† Corresponding author. E-mail:
Project supported by the National Key Research and Development Program of China (Grant No. 2016YFC1401007), the Global Change Research Program of China (Grant No. 2015CB953901), the National Natural Science Foundation of China (Grant No. 41776181), and the Postgraduate Research & Practice Innovation Program of Jiangsu Province, China (Grant No. KYCX18 1012).
In the global climate system, the polar regions are sensitive indicators of climate change, in which sea ice plays an important role. Satellite remote sensing is a significant tool for monitoring sea ice. The use of synthetic aperture radar (SAR) images to distinguish sea ice from sea water is one of the current research hotspots in this topic. To distinguish sea ice from the open sea, the polarization ratio characteristics of sea ice and sea water are studied for L-band and C-band radars, based on an electromagnetic scattering model of sea ice derived from the integral equation method (IEM) and the radiative transfer (RT) model. Numerical experiments are carried out based on the model and the results are given as follows. For L-band, the polarization ratio for sea water depends only on the incident angle, while the polarization ratio for sea ice is related to the incident angle and the ice thickness. For C-band, the sea water polarization ratio is influenced by the incident angle and the root mean square (RMS) height of the sea surface. For C-band, for small to medium incident angles, the polarization ratio for bare sea ice is mainly determined by the incident angle and ice thickness. When the incident angle increases, the RMS height will also affect the polarization ratio for bare sea ice. If snow covers the sea ice, then the polarization ratio for sea ice decreases and is affected by the RMS height of snow surface, snow thickness, volume fraction and the radius of scatterers. The results show that the sea ice and the open sea can be distinguished by using either L-band or C-band radar according to their polarization ratio difference. However, the ability of L-band to make this differentiation is higher than that of C-band.
Sea ice covers about 7% of the earth’s surface and 12% of the world’s oceans. Much of the world’s sea ice is enclosed within the polar ice packs. Although sea ice is mainly distributed in the cold regions far from human settlements, sea ice has attracted close attention because it is an important part of the climate system. Changes in sea ice have significant influences on weather and climate through complex feedback processes, and thus have become an important indicator of global climate change. Therefore, the use of space-based radar to detect sea ice and change in sea ice in polar region is of great scientific significance and application value.
Synthetic aperture radar (SAR) is an active earth observation system that can work day and night in almost all-weather, without limitation due to clouds, but has limited capability to penetrate the surface beneath. Since its initial launch on Seasat in 1978, SAR has been used for earth observation, especially ocean observations, including sea surface wind field retrieval,[1] oil spill monitoring,[2,3] target identification,[4] sea ice monitoring,[5] ocean current inversion,[6–8] and other fields. Among them, the sea ice monitoring by remote sensing has attracted increasing attention, and focuses mainly on the identification and classification of sea ice. At present, image texture analysis,[9–12] neural network methods,[13,14] support vector machine methodologies[15,16] are mainly used. These methods usually rely on the remote sensing imagery itself; few of them identify and classify sea ice based on the electromagnetic scattering mechanism.
In this work, our study is based on the electromagnetic scattering model for remote sensing parameter retrieval and the scattering characteristics of sea ice. The traditional random rough surface scattering models mainly include the small disturbance model (SPM) and the Kirchhoff model (KA).[17–20] The SPM is suitable for slightly rough surfaces, while the KA is mainly used for rough surfaces with large surface curvature. Based on these models, a series of models has been developed, including small slope approximation (SSA)[21–24] and integral equation model (IEM).[25–27] For slightly rough surfaces, the IEM can be simplified into SPM; under the Kirchhoff approximation, the IEM is equivalent to the KA. Compared with other scattering models, IEM has high precision and wide application range, and has received much attention.
According to the IEM and radiative transfer (RT) model, some researchers have studied the electromagnetic scattering properties of sea ice. Wakabayashi et al.[28] studied the polarimetric characteristics of sea ice in the Sea of Okhotsk observed by airborne L-band SAR. They found that the backscattering coefficients, and particularly the vertical (VV) to horizontal (HH) backscattering ratio, are highly correlated with ice thickness. A simple simulation using IEM was conducted to study the relations among these variables, by using the physical parameters of typical sea ice. Liu et al.[29] developed a sea-ice microwave scattering mechanism for thin sea ice with slight roughness in the Bohai Sea as observed in the winter of 2012 according to the backscattering coefficients, which were measured under the condition of the three bands (L, C, and X), two polarizations (HH and VV), and incident angles of 20°–60°, by using a ground-based scatterometer. Theoretical results were obtained based on IEM and RT models and compared with the measurements. Nakamura et al.[30] carried out sea ice thickness retrieval in the Sea of Okhotsk by using dual-polarization SAR data. They developed an algorithm for retrieving the ice thickness, which is based on the backscattering ratio and IEM. In Syahali and Ewe’s work,[31] multiple-surface scattering, based on IEM that calculates surface scattering and additional second-order surface-volume scattering, was added into the model (based on prior studies) for improvement in the backscattering calculation.
The SAR-based sea ice classification and identification are very important in the research of ocean and polar remote sensing. Traditional classification methods concentrate on the SAR images themselves, ignoring the physical mechanism. In our work, we will investigate the polarization ratio characteristics of sea water and sea ice (with or without snow-cover) based on the electromagnetic scattering mechanism. The results will demonstrate that the difference between the polarization ratios of sea water and sea ice can be used to discriminate sea ice from sea water.
In the second section of this paper, an electromagnetic scattering model for a random rough surface is derived from IEM and RT theory, based on a geometric electromagnetic scattering model for a layered medium. In the third section, based on the above model, the polarization ratios for sea ice and water under different conditions are numerically simulated. The polarization ratio characteristics of sea water and sea ice (bare and snow-covered) for C-band and L-band are studied. The last section draws the conclusions from the present study.
