SAR Polarimetry for Classification of Sea Ice: a Comparison of Physical Based Algorithms on ICESAR Data
Marino, Armando1; Dierking, Wolfgang2; Hajnsek, Irena3; Wesche, Christine2
1ETH Zurich, SWITZERLAND; 2AWI, GERMANY; 3ETH Zurich & DLR, GERMANY

INTRODUCTION:
The observation of sea ice is a major topic in remote sensing due to the difficulty to gather in situ data at the required spatial and temporal density [1, 2]. Monitoring of sea ice is important for many environmental issues [1]. First of all, it is a sensitive climate indicator and it plays an important role in the global climate system. It restricts the exchange of heat and chemical constituents between ocean and atmosphere, acting as an insulator. Moreover, it influences the global climate system because of effects related to the albedo, reducing the amount of solar radiation absorbed at the Earth's surface. On the other hand, sea ice affects oceanic circulation directly by the rejection of salt to the underlying ocean during ice growth, which may contribute to deep water formation. Besides these, the possibility and safety of navigation in Polar Regions is severely influenced by the presence of sea ice.

SAR POLARIMETRY:
Microwave sensors and Synthetic Aperture Radar (SAR) are very valuable for monitoring of sea ice since they can acquire information in absence of solar illumination (i.e. during Polar nights) and with almost any weather conditions. Unfortunately, the description of the backscattering behaviour of sea ice is particularly challenging. For this reason, many scientists moved toward systems able to increase the amount of information acquired. In this context, polarimetry plays useful role, because it is able to enhance the discrimination capability of the observed target [3, 4]. A scattering (Sinclair) matrix [S] can be used to characterise the polarimetric behaviour of deterministic targets [3]. A scattering vector k can be obtained rearranging the elements of the scattering matrix. A target that pixel per pixel changes its polarimetric behaviour (i.e. scattering matrix) is defined ''partial'' and can be characterised with the second order statistics of the scattering vector. The latter are generally arranged in a covariance matrix [C].

SEA-ICE CLASSIFICATION WITH SAR POLARIMETRY:
This paper will compare three different classification methodologies that make advantage of polarimetric SAR data, in order to understand which the best methodology for the different situations is. 1) The first considers the estimation of ''polarimetric observables'' (ratios and coherences between polarimetric channels) to build a feature vector able to separate the different ice types (and open water) on a multidimensional space. This approach was largely adopted in the literature and its value is a consequence of the choice of observables that physically should capture the different behaviour of ice types and open water [2,5,6]. 2) The Wishart classifier for the covariance matrix [C]. This approach is based on the statistical distance (in the covariance matrix space) of the pixel from the different classes. The supervised version makes use of a first step where a Cloude-Pottier decomposition is performed [3]. The latter was already exploited in some occasions for sea ice classification [7]. 3) The classifier based on the perturbation analysis [8]. In some conditions, this recent classifier showed improvements compared to the Wishart classifier. Specifically, the overall intensity of the backscattering is neglected and this is beneficial in situations where a modulation of the intensity may not be related to physical but rather geometrical phenomena. Even though the intensity was demonstrated to be an essential classifier for sea ice the possibility to decouple intensity to polarimetric information may be beneficial.

DATASET USED:
The dataset exploited in this study was acquired during the ICESAR campaign in 2007 by the E-SAR airborne system of DLR (German Aerospace Agency). The sea ice acquisitions were carried out together with the AWI (Alfred Wegener institute for polar and marine research) around Svalbard over three different sea ice regimes: Fram Strait, Storfjord and Barents Sea [9-10]. In this analysis only L-band acquisitions are used, since they are the only one presenting quad-polarimetric data. The resolution of the system is 2.12m in slant-range and about 1m in. The NESZ goes from -30dB to -35dB, while the incidence angle from 26 to 65 degrees. Aerial photographs of the area during the acquisition show that in most cases different forms of first year ice (level, deformed, rafted) were present, with areas of brash and lead ice between the floes. Sometimes, open water patches were observed in the ice.

DISCUSSION:
As a final remark, the results that will be presented show that polarimetry could help the data analysis solving eventual ambiguities. However, in many instances, the refrain in exploiting polarimetric modes is the impossibility to achieve very large swath (as ScanSAR images) that in many sea-ice applications are needed to cover vast areas in short time. Fortunately, in the next generations of SAR satellites this inconvenient may be bypassed by the possibility to use compact polarimetry (as for the RADARSAT constellation) or dual polarimetry (as for the Sentinel constellation) with ScanSAR modes. In particular, compact polarimetry somehow allows reconstructing quad-polarimetric data, although part of the information is clearly lost.

REFERENCES:
[1] S. Sandven, Johannessen, O.M., and K Kloster, Sea Ice Monitoring by Remote Sensing, John Wiley & Sons, 2006.
[2] C.R. Jackson and J.R. Apel, Synthetic Aperture Radar Marine User's Manual: National Oceanic and Athmospheric Administration (NOAA), NOAA, 2004. [3] S. R. Cloude, Polarisation: Applications in Remote Sensing, Oxford University Press, 2009.
[4] S.V. Nghiem, R. Kwok, S.H. Yueh, and M.R. Drinkwater, ''Polarimetric signatures of sea ice 1. theoretical model,'' Journal of Geophysical Research, vol. 100(13), pp. 665-679, 1995.
[5] Drinkwater, M., Kwok, R., Rignot, R., Israelsson, H., Onstott, R.G., & Winebrenner, D.P. (1992). Chapter 24. Potential Applications of Polarimetry to the Classification of Sea Ice. Geophysical Monograph Series, 68, 419 430.
[6] H. Wakabayashi, T. Matsuoka, K. Nakamura, and F. Nishio, ''Polarimetric characteristics of sea ice in the sea of Okhotsk observed by airborne L-Band SAR,'' IEEE Transaction on Geoscience and Remote Sensing, vol. 42, 2004.
[7] Scheuchl, B., Hajnsek, I., & Cumming, I.G. (2003). Classification Strategies for Polarimetric SAR Sea Ice Data. Proceedings on POLinSAR.
[8] Armando Marino, Shane R. Cloude & Iain H. Woodhouse (2012), Detecting Depolarizing Targets using a New Geometrical Perturbation Filter, IEEE TGRS, 50(10), 2012.
[9] Dierking, W. (2008), ''Technical assistance for the deployment of airborne SAR and geophysical measurements during the ICESAR 2007'', Final Report - Part 2: Sea Ice, ESA-ESTEC Contract No. 20655/07/NL/CB, 80 pp.
[10] Dierking, W. (2010), ''Mapping of different sea ice regimes using images from Sentinel-1 and ALOS synthetic aperture radar'', IEEE TGRS, 48(3), 1045-1058.