1. WRAP Technical Work Overview
Regional Planning Organizations’ Meeting
Denver, CO
June 9, 2005
Gail Tonnesen, University of California Riverside
Zac Adelman, University of North Carolina/CEP
Ralph Morris, ENVIRON Corporation
Tom Moore, WRAP

2. Emissions Emissions Inventories
Multiple iterations on emissions and visibility model as new emissions data become available
To date all 2002 modeling uses Preliminary 2002 emissions
Final 2002 emissions inventory is being prepared now
Emissions from EDMS will be used in next iteration.
Emissions data from other RPOs are being included as they becomes available.

3. Visibility Modeling Key issues to be decided:
How do we judge acceptable model performance? (need guidance from EPA.)
Which visibility model to use, CMAQ or CAMx?
Use 36-km or 12-km for control strategy modeling?
Attribution of Haze and Source Apportionment

4. CMAQ 36-km Simulations CMAQ v4.4 36-km grid, 112x148x19 Annual Run
CB4 chemistry
Evaluated using: IMPROVE, CASTNet, NADP, STN, AIR/AQS
BC from 2001 GEOS-CHEM global model (Jacob et al)

5. Summary of Visibility Modeling
Preliminary 2002 version C:
Used MM5 Run 0 for version C and earlier CMAQ runs
Emissions Pre02c
CMAQ version 4.4 beta
Preliminary 2002 version D
Used new MM5 run
Emissions Pre02d.
CMAQ version 4.4 final
CMAQ evaluation results are posted on the RMC web page:

6. Model Performance Evaluation
Will not get “good” performance for long term modeling for all species at all sites for all days:
Errors and uncertainty in emissions, met, and air quality models.
Effects of grid resolution
Error and uncertainty in ambient data.
Recommend a semi-qualitative approach
if it cannot simulate reality, the model should create a reasonable version of reality.
Relative reduction factors are used to calculate progress
Evaluation should also include unpaired-in-space and unpaired-in-time analysis.

7. Model Performance Evaluation
Experimented with many metrics and approaches for presenting results:
Compute and Tabulate over 20 performance metrics
Scatter-plots & time-series plots
Soccer Goal plots
Stacked bar time-series plots
Monthly bias and error time-series
Average of model and data for best &worst 20% days
Bugle plots
PAVE spatial plots with ambient data overlay

8. MPE – Allday Onesite FB FE SO4 16 45 NO3 91 123 OC 34 58 EC -44 66
SOIL
-20 75 CM
-103

9. MPE – Allday Onesite FB FE SO4 23 54 NO3 -20 111 OC 66 103 EC -35 69
SOIL 92 CM
-109
140

10. MPE – Allday Onesite FB FE SO4 -47 70 NO3 -25 120 OC -28 67 EC -66 80
SOIL
-36 91 CM
-165
165

11. MPE – Allday Onesite FB FE SO4 21 44 NO3 45 138 OC 64 75 EC -4 49 SOIL
-11 71 CM
-109
121

12. MPE – Allday Onesite FB FE SO4 -3 37 NO3 63 132 OC 15 36 EC -64 73
SOIL
-69 90 CM
-152
154

13. MPE – Allday Onesite FB FE SO4 16 40 NO3 96 OC 66 71 EC 56 64 SOIL 77CM
-112
118

14. MPE – Allday Onesite FB FE SO4 66 72 NO3 93 104 OC 105 109 EC 135 SOIL
126
128 CM
-41

15. MPE – Allday Onesite FB FE SO4 34 47 NO3 49 91 OC 26 45 EC -29 56 SOIL24 80 CM
-82
114

16. Average Best and Worst 20% Days
Observed and estimated extinction (Bext) calculations at each WRAP IMPROVE sites
Site-specific f(RH) adjustment factors
Rank days by observed total extinction (Mm-1)
BTot = BSO4 + BNO3 + BOC + BEC + BSoil + BCM + BRay
Types of plots:
Scatter plots showing all sites
Bar plot for a single site

17. Model Performance for Average of Worst 20% Days at WRAP IMPROVE Sites
Preliminary 2002 CMAQ Simulation
SO4
NO3 OC EC

18. Grand Canyon NP, AZ
Chiricahua NM, AZ
Extinction (Mm-1) model performance for average of Worst 20% observed days
Grand Canyon
Chiricahua

19. Yellowstone NP, WY
Grand Canyon NP, AZ
Extinction (Mm-1) model performance for average of Best 20% observed days
Grand Canyon
Yellowstone

21. Monthly SO4 Fractional Bias

22. Model outputs vs. IMPROVE ambient overlay plots

23. Choice of Grid Resolution
36-km: 148x112 cells
12-km (nested) 207x186 cells
4-km (not nested) 144x225 cells

24. Emissions Density Plots: NOx Area Jan

25. Emissions Density Plots: NOx Area Jan

26. CMAQ 36km vs. 12km vs. 4km: Riverside County

27. Model Performance Evaluation; IMPROVE SO4
Jan. 36km vs. 4km
Jan. 36km vs. 12km

28. Model Performance Evaluation; IMPROVE SO4
July 36km vs. 4km
July 36km vs. 12km

29. Grid Selection Conclusions
There is no improvement in objective model performance (using bias and error metrics) at the finer grid resolutions.
Factors other than model performance metrics should be using in selecting what grid resolution to use.

30. Use CMAQ and CAMx? CAMx V4.20beta vs. CMAQ V4.4
Using Preliminary 2002 vers D emissions
Compare for February and July

31. SO4 July 2002
CMAQ vs. CAMx
SO4 IMPROVE
SO4 CASTNet
IMPROVE

32. SO4 January 2002
CMAQ vs. CAMx
SO4 Jan IMPROVE
SO4 Jan CASTnet

33. Conclusions: CMAQ vs. CAMx Performance
Both models exhibit very similar, good performance for SO4 in summer
Slight SO4 overestimation in winter, CAMx overestimation greater than CMAQ
Both models have poor NO3 performance
Summer underestimation (CMAQ worse than CAMx)
Winter overestimation (CAMx worse than CMAQ)
OC, EC, TCM, Soil and CM performance mixed
Comparison will be repeated with Final EI
Likely that choice will be based on factors other than model performance evaluation.

34. Source Apportionment Algorithms
Use air quality model to track the contribution of selected emissions sources to PM formation at receptor sites.
Alternative to sensitivity simulations or CALPUFF.
Implemented in CMAQ tagged species source apportionment (TSSA) but has mass conservation errors:
Porting TSSA to new CMAQ release, ready late 2005.
CAMx has PM source apportionment technology (PSAT)
Will run CAMx/PSAT with final 2002 EI
WRAP is using a combination of mass tracking and back trajectory modeling for source attribution.

35. Rocky Mtn. NP, Colorado Day 185 (07/04/05) 14th Worst Day of 2002
PSAT TSSA
Pt-CO Pt-CO
NR&Ar-CO NR&Mob-CO UT_Fire Other
East-Pt East-Pt
Mob-CO
Pt-WY Pt-WY
Pt-NV Pt-NV
Pt-UT Pt-UT
Pt-NM Pt-NM

36. Schedule Final Emissions Inventory to be completed in summer of 2005
Final model selection Fall 2005
Additional source apportionment Fall 2005
Control strategy and emissions sensitivity simulations winter 2005.