Soil Moisture Content

The assessment of soil moisture content and surface roughness from remotely sensed data is of primary importance for improving agricultural techniques of conservation farming such as yield forecasts, scheduling irrigations, fertilization. SAR (Synthetic Aperture Radar) remotely sensed data may provide a powerful tool for indirectly retrieving these agricultural surface parameters over large areas with frequent coverage. Nevertheless, one major source of error in the quantitative estimate of such geophysical parameters is the presence of the speckle within the scene. To overcome this difficulty many speckle filtering techniques have been developed for reducing this multiplicative noise. However, up to now, few works analyze the performance of these filters on the retrieval of spatially and physically accurate information useful for the estimates of these soil properties. In this context, this paper outlines the sensitivity of several speckle filtering methods on the assessment of these two agricultural surface parameters. Results stressed that depending on the speckle filtering method used, significant deviations were obtained on the estimates of soil properties. INTRODUCTION Soil moisture content and surface roughness play a critical role in the hydrological processes. They control the distribution of rainfall into runoff, evapotranspiration, and infiltration, which must be considered in water and energy balance [1]. Thus these two soil parameters need to be measured consistently on a spatially distributed basis. Remotely sensed SAR (Synthetic Aperture Radar) data have the potential to provide spatial and multitemporal estimates of these surface parameters, depending upon the sensor configuration and field condition. Nevertheless, the strong radiometric variability of extended surface targets within the SAR scene make the quantitative estimate of soil properties difficult. This variability is due to the presence of a signal-dependant multiplicative noise (so-called speckle), and directly results from the coherence of microwave radiations which induce unpredictable interference phenomena within SAR cell resolution. Many techniques have been developed to attempt to reduce speckle within SAR images. The first category of filters, namely heuristic filters, does not consider the distribution of radar reflectivity within the scene (e.g., median filter) and the second, the adaptive filters, (e.g., Lee filter [2]) incorporate as A Priori knowledge statistical description of the scene and of the speckle. Both methods enhance radiometric resolution at the expense of spatial resolution. Recent more sophisticated filters (e.g., Gamma-MAP filter [3], Wavelet filter [4]) try to preserve pertinent details and spatial resolution keeping a strong speckle reduction in homogenous area. The Speckle filtering of SAR images is a primordial step in extracting the useful signal (i.e., the underlying scene radar reflectivity or backscattering coefficient, σ) to be inverted for the retrieval of physical properties of the ground target. Although these filtering approaches have been tested with image processing criteria, such as the preservation of edge gradient value and the smoothing degree of homogeneous areas, a main issue is to know how these SAR filters influence spatially and physically the signal useful to the extraction of surface parameters. In this paper, we attempted to reply to this question for a case study on an agricultural site in Normandy (France). This was conducted using RADARSAT-SGC time-series for which a set of filtering techniques (i.e., 'box' and 'median' filters, and more sophisticated filters using wavelet representation and simulated annealing technique) was applied. Results stressed the significant impact of SAR filtering technique in the assessment of soil moisture content and surface roughness. SPECKLE REDUCTION TECHNIQUES Speckle filtering methods can be separated into two categories: the non-adaptive approaches (box filter, median, etc.) and adaptive techniques. Speckle reduction can be achieved using a simple box filter which enhances radiometric resolution without any consideration on the target nature. Consequently, edges and other significant details are strongly degraded. The median filter allows to better preserve edge properties but with a possible bias in the radar reflectivity estimation. By using adaptive filters a compromise exists between the radiometric enhancement in the homogeneous areas and the preservation of the spatial resolution within the heterogeneous areas (i.e., the textural area or edges). Depending on the local heterogeneity degree of the target, pixels are weighted with a value ranging from the local mean to the raw intensity values. Filters differ from the local weight determination depending on the A-Priori hypothesis applied on the probability density function (pdf) of both speckle and radar reflectivity. Thus, filters can be distinguished by the estimation strategy used such as the Minimum Mean Square Error (MMSE) or the Maximum A Posteriori (MAP). As an example, whereas Lee’s filter [2] assumes Gaussian pdf and apply a