Soil roughness retrieval from TerraSar-X data using neural network and fractal method
Introduction
Surface roughness is a key parameter in describing the soil surface properties in many applications such as agricultural and hydrology. This parameter is an index of soil sensitivity to wind erosion in agricultural areas, and plays an important role in infiltration and storing water (Beckmann and Spizzichino, 1987). Runoff and the soil erosion are two important factors that cause the damage to the agricultural field. Runoff occurs when rainfall’s intensity is more than infiltration of water (Zobeck and Onstad, 1987, Brun et al., 1990, Boiffin et al., 1988, Le Bissonnais, 1990). Soil surface roughness plays an important role in water absorption, more infiltration and reducing runoff. Trapping water in rough surfaces is more than smooth surface (Le Bissonnais et al., 1998). Surface roughness is defined as variation and disturbances of soil surface that is caused by many factors such as soil texture, grain size, and some activity that done on the ground like plowing (Amoah et al., 2013). Surface roughness definition is based on a calculation of auto-correlation function, from which two parameters, the standard deviation of height (rms) and the correlation length are computed. Standard deviation is the variation of height in vertical and correlation length is the horizontal variation of height (Baghdadi et al., 2018). In more than 95% of the agricultural field, the standard deviation (rms) is between 0.25 cm for smooth soils to 4 cm for rough soils and a correlation length is between 2 cm and 20 cm (Baghdadi et al., 2011a, Baghdadi et al., 2011b, Baghdadi et al., 2011c).
There are several methods for estimating of soil roughness. These methods include needle and laser profilometer, shadow analysis, photogrammetric method and laser scanners (Römkens et al., 1986, Zribi et al., 2000, Darboux and Huang, 2003, García Moreno et al., 2008). These methods in a widespread area require a lot of time and effort and practically is impossible.
In recent decades, Efforts have been made by many researchers for estimating of surface parameters (moisture and roughness) using Synthetic Aperture Radar (SAR) images (Schmugge, 1978, Jackson et al., 1981, Dobson and Ulaby, 1986, Fung, 1994).
The backscattering coefficient of the radar signal depends on the surface parameters (roughness and dielectric constant) and sensor parameters (wavelength, polarization, and incidence angle). Different physical and statistical models have been introduced to define a relationship between the backscattering coefficient and the sensor and target parameters. Surface parameters are estimated using the inversion of these models (Baghdadi et al., 2016, Hajj et al., 2017). Unlike statistical models, physical models for soil moisture and roughness estimates do not require site-specific calibration and could always be used to simulate the backscattering coefficients from the radar configuration (frequency, polarization, and incidence angle) and soil parameters (Baghdadi et al., 2002). The Integral Equation Model (IEM) is the most well-known physical model used to simulate the backscattering coefficients (Fung, 1994). The IEM developed by Fung (Fung, 1994) has a significant difference between simulated and actual SAR data, which results is an inaccurate estimation of roughness and soil moisture content (Gorrab et al., 2015, Panciera et al., 2014). To improve the accuracy of simulating backscattering values, Baghdadi et al. developed the semi-empirical calibration for the IEM at the X-band (Baghdadi et al., 2011a), C-band (Baghdadi et al., 2011b), and L-band (Baghdadi et al., 2015). Since the measurement of correlation length is very difficult, in the calibrated model, it is defined as a function of rms and incidence angle. Physical models have shown that the backscattering coefficient is more sensitive to surface roughness than soil moisture, at the bare soil, especially at high incidence angles (Baghdadi et al., 2016, Ulaby et al., 1982).
The inversion of IEM has been widely used for roughness and soil moisture estimates from SAR data at X and C bands (Baghdadi et al., 2011a, Baghdadi et al., 2011b). The Neural network is one of the most popular methods for inversion the IEM. Satalino et al. (2002) presented an inversion algorithm for IEM based on the neural network to estimate roughness and moisture using the ERS-SAR data in the VV-23° polarization.
Baghdadi et al. used polarimetric data at band C to estimate soil moisture and roughness. At first, they simulated data for (0.3 cm < rms < 3.6 cm) and moisture between 5 and 42 vol% using the IEM. They trained the neural network using simulated data. The neural network was evaluated based on real data the accuracy of the moisture content was about 7% (Baghdadi et al., 2012a, Baghdadi et al., 2012b).
Baghdadi et al. used the neural network to inverse the IEM. They simulated a roughness range between 0.5 and 3.8 cm and moisture between 2 and 40 vol% using the IEM. The model was evaluated based on sentinel-1 data at C band in VV Polarization. The accuracy obtained for roughness was about 0.8 cm and moisture was about 6 vol%. They concluded that the C band in VV polarization did not yield reliable results for soil roughness estimation. In another study, Baghdadi et al developed a neural network model for estimating the moisture and roughness from the RadarSat-2 data. Their results showed that the roughness accuracy was about 0.5 cm. The accuracy of roughness estimation for data with rms lower than 2 cm was better than rms higher than 2 cm. In roughness higher than 2 cm, the underestimation and roughness lower 2 cm overestimation was observed (Baghdadi et al., 2012a, Baghdadi et al., 2012b).
In order to soil roughness estimates, choosing the appropriate band and polarization is very important. Gorrab et al. (2015) argue that the HH polarization and high incidence angles are proper for classification of soil roughness.
