3.1. Development of the algorithm

In winter 1996, a field sea-ice observation experiment, the Sea Ice Observation Program for the Sea of Okhotsk (SIPSO), was carried out. This experiment used a National Space Development Agency of Japan (NASDA)-developed Airborne Microwave Radiometer (AMR) mounted on a Beach-craft-200 airplane (B-200) to collect surface measurements in Lake Saroma, northern Hokkaido, Japan, and the Sea of Okhotsk. The AMR has 6.925, 10.65, 18.7, 23.8, 36.5 and 89.0 GHz channels, each of which can measure vertical and horizontal polarization. This instrument was developed for a ground experiment related to the Advanced Microwave Scanning Radiometer (AMSR) on the NASDA ADEOS-II satellite, which will be launched in 2001. A videotape recorder (VTR) camera and an infrared radiometer were also mounted on the airplane.

Lake Saroma is a 150.4 km² salt lagoon connected to the Sea of Okhotsk by two estuaries, and freezes up entirely with about 30 cm thick ice. Reference Tateyama, Enomoto, Takahashi, Shirasaki, Hyakutake and NishioTateyama and others (2000) described the development of a new algorithm based on the SIPSO experiment, the SIPSO/Kitami Institute of Technology (S/KIT) algorithm. The VTR images were taken by the B-200 and used as the distribution data for the ice-surface conditions. From the VTR images the ice-thickness distribution was estimated by observing the surface color, size, snow cover and edge ridging.

The distribution of the brightness temperature (T_B) of the vertically polarized 89.0 GHz channel (T_B89V) of the AMR varied with ice thickness and the ice surface. This variation was from about 235 K on snow-covered thick ice to about 270 K on the bare thin ice. In contrast, the T_B from the vertically polarized 36.5 GHz channel (T_B37v) was 250−265 K. T_B89V shows similar results to the infrared measure of surface ice temperature. T_B37V comes from about 3 cm depth and shows a constant value compared to T_B89V. T_B89V cannot differentiate between open water and ice, as it indicates almost the same brightness temperature because the emissivity of open water rises with frequency. Hence, the contrast between sea ice and open water in the 85 GHz channel is too small to differentiate them. To obtain higher spatial resolution, to discriminate among ice types according to thickness and to detect open water, a designated ratio between T_B37V and T_B89V from the AMR, R_37V/89V, was used. When the R_37V/89V is low in the packed-ice area, the temperature of ice in the upper layer is assumed to be low, meaning that the ice is thick, and vice versa. Although T_B89V cannot detect open water, the R_37V/89V can clearly distinguish between open water and ice by using T_B37V, which is sensitive to the different dielectric constant between the ice and water. This difference between T_B37V and T_B89V in open water is larger than that obtained using T_B37V or any other single channel, so the R_37V/89V is better at finding open water.

3.2. Ice-classification ability

Figure 2 shows the results of ice classification using AMR observations for the area covered by new ice, young ice and floe, off the eastern coast of Hokkaido. Figure 2a uses the NASA Team algorithm PR GR plane (Reference CavalieriCavalieri and others, 1991); Figure 2b uses R_37V/_89V instead of GR. PR and GR are given by

Fig. 2. Ice-signal distribution, taken by AMR on 17 february 1996 in the sea of okhotsk, (a) the nasa team algorithm (PR-GR plane); (b) PR-R_37y/8v plane.

PR is not only a good water/ice discriminator, but is also largely independent of the physical temperature from the radiating medium since it is the ratio of the observed radiance. GR is defined as a discriminator of ice type by using the differences of slope between the T_B at 36.5 and 18.7 GHz. PR mainly gives ice concentrations. GR gives a fraction between the two ice types, which are first-year ice and multi-year ice in the Arctic. In the Antarctic, this fraction consists of first-year ice and heavy snow-covered ice (Reference Cavalieri, Gloersen and CampbellCavalieri and others, 1984; Reference Markus, Cavalieri and JeffriesMarkus and Cavalieri, 1998). The NASA Team algorithm in the PR-GR plane is represented by a curvilinear triangle whose vertices are defined by tie points, which consist of the T_B values for open water, first-year ice and new ice (Reference CavalieriCavalieri, 1994).

