3dvar assimilation of modis aod for regional air- quality...

3DVAR assimilation of MODIS AOD for regional air-quality analysis and prediction using WRF/Chem

Zhiquan Liu ([email protected]) NCAR/NESL/MMM

NCAR/MMM: Hui-Chuan Lin, Craig S. Schwartz JCSDA: Quanhua Liu

CMOS/AMS NWP, Montreal 1 06/01/2012

Outline

• Methodology

• Results for a dust storm event over East Asia – Only assimilate MODIS AOD

• Results over North America

– Assimilate both MODIS AOD and AIRNow PM2.5

• Summary and Future work

CMOS/AMS NWP, Montreal 06/01/2012 2

GOCART and WRF/Chem

• The GOCART aerosol module is available within the WRF/Chem model and produces forecasts for 14 aerosol species:

•Hydrophobic and hydrophilic organic carbon (OC1, OC2) •Hydrophobic and hydrophilic black carbon (BC1, BC2) •Sulfate •Dust in 5 particle-size bins [dust{1,2,3,4,5] •Sea salt in 4 particle-size bins [seas{1,2,3,4}]

• WRF-Chem “P25” aerosol variable also an analysis variable •P25 is unspeciated aerosols contributing to PM2.5

15 aerosol variables (mass concentration) to be analyzed


3DVAR aerosol data assimilation

• 3DVAR is to minimize a cost function (in a least square sense) which measures the weighted distance of the model state x to the model

“background” xb and the observations y. In our case for aerosol data assimilation: X are 15 aerosol species mass concentration in 3D space. Xb the “background” of X, short-term forecast from WRF/Chem. Y can be any aeros0l-related observations (in our case, MODIS AOD and surface

PM2.5). H is “observation operator”, which transforms the model state to observation

space. The background error covariance B (having spatial correlation) and observation

error covariance R (no spatial correlation). CMOS/AMS NWP, Montreal

J(x) =12

(x - xb )T B-1(x - xb ) +12

[H(x) - y]T R-1[H(x) - y]

4 06/01/2012

Observation Operators

• MODIS AOD: – Use Community Radiative Transfer Model (CRTM) of Joint Center for

Satellite Data Assimilation (JCSDA) as the observation operator.

• Linear formula for model-simulated PM2.5 (from WRF-Chem) – PM2.5 = ρ[p25 + bc1 + bc2 + 1.8(oc1 + oc2) + dust1 + 0.286*dust2 + seas1 +

0.942*seas2 + 1.375*sulf]


Outline

• Methodology





CMOS/AMS NWP, Montreal

Liu, Z., Q. Liu, H.-C. Lin, C. S. Schwartz, Y.-H. Lee, and T. Wang, 2011: Three-dimensional variational assimilation of MODIS aerosol optical depth: Implementation and application to a dust storm over East Asia. J. Geophys. Res., 116 , D23206, doi:10.1029/2011JD016159.

06/01/2012 6

East Asia domain


261x222 @27 km 45L with top @50 hPa Validation observations: 7 AERONET sites CALIPSO AOD chem_opt=301: GOCART+RACM Emissions: Online biogenic RETRO+”Streets” anthropogenic GOCART dust emission LBC: NCAR CAM-Chem 6-hr cycling DA/FC experiment: 17~24 March, 2010. MET fields updated from GFS. Aerosol fields updated from AOD DA.

06/01/2012 7

L2 MODIS [email protected]μm coverage


0600 UTC, 21 March 2010

purple: dark-surface retrievals from Aqua; gold: dark surface from Terra; blue: deep-blue produced from Aqua.

Data only available at day time (00Z and 06Z), visible band.

0000 UTC, 21 March 2010

06/01/2012 8


Column dust vs. MODIS true color image. 2010032003

06/01/2012 9

Verify @550nm at other 6 AERONET sites



Verify vs. CALIPSO AOD

06/01/2012 11

Vertical distribution of AOD


2010-03-19 17:00 2010-03-20 20:00

06/01/2012 12

Outline

• Methodology






Schwartz, C. S., Z. Liu, H.-C. Lin, and S. McKeen, 2012: Simultaneous three-dimensional variational assimilation of surface fine particulate matter and MODIS aerosol optical depth. J. Geophys. Res., in press.

North America domain


246x164 @20 km 41L with top @50 hPa Observation coverage: AERONET sites (not assimilated) AIRNow PM2.5 sites (assimilated) chem_opt=300: GOCART w/o chemistry Area within the box excluded for PM2.5 verification score calculation: desert area, bad MODIS AOD data

14 06/01/2012

MODIS AOD coverage (only day time)

CMOS/AMS NWP, Montreal 15

NASA Level 2 10km x 10km ocean & land total AOD @ 550 nm from Aqua and Terra. Deep blue product not used in this study.

15Z – 21Z, 17 June 2010 21Z 22 ~ 03Z 23, June 2010

06/01/2012

• Four experiments 1) No data assimilation (continuous WRF-Chem forecast) 2) PM2.5 DA 3) AOD DA 4) AOD+PM2.5 DA

• Cyclic data assimilation with 6-hr cycles beginning 0000 UTC 01 June, ending 1800 UTC 14 July, 2010. (~45 days)

• PM2.5 observations assimilated each cycle, but AOD observations primarily available only at 1800 UTC • All 1800 UTC analyses initialized 48-hr forecasts

• Meteorological fields updated every 6-hrs from 20-km NAM grids


Experimental design

16 06/01/2012

Domain-averaged total aerosol mass concentration @ 1800 UTC

before/after data assimilation


PM2.5 forecast verification • PM2.5 obs impact quickly decreased in the first 1-hour. • MODIS AOD is more efficient than PM2.5 data to correct low model bias.

CMOS/AMS NWP, Montreal 18

Forecast Range (hours) 06/01/2012

Mean AOD bias vs. AERONET


Summary

• A 3DVAR aerosol DA framework allows simultaneous assimilation of aerosol-related observations from multiple sources (both space-borne and ground-based).

• Promising results for both dust storm and general air-quality applications

• Assimilating AOD observations is more efficient than PM2.5 data to reduce model low PM2.5 bias

• Simultaneous assimilation of surface PM2.5 and MODIS AOD produced the best analysis and forecast of PM2.5 and AOD.


Ongoing and Future work

• Direct MODIS radiance DA for aerosol analysis

• Assimilate multi-spectral/sensor/angle AOD products and other aerosol observations

– GOES, AVHRR, SeaWiFS, MISR, future GOES-R/VIIRS … – PM10, Visibility, Lidar ext. coeffs. profiles (both ground- and satellite-based)

• Develop 4DVAR and/or EnDA approaches for aerosol analysis

• Simultaneous assimilation of meteorological and aerosol observations

48th Oholo Conf., Eilat, 6-10 Nov. 2011

3dvar assimilation of modis aod for regional air- quality...

Documents