EXTRACTION OF LATE SUMMER SEA ICE PROPERTIES FROM POLARIMETRIC SAR FEATURES IN C- AND X- BAND

Transcription

1 EXTRACTION OF LATE SUMMER SEA ICE PROPERTIES FROM POLARIMETRIC SAR FEATURES IN C- AND X- BAND Ane S. Fors 1, Camilla Brekke 1, Sebastian Gerland 2, Anthony P. Doulgeris 1, and Torbjørn Eltoft 1 1 Department of Physics and Technology, University of Tromsø - The Arctic University of Norway, 9037 Tromsø, Norway, ane.s.fors@uit.no, camilla.brekke@uit.no, anthony.p.doulgeris@uit.no and torbjorn.eltoft@uit.no 2 Norwegian Polar Institute, FRAM Centre, 9296 Tromsø, Norway, gerland@npolar.no ABSTRACT In this study we examine the potential use of six polarimetric features for interpretation of late summer sea ice types. Five high-resolution C and X-band scenes were recorded in the Fram Strait covering fast first-year and old sea ice. In addition sea ice thickness, surface roughness and melt pond fraction were collected during a helicopter flight at the study area. From the SAR scenes, six polarimetric features were extracted. Along sections of the track of the helicopter flight, the mean of the SAR features were compared to mean values of the properties measured during the helicopter flight. The results reveal relations between several of the SAR features and the geophysical properties measured in C-band, and weak relations in X-band. 1. INTRODUCTION With the decline in sea ice extent and a lengthening of the summer melt season observed in the Arctic [1], a need of more detailed monitoring of the summer sea ice cover is required both for shipping, oil- and gas industries, and for climate science. Synthetic aperture radar (SAR) offers all-weather highresolution imagery of the Arctic region. SAR polarimetry is known to provide information about scattering mechanisms, which may help interpreting microwave signatures of sea ice. Interpreting SAR scenes in the melt season is however difficult. Temperatures varying around freezing point leads to large changes in the dielectrical properties of the sea ice, and high moisture contents in the top layers of the ice mask ice type differences. Even if some studies have been performed ([2, 3, 4, 5]), interpretation of SAR polarimetric signatures of summer sea ice is still a challenge. In this study we examine the relationship between six polarimetric SAR features and sea ice thickness, sea ice surface roughness and melt pond fraction. The study is a first step in exploring the SAR features potential for late summer sea ice type extraction. Five high-resolution C and X- band scenes from Radarsat-2 and TerraSAR-X are used in the study. These scenes were acquired during a week in the melt-freeze transition in late summer in the Fram Strait in POLARIMETRY The six polarimetric SAR features investigated in this study are all related to the covariance matrix, C. The Lexiograpic feature vector, s, forms the basis of C. For three polarimetric channels (d = 3) s is given by s = [ S HH 2SV H S V V ] T, (1) where T denotes transpose. Reciproxivity is assumed (S HV = S V H ). The 3 3 covariance matrix C is then given by C = 1 L L i=1 s i s T i, (2) where s i is the single look complex vector corresponding to pixel i, L is the number of scattering vectors in a local neighbourhood and denotes the complex conjugate. The six features investigated in this study are presented in Table 1. A combination of the features have previously shown promising results in segmentation of both a Radarsat-2 winter sea ice scene [6] and the three Radarsat-2 late summer sea ice scenes included in this study (Fors et.al (2014), manuscript submitted). Further descriptions of the features can be found in [6]. Note that as the TerraSAR-X scenes are dual-polarimetric, hence the covariance matrix reduces to a 2 2 matrix and all features cannot be retrieved for these scenes (see Table 1). In our study a neighborhood of 7 7 pixels (L = 49) is used, applied with a non-overlapping stepping window.

