Use of Ocean Wave Imaging to Detect the Marginal Ice Zone in ERS-SAR Images

Abstract

The discrimination between open water and sea ice in the marginal ice zone (MIZ) imaged by ERS-SAR is often not unique, especially if it is carried out by using only the normalized radar cross section (NRCS). Dependent on the environmental conditions and the ice type, similar values for the NRCS of sea ice and open water may occur. SAR imaging of ocean surface waves traveling into the MIZ offers the possibility to detect the transition from open water to sea ice. Sea ice impacts directly on the dynamics of a wave system. The wave amplitude is damped, the wavelength is shortened, and the traveling direction is changed according to Snell's law. Furthermore the azimuthal low pass filtering of the nonlinear imaging of ocean waves by SAR, especially for SAR systems with high R/V-ratios like ERS-1/2, is strongly affected. Both effects are clearly visible in SAR image spectra. From the second moments of SAR image spectra three parameters have been derived describing quantitatively the distribution of spectral energy in wavenumber space. The parameters are used to characterize SAR image spectra and their variation due to the influence of sea ice. Thus a change of the parameters reveals the presence of an ice cover on the ocean surface. The parameters can serve for the definition of an ice edge within the MIZ in the sense of an outer border of the influence of sea ice on ocean wave imaging.

Keywords: SAR, Sea ice, Ocean waves, Marginal ice zone

Introduction

Imaging of the marginal ice zone (MIZ) by active space borne microwave sensors like the ERS SAR is an important remote sensing tool for the sea ice coverage in the polar oceans. Independent from daylight and weather conditions it allows the study of sea ice and the transition zone between ice covered and open water surface under different environmental conditions. For the operational mapping and monitoring the ice coverage passive microwave systems like SSM/I (Special Sensor Microwave /Imager) are commonly used. But ice concentration calculations for the MIZ from passive microwave data exhibits an error up to 20% depending on weather and seasonal conditions and ice type. The analysis of ERS SAR images allows to improve and validate algorithms developed for the detection of the ice edge and the calculation of ice concentrations from passive microwave data (Beaven et al., 1996). In this context the information about the MIZ coming from ERS SAR take place of in situ measurements which are almost lacking in this regions due to weather and ice conditions, high costs, and restricted potentialities of ship expeditions.

Several methods for the classification of different ice types in ERS SAR images have been developed and are well established (Fetterer et al.,1994). Giving sufficient results for the discrimination between multiyear (MY) and first year (FY) sea ice within a closed ice pack the methods are affected by the ocurrance of open water which becomes a severe problem in the MIZ. Ice types are distinguishable in SAR images by their radar backscattering properties described by the normalized radar cross section (NRCS). The fact that the NRCS for both surface types becomes similar depending on ice type and the wind conditions above the water surface makes the discrimination between sea ice and open water difficult which is necessary for the detection of the ice edge within the MIZ . Thus in many ERS SAR images taken in the MIZ a clear distinction between sea ice and open water is not possible.

A completely different approach for the detection of the ice edge in ERS SAR images is the analysis of the imaging of ocean surface waves propagating into the MIZ. Sea ice changes directly the dynamics of the waves, and the wave imaging by SAR is modified in a characteristic fashion. These effects can clearly be seen in SAR image spectra and has been shown by many authors (e.g. Liu et al., 1994; Shuchman et al., 1994). In contrast to the imaging of ocean waves in open water which is well understood the imaging of waves in sea ice is still under discussion. Despite this situation in our paper we will present a method based on the analysis of SAR image spectra for the detection of the ice edge. It will be shown that the impact of sea ice on ocean wave imaging can be used as an unique indicator for the presence of an ice coverage.

Ocean Waves Propagating into Sea Ice

Ocean surface waves propagating in sea ice and in open water appear different in ERS SAR images due to different wave dynamics and wave imaging. For a wave propagating into sea ice its imaging is changed which can be seen in SAR image spectra calculated from parts of an SAR image where waves are travelling in open water and in a closed ice pack, respectively.