The results show that it is hard to distinguish young ice from floes by using the PR-GR plane (Fig. 2a). When R37V/89V is used instead of GR, discrimination between ice floes and young ice is possible by using the R_37V/89V = 1−00 line (Fig. 2b). Young ice and newly formed ice can be divided by R37V/89V = 0.92, and new ice and open water can be divided by R37V/89V = 0.86. The new ice, young ice and floes have respective thicknesses of < 10,10−40 and >40 cm.

3.3. Comparison with ship-based measurements

Ice and oceanographic observations were carried out by the Institute of LowTemperature Science, Hokkaido University, and the Ship Research Institute, Ministry of Transport, using the Japanese Coast Guard ice-breaker soya off the eastern coast of Hokkaido in February 1996 and 1997 (Reference Toyota, Ukita, Ohshima, Wakatsuchi and MuramotoToyota and others, 1999). Albedo, ice concentrations, ice thicknesses and snow depths were estimated by video analysis. The ice-thickness parameter R_37V/89V derived from the SSM/I data with the S/KIT algorithm was validated using the ship-based ice-measurement data. Figure 3a shows the relationship between the ship-based measurements and R_37V/89V. The horizontal axis is the R₃7V/89V algorithm, and the vertical axis is the ship-based ice measurements. The ship data were converted to the stereographic grids for SSM/I, and averaged about 62 samples for every one of the gridcells. The histogram for ice depths derived from the ship’s observations ranged from 18 to 82 cm, the most frequent range being 60−70 cm thick and the second most frequent 30−40 cm thick. There were no samples of new ice 1−10 cm thick from these ship data. For the thinner-ice validation, AMR experimental data and satellite images were used (Reference Tateyama, Enomoto, Takahashi, Shirasaki, Hyakutake and NishioTateyama and others, 2000). The results are indicated by the solid box in Figure 3.

Fig. 3. The ship-based ice thickness vs thickness parameters. (a) R_37V/85V of the s/KIT algorithm; (b) pr of the MNT algorithm.

The effect of snow cover over sea ice on the estimation of ice thickness cannot be neglected, because the 85 GHz channel is sensitive to surface changes. From ship observations the snow cover normally does not exceed 20% of ice thickness in large ice-drifting regions. On the thin ice, snow will be blown away by wind or melted by surface brine. The snow cover was taken into account in an emphasis factor of ice thickness in this study. In other words, snow cover is deep when ice is thick, and vice versa.

Scatter plots are shown in Figure 3. Black dots and gray triangles indicate 100% and 80−99% packed-ice samples in a concentration calculated by the tie-points modified NASA Team thin-ice (MNT) algorithm for the Sea of Okhotsk (Reference EnomotoEnomoto, 1996). The areas surrounded by boxes indicate ice-type classification from the S/KIT algorithm. From this result, ice classifications were summarized so that floe ice, which ranged from 35 to 85 cm in the ship-based ice measurement, was classified between 1.00 and 1.10 in the R37V/89V algorithm. Young ice 11−35 cm thick was scattering in 0.97−0.99. Ice 30−40 cm thick tended to indicate concentrations of 80−99%.

Figure 3b shows the relationship between the ship-based measurements and the PR. Ice types were categorized for PR ranges of 0.12−0.17, 0.05−0.12 and 0.02−0.05 for new ice, young ice and thick first-year ice, respectively (Reference CavalieriCavalieri, 1994). The PR variation tends to classify the thin-ice type as "loose packed ice". The ice classifications for 100% concentration were improved by using the thickness parameter R_37V/89V, although there were a few mis-categorized points.

