LANDFLUX Assessment and Organization
Workshop , June, 2007, Toulouse, France
Atmospheric Correction in the
Reflective Domain of MODIS and
AVHRR Data
Uncertainties Analysis and Evaluation
of AERONET Sites
Eric F. Vermote
Department of Geography, University of Maryland,
and NASA GSFC code 614.5
Crystal Schaaf
Department of Geography, Boston University
and many contributors
Surface Reflectance (MOD09)
The Collection 5 atmospheric correction algorithm is used to produce
MOD09 (the surface spectral reflectance for seven MODIS bands as it
would have been measured at ground level if there were no atmospheric
scattering and absorption).
Goal: to remove the influence of
• atmospheric gases
- NIR differential absorption
for water vapor
- EPTOMS for ozone
• aerosols
- own aerosol inversion
Home page: http://modis-sr.ltdri.org
Movie credit: Blue Marble Project (by R. Stöckli)
Reference: R. Stöckli, E. Vermote, N. Saleous, R.
Simmon, and D. Herring (2006) "True Color Earth Data
Set Includes Seasonal Dynamics", EOS, vol. 87(5), 49-55.
www.nasa.gov/vision/earth/features/blue_marble.html
2
Basis of the MODIS Atmospheric
Correction algorithm
The Atmospheric Correction algorithm relies on
 the use of very accurate (better than 1%) vector radiative transfer
modeling of the coupled atmosphere-surface system (6SV)
 the inversion of key atmospheric parameters (aerosol, water vapor)
 atmospheric correction algorithm look-up tables are created on the basis of
vector RT simulations performed by the 6SV (Second Simulation of a Satellite
Signal in the Solar Spectrum, Vector) code, which enables accounting for
radiation polarization.
May 2005: the release of a β-version of the vector 6S (6SV1.0B)
................................................
....extensive validation and testing.....
................................................
May 2007: the release of version 1.1 of the vector 6S (6SV1.1)
3
6SV Features
Spectrum: 350 to 3750 nm
Molecular atmosphere: 7 code-embedded + 6 user-defined
models
Aerosol atmosphere: 6 code-embedded + 4 user-defined (based
on components and distributions) + AERONET
Ground surface: homogeneous and non-homogeneous
with/without directional effect (10 BRDF + 1 user-defined)
Instruments: AATSR, ALI, ASTER, AVHRR, ETM, GLI, GOES,
HRV, HYPBLUE, MAS, MERIS, METEO, MSS, TM, MODIS,
POLDER, SeaWiFS, VIIRS, and VGT
5
6SV Web page
http://6S.ltdri.org
8
6SV Users (over the World)
Total: 898 users
10
6SV Interface
We provide a special Web interface which can help an inexperienced user
learn how to use 6SV and build necessary input files.
This interface also lets us track the number and location of 6SV users
based on their IP addresses.
9
6SV Validation Effort
The complete 6SV validation effort is summarized in two manuscripts:
 S. Y. Kotchenova, E. F. Vermote, R. Matarrese, & F. Klemm, Validation of a
vector version of the 6S radiative transfer code for atmospheric correction of
satellite data. Part I: Path Radiance, Applied Optics, 45(26), 6726-6774, 2006.
 S. Y. Kotchenova & E. F. Vermote, Validation of a vector version of the 6S
radiative transfer code for atmospheric correction of satellite data. Part II:
Homogeneous Lambertian and anisotropic surfaces, Applied Optics, in press, 2007.
6
Code Comparison Project (1)
Dave Vector
RT3
Vector 6S
SHARM
(scalar)
Monte Carlo
(benchmark)
Coulson’s
tabulated
values
(benchmark)
All information on this
project can be found at
http://rtcodes.ltdri.org
13
Effects of Polarization
Example: Effects of polarization for the mixed (aerosol (from AERONET)
+ molecular) atmosphere bounded by a dark surface.
The maximum
relative error is
more than 7%.
7
Input Data for Atmospheric
Correction
Key atmospheric
parameters
Vector 6S
 surface pressure
 ozone concentration
 column water
 aerosol optical thickness (new)
LUTs
AC algorithm
coarse resolution
meteorological data
MODIS calibrated data
http://modis-sr.ltdri.org
Reference: Vermote, E. F. & El Saleous, N. Z. (2006). Operational atmospheric correction of
MODIS visible to middle infrared land surface data in the case of an infinite Lambertian target,
In: Earth Science Satellite Remote Sensing, Science and Instruments, (eds: Qu. J. et al), vol. 1,
chapter 8, 123 - 153.
15
Calibration Uncertainties
(MODIS)
We simulated an error of ±2% in the absolute calibration across all 7 MODIS bands.
Results: The overall error stays under 2% in relative for all τaer considered.
(In all study cases, the results are presented in the form of tables and graphs. )
Table (example): Error on the surface
reflectance (x 10,000) due to uncertainties in
the absolute calibration for the Savanna site.
Impact of Calibration uncertainties (+/-2%)
Aeroaol optical depth retrieved
0.8
0.7
0.6
0.5
550nm
0.4
470nm
645nm
0.3
870nm
0.2
0.1
0
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Aerosol optical depth true
17
Uncertainties on Pressure and
Ozone
The pressure error has impact on
Impact of ozone uncertainties (+/-0.2cm.atm)
 molecular scattering (specific band)
0.5
Aeroaol optical depth retrieved
 the concentration of trace gases
(specific band)
 τaer (all bands)
Impact of pressure uncertainties (+/-10mb)
Aeroaol optical depth retrieved
0.45
0.45
0.4
0.35
0.3
550nm
0.25
470nm
645nm
0.2
870nm
0.15
0.1
0.05
0
0
0.4
0.1
0.2
0.3
0.4
0.5
Aerosol optical depth true
0.35
0.3
550nm
0.25
470nm
645nm
0.2
870nm
0.15
The ozone error has impact on
 the band at 550 nm (mostly)
0.1
 the band at 470 nm
0.05
τaer
0
0
0.1
0.2
0.3
0.4
the retrieval of
all bands
0.5
Aerosol optical depth true
18
Uncertainties on Water Vapor
Retrieval of the column water vapor content:
 if possible, from MODIS bands 18 (931-941 nm) and 19 (915 – 965 nm) by using
the differential absorption technique. The accuracy is better than 0.2 g/cm2.
 if not, from meteorological data from NCEP GDAS
Impact
of water
vapor
uncertainties(+/-0.2g/cm
(+/-0.2cm) 2)
Impact
of water
vapor
uncertainties
0.5
Aeroaol optical depth retrieved
Table (example): Error on the surface
reflectance (x 10,000) due to uncertainties in
the water vapor content for the Semi-arid site.
0.45
0.4
0.35
0.3
550nm
0.25
470nm
645nm
0.2
870nm
0.15
0.1
0.05
0
0
0.1
0.2
0.3
0.4
0.5
Aerosol optical depth true
19
Uncertainties on the Aerosol
Model
In the AC algorithm, an aerosol model is prescribed depending on the geographic
location. We studied an error generated by the use of an improper model.
Prescribed: urban clean
Additional: urban polluted, smoke low absorption, smoke high absorption
The choice of the aerosol model
is critical for the theoretical
accuracy of the current product
(in particular, for the accuracy
of optical thickness retrievals).
21
Retrieval of Aerosol Optical
Thickness
Original approach: “dark and dense vegetation (DDV) technique”
a linear
relationship between ρVIS and ρNIR
limitation to the scope of dark targets
Current approach: a more robust
“dark target inversion scheme”
a non-linear relationship derived
using a set of 40 AERONET sites
representative of different land
covers
can be applied to brighter targets
20
MODIS Collection 5 Aerosol
Inversion Algorithm
Pioneer aerosol inversion algorithms for AVHRR, Landsat and MODIS (Kaufman et al.)
(the shortest λ is used to estimate the aerosol properties)
Refined aerosol inversion algorithm
 use of all available MODIS bands (land + ocean e.g. 412nm)
 improved LUTs
 improved aerosol models based on the AERONET climatology
 a more robust “dark target inversion scheme” using Red to predict the blue
reflectance values (in tune with Levy et al.)
 inversion of the aerosol model (rudimentary)
22
Error Budget
Goal: to estimate the accuracy of the atmospheric correction under
several scenarios
Input parameters
Values
Geometrical conditions
10 different cases
Aerosol optical thickness
0.05 (clear), 0.30 (average), 0.50 (high)
Aerosol model
Urban clear, Urban polluted, Smoke low
absorption, Smoke high absorption (from
AERONET)
Water vapor content
(g/cm2)
1.0, 3.0, 5.0 (uncertainties ± 0.2)
Ozone content (cm ·
atm)
0.25, 0.3, 0.35 (uncertainties ± 0.02)
Pressure (mb)
1013, 930, 845 (uncertainties ± 10)
Surface
forest, savanna, semi-arid
16
MODIS Overall Theoretical
Accuracy
Overall theoretical accuracy of the atmospheric correction method considering
the error source on calibration, ancillary data, and aerosol inversion for 3 τaer =
{0.05 (clear), 0.3 (avg.), 0.5 (hazy)}:
Reflectance/ value
VI
r3 (470 nm)
0.012
r4 (550 nm)
0.0375
r1 (645 nm)
0.024
r2 (870 nm)
0.2931
r5 (1240 nm) 0.3083
r6 (1650 nm) 0.1591
r7 (2130 nm)
0.048
NDVI
0.849
EVI
0.399
Forest
Savanna
Semi-arid
Aerosol Optical Depth
value
Aerosol Optical Depth
value
Aerosol Optical Depth
clear
avg
hazy
clear
avg
hazy
clear
avg
hazy
0.0052 0.0051 0.0052
0.04 0.0052 0.0052 0.0053
0.07 0.0051 0.0053 0.0055
0.0049 0.0055 0.0064 0.0636 0.0052 0.0058 0.0064 0.1246 0.0051
0.007 0.0085
0.0052 0.0059 0.0065
0.08 0.0053 0.0062 0.0067
0.14 0.0057 0.0074 0.0085
0.004 0.0152 0.0246 0.2226 0.0035 0.0103 0.0164 0.2324 0.0041 0.0095 0.0146
0.0038
0.011 0.0179
0.288 0.0038 0.0097 0.0158 0.2929 0.0045 0.0093 0.0148
0.0029 0.0052 0.0084 0.2483 0.0035 0.0066 0.0104 0.3085 0.0055 0.0081 0.0125
0.0041 0.0028 0.0042
0.16
0.004 0.0036 0.0053
0.28 0.0056
0.006 0.0087
0.03
0.034
0.04
0.471
0.022
0.028
0.033
0.248
0.011
0.015
0.019
0.005
0.006
0.007
0.203
0.003
0.005
0.005
0.119
0.002
0.004
0.004
The selected sites are Savanna (Skukuza), Forest (Belterra), and Semi-arid (Sevilleta).
The uncertainties are considered independent and summed in quadratic.
27
Performance of the MODIS C5
algorithms
To evaluate the performance of the MODIS Collection 5 algorithms, we analyzed 1
year of Terra data (2003) over 127 AERONET sites (4988 cases in total).
Methodology:
Subsets of Level 1B
data processed using
the standard surface
reflectance algorithm
comparison
Reference data set
Atmospherically
corrected TOA
reflectances derived
from Level 1B subsets
If the difference is within
±(0.005+0.05ρ), the
observation is “good”.
Vector 6S
AERONET measurements
(τaer, H2O, particle distribution)
http://mod09val.ltdri.org/cgi-bin/mod09_c005_public_allsites_onecollection.cgi
28
Validation of MOD09 (1)
Comparison between the MODIS band 1 surface reflectance and the reference data set.
The circle color indicates the % of comparisons within the theoretical MODIS 1-sigma error bar:
green > 80%, 65% < yellow <80%, 55% < magenta < 65%, red <55%.
The circle radius is proportional to the number of observations.
Clicking on a particular site will provide more detailed results for this site.
29
Validation of MOD09 (2)
Example: Summary of the results for the Alta Foresta site.
Each bar: date & time when coincident
MODIS and AERONET observations
are available
The size of a bar: the % of “good”
surface reflectance observations
Scatter plot: the retrieved surface
reflectances vs. the reference data
set along with the linear fit results
30
Validation of MOD09 (3)
In addition to the plots, the Web site displays a table
summarizing the AERONET measurement
and geometrical conditions, and shows browse
images of the site.
MOD09-SFC
Similar results are available for all
MODIS surface reflectance products
(bands 1-7).
Percentage of good:
band 1 – 86.62% band 5 – 96.36%
band 2 – 94.13% band 6 – 97.69%
band 3 – 51.30% band 7 – 98.64%
band 4 – 75.18%
31
Validation of MOD13 (NDVI)
Comparison of MODIS NDVI and the reference data set for all available AERONET
data for 2003. Globally, 97.11% of the comparison fall within the theoretical MODIS
1-sigma error bar (±(0.02 + 0.02VI)).
green > 80%, 65% < yellow <80%, 55% < magenta < 65%, red <55%
32
Validation of MOD09 (EVI)
Comparison of MODIS EVI and the reference data set for all available AERONET
data for 2003. Globally, 93.64% of the comparison fall within the theoretical MODIS
1-sigma error bar (±(0.02 + 0.02VI)).
