1. Introduction
The ocean is playing an increasingly important role in politics, economy, culture, and other areas worldwide. Accordingly, the studies on electromagnetic (EM) scattering simulations of marine environments are of considerable value, with applications in marine remote sensing, as well as civil and military fields [
1,
2,
3,
4,
5,
6]. Due to the constraints imposed by physical and financial resources, experimental data are limited in terms of the variation in sea conditions, radar parameters and viewing geometry, which has restricted the development of modern radar techniques, such as high-resolution imaging and target detections. As such, sea scattering simulation based on the EM scattering model has become a viable alternative, due to its low cost and easy implementation.
However, EM scattering faces challenges when applied to rough surfaces, so a number of approaches have been introduced in previous studies to address this. These approaches can be divided into two categories: numerical methods and theoretically approximate approaches. Numerical methods, such as method of moment (MoM) [
7] and finite-difference time-domain method (FDTD) [
8], usually provide the most accurate results. However, the acquisition of these accurate results comes at the cost of considerable computation requirements, which constrain the ability of numerical methods to be applied to electrically small objects or one-dimensional problems. As a comparison, theoretically approximate approaches [
9,
10], like the Kirchhoff approximation (KA) method, small perturbation method (SPM), two-scale model (TSM), and small slope approximation (SSA) [
11], usually accomplish the calculation with their corresponding assumptions, with which the original problems can be greatly simplified, resulting in a considerable decrease in computation demand. Armed with these approaches, EM scattering of two-dimensional (2D) surfaces and some other practical problems can be solved efficiently with precise or acceptable results. Comparisons of the effectiveness and validity of these models have been discussed in detail [
12,
13]. Among the theoretically approximate approaches, SSA is an effective candidate that bridges the gap between the KA and SPM models, and can be degenerated when appropriate. Moreover, as SSA avoids the empirically scale partition in the TSM model, is more accurate, and has larger application scope, it has been widely applied to the practical rough surface scattering problems, and especially in sea surface scattering studies, where SSA has been proven to be reliable compared with the other methods and experiment data [
14,
15,
16,
17]. Additionally, the SSA method has been developed into a deterministic model, which is favorable for some other application areas, such as composite scattering problems for rough surfaces with targets [
18], synthetic aperture radar (SAR) image simulation of maritime scenes, and dynamic scattering of the sea surface [
19,
20].
However, for SSA in the EM scattering calculation, the spatial sample interval for the sea surface should be around or even smaller than one-eighth of the incident wavelength to ensure accuracy of the integral operation in the SSA formula. On the other hand, the size of the simulated sea surface must be larger than its dominant wavelength to reflect the time-variation and modulation effects of the maritime scenes. This directly results in the huge number of sample points and the tremendous amount of computation required for corresponding simulation applications. Particularly, for the X-band or higher microwave bands, the spatial interval can reach the millimeter level, and the total number of sample points will be in the billions or even more [
20]. To solve this problem, some assumptions or techniques have been proposed to simplify and implement the numerical simulation. Li et al [
21] replaced the capillary wave with a specific sinusoidal wave according to the Bragg scattering mechanism, to simplify the SSA. Jiang et al. [
22,
23] applied the spectrum decomposition method to split the sea profile into two kinds of scales and filled the large scales with one small scale sample, which provided a solution for the storage issues encountered with high microwave band and large surface area. Wang and Xu [
24] decomposed the spectrum of a 2D sea surface into multiple blocks and combined computer memory and external storage to accomplish the Doppler simulation of a 2D sea surface up to Ku-band. Regardless, the above approaches either lose the ability to accurately describe the sea surface or do not improve the computation efficiency during the numerical simulation.
In this paper, a novel realization approach for SSA is proposed to significantly decrease the computation amount and computer memory requirements for sea surface scattering simulation. The architecture of the proposed method is displayed in
Figure 1. First, the sea surface is decomposed into two scales, and each scale has its own spatial sample interval. Then, the inclination state of the large-scale sea surface is determined under a specific wind speed. After that, scattering calculations for a typical surface cell with finely sampled structure are carried out and saved in all possible inclination situations. Finally, the scattering results of all the cells from a concrete sea surface are extracted from the saved cell scattering dataset, and the total field of the whole sea surface is obtained by summing the individual cell results.
The remainder of this paper is organized as follows:
Section 1 briefly presents the SSA theory for sea surface scattering under the tapered incident wave;
Section 2 outlines the detailed process for the novel implementation of SSA, including the sea surface decomposition, the scattering calculation of a typical surface cell in all possible inclination situations, and the interpolation and extraction operations.
Section 3 provides the numerical simulation of backscattering scattering, bistatic scattering, and comparisons to display the feasibility and accuracy of the proposed approach. Finally,
Section 4 concludes this paper.
