Will ensemble approach improve surface PM2.

5
estimate from space? 

"
Jun Wang
Yun Yue, Xiaoguang Xu

Yang Liu

Robert Levy Suomi-NPP launch

28 Oct. 2011

James J. Szykman
"
In collaboration with

Lina Balluz VIIRS,13:30 Local Time

Chaoyang Li Robert Holz" 14.1 revs/day
 Work/Data Flow & Approaches"

Black: datasets & model already in place; green: existing model capability and
data flow that will be improved; red: the data and data flow will be created
National Ambient Air Quality Standards 

NAAQS as of Oct. 2011"

12 μg/m3, "
FR, 15 Jan. 2013"
Existing PM2.5 ground monitoring in continental U.S."
20 June 2013" 21 June 2013"

Blue: ~1000 stations using Federal Reference Method (FRM) as part of Air Quality
System (AQS). Measure daily PM2.5 at daily, every 3rd or 6th day frequency. !
!
Red: ~600 stations using a variety of techniques to provide continuous (hourly
resolution) PM2.5 mass, in support of the AIRNow program.!
!
Still, many areas remain unmonitored. "
 Use  of  EPA  Remote  Sensing  Informa4on  Gateway  
 to  deliver  VIIRS  AOD/PM2.5  data  products
•  Current  satellite  WCS:  
Ø  MODIS  C6  (10  km,  3  km,  DB)  
Ø  CALIOP,  GASP  (GOES  AOD)  
Ø  Prototype  NOAA-­‐VIIRS  

•  Establish  OGC  compliant  Web  

Coverage  Service  (WCS  )  between  
PEATE  and  RSIG  to  add  NASA-­‐  VIIRS  
data  (This  project).  

•  GEOS-­‐Chem  scaling  factors  used  to  

create  a  daily  Look-­‐Up-­‐Table  (LUT)  
of  the  spa4al  varying  rela4on  of  
AOD  and  PM2.5  (van  Donkelaar    et.  
al.,  2012,  ES&T)  .  

•  Prototype  use  of  AOD-­‐to-­‐PM2.5  

scaling  factors  via  regional  models  
(WRF-­‐CMAQ  &  WRF-­‐CHEM)  and  
explore  ensemble  type  approach    
http://ofmpub.epa.gov/rsig/rsigserver?index.html" (This  project).  
 SATELLITE MEASUREMENTS OF AEROSOL MASS AND
TRANSPORT
Atmospheric Environment, 1984.
ROBERT S. FRASER
Laboratory for Atmospheric Sciences,NASA/Goddard Space Flight Center, Greenbelt, MD 20771,U.S.A.

YORAMJ. KAUFMAN
University of Maryland in collaboration with Goddard Laboratory for Atmospheric Sciences,
NASA/Goddard Space Flight Center, Greenbelt, MD 20771, U.S.A.
Satellite measurements of aerosol mass and transport 251
and

ALGORITHM
R. L. MAHONEY
Science Systems and Applications, Inc. 10210Greenbelt Road, Seabrook, MD 20706, U.S.A.

Retrieval algorithm
(First received 29 September 1983)

SURFACE
Abstract-The aerosol optical thickness over land is derived from satellite measurements of the radiance of
Aerosol Information
scattered sunlight. REFLECTANCE
Satellite data These data are used to estimate the columnar mass density of particulate sulfur on a day
with a large amount of sulfur. The horizontal transport of the particulate sulfur is calculated using wind
vectors measured with rawins. t AOIPS CONTOURS
AOIPS SITES OPTICAL THICKNESS
GOES MODEL
w&d index:- Satellite observations,
Key COUNTS CLoUDS -
aerosol ATMOSPHERE
optical thickness air pollution,MASS TRANSPORT
remote sensing.
CALIBRATION
VISIBILITY
c

Fig.1. 3.
INTRODUCTION seem to
Algorithm for deriving aerosol properties havesatellite
from been reported. Satellite measurements o
observations.
aerosol optical thickness over land and estimates of the
Air pollution studies involving the total particulate mass of particulate S and its transport are presented
 Atmospheric loading of
particulate sulfur
(gm-2) on 31 July 1980.

