Introduction

In winter the Sea of Okhotsk is covered with various types of sea ice whose thickness ranges from a few centimeters to several meters (Reference Fukamachi, Mizuta, Ohshima, Toyota, Kimura and WakatsuchiFukamachi and others, 2006; Reference Uto, Toyota, Shimoda, Tateyama and ShirasawaUto and others, 2006). Ice thickness plays a crucial role in the heat exchange between the atmosphere and ocean, in particular ice growth processes, and then the climate on a global scale. Therefore ice-thickness distribution is one of the most important parameters for understanding the effect of sea ice on the climate system. Since spaceborne remote sensing is an ideal tool for obtaining data on a global scale, enormous efforts have been made to develop algorithms to retrieve ice-thickness distribution. For relatively thin ice (<0.2 m), the thickness algorithms for coastal polynyas in Arctic and Antarctic regions, combined with US National Oceanic and Atmospheric Administration (NOAA) Advanced Very High Resolution Radiometer (AVHRR) data, were developed (Reference Martin, Drucker, Kwok and HoltMartin and others, 2004; Reference Tamura, Ohshima, Markus, Cavalieri, Nihashi and HirasawaTamura and others, 2007). For thick ice (>1 m), the measurement of the freeboard has proven to be effective with the assumption of isostatic balance, from field observations in the Arctic (Reference Comiso, Wadhams, Krabill, Swift, Crawford and TuckerComiso and others, 1991) and from satellite altimetry (radar (Reference Laxon, Peacock and SmithLaxon and others, 2003) and laser (Reference Kwok, Zwally and YiKwok and others, 2004, for the Arctic; Reference Zwally, Yi, Kwok and ZhaoZwally and others, 2008, for the Antarctic)). This allowed estimation of the seasonal and interannual variation of ice volumes in the Arctic Ocean (Reference Kwok and CunninghamKwok and Cunningham, 2008). To reduce the uncertainty caused by snow, Reference KurtzKurtz and others (2009) attempted to combine laser altimetry and passive microwave radiometry. However, while the algorithm for level thin ice or thick ice has been explored, relatively thick sea ice (0.2–1 m) in the seasonal ice zone (SIZ) remains a big issue.

In the SIZ, the degree of surface roughness is expected to be a useful surface property related to ice-thickness distribution because deformation processes in the SIZ, usually accompanied by surface roughness, play an essential role in the development of ice thickness except for landfast sea ice (Reference Worby, Jeffries, Weeks, Morris and JaWorby and others, 1996; Reference Toyota, Takatsuji, Tateyama, Naoki and OhshimaToyota and others, 2007). To extract surface roughness as discussed by Reference Dierking and BuscheDierking and Busche (2006), L-band synthetic aperture radar (SAR) (0.15–0.30m wavelength) data appear to be better suited than the C-band (0.03–0.07m wavelength) data mainly used from satellite SAR so far. This is because L-band is closer to the horizontal scale of surface roughness (~tens of cm to a few meters) than C-band (Reference Lubin and MassomLubin and Massom, 2006). To verify this idea, a validation experiment was conducted in the Sea of Okhotsk in February 2005, in coordination with airborne polarimetric and interferometric SAR (Pi-SAR), and promising results were obtained (Reference Toyota, Nakamura, Uto, Ohshima and EbuchiToyota and others, 2009). Here we further examine the feasibility of spaceborne L-band SAR (Advanced Land Observing Satellite (ALOS)/Phased Array-type L-band SAR (PALSAR)) with coarser resolution as a tool for estimating ice-thickness distribution in the SIZ. To do this, we conducted concurrent measurements of surface roughness and ice thickness, coordinated with the PALSAR orbit in February 2008. Although the relationship between ice thickness and airborne L-band SAR backscatter data has been presented before (e.g. Reference Nakamura, Wakabayashi, Naoki, Nishio, Moriyama and UratsukaNakamura and others, 2005; Reference Kern, Gade, Haas and PfafflingKern and others, 2006), this is the first trial to validate it with surface roughness data and apply it to satellite L-band SAR. the main aim of this experiment is to improve the estimation of surface roughness by means of helicopter observations and then to examine the relationship of backscatter data with surface roughness and ice thickness, and the feasibility of spaceborne L-band SAR for estimating the ice-thickness distribution in the SIZ.

Methods

Measurements

The in situ observations were carried out on board Soya along a longitudinal line ~40km long from 143.925^˚E to 144.375^˚ E at 45.2^˚N in the southern Sea of Okhotsk on 12 February 2008 (Fig. 1). the sea-ice extent in the Sea of Okhotsk (110.69 ×10⁴ km²) was close to normal (105.08×10⁴ km² for the period 1971–2000) according to the Japan Meteorological Agency (Fig. 1a), and it is seen in Figure 1b that variable ice conditions were present in the study area.

