|
Titel |
Estimation of soil moisture using radar and optical images over Grassland areas |
VerfasserIn |
Mohammad El Hajj, Nicolas Baghdadi, Mehrez Zribi, Gilles Bellaud, Bruno Cheviron, Dominique Courault, Olivier Hagolle, Francois Charron |
Konferenz |
EGU General Assembly 2015
|
Medientyp |
Artikel
|
Sprache |
Englisch
|
Digitales Dokument |
PDF |
Erschienen |
In: GRA - Volume 17 (2015) |
Datensatznummer |
250102474
|
Publikation (Nr.) |
EGU/EGU2015-1794.pdf |
|
|
|
Zusammenfassung |
The purpose of this study was to develop an inversion approach to estimate soil moisture over
Grassland areas by coupling SAR and optical data. A time series of radar (TerraSAR-X and
COSMO-SkyMed) and optical (SPOT 4/5, LANDSAT 7/8) images were acquired over an
agriculture region in southeastern France. In most cases, the optical and radar images were
not separated by more than four days. Ground-truth measurements of volumetric soil
moisture and vegetation descriptors were performed simultaneously with image
acquisitions.
In this study, the semi-empirical water-cloud model (WCM) was used to model the total
backscattering coefficient in HH and HV polarizations as a function of soil moisture and
vegetation descriptor. WCM was fitted against in situ measurements to estimates WCM
parameters according to each vegetation descriptor and each radar polarization (HH and HV).
The parameterized water cloud model was then used to generate one synthetic dataset using
wide range of soil moisture and NDVI values. Next, a noise was applied to both simulated
radar responses and NDVI.
An inversion technique based on Multi-Layer Perceptron (MLP) neural networks (NN)
were used to invert the radar signal in order to estimate the soil moisture. Three inversion
configurations were defined using in addition to one vegetation descriptor: (1) HH
polarization, (2) HV polarization, and (3) both HH and HV polarizations. The neural
networks were trained and validated on the noisy synthetic dataset. Next, the inversion
approach was then conducted on real dataset comprised observed SAR data, soil moisture and
NDVI.
The best soil moisture estimates were obtained with the use of HH (in addition to one
vegetation descriptor). For NDVI (Normalized Difference vegetation index) lower than about
0.8 the RMSE (Root Mean Square Error) between measured and estimated soil moisture is
about 0.05 cm3/cm3. However, for NDVI greater than about 0.8 the RMSE on estimated soil
moisture is about 0.08 cm3/cm3. |
|
|
|
|
|