1. Daytime variations of AOD and PM2
Daytime variations of AOD and PM2.5 relationship in the US: Implications for geostationary satellite data applications
Mian Chin1, Alex Coy2, Qian Tan3,4, Tom Kucsera1,5, and Hongbin Yu1
1NASA Goddard Space Flight Center, MD, USA
2Montgomeray Blair High School, MD, USA
3NASA Ames Research Center, CA, USA
4Bay Area Environmental Research Inc., CA, USA
5Universities Space Research Association, MD, USA

2. Introduction
Aerosol particles with diameter less then 2.5 μm (aka PM2.5) is a major pollutant that determines the air quality and affects human health
In recent years, many studies have presented the use of satellite remote sensing of column aerosol optical depth (AOD) to estimate the surface PM2.5 concentrations, but they mostly involve using the polar-orbiting satellite data (once-a-day, “snapshot” observations). There is a lack of knowledge on how the column AOD and surface PM2.5 vary during the day for practical air quality applications
This project examines the daytime variations of AOD and PM2.5 using AOD data from AERONET and surface PM2.5 data from the ground measurement networks at four nearly collocated AOD-PM2.5 sites in the U.S.
Results have implications for using geostationary satellite AOD data over the U.S., e.g., GOES-16, TEMPO, to estimate daytime variations of PM2.5

3. Factors to be considered in interpreting AOD-PM2.5 relationship
Aerosol vertical profile:
AOD is an remote-sensing optical measurement for the entire atmospheric column, whereas PM2.5 is an in-situ measurements of mass at the surface. -> PM2.5 depends on aerosol vertical profile (including PBL or mixed layer height), while AOD corresponds to total aerosol amount in the atmospheric column
Relative humidity (or water vapor):
AOD value increases with the increase of RH, whereas PM2.5 usually refers to the aerosol dry mass -> AOD corresponds to RH but PM2.5 does not
Aerosol composition and particle size:
Different aerosol species and size have different mass extinction efficiency -> even under dry conditions, mass-to-extinction conversion depends on aerosol composition and particle size
Due to the data limitation, we only examine the influence of water vapor. Other factors will be examined either using a reanalysis product or other proxies

4. AOD, PM2.5 and related data Data:
Column 440 nm AOD: AERONET
Column water vapor: AERONET
Surface PM2.5: USA EPA AQS
Hourly average
Pair of AOD-PM2.5 sites selection criteria: AOD and PM measurement sites within 5 km in distance
Near Washington DC: GSFC AOD and Prince George’s County PM2.5 (3.9 km apart), 2012, 2013, 2014
Fresno AOD and PM2.5 (0.1 km apart), 2012, 2013, 2014
St. Louis: St. Louis University AOD and St. Louis City PM2.5 (3.7 km apart), 2013, 2014, 2015
Houston: Univ. Houston AOD and Harris County PM2.5 (2.9 km apart), 2012

5. 1. How closely are AOD and PM2
1. How closely are AOD and PM2.5 correlated each day in different locations?

6. Examples of daytime AOD-PM2
Examples of daytime AOD-PM2.5 correlation and AOD-water vapor Near Washington DC, 2012: “good” AOD-PM2.5 correlation days
Dotted line: PM2.5 data without coincidental AOD data
AOD-PM2.5
AOD-H2Og
AOD varies with both PM2.5 and water vapor
AOD varies with both PM2.5 and water vapor
AOD varies mainly with PM2.5
AOD varies with PM2.5 but not water vapor

7. Examples of daytime AOD-PM2
Examples of daytime AOD-PM2.5 correlation and AOD-water vapor Near Washington DC, 2012: “bad” AOD-PM2.5 correlation days
Dotted line: PM2.5 data without coincidental AOD data
AOD-PM2.5
AOD-H2Og
AOD does not vary with either PM2.5 or water vapor
AOD varies with water vapor but not PM2.5
AOD varies with water vapor but not PM2.5
AOD does not vary with either PM2.5 or water vapor

8. Examples of daytime AOD-PM2
Examples of daytime AOD-PM2.5 correlation and AOD-water vapor over Fresno, CA, 2013: “good” AOD-PM2.5 correlation days
Dotted line: PM2.5 data without coincidental AOD data
AOD-PM2.5
AOD-H2Og
AOD varies with both PM2.5 and water vapor
AOD varies with both PM2.5 and water vapor
AOD varies with PM2.5 but not water vapor
AOD varies with both PM2.5 and water vapor

9. Examples of daytime AOD-PM2
Examples of daytime AOD-PM2.5 correlation and AOD-water vapor over Fresno, CA, 2013: “bad” AOD-PM2.5 correlation days
Dotted line: PM2.5 data without coincidental AOD data
AOD-PM2.5
AOD-H2Og
AOD does not vary with either PM2.5 or water vapor
AOD does not vary with PM2.5 but with water vapor
AOD does not vary with PM2.5 but with water vapor
AOD does not vary with PM2.5 but with water vapor