The physical model for sea ice includes both internal structure (such as salinity, brine and bubbles, including their size, distribution, and concentration) and external characteristics (such as surface roughness, thickness, and related characteristics) that influence the microwave scattering characteristics of sea ice as illustrated in Fig.
The sea ice electromagnetic scattering includes several components: (i) top surface scattering from the air–ice interface, (ii) volume scattering from the scatterers in the sea ice, (iii) bottom surface scattering from the ice–water interface, and (iv) scattering from the interaction of the surfaces and the scatterers as shown in Fig.
Here, σts is derived from IEM and has the following form:[27]
Top surface scattering can be expressed as
The volume scattering term is derived from the RT model:[27]
The interaction term is also derived from the RT model,[27]
The polarization ratio of sea ice is defined as follows:
The ability of microwave radar to penetrate sea water is weak, and the penetration depths at C-band and L-band are about 0.4 cm and 3 cm, respectively.[29] Therefore, microwave electromagnetic scattering mainly acts on the sea surface, generating surface scattering, which can be calculated by the integral equation method (IEM). The electromagnetic scattering of seawater is mainly related to incident angle, electromagnetic wave frequency and wind speed. The wind speed affects the sea surface roughness (characterized by the RMS height and the correlation length), which in turn affects the radar backscattering cross-section.
For L-band and C-band radars, the variations of the polarization ratio of seawater with the RMS height of the sea surface are shown in Figs.
In this subsection, the polarization ratios of bare sea ice under different physical conditions are simulated. Simulation parameters used in the model are given in Table
As shown in Fig.
The variations in bare sea ice polarization ratio with ice thickness at three typical incident angles are shown in Figs.
The scatterers in the sea ice (mainly brines and bubbles) are the source of the volume scattering and bottom-volume scattering interaction. Since the interaction is weak, only ‘single-bounce’ interaction is considered. In the experiment, we only consider brines. Figures
Snow may have a significant influence on the polarization ratio characteristics of sea ice. Similarly, the total backscattering now consists of the top surface scattering from the air-snow interface, volume scattering in the snow, bottom surface scattering from the snow-ice interface and bottom-volume scattering interaction. To study the effect of snow on the sea ice polarization ratio, numerical experiments were carried out based on the above scattering model. The parameters set in the model are shown in Table
As shown in Fig.
The variations of sea ice polarization ratio with snow thickness at three typical incident angles are shown in Fig.
As shown in Fig.
For the L-Band, according to this analysis, we find that the seawater polarization ratio only depends on incident angle, while the sea ice polarization ratio is mainly affected by incident angle and ice thickness. The influence of other factors on it is negligible. Figure
For the C-band, the polarization ratio for radar scattering on bare sea ice is mainly related to the incident angle, RMS height and ice thickness, and decreases with the ice thickness. When the ice thickness reaches 10 cm, the sea ice polarization ratio no longer changes because the ice thickness is generally more than 10 cm in polar areas, and the effect of ice thickness is negligible. Therefore, it is only necessary to compare the sea ice polarization ratios under different incident angles and RMS heights. As shown in Fig.
The influence of polar sea ice on the climate has increasingly received attention. The wide-area monitoring of sea ice in the polar region has become the key to studying sea ice. Although the relevant research on sea ice monitoring by SAR has made some progress in recent years, there is still a technical bottleneck in distinguishing the automatic images of sea ice and seawater. Therefore, it is necessary to study the electromagnetic scattering mechanism for each of sea ice and sea water. Based on the electromagnetic scattering model derived from IEM and RT, the polarization ratios of seawater, bare sea ice and snow-covered sea ice are numerically simulated under different physical conditions. Their polarization ratio characteristics are analyzed.
Our numerical results indicate that the L-band seawater polarization ratio is only related to the incident angle and increases with incident angle increasing. The L-band sea ice polarization ratio is mainly affected by the incident angle and ice thickness, and decreases as ice thickness increases. After the ice thickness reaches a certain critical value, it no longer changes. This is mainly related to the L-band penetration depth for sea ice. Moreover, the sea ice penetration depth at L-band is larger than that of C-band. At C-band, the seawater polarization ratio is affected by incident angle and RMS height of sea surface, and increases with these two variables increasing; the bare sea ice polarization ratio is mainly related to incident angle, RMS height and ice thickness. When the incident angle exceeds 45°, the effect of RMS height on bare sea ice polarization ratio is negligible. When there is snow on the sea ice, the sea ice polarization ratio decreases, and the L-band sea ice polarization ratio is only related to the incident angle. By comparison, the C-band sea ice polarization ratio is affected by incident angle, RMS height of snow surface, snow thickness, volume fraction and radius of scatterers.
For the L-band radar, the difference between the polarization ratios of sea water and sea ice is obvious and increases with incident angle increasing. For the C-band radar, when the incident angle exceeds 40°, the difference between the polarization ratios of seawater and sea ice is also obvious and increases with incident angle increasing. Therefore, the sea ice and sea water can be distinguished using this property, which means that the recognition and classification algorithm for sea ice and open water does not rely on the remote sensing image itself. This provides theoretical support and technical means for further exploring the sea ice classification mechanism.
[1] | |
[2] | |
[3] | |
[4] | |
[5] | |
[6] | |
[7] | |
[8] | |
[9] | |
[10] | |
[11] | |
[12] | |
[13] | |
[14] | |
[15] | |
[16] | |
[17] | |
[18] | |
[19] | |
[20] | |
[21] | |
[22] | |
[23] | |
[24] | |
[25] | |
[26] | |
[27] | |
[28] | |
[29] | |
[30] | |
[31] | |
[32] |