PM Model Performance in Southern California Using UAMAERO-LT

1
PM Model Performance in Southern California Using
UAMAERO-LT

• Joseph Cassmassi
• Senior Meteorologist
• SCAQMD
• February 11, 2004

2
Particulate Modeling in the South Coast Air
Basin Historical Perspective

• 1991 1994 AQMPs Annual PM10 simulated using
  PIC for SO4 and NO3 with CMB and speciated
  rollback
• 1997 AQMP Annual PM10 simulated using UAMLC for
  SO4 and NO3 with CMB and speciated
  rollback gt Simulated PM10 episode using
  UAMAERO
• 2003 AQMP Annual PM10 and PM2.5 using
  UAMAERO-LT gt UAMAERO-LT developed by STI to
  incorporate CBIV gas chemistry and
  empirical partition PM model gt PM
  partitioned into coarse and fine modes based
  on empirical data

3
Establishment of Performance Criteria

• No formal criteria recommended by EPA
• Established 30 error margin for annual average
  in 1997 PM10 modeling protocol (predict
  ed observed) / observed 30
• Error calculated for each species NH4, NO3, SO4,
  OC, EC, Other (Crustal)
• Error averaged by species for PM sites simulated
• Bias reviewed by species and sites simulated
• Complementary quarterly analysis

4
Performance Indicator Debate

• Several indices and sub-regional analyses used
  for gases gt Peak predicted /
  observed gt Bias gt Error
• Peak predicted / observed used for ozone in 2003
  AQMP
• Advisory group recommendations used RRF to assess
  different models/chemical mechanisms

5
Game Plan For 2003 PM

• Original concept annual model Basin for 1995
  and evaluate output for 5 speciated sites
• Requested by EPA to extend analysis beyond 5
  sites to enhance spatial resolution
• Incorporate SSI Hi-Vol data in the analysis
  (evaluate simulation of PM10 mass)
• Conduct grid level analyses to evaluate emissions
• Conduct temporal (daily) evaluations

6
Time Considerations

• Model performance indicators for particulates
  need to be comprehensive because of model
  simulation time requirements
• Annual simulation using UAMAERO-LT including set
  up and post processing gt Xeon Linux Dual
  Processor 3 Days
• gt 5 vertical layers, 65 X 40 grid
• Speciated rollback can be used a quick
  confirmation analysis
• Episodic simulations variable dependent upon
  chemistry and dispersion platform

7
Questions Asked of the Annual Average PM10
Performance Evaluation

• Concentration gt within 30 error? gt
  species proportions reasonable?
• Are predictions at SSI sites reasonable
  (inferring Basin emissions totals in ballpark)?
• Does spatial distribution match observations?
• Are concentrations peaking during the correct
  seasonal?
• Can emissions errors/anomalies be detected?

8
PM2.5 Performance EvaluationExtending the PM10
Criteria to PM2.5

• PM2.5 ratio of PM10 set by empirical
  analysis Species PM2.5/PM10
• NH4 0.90 NO3 0.74 SO4 0.80 OC
  0.73 EC 0.88 Primary Variable
• Use same criteria as PM10 30 absolute error
  for individual PM2.5 species averaged over five
  stations
• Report bias tendency
• Small concentrations exaggerate statistics

9
Model Simulated 1995 Annual PM2.5 (ug/m3)
NH4 NO3 SO4 OC EC OTR
Anaheim 4.2 10.6 2.3 5.1 2.0 5.7
Diamond Bar 4.2 10.9 2.1 4.4 1.7 5.1
Fontana 5.1 12.4 2.7 4.7 2.0 6.4
Los Angeles 3.3 8.0 2.1 4.9 1.9 5.8
Rubidoux 5.4 13.5 2.3 5.2 1.9 6.1
1995 Measured Annual PM2.5 (ug/m3)
NH4 NO3 SO4 OC EC OTR
Anaheim 4.3 8.2 4.2 6.2 2.7 0.7
Diamond Bar 4.6 9.3 4.0 6.4 3.3 0.1
Fontana 4.7 10.8 3.8 7.2 3.3 2.0
Los Angeles 3.9 7.5 3.7 6.5 3.4 0.1
Rubidoux 6.1 14.7 3.2 6.4 2.8 2.6
10
Annual PM2.5 Component Bias (ug/m3) By
Station
NH4 NO3 SO4 OC EC OTR
Anaheim 0.0 2.5 -1.9 -1.0 -0.7 5.0
Diamond Bar -0.3 1.7 -1.9 -2.1 -1.7 5.0
Fontana 0.4 1.6 -1.1 -2.6 -1.4 4.4
Los Angeles -0.6 0.5 -1.5 -1.6 -1.5 5.7
Rubidoux -0.7 -1.2 -0.9 -1.2 -0.9 3.5

Average -0.3 1.0 -1.5 -1.7 -1.2 4.7
Annual PM2.5 Component Percent Absolute Error By
Station
NH4 NO3 SO4 OC EC OTR
Anaheim 0.9 30.2 44.2 16.3 27.1 686.5
Diamond Bar 7.2 18.2 47.4 31.9 50.2 4958.3
Fontana 7.5 15.1 28.0 35.4 41.0 219.7
Los Angeles 15.8 6.1 42.1 24.1 43.7 5658.3
Rubidoux 11.8 7.9 29.3 18.9 30.7 136.4

Average 8.6 15.5 38.2 25.3 38.5 2331.8
11
Graphical Evaluation

• Time Series gt use PM10 analysis for
  estimate of PM2.5 gt at least 75 PM10 is PM2.5
  for each species
• Evaluate SSI Hi-Vol NO3 SO4
• Bivariate plots (Predicted vs. Observed)
• Spatial Mapping gt grid cell analysis above
  threshold gt map particulate emissions

12
Rubidoux
13
Rubidoux
14
Simulated 1995 Annual Average PM10
15
1995 Annual Average PM10
16
Simulated 1995 Annual Average PM2.5(Same Scale
as PM10)
17
Simulated 1995 Annual Average PM2.5(Half Scale
as PM10)
18
31.8
25.1
27.7
35.8
26.3
(1995 PTEP Sites Annual Average PM2.5
Superimposed)
19
Other Model Performance Indicators(1995 Rubidoux
As An Example)

• Annual Average Ozone gt Predicted 1.1
  pphm gt Observed 2.9 pphm
• 24-hr Average Ozone gt Predicted 3.3
  pphm gt Observed 7.4 pphm
• Nitric acid gt Predicted 1.9 µg/m3 gt
  Observed 1.1 µg/m3

20
Uncertainties Contributing to Performance
Evaluation

• Primary emissions are grid specific and
  contribute to several PM2.5 categories EC, OC,
  SO4 crustal
• Ammonia emissions variable
• NOx impact to particulate formation non-linear
• Specification of boundary conditions
  (2003 AQMP used monthly values)

21
Assessment of PM10/2.5 Modeling

• Need advise on rank of importance of the
  different performance measures to model
  acceptance
• Need better meteorology and dispersion platform
• Evaluate and export LT linear chemistry to other
  platforms
• Evaluate full aerosol chemistry