1. Introduction
In recent years, global climate change has drawn increasing attention [
1]. Accelerated melting of the Arctic sea ice has led to a decline reflection of the Sun, allowing more solar radiation to be absorbed by the Arctic Ocean, further accelerating the ablation of the Arctic sea ice, and thus climate feedback. This “Arctic amplification” effect makes the Arctic Ocean a sensitive area influencing local and global climate changes [
2].
Remote sensing is an important tool to monitor sea ice over a wide region. Passive microwave (PM) remote sensing can distinguish between sea ice and sea water, which are widely used to study the variation of sea ice extent at the Arctic scale [
3,
4,
5,
6]. Sea ice concentration (SIC) describes the relative amount of area covered by ice when compared to a satellite grid of a certain size [
7]. However, a PM SIC threshold is needed to define whether a satellite cell is “ice-covered” or “open water”, to calculate the sea ice extent for trend analysis. For data cells with a SIC greater than or equal to the representative threshold value, the cell is regarded as an ice cell, while the others are excluded as open water.
A typical SIC threshold is 15%, which was first used by Parkinson et al. to define the location of the sea ice edge for measuring the Arctic sea ice extent [
3]. According to the 15% SIC threshold, the sea ice extent in the northern hemisphere was analyzed over the period 1973–1987 using data derived from the Nimbus 5 Electrically Scanning Microwave Radiometer (ESMR) and the Nimbus 7 Scanning Multichannel Microwave Radiometer (SMMR) [
8]. Cavalieri et al. followed the use of the 15% special sensor microwave imager (SSM/I) SIC threshold as the threshold of the location of sea ice edge for mapping Arctic sea ice [
9]. However, Remund and Long compared the Ku band NSCAT scatterometer data with SSM/I radiometer derived SIC. They found that the NSCAT resulting edge closely matched the SSM/I derived 30% ice concentration edge [
10]. Worby and Comiso also questioned the reliability of the sea-ice edge derived from 15% SIC after comparing Antarctic sea ice edge locations observed by SSM/I, ships, and a variety of remote sensing images [
11].
In addition, the SIC threshold at the sea ice edge used in the released sea ice products is not always consistent. A threshold of 15% is used in the product of “Sea Ice Index” [
12], and the product of “Sea Ice Trends and Climatologies from SMMR and SSM/I-SSMIS” [
13]. However, some sea ice products used different SIC threshold values where a value of 20% is used in the product of “Arctic Sea Ice Freeboard and Thickness” [
14], 30% is used in the product of “March through August Ice Edge Positions in the Nordic Seas” [
15], and in the product of “CryoSat-2 Level 4 Sea Ice Elevation, Freeboard, and Thickness” [
16], 35% is used in the product of “Global Products for Ice Concentration, Ice Edge, Ice Type, Sea Ice Drift” [
17].
In these contexts, this paper focused on exploring a representative ice-water discrimination threshold for the SSMIS (with a spatial resolution of 25 km) in the Arctic in recent years.
Section 2 describes the data used for comparing the SIC at the Arctic sea ice edge.
Section 3 gives an overview of the sea ice edge extraction steps from satellite and ship-based sea ice observations. The comparison results are given in
Section 4 and discussed in
Section 5. The final section contains the conclusions drawn from the analysis.
5. Discussion
Currently, much attention has been paid to the variation and the trend of Arctic sea ice using passive microwave remotely sensed data [
28,
29], but studies on sea ice concentration thresholds based on these data were reported in a few cases [
10,
30]. The objective of this study was to explore the representable value of the PM SIC threshold corresponded on average to the position of the Arctic sea ice edge during summer in recent years. For this connection, the MODIS sea ice product (MOD29), MODIS images, as well as ship observational data from different Arctic marginal sea regions and different years were used to extract Arctic sea ice edges, and conduct an overlay analysis of the responding PM sea ice concentration threshold in the context of a rapid changing Arctic.
Overall, the thresholds obtained from the comparisons were rather similar; different comparison results indicated that the representable threshold value of Arctic sea ice edge during the summer was higher than the well-known 15% (
Figure 5,
Table 4 and
Table 5). The mean threshold of 30% calculated in this study was consistent with that in [
30], which compared NSCAT scatterometer data with radiometer derived SIC. The higher threshold value can be attributed to the following reasons. First, during the summer season, PM SIC at the ice edge with course resolution may rise from 0% at one pixel to 100% at its neighboring pixel, which result in mean SIC at the ice boundaries extracted from MOD29, MODIS images, and ship-based observational data were high at 30%, even at 40% due to the mixed pixel problem, especially for the declining and thinning Arctic sea ice in recent years. Second, the image of MODIS was a snapshot of the day, and the ship-based observations were at an hourly time-scale, however, the PM SIC was the daily average product. This may have caused the mismatch of the sea ice edges, especially on windy days when the sea ice edge moves at a high speed.
Although the threshold values from the MOD29 and MODIS images were both higher than 15%, MOD29 (36% on average) was generally higher than that from the corresponding MODIS images (35% on average).