3. Numerical Simulations and Discussion
In this section, some numerical simulations are illustrated for sea surface scattering by applying the proposed method, including back and bi-static scattering coefficients varying with incident or scattering angles. Moreover, in order to demonstrate the validity of the proposed method it is compared to statistical SSA.
First, the backscattering simulation of sea surface was carried out at the Ku-band (14.0 GHz) with a wind speed of 5 m/s, which was chosen for the comparison with the statistical SSA results reported previously [
27]. In this simulation, the large mesh grids were set to 0.1 × 0.1 m
2, and the small grids were 0.001 × 0.001 m
2 to accurately estimate the Bragg scattering contribution. During the scattering field dataset construction, the typical cell included 100 large grid cells of the all possible inclinations for EM scattering calculation. In the implementation process, there were 1024 sampling points in both x and y directions, and the simulation area of the sea surface was 102.4 × 102.4 m
2. The Elfouhaily model [
28] was adopted for the sea roughness spectrum, which was developed based on available field and wave-tank measurements. The Elfouhaily model is supported by strong physical arguments, contrary to other spectra that are mostly empirical. The relative dielectric constant of the sea water was calculated using the Klein dielectric constant model [
29] at 20 °C and 35 ‰ salinity. With a 5 m/s wind speed, the variation ranges of the normal vector inclination angles,
and
versus the
x- and
y-axis, were 61–117 and 67–113°, respectively. After the dataset of the typical cell of the all possible inclinations was established for EM scattering calculation, interpolation was performed to obtain a much more delicate variation description with the interval of 0.1°. Finally, the scattering results of the entire large sea surface were derived through extraction and synthesis.
Figure 8 compares the backscattering coefficient results between the statistical SSA (theoretical results of SSA) and the proposed approach for both VV and HH polarizations with the incident azimuth of
(upwind direction). The results of the proposed approach were averaged over 30 samples. As shown in
Figure 8, good agreement exists between the corresponding results for all the incident angles, which suggests a good backscattering coefficient estimation of the sea surface at Ku-band with a wind speed of 5 m/s when using the proposed approach.
For a further verification, the backscattering results obtained using the two methods with a 15 m/s wind speed were compared (
Figure 9). In this simulation, the other parameters and the operations remained the same with the above comparison, except for the wind speed-related values, such as the variation in the ranges of the normal vector inclination angles. With a 15 m/s wind speed, range in values changed to 56–124 a 61–121° for
and
, respectively. Again, the results of the proposed approach were averaged over 30 samples. From the comparison, the results obtained from both methods matched well for all incident angles and both VV and HH polarizations. This good agreement indicates the good performance of the proposed method for the EM scattering estimation of the sea surface under different sea conditions.
The above results only involve the backscattering calculations. The following simulations expand the sea surface scattering estimation to bistatic scattering to further validate the proposed method. First, a general forward-backward configuration of the bistatic scattering was considered, where the
z-axis, the incident wave vectors, and the scattered wave vectors were in the same plane. The bistatic scattering simulation of sea surface was carried out at the Ku-band (14.0 GHz) with a wind speed of 5 m/s. Parameters such as sea spectrum, dielectric constant, and sampling intervals, were kept the same. The other parameters were fixed as follows: the transmitter incident and azimuth angles were 50° and 0°, respectively; the azimuth relative to wind direction was equal to 0 (upwind case); the receiver azimuth was set to180°; and the receiver incident angle
varied from −90° to 90°.
Figure 10 compares the forward-backward bistatic scattering results between the proposed method and the statistical SSA, which is derived from Awada et al. [
12]. As observed in the figure, the maximum energy was received around the specular direction
, which is a logical result because this is the true specular direction as given by Snell’s law. This maximum decreased when the wind speed increased. On the other hand, good agreement was observed between the bistatic scattering results from both methods for all incident angles and both HH and VV polarizations. This agreement suggests that the proposed approach has good estimation ability for sea surface bistatic scattering at Ku-band at a wind speed of 5 m/s.
To further illustrate the performance of the proposed approach for bistatic scattering simulation, another bistatic scattering simulation configuration was examined. In this simulation, parameters were the same as for the above example, except that the wind speed was changed to 15 m/s, the transmitter incident and azimuth angles were 40° and 0°, respectively, the receiver incident was set to 40°, and the receiver azimuth angle
varied from 0° to 180°.
Figure 11 compares the average NRCS between statistical SSA and the proposed approach. From the comparison, the results obtained with both methods were similar for all azimuth angles and both VV and HH polarizations. This good performance again shows the effectiveness of the proposed approach in bistatic sea surface scattering calculation for different scattering angles and sea conditions.
