The MODIS Aerosol Retrieval Algorithms

1
The MODIS Aerosol Retrieval Algorithms

• Yoram Kaufman, Didier Tanre, Shana Mattoo
• Lorraine Remer, Rob Levy, Allen Chu, Vanderlei
  Martins, many others

2
The Importance of Aerosols (a brief list)

• Environmental
• Radiation budget
• Cloud formation and rainfall
• Visibility
• Human Health
• Respiratory ailments
• Heart disease

3
What do we want to know about aerosol to
determine its importance?

• Total Amount
• Size distribution
• Geographic location and transport
• Spatial distribution in the horizontal and
  vertical
• Shape
• Optical properties
• Chemical properties

4
MODIS on its own can significantly help us to
determine

• Total Amount
• Size distribution - over ocean
• Geographic location and transport
• Spatial distribution in the horizontal
• Optical properties - light scattering

• MODIS on its own tells us little or nothing
  about
• Size distribution - over land
• Spatial distribution in the vertical
• Shape
• Optical properties - absorption
• Chemical properties

5

• MODIS makes no direct measurements of
• the physical properties of aerosols.
• MODIS measures only reflected radiation and this
  signal is used to derive the physical properties
  of the aerosols!

6
Two MODIS Algorithms

• Ocean and Land algorithms are totally separate
  but use similar techniques to derive aerosol
  properties.
• To increase the signal to noise ratio in the data
  we use a statistical approach that groups many
  pixels.
• Both algorithms are applied to individual boxes
• of 20 x 20 pixels at 500 Meter resolution
• Both algorithms produce 10 Km products.

7
Land or Ocean

• If all pixels in the 10 x 10 kilometer box are
  ocean the Ocean Algorithm is used.
• If any land pixels are observed in the 10 Km box
  the Land Algorithm is used.
• The MOD35 cloud mask product is used to
  determine if each pixel is land or ocean.

8
MODIS retrieval algorithms make use of a
carefully calibrated and properly geo-located
signal.
9

• We can think of any remote sensing retrieval
  algorithm as having three phases
• Removing distortion from the signal
• Separating signal from noise
• Correctly interpreting the signal

10
Removing Distortion

• GAS Correction
• Most earth science remote sensors have to take
  into account changes in the signal due to
  atmospheric gases - water vapor, ozone and CO2.

11
Removing DistortionGas Correction

• To correct the top of atmosphere (TOA)
  reflectance signal for the effects of atmospheric
  gasses MODIS uses
• Ancillary (outside) Data
• NCEP, GDAS - meteorological data for water
  vapor
• TOVS or TOAST - ozone analysis.
• MODIS level two product
• MOD07 - atmospheric profile to determine water
  vapor
• If these are not available climatologically data
  is used as a first guess assumption for these
  values.

12
Ocean Algorithm

• Separating signal from noise
• Cloud Masking
• Sediment Masking
• Statistical treatment of pixels
• Glint Masking

13
Cloud Masking

• Spatial Variability Tests
• Visible channel brightness
• Cirrus Cloud Removal
• -- Near IR Tests
• MOD35 Cloud Product Tests
• -- Infrared Tests

14
Spatial Variability

• Areas of aerosol are usually quite uniform and
  look smooth to our eyes.
• Clouds are generally much less uniform and
• look bumpy.
• 3 x 3 boxes of pixels within the 10 Km Box
• are evaluated in the 0.55 channel to see if they
  are
• uniform or variable.

15
Spatial Variability

• Any group of 9 pixels with
• Standard deviation gt 0.0025 is identified as
  cloud
• The upper left pixel of this group is discarded.
• Information is used from neighboring pixels to
  the right
• and below the 10 Km box to evaluate the status of
  pixels
• near the edge.
• The 10 Km boxes at the right hand edge of the
  swath
• and at the end of the granule are discarded.

16
Cloud
Cloud
Dust Plume
Dust Aerosol
Cloud
Cirrus Cloud
Cloud Shadow
17
Results of 3 x 3 Variability cloud screening
Dust Plume
Center of large smooth clouds not detected
18
This is a spectral test using the ratio of ? 0.47
/ ? 0.66 lt 0.75
We see this naturally since the absorption in
the blue makes dust look brown to our eyes
19
(No Transcript)
20
There are 3 tests we apply to identify Cirrus
clouds
If any one of these tests indicate cirrus is
present we label the pixels as cloudy and mask
them
In this case we label the pixel as
non-cloudy but reduce the quality of the
retrieval for the whole 10 x 10 Km box
21

• After screening for clouds we look for and
  eliminate
• Sediments
• Glint to within 40 degrees of the specular
  reflection

• If there are
• at least 10 remaining pixels in
• the 0.86?m channel
• and 30 remaining pixels in
• all of the other channels
• we will attempt to make an aerosol retrieval

We further eliminate the brightest 25 and
darkest 25 of the remaining pixels in the 0.86
channel. This is done to eliminate any
remaining outliers so that we can arrive at the
true mean reflectance values for the box.
If the pixels that are eliminated in this
final step include those that reduce the quality
of the retrieval, the high quality flag is
restored.
22
Assumptions

• We need to make several assumptions to correctly
  infer the aerosol characteristics from the
  remaining signal. Assumptions include
• Surface reflectance values
• A bi-modal aerosol distribution
• A set of aerosol properties (models)
• for each of the aerosol modes.

23
Separation of signal and noise

• The spectral reflectance's measured by the
  satellite from the remaining pixels contains
  elements of both the ocean surface and
  atmospheric aerosols.
• We still need to remove surface effects.

24
Ocean Surface

• Contributions to the total signal due to
• the ocean surface include
• Sun glint reflection from surface waves
• Reflection from whitecaps
• Lambertian reflectance from underwater scattering.

25

• After removing as much of the noise as possible
  due to the surface signal we are left with an
  optical signal representing aerosols that must be
  correctly interpreted.
• We are still very limited in our ability to
  correctly infer the aerosol properties from this
  signal.
• We use our knowledge and experience to construct
• models of aerosols which represent real world
  conditions.
• With the help of these models we can correctly
• Interpret the measured signal.

26
Aerosol models over oceanAre represented by
theoretical modes
ATBD
27
MODIS aerosol retrieval over ocean
Find one coarse mode and one fine mode that
combine to match the observed spectral
reflectance's
Radius (µm)
28
Look Up Tables - LUT

• The radiative transfer code is computationally
  time
• consuming.
• To make the radiative transfer code run more
  efficiently
• we make use of a set of pre-computed tables
• for the various sets of possible angles and
  amounts of
• aerosol.
• We interpolate from the values in the table for
  angles
• and aot values not in the LUT

29
Creating the LUT
We must consider which atmospheric scenarios
(combination of aerosol Rayleigh surface
other) are representative of what MODIS observes
(including appropriate geometry, MODIS bandwidth
information, etc). LUT utilizes vector
radiative transfer (vRT) code to simulate

• ??a,T?, s? (path radiance, transmission,
  backscattering)
• Combination of Rayleigh Aerosol
• ??s? (surface reflectance)
• Combination of foam / whitecaps (assuming V6
  m/s) water leaving radiance (nonzero at 0.55
  ?m only)