1. Introduction
Since late 1978, passive microwave sensors have been providing a history of sea ice concentration from satellite data suitable for tracking climate change and variability in the polar regions. Sea ice extent (the area within the 15% concentration contour) and area (the area-integrated concentration) have long been considered as key climate indicators and have been included in numerous national and international climate assessment reports (e.g., [
1,
2]). However, these two parameters provide only limited information about the character of the sea ice; in addition, they have limited skill as indicators of future sea ice conditions, both seasonally and inter-annually [
3].
A suite of interrelated sea ice climate indicators that go beyond extent and area to describe the seasonal evolution of the Arctic sea ice cover from spring through autumn has been derived based on the methods described in [
4,
5], using a long-term satellite-based climate data record. In this paper, we describe the temporal means and variability of the following indicators: snow and ice melt onset dates, sea ice opening, retreat, advance, and closing dates, seasonal ice loss and gain periods, inner and outer ice-free periods (as defined in
Table 1 and described below.)
The day of opening (DOO) captures the start of the seasonal ice loss period, while the day of retreat (DOR) captures the end of the ice loss period [
5]. The 80% sea ice concentration (SIC) threshold typically marks the date when SIC declines toward the annual minimum [
4,
6] and is often used as the upper SIC threshold to define the marginal ice zone boundary. The 80% concentration threshold is also used by operational ice centers (e.g., U.S. National Ice Center [
7]) as the boundary between the marginal ice zone and the pack ice. The difference between DOO and DOR is referred to as the seasonal loss of ice period (SLIP). Similarly, the day of advance (DOA) and day of closing (DOC) denote the beginning and ending of the seasonal ice gain period, respectively. The difference between the two is referred to as the seasonal gain of ice period (SGIP). The inner ice-free period (IIFP) is defined as the difference between DOA and DOR, which captures the period of open ocean when the grid cell ice fraction is less than 15%. The outer ice-free period (OIFP) is the difference between DOC and DOO, which captures both open ocean and ice transition periods. A schematic diagram is shown in
Figure 1, which illustrates the temporal relationship between all dates (DOx) and periods.
Arctic snow and ice melt onset (MO) may be triggered by positive anomalies of air temperature and water vapor [
8]. The timing of MO on Arctic sea ice influences the amount of solar radiation absorbed by the ice–ocean system throughout the melt season by reducing surface albedos in the early spring [
9]. The timing of Arctic MO can be influenced by synoptic conditions [
10]. Significant trends of ~2–3 days/decade in early MO for the period of 1979–2011 were observed by [
11]. They pointed out that the variability in MO are largely driven by spring surface air temperature. Bliss et al. [
9] compared two long-term time series of MO on Arctic sea ice derived from passive microwave brightness temperatures (Tbs) for the period of 1979–2012 and found large mean differences in trends between two different MO algorithms, with the largest uncertainty in thin-ice regions. Depending on the retrieval algorithms and correction methods, the Arctic MO decadal trends were found to range from −1.6 to −4.5 days/decade [
9].
Stammerjohn et al. [
12] identified trends in the DOR, DOA, and the duration of the sea ice season in the Kara and Barents Sea regions and for the East Siberian Sea, Chukchi Sea, and the western Beaufort Sea. Trends in the DOR and DOA for both regions ranged from 1 to 1.9 days per year (i.e., about 10–19 days per decade); with DOR occurring earlier in the year and DOA occurring later. The trends in DOR and DOA resulted in a shortening of the sea ice season by ~2.8 days per year (28 days per decade). Changes in the duration of the sea ice season were first reported by [
13] who later identified a reduction in the length of the Arctic ice season by at least 5 days per decade [
14]. A strong relationship between the timing of ice retreat and that of maximum sea surface temperature (SST) was found by Steele and Dickinson [
5]. Stroeve et al. [
15] investigated the predictability of DOA timing in response to timing of the prior DOR using several SIC thresholds and also found inverse correlations between DOR and DOA for most of the Arctic, as would be expected. However, some positive correlations were found in areas primarily in the eastern Arctic and along the ice edge near Franz Josef Land where the IIFP is short and can occur at varied times of the melt season from year to year.
