Office of Research and DevelopmentAtmospheric Modeling and Analysis Division, National Exposure Research Laboratory

Use of Remote Sensing Information in Regional Air Pollution Modeling: Examples and Potential Use of VIIRS Products

Rohit MathurAtmospheric Modeling and Analysis Division, NERL, U.S. EPA

VIIRS Aerosol Science and Operational Users Workshop, November 21-22, 2013, College Park, MD

Acknowledgements: George Pouliot, Xing Jia, Robert Gilliam, Jon Pleim

Office of Research and DevelopmentAtmospheric Modeling & Analysis Division, National Exposure Research Laboratory

2

Motivation• Applications of regional AQ models are continuously being extended to

address pollution phenomenon from local to hemispheric spatial scales over episodic to annual time scales

• The need to represent interactions between physical and chemical processes at these disparate spatial and temporal scales requires use of observational data beyond traditional surface networks

Office of Research and DevelopmentAtmospheric Modeling & Analysis Division, National Exposure Research Laboratory

Use of Remote Sensing Information in Regional AQMs

• Evaluation/Verification of model results– High spatial resolution over large geographic regions of remote sensing data

is attractive• Improve estimates of model parameters

– Emissions (e.g, wildland fires, trends/accountability)– Key meteorological parameters (e.g., SST)– Lateral Boundary conditions (LRT effects)– Location and effects of clouds (e.g., photolysis)

• Chemical data assimilation– Improving short-term air quality forecasts– Identification of model deficiencies

• Data Fusion/Reanalysis: combining model and observed fields– For use in health, exposure and ecological studies (2012 NRC Report on

Exposure Science in the 21st Century)

Office of Research and DevelopmentAtmospheric Modeling & Analysis Division, National Exposure Research Laboratory

4

Improving Model Parameter Estimates: Fire Emissions

Courtesy: A. Soja

Surface PM2.5 : June 10-17, 2008

Fire detects have greatly helped with more accurate spatial allocation of emissions, but challenges remain:• Injection height/vertical distribution• Emission factors (the new approach used soil carbon

content) • Ground fire detection

Courtesy: George Pouliot

ObservedNo FireNEI-SmartfireNew Estimate

Office of Research and DevelopmentAtmospheric Modeling & Analysis Division, National Exposure Research Laboratory

5

Significant under-estimation:July 19-24

Large under-estimation (>2x)in OC in mid-July

Diagnosing Model Performance

Evidence of Long-range transport from outside the modeled domain

Model picks up spatial signatures ahead of the front, but under-predictions behind the front (LBCs)

Office of Research and DevelopmentAtmospheric Modeling & Analysis Division, National Exposure Research Laboratory

7

Distribution of measured carbonaceous aerosol at STN sites within domain Regional enhancement in TCM on July 17-20 suggests influence of wildfires on air masses advected into the domain

Further Evidence

7/14/04 7/15/04 7/16/047/13/04

Long Range Transport of Alaskan Plume

7/17/04

Office of Research and DevelopmentAtmospheric Modeling & Analysis Division, National Exposure Research Laboratory

8

Estimating the Impacts of Alaskan fires through Assimilation of Satellite AOD Retrievals

Methodology1. Model based correlation between AOD and column

PM burden (July-August, 2004 data):– [PM]Col.Burden = f(AOD)

• [PM]col. Burden = 9.065 AOD + 0.18 (r2 =0.9)2. Estimate inferred PM2.5 burden:

– [PM]infer = f(AODMODIS)3. Estimate Difference in PM mass loading:

– [PM]infer – [PM]BaseModel

4. Distribute PM2.5 mass difference vertically between predefined altitudes

• Above BL: 2.2 – 4 km (based on Regional East Atmospheric Lidar Mesonet (REALM) data); layers 14-16

5. Speciation: EPA AP-42 emission factors for wildfires: OC (77%), EC(16%), SO4

2- (2%), NO3- (0.2%), Other(4.8%)

• CO/PM2.5 = 10

Adjust Model Initial Conditions 16Z on July 19, 2004

Mathur, 2008 (JGR)

CO Comparisons with NASA DC-8 Measurements during ICARTT

Assim-Base; 1700Z

Enhanced CO associated with concurrently enhanced acetonitrile (CH3CN) – chemical marker for BB

Assimilation helps improve the model predicted CO distributions

Representing the 3D Transport Signature of the Alaskan Plume

Office of Research and DevelopmentAtmospheric Modeling & Analysis Division, National Exposure Research Laboratory

10

MODIS AOD Assimilation: Impact on Surface PM2.5 Model Performance

• Reduced Bias/Error• Improved Correlation

AIR

NO

WJu

ly 1

9-23

, 04

STN

July

20,

04

Domain median surface levels enhanced by 23 - 42% due to Alaskan fires on different days

Office of Research and DevelopmentAtmospheric Modeling & Analysis Division, National Exposure Research Laboratory

Air Quality – Climate InteractionsEstablishing Confidence in Simulated Magnitude and Direction of Aerosol Feedbacks

• Large changes in emissions and tropospheric aerosol burden have occurred over the past two decades– Title IV of the CAA achieved significant

reductions in SO2 and NOx emissions– Large increase in emissions in Asia over the

past decade

11

• Can models capture past trends in aerosol loading and associated radiative effects?