Change of Physical Properties

At the transition from open water into a sea ice covered water surface the phase velocity of a wave is decreased because the ice acts like a continuum of non-interacting mass points which exert a pressure upon the the water surface (Wadhams et al., 1995). This results in a reduction of the wavelength or a shift of the spectral energy in the wavenumber space of the SAR image spectrum towards higher wavenumbers or shorter wavelengths, respectively. Furthermore, the amplitude of the wave is increased leading in combinition with the reduction of the wavelength to a greater wave steepness. According to Snell's law the wave underlies a refraction towards the normal of incidence of the ice edge (Shuchman et al.,1994 ; Wadhams et al., 1995). This results in a rotation of the energy in the SAR image spectrum depending on the orientation of the wave propagation direction in open water with respect to the ice edge.

Change of Wave Imaging

The imaging of an ocean surface wave propagating in open water can be described by the velocity bunching model which is an excellent approximation including all wave motion induced effects of the water surface up to the second order in time (Alpers and Rufenach, 1979). Wave imaging by SAR is a nonlinear process and the degree of the nonlinearity depends on the amplitude , the wavenumber and the propagation direction with respect to the azimuth direction (i.e the flight direction of the SAR sensor) if the R/V-ratio (R is the distance between the SAR antenna and the imaged surface and V is the velocity of the antenna platform) and incidence angle of the radar are fixed. It increases with increasing amplitude and increasing wavenumber and it decreases if the propagation direction changes from azimuth to range direction (i.e the look direction of the antenna). Despite the fact that a linear or weak nonlinear imaging of ocean waves is possible under appropriate wave conditions, in many practical cases the imaging of ocean waves by ERS SAR is strongly nonlinear because of its high R/V-ratio of 130s. Furthermore the imaging of ocean waves is affected by a degradation of the azimuthal resolution caused by the motion of the water surface induced by waves of a smaller scale than the nominal spatial resolution of the ERS SAR of 25m (subscale waves). The velocity bunching effect of the nonlinear ocean wave imaging causes a shift and a rotation of the energy in a SAR image spectrum compared to the energy distribution of the ocean wave spectrum (Bruening et al., 1990). The degradation of the azimuthal resolution results in a narrowing of the energy distribution of the SAR image spectrum in its azimuthal extension in wavenumber space because waves shorter than two times of the degraded azimuthal resolution are not imaged. This effect is very often visible in SAR image spectra calculated from ERS SAR images. The imaging of ocean waves in sea ice is less affected by the degradation of the azimuthal resolution because the subscaled waves are damped out very rapidly and waves with shorter wavelength (higher wavenumbers) are imaged. This results in a broadening of the SAR image spectrum in azimuth compared to that one in open water. The physical change and the change of the nonlinear imaging act together in modifying the energy distribution of the SAR image spectra of wave systems propagating in sea ice compared to the same wave system propagating in open water. This can be used as an indication for the appearance of sea ice on the water surface.

Use of SAR Images Spectra

As a result of the combination of the physical change of the waves and the wave imaging SAR image spectra of waves in ice differ in a characteristic fashion from, SAR image spectra of waves in open water. Figure 1 shows an ERS-1 SAR PRI intensity image taken in the Bellingshausen Sea. In the upper left part of the image an area covered by sea ice is visible but even if the transition zone between open water and sea ice can be identified by human eye the automatic detection of the ice edge using the NRCS is difficult. In areas with open water an ocean surface wave field is imaged which is not visible in this reproduction of the SAR image. The wave system is propagating into the ice covered region.

Figure 1: ERS -1 SAR PRI image (orbit 22270, frame 4869) of the ice edge in the Bellingshausen Sea taken at October 18, 1995, 12:59 UTC, center coordinates: 61.49^o S, 59.39^o W, the flight direction is 205^o to N

Spectral Energy Distribution

Figures 2 and Figure 3 show SAR image spectra calculated from the areas 1 and 2 marked in Figure 1. The energy distribution of the SAR image spectra is presented in wavenumber space (or k-space) in an absolute scale in dB, the k_x-axis is parallel to the flight or azimuth direction of the SAR and the k_y-axis is parallel to the antenna look or range direction respectively. The circles represent the wavelength of 50m, 100m, 150m, 200m, 250m, and 300m from the outer to the inner. The spectrum in Figure 2 shows how the wave system propagating in open water is imaged by the ERS-1 SAR. The spectral maximum is at a wavelength of 320m and its propagation direction is 53^o with respect to the flight direction . The angle is counted clockwise because the ERS SAR is a right looking system. Due to the point symmetry of SAR image spectra a direction of 223^o is also possible but not consistent with the surrounding distributuion of fast ice and land areas which prevent waves travelling in this direction. These values describe the length and orientation of the dominant wave as it appears in the ERS SAR image and not the physical values for this wave on the water surface.