As wind direction is also an important factor in sea surface scattering, investigating the performance of the proposed method under different wind directions was necessary. Given the relative geometry relationship between the incident azimuth and wind direction, the variation in backscattering coefficients as a function of azimuth angle of incident wave was calculated.
Figure 12 compares the backscattering results as a function of azimuth angle of incident wave between the statistical SSA and the proposed method, where the incident wave was 14.0 GHz, the wind speed was 5 m/s, and the incident pitch angle was 60°. From the comparison, we observed that backscattering coefficients change with wind directions in a range of about 4 dB, and the results obtained with both methods were well-matched.
Comparisons of the cross-polarization bistatic scattering coefficients between the statistical SSA (data derived from Awada et al. [
12]) and the proposed method were also completed (
Figure 13). These comparisons were carried out in two different cases: (1) the transmitter incident and azimuth angles were 40° and 0°, respectively; the receiver azimuth was set to 45°; and the receiver incident angle
varied from 0° to 90°; and (2) the transmitter incident and azimuth angles were 40° and 0°, respectively; the receiver incident angle was set to 40°; and the receiver azimuth
varied from 0° to180°. The cross-polarization bistatic scattering coefficients derived from the two approaches were also well-aligned. Moreover, in
Figure 13b, as
and
had the same value, the curves for VH and HV polarizations coincide due to the theoretical reason of SSA.
Through the above scattering result comparisons, we demonstrated that the proposed novel realization of SSA has the same accuracy for sea surface scattering for backscattering, bistatic scattering, and both polarizations. In addition to the accuracy attained, the computation efficiency of the scattering calculation for a large sea surface was significantly improved due to the combination of the dataset and field synthesis. In the above simulations, a total of 29 × 24 and 34 × 30 inclination situations were used for the 5 m/s and 15 m/s wind speeds, respectively, with an angle sampling interval of . For each situation a 1 × 1 m2 square area was calculated by the SSA. Accordingly, the total computation amount was equivalent to the EM scattering computation amount for the sea surface with an area of 29 × 24 m2 and 34 × 30 m2, respectively. For a 100 × 100 m2 area sea surface scattering simulation, the computation amount decreased to 7–10% of the traditional SSA realization. Importantly, this total computation amount was almost independent of the scale of the maritime scene to be simulated, and the performance of the efficiency improvement increased with the scale of the simulated sea surface. On the other hand, as the proposed approach uses the cooperation between large and small mesh grids, the one-eighth of wavelength sampling is no longer required and the computation memory is then greatly reduced, especially for high microwave bands and large sea surfaces. These improvements give the proposed method an outstanding advantage in large sea surface EM scattering simulation applications. Notably, the proposed method in this paper is only an efficiency- and memory-related improvement of the traditional numerical simulation of SSA. That is to say, this novel realization has the same accuracy and methodology limits as first-order SSA. The related simulations using SSA have drawbacks when tackling sea surface scattering under grazing angles and high sea states. In these relatively extreme cases, some extra effects should be considered, such as spiky and breaking waves, multiple scattering and the sheltering effect. These are all challenging problems to be tackled in the following work.
4. Conclusions
In this paper, a novel approach for SSA is proposed to significantly decrease the computation and computer memory requirements for sea surface scattering simulation. For this realization, the sea surface was decomposed into two scales, and each scale had its own spatial sample interval. Then, the scattering of the meshed facets of the cell surface was calculated under all the possible inclinations and a corresponding database was established. Once the dataset of the typical cell for EM scattering calculation of the all possible inclinations was established, the scattering results of the entire large sea surface were derived through extraction and synthesis operations. In the numerical simulations, backscattering and bistatic scattering of the sea surface were simulated for different wind speeds, incident and scattering angles, and polarizations. From the comparisons with statistical SSA results, we demonstrated that the proposed approach possesses the same accuracy for relative sea surface scattering applications. In addition to the accuracy, the computation efficiency of the scattering calculation for a large sea surface was significantly improved, due to the combination of a database and field synthesis. Moreover, one-eighth of the wavelength sampling is no longer needed and the computation memory is then greatly reduced. With these improvements, this proposed approach has prominent advantages in sea surface EM scattering simulation applications, especially for high microwave bands and large sea surfaces. Unlike statistical SSA, which is an analytical method and can only be applied to the prediction of mean scattering coefficients from sea surfaces, the proposed method is an improved Monte-Carlo simulation of sea surface scattering, which includes sea surface generation and the EM scattering calculation processes from every surface section. This model can not only be used for mean scattering coefficients prediction, but can also be applied to relative simulations about the specific sea surface profile, such as composite scattering problems of sea surfaces with targets, synthetic aperture radar (SAR) image simulation of maritime scenes, and the dynamic scattering characteristics of the sea surface.