Use of Satellite Remote Sensing for Public Health Applications: The HELIX-Atlanta Experience

Bill CrossonBill CrossonMohammad Al-Hamdan, Maury Estes, Ashutosh Limaye, Mohammad Al-Hamdan, Maury Estes, Ashutosh Limaye,

Dale Quattrochi, Doug RickmanDale Quattrochi, Doug RickmanNASA/MSFC/NSSTCNASA/MSFC/NSSTC

Partners:Partners:Centers for Disease Control and PreventionCenters for Disease Control and Prevention

Kaiser-Permanente GeorgiaKaiser-Permanente GeorgiaU.S. Environmental Protection AgencyU.S. Environmental Protection Agency

Georgia Environmental Protection DivisionGeorgia Environmental Protection DivisionGeorgia Division of Public HealthGeorgia Division of Public Health

Emory UniversityEmory UniversityGeorgia Institute of TechnologyGeorgia Institute of Technology

HELIX-Atlanta Overview

Health and Environment Linked for Information Exchange in Atlanta (HELIX-Atlanta) was developed to support current and future state and local environmental public health tracking (EPHT) programs to implement data linking demonstration projects which could be part of the EPHT Network.

HELIX-Atlanta is a pilot linking project in Atlanta for CDC to learn about the challenges the states will encounter.

NASA/MSFC and the CDC are partners in linking environmental and health data to enhance public health surveillance.

The use of NASA technology creates value – added geospatial products from existing environmental data sources to facilitate public health linkages.

‘The Process is the Product’ – Proving the feasibility of the approach is the main objective

HELIX-Atlanta Timeline

October 2003: Initiate HELIX-Atlanta

January 2004: Gather information on existing systems of potential in a tracking network

April 2004: Select initial projects

September 2004: Begin implementation

September 2005: Signed Business Associate Agreement between NASA/MSFC and Kaiser-Permanente for exchange of Protected Health Information

February 2006: Evaluate initial projects, recommend next steps

HELIX-Atlanta Challenges

Sharing data between agencies with different missions and mindsets Protecting confidentiality of information Ensuring high quality geocoded data Ensuring appropriate spatial and temporal resolutions of environmental data Developing sound resources and methods for conducting data linkages and data analysis

HELIX-Atlanta Respiratory Health Team

RH Team Pilot Data Linkage Project:Link environmental data (MODIS) related to ground-level

PM2.5 with health data related to asthma

Goals: 1. Produce and share information on methods useful for

integrating and analyzing data on asthma and PM2.5 for

environmental public health surveillance. 2. Generate information and recommendations valuable

to sustaining surveillance of asthma with PM2.5 in the

Metro-Atlanta area.

Time period: 2003Domain: 5-county metropolitan Atlanta

Sources of PM2.5 data: EPA AQS

EPA Air Quality System (AQS) ground measurements National network of air pollution monitors Concentrated in urban areas, fewer monitors in rural areas Time intervals rangefrom 1 hr to 6 days(daily meas. every 6th day) Three monitor types:• Federal ReferenceMethod (FRM)• Continuous• Speciation FRM is EPA-acceptedstandard method;processing time 4-6 weeks

FRM sitesNon-FRM sites

AQS PM2.5 Data ExamplesObserved PM 2.5 - 2002

0

5

10

15

20

25

30

35

40

Date

mic

rog

ram

s/m

3

S. DekalbS. Dekalb (Daily data)DoravilleE. Rivers

Observed PM 2.5 - 2003

05

101520253035404550

Date

mic

rog

ram

s/m

3

S. DekalbGwinnett TechS. Dekalb (Daily data)DoravilleE. Rivers

• Excellent agreement between measurements at multiple sites each day• Small seasonal variations• Large day-to-day variations

Sources of PM2.5 data: MODIS

MODIS Aerosol Optical Depth (AOD) Provided on a 10x10 km grid Available twice per day (Terra ~10:30 AM, Aqua ~1:30 PM) Clear-sky coverage only Available since spring 2000 Used in NOAA/EPA AirQuality Forecast Initiative toproduce air quality forecastsfor northeastern US;forecasts for entire US in 2009

MODIS Aerosol Optical Depth (AOD) Data Examples

Smoothed MODIS AOD - 2002

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

Date

MO

DIS

AO

D

S. DekalbDoravilleE. RiversMean

Smoothed MODIS AOD - 2003

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

Date

MO

DIS

AO

DS. DekalbGwinnett TechDoravilleE. RiversMean

• 31-day averages at each site and mean of all sites• 2003 shows higher summer values and more seasonal variation