Derived from GOES visible

reflectance are
31 JULY 80
13002 •  Aerosol optical thickness
(AOT)/depth (AOD)
51 I
31 JULY 80
13002
•  Columnar amount of sulfur
I
97 90 80 70 64

Fig. 5. The columnar mass density of particulate 7/31/1980!

sulfur. The units are g m-‘. The transport data on
51 I I
Fig. 7 is computed through the boarders shown here.
97 90 80 70 64

Fig. 5. The columnar mass density of particulate sulfur. The units are g m-‘. The transport data on
Table 1. Comparison Fig. 7of
is computed
columnar through
massestheofboarders shown here.
sulfur derived from ground-based and satellite observations. The satellite
observations were made at 13OOGMT on 31 July 1980
1. Comparison of columnar masses2 of sulfur derived
3 from ground-based
4 and satellite
5 observations. 6The satellite 8
1 7
observations were made at 13OOGMT on 31 July 1980
Place Latitude Longitude Particulate Columnar Reference Satellite Ratio columns
2 (deg.
3 N) 4(deg. W) sulfate
5 mass 6 sulfur mass7 8 sulfur mass 7 and 5
Latitude Longitude Particulate @g
Columnar m-“) Reference (pm-*)
Satellite Ratio columns (g m-‘)
(deg. N) (deg. W) sulfate mass sulfur mass sulfur mass 7 and 5
Virginia 38.7 @g m-“)78.3 (pm-*) 38 0.018(g m-‘) Ferman et 0.040 2.3
al. (1981)
ia Virginia38.7 78.338.7 38 78.3 0.018 38 Ferman et 0.0180.040 Stevens 2.3
et 0.040 2.3
al. (1981)
ia 38.7
al. (1984)
78.339.3 38 0.018 Stevens et 0.040 2.3
Near Baltimore 76.4 24 al. (1984) 0.014 Tichler et 0.017 1.2
Baltimore 39.3 76.4 24 0.014 Tichler et 0.017 al. (1981)
1.2
al. (1981)
 NASA Earth Observation System"

18 Dec. 1999"

Updated Jan. 2015"

 Satellite Remote Sensing of Aerosol Transport
09/10/02
09/10/02

09/11/02

09/11/02

09/12/02
09/11/02

MODIS Aerosol Optical Thickness &

700mb Geopotential Height

PM2.5 = A*AOD + B

Wang & Christopher, 2003;

 Past studies on AOD vs. surface PM concentration"
(from Hoff and Christopher, 2009, JAWMA)

2003

2009

multivariate regression, Kriging, neutral network, etc… "

 AOD vs. surface PM is non-linear"

~200 sites over continental U.S.

CTM has been used to provide

ancillary information needed to derive
surface PM2.5 from AOD.!
(Liu et al., 2004; Wang et al., 2010;
van Donkelaar et al., 2010).!
!
PM vs AOT linear correlation coefficient !

Engel-Cox et al., 2004, JAWMA.

 Challenges & Strategies"
•  Challenges: "
•  Vertical distribution, particle size distribution, aerosol composition,
sampling bias !
•  Cloudy conditions!
•  Strategies: !
•  Ensemble modeling using WRF-Chem, WRF-CMAQ, and GEOS-Chem!
•  Spatial & statistical modeling!
!

Satellites"

Models"

aircraft, ships,"
Surface sites" sondes, lidars"
 Direct Use of Reflectance to constrain CTM Model

UNL-VRTM Reflectance
surface reflectance ρGEOS-chem &
algorithm (this work) Jacobian ∂ρ/ ∂τ
GEOS-chem
Harvard
Source,
Sinks MOD04
Chemical MYD04 Reflectance
reactions ρMODIS
Levy et al.
(2007) Keep same dark cloud-
MODIS/Terra
MODIS/Aqua free pixels as MOD04.