Figure 2: SAR image spectrum calculated from area 1 in Figure 1. The spectral energy is presented in an absolute scale in dB. The circles represent from the outer to the inner 50m, 100m, 150m, 200m, 250m, and 300m wavelenght. The spectral maximum is 7.4 dB at a wavelength of 320m oriented in 53^o with respect to the azimuth direction.

Figure 3 shows a SAR image spectrum of the same wave system propagtion in sea ice. The wavelength of the spectral maximum has changed to 223m and the direction to 4^o. We interpret this effect as predominantly caused by the physical change of the wavesytem due to the refraction at the ice edge. The distortions caused by the nonlinear wave imaging seem to play a minor role for the spectral maximum in this case. In this SAR image spectrum energy appears at azimuthal wavenumbers greater than k_x=0.031 rad/m (i.e a wavelenght of about 200 m) which is not visible in the spectrum presented in Figure 2. The interpretation is that the damping of waves with wavelength shorter than the nominal resolution results in a reduction of the degradation of the azimuthal resolution. Waves which are smeared out by the degraded azimuthal resolution in open water are imaged in ice .

Figure 3: SAR image spectrum calculated from area 2 in Figure 1. The presentation is the same as in Figure 2. The spectral maximum is 6.8 dB at a wavelength of 228m oriented in 4^o with respect to the azimuth direction.

Parameters from Spectral Moments

The SAR image spectra in Figures 2 and 3 show how the differences of the imaging of ocean waves by ERS-SAR reveals the occurance of sea ice. But nevertheless they are calculated from two small areas of the full SAR image only. We calculate SAR image spectra within the whole SAR image in boxes of 256 x 256 pixels (i.e. 1600m x 1600m) with an overlap of 128 pixels (i.e. 800m) in both directions. Because it is impossible to compare so many spectra to each other, parameters have been derived describing the characteristic properties of the spectral energy distribution of a SAR image spectrum quantitatively. For this purpose from the SAR image spectrum

the spectral moments

are calculated. The moment M₀₀ represents the total energy of the spectrum. The moments M₀₁=M₁₀ vanish due to the point symmetry of the spectrum. The moments M_20, M₁₁, and M₀₂ describe the deviation of the spectral energy distribution from rotational symmetry to the second order and form a symmetric 2 x 2 tensor, which can be represented by its eigenvalues and its angle of principal axis by applying a principal axis transform (diagonalization):

From these variables we derived spectral parameters which describe the characteristic porperties of the energy distribution of SAR image spectra:

1. Directionality:

The directionality describes how strong the spectral energy is concentrated in one spectral maximum (or due to the ambigiuty of the SAR image spectra in two maxima located on opposite positions in wavenumber space). For a single pronounced spectral maximum the value of the directionality is close to 1, for a broad maximum or a distribution with two or more clearly separated maxima its value is close to 0.

2. Angle of principal axis :

The angle of principal axis describes the orientation of the spectral energy distribution in wave number space. It is a measure for the propagation direction of a wave system as it appears in the SAR image only if the spectral energy is concentrated in a single narrow maximum. In general it describes the globaly preferenced orientation of the spectral energy.

3. Normalized difference of the spectral energy extension between azimuth and range, short we normalized extension difference, NED:

The NED measures how strong the spectral energy is concentrated in a small band in the range direction in wavenumber space due to the degradation of the azimuthal resolution. Waves propagating in open water are often characterized by an NED below -0.25 due to this effect.