green > 80%, 65% < yellow <80%, 55% < magenta < 65%, red <55%
33
A 0.05 degree global
climate/interdisciplinary long term data
set from AVHRR, MODIS and VIIRS
• NASA: Ed Masuoka (MODAPS), Nazmi Saleous, Jeff Pedelty
(Processing), Sadashiva Devadiga (Quality Assessment), Jim Tucker
& Jorge Pinzon (Assessment)
• UMD: Eric Vermote (Science), Steve Prince (Outreach), Chris Justice
• NOAA: Jeff Privette (Land Surface Temperature)
• South Dakota State University: David Roy (Burned Area)
• Boston University: Crystal Schaaf (BRDF/Albedo)
AVHRR and MODIS Production Systems
AVHRR GAC L1B
1981 - present
MODIS coarse resolution
surface reflectance
2000 - present
MODIS Level 0
2000 - present
-Geolocation
-Calibration
-Cloud/Shadow Screening
-Atmospheric Correction
Land products
Land products
Gridding
AVHRR products
Gridding
MODIS products
List of potential products:
Surface Reflectance, VI,
Land surface temperature/emissivity,
Snow, BRDF/Albedo, Aerosols,
burned area, LAI/FPAR
-Geolocation
-Calibration
-Cloud/Shadow Screening
-Atmospheric Correction
MODIS standard products
Full resolution and
Climate Modeling Grid
(CMG)
Format:
HDF-EOS
Geographic projection 1/20° resolution
Daily, multi-day, monthly
Production of the Beta (Version 1)
Data Set
- Algorithms:
-Vicarious calibration (Vermote/Kaufman)
-Cloud screening: CLAVR
-Partial Atmospheric Correction:
-Rayleigh (NCEP)
-Ozone (TOMS)
-Water Vapor (NCEP)
-Products:
-Daily surface reflectance (AVH09C1)
-Daily NDVI (AVH13C1)
-16-day composited NDVI (AVH13C3)
-Monthly NDVI (AVH13CM)
-Format:
-Linear Lat/Lon projection
-Spatial resolution: 0.05º
-HDF-EOS
-Time Period:
- 1981 – 2000 completed (Beta = ver 1)
-Distribution:
-ftp and web
NOAA-11 - 1992193 (7/11/1992) : Ch1,
Ch2 and NDVI
LTDR Web Page
http://ltdr.nascom.nasa.gov/ltdr/ltdr.html
The calibration of the AVHRR has been
thoroughly evaluated
The coefficients
were consistent within
less than 1%
AVHRR Overall Potential
Theoretical Accuracy
Overall theoretical accuracy of the atmospheric correction method considering
the error source on calibration, ancillary data, and aerosol inversion for 3 τaer =
{0.05 (clear), 0.3 (avg.), 0.5 (hazy)}:
Forest
RMS Errors at
Aerosol Optical
Depth
Reflectance/
Vegetation
Index
Savanna
RMS Errors at
Aerosol Optical
Depth
Semi-arid
RMS Errors at
Aerosol Optical
Depth
Value
Clear
Avg
Hazy
Value
Clear
Avg
Hazy
Value
Clear
Avg
Hazy
r Ch1 (VIS)
0.045
0.010
0.010
0.010
0.086
0.010
0.010
0.010
0.143
0.011
0.010
0.010
r Ch2 (NIR)
0.237
0.009
0.013
0.019
0.196
0.008
0.010
0.014
0.217
0.008
0.010
0.013
r Ch3 (MIR)
0.045
0.001
0.002
0.003
0.086
0.002
0.002
0.003
0.143
0.003
0.003
0.004
NDVI
0.682
0.056
0.058
0.064
0.392
0.043
0.047
0.054
0.206
0.030
0.033
0.038
Accuracy of Long Term AVHRR-NDVI Datasets, Jyoteshwar R. Nagol
, Eric F. Vermote, Stephen D. Prince, Submitted to RSE.
27
Validation of AVHRR LTDR (1)
Name of data
PAL
GIMMS-G
GVI
LTDR (Beta release)
Spatial Resolution
8 km
8 km
Frequency
10 day & monthly
Bimonthly
Availability
1981-2001
1981-2003
16 km
10 day &
monthly
1985-2000
0.05 degree
Daily, multi-day,
monthly
1981- Present
Format
8 bit
Cloud Screen
CLAVR
(Stowe et al. 1995)
None
CLAVR-3
(Stowe et al. 1995)
No
Yes
Raleigh/ ozone
Water Vapor
correction
Yes
8 bit
Temperature (T5)
(James and Kalluri
1994)
No
No
No
No
Yes
Aerosol correction
No
No
No
Yes
(not included in Beta
release)
Calibration
(Rao and Chen
1996)
(Los 1998; Rao and
Chen 1996)
(Rao and
Chen 1996)
(Vermote and
Kaufman 1995)
No
Yes, (Pinzon et al.