10. Daily AOD-PM2.5 correlation coefficients in the four locations
Near Washington DC, 2012
Near Washington DC, 2012:
Total available days: 207
16% ( 33): R ≥ 0.7
15% ( 32): 0.5 ≤ R < 0.7
13% ( 26): 0.3 ≤ R < 0.5
14% ( 30): 0.0 ≤ R < 0.3
42% ( 86): R < 0.0
R (AOD-PM2.5)
Fresno, CA, 2013
Fresno, 2013:
Total available days: 225
19% ( 42): R ≥ 0.7
19% ( 42): 0.5 ≤ R < 0.7
11% ( 26): 0.3 ≤ R < 0.5
20% ( 46): 0.0 ≤ R < 0.3
31% ( 69): R < 0.0
R (AOD-PM2.5)
St Louis, MO, 2014
St. Louis, 2014:
Total available days: 217
17% ( 38): R ≥ 0.7
11% ( 23): 0.5 ≤ R < 0.7
12% ( 26): 0.3 ≤ R < 0.5
14% ( 31): 0.0 ≤ R < 0.3
46% ( 99): R < 0.0
R (AOD-PM2.5)
Houston, TX, 2012
Houston, 2012:
Total available days: 181
15% ( 27): R ≥ 0.7
7% ( 13): 0.5 ≤ R < 0.7
8% ( 14): 0.3 ≤ R < 0.5
21% ( 38): 0.0 ≤ R < 0.3
49% ( 89): R < 0.0
R (AOD-PM2.5)

11. Daily AOD-water vapor correlation coefficients
Near Washington DC, 2012
Near Washington DC, 2012:
Total available days: 220
52% (114): R ≥ 0.7
11% ( 24): 0.5 ≤ R < 0.7
9% ( 19): 0.3 ≤ R < 0.5
8% ( 17): 0.0 ≤ R < 0.3
20% ( 45): R < 0.0
R (AOD-H2Og)
Fresno, CA, 2013
Fresno, 2013:
Total available days: 228
31% ( 71): R ≥ 0.7
11% ( 24): 0.5 ≤ R < 0.7
12% ( 28): 0.3 ≤ R < 0.5
16% ( 36): 0.0 ≤ R < 0.3
30% ( 69): R < 0.0
R (AOD-H2Og)
St Louis, MO, 2014
St. Louis, 2014:
Total available days: 224
39% ( 88): R ≥ 0.7
15% ( 34): 0.5 ≤ R < 0.7
10% ( 22): 0.3 ≤ R < 0.5
11% ( 25): 0.0 ≤ R < 0.3
25% ( 55): R < 0.0
R (AOD-H2Og)
Houston, TX, 2012
Houston, 2012:
Total available days: 198
55% (108): R ≥ 0.7
11% ( 22): 0.5 ≤ R < 0.7
9% ( 17): 0.3 ≤ R < 0.5
9% ( 17): 0.0 ≤ R < 0.3
17% ( 34): R < 0.0
R (AOD-H2Og)

12. Summary of R between AOD and PM2.5 for multiple years over the 4 sites
Location
Year
Total days with ≥ 3 hours obs.
% days (days) with R ≥ 0.7
% days (days) with 0.5 ≤ R < 0.7
% days (days) with
0.3 ≤ R < 0.5
% days (days) with 0 ≤ R < 0.3
% days (days) with R < 0
Near Washington DC
(GSFC-PG County, 3.9 km apart)
2012
207
16% (33)
15% (32)
14% (28)
14% (30)
42% (86)
2013
183
13% (24)
11% (20)
11% (21)
19% (34)
46% (84)
2014
170
15% (26)
14% (24)
14% (23)
16% (28)
41% (69)
Fresno (Fresno_2-Fresno, 0.1 km apart)
213
20% (42)
17% (37)
12% (25)
22% (46)
30% (63)
227
19% (42)
11% (26)
20% (46)
31% (69)
169
22% (37)
17% (28)
34% (58)
St. Louis
(St. Louis U.-St. Louis City, 3.7 km apart)
113
11% (12)
11% (13)
57% (64)
217
17% (38)
11% (23)
12% (26)
14% (31)
46% (99)
2015
137
14% (19)
9% (13)
13% (18)
30% (41)
34% (46)
Houston
(U. Houston-Harris, 2.9 km apart)
181
15% (27)
7% (13)
8% (14)
21% (38)
49% (89)
Only 13-22% days when daytime AOD and PM2.5 are correlated with R ≥ 0.7
But 30-57% days when daytime AOD and PM2.5 are un-correlated with R < 0