Fig. 1.

Ice conditions and location map of the observation on 10 February 2008. (a) Sea-ice extent in the Sea of Okhotsk (based on Japan Meteorological Agency information (http://www.data.kishou.go.jp/kaiyou/db/seaice/okhotsk/okhotsk_extent.html)). (b) MODIS image within the frame of (a) with the observation line (thick solid line).

Ice thickness was measured with the ship-borne monitoring systems of both video and EM. the EM system was the same as used for the Pi-SAR experiment in 2005, while the video monitoring system measurement was done for ice floes which were broken at the bow and turned into side-up positions (Reference Toyota, Kawamura, Ohshima, Shimoda and WakatsuchiToyota and others, 2004). the measurement accuracy of the EM system is ±0.1m for undeformed ice (Reference Haas, Thomas and DieckmannHaas, 2003) and 10% for deformed ice in this region (Reference Toyota, Nakamura, Uto, Ohshima and EbuchiToyota and others, 2009). As for the video system, although the measurement accuracy itself is less than a few centimeters, the problems are its representativeness and subjectivity because this is a sporadic measurement. According to Reference Toyota, Kawamura, Ohshima, Shimoda and WakatsuchiToyota and others (2004), a 10 min running mean (~3 km) gives a general trend of ice thickness that can be regarded as representative. Moreover, since they showed that the difference in mean values between observers is only a few centimeters, it seems to be free from subjectivity. Therefore, the accuracy of this method is considered to be a few centimeters when the 10 min averaged data are concerned. Since the EM system was not working that day, the video monitoring data were used for validation. It was confirmed that the video data coincide with the EM data for the Okhotsk sea ice except for significantly ridged ice (Reference Toyota and JedlovecToyota, 2009), so they are deemed a useful alternative to EM for validation. the total number of samples along the observation line (Fig. 1b: ~35 km) was 131 (one per 0.27 km on average).

A helicopter-borne laser profiler (RIEGL, LD90-3100HS) was used to improve the accuracy of surface roughness, as suggested by Reference Toyota, Nakamura, Uto, Ohshima and EbuchiToyota and others (2009). As a helicopter observation is free from the effect of ice surface conditions, we can expect it to become easier to discern surface topography from the vertical motion of the helicopter in the time series of the profiles. Due to cloudy conditions at the observation time of PALSAR (1004 h local sidereal time (LST)), the helicopter operation was started at 1400 h. For the measurement, a downward-looking profiler was attached to a pole extended through the door of the helicopter. the helicopter flew along the cruise track at an altitude of ~50m (Fig. 4, further below), and the profiler recorded the height above the surface at 10 Hz (i.e. 2m spatial resolution). During the flight, surface conditions beneath the helicopter were also monitored with a high-resolution video camera through the other side of the door. the images obtained were used to distinguish between sea-ice and open-water areas. In fact, whereas a heliborne profiler measures the roughness of the snow surface, L-band SAR backscatter corresponds to the snow/ice interface roughness. Even so, our measurement is still valid since Reference Andreas, Lange, Ackley and WadhamsAndreas and others (1993) reported that snow surface roughness is strongly correlated with snow/ice interface roughness for Antarctic sea ice.

Analytical procedures

The distribution of PALSAR backscatter coefficient is shown in Figure 2a, where the satellite path is on the eastern side of the image (north-northeast to south-southwest) and the incident angle ranges from 18^˚ to 43^˚ from near to far. the figure indicates that backscatter depends significantly on the incident angle. Figure 2b shows the profile of backscatter coefficients along the A–B cross section in Figure 2a with a regression line derived by a least-square method. To remove the bias caused by the incident angle, the backscatter data at each pixel were normalized to that at the incidence angle (30.4^˚) of the center of the image using the regression obtained in Figure 2b.

Fig.2.

PALSAR backscatter coefficients observed on 12 February 2008. (a) Geographical distribution in the study area (© JAXA). Square area corresponds to the region depicted in Figure 4. (b) Profile of backscatter coefficient along the A–B cross section in (a), crossing the center of the image. the regression line is shown with a dashed line.