2 Table 1: Polarimetric SAR features included in the study. Polarimetric feature Relative kurtosis Geometric brightness Cross-polarisation ratio Definition PL 1 T 1 si ]2 RK = L1 d(d+1) i=1 [si C B = d detc <S SV H> RV H/V V = <SVV H > VS Extracted for scene All scenes All scenes R1, R2, R3, T2 RV V /HH = VV Co-polarisation ratio Co-polarisation correlation magnitude Co-polarisation correlation angle <SV V SV V> <SHH SHH > <SHH SV V > ρ = > <SHH SHH ><SV V SV V 6 ρ = 6 (< SHH S V V >) Figure 1: Map of the Fram Strait showing the location of the satellite scenes included in the study (black frames for Radarsat-2 and green frames for TerraSAR-X) and the track of the helicopter flight collecting airborne measurements for the study (blue dashed line). At the time of the flight, R/V Lance was slightly north of this map section. 3. THE DATASET The data used in this study were collected from a ship, helicopter and satellite-borne campaign in the Fram Strait in late summer The study site is situated in an area with iceberg-fast sea ice, with both first-year and old sea ice in different stages of development. Three C-band scenes from Radarsat-2 (RS-2) and two X-band scenes from TerraSAR-X (TS-X) were acquired during the campaign, all partly covering the same area on the ground (see Figure 1 and Table 2). All scenes were recorded in ascending satellite orbit. Airborne observations were collected during a helicopter flight at the study site (see Figure 1, Figure 2 and Table 2). Sea ice thickness was recorded with an electromagnetic induction sounder and a laser altimeter towed under the helicopter. Details about the technique can be found in [7]. The laser altimeter was also used to find the surface elevation relative to level ice [8]. Helicopter altitude variations were removed as described in [9]. Surface roughness was calculated as the standard deviation of the profile surface elevation about the mean (root mean square N Figure 2: Position of the recorded measurements from the helicopter flight used in this study displayed on the Radarsat-2 scene from 31 August 2011 (R2). The polarimetric image is a Pauli composite, the intensity channel combinations HH V V, 2 HV and HH + V V are assigned to the RGB channels, respectively. The dark area to the left is open water, the light purple area in the middle of the scene consist of thin first year ice, the remaining purple areas consists of thick first-year ice and old ice and the brighter areas to the right consists of heavily deformed old ice. height). A digital camera (GoPro YHDC5170) was taking downward looking photographs of the sea ice surface from the helicopter. The area coved by each image is about 100 m (length) 140 m (width), and the sampling rate was 0.5 Hz. The images was classified with a semiautomatic classification algorithm to retrieve the fraction of sea ice covered by melt ponds (melt pond fraction) in each image [10]. Whether the melt ponds are open or refrozen during the period of the campaign is not known, as no ground-based information could be retrieved from the study site. Air temperature was recorded at R/V Lance, sailing within 100 km north and west of the study site. The temperature was fluctuating around zero degrees Celsius during the campaign.

3 Table 2: Properties of the satellite SAR scenes and the helicopter flight. Date Time (UTC) Scene ID Satellite and Mode Polarization Incidence angle [ ] Pixel spacing [m] (azimuth slant range) 29/08/ :41 R1 Radarsat-2, Fine Quad HH,HV,VH,VV x 5.0 m 30/08/ :23 T1 TerraSAR-X, StripMap HH,VV x 1.9 m 31/08/ :23 R2 Radarsat-2, Fine Quad HH,HV,VH,VV x 5.1 m 03/09/ :09 Helicopter fligth 04/09/ :07 R3 Radarsat-2, Fine Quad HH,HV,VH,VV x 6.8 m 05/09/ :00 T2 TerraSAR-X, StripMap VH,VV x 2.1 m In this study, mean values of the SAR features and the geophysical properties are calculated from co-located sections of 400 meters length along the track of the helicopter flight, and are compared in scatter plots. Pearsons correlation coefficient (R) is used together with visual inspection of the plots to examine the relationships between the SAR features and the measured geophysical properties. The SAR features showing the most promising relationship to each of three geophysical properties are further investigated in the following section. 