Deriving Microwave Vegetation Indices with AMSR-E

1
Deriving Microwave Vegetation Indices with AMSR-E
Jiancheng Shi ICESS University of California,
Santa Barbara
2
Importance in Monitoring Vegetation

• Land-atmosphere interaction Boundary Conditions
  in Climatic and Hydrological Modeling
• Albedo Characteristics of radiation cycle
  balance
• Water storage and evaportranspiration
  Characteristics of water cycle and balance
• Major Information in Terrestrial Carbon Cycle
  (Biogeochemical cycle)
• Storage and passway
• Ecology primary productivity et al.,

3
Optical Sensor Derived Vegetation Information
Current most commonly used tools from optical
sensors
NDVI a measure of vegetation greenness from
visible and near infrared spectrum LAI a
measure of total surface area of all leaves
contained in a canopy over an unit area a
geometric parameter
Major effects Atmospheric contaminations and
background soil color. Vegetation type also
significantly impact on accuracy of estimating
LAI.
4
Contributions to Satellite Observed Tb

• The observed brightness temperature (Tb) depends
  on
• vegetation properties
• Optical depth vegetation water content, height,
  density, scatter size and orientation
• Scattering albedo vegetation water content,
  scatter size and orientation
• Vegetation temperature

• surface emissivity
• Surface dielectric constant lt- soil moisture
• Surface temperature
• Soil type
• Surface roughness
• Atmospheric properties

5
Available Passive Microwave Techniques

• Current Techniques
• Microwave polarization difference index MPDI
  Tbv - Tbh
• Normalized polarization index PI
• C(Tbv Tbh)/ (Tbv Tbh)
• Normalized frequency difference index for forest
  EDI
• C(Tbp(f1) Tbp(f2))/ (Tbp(f1) Tbp(f2))

Major problems Greatly affected by surface
soil moisture, roughness, and temperature
Impacted by surface soil moisture and roughness
Applicable to dense forest only
6
Current Problems Outline

• Facing major problems in vegetation monitoring
• Atmospheric effects optical sensors gt microwave
  sensors
• Background effects - microwave sensors gt optical
  sensors
• This study a new algorithm for estimation of
  microwave vegetation indices
• Theoretical base of the algorithm
• Comparisons with NDVI and LAI data globally.

7
AMSR-E on AQUA Satellite
AMSR-E Advanced Microwave Scanning Radiometer
AQUA Satellite
Sensor Specifications

• Launched on May 4, 2002
• Sun-synchronous orbit
• Equatorial crossing at 1330 LST (ascending)

• 12 channel, 6 frequency conically scanning
  passive microwave radiometer
• Earth incidence angle of 55
• Built by the Japan Aerospace Exploration Agency
  (JAXA)

8
Microwave RT Model (?-t model)
At pixel scale, fraction of vegetation cover need
to be considered
Four component emission model
Re-range
For Tb
Ve
Vatt
Subscript p polarization Lp one-way
attenuation factor Superscripts t, v, and s are
for total, vegetation, and surface terms Tv and
Ts is temperatures for vegetation and ground
surface fv is vegetation fraction cover
9
Characteristics of Surface Components at
Different Frequency
AIEM simulations with wide range soil moisture
roughness

• Surface component summary
• Surface emissivity increases as frequency
  increases due to frequency dependence of water
  part of dielectric properties
• Surface emissivities at two adjacent AMSR-E
  frequencies can be well correlated for all soil
  moisture and surface roughness conditions
• They can be described as a linear function with
  the parameters are polarization independent.