τ = τ + ∂ρ/∂ τ × Δρ ρ MODIS =
ρGEOS-chem ?
scale the aerosol mass

Wang et al., 2010, RSE. AOT(τ) and 3D aerosol mass

 Results for April 2008 over China
Wang et al., 2010!
This work"
Model only" Model + MODIS product" Model + MODIS Ref."
A B C

D E F
 Surface PM2.5 climatology"

PM2.5 averaged during 1/1/2001 – 12/31/2006, 10x10 km2"

MODIS & MISR AOD + a CTM (GEOS-Chem)"

van Donkelaar et al. 2010 !

 

An ensemble approach

multiple AOD products + multiple models"

•  Hypothesis: "
–  each satellite AOD product has its unique strengths and
weaknesses, and a combination of them can yield a better AOD
product than any individual product"
•  Questions:"
–  if the climatology of PM2.5-AOD ratio can be better represented by
the ensemble mean of multi-models (instead of one model, GEOS-
Chem, that is currently used); "
–  if the combination of AOD from different sensors and algorithms
together with PM2.5-AOD ratio from (a) can yield the best estimate
of PM2.5 than from each individual source of AOD, and "
–  the cost-and-benefits of using hindcast to estimate surface PM2.5
from AOD. "
 JGR, 2007.!
The ensemble PM2.5 forecast, created by combining six separate forecasts
with equal weighting, is also evaluated and shown to yield the best
possible forecast in terms of the statistical measures considered. "
 Focusing on June 2012"

Monthly GEOS-Chem PM2.5 (ug/m3) overlaid with 650 EPA sites

50oN

40oN

30oN

20oN
120oW 105oW 90oW 75oW 60oW

0 5 10 15 20
6/26/2012 " A case study"
 WRF-Chem" GEOS-Chem"
Correlation coefficient (one-tailed 10% significance level): 412 EPA sites
50oN
Bias (ug/m3) of GEOS-Chem Monthly PM2.5: 650 EPA sites
50oN
40oN

40oN
30oN bias"
30oN
20oN
120oW 105oW 90oW 75oW 60oW

20oN
120oW 105oW 90oW 75oW 60oW
-1.0 -0.5 0.0 0.5 1.0

-10 -5 0 5 10

Correlation coefficient (one-tailed 5% significance level): 332 EPA sites

o
50 N
Bias (%) of GEOS-Chem Monthly PM2.5: 650 EPA sites
o
50 N
40oN

40 o
N
R"
30oN

o
30
20N o
N
120oW 105oW 90oW 75oW 60oW

20oN
o -0.5oW 0.090oW 0.5 75oW
120-1.0
W 105 1.0 60oW

It appears that WRF-Chem does a better job -100in simulating

-50 0 surface
50 100
PM2.5 over Texas, albeit its large positive bias over eastern part of U.S.!
30oN
Correlation coefficient (one-tailed 10% significance level): 262 EPA sites
30ooN
50 N
20oN
120oW 105oW 90oW 75oW 60oW
20ooN
40 N 120oW 105oW 90oW 75oW 60oW
< -1.0 -0.5 0.0 0.5 1.0

30oN < -1.0 -0.5 0.0 0.5 1.0

o
R!
Save as above but with Satellite AOD applied: 278 EPA sites
20 N
50oN 120oW 105oW 90oW 75oW 60oW
Save as above but with Satellite AOD applied: 278 EPA sites
50oN

40oN < -1.0 -0.5 0.0 0.5 1.0

40oN

30oN
Save as above but with Satellite AOD applied: 278 EPA sites
30oN
50ooN
20 N
R after
120oapplying
W MODIS
105 o
W AOD!
90oW 75oW 60oW
20oN
40oN 120oW 105oW 90oW 75oW 60oW
< -1.0 -0.5 0.0 0.5 1.0
 Change of Bias (ug/m3) by Applying Satellite AOD: 476 EPA sites
50oN

40oN

30oN

Applying MODIS AOD!