2002

2003

Estimating PM2.5 from MODIS data

For 2002-2003, obtain MODIS AOD and EPA AQS PM2.5 data

Extract AOD data for AQS site locations

Calculate daily averages from hourly AQS PM2.5 data

Using daily PM2.5 averages from all 5 Atlanta AQS sites, determine statistical regression equations between PM2.5 and MODIS AOD

Apply regression equations to estimate PM2.5 for each 10 km grid cell across region

MODIS AOD - PM2.5 RelationshipPM 2.5 and MODIS AOD - 2002

0

5

10

15

20

25

30

35

40

1/1/

2002

2/1/

2002

3/1/

2002

4/1/

2002

5/1/

2002

6/1/

2002

7/1/

2002

8/1/

2002

9/1/

2002

10/1

/200

2

11/1

/200

2

12/1

/200

2

Date

mic

rog

ram

s/m

3

0.00

0.15

0.30

0.45

0.60

0.75

0.90

1.05

1.20

MO

DIS

AO

D

Mean PM 2.5

Mean MODIS AOD

PM 2.5 and MODIS AOD - 2003

0

5

10

15

20

25

30

35

40

45

Date

mic

rog

ram

s/m

3

0.00

0.20

0.40

0.60

0.80

1.00

1.20

1.40

1.60

1.80

MO

DIS

AO

D

Mean PM 2.5

Mean MODIS AOD

• Daily 5-site means of

observed PM2.5 and MODIS

AOD• MODIS data not available every day due to cloud cover• MODIS AOD follows

seasonal patterns of PM2.5

but not the day-to-day variability in fall and winter

2002

2003

Observed (AQS) and Predicted PM 2.5

Doraville

0

10

20

30

40

4/1/00 10/1/00 4/1/01 10/1/01 4/1/02 10/1/02 4/1/03 10/1/03

Date

PM

2.5

(m

icro

gra

ms/

m3) 31-day mean Observed PM 2.5

31-day mean MODIS-Estimated PM 2.5

South Dekalb

0

10

20

30

40

4/1/00 10/1/00 4/1/01 10/1/01 4/1/02 10/1/02 4/1/03 10/1/03

Date

PM

2.5

(m

icro

gra

ms/

m3)

31-day mean Observed PM 2.5

31-day mean MODIS-Estimated PM 2.5 • MODIS-based predictions

follow seasonal PM2.5 patterns• MODIS AOD is nearly constant in fall and winter,

while observed PM2.5 is not.

Some major events are not captured by MODIS estimates.

PM 2.5 – MODIS AOD Correlations

• Correlations between PM2.5 and MODIS AOD are

generally high (> 0.55) for the warm season.• The lower correlation for MODIS-Aqua in 2002 is for July-September only.

MODIS-Terra MODIS-Aqua2000 --> 0.5792001 --> 0.6432002 --> 0.559 0.4012003 --> 0.661 0.727

April - SeptemberApril - September

0 20 40 60 8010Miles ®

PM2.5 Estimated from MODIS-Terra

Legend

PM2.5

Value

9-11

11-13

13-15

15-17

17-19

19-21

21-23

23-25

25-27

June 24, 2003

Data Linkage

3HealthData

MODISAQS

Linkage

LinkedData

Acute Asthma Visits

Aggregationemailemail

EHTB

NCEH

HELIX - Atlanta Team

EnvironData

EPA NASA

Quality Control Procedure for AQS PM2.5 data

Adapted from CHARM rain gauge network Eliminates anomalous measurements based on ‘Corroborative Neighbor Statistic’ Applied to all daily AQS PM2.5 measurements

before spatial surfaces are built

Suspicious value

Before After

H igh : 50 g/m 3

Low : 0 g/m 3

P M 2 .5 C oncentra tion

E P A s ites

H igh : 50 g/m 3

Low : 0 g/m 3

H igh : 50 g/m 3

Low : 0 g/m 3

P M 2 .5 C oncentra tion

E P A s ites

Spatial surfacing - AQS PM2.5 data

Algorithm adapted from CHARM rain gauge network

1st degree recursive B-spline in x- and y-directions

Daily surfaces created ona 10x10 km grid

Variable number ofmeasurements availableeach day

H igh : 50 g/m 3

Low : 0 g/m 3

P M 2 .5 C oncentra tion

E P A s ites

H igh : 50 g/m 3

Low : 0 g/m 3

H igh : 50 g/m 3

Low : 0 g/m 3

P M 2 .5 C oncentra tion

E P A s ites

Data-rich day

Cross-Validation

a.k.a. ‘bootstrapping’ or ‘omit-one’ analysis Objective: Estimate errors associated with daily spatial surfaces Procedure:

1. Omitting one observation, create surface using N-1 observations2. Compare value of surface at location of omitted observation with

the observed value3. Repeat for allobservations 4. Calculate errorstatistics by day or site

Cross-ValidationDaily Error Statistics

Bootstrap-Observed

0

1

2

3

4

5

6

7

8

9

10

1 26 51 76 101 126 151 176 201 226 251 276 301 326 3512003 Day of Year

RM

SD

(m

icro

gra

ms

/m3 )

RMSD = 2.7 g/m3

Bootstrap-Observed

0

1

2

3

4

5

6

7

8

9

10

235 362 254 250 320 93 282 207 242 337 171 77 63 322 3452003 Day of Year

RM

SD

(m

icro

gra

ms

/m3 )

Rank Order

Time Series

Cross-ValidationError Statistics by Site

Bootstrap-Observed

0

1

2

3

4

5

6

7

1 11 21 31 41 51 61 71 81 91 101 111 121 131 141Site Number

RM

SD

(m

icro

gra

ms

/m3 )

Bootstrap-Observed

0

1

2

3

4

5

6

7

71 72 5 4512513

8 29100 1514

013

7 1 14136 30 64 87 37 33 82 81 5910

1 35 11710

9 17102 89

Site Number

RM

SD

(m

icro

gra

ms/

m3 )

Rank Order

RMSD by Site

MODIS PM2.5 Bias Adjustment

High : 50 g/m 3

Low : 0 g/m 3

PM 2.5 Concentration

EPA sites

H igh : 50 g/m 3

Low : 0 g/m 3

High : 50 g/m 3

Low : 0 g/m 3

PM 2.5 Concentration

EPA sites

65 g/m3

0 g/m3

Assumption: AQS measurements are unbiased relative

to the local mean, but MODIS PM2.5 estimates may have biases. Prefer a ‘regional-scale’ bias adjustment (as opposed to local scale) Procedure:

1. Use a two-step B-spline algorithm to create highly smoothed

versions of the MODIS and AQS PM2.5 daily surface

2. Compute the ‘Bias’ as the difference between the smoothed fields

3. Subtract the bias from the MODIS PM2.5 daily surface to give the

‘bias-corrected’ MODIS daily surfaceSmooth AQSSmooth MODIS

MODIS Bias Legend

Bias

Value

High : 10.59

Low : -22.92

10.6 g/m3

-22.9 g/m3

Merging MODIS and AQS PM2.5 Data

MODIS and AQS data have been merged to produce

final PM2.5 surfaces. MODIS data are weighted lower than AQS. Weights derived through a simplified, time-invariant Kalman filter

approach have been derived and applied to the MODIS PM2.5 estimates.

AQS only

Unadjusted MODIS

Merged(weight = 0.1)

Bias-adjusted MODIS

Linkage of Environmental and Health Data

MembersLON LAT ID AGE GENDER YEAR/MO-84.207 99.200 1 Child M 200301-84.802 99.359 2 Adult M 200301-83.798 99.993 4 Child F 200301

Acute asthma doctor’s office visitsID AGE LON LAT GENDER DATE1811 Child -84.179 99.118 F 1/1/200354767 Adult -84.625 99.802 F 1/1/200384580 Adult -84.679 99.691 F 1/1/2003

Health Data Set

Linkage of Environmental and Health Data

PM2.5 for each visitDate ID Member Lat/Lon Cell Cell Lat/Lon County State Gender Age PM2.51 1 1811 99.572 -84.251 1944 99.552 -84.284 Coweta GA F Child 21.741 2 15299 99.063 -83.860 1608 99.104 -83.806 Upson GA F Child 12.79 1 2 15879 99.727 -84.369 2079 99.731 -84.403 Fulton GA M Child 12.21

Visit counts by grid cell Date Cell Lat Lon County State FC MC FA MA200301 1 99.045 -88.940 Perry MS 1 0 2 0200301 2 99.045 -88.821 Perry MS 0 0 0 0200301 3 99.045 -88.701 Greene MS 0 1 0 1

Data Linkage Outputs

Successes

Negotiated a Business Associate Agreement with a health care provider to enable sharing of Protected Health Information Developed algorithms for QC, bias removal, merging MODIS and AQS PM2.5 data, and others…

Proven the feasibility of linking environmental data (MODIS PM2.5 estimates) with health data (asthma)

Remaining Challenges

Build computer infrastructure to enable public health surveillance Identify and develop environment data sources from NASA or elsewhere that are better suited for public health surveillance Coordinate with state and local agencies to develop public health surveillance networks in their locales