10
Validation and Description of Surface Emission
Relationship at Different Frequency
Experimental Data (BARC 1978-1982)
C X at 50 60 V pol
C X at 50 60 H pol
Linear function for surface emissivity at two
adjacent AMSR-E frequencies
Ep(f1) a(f1,f2)b(f1,f2)Ep(f2)
Relative error in of above equation from AIEM
simulated data
XKu RMSE0.9
CX RMSE 0.5
11

Relationship of TBs at Two Frequencies
With two frequencys surface emission relation
Bp
Ap
Relationship of Tbs at two adjacent frequencies
can be described as a linear function with its
slope and intercept depending only on the
vegetation properties.
A and B are defined as the Microwave Vegetation
Indices MVIs. They are independent of ground
emission signals.
12
Deriving Microwave Vegetation Indices
Assumption no polarization dependence in
vegetation components Ve and Vatt
Directly solve two vegetation related components
(slope and intercept) in the linear equations by
the measurements
There are four Microwave Vegetation Indices that
can be derived by AMSR-E measurements with a low
frequency pair A(C,X) and B(C,X), and with a
high frequency pair A(X,Ku) and B(X,Ku).
13
Characteristics of Microwave Vegetation Indices
of A and B Parameters
A is a function of vegetation fraction cover,
temperature, water content, wet biomass, and
scatter shape and sizes B is a function of
vegetation fraction cover, water content, wet
biomass, and scatter shape and sizes, but not
temperature.
For penetrative vegetation surfaces A gt 0 and B
lt 1 A(C,X) lt A(X,Ku) Lower frequency pair lt
higher frequency pair B(C,X) gt B(X,ku) Lower
frequency pair gt higher frequency pair For bare
surfaces For snow covered
surfaces A a (0) and B b (1)
A lt 0 and B lt 1
14
Effects of Terrain

1. Interpreting measurements as from mean incidence
2. Polarization rotation correction based on sum and
   difference

Terrain topography does not impact on deriving
Microwave Vegetation Indices
15
Distribution Pattern of RFI and MVIs of AB
(6.9-10.65 GHz)
Li and Njoku 2003 (left) Percentage of MVIs in
their physical range in July 2005 calculated by
ascending data (right)

• Similar pattern with identified RFI
• They are well associated with urban areas
• Large areas can be still used for land
  monitoring.

16
Data Processing

• Strong RFI or snow effects
• A lt 0 and/or B gt 1
• Weak RFI and Atmospheric condition effects (rain
  and clouds)
• A and B Fluctuations
• Data processing
• Remove A lt 0 or B gt 1
• Run a median filter

Median Filter
17
Data Analyses

• Comparisons of MVIs with NDVI 16-day composite
  and LAI 8-day composite derived from MODIS
• Do MVIs have similar global pattern as NDVI and
  LAI in different seasons?
• Do MVIs provide vegetation phenology (seasonal
  variation) information and how do they compared
  with NDVI and LAI?
• What are the correlations of MVIs with NDVI and
  LAI in different land cover types?
• Do MVIs provide the complementary vegetation
  information to NDVI and LAI?

18
Comparison of Monthly Mean MVIs (AMSR-E) with
NDVI LAI (MODIS) in April, 2003
NDVI
LAI
B(C,X)
A(C,X)
B(X,Ku)
A(X,Ku)
19
Comparison of Monthly Mean MVIs (AMSR-E) with
NDVI LAI (MODIS) in July, 2003
NDVI
LAI
B(C,X)
A(C,X)
B(X,Ku)
A(X,Ku)
20
NDVI and MVI B(X,Ku) During 2003
NDVI 16-Day Composite
The intercept parameter A is well positively
correlated to NDVI and B is inversely correlated
to NDIV
The intercept parameter A is not only affected by
the vegetation properties but also by temperature
and B is only affected by the vegetation
properties
B(X,Ku) 16-Day Mean

• A good agreement on overall global pattern at
  different seasons
• The difference reflects the senors measurement
  sensitivity to different vegetation components
• Optical leafy part of greenness
• Microwave both leafy and woody parts of
  vegetation.