1989-1991 2007-2009

• Can the associated increase in surface solar radiation be detected in the measurements (“brightening effect”) and be used to constrain model results?

0

10

20

30

40

50

1990 1995 2000 2005

SO2

annu

al e

mis

sion

(Tg)

USChinaOECD+Central Europe

• Is the signal (magnitude and direction) detectable in the observations?

Office of Research and DevelopmentAtmospheric Modeling & Analysis Division, National Exposure Research Laboratory

Trend in Aerosol Optical Depth (AOD)2000 2009

MODIS+ SeaWiFS

WRF-CMAQ (sf)

JJA-average

MODIS - level 3 Terra SeaWiFS - level 3 Deep

Blue Missing value in MODIS

(mostly in Sahara Desert) was filled by SeaWiFS (550nm)

533nm

Air Quality – Climate Interaction

Trend in Aerosol Optical Depth (AOD)

MODIS+ SeaWiFS WRF-CMAQ(sf)

East China East US Europe

JJA-average

(2009 minus 2000)

from 1990 to 2009

Trend in clear-sky shortwave radiation

CERES

East China East US Europe

JJA-average (2009 minus 2000)

WRF-CMAQ(sf)

WRF-CMAQ(nf)

brig

hten

ing

Dim

min

g

from 1990 to 2009

Office of Research and DevelopmentAtmospheric Modeling & Analysis Division, National Exposure Research Laboratory

RMSE Change

63%

July 1-31: GHRSST - PathFinder

Red

uctio

n in

Er

ror

Incr

ease

in

Erro

r

T-2mRMSE Change

RMSE and bias reduced with GHRSST. Reduction is even greater compared to NAM 12-km SST data. Implications for representing Bay Breeze and

pollutant transport

GHRSST• 1-km horizontal resolution

global dataset• Daily

Improving Model Parameter Estimates: Sea Surface Temperature

Office of Research and DevelopmentAtmospheric Modeling & Analysis Division, National Exposure Research Laboratory

• Added diagnostic tracers to track impact of lateral boundary conditions: surface-3km (BL) and 3km-model top (FT)– Quantify modeled “background” O3

“FT” contribution to model background

Average: July-August, 2006

Modeled “background” O3

Accurate representation of aloft pollution critical for simulation of surface “background”

Long-Range Transport and “Background” Pollution Levels

• Significant spatial variability• Background could constitute a sizeable fraction of more stringent NAAQS

Office of Research and DevelopmentAtmospheric Modeling & Analysis Division, National Exposure Research Laboratory

Representing Impacts of Long-Range TransportTransport of Saharan Dust: Summer 2006

17

Surface PM concentration in the Gulf states impacted by LRT during July 30-Aug 3

Texas Sites

Dust Transport: 850 mb

Regional Model Driven by Hemispheric LBCs

Office of Research and DevelopmentAtmospheric Modeling & Analysis Division, National Exposure Research Laboratory

Representing Impacts of Long-Range TransportImpact on Model Performance: July 30-August 3, 2006

18

Bias Difference:Base LBCs - Hemis. LBCs

Lower bias in Hemis.

Lower bias in Base

Vertically varying (time-dependent) LBCs are needed to accurately quantify impacts of LRT on episodic regional pollution as well as “background” pollution

Office of Research and DevelopmentAtmospheric Modeling & Analysis Division, National Exposure Research Laboratory

19

Summary• Air quality remote sensing data is useful for model evaluation and

improvements– What level of quantitative agreement is acceptable?– Need for harmonization between assumptions used in retrieval and

CTM process algorithms (e.g., AOD, NO2 columns) for more rigorous quantitative use

• Columnar distributions are a good starting point, but there is a need for better vertical resolution– Discern between BL and FT• Measurements to characterize transport aloft (and subsequent

downward mixing next morning) are needed• Improving the characterization of FT predictions in regional AQMs

will result in improvements in surface-level predictions

• Potential for use in chemical data assimilation– Simultaneous information on multiple chemical species– Combining model and observed information on the chemical state of

the atmosphere has potential for both human-health and climate relevant endpoints