Background & Summary

Polynyas play a crucial role in global ocean circulation and the Antarctic ecosystem1. They facilitate active ocean-atmosphere heat exchange and abundant sea ice production, leading to the formation of dense water, a precursor to Antarctic Bottom Water2,3. This dense water drives the low limb of the thermohaline circulation, crucial for the climate system4,5,6,7,8. Polynyas also benefit the Antarctic ecosystem by allowing increased penetration of solar radiation, thereby promoting phytoplankton blooms9,10,11, higher net primary productivity12,13,14,15, and acting as significant carbon dioxide sinks16,17,18. Against the current background of global warming, polynyas are attracting increasing attention from scientific communities, out of which evidence has emerged of rapidly changing air–sea interaction19,20. As part of this work, an exhaustive survey of Antarctic polynyas is essential to improve our understanding of their mechanisms and impacts.

Spatially, polynyas are defined as areas of open water or thin ice surrounded by thicker sea ice, meanwhile, temporally, they last for weeks and recur over years1. Previous studies have integrated daily open-water maps to obtain the spatial distribution of the climatological probability of ice-free conditions, and then artificially identified major polynyas using the spatial definition. Whilst these works have been good at identifying frequent polynyas and providing valuable insights into the impacts of polynyas on local ecosystems9,16,21 and the production of sea ice22,23,24,25, due to the methodological limitation whereby temporal integration causes excessive smoothing, they have failed to reveal infrequent polynyas. This is important because infrequent polynyas are a crucial part of the Antarctic climate system as formation regions for dense shelf water (the precursor of Antarctic Bottom Water) and carbon sinks26,27,28. Compared with the relatively better-studied frequent Antarctic polynyas, infrequent polynyas open up more irregularly and hence can be more sensitive to climate change29,30. In this respect, employing temporal definitions of polynyas directly, rather than through muti-year integration, is the key to more efficiently extracting information on infrequent polynyas1,31,32,33. Moreover, existing maps only cover up to the year 20119,22,23, which neglects the most recent decade when the Antarctic sea-ice extent has changed drastically. Therefore, it is necessary to develop a new algorithm and dataset to accurately depict both infrequent and frequent polynyas and extend the temporal coverage for studying them.

In this work, we developed a novel method for identifying polynyas. Previous identification methods are based on multi-year integrated climatological maps, which can lead to missing infrequent polynyas with short lifespans. Inspired by an existing cyclone track algorithm34, our new method uses daily maps of sea ice concentration (SIC), traces each area of open water to capture its characteristics, and selects polynyas among them according to spatiotemporal definitions of polynyas. Using this approach, we have created a dataset, called the Daily Extent of Each Polynya in Antarctica (DEEP-AA), which spans from 2003 to 2022. With a temporal resolution of 1 day and a spatial resolution of 6.25 km, DEEP-AA provides detailed information on the area of each polynya in Antarctica. To ensure reliability, we evaluated DEEP-AA by comparing it with various other datasets, including visible/near-infrared remote sensing images and ship-based records35. For frequent coastal polynyas, DEEP-AA can accurately identify them in a manner consistent with previous studies35,36. For infrequent polynyas, we successfully identified both coastal and open-ocean polynyas, including those studied previously (e.g., Muad Rise polynya37,38,39) and other newly found cases (e.g., South Drygalski Ice Tongue polynya). Benefiting from our approach that allows the identification of infrequent polynyas, the total number of polynyas identified using our method is three times more compared with previous studies9,21,23. Lastly, as an example of use, we provide preliminary statistics on the characteristics of infrequent and frequent polynyas, revealing that they have different seasonal cycle patterns. The area of infrequent polynyas begins to expand in August, one month earlier than frequent polynyas. We obtained multiple polynya identification results based on different parameter settings for sensitivity analysis. The results of all sensitivity analyses are available online as components (subsets) of DEEP-AA. For general use, we recommend the subset named “SIC60_6.25km_20d”, which has been parameter-tuned. All analyses presented in this paper are also based on this subset. Our dataset provides a foundation for studying the common characteristics, overall impacts, and regional and type differences of Antarctic polynyas.

Methods

Our method take inspiration from an existing cyclone tracking algorithm (TRACK)34,40. Similar to TRACK, we filter out noise first, and then trace and select polynyas. However, the specific algorithms for each step are redesigned according to the definition of polynyas, and all code is rewritten. The detailed steps of our novel method are as follows:

Step 1—Remove open-sea regions in daily open-water maps

Firstly, we obtain the open-water areas based on SIC data. Following Campbell et al.37, we define regions where SIC is below 60% as open water. Then, we mask the open waters not enclosed by sea ice, i.e., the open sea. Here it is defined as the open waters connected to the northern boundary of the study area (Fig. 1, Step1, and Supplementary Figure 3). Note that the SIC dataset used in this study provides fake low concentration over the landfast ice41. Thus, we mask landfast ice using Fraser’s dataset42. A few large tabular icebergs and many small grounded icebergs are also included in the landfast ice dataset.