21
Difference in Vegetation Effects on Optical
Microwave Measurements
Vegetation emissivity
Attenuation factor
Lpexp(-t/cos(?))
Components of the optical thickness of vegetation
canopy
Relationship of LAI to the optical thickness of
vegetation canopy
Ulaby, et al., 1986
The microwave measurements should be greatly
affected not only by the canopys absorption
properties but also by their scattering
properties in both leafy and woody components.
22
Coefficient of Variation of NDVI and MVIs
Seasonal Variability
LAI
NDVI
0 1.0
0 1.2
B(C,X)
B(X,Ku)
0 - 0.3
0 - 0.3
A(C,X)
A(X,Ku)
0 1.0
0 1.0
23
Correlations of MVIs with NDVI LAI
NDVI
LAI
B(C,X)
B(X,Ku)
24
Preliminary Analyses
25
Preliminary Analyses
Landcover B(C,X) B(X,Ku) Landcover B(C,X) B(X,Ku)
Evergreen Needleleaf Forest 0.029 0.076 Closed Shrubland 0.157 0.232
Evergreen Broadleaf Forest 0.143 0.179 Open Shrubland 0.756 0.717
Deciduous Needleleaf Forest 0.518 0.458 Grassland 0.667 0.676
Deciduous Broadleaf Forest 0.563 0.516 Cropland 0.409 0.002
Mixed Forest 0.141 0.208 Tundra 0.511 0.468
Woodland 0.456 0.504 Bare Ground 0.177 0.064
Wooded Grassland 0.430 0.389
26
Examples of Correlations between MVIs and NDVIs
Correlations of the grass land in Mongolia
Correlations of the shrub land in Tibet
A(C,X)
A(C,X)
B(C,X)
B(C,X)
NDVI
NDVI
27
Time Series Plots from Shrub-land Examples
Southern Hemisphere
Northern Hemisphere
Top row samples from North America 1) before
green vegetation grow, significant trend in B
parameters decreasing due to frozen -gt thaw
transition 2) The overall sensitivity
differs. Bottom row fully woody shrub 1) no
significant change but with different magnitudes
in B(C,X)s samples 2) Some B(X,Ku) show good
seasonality bu some not.
28
Time Series Plots from Forests Examples
Evergreen Broadleaf Forest


Evergreen Needleaf Forest
Deciduous broadleaf Forest
29
Microwave Technical Summary

1. Based on the characteristics of frequency
   dependence of surface emission signal, we found
   the relationship of Tbs at two adjacent AMSR-E
   frequencies can be described as a linear function
   with its slope and intercept depending only on
   the vegetation properties
2. These slopes and intercepts are the Microwave
   Vegetation Indices (MVIs) that we defined in this
   study. They are independent of ground emission
   signals and depend only on the vegetation
   properties
3. They can be solved directly from the measurements

30
Summary on Physical Measurements

1. The intercept parameter A derived from AMRS-E is
   well positively correlated to NDVI and LAI
2. The intercept is a function of vegetation
   fraction cover, temperature, water content, wet
   biomass, and scatter shape and sizes

1. The slope parameter B derived from AMRS-E is also
   well inversely correlated to NDVI and LAI
2. The slope is a function of vegetation fraction
   cover, water content, wet biomass, and scatter
   shape and sizes

• Microwave derived MVIs provide both leafy and
  woody information
• NDVI reflects greenness and LAI measures leaf
  surface area.
• The measurements are sensitive to different
  vegetation properties While optical sensors
  green biomass Microwave sensors both leaf and
  woody part of scattering and absorption
  properties

31
Summary

• Comparison with NDVI and LAI measurements
• A similar global patterns in difference seasons
• Provide vegetation phenology (seasonal variation)
  information
• Good correlations in some land cover types,
  especially grass and shrub land
• Provide woody part vegetation information.
• Our newly developed microwave vegetation indices
  provide the potential of complementary dataset
  for global vegetation monitoring and
  applications.
• This technique can be also applied to other
  passive microwave satellites like WINSAT,
  TMI/TRMM, SSM/I and SSMR.