13. Correlations are not the only thing; slopes and intercepts of AOD-PM2
Correlations are not the only thing; slopes and intercepts of AOD-PM2.5 show large variations on the days when AOD-PM2.5 are correlated at R ≥ 0.7 – Example 1: Near Washington DC, 2012
R (AOD-PM2.5)
Slope (b) of linear fitting PM2.5 = a + b AOD when R ≥ 0.7
Intercept (a) of linear fitting PM2.5 = a + b AOD when R ≥ 0.7
R (AOD-PM2.5)
Slopes vary by a factor near 20 from the lowest to the highest slopes
No clear seasonal pattern
Slope
Intercepts vary with a range of [-45, +10]
No clear seasonal pattern
Intercept
Days in 2012

14. Correlations are not the only thing; slopes and intercepts of AOD-PM2
Correlations are not the only thing; slopes and intercepts of AOD-PM2.5 show large variations on the days when AOD-PM2.5 are correlated at R ≥ 0.7 – Example 2: Fresno CA, 2013
R (AOD-PM2.5)
Slope (b) of linear fitting PM2.5 = a + b AOD when R ≥ 0.7
Intercept (a) of linear fitting PM2.5 = a + b AOD when R ≥ 0.7
Days in 2013
R (AOD-PM2.5)
Slopes vary by a factor near 30 from the lowest to the highest slopes
No clear seasonal pattern
Slope
Intercepts vary with a range of [-170, +25]
No clear seasonal pattern
Intercept

15. Conclusions #1
Daytime variations of AOD and PM2.5 are not always correlated. From the four sites we examined:
Only about 10-20% days when daytime AOD and PM2.5 are correlated at R ≥ 0.7
In comparison, about 30-60% days when daytime AOD and PM2.5 are un-correlated with R < 0
Even in the “good correlation” days when R ≥ 0.7, the linear fitting of PM2.5 = a + b*AOD exhibits significant day-to-day differences in slopes and intercepts, making quantitative estimates of PM2.5 from AOD observations difficult
AOD is much more frequently correlated with the column water vapor than with PM2.5, indicating that the humidification effects on AOD (not on PM2.5) complicate the interpretation of AOD-PM2.5 relationship
Bottom line: Satellite AOD will provide information on possible pollution levels at the surface, but direct use for quantitative estimation of PM2.5 is difficult

16. Example over Washington DC in 2012 and Fresno in 2013
2. What are the errors in estimated daytime PM2.5 based on observed AOD, if we use the monthly statistics of AOD-PM2.5 relationship?
Example over Washington DC in 2012 and Fresno in 2013

17. Near Washington DC: Hourly AOD-PM2.5 relationship in each month 2012
Annual average of co-existing AOD and PM2.5: coincidental AOD_440nm =0.19, PM2.5=10.7 μg m-3 (avg all=11.4 μg m-3)
AOD and PM2.5 are positively correlated on monthly statistics but with large variations (R = , slope = 15-78, intercept = )
Errors in hourly PM2.5 using PM2.5 = a + b・AOD for each month
If we use the monthly linear fitting of AOD and PM2.5 from above, the error in estimated hourly PM2.5 can be as high as >20 μg m-3 for near Washington DC in 2012, although the 80% of the data are within 10 μg m-3

18. Fresno: Hourly AOD-PM2.5 relationship in each month 2013
Annual average of co-existing AOD and PM2.5: coincidental AOD_440nm =0.16, PM2.5=15.4 μg m-3 (avg all=19.3 μg m-3)
AOD and PM2.5 have weak correlations for all months (R = )
Errors in hourly PM2.5 using PM2.5 = a + b・AOD for each month
If we use the monthly linear fitting of AOD and PM2.5 from above, the error in estimated hourly PM2.5 can be as high as >80 μg m-3 for near Fresno in 2013, even though 80% of the data in most months are within 15 μg m-3.

19. Conclusions #2
On monthly basis, the hourly AOD and PM2.5 are mostly positively correlated near Washington DC, but R, slope, and intercept for the monthly data also have large variations (slope and intercept varying > 4x) from month to month. Correlations are much weaker in Fresno
Using the linear fit statistics within a month, the errors in estimated hourly PM2.5 can be as high as > 20 μg m-3 near Washington DC in For Fresno, the monthly correlations are worse; although most errors are within 20 μg m-3, the maximum can be 80 μg m-3. Using the same statistics for a different year would likely result in larger errors.