MMSE criterion, Gamma-MAP filter [3] assumes gamma pdf and uses a local MAP estimation. Oliver proposes a global MAP estimation based on the Metropolis algorithm [5]. Integration of both optimal target and edge detection with an adaptive size of window strategy can improve filtering results [2]. Wavelet filters are based on a multiscale representation of the image where high frequencies (Wavelet coefficients) are denoised using a MAP criterion and gamma pdf assumptions [4]. On the contrary, in the case of an agricultural parcel, a non adaptive filter will be to take the parcel mean as input for the assessment of the soil parameters. DATA SET RADARSAT-SGC time-series (4 dates) were acquired over an experimental agricultural site in Normandy (Blosseville, France) under several incidence angles (standard beam S4, S5, S6 and S7, which correspond to an incidence angle of 37, 39, 43.5 and 47, respectively) during March 1998. These SAR data were over-sampled, reducing the 12.5 m nominal resolution to 25 m, with a number of looks around 4. Moreover, a SPOT-XS image from 1997 completed our remote sensing database. METHODOLOGY Speckle filtering methods Five speckle reduction methods were applied on all the calibrated SAR images using the following specifications: (1) box filter (with a window size of 11 x 11 pixels); (2) median filter (5 x 5); (3) GammaMAP filter (11 x 11); (4) simulated annealing (ACMAP [5]) using 50 iterations and (5) wavelet filter (3 levels of decomposition) [4]. Co-registration of SAR images Standard techniques were employed to extract control points in all images (30 points were found for both optic and SAR data) and, subsequently deduce bilinear polynomial transformations using standard imageprocessing techniques. Residuals were monitored in the process of selecting the polynomial degree: it was found that simple linear polynomials were sufficient to achieve residual (sub-pixel) levels and that these residuals were independent of the application of higherorder polynomials. Parcel sampling Samples (~ 300 relatively homogeneous parcels) were selected within the SPOT scene, thus allowing a better control of the sampling quality for the sensitivity study of filtering SAR data in the estimate of physical soil parameters. SAR backscattering models 2 SAR backscattering models were used for retrieving surface roughness and soil moisture content over each of all selected parcels. The first is a semi-empirical model from Dubois [6] which presents some built-in limitations (the imaginary part of the dielectric constant ε is not taken into account, no dependence on the surface correlation, cf. [7]). Angular dependence of the backscattering coefficient for HH polarization is given by : σo hh = 10 . cos θ sin θ . 10 0.028.ε.tan θ . (k.h) . sin θ . λ (1) where θ is the incidence angle, k the wave number, ε the real part of the dielectric constant, h the rms height of the surface (cm) and λ the wavelength (cm). The validity domain of this relationship corresponds to rms roughness (k.h) values within [0.3 ; 3] and incidence angles between 30° and 65°. The second is the analytical integral equation model (IEM) specially adapted to roughness values typical of agricultural soil, and convenient for a large range soil status conditions [8]. The SAR backscattering coefficient (σ) is expressed as: σ 0 pp= k 2 .| fpp| .exp(-4.K0). Σ n=1 +∝ (4.K0) n n! .W (2.k.sinθ,0) + k 2 .Re(f * pp.Fpp).exp(-3.K0). Σ n=1 +∝ (2.K0) n n! .W (2.k.sinθ,0) + k 8 .| Fpp| .exp(-2.K0). Σ n=1 +∝ (K0) n n! .W (2.k.sinθ,0) (2) with: fhh = 2.R⊥ cosθ fvv = 2.R// cosθ Fhh = γ.[4.R⊥ (1 1 εr ) . (1+R⊥) ], Fvv = γ . [(1 εr.cosθ μr.εr sinθ ) . (1-R//) 2 + (1 1 εr ) . (1+R//) ], and K0=(k.h) .cosθ γ = 2. sinθ cosθ . where pp stands for HH or VV polarization and R// and R⊥ are the Fresnel reflection coefficients dependent on dielectric constant for vertically and horizontally polarized waves. Re means the real part of the complex number. W(2.k.sinθ,0) characterizes the surface roughness spectrum [9], and is a function of the rms height h of the surface and of its correlation length l for a surface with an exponential-distributed roughness values. For each of all selected samples a least-square fit to (1) was applied on the 4 corresponding angular SAR backscattering coefficients for estimating ε and (k.h) values. The SAR multi-angular fit to (2) for retrieving (k.h), (k.l) and ε parameters relies on the optimized simplex method [10]. These 2 SAR multi-angular regression methods were performed on a pixel per pixel basis for each set of angular SAR data (i.e., σ intensity resulting from parcel averages and σ estimated by each of 5 filtering techniques). ε is converted to volumetric soil moisture m through empirical curves [11]. Samples with a too small number of pixels (n < 200) and for which the number of realistic estimates is lower than 75% were discarded from our analysis. A degree of heterogeneity level for each parcel (sample) was evaluated using the normalized standard deviation coefficient of the underlying radar reflectivity (CR) [5]. The statistical validity of the regression is evaluated using the following probability: