Microwave Vegetation Indices and VWC in NAFE

1
Microwave Vegetation Indices and VWC in NAFE06
T. J. Jackson1, J. Shi2, and J. Tao3
1 USDA ARS Hydrology and Remote Sensing Lab,
Maryland, USA 2 University of California at Santa
Barbara, Santa Barbara, CA 3 Beijing Normal
University, Beijing, China

2
Introduction MVI Motivation

• Soil moisture retrieval using microwave remote
  sensing requires a correction for vegetation
  effects.
• One approach is to use vegetation water content
  (VWC).
• VWC is usually estimated from vegetation indices
  (NDVI, NDWI) derived from optical satellite
  sensors.
• Issues
• Atmospheric effects
• Primarily responsive to the leafy part of the
  vegetation
• Merging multiple data sources in real time

3
MVI Approach

• Basis The effect of vegetation on surface
  emissivity (soil moisture) varies with microwave
  frequency.
• Can this fact be used to derive information on
  both the leafy and woody parts of the vegetation
  canopy, especially if combined with traditional
  indices?
• Objective Evaluate the possibility of monitoring
  vegetation, in particular VWC, using microwave
  vegetation indices (MVIs).

4
Outline

• RT model
• Simulation datasets and computation of indices
• AMSR-E
• Qualitative evaluation of global and seasonal
  patterns
• Comparison to other MW indices
• Preliminary quantitative verification of VWC
  estimates during NAFE06

5
Microwave RT Model (t-w model)
Three component emission model Full canopy
Symbols p polarization ffrequency vvegetation s
soil surfsoil surface qangle wsingle
scattering albedo goptical depthexp(tsec(q)) TB
brightness temperature eemissivity Ttemperature
toptical depth
1 2 3
At the satellite footprint scale, the fraction
of vegetation cover needs to be considered, the
following presentation is for 100 cover. Here it
is assumed that incidence angle is constant.
6
Radiative Transfer Equation (rearranged)
Vegetation Emission Component
Vegetation Attenuation Component
7
Radiative Transfer Equation (rearranged)
Vegetation Emission Component
Vegetation Attenuation Component

• Implies that the measured brightness temperature
  at a given frequency (f) and polarization (p) can
  be linearly related to the soil surface
  emissivity.
• Bare soil TBesurfTsoil
• Dense canopy TBTveg
• Both the slope and intercept are functions of the
  vegetation fractional cover, temperature and
  other physical properties including biomass,
  water content, and characteristics of the scatter
  size, shape and orientation of vegetation canopy.
• The RT equation has too many unknowns for
  solution without some assumptions.

8
Frequency Dependence of the Surface Emission

• If the surface component behaves in a predictable
  manner with freq., it will be possible to reduce
  the number of unknowns when using multifrequency
  observations
• Characteristics of bare surface emission at the
  different AMSR-E frequencies were evaluated using
  a simulated surface emission database for the
  sensor parameters
• Advanced Integral Equation Model (AIEM)
• Frequencies 6.925, 10.65, and 18.7 GHz, V and H,
  and q55
• Soil moistures (2 to 44, 2 interval) surface
  roughness parameters, root mean square heights
  (0.25 cm to 3 cm, 0.25 cm interval) and
  correlation lengths (2.5 cm to 30 cm, 2.5 cm
  interval).
• 2,904 simulated emissivities for each frequency
  and polarization.
• Examined the relationships of frequency pairs

9
Characteristics of Surface Emissivity at
Different Frequencies/Polarizations
Emissivity
Emissivity

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

10
Predict Surface Emissivity at One Frequency from
Another Frequency for a Specific Sensor
Configuration

• A general linear model was fit to the simulated
  data set
• This yielded the following results for the AMSR-E
  channels considered
• This result is key to the vegetation index
  analysis it allows us to minimize the effects of
  the surface

11
Derivation of the MVIs

• Start with the t-w model and re-arrange terms
• Specify the data source (AMSR-E)
• Reduce dimensionality by establishing the
  frequency dependence of surface emissivity
  equations and calibrate these relationships using
  numerical simulation
• Further reduce dimensionality by assuming that at
  satellite scales the VE and VA terms are
  independent of polarization
• Re-arranging terms results in

12
Derivation of the MVIs

• Starting with the t-w model and re-arrange terms
• Specify the data source (AMSR-E)
• Reduce dimensionality by establishing the
  frequency dependence of surface emissivity
  equations and calibrate these relationships using
  numerical simulation
• Further reduce dimensionality by assuming that at
  satellite scales the VE and VA terms are
  independent of polarization
• Re-arranging terms results in
• B is the Microwave Vegetation Index (MVI)

Ratio of Polarization Differences
13
Derivation of the MVIs (2)

• With the three AMSR-E low frequencies available,
  two MVIs can be derived
• Low frequency 6.925, 10.65 GHz
• High frequency 10.65, 18.7 GHz

14
Preliminary Qualitative Interpretation of the MVIs

• B is affected by biomass, water content, and
  characteristics of the size, shape and the
  orientation of scatters in the vegetation canopy.
  