Results

The results of the calculation of spectral parameters for the ERS-1 SAR image in Figure 1 are presented in Figures 4, 5, and 6. Figure 4 shows the values for the directionality, Figure 5 for the angle of principal axis, and Figure 6 for the NED. In Figure 4 the open water surface is characterized by high values for the directionality from 0.75 to 1.0 (bright area). This is consistent with the appearance of a single spectral maximum in the SAR image spectrum in Figure 2 indicating that a unimodal wave system is imaged. In the upper left part of the image where sea ice is present, a bright area (high directionality) is separated by a dark band from the area of open water. This can be explained by the diffuse structure of the transition zone between ice and open water. The waves are refracted at the ice covered areas within this zone and propagate undisturbed in the ice free zones. This can be seen in SAR image spectra calculated in this zone (not presented here), where at least two spectral peaks of similar intensity appear resulting in a low directionality of less than 0.75. In the adjacent bright area the wavesystem changed its wavelength and propagation direction and only one spectral maximum appears in the SAR image spectrum (see Figure 3). This leads to values for the directionality above 0.75. In the upper left corner of the image the wave system disappears due to the dissipation of wave energy by damping. Consistent to the interpretation of the directionality in Figure 4 are the values for the angle of pricipal axis shown in Figure 5. The spectral energy appears in the image with orientation angles between 60^o and 90^o with respect to the flight direction in open water. In the transition zone the orientation changes succesively from 60^o to 0^o due to the refraction. In the upper left corner of the image the values for the angle varies between -90^o and 90^o because waves are absent and do not dominate the orientation of the spectral energy distribution. Figure 6 shows that the NED is below -0.25 for open water indicating that the spectral energy distribution is narrow in its azimuthal extension in wavenumber space due to the degradation of the azimuthal resolution. This effect is reduced for the waves propagating in sea ice in the upper left part of the image rising the value for the NED up to a range between 0 and 0.75.

Figure 4: Image presenting the values for the directionality calculated from the image in Figure 1.

Figure 5: Image presenting the values for the angle of pricipal axis calculated from the image in Figure 1.

Figure 6: Image presenting the values for the normalized extension difference, NED, calculated from the image in Figure 1.

Conclusions

The imaging of ocean surface waves by ERS SAR allows the detection of the ice edge in the marginal ice zone devoiding the use of the backscattering properties described by the normalized radar cross section (NRCS). Due to the differences in wave dynamics and in the imaging by SAR for ocean surface waves propagating in open water and in sea ice the analysis of SAR image spectra reveals the existence of an ice covered water surface. Three spectral parameters have been derived from SAR image spectra describing essential properties as concentration, orientation, and azimuthal extension of the energy distribution in SAR image spectra in wavenumber space. The changes of the wave imaging at the transition from open water to sea ice are clearly detectable by these spectral parameters and are used for the predefinition of the ice edge in ERS SAR images.The ice edge in ERS SAR images is defined as the outer border of influence of sea ice on the imaging of ocean surface waves measurable by the spectral parameters.

Acknowledgement

This work was supported by ESA under contract AO2.D146-3 and by the Bundesministerium fuer Bildung, Wissenschaft und Technologie (BMBF) under contract 03PL018A.

References

Alpers, W., Rufenach, C. L., 1979

The effect of orbital motion on synthetic aperture radar imaging of ocean waves. IEEE Trans. AP, 27, pp. 685-690.

Beaven, S. et. al., 1996

Fusion of Satellite Active and Passive Microwave Data for Sea Ice Type Concentration Estimates. IEEE Trans. GRS, 34, pp. 1172-1183.

Bruening , C. et. al., 1990

Monte-Carlo simulation studies of the nonlinear imaging of a ocean wave field by a synthetic aperture radar. Int. J. Rem. Sen., 11, pp. 695-1729.

Fetterer, F. et. al., 1994

Sea ice type maps from Alaska Synthetic Aperture Radar Facility imagery: An assessment.

JGR, Vol. 99, No. C11, pp. 22443-22458.

Liu, A. et. al., 1994

Ocean-ice interaction in the marginal ice zone using synthetic aperture radar imagery. JGR, Vol. 99, No. C11, pp. 22391-22400.

Shuchman, R . et .al., 1994

Extraction of marginal ice zone thickness using gravity wave imagery. JGR, Vol. 99, No. C1, pp. 901-918.

Wadhams, P. et. al., 1995

Wave Spectra of SAR Imaging of the Odden Ice Tongue. Proc. IGARSS'95, pp. 630-633

Keywords: ESA European Space Agency - Agence spatiale europeenne, observation de la terre, earth observation, satellite remote sensing, teledetection, geophysique, altimetrie, radar, chimique atmospherique, geophysics, altimetry, radar, atmospheric chemistry