CMAQ and REMSAD- Model Performance and Ongoing Improvements

1
CMAQ and REMSAD- Model Performance and Ongoing
Improvements

• Brian Timin, Carey Jang, Pat Dolwick, Norm
  Possiel, Tom Braverman
• USEPA/OAQPS
• December 3, 2002

2
Introduction

• USEPA has performed an annual simulation of CMAQ
  and REMSAD for a 1996 base year
• An operational evaluation has been completed for
  both models
• Performance evaluations have uncovered some
  weaknesses in the model formulation and
  inventories
• OAQPS has identified a list of CMAQ model
  improvement priorities

3
1996 National CMAQ and REMSAD- Model Setup

• CMAQ- May 2001 release w/MEBI solver
• REMSAD- Version 7.01
• Model Setup
• Domain
• CMAQ and REMSAD 36km, 12 layers, 38 m surface
  layer
• Emissions
• CMAQ and REMSAD 1996 NEI w/adjustments,
  processed via SMOKE
• Meteorology 1996 MM5
• Chemistry
• CMAQ CB-IV chemical mechanism w/ fast solver
  (MEBI)
• REMSAD micro-CB-IV chemical mechanism

4
Nationwide Modeling Domains
CMAQ Modeling Domain
REMSAD Modeling Domain
CMAQ National domain is a Lambert conformal
projection from 100W, 40N REMSAD uses a
lat-long projection
5
Notes on Emission Inventory

• Base Year 1996 NEI w/adjustments
• Removal of wildfires, wind blown dust, and
  residential on-site incineration
• PM Transport Factor
• 75 reduction in fugitive dust sources
• Adjusted CA NOx and VOC (non-EGU)
• Revised Temporal Data
• Prescribed burning
• Animal husbandry
• Used results from ORD inverse modeling (monthly
  reductions of 20-60)
• Annual NH3 inventory reduced by 30
• Biogenic Emissions
• BEIS 3.09

6
CMAQ and REMSAD Model Performance

• Completed statistical comparison against
  observations for 12 layer REMSAD and CMAQ
• Data sources IMPROVE network CASTNET dry dep.
  Network NADP wet deposition network CASTNET
  visibility network
• All comparisons paired in time/space
• Statistics and scatterplots for seasonal and
  annual averages
• Calculated performance statistics by year and
  season for each monitoring site
• Thousands of individual numbers only presenting
  gross summary
• Limited data base (in 1996) makes conclusive
  statements re model performance difficult

7
IMPROVE Annual Average Performance Statistics
Annual mean predicted/annual mean observed
REMSAD
CMAQ
8
Seasonal Average Sulfate Performance
9
July Average Sulfate
10
Seasonal Average Particulate Nitrate Performance
11
January Average Particulate Nitrate
12
Seasonal Average Organic Aerosols Performance
13
July Average Organic Aerosols
AE2 Aerosol
14
Seasonal Average Crustal/Other PM2.5 Performance
15
July Average Crustal/Other PM2.5
16
Winter Average NitrateCMAQ 1996 vs. Observed
2001-2002 (IMPROVE and Urban Speciation)
Qualitative comparison of spatial patterns with
more recent urban speciation data
17
Model Performance- Summary of Individual Species

• CMAQ tends to predict higher concentrations than
  REMSAD especially in the West
• REMSAD slightly underpredicts sulfate in the
  East CMAQ slightly overpredicts sulfate
• Nitrate is overpredicted in the East
• Total nitrate (particulate nitric acid) is
  overpredicted in all seasons
• Indicates an overestimation of nitric acid
• REMSAD underpredicts organic carbon CMAQ is
  relatively unbiased
• Large uncertainty in the primary organic
  inventory (no wildfires), the organic
  measurements, and the secondary organic chemistry
• CMAQ is predicting much more biogenic SOA but it
  is using an aerosol yield approach (AE2)
• Much of the biogenic SOA in REMSAD is being
  partitioned into the gas phase

18
Model Performance- Individual Species

• Elemental carbon is generally unbiased
• Large uncertainty in measurement of elemental
  carbon (EC/OC split)
• IMPROVE sites have very low EC concentrations
• Soil/other concentrations are overpredicted
• Inventory issues
• Fugitive dust, unspeciated emissions from
  construction, paved roads, etc. in urban areas
• NADP wet concentration comparisons
• Sulfate
• CMAQ overpredicts in the East REMSAD
  underpredicts
• Nitrate
• Both models overpredict in the East REMSAD
  underpredicts in the West
• Ammonium
• REMSAD underpredicts CMAQ slightly overpredicts
  in the East

19
Next Steps

• Additional evaluation techniques can be applied
• Further comparisons to more recent urban
  speciation data
• Closer look at individual sites, days, seasons,
  regions
• Time series plots
• 20 best/worst days for visibility
• Plan to model 2001 base year
• Significantly more ambient data available
• Continue to look at PM monitoring issues and how
  they affect model performance evaluation
• Uncertainty in nitrate observed data
• EC/OC split
• Monitoring network protocol differences

20
(No Transcript)
21
OAQPS CMAQ Model Improvement Priorities
(non-inventory)

• Winter nitrate overprediction (general nitric
  acid overprediction)
• Chemistry
• Dry deposition
• SOA overpredictions (biogenic) with 2002 release
• Emission factors
• Aerosol yields
• Gas/particle partitioning
• Horizontal diffusion
• Relatively low explicit diffusion
• Run times
• Decreased run time will allow more refined
  modeling of longer time periods