4. RESULTS AND DISCUSSION Both C and X- band scenes were investigated in this study. The results are presented and discussed separately for the two frequency-bands in this section C-band Among the six investigated SAR features in C-band, cross-polarisation ratio was found to show the strongest relationship to sea ice thickness, increasing with increasing sea ice thickness (see Figure 3 (a)-(c)). Crosspolarisation ratio is a measure of depolarisation. Its relationship to sea ice thickness could be explained from difference in surface roughness between young (thin) and older (thicker) sea ice. Difference in sea ice structure, such as grain size and size and amount of air pockets within the sea ice can also cause difference in depolarisation. The salinity of first-year and old sea ice is expected to be similar at the end of the melt season as melt water flushes most of the brine inclusions from the firstyear ice during summer. Salinity is therefor not expected to cause differences in depolarisation between first-year ice and old ice in late summer. The relationship between cross-polarisation ratio and sea ice thickness is possibly dependent on incidence angle. The weakest correlation (R = 0.17) occurs in the scene with the lowest incidence angle (R1), and the strongest correlation (R = 0.80) occurs in the scene with the highest incidence angle (R2). However, changes in geophysical conditions at the ground could also cause these differences. Sea ice thickness can not be measured directly with C- band SAR. The relationship between sea ice thickness and any C-band SAR feature would therefor always be dependent on the location of the ice, the time of year and the growth history of the ice. Relative kurtosis was found to have the strongest relationship to melt pond fraction among the SAR features, decreasing with increasing melt pond fraction (see Figure 3 (d)-(f)). This relationship appears to be independent of incidence angle. Relative kurtosis describes the intensity distribution of the SAR backscatter signal. A relative kurtosis of unity points towards Gaussian distributed data, higher relative kurtosis indicate a sharper distribution with heavier tails [6]. The lower relative kurtosis of areas heavily ponded could be caused by a more even mixture between ponds and ice, but could also be related to backscatter from the edges of the ponds. There is a need of further investigation to better understand this relationship. Relative kurtosis was also found to show the strongest relationship to surface roughness among the SAR features, increasing with increasing roughness (see Fig.3 (g)-(i)). Deformed areas and inhomogeneous areas are expected to produce a higher relative kurtosis [6], hence this result is expected. The relationship is however weak for two of the scenes (R < 0.42), and could also be a result of the strong existing correlation between melt pond fraction and surface roughness X-band In X-band the relationships between the six investigated SAR features and sea ice thickness, melt pond fraction and surface roughness were weak. Several factors could explain this lack of relation, compared to the C-band scenes. The incidence angle was much lower in the TS-X scenes than in the RS-2 scenes. The TS-X scenes were also dual-polarimetric, containing less polarimetric information than the RS-2 scenes. In addition, one of the TS-X scenes, T1, was acquired at a time with temperatures above zero degrees Celsius and high relative humidity. These conditions are probably at the limit of conditions suitable for sea ice type discrimination with SAR. Investigation of more TS-X scenes are necessary, to reveal their potential in sea ice type extraction.

4 (a) R1 (b) R2 (c) R3 (d) R1 (e) R2 (f) R3 (g) R1 (h) R2 (i) R3 Figure 3: Scatter plots of geophysical properties vs SAR features for the three RS-2 scenes included in the study. (a)-(c) Cross-polarisation ratio ( (R V H/V V )) vs sea ice thickness. (d)-(f) Relative kurtosis (RK) vs melt pond fraction. (g)-(i) Relative kurtosis (RK) vs surface roughness (root mean square height). Pearsons correlation coefficient, R, is a measure of the linear relationship.