For application of the S/KIT algorithm to the SSM/I datasets over the whole Sea of Okhotsk, visible and near-infrared images from the Advanced Very High Resolution Radiometer (AVHRR) on board the U.S. National Oceanic and Atmospheric Administration (NOAA) satellite and Advanced Visible and Near Infrared Radiometer (AVNIR) images from the Advanced Earth Observation System (ADEOS) satellite were used for tuning the threshold values of ice types (Table 1). The S/KIT algorithm has a higher resolution of 12.5 × 12.5 km₂, and the data contamination caused by "weather and land effects" was reduced as described in the next subsection.

3.4. Weather and land effects

It is necessary to remove contamination from the microwave ice signals. This contamination of the microwave sea-ice signal is due to water vapor, clouds, rainfall and snowfall. A "land-proximity effect" also causes a false ice signal because of the side lobe of the antenna pattern when a field of view includes land. This has a serious effect on observations of the coastal sea-ice distribution. Although it is possible to remove the atmospheric-signal contamination by using weather filters, it is impossible to separate the land contamination from the ice signal merely by means of an algorithm, because of the side lobe and its spatial resolution. However, the area affected by land noise can be reduced by increasing the resolution.

This study uses a weather filter suggested by J. C. Comiso (personal communication, 1994), but with a slightly lower value in the following threshold. The concentration is to be set at 0% when

The weather filter for the Sea of Okhotsk, suggested by Reference Cho, Sasaki, Shimoda, Sakata and NishioGho and others (1996), is also used. If GR >0.05 and GR’ > 0.03, where GR’ is given by

then this concentration is substituted for 0%. The T_B of the vertically polarized 22 GHz (T_B22V) channel is used to measure atmospheric effects because this channel is located in an absorption band of the atmosphere.

A significant reduction in land contamination was obtained by increasing the resolution. Figure 4 compares the seasonal ice extent estimated by the MNT algorithm and the S/KIT algorithm for the Sea of Okhotsk during January-December 1996. Since there is no sea ice between June and the beginning of November, ice signals in this period are always false. The false ice signals from the MNT algorithm in the summer reach 10% of the average winter ice area, whereas the S/KIT algorithm shows < 3% false signals. Especially in summer, the S/KIT algorithm has less false ice, with a mean value of 60% fewer and a maximum value of 80% fewer false signals than the MNT algorithm. Even in December and May, when there are false signals due to the land-proximity effect, the S/ KIT algorithm reduces these by applying a higher-spatial-resolution channel, while confirming the ice signal with the pixels of the lower-spatial-resolution channel (i.e. 37 GHz) along the coast and ice edge.

Fig. 4. A tine series of sea-ice extent in the sea of okhotsk during 1 january-31 december 1996 (10 days interval). the solid line shows the S/KIT algorithm, and the dashed line shows the MNT algorithm.

These weather filters (Equations (3) and (4)) are not designed to differentiate between sea ice and weather contamination over sea ice. Therefore our ice products are assumed to be free of weather effects. Despite the absence of effective filters for weather contamination over sea ice, pending further study, the S/KIT-algorithm-derived data appear to be more reliable. A considerable number of ice maps were examined using the S/ KIT algorithm comparing cloud-cover information derived from NOAA AVHRR satellite images, but there was little apparent weather effect over sea ice. Sea-ice-type signals were quite stable in the successive daily images, although skies were frequently overcast.

4.1. Calculation of ice thickness and ice volume

The SSM/I data are available from 1987, but the 85 GHz channels were too noisy to use between 1989 and 1991. Ice-thickness and −volume estimates using the S/KIT algorithm were carried out from December 1991.