2002)
No
Yes
(not included in Beta
release)
Satellite Drift
(SZA)
8 bit
Accuracy of Long Term AVHRR-NDVI Datasets, Jyoteshwar R. Nagol, Eric F. Vermote, Stephen D. Prince, Submitted to RSE.
29
Validation of AVHRR LTDR (2)
Spatial distribution of AERONET sites used to produce
reference data. After masking for cloud, cloud shadow,
and view zenith angle greater than 42 only 580 data
points from 48 AERONET sites remain in the reference
data for 1999.
Accuracy of Long Term AVHRR-NDVI Datasets, Jyoteshwar R. Nagol, Eric F. Vermote, Stephen D. Prince, Submitted to RSE.
29
Validation of AVHRR LTDR (3)
1
1
1 to 1 line
1 to 1 line
PAL_rho1
0.8
PAL_rho2
0.8
Linear (PAL_rho2)
PAL Channel 2
PAL Channel 1
Linear (PAL_rho1)
0.6
0.4
0.6
0.4
0.2
0.2
0
0
0
0.2
0.4
0.6
0.8
1
0
0.2
Reference data - Channel 1
1
0.6
0.8
1
1
1 to 1 line
1 to 1 line
LTDR_rho2
0.8
Linear (LTDR_rho2)
NDVI Statistics
Compared to reference
LTDR_rho1
0.8
Linear (LTDR_rho1)
LTDR Channel 1
LTDR Channel 2
0.4
Reference data - Channel 2
0.6
0.4
0.2
0.6
0.4
Datasets
(VZA < 42)
RMSE
TOA
0.115
PAL
0.066
LTDR
0.039
RMSE
R-square
RMSE
(NDVI < 0.3)
RMSE
(NDVI > 0.3)
0.92
0.062
0.164
0.93
0.058
0.078
0.97
0.019
0.057
0.2
0
0
0
0.2
0.4
0.6
0.8
Reference data - Channel 2
1
0
0.2
0.4
0.6
0.8
1
Reference data - Channel 1
Accuracy of Long Term AVHRR-NDVI Datasets, Jyoteshwar R. Nagol, Eric F. Vermote, Stephen D. Prince, Submitted to RSE.
29
LTDR QA Home Page
AVHRR BRDF/Albedo Product Using the MODIS
Algorithm: Broadband Black-Sky Albedo (July 1999)
Narrowband to broadband conversion:
Liang, S. L. et al, 2002, Remote Sensing of Environment, 84: 25-41
Albedo evaluation
2006 activities
• Produced an AVHRR surface reflectance and NDVI
beta data set using Pathfinder 2 algorithms (vicarious
calibration; Rayleigh, ozone and water vapor
correction).
• Set up a web/ftp interface for data distribution.
• Identified a set of validation sites for use in the
evaluation of the products.
• Evaluated data set and started operational QA
activity (global browse, known issues, time-series
monitoring and trends).
– Identified problems with geolocation, cloud screening, water
vapor correction, QA bits, etc. in Beta (version 1) data set
– Fixes have been made and will be incorporated in version 2
data set.
Outreach workshop January 2007
- LTDR workshop held January 18, 2007 at the UMUC Conference Center
- Held in conjunction with MODIS Collection 5 workshop
- Most in C5 workshop stayed for LTDR Outreach Workshop
- Goal was to present project status, receive feedback on products/schedule
- Approximately 140 attendees, including MODIS/AVHRR project personnel.
- Presentations from LTDR folks (algorithms, science, QA, data formats, evaluation,
intercomparisons with existing AVHRR products)
- Also presentations from international AVHRR experts
- A. Trischenko (CCRS) “Developing the AVHRR and MODIS Long Term
Data Records at the CCRS”
- P. Frost (CSIRO) “Integration of Sensors Applied on South African
Ecosystems (ISAFE)”
- M. Leroy (CESBIO) “African Monsoon Multidisciplinary Analysis (AMMA)”
- Good interaction and feedback.
2007 and beyond
• Produce improved (version 2) surface reflectance and NDVI
data set for 1981-1999 and 2003
• Produce preliminary aerosol-corrected data set for 1999
and 2003
– Use coincident MODIS and AVHRR data to improve aerosol
retrieval and correction in AVHRR
• Release aerosol-corrected surface reflectance and NDVI
data set (version 3)
• Produce BRDF/Albedo
• Produce/Release Land Surface Temperature
• Produce Burned Area
• Release version 4 surface reflectance/NDVI data set
incorporating fixes identified since vers 3 release
– Another workshop will be held in conjunction with version 4 release.