22
Model Improvement Priorities(modeling inventory)

• Ammonia inventory
• Currently using adjusted 1996 inventory based on
  ORD monthly inverse modeling estimates
• Need long term methodological improvements
• Primary organic carbon
• Need improved fire emissions
• May be missing some organic sources
• Primary semi-volatiles?
• Primary unspeciated PM2.5 (PM-Other)
• Modeled concentrations are grossly overestimated
• Unspeciated fraction in certain speciation
  profiles is very high
• Solid waste combustion (89 unspeciated)
• Coal combustion (85 unspeciated)
• Wood waste combustion (65 unspeciated)

23
Primary PM2.5 Emissions CMAQ- Partial
Solution

• The primary PM emissions in the 2001 CMAQ release
  were emitted in the wrong module
• Emitted in the AERO module
• Should be emitted in the VDIFF module
• Problem corrected in the 2002 release
• Primary PM2.5 concentrations reduced by 5-35

24
July Average PM-Other Concentrations
PM2.5-Other 2001 Release
Ratio of 2002/2001 Release
25
Winter Nitrate- CMAQ vs. REMSAD

• Much of the difference in winter nitrate
  predictions between CMAQ and REMSAD can be traced
  to different implementations of the dry
  deposition routines
• Nitrate concentrations were found to be sensitive
  to dry deposition of NH3, HNO3, and NO2
• Improvements and adjustments are needed in both
  CMAQ and REMSAD, particularly in the areas of
• Treatment of snowcover and freezing temperatures
• Specification of land use and surface roughness
• Treatment of soluble species when canopies are
  wet
• January nitrate concentrations agreed to within
  25 after the dry deposition routines were made
  more similar to each other through a series of
  sensitivity runs (with REMSAD)

26
January Nitrate Comparison After Dry Deposition
Sensitivities
27
Dry Deposition- CMAQ

• CMAQ contains 2 dry deposition routines RADMDRY
  and M3DRY
• M3DRY is a new routine
• Many improvements over the old Wesely routine
  (RADMDRY)
• MM5-PX (Pleim/Xiu land surface model) output is
  needed to take advantage of many of the
  improvements in M3DRY
• Most significant change is enhanced deposition
  velocities for soluble species when canopy is wet
• M3DRY does not currently have a temperature
  function or a specific treatment for snow or
  frozen ground
• ORD is working on improvements to M3DRY
• Adding freezing temperature and snowcover
  treatment
• M3DRY may increase dry deposition of soluble
  species (e.g NH3)

28
Additional Issues- CMAQ 2002 Release

• CMAQ 2002 release contains new AE3 aerosol
  mechanism
• Includes ISORROPIA nitrate partitioning and SOA
  gas/particle partioning
• Ran sensitivity test of 2002 release with AE3 for
  January and July 1996
• Particulate nitrate increases due to
  heterogeneous chemistry
• Gas phase N2O5 rate constant lowered
• Added a heterogeneous N2O5 reaction to aerosol
  mechanism
• N2O5---gt HNO3 (particulate nitrate)
• Biogenic SOA increases by a factor of 3 to 4
• AE3 biogenic SOA (July) is too high in parts of
  the country (especially the West)
• Aerosol yields increased by a factor of 4 (in new
  release)
• SOA partitioning is dominated by particle phase

29
AE2 vs AE3January Average Particulate Nitrate
CMAQ 2001- AE2
CMAQ 2002- AE3
AE3 includes both effect of ISORROPIA and
heterogeneous chemistry
30
AE2 vs AE3July Average Biogenic SOA
CMAQ 2001- AE2
CMAQ 2002- AE3
31
SOA Gas/Particle PartitioningJuly Average
Biogenic SOA in Particle Phase
32
Horizontal Diffusion

• Kh in CMAQ may be too low, especially at 36km
  resolution
• CMAQ Kh is indirectly proportional to grid cell
  size
• REMSAD, UAM-V, and CAMx Kh is directly
  proportional to grid cell size
• At 36km resolution the Kh in CAMx is 17,000
  m2/sec and the Kh in CMAQ is 25 m2/sec (both
  using PPM advection)
• Which methodology is more scientifically correct?

33
Summary of OAQPS CMAQ Model Improvement
Priorities (non-inventory)

• Winter nitrate overprediction (general nitric
  acid overprediction)
• Gas phase chemistry (daytime and nighttime)
• Daytime NO2 OH rate constant
• SAPRC
• CB-IV 2002
• Nighttime
• N2O5 gas phase rate constant and heterogeneous
  reaction
• Dry deposition (M3DRY routine)
• Snowcover and freezing temperatures
• Wet canopy
• AE3 SOA overpredictions (biogenic)
• Terpene emission factors
• Aerosol yields
• Gas/particle partitioning

34
Summary of OAQPS CMAQ Model Improvement
Priorities (non-inventory)

• Horizontal diffusion
• Is current methodology OK?
• Does CMAQ need more explicit diffusion when using
  accurate advection schemes (PPM and Bott)?
• Run times
• Can CMAQ be made to run faster?
• 2001 release is 3 times slower than REMSAD
• 2002 release (with CB-IV) is almost 4 times
  slower than REMSAD
• SAPRC will slow it down even more
• OAQPS is working with ORD to address all of the
  above issues