Soil moisture retrieval over periodic surfaces using PolSAR data

Transcription

1 Soil moisture retrieval over periodic surfaces using PolSAR data Sandrine DANIEL Sophie ALLAIN Laurent FERRO-FAMIL Eric POTTIER IETR Laboratory, UMR CNRS 6164, University of Rennes1, France

2 Contents Soil moisture retrieval over plowed fields: classical methods Time-Frequency analysis Polarimetric analysis Rough periodic surface scattering model Soil moisture retrieval

3 Test site Organic matter content Ikonos image L band Quad pol data set DLR E-SAR sensor March 2000, Alling, Germany RGB Herold M. et al., Acquisition and evaluation of field measurements from the Alling-SAR 2000 campaigns, 2001.

4 Test site Organic matter content Ikonos image Slightly wet area Mainly rough and flat fields Some are plowed RGB Herold M. et al., Acquisition and evaluation of field measurements from the Alling-SAR 2000 campaigns, 2001.

5 Classical retrieval methods Objective - Estimate soil moisture over all unvegetated agricultural fields FLAT and PLOWED fields Classical soil moisture retrieval schemes - Oh s method Co and Cross polar ratio analysis p σ = σ HH VV q σ = σ HV VV -H/α method Entropy and α angle analysis Polarimetric degree of randomness Nature of scattering mechanisms

6 Oh s s method Plowed fields p (db) q (db) Oh s model co- and cross-polar ratios Θ=45 ks: roughness mv: moisture content Oh Y., Quantitative Retrieval of Soil moisture Content and Surface Roughness From Multipolarized Radar Observations of Bare Soil Surfaces,2004.

7 Oh s s method Plowed fields p (db) q (db) Oh s model co- and cross-polar ratios Θ=45 Soil estimation: very smooth and very wet

8 Oh s s method Ground truth: plowed fields and slightly wet Plowed fields p (db) q (db) Oh s model co- and cross-polar ratios Θ=45 Soil estimation: very smooth and very wet

9 H/α method Plowed fields H α ( ) α ( ) Roughness Θ=45 Soil estimation: very smooth and very wet ε H Hajnsek I., Pottier E., Cloude S.R., Inversion of surface parameters from polarimetric SAR, 2003.

10 H/α method Plowed fields H α ( ) Soil estimation: very smooth and very wet Ground truth: plowed fields and slightly wet α ( ) Roughness Θ=45 ε H

11 Classical retrieval methods Ferro-Famil L et al., Nonstationary natural media analysis from polarimetric SAR data using a two-dimensional time-frequency decomposition approach, Erroneous estimates over some plowed fields Classical methods are not adapted Low values of entropy and α angle may be observed over anisotropic fields * Non-stationary scattering pattern investigation Time-frequency analysis

12 Time-Frequency analysis Principle of SAR One scene is observed under different azimuth look angles f az = 2 f c V c SAR sinφ o Doppler spectrum Antenna azimuth aperture + φ dmax f dmax + fdmax f f m m f d f M f M φ m φ d φ M φ M φm φ dmax Ferro-Famil L et al., Scene Characterization Using Subaperture Polarimetric SAR data, 2003.

13 Time-Frequency analysis Azimuth frequency domain FFT Full resolution Φ 1 Φ 2 Φ 3 Φ 4 Ferro-Famil L et al., Scene Characterization Using Subaperture Polarimetric SAR data, 2003.

14 Time-Frequency analysis Azimuth frequency domain FFT Full resolution Φ 1 Φ 2 Φ 3 Φ 4 Bragg resonance Ferro-Famil L et al., Scene Characterization Using Subaperture Polarimetric SAR data, 2003.

15 Time-Frequency analysis Ferro-Famil L et al., Nonstationary natural media analysis from polarimetric SAR data using a two-dimensional time-frequency decomposition approach, Azimuth frequency domain FFT Full resolution Φ 1 Φ 2 Φ 3 Φ 4 Nonstationary field detection Polarimetric representations statistics Over homogeneous areas n-look sampled matrix S (φ, ( 0 ) i ) ~ N C Σ i T (φ ) ~ W, i C ( n Σ ) i Maximum-likelihood detection Hypothesis: T φ T φ ) ( 1 ), K, ( R R ni T( φi ) i= = 1 n Λ ML ratio test: with T t t follow the same distribution, i.e. n t n i = R i= 1 and T t Σ 1 =... = Σ R R nit( φi ) i= = 1 n t

16 Time-Frequency analysis Azimuth frequency domain FFT Full resolution Φ 1 Φ 2 Φ 3 Φ 4 Nonstationary field detection Ferro-Famil L et al., Scene Characterization Using Subaperture Polarimetric SAR data, 2003.

17 Time-Frequency analysis Azimuth frequency domain FFT Full resolution Φ 1 Φ 2 Φ 3 Φ 4 Nonstationary field detection Polarimetric analysis Ferro-Famil L et al., Scene Characterization Using Subaperture Polarimetric SAR data, 2003.