The climatological means and temporal variability of these parameters will inform a variety of stakeholders. For example, the timing of melt onset and the date of sea ice retreat can provide useful information on the evolution of the ice cover through the rest of the summer, and thus be useful to efforts to improve seasonal sea ice forecasts (e.g., [
15]). MO is also an indicator for summer minimum extent, since it plays a key role in the amount of solar energy absorption, and along with freeze-up, the overall energy balance of the Arctic climate system (e.g., [
16,
17]). The length of the open water season (as derived from retreat/advance dates) is important for human activities (navigability of waters, access to natural resources, hunting and transportation by indigenous populations) and for wildlife (e.g., polar bear access to food sources). The long-term time series will inform future planning of military, civilian, and commercial infrastructure (buildings, marine vessels).
Therefore, it is beneficial to provide users with the basic characteristics of these parameters from a consistent long-term time series. In this paper, we describe temporal mean and variability of Arctic climate indicators derived from a long-term satellite-based sea ice concentration climate data record. This is the first time that the basic statistics and temporal variability of all dates and periods are computed and described. By utilizing our integrated data set with all of these parameters, this study provides a new way to holistically look at the changing seasonality of the Arctic sea ice cover.
The paper is organized as follows.
Section 2 outlines the datasets used in this study.
Section 3 describes the temporal means and variability of these dates and periods. Discussions on sensitivity of different averaging choices and methods on the decadal trend of MO and on the means of dates and periods are provided in
Section 4, followed by the conclusions in
Section 5.
4. Discussion
The total number of valid dates cells varies on a year-to-year basis. As mentioned previously, there are two choices for computing regional averages: (i) average in each year only over the area for which a parameter is valid for ALL years (i.e., the minimal intersection of all years), or (ii) average in each year over the full area valid for that parameter in that year. Large difference of more than 11 days was found in regional averages of MO dates between these two different averaging choices. So was their decadal trend difference of about 3 days per decade, which is as high as the trend computed from the time series obtained from the first averaging choice (see
Figure 2 and
Table 4). However, the differences are greatly reduced if the long-term means are removed at each valid date cell before the regional averaging is carried out. The biases are now only 0.09 day between these time series of MO anomalies and 0.7 day per decade for their decadal trends. While the trend computed for the first averaging choice remains the same for both time series of MO and MO anomaly, the trend computed from the second averaging choice using the MO anomalies is very close to that from the first averaging choice. This demonstrates that the spatial-varied mean state of valid dates has a big impact on the trend analysis. If the second averaging choice is preferred to focus on temporal variability, one should use the anomaly fields to compute regional time series. This may, in part, explain the differences in trend values of dates and periods between our results and the previous results, such as those from [
12,
14].
After the initial melting date, additional freeze/thaw cycles may occur. Then, the question is whether continuous melt onset (CMO) will be a better indicator. The passive microwave CMO data available rely on a sea ice concentration threshold in cases when a clear melting signal cannot be determined from the brightness temperatures [
6]. In these cases, the algorithm for continuous melt identifies the day when the SIC falls below 80% for the last time before the location remains ice-free until autumn freeze-up. The threshold is primarily used to identify continuous melt onset in the peripheral ice regions [
9] where we obtain our dates and periods of retreat/advance. This threshold conflicts with the DOO indicator, which also uses an 80% SIC threshold. In this case, we chose to use the initial MO date to identify the earliest changes in the sea ice cover at the beginning of the melt season. The earlier MO dates are more closely tied to the spring atmospheric conditions that initiate melt (as used by Mortin et al. [
8] and Liu and Schweiger [
10]) and therefore give additional information about the sea ice melt and retreat period that the continuous MO date would not provide. Bliss et al. [
9] have compared the initial, early and continuous MO dates. The annual evolutions of those MO dates for our record periods are included in
Figure 9 as reference.