5 5. CONCLUSIONS In this study we investigated the relationship between six polarimetric SAR features and sea ice thickness, surface roughness and melt pond fraction, as a first step in examining the features potential in sea ice type extraction. The study was performed both on X and C-band SAR scenes. In C-band, three relationships between the SAR features and the measured geophysical properties were discussed. Cross-polarisation was found to have the strongest relationship to sea ice thickness among the investigated features. Relative kurtosis was found to have the strongest relationship to both melt pond fraction and sea ice surface roughness. In X-band, the relationships between the SAR features and the measured geophysical properties were weak. The scenes had lower incidence angle, fewer polarimetric channels and more challenging temperature conditions on the ground than the C-band scenes, hence the lack of relationships in X-band is not necessarly a result of the frequency. More scenes and further investigations are needed to fully reveal the potential of the six investigated polarimetric features for late summer sea ice type extraction in C and X-band. ACKNOWLEDGMENTS The authors would like to thank the captain, crew and scientist from the Norwegian Polar Institute onboard R/V Lance in the Fram Strait 2011 for data collection. A special thank goes to Angelika H.H. Renner for all help with collecting and preprocessing the helicopter measurements and for good discussions. Thanks also to Justin Beckers at University of Alberta, Canada, for preprocessing the laser altimeter measurements. Radarsat-2 data are provided by NSC/KSAT under the Norwegian-Canadian Radarsat agreement 2011 and TerraSAR-X data are provided by InfoTerra. This project was supported financially by the project Sea Ice in the Arctic Ocean, Technology and Systems of Agreements ( Polhavet, subproject CASPER ) of the Fram Centre, and by the Centre for Ice, Climate and Ecosystems at the Norwegian Polar Institute. This project was also funded financially by Regional Differensiert Arbeidsgiveravgift (RDA) Troms County. of freeze-up and melt processes on microwave signatures. In F. D. Carsey, editor, Microwave Remote Sensing of Sea Ice, number 68 in Geophysical Monograph, pages AGU, [3] D. Isleifson, A. Langlois, D. G. Barber, and L. Shafai. C-Band Scatterometer Measurements of Multiyear Sea Ice Before Fall Freeze-Up in the Canadian Arctic. IEEE Transactions on geoscience and remote sensing, 47(6): , [4] K. Warner, J. Iacozza, R. K. Scharien, and D. G. Barber. On the classification of melt season firstyear and multi-year sea ice in the Beaufort Sea using Radarsat-2 data. International Journal of Remote Sensing, 34(11): , June [5] R. K. Scharien, K. Hochheim, J. Landy, and D. G. Barber. First-year sea ice melt pond fraction estimation from dual-polarisation C-band SAR Part 2: Scaling in situ to Radarsat-2. The Cryosphere, 8(6): , November [6] M.-A. Moen, A. P. Doulgeris, S. N. Anfinsen, A. H. H. Renner, N. Hughes, S. Gerland, and T. Eltoft. Comparison of feature based segmentation of full polarimetric SAR satellite sea ice images with manually drawn ice charts. The Cryosphere, 7(6): , November [7] C. Haas, J. Lobach, S. Hendricks, L. Rabenstein, and A. Pfaffling. Helicopter-borne measurements of sea ice thickness, using a small and lightweight, digital EM system. Journal of Applied Geophysics, 67(3): , March [8] J. Beckers, A. H. H. Renner, G. Spreen, S. Gerland, and C. Haas. Sea ice surface roughness estimates from airborne laser scanner and laser altimeter observations in Fram Strait and north of Svalbard. Annals of Glaciology, 56(69), [9] W. D. Hibler. Removal of aircraft altitude variation from laser profiles of the arctic ice pack. Journal of Geophysical Research, 77(36): , December [10] A. H. H. Renner, M. Dumont, J. Beckers, S. Gerland, and C. Haas. Improved characterisation of sea ice using simultaneous aerial photography and sea ice thickness measurements. Cold Regions Science and Technology, 92:37 47, August REFERENCES [1] W. N. Meier, G. K. Hovelsrud, B. E. H. van Oort, J. R. Key, K. M. Kovacs, C. Michel, C. Haas, M. A. Granskog, S. Gerland, D. K. Perovich, A. Makshtas, and J. D. Reist. Arctic sea ice in transformation: A review of recent observed changes and impacts on biology and human activity. Reviews of Geophysics, 52(3): , September [2] S. P. Gogineni, R. K. Moore, T. C. Grenfell, D. G., Barber, S. Digby, and M. Drinkwater. The effects