Three-State Air Quality Study (3SAQS) Three-State Data Warehouse (3SDW) 2008 CAMx Modeling Model Performance Evaluation Summary University of North Carolina.

Three-State Air Quality Study (3SAQS)

Three-State Data Warehouse (3SDW)

2008 CAMx ModelingModel Performance Evaluation Summary

University of North Carolina (UNC-IE)Cooperative Institute for Research in the Atmosphere (CIRA)

ENVIRON International Corporation (ENVIRON)

June 19, 2014

2

Background• Objective: Develop a 2008 base year air quality modeling

platform for use in NEPA analyses for oil and gas development projects in CO, UT, and WY

• 3SAQS 2008 Base version A (3SAQS_CAMx_Base08a) is developed directly from the WestJumpAQMS Final Base (Base08c) platform

• Key differences in 3SAQS_CAMx_Base08a from WestJumpAQMS Base08c:– Updated MOVES on-road mobile emissions– Updated ancillary emissions data for livestock, on-road mobile, off-road

mobile, nonpoint, and residential wood combustion sources

• All other modeling parameters are exactly the same between WestJumpAQMS and 3SAQS

3

3SAQS CAMx Base08aSummary of Inputs

• WestJumpAQMS WRF 2008 36/12km meteorology• MOZART4 IC/BC• 2008 Emissions– WRAP Phase III 2008 Oil and Gas EI

• 2008 NEI O&G outside of WRAP basins

– MEGAN biogenics– DEASCO3 fires– 2008 NEI (2007v5 and 2008v2 platforms)– Hourly 2008 CEMs for sources reporting to CAMD

4

Model Performance Evaluation• Model species

– Gases: Ozone, CO, NOx– PM: Total PM2.5, EC, OC, SO4, NO3, NH4– Deposition: total N and S (not presented here)

• Performance statistics– Based on recent EPA publications and guidance– Fractional Bias (FB) and Error (FE)– Normalized Mean Bias (NMB) and Error (NME)– Coefficient of determination (R2)

• Performance displays– Scatter plots, soccer plots, time series, and tile plots

5

Ozone Model Performance• Daily Maximum Ozone• Maximum Daily Average 8-hour Ozone (MDA8)• CAMx 12km grid cells paired in space and time

with AQS (urban) and CASTNet (rural) monitors• With and without a 60 ppb floor on

observations• Focus on O3 and precursor performance in 12-

km domain, CO, UT, and WY

12-km Domain Hourly and MDA8 O3

• Biases switch from positive to negative and errors decrease with the 60 ppb threshold

• Model tends to overpredict low observed values and underpredict high observed values

12-km Domain Hourly and MDA8 O3

Colorado Sites Hourly and MDA8 O3

Colorado AQS Sites: Hourly O3, NO2, CO, SO2

Colorado AQS Sites: Monthly Average Hourly

O3 Diurnal ProfilesFa

llSu

mm

erSp

ring

Win

ter

AQS Obs --CAMx --

Utah Sites Hourly and MDA8 O3

Utah AQS Sites: Hourly O3, NO2, CO, SO2

Fall

Sum

mer

Sprin

gW

inte

r

Utah AQS Sites: Monthly

Average Hourly O3 Diurnal

Profiles

AQS Obs --CAMx --

Wyoming Sites Hourly and MDA8 O3

Wyoming AQS Sites: Hourly O3, NO2, CO, SO2

Fall

Sum

mer

Sprin

gW

inte

r

Wyoming AQS Sites: Monthly Average Hourly

O3 Diurnal Profiles

AQS Obs --CAMx --

Utah AQS Sites: December 2008 O3 and NO2

AQS Obs --CAMx --

• Diurnal patterns match well, but mismatch on magnitudes• NO2 diurnal patterns indicate a strong signal from onroad mobile• Under predicting NO2 and concentration spikes related to rush hour traffic

bring O3 closer to observations; indicates need for more NOx in the model• Appears to be an emissions issue, although mixing may be playing a role

overnight

18

Ozone Performance Summary• Seasonality:

– Spring and summer 1-hr and 8-hr shows low bias in all three states at urban and rural sites

– Low bias in all months at all CASTnet sites in three states– Positive bias in fall and December for CO urban sites– Positive bias in fall and winter for UT urban sites– Negative bias in February for WY urban sites

• Diurnal Patterns: – Generally good match with observed diurnal variability, missing the

magnitude– Across the board over estimates in the early morning hours (0-0700 LST)– Lowest biases during peak photochemical hours (1000-1600 LST)

19

NO2, CO, SO2 Performance Summary

• Colorado– Positive NO2 biases for all months other than January– Negative CO biases for all months– Positive SO2 biases for all months (with errors > 100%)

• Utah– Mostly negative biases for all species.– Positive bias for SO2 in February

• Wyoming– Negative biases for all species– Errors > 100% for SO2 in almost all months (exceptions are Jan and Feb)

20

PM Model Performance• Combine program normalizes model and obs PM species• AMET matches the model output for particular locations to the

corresponding observed values from one or more networks of monitors.

• Comparisons of total PM2.5 and constituents by state and over the whole 12-km domain

• Scatter plots of modeled vs. observed PM2.5 at the IMPROVE and CSN monitor locations by season, for each state, and the 12-km domain (note: no total PM2.5 in CASTNET, no CSN monitors in WY)

• Soccer plots of performance measures for PM constituents by month, for each state, and the 12-km domain

21

PM2.5 Performance by Season - 12-km Domain

Winter Spring

Summer Fall

22

PM2.5 Performance by Season - CO

Winter Spring

Summer Fall

23

PM Constituents CAMx vs. IMPROVE – CO SO4 NO3

OC EC

24

SO4 NH4 NO3

OC EC

PM Constituents CAMx vs. CSN – CO

25

PM2.5 Performance by Season - UT

Winter Spring

Summer Fall

26

SO4 NO3

OC EC

PM Constituents CAMx vs. IMPROVE – UT

27

PM2.5 Performance by Season - WY

Winter Spring

Summer Fall

28

PM Constituents CAMx vs. IMPROVE – WYSO4 NO3

OC EC

29

SO4 NH4 NO3

OC EC

PM Constituents CAMx vs. CSN – UT

30

PM Performance Summary• Seasonality: – Winter and summer better correlated (at 12-km and state-

level)– Low bias in summer in all three states

• PM Composition: – The inorganic constituents (esp. NH4 and NO3) are under-biased

in urban sites in UT. – SO4 and OC contribute most to the PM2.5 overprediction in

winter in CO at rural sites.– Considerable overprediction in OC in fall months and EC in

August.

31

PM Performance Summary• Intra-regional differences– Model biased higher in CO than in other two states in all

seasons except in summer in urban locations– Model performance in urban UT sites shows low bias in

summer, and slightly smaller but low bias in fall

• Near- vs. remote-from-source differences:– Better correlation of rural sites with model than urban sites in

winter and spring; opposite trend in summer and fall– Low bias in urban sites in all states in summer