Because the periods are inter-related with sea ice dates, calculation of a period mean includes only date cells that are valid to both the dates used to compute the period. Due to the fact that not all valid date cells undergo a complete melt, retreat, advance, and freeze-up cycle, it is expected to see some sensitivity of the resultant mean values to which valid date cells are used to compute the means. For Case allv, where all valid date cells are included when calculating the Arctic means of individual dates and periods, the DOC mean is 309.3, which is actually short of the DOA mean of 309.8, while the SGIP mean is 14.6 days (
Table 5). This may seem to be counter-intuitive but it is in fact plausible as it is not unusual to have large areas of very early DOC in the north, namely, right after the sea ice minimum. This happens because large areas of the interior drop below 80% at the end of summer (namely, very late DOO), and then close right back up again just after the sea ice minimum (namely, very early DOC). The calculation of SGIP mean uses only those date cells that are valid for both DOA and DOC, which will mask out those cells. Additional study is needed to confirm this, which is beyond the scope of this study. We have, however, considered two other cases to help demonstrate the potential validity of our speculation. Case mask is when the cells are masked out by that of DOR. Case mask implies that only cells that experienced ice retreat are selected. Case comm is only used for cells, of which all dates are valid for the year. Case comm implies that all the selected cells have undergone the complete melt, retreat, advance, and freeze-up cycle. The total records for computing DOC means are reduced in both Case mask and Case comm, compared to that of Case allv (
Table 5). The DOC means for Case mask and Case comm are 323.1, which is 13.3 and 14.6 days more than the DOA means, respectively.
5. Conclusions
Temporal means and variability of a set of climate indicators that track the seasonal evolution of the Arctic snow and ice cover have been described. These climate indicators, derived from a well-managed and documented satellite CDR, include snow and ice melt onset dates, sea ice opening and retreat dates, sea ice advance and closing dates, seasonal ice loss or gain periods, inner and outer ice-free periods.
Significant trends calculated over the period of March 1979–February 2017 are observed for all dates (MO, DOO, DOR, DOA, and DOC). The dates of melt onset, opening, and retreat are getting earlier in the year at a rate of more than 5 days per decade. The dates of ice advance and closing are getting later in the year at a rate of more than 5 days per decade. As the result, both inner ice-free period (i.e., potential open ocean) and outer ice-free period show a significant upward trend of nearly 12 days per decade. Seasonal ice loss period (SLIP) and seasonal ice gain period (SGIP) exhibit a slight but significant downward trend of −1.48 days per decade and −0.53 days per decade, respectively. All trends of dates and periods are significant at the 99% confidence level.
If these significant trends persistent, there will be potential impact on human activities and for wildlife due to the increased length of ice-free periods. Earlier melt onset and prolonged open water reduce surface albedos which could have a large influence on the total amount of solar radiation absorbed by the Arctic ice-ocean system [
24].
Our analysis has demonstrated the sensitivity of computing long-term means of dates and periods, and that of the MO trends to regional averaging methods and valid cells masks due to the transient valid date cells. This is exacerbated by recent accelerated Arctic sea ice melting. It is our hope to start a community-wide dialogue to standardize how they should be computed for the purpose of climate monitoring of these indicators.
Sea ice coverage varies on the regional scales (e.g., [
35]). Variability in trends for dates and periods does occur on the regional scales (see
Figure 6; also, [
6,
12,
36]). The regional variability of the climate indicators will be examined in more detail by [
37], using the same datasets utilized in this paper.
The retrieval algorithms of passive microwave sea ice concentrations are sensitive to emissivity and surface temperature of sea ice, atmospheric effects, melt ponds, and thin ice, etc. They are also prone to land contamination in the coastal areas. Large uncertainty in SIC retrievals are found in the marginal ice zone, especially in summer, due to atmospheric and wind roughness efforts of open water areas [
38]. (See [
39] for a comprehensive review of accuracy and precision of sea ice concentration algorithms in various sea ice and atmospheric conditions and inter-comparison of the SIC retrieval algorithms.) Methods have been developed to minimize the impact of those error sources [
20]. The error sources and uncertainties associated with SIC algorithms will nonetheless adversely affect the accuracy and precision of the DOx algorithms.
We here use a consistent and authoritative SIC data set (the CDR), which ensures the best possible inputs to our DOx algorithms. The existing DOx algorithms differ slightly among the small (but growing) number of researchers investigating the seasonal timing of sea ice retreat and advance. Going forward, an inter-comparison of these algorithms, including an in-depth study of a few significant years in sea ice history, for example, 1998, 2007, 2012, etc., would be useful.