20. Next steps
Examine AOD-PM2.5 relationship in other locations (e.g., DISCOVER-AQ data over US, and KORUS-AQ data over Korea)
Examine AOD-PM2.5 relationship at longer time scales (e.g., daily mean, monthly mean)
Explain physical reasons of AOD-PM2.5 relationships in addition to the humidification effects on AOD (including mixing layer depth, aerosol vertical profile, aerosol composition and particle size) with a model and/or with available meteorological data
Acknowledgment:
NASA GEO-CAPE study funding
NASA Space Club internship funding
AERONET site managers and calibration teams
EPA AQS site managers and calibration teams

21. Bonus: AOD-PM2.5 relationship in Beijing, China
Beijing’s pollution level is much higher than US’ by a factor of ~8 on annual average
High PM2.5 and high AOD are often observed together
Sites:
Beijing 1: CAMS AOD and Guan Yuan PM2.5 (1 km apart), 2014, 2015, 2016
Beijing 2: Beijing AOD and Olympic Center PM2.5 (1.5 km apart), 2014, 2015, 2016
(Surface PM2.5 data from China Environmental Monitoring Center)

22. Examples of “good” AOD-PM2.5 correlation days in Beijing, 2015
Dashed line/open circles: PM2.5 data with no coincidental AOD data
AOD varies with PM2.5 but not with water vapor

23. Examples of “bad” AOD-PM2.5 correlation days in Beijing, 2015
Dotted line: PM2.5 data without coincidental AOD data
AOD varies with water vapor but not with PM2.5

24. Daily correlation coefficients in Beijing, China, 2015
Daily AOD-PM2.5 vapor correlation coefficients, 2015
Beijing, China, 2015
Beijing, AOD-PM2.5, 2015:
Total available days: 212
36% ( 92): R ≥ 0.7
14% ( 35): 0.5 ≤ R < 0.7
10% ( 25): 0.3 ≤ R < 0.5
15% ( 38): 0.0 ≤ R < 0.3
25% ( 64): R < 0.0
R (AOD-PM2.5)
Daily AOD-water vapor correlation coefficients, 2015
Beijing, China, 2015
Beijing, AOD-water, 2015:
Total available days: 256
50% (127): R ≥ 0.7
11% ( 29): 0.5 ≤ R < 0.7
6% ( 16): 0.3 ≤ R < 0.5
10% ( 26): 0.0 ≤ R < 0.3
22% ( 58): R < 0.0
R (AOD-H2Og)
For all three years ( ) and two pairs of Beijing sites, 30-40% days AOD and PM2.5 are correlated with R ≥ 0.7 and 25-33% days R < 0 – better correlations than sites in the US
For AOD-water vapor, 40-60% days they are correlated with R ≥ 0.7 and 20-25% days R < 0 – Comparable the sites in the US

25. Summary of R between AOD and PM2.5 for multiple years in Beijing
Location
Year
Total days with ≥ 3 hours obs.
% days (days) with R ≥ 0.7
% days (days) with 0.5 ≤ R < 0.7
% days (days) with
0.3 ≤ R < 0.5
% days (days) with 0 ≤ R < 0.3
% days (days) with R < 0
Beijing 1
(CAMS-Guan Yuan, 1.0 km apart)
2014
158
31% (49)
17% (27)
10% (16)
14% (22)
28% (44)
2015
254
36% (92)
14% (35)
10% (25)
15% (38)
25% (64)
2016
198
28% (55)
13% (26)
11% (21)
15% (30)
33% (66)
Beijing 2
(Beijing-Olympic Ctr., 1.5 km apart)
197
34% (67)
10% (20)
9% (18)
16% (31)
31% (61)
212
40% (84)
9% (19)
10% (22)
32% (68)
140
31% (44)
12% (17)
14% (19)
14% (20)
29% (40)
28-40% days when daytime AOD and PM2.5 are correlated with R ≥ 0.7
25-33% days when daytime AOD and PM2.5 are un-correlated with R < 0

26. slopes and intercepts of AOD-PM2
slopes and intercepts of AOD-PM2.5 also show large variations on the days when AOD-PM2.5 are correlated at R ≥ 0.7
R (AOD-PM2.5)
Slope (b) of linear fitting PM2.5 = a + b AOD when R ≥ 0.7
Intercept (a) of linear fitting PM2.5 = a + b AOD when R ≥ 0.7
R (AOD-PM2.5)
Slopes vary by a factor near 90 from the lowest to the highest slopes
Slope
Intercepts vary with a range of [-600, +100]
Intercept
Days in 2015

27. Beijing: Hourly AOD-PM2.5 relationship in each month 2015
Annual average of co-existing AOD and PM2.5: coincidental AOD_440nm =0.65, PM2.5=65.3 μg m-3 (avg all=82.6 μg m-3)
AOD and PM2.5 are positively correlated on monthly statistics but with large variations (R = , slope = , intercept = )
Errors in hourly PM2.5 using PM2.5 = a + b・AOD for each month
If we use the monthly linear fitting of AOD and PM2.5 from above, the error in estimated hourly PM2.5 can be as high as >300 μg m-3 for Beijing in 2015, although 80% of the data are within 70 μg m-3