Test site

Figure 1shows amap of the test site. The airborne SAR and the moored sonar observation data were obtained simultaneously off the Okhotsk coast of Hokkaido. The SAR observations were carried out twice, at 16:30 Japanese Standard Time (JST) on 23 Februaryandat12:45 JST on 24 February 1999. The weather conditions during this period were relatively calm. The average wind speed was 3.5 ms^–1comingmainly from the west. The air temperature ranged from –12³to 0°C at the nearest ground meteorological station, which was 12km from the IPS.

Fig. 1.

Map of test site. The two rectangles in the enlarged map are 23 and 24 February 1999 Pi-SAR observation areas. The marked mooring point is the IPS and ADCP location.

Airborne SAR

The dual-frequency, multi-polarization airborne SAR (Pi-SAR) was developed by the Communications Research Laboratory and National Space Development Agency of Japan. The frequencies are X- and L-band with resolutions of 1.5 and 3.0 m. The ground-range swath is approximately 15 km, and typical incidence angles range from 20³to 60°. The Pi-SAR has polarimetric (X- and L-band) and interferometric (X-band) functions (Reference KobayashiKobayashi and others, 2000). In this study we discuss the backscattering magnitude of HH, VV and HV polarization channels since polari-metric calibration is still ongoing. A polarimetric analysis will be carried out in the near future.

The observation areas are shown as rectangles in the enlarged map in Figure 1. Figure 2 shows the Pi-SAR images obtained on 23 and 24 February, as a color composite of the HH (red), VV (green) and HV (blue) channels. The backscattering power (brightness in the image) from thin sea ice and first-year ice is dominated by the ice surface roughness related to the radar wavelength and the dielectric constant that was determined by the salinity of the sea ice.

Fig. 2.

Pi-SARimages (7 km in ground range, 10 km in azimuth) obtained on 23 and 24 February, as a color composite of the HH (red),VV (green) and HV (blue) channels. Numbers in the center are the incidence angles. Moored IPS location (red point) and estimated track of sea ice that passed over the IPS (yellow lines) are superimposed in (b) and (d).

Although the sea ice drifted eastwards about 7.5 km during the 2 day observation (20 hours), the relative arrangement of the ice barely changed because of the calm weather during this period.

IPS and ADCP

The IPS and ADCP were moored about 10 km off the Okhotsk coast, where the water is about 60m deep. The moored IPS was located around 44°20’N, 144°39’ E, and the time series of the sea-ice draft data was obtained from early February to late March. The horizontal spot on the IPS beam was about 1.5m, and the ice-draft uncertainty was 0.1–0.2 m. This value depends on the accuracy of the temperature-dependent sound-speed estimation in the water (Reference Melling, Johnston and RiedelMelling and others, 1995). The time series of the sea-ice velocity data was obtained by the ADCP. The accuracy of the ice velocity was 0.2 cm s^–1. The track of the sea ice passing over the IPS was estimated using the ADCP ice-velocity data, and the track between the two SAR observations is superimposed on Figure 2b and d. The total track distance measured over 20 hours was 7590 m, and the average ice velocity was 10 cm s^–1.

IPS and SAR profiles

Figure 3a shows the topographic profile of the ice draft, which was converted from equal time intervals to equal space intervals using the ADCP ice-velocity data. The average ice draft was about 0.49 m, and profile maximum is 4.77 m. The standard deviation profile of the ice draft is shown in Figure 3b. It was derived from ice-draft data from five neighboring points at 0.5 m intervals. The profile shape had good agreement with the shape of the ice-draft profile. Large values for standard deviations and thick ice (>1m) are often observed at the rim of ice floes. We concluded from these results that thick ice was formed by rafting and ridging at the rims of the ice.

Fig. 3.

A 7.6 km section of profiles from the observations by moored sonar and Pi-SAR: (a) IPS profile of ice draft; (b) profile of standard deviation of ice draft; (c) L-band HH, (d) L-band VV, (e) L-band HV, (f) X-band HH, (g) X-band VV, and (h) X-band HV polarization SAR profiles of backscattering coefficient. Hatched areas indicate thin-ice areas.

Figure 3c–h show the backscattering coefficient (σ°) profile of the estimated ice track (solid lines in Figure 2b (0–3600m) and d (3600–7600m)). The backscattering coefficient has incidence-angle dependence. We corrected for the dependence by comparing the backscattering coefficients of the observed locations that had different incidence angles in 2 day SAR observations. The correction coefficients were 50.4 dB per degree in all polarization channels of the L-band and 50.2 dB per degree in all polarization channels of the X-band around an incidence angle of 45°.

