The measurement of Soil Moisture from ENVISAT ASAR data

1
The measurement of Soil Moisture from ENVISAT
ASAR data
User workshop- Vienna, June 13, 2005
2
Scattering from terrain
smooth soil permittivity (SMC)
Scattering pattern
Vegetated soil attenuation effect of vegetation
rough soil permittivity surface
roughness (SMC, HStD, correlation length)
3
Physical principle of SMC measurement
High contrast between microwave permittivity of
dry matter and water
4
Sensitivity to spatial variations of SMC
Experimental results
C-band ? 25o
Bare vegetated soils
5
Sensitivity to tremporal variations of SMC
Experimental results
C-band, ? 25o ,VV pol
Bare vegetated soils
Site 2 Les Alpilles (France)
Site 1 Montespertoli (Italy)
Correlation coefficient Montespertoli R2
0.98 Les Alpilles R2 0.57 Total R2 0.66
6
Development of algorithms for retrieval of bare
soil parameters

• Inversion algorithms
• Regression
• Iteration (Nelder)
• Bayesian
• Neural Network

7
Comparison of inversion algorithms Simulated data

• Training and test with simulated data (IEM
  noise)
• Data set of 3000 simulated values of so
  generated by multiplying outputs of IEM (at
  Envisat observation parameters) for a noise
  random variable with mean and variance derived
  from the experimental dataset.
• SMC values ranged between 5 and 35

Simulated backscattering
8
Comparison of inversion algorithms Simulated data
SMC

• Classification
• 5 classes of SMC

9
Retrieval test with experimental data
10
Evaluation of inversion algorithms
11
The Envisnow Soil moisture Algorithm
12
Validation of the algorithm
13
Validation of the algorithm Scrivia
3 ENVISAT ASAR images (HH pol., ? 23?) taken in
November 2003, April 2004 and June 2004.
14
Validation of the algorithm Scrivia
15
Validation of the algorithm Cordevole
Cherz plateau
16
Validation of the algorithm Cordevole
ANN1 Training with a subset of exp data ANN2
Archive data correction for vegetation
17
Validation of the algorithm Cordevole
18
SMC time variation (Cherz)
Resolution 150 m
19
Soil moisture summary

• 3- 5 levels of SMC can e detected between 10
  and 40
• Iteration (Nelder) is the most accurate but slow
• Bayes is the most stable but very slow
• Regression is the fastest but less accurate
• ANN gives the best compromise
• Problem availability of ASAR HH pol data