In estimating surface roughness from the laser data, we adopted the same procedure as was used for the Pi-SAR experiment. To extract surface elevation from airborne laser profiles in the polar pack ice, the method of Reference HiblerHibler (1972) which determines a sea level by connecting minimum points is often used (e.g. Reference Comiso, Wadhams, Krabill, Swift, Crawford and TuckerComiso and others, 1991). However, in our case where surface topography is relatively small and sometimes comparable to the measurement accuracy (~a few centimeters), our statistical procedure seems to work more effectively. Although the sampling interval (~2m) for heliborne measurement is coarser than the interval of 0.5 m for shipborne measurement, it is found that the accuracy was much improved. an example is shown in Figure 3. Since the variation of height due to the heli-operation had quite different scales from that due to surface topography in both magnitude and time (Fig. 3a), low-pass filtering worked more effectively to extract surface topography compared with the shipborne profiles. the estimated surface elevation ranged from 0.1 to 0.5 m (Fig. 3b) and seems to match well with the ridge sail height estimated by visual observation from the ship.

Fig. 3.

Example of the time series of heliborne laser profiling data, obtained from 1408 to 1409 h on 12 February 2008. (a) Raw data (b) Estimated surface elevation.

The difference in observation time among SAR, ice thickness and surface roughness is also an important issue since in this area the southward ocean current is as high as a few tens of centimeters per second (Reference Ohshima, Wakatsuchi, Fukamachi and MizutaOhshima and others, 2002) and ice conditions are variable (Fig. 1b). Along the observation line, the shipborne and heliborne measurements were conducted from 0700 to 1600 h and from 1401 to 1445 h LST, respectively. Thus the maximum time difference is 6 hours, corresponding to several kilometers. To correct this, the ice-drift speed was estimated using the video images monitored from the helicopter. We checked the location and time of the ship trails which appeared in the video images and calculated their difference from the shipborne GPS record. the estimated ice-drift speed was (u, v) = (0.163, –0.188)ms^–1 near the center of the observation line and (u, v) = (0.178, –0.196)ms^–1 near the eastern edge. Since these two values are very close, the averaged value (0.17, –0.19)ms^–1 was used, assuming that the drift speed was invariant. the drift speed due to wind forcing is estimated as (0.13, –0.02)ms^–1, assuming that sea ice drifts at a rate of 2% of wind (Reference Kimura and WakatsuchiKimura and Wakatsuchi, 1999). Therefore the ocean current is estimated as (0.04, –0.17)ms^–1, which almost coincides with the value reported by Reference Ohshima, Wakatsuchi, Fukamachi and MizutaOhshima and others (2002). With this ice-drift speed and the difference of time at each point, we located the pixels on the backscatter map that should be used for validation (Fig. 4).

Fig. 4.

The pixels of ice thickness (solid black line) and surface roughness (solid red line) used for validation, shown on the map of the backscatter coefficient distribution normalized to the incidence angle of 30.4^˚ (© JAXA). the tracks of ship and helicopter are also depicted with dashed black and red lines, respectively. the region depicted corresponds to the square area in Figure 2a.

Results

To compare SAR backscatter with ice thickness and surface roughness along each observation line in Figure 4, we sectioned these observation lines into segments every 0.025^˚ in the longitudinal direction (~2 km), and averaged the data for each segment. Although the length of segment might be too large compared with the horizontal resolution of SAR, we selected this value so as to contain as many thickness data as possible for representativeness. of the 18 segments in total, 4 segments where the number of ice-thickness data was less than three and the 2 segments where open water dominated were excluded from the analysis of ice thickness. As for the surface roughness, the 4 segments where open water dominated were excluded. As a result, it was found that backscatter has a significant correlation with both ice thickness and surface roughness at >99% level (Fig. 5). the correlation coefficient is 0.86 between ice thickness and backscatter, and 0.70 between backscatter and surface roughness. From Figure 5a the regression line is estimated as

Fig. 5.

Scatter plots of the values averaged for 0.025^˚ segments between (a) normalized backscatter coefficient at HH polarization (incidence angle 30.4^˚) and ice thickness with a regression line (r = 0.86, significant at >99% level); and (b) normalized backscatter coefficient at HH polarization (incidence angle 30.4^˚) and surface roughness (r = 0.70, significant at >99% level).

(1)

with a root-mean-square (rms) error of 0.04 m, where H _i is the ice thickness (m) and BS is the normalized backscatter coefficient at HH polarization (dB). Since the rms error is comparable to the measurement accuracy of the video system, the error from the regression can be explained mainly by the measurement error, indicating the validity of the regression. This result supports our hypothesis that the L-band SAR data correspond well with surface roughness and therefore L-band SAR has a good correlation with ice thickness. However, it should be noted that H _i in Equation (1) is the thickness of less deformed ice due to the measurement method. To estimate the ice thickness including significantly deformed ice, the slope of the regression may be somewhat larger, especially for ice thicker than 0.4 m where the ridging activity begins to become significant (Reference Toyota, Takatsuji, Tateyama, Naoki and OhshimaToyota and others, 2007). the lower limit for Equation (1) seems to be ~0.2m since below this value the ice-thickening processes are less associated with surface roughness.

