Procedure

• Equations
• express parameters of lumped semivolatile group
  (a, MW, sj, po, DH) as functions of the
  individual semivolatile component parameters (ai,
  MWi, sji, poi, DHi)
• Criteria
• Specify some basis for dividing mixture
  components into lumped groups
• for HCs, SAPRC uses compound classes (alkane,
  alkene, aromatic, etc.) and reactivity
• for SVs, we use vapor pressure (additional
  classification by chemical structure, polarity,
  etc. possible)

13
Lumping Equations
derived by setting Ylumped Ydetailed at
expected conditions T, Mo, MW, gi1

• (mass conserved)

weighted average
weighted average (not a real structure)

• fit of ln(po) vs. 1/T at Tlow and Thi
• (weighted average has large error)

14
LCA Lumping

• Individual SV products from 5 HC precursor types
  were lumped together
• 2 or 3 lumped groups for each precursor
• Partitioning properties of individual SV products
  determined from experimental data

15
Missing Data?

• But we dont have detailed semivolatile product
  information for most individual SOA precursors!
• Used a combination of approaches
• surrogates (e.g. limonene has same SV products as
  a-pinene)
• empirical (adjust stoichiometric coefficients to
  match chamber data)
• wild guesses (long chain alkanes form an
  arbitrary mixture of long chain acids and diacids)

16
CMAQ-LCA Simulations
CMAQ Scenario

• Input files
• MM5 meteorological files
• SMOKE emission files
• 32 km / 8 km grids, 18 vertical layers
• Default BC and IC files for coarse domain
• Coarse domain outputs BC and IC for fine grid

• SOS99 Episode
• 3-16 July 1999
• Southeastern U.S.
• SEARCH, IMPROVE, SOS99 monitoring sites

17
LCA Simulations
CMAQ Scenario
Compared 6 versions of the LCA aerosol module
18
Comparison to CMAQv4.4
CMAQ Scenario
LCA-Ideal vs. CMAQ vs. Observations

• SOA concentration predictions at JST monitoring
  site similar with CMAQ and LCA-Ideal
• Both models underpredict daily average SOA
  concentrations at JST and other monitoring sites

19
Composition Effects
CMAQ Scenario
LCA-Ideal vs. LCA-Wood vs. LCA-Diesel

• Wood smoke predictions same as Ideal results
• ideal behavior a good assumption when SOA POA
  are similar
• Diesel soot simulation predicts much less SOA
• ? composition effects important when SOA POA
  are different

20
Model Sensitivity
CMAQ Scenario
LCA-Wood vs. LCA-DH vs. LCA-H2O vs. LCA-10
Temperature Dependence lowering DHvap reduces
mean SOA Water Uptake by Organics including
water uptake increases mean SOA of SOA
Products changing the number of lumped groups
alters predictions of mean SOA
21
Regional Statistical Evaluation

• All LCA versions tend to underpredict mean OM2.5
  concentrations
• LCA-Diesel and LCA-Wood-DH by 45
• LCA-Ideal and LCA-Wood by 25
• LCA-Wood-10 by 35
• LCA-Wood-H2O by 15
• better performance against SEARCH data than for
  IMPROVE and SOS99 sites
• Significant model and measurement uncertainties
  remain

22
Summary

• Conclusions
• New LCA module incorporates composition effects
  on semivolatile organic partitioning
• Predictions of SOA production depend strongly on
  model formulation
• Composition effects need to be included in
  atmospheric modeling applications
• Current/Future Work
• Expand composition representation of POA
  emissions
• Include organic partitioning to both organic and
  aqueous phases
• Other precursors, organic aerosol aging, external
  mixtures,
• Funding
• National Science Foundation