Figure 3e and g show the backscattering coefficient profiles of the L-band HV and the X-band VV at a corrected 45³incidence angle. The SAR data were averaged to 12.5m (5 pixels for the L-band, 10 pixels for the X-band) for the direction perpendicular to the ice track, and to a 12.5m running mean with 2.5 m intervals for the direction of the ice track. The L-band profile had good agreement with the ice draft and the standard deviation. The correlation coefficient between the backscattering coefficient of the L-band HV channel and the ice-draft profile reached 0.64, which was the highest correlation with the draft data for the three L-band channels (0.46 at HH, 0.42 at VV, 0.64 at HV). There were much higher correlations between the profiles of SAR backscattering and ice-draft data near the start and the end than in the middle of the estimated ice track. This suggests that the difference in the data-acquisition points of SAR and IPS was largest in the middle of the estimated ice track.

The backscattering of cross-polarization is mainly caused by surface multiple scattering and volume scattering from sea ice and is not related directly to the ice draft. This suggests that the surface roughness of thick sea ice resulting from ridging and rafting caused the large backscattering.

The profile of the backscattering coefficient of X-band VV had almost no correlation with the ice-draft data. However, there was good correlation between them in the open-water area, where the X-VV backscattering was extremely small. Interestingly, large backscattering was clearly observed in the thin-sea-ice areas (hatched areas in Fig. 3), which were not distinguished from the open water in the L-band SAR and IPS data.

Ice-draft estimation using SAR data

The good correlation between the ice-draft profile and the L-band HV SAR backscattering coefficient profile (Fig. 3a and e) enables us to estimate the horizontal distribution of the ice draft using the SAR data. Figure 4 shows the scatter plot of the backscattering coefficient in decibels vs the ice-draft data on a log scale. The backscattering coefficient was simply approximated by the regression line in Figure 4. The regression line is described as follows:

Fig. 4.

Scatter plot of L-band HV polarization backscattering coefficient in decibels vs ice-draft data on log scale. Dashed line is regression line. The noise floor was less than –40 dB (dotted line).

(1)

where d is the ice draft in meters. The dispersion of the backscattering values from the regression line was mainly caused by small differences in the data-acquisition points of the SAR and IPS. The backscattering coefficient for the change in the ice draft ranged more widely in the L-band HV channel than in the other polarization channels.

We converted the SAR backscattering image of the L-band HV channel in Figure 2b into the estimated ice-draft image that is shown in Figure 5a using Equation (1). This image suggests that the rim of ice floes has large ice-draft values. Ice thicker than 1 m in this image reached 15% of the area, containing more than one-third of the fresh-water volume in this area.

Fig. 5.

(a) Estimated ice-draft image converted from the backscattering coefficient image of the L-band HV channel. (b) Classification of three ice categories: OWis open water, N-GI is nilas and gray ice, and GWI-FYI is gray-white ice and first-year ice.

Classification of ice types

Distinguishing open water and thin ice <15 cm thick using L-band SAR data is still difficult because the backscattering power from open water was comparable to that from thin ice. Also, the values were near the noise floor at a relatively large incidence angle of 45°. The noise floors in the SAR observations were estimated to be less than –40 dB at L-HV, –34dB at L-HHand L-VV, –38dB atX-HV and –28 dB at X-HHandX-VVin backscattering coefficient. However, the X-bandVVchannel backscattering profile shown in Figure 3g suggests that there was large backscattering from the thin-ice area (over 18 dB larger than the backscattering from open water). Also, a large polarization ratio (VV/HH) in X-band backscattering from thin ice was reported by Reference MatsuokaMatsuoka and others (2001) in the same Pi-SAR experiments.

In the case of thin ice and first-year ice, the surface roughness and the salinity play the dominant role in determining the backscattering properties. Previous research showed that the measured rms roughness of sea ice thinner than 15 cm was mostly 50.1 cm (Reference Onstott and CarseyOnstott, 1992). According to the Fraunhofer criterion for a surface to be considered smooth (Reference Ulaby, Moore and FungUlaby and others, 1982), the value of 0.1 cm is smooth enough to reduce the backscattering power for the large wavelength of the L-band and is not for the X-band. First-order model calculations for the relatively smooth surface sea ice showed that the polarization ratio increases as the rms roughness decreases and as the salinity increases (Reference Winebrenner, Tsang, Wen and WestWinebrenner and others, 1989).

These results allowed classification of the ice types using the X-band VV and HH data together with the L-band data. Ice types were classified into the World Meteorological Organization’s standard categories, based on the ice thickness (WMO, 1970). The classification algorithm derived from comparing the ice-draft data and the SAR data is shown in Figure 6. The SAR channels used in classification and the threshold values were chosen by picking the training sets along the estimated ice track. The estimated accuracy of the classification was 490% in each ice (surface) type. Open water was detected using the X-band VV channel; gray-white ice and first-year ice thicker than 15 cm were detected using the X-band HH and L-band HH channels; and thin ice, such as nilas and gray ice less than 15 cm, was detected using the X-band polarization ratio. Figure 5b shows the classified ice. We can clearly distinguish the thin-ice areas between the open water and first-year ice. The image showed 13% open water, 8% thin ice (nilas and gray ice) and 79% first-year ice (gray-white ice and first-year ice).

Fig. 6.

Example block diagram of ice-type classification algorithm.