20oN
120oW 1053o)Wby Applying o
90VIIRS
W AOD: 75
o o
420WEPA sites 60 W
|b2| - |b1|! Change of Bias (ug/m
50oN
< -6 -3 0 3 6

40oN
MODIS-type
VIIRS AOD leads
to significant 30oN
overestimation"
Applying MODIS-type VIIRS AOD!
20oN
120oW 105oW 90oW 75oW 60oW
 80 GC GC
80
Distribution
MODIS-GC of sorted relative bias
VIIRS-GC and correlation
%)" Bias (%)

Sorted Bias (%)

60 coefficient for 60each station"
biasSorted

100
40 100
40

GC GC
Applying "
80
20 80
MODIS-GC
20
VIIRS-GC VIIRS"
Sorted Bias (%)

Sorted Bias (%)

600 60
ABS (relative

0
0 100 200 300 400 500 0 100 200 300 400 500
Sorted Number Sorted Number
40 40
Applying "
20 20
MODIS"
0 0
0 100 200 300 400 500 0 100 200 300 400 500
1.0 Sorted Number 1.0 Sorted Number
0.8 GC GC
MODIS-GC
0.8
VIIRS-GC
R"

0.6 0.6
Sorted Bias (%)

Sorted Bias (%)

Correlation

0.4 0.4
0.2 0.2
1.0 1.0
0.0 GC 0.0 GC
0.8 0.8
MODIS-GC VIIRS-GC
-0.2 -0.2
0.6 0.6
Sorted Bias (%)

Sorted Bias (%)

-0.4 -0.4
0.4 0.4
0 100 200 300 400 500
0.2 Sorted Number 0.20 100 200 300 400 500
Sorted Number
0.0 0.0
 Use  of  EPA  Remote  Sensing  Informa4on  Gateway  
 to  delivery  NASA-­‐VIIRS  AOD/PM2.5  data  products

Ø  User  Interface  for  

New  RSIG  3-­‐
Dimensional  
applica4on.  

Ø  Delivers  data/

products  to  user  
via  web  coverage  
services.  

Ø  Beta  tes4ng  to  

begin  summer  
2015.  
Ø  Early  morning  CALIOP  overpass  shows  extremely  high  
ex4nc4on  (532  nm)  in  the  lowest  2  km  associated  with  
smoke  from  the  RIM  fire  (Aug  2013).  
Ø  VIIRS  AOD  (NOAA-­‐EDR)  captures  high  AOT  covering  most  of  MT  and  moving  into  the  
Dakotas,  both  areas  associated  with  sparse  surface  PM2.5  monitoring  coverage.  
 Bias (ug/m3) of GEOS-Chem Monthly PM2.5: 476 EPA sites
50oN

Applying MODIS AOD reduces biases in GEOS-Chem PM2.5"

40oN

Bias (ug/m3) of GEOS-Chem Monthly PM2.5: 476 EPA sites

o
50oN
30

o
40oN
20
120oW 105oW 90oW 75oW 60oW
Change of Bias (ug/m3) by Applying Satellite AOD: 476 EPA sites
30oN 50oN
< -10 -5 0 5 10

b1!
o
20 N 40oN
o o o o o
120 W 105 W 90 W 75 W 60 W
Save as above but with Satellite AOD applied: 476 EPA sites
o 30oN
50 N
< -10 -5 0 5 10
|b2| - |b1|!
20oN
40oN 120oW 105oW 90oW 75oW 60oW

Save as above but with Satellite AOD applied: 476 EPA sites
30o
o
50 N < -6 -3 0 3 6

b2!
o
40oN
20
120oW 105oW 90oW 75oW 60oW

30oN
< -10 -5 0 5 10

20oN
o o o o o
Change of Daily Absolute-Bias (ug/m3) by Applying Satellite AOD: 476 EPA sites
50oN

40oN

30oN

20oN
120oW 105oW 90oW 75oW 60oW

< -6 -3 0 3 6