Discussion and Conclusion

Our validation field experiment, coordinated with the spaceborne ALOS/PALSAR overpass, successfully showed that the L-band SAR backscatter coefficients (HH) with a horizontal resolution of 100m and those obtained with a horizontal resolution of 3m by the airborne Pi-SAR experiment are similarly correlated with ice thickness and surface roughness. This indicates that satellite L-band backscatter data are a promising tool to estimate ice-thickness distribution in the SIZ, where surface roughness is closely related with the ice-thickness distribution through deformation processes. Additionally, it was found that heliborne laser profiling can extract surface topography more effectively than a shipborne measurement. Since it has been shown that sea-ice growth processes in this area are similar to those in the Antarctic seas (Reference Toyota, Takatsuji, Tateyama, Naoki and OhshimaToyota and others, 2007), we expect that our results can be applied on the Antarctic sea ice.

To see how the satellite L-band SAR data can represent the ice-thickness distribution in this region, regression (1) is applied to the PALSAR data obtained in mid-February from 2007 to 2009 (Fig. 6). For each image, the dependence of backscatter data on incident angle was calculated individually in the same way, and then the data normalized to the center of the image (30.4^˚) were supplied to Equation (1). In each figure, the Advanced Scanning Microwave Radiometer (AMSR)-derived ice concentration data were used to extract sea-ice area. We defined the area with ice concentration greater than 15% as sea-ice area, and depicted ice-thickness distribution only for the sea-ice area. Figure 6 shows that in each year ~0.2–0.6m thick sea ice is dominant and that ice thickness appears to be relatively thicker in 2008 than in 2007 and 2009. the calculated average thicknesses in the southern area (44.3–45.5^˚ N, 142–145^˚E) are 0.33±0.12, 0.42±0.11 and 0.37±0.11 m. the histogram of ice thickness in each year shows these features more clearly (Fig. 7). This result almost coincides with the observational results. the monitoring of ice-thickness distribution has been conducted with the same video system in the wide area of the southern Sea of Okhotsk in every February since 1996. According to the results of this observation, the average thicknesses are 0.43 ±0.16, 0.42±0.19 and 0.24±0.16m in 2007, 2008 and 2009, respectively, indicating significantly thinner ice in 2009. the reason why PALSAR-derived ice thickness in 2007 is much smaller than the observational result may be in part that the thicker area in the eastern side was outside the range in the PALSAR image. Thus PALSAR can be expected to be useful for monitoring the interannual variation of ice thickness, especially for less deformed ice. In addition, characteristic patterns are brought into relief by these figures. on a large scale the thick developed ice extends southward from east of Sakhalin, while on a small scale ice bands near the ice edges in each year and patches of ~0.8m thick ice floes in the western part in 2008 can be found. These properties will serve to make the ice-thickening processes in each area better understood. We conclude that the retrieval of ice-thickness distribution from PALSAR data seems to be successful in this area.

Fig. 6.

PALSAR-derived ice-thickness distributions for (a) 14 February 2007, (b) 12 February 2008 and (c) 14 February 2009 (# JAXA). Only the sea-ice areas (AMSR-derived ice concentration >15%) are depicted.

Fig. 7.

PALSAR-derived ice-thickness distributions on 14 February 2007, 12 February 2008 and 14 February 2009.

However, some problems still remain unsolved. One is the effect of open water. Under turbulent conditions, the ocean surface becomes rough and also has high backscatter coefficients even within the sea-ice area. In such a case it is difficult to distinguish between open water and the ice surface with only single polarization SAR data. While in this study we could do this with a video camera, additional information would be required to apply our result to PALSAR data operationally. Moreover, when we extend this result to the polar region, the presence of multi-year ice should be taken into account. Multi-year ice has significantly different properties, in particular in the Arctic. Due to desalination in summer, the penetration depth increases and then the volume scattering becomes more significant at L-band frequencies. In this case, it is known that C-band is more appropriate for distinguishing between multi-year ice and deformed first-year ice (Reference Rignot and DrinkwaterRignot and Drinkwater, 1994). Therefore, the combination of C- and L-band is desirable. This future work is expected to be accomplished by using L-band SAR effectively together with other satellite sensors.