• Qualitative evaluation of global patterns and
  seasonal patterns in specific regions.
• Data set 2003 AMSR-E and MODIS NDVI

15
Global NDVI (April 2003)

• NDVI features
• Source 16 day composites, average of two per
  month
• Averaged up from 1 km to 25 km
• Deserts, Tropical forests show expected patterns.
• MVI products
• Calculated on a daily basis, median filtered over
  5 days, and averaged over the same 16 day period
  of the NDVI products

NDVI
A(6.925,10.65)
A(10.65,18.7)
B(10.65,18.7)
B(6.925,10.65)
16
Global MVIs and NDVI (April 2003)

• The higher frequency MVI shows increased
  vegetation effects
• Somewhat similar to NDVI Differences in the
  tropics, northern mid latitudes

NDVI
A(6.925,10.65)
A(10.65,18.7)
B(6.925,10.65)
B(10.65,18.7)

• Low values of B indicate different pol.
  differences for the two freq., high values
  indicate similar differences.
• Interpretation issues atmosphere, non-vegetated,
  dense vegetation?

17
Seasonal Patterns (April vs. July 2003)
NDVI
B(10.67, 18.7 GHz)
April
July

• NDVI Seasonal increases in the northern
  hemisphere and decreases in the south
• B Seasonal patterns are not as variable as NDVI
  and there are differences in the tropics,
  northern mid latitudes

18
Quantitative Comparison to VWC National Airborne
Field Experiment 2006 (NAFE06)

• 29 Oct 20 Nov 2006, the Murrumbidgee watershed
  in southeastern Australia.
• Lat 35.159S to 34.683S, Lon 145.640W to
  145.900W
• Vegetation types dry pasture, irrigated rice,
  irrigated wheat, irrigated pasture, dry wheat and
  fallow
• Ground Data
• MSR spectral signatures for different land covers
• Survey of land cover
• Observed vegetation water content
• Satellite Data
• Landsat 5 TM data (6 Oct, 7 Nov and 23 Nov)
• AMSR-E L3 data (From 6 Oct to 23 Nov)

19
NDWI VWC Products

• Normalized Difference Water Index (NDWI) is
  sensitive to liquid water molecules in vegetation
  canopies.
• A linear relationship was calibrated using the
  ground and satellite observations for specific
  land covers.
• This is applied with landcover to each Landsat
  data set (Oct. 6, Nov. 7, and Nov. 23).
• Daily VWC maps are derived on specific days by
  interpolation.

20
Deriving MVIs Using AMSR-E Data

• MVI were derived using daily brightness
  temperature data at 6.925 GHz, 10.65 GHz and 18.7
  GHz from AMSR-E (25 km x 25 km gird data) .
• Flags

Criteria Test Criteria Function
1 TBv lt TBh RFI in H but not V
2 TBp(high frequency) - TBp(low frequency)lt -5 RFI in low frequency
3 A lt 0 or B gt 1 Test A and B in physical range
21
Comparison of MVIs and VWC

• Landsat based VWC values were averaged over 4
  AMSR-E EASE-GRID pixels within the Yanco area
• For each day with AMSR-E data, the MVIs of the 4
  pixels were compared to VWC

22
Results Comparison of MVIs and VWC

• N23

N23
23
Results Comparison of MVIs and VWC

• Each grid cell exhibits a good correlation
  between B and VWC, however, (3,4) are different
  than (1,2).
• Using the individual regressions, the SEE for VWC
  is 0.087 kg/m2.
• The different slopes are likely associated with
  landcover.

24
Summary

• Described a new Microwave Vegetation Index (MVI)
  that utilizes multifrequency observations
• The MVI conveys somewhat different information
  than NDVI. The two may be complementary.
• Evaluated the potential of the MVI in predicting
  vegetation water content in a case study.
• Further development and evaluation is ongoing
  global responses and additional VWC data sets.