Figure 6 is the MOD29 sea ice product overlapping the sea ice edges interpretation from the MODIS image. From the purple line, which is the sea ice edge visual interpretation from the MODIS image, and the white, the blue, and the yellow parts which present the sea ice, seawater, and cloud from the MOD29 data, respectively, we can conclude that the MOD29 sea ice product would misjudge part of the seawater as sea ice, which results in a higher sea ice concentration at the sea ice edge corresponding to that extracted from the MODIS image.
The correlation between the SIC of the ship-observed sea ice edge points and the corresponding SSMIS was moderate, which may be due to the following factors. First, there were some errors in the SSMIS SIC due to the SIC retrieval algorithm, as well as the existing errors for the sea ice ship-based observations due to the influence of in-situ human observations. Second, the spatial resolution between the SSMIS and the ship-based observation SIC was different as the resolution of the SSMIS data was 25 km and the ship observational data were 1 km.
It should be noted that different SIC algorithms, with the spatial resolution of each channel that enters into these SIC algorithms may influence the threshold results. Although Comiso and Parkinson [
31], Heygster et al. [
32], and Beitsch et al. [
33] all reported that there were not significant biases of derived SIC among the different algorithms and sensors, however, these would be of great value to study in depth. We made the statistical analysis of SSMIS SIC based on the Bootstrap algorithm [
34] and the ASI algorithm [
35] at the ice-water boundary lines from a visual interpretation of the MODIS images. Comparing the results in
Table 4 and
Table 6, we found that the different SIC algorithms and different spatial resolution (e.g., 6.25 km for the ASI algorithm using a high-frequency channel) had some influence. However, the representable values of the PM SIC thresholds corresponding on average to the position of the Arctic sea ice edge were rather similar, which were all higher than the commonly used threshold of 15%.
Since a representable SIC threshold of 15% or 30% would influence the exact Arctic sea ice extent, we compared the difference of the calculated Arctic sea ice extent based on a 15% and 30% SIC threshold for both the NT SIC algorithm and Bootstrap SIC algorithm with a spatial resolution of 25 km for SSM/I and SSMIS data, and the ASI SIC algorithm with a spatial resolution of 6.25 km for AMSR-E and AMSR2 PM data.
Figure 7 shows that there was a sea ice extent difference when using a 15% SIC threshold and 30% SIC threshold, especially during the summer season.
The maximum differences of the calculated sea ice extent based on 15% and 30% SIC thresholds were 10.33 × 105 km2 and 6.24 × 105 km2 for the NT and Bootstrap SIC algorithms, and 3.99 × 105 km2 for the ASI SIC algorithm with a spatial resolution of 6.25 km. The averaged relative differences at the minimum and maximum extents from 2002 to 2016 based on 15% and 30% SIC thresholds were 5.63 × 105 km2 and 1.24 × 105 km2 for the NT and Bootstrap SIC algorithms, and 1.48 × 105 km2 for the ASI SIC algorithm. We deduce that the improvement of the SIC algorithms and the satellite sensors’ spatial resolution will reduce the impact of choosing different SIC thresholds on the calculated sea ice extent. However, these influences should not be ignored when using different SIC products for the long-term trend of Arctic sea ice extent.
6. Conclusions
This paper explored the representable value of the SSMIS sea ice concentration threshold at the Arctic sea ice edge during summer in recent years. The MOD29 sea ice product, MODIS images, and ship-based sea ice observations from CHINARE-2012 and CHINARE-2014 were used to extract Arctic sea ice edges. Based on these extracted ice edge lines and points, we made a comparison and statistical analysis of the SSMIS sea ice concentration at the Arctic sea ice edge in the summer of 2012 and 2014.
Overlay analysis of the sea ice edges extracted from the MOD29 sea ice product and SSMIS SIC showed that the average SIC threshold of the four Arctic marginal seas was 33%. The average SIC threshold (42%) of the Chukchi Sea was higher than that in other marginal seas, 34% for the Laptev Sea, 29% for the East Siberian Sea, and 29% for the Greenland Sea. The average SIC thresholds in 2012 and 2014 were 32% and 34%, respectively, with an average value of 33%. Therefore, the commonly used 15% SIC threshold at the Arctic sea ice edge could be adjusted to 30% in the recently rapid changing Arctic.
A comparison of four scenes of MODIS visual interpretation sea ice edges and the corresponding SSMIS SIC indicated that the average SIC threshold value was 33% on 9 June 2012 for the Chukchi Sea, 37% for the Greenland Sea ice on 15 July 2012, 34% for the Laptev Sea on 30 June 2014, and 34% for the East Siberia Sea on 10 July 2014. All these values further confirmed that using 30% as the sea ice edge threshold with respect to the 15% was more reasonable.
The threshold SIC values from overlying the MOD29 and MODIS visual interpretation of the sea ice edge lines on the same date and the same sea region were approximately similar. MOD29 was slightly higher than that of the MODIS visual interpretation as the MOD29 sea ice product misjudged sea water as sea ice in some areas, resulting in a higher SIC at the sea ice edge.
The average SIC at the Arctic sea ice edge points extracted from ship-based observations during CHINARE-2012 and CHINARE-2014 was 31%, which also showed that choosing around 30% as the SIC threshold is recommended for sea ice extent calculation based on SSMIS SIC data. These results and conclusions can provide a reference for further development and improvement of PM SIC products and studying the variation of sea ice under the rapidly changing Arctic.