18 Polarimetric Analysis Nonstationary fields Z 1 Z 2 Stationary fields Z 3 Z 4 Co-polar ratio low variations between sub-images (±1.5db) may be used for soil moisture retrieval

19 Polarimetric Analysis Nonstationary fields Z 1 Z 2 Stationary fields Z 3 Z 4 Cross-polar ratio strong variations between sub-images (±6db) lower depolarization in presence of Bragg resonance

20 Polarimetric Analysis Nonstationary fields Z 1 Z 2 Stationary fields Z 3 Z 4 Co and Cross-polar ratios Not adapted for soil moisture retrieval over nonstationary fields

21 Polarimetric Analysis Nonstationary fields Z 1 Z 2 Stationary fields Z 3 Z 4 Entropy / α strong variation between sub-images (±0.4 for H and ±10 for α) one main scattering mechanism in presence of Bragg resonance Not adapted for soil moisture retrieval over nonstationary fields New inversion parameters are needed: development of a new rough periodic surface scattering model

22 Rough periodic surface Surface characterization scattering Model f 2π x ( x, y) = Bcos + ξ ( x, y) P Agricultural field periodic component random component New rough periodic surface scattering model based on the Kirchhoff model with scalar approximation adapted to rough periodic surfaces Yueh H. A. et al., Scattering from randomly perturbed periodic and quasi-periodic surfaces, 1988

23 Rough periodic surface scattering Model Backscattering coefficients σ + pq = σ pq + σ c pq n σ pq s Coherent component Incoherent component Slopes components negligible Rough Surface Rough periodic Surface Bistatic coherent component Monostatic coherent component Monostatic Incoherent component No coherent part in monostatic

24 Rough periodic surface Incoherent backscattering coefficients HH polarization VV polarization scattering Model l = 100cm l = 50cm l = 10cm Correlation lengths Rough surface behavior σ VV > σ HH l influences the coefficient shapes which depend on the Floquet modes Floquet modes From Bragg resonance condition λ sinθ cosφo = n 2P Surface characteristics: P = 100cm B = 10cm rms height = 1cm ε = 6 F = 1.3GHz φ o = 0

25 Rough periodic surface scattering Model Bragg resonance conditions depends on incidence angle: θ P=0,6 m n=3 azimuth look angle: Φ 0 sinθ cosφ = n o λ 2P Floquet mode: n P=0,5 m to 1 m n=3 period: P

26 Rough periodic surface Incoherent backscattering coefficients HH polarization VV polarization scattering Model l = 100cm l = 50cm l = 10cm Correlation lengths Floquet modes From Bragg resonance condition λ sinθ cosφo = n 2P nπ θ = asin Pkcos( φ o ) Locations and amplitudes of the Floquet modes: 2 ( kdz. B) = f ( θ, φ, P, B k) Jn o, Surface characteristics: P = 100cm B = 10cm rms height = 1cm ε = 6 F = 1.3GHz φ o = 0 Chuang S.L et al., Scattering of waves from periodic surfaces, 1981.

27 Rough periodic surface Incoherent backscattering coefficients HH polarization VV polarization scattering Model l = 100cm l = 50cm l = 10cm Correlation lengths Floquet modes From Bragg resonance condition λ sinθ cosφo = n 2P nπ θ = asin Pkcos( φ o ) Polarimetric analysis Surface characteristics: P = 100cm B = 10cm rms height = 1cm ε = 6 F = 1.3GHz φ o = 0 Chuang S.L et al., Scattering of waves from periodic surfaces, 1981.

28 α 1 angle Analysis Nonstationary fields Z 1 Z 2 Stationary fields Z 3 Z 4 α 1 angle: low variation between sub-images (±2 ) depends on soil moisture and incidence angle May be used for soil moisture retrieval even over nonstationary fields

29 Soil moisture retrieval α 1 inversion method mean( α1 ) data surface α 1 = α corrected 1 adapted for each θ iem mean( α ) 1 iem ε estim ε gd meas ε estim ε gd meas Allain S. et al, Two novel surface model based inversion algorithms using multi-frequency polsar data, 2004.

30 Soil moisture retrieval α 1 inversion method mean( α1 ) data surface α 1 = α corrected 1 adapted for each θ iem mean( α ) ε estim ε gd meas mean error(ε)= 13% 1 iem ε estim ε gd meas Allain S. et al, Two novel surface model based inversion algorithms using multi-frequency polsar data, 2004.

31 Soil moisture retrieval Organic matter content Dielectric constant retrieval Very wet area 5 0

32 Soil moisture retrieval Organic matter content Dielectric constant retrieval Slightly wet area

33 Soil moisture retrieval Azimuthal look angle variations Dielectric constant retrieval

34 Conclusions Classical retrieval methods anomalous behaviors may appear over periodic surfaces Time-Frequency analysis identify nonstationary fields confirms the dependence on the azimuth look angle New rough periodic surface scattering model α 1 parameter: remains constant in presence of resonance phenomena highly sensitive to soil moisture Application over real SAR data acquired at L band

35 Outlook rg az az rg AGRISAR 2006 L band Quad pol data set DLR E-SAR sensor

36 Outlook rg az Bragg phenomenon az rg

37 Grazie