The purpose of this study is to evaluate the potential of TerraSar-X data in HH polarization for roughness estimation over bare soils. The calibrated IEM is used to simulate data at X band in HH polarization and inversion method is carried out by the neural network. The validation of the neural network is based on simulated and real data. Section 2, describes methods and materials, including the neural networks and datasets (simulated and real). The method of calculating in situ roughness by the fractal method is also discussed. The performance of the neural network and accuracy of results are explained In Section 3. Section 4, presents the conclusions.
Section snippets
Methodology
In this paper, a method for estimating of soil roughness from Synthetic Aperture Radar data (SAR), TerraSar-X, in HH polarization based on the multilayer perceptron neural network is presented. Baghdadi et al. concluded that the accuracy of the estimation of bare soil roughness is higher in HH polarization and high incidence angles (>45°) (Gorrab et al., 2015, Baghdadi et al., 2008). TerraSar-X data in this study is a 47° incidence angle and HH polarization.
Backscattering of radar’s signal
Results and discussion
The main objective of this study is to estimate the surface roughness based on synthetic and real data at X-band in HH polarization. The neural network is used to inversion the surface parameters (roughness and moisture). In this study the Levenberg-Marquardt algorithm (Lourakis, 2005) is used for training the neural networks. The optimum number of hidden layer is selected based on maximum performance by applying minimization the mean square error. The number of hidden layer is two for the
Conclusions
The purpose of this study was to estimate the roughness of bare soil surface using X-band data in HH polarization. Neural network method has been used in this study as an inversion method. Several neural networks were trained using the simulated data that generated from calibrated IEM. Estimation of roughness was based on two steps. In the first step, the neural network was trained with backscattering coefficient and incidence angle and the moisture was output. In the next step, based on a
Acknowledgements
“TSX Data” were provided by the European Space Agency, Project Proposal id 34722, © DLR, distribution Airbus DS Geo GmbH, submitted in relation to the project ALPSMOTION coordinated by Eurac Research-Institute for Earth Observation and funded by the Autonomous province of Bolzano, Alto Adige, “Ripartizione Diritto allo Studio, Università e Ricerca Scientifica”. M. Maleki passed six months internship in Eurac Research-Institute for Earth Observation. The authors would like to thank institute of
References (38)
Analysis of TerraSAR-X data and their sensitivity to soil surface parameters over bare agricultural fields
Remote Sens. Environ.
(2008)- et al.
Predicting rms surface roughness using fractal dimension and psd parameters
J. Terramech.
(2011) Shadow analysis: a method for measuring soil surface roughness
Geoderma
(2008)Crusting, runoff and sheet erosion on silty loamy soils at various scales and upscaling from m2 to small catchments
Soil Tillage Res.
(1998)- et al.
An automated non-contact surface profile meter
Soil Tillage Res.
(1986) - et al.
Description of surface roughness as an approximate self-affine random structure
Surf. Sci.
(1995) - et al.
Tillage and rainfall effects on random roughness: a review
Soil Tillage Res.
(1987) Characterization of the soil structure and microwave backscattering based on numerical three-dimensional surface representation: analysis with a fractional Brownian model
Remote Sens. Environ.
(2000)- et al.
Quantifying watershed surface depression storage: determination and application in a hydrologic model
Hydrol. Process
(2013) Potential of Sentinel-1 images for estimating the soil roughness over bare agricultural soil
Water
(2018)
Comparison between backscattered TerraSAR signals and simulations from the radar backscattering models IEM, Oh, and Dubois
IEEE Geosci. Remote Sens. Lett.
Estimating surface soil moisture from TerraSAR-X data over two small catchments in the Sahelian Part ofWestern Niger
Remote Sens.
A potential use for the C-band polarimetric SAR parameters to characterize the soil surface over bare agriculture fields
IEEE Trans. Geosci. Remote Sens.
Semi-empirical calibration of the integral equation model for co-polarized L-band backscattering
Remote Sens.
A new empirical model for radar scattering from bare soil surfaces
Remote Sens.
Semiempirical calibration of the integral equation model for SAR data in C-band and cross polarization using radar images and field measurements
IEEE Geosci. Remote Sens. Lett.
Estimation of soil parameters over bare agriculture areas from C-band polarimetric SAR data using neural networks
Hydrol. Earth Syst. Sci.
Potential of ERS and Radarsat data for surface roughness monitoring over bare agricultural fields: application to catchments in Northern France
Int. J. Remote Sens.
The Scattering of Electromagnetic Waves from Rough Surfaces
Cited by (7)
Low-thrust spacecraft trajectory optimization via a DNN-based method
2020, Advances in Space ResearchCitation Excerpt :Over the past few years, the applications of artificial intelligence in aerospace engineering have been of considerable interest to the space community (Izzo et al., 2018). It is well known that deep neural networks (DNNs) (Schmidhuber, 2015; Lan et al., 2020; Maleki et al., 2019; Cavus et al., 2020; Huang et al., 2019) perform well in tasks such as image recognition (Yang et al., 2017) and speech processing (Collobert et al., 2011). In this study, DNNs are expected to help the trajectory optimization and flight control of continuous low-thrust spacecraft.
Coupling Model-Driven and Data-Driven Methods for Estimating Soil Moisture Over Bare Surfaces With Sentinel-1A Dual-Polarized Data
2023, IEEE Journal of Selected Topics in Applied Earth Observations and Remote SensingReview of remote sensing application in soil degradation
2021, Journal of Agricultural Resources and EnvironmentDigital terrain model generalization for multiscale use
2020, International Review of Civil EngineeringInvestigation of potential dust sources using sentinel-1 and neural network: A case study from bandar-e emam-omidiye
2019, International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences - ISPRS Archives