Ice volume was calculated as the product of the ice thickness and ice-covered area, which has been defined as grid-cells containing ice weighted by the ice concentration. Ice of several thicknesses was assumed from the parameter R37V/85V. This parameter varies with ice types as shown in Table 1. Fast ice, which appears to have a smooth surface and is a brighter white than ice floes in AVHRR images (e.g. around the estuary of the Amur River), was defined as 86−120 cm thick. Ice thickness h (cm) is calculated using the following equation:

where h₀ is the minimum thickness of each ice type. New ice, young ice, floes and fast ice have h₀ values of 1, 11, 35 and 86 cm thick, respectively. These minima are derived from ship-based measurements. c is a constant that is calculated by the range in ice thicknesses divided by R₃₇v/85V (maximum) minus the threshold values TH of each of the ice types. TH is from the lower threshold value of R₃₇v/85V in Table 1. In the case of floe ice, for example, h₀, c andTH are 35, 454.5 and 1.00, respectively. For ice of low concentrations, thicknesses ranged from 1 to 50 cm because low concentrations of ice can always be seen in the marginal ice zone. In this area, sea ice consists of new ice and crushed ice, so ice thicknesss can generally be regarded as thin. The contribution of areas of low concentration to total ice-covered area is low in the maximum ice-covered season.

Total ice volume V_total was calculated by

where A_local and h_local are ice area and ice thickness, respectively, in each of the gridcells.

Spatial mean ice thickness is calculated by dividing daily total ice volume by daily ice area. This value indicates the average ice thickness over the whole study area.

4.2. Time series of ice area, thickness and volume

Table 2 shows the yearly mean ice area, thickness and volume as well as the maximum two and minimum two measures for this study period. Intra-seasonal fluctuations of ice area, thickness and volume are shown in Figures 5–7, respectively. These figures show the daily ice products given by the S/KIT algorithm over the study period.

Fig. 5. Daily sea-ice area during 1992−99 winters (december-april) in the sea of okhotsk and the bering sea.

Fig. 6. Daily spatial mean sea-ice thickness during 1992−99 winters (december-april) in the sea of okhotsk and the bering sea.

Fig. 7. Daily sea-ice volume during 1992−99 winters (december-april) in the sea of okhotsk and the bering sea.

In the Sea of Okhotsk the two years with the minimum yearly area average were 1996 and 1997, with 62.9% and 68.6%, respectively, of the maximum yearly area average (1998). In the Bering Sea the minimum winter average area was observed in 1996 and was 68.6% of the maximum winter average area (1995). Differences between the largest and smallest ice volume were more marked. In the Sea of Okhotsk the winter average volumes in 1996 and in 1997 were 52.2% of the maximum winter average volumes in 1993 and 1995. In the Bering Sea the ice volume for 1996 was the smallest and was 62.5% of the maximum in 1995.

Except for 1995 and 1997 the daily ice areas of the Sea of Okhotsk were larger than those of the Bering Sea, but the winter average areas of the Sea of Okhotsk were always 10−30% smaller than those of the Bering Sea. Mean winter ice thicknesses in the Bering Sea were always thicker than in the Sea of Okhotsk. The winter average ice thickness during this period in the Sea of Okhotsk was 0.28 m. In 1996 and 1997 the spatial mean ice thicknesses were thin, 0.22 and 0.23 m, respectively. In 1998, although the winter average ice area was the largest for the period of study, the mean ice thickness was not thick, with the result that the winter average ice volume was not the largest. Likewise, in the Bering Sea, the winter average ice volume in 1998 was below average, although the ice area covered was the second largest, because the winter mean ice thickness was the thinnest. These results suggest that ice area and ice thickness do not vary together but can vary out of phase. One possible reason for this is that ice area is much affected by wind, while ice thickness is much affected by surface air temperature, but this must await further study.

The intra-winter variation was greater for ice thickness and volume than for variation in ice area. A negative correlation can be seen between these two seas. This out-of-phase "see-saw" response showed the two different time-scales, which are short (1 week) and long (2−4 weeks). The short-time-scale variations, which look like negative and positive spikes, are thought to be caused by the movement of low-pressure weather systems. A low approaching the ice-covered area may make the ice area smaller and looser due to warm-air advection (negative spike), and a departing low may make it larger due to cold-air advection (positive spike). Ice thickness clearly showed not only a short- but also a long-term variation, which may be caused by changes in the strength of the Aleutian low. The ice area and thickness is smaller and thinner in the Sea of Okhotsk when the Aleutian low strengthens (Reference Tateyama and EnomotoTateyama and Enomoto, 2000).