Step 2—Filter high-frequency noise

In Step 1, we obtain daily maps containing thousands of areas of open water surrounded by sea ice, but most of them may be created by synoptic-scale meteorological events (e.g., cyclones43), not polynyas. These huge amounts of noise will increase the computational cost and make the algorithm more complex, as we need to trace them accurately. To address this issue, we implement a preprocessing step to remove the noise signals that persist for particularly short durations and are unlikely to be polynyas. We employ a relaxed temporal filter by calculating the probability of open water for each pixel within a moving window (14 days in this work) and masking out those pixels with a low probability (<70% in this work). This step enables us to eliminate a large portion of the noise, thereby facilitating subsequent tracing efforts by focusing on more reliable open-water features.

Step 3—Trace areas of open water on a daily scale

Constrained by the topography and landfast ice, for example, the location of a polynya is approximately fixed over time1. Thus, we first link the areas of open water that overlap in space during their temporal evolution to identify multiple open-water sequences (Fig. 1, Step 3). Note that multiple areas of open water may merge into one, or split into multiple areas, which causes sequence branching (e.g., the yellow sequence in Fig. 1, Step 3). These branches indicate two possible situations: (a) a temporary connection between independent polynyas; or (b) an unusual division of one polynya. Based on the duration of the branch, our algorithm can determine for each one which of the above situations it belongs to, i.e., the branches belong to one or multiple polynyas. The details are shown in Supplementary Text 1. Moreover, some polynyas may reopen after being lost for days or months. The reappearance of polynyas is also taken into account in our algorithm (see Supplementary Text 1).

Fig. 1
figure 1

Diagram of the algorithm and the variables contained in the dataset. A detailed diagram of the “open sea removing” algorithm in Step 1 is provided in Supplementary Figure 3. The specifics of the tracing algorithm used in Step 3 are described in Supplementary Text 1.

Full size image

Step 4—Select open-water sequences

We use the durations of open-water sequences traced in Step 3 to identify polynya occurrences. We calculate the persistence in days of each open-water sequence and define those with short lifespans (<20 days) as not polynyas. Extremely small areas of open water (maximum area <100 km2) are also removed. Additionally, in summer, sea ice in Antarctica mostly melts, causing polynyas to become highly unstable, rapidly expand, and disappear44, which makes it impossible to trace areas of open water. Therefore, we restrict the tracing to the period from April to October, and for each polynya when its daily-mean air temperatures (from the ERA5 reanalysis dataset) are above freezing, it will also be masked. Finally, we perform integration within the lifespan of each open water sequence to obtain their respective probability maps. We refer to the area on each map with a probability exceeding 20% as the “typical area” of this particular open-water sequence.

Step 5—Trace and select open-water sequences on a yearly scale

Polynyas are defined as areas of open water or thin ice, which can recur multiple times in different years. Thus, in this step, we link the typical areas of open-water sequences in different years obtained in Step 4. This step is the same as Step 2, except with certain thresholds applied (Supplementary Table 1). Considering that polynyas should “recur from year to year”1, in our dataset, we remove open water regions that do not reappear.

Step 6—Map daily each polynya’s edge

Via the above steps, we successfully extract all Antarctic polynyas. In this step, based on the year and location where it is discovered in the dataset, we give each polynya sequence a unique identification code (ID). Since the sequence contains daily information about each polynya, we can map the areas of polynyas for each day, i.e., label each polynya by their IDs on the daily open-water maps. We also tag whether the polynya is coastal or open-ocean by checking whether the minimum distance from the polynya to the coast is always less than 50 km. The non-polynya open waters surrounded by sea ice (removed in Steps 2–5), open sea, and landfast ice/land are also provided in DEEP-AA.

The following is a list of the data used to build our DEEP-AA dataset. The SIC data used in this work are based on the Advanced Microwave Scanning Radiometer (AMSR) data provided by the University of Bremen (https://doi.org/10.1594/PANGAEA.91977845, https://doi.org/10.1594/PANGAEA.89840046) with a resolution of 6.25 km/1 d47. The meteorological data are derived from ERA5 (https://doi.org/10.24381/cds.adbb2d4748) with a spatial resolution of 0.25°49. The landfast ice masks come from the Circum-Antarctic landfast sea ice extent, 2000–2018, version 2.2 (https://doi.org/10.26179/5d267d1ceb60c50), with a resolution of 1 km/half-month42, and the U.S. National Ice Center Arctic and Antarctic Sea Ice Concentration and Climatologies in Gridded Format, Version 1 (https://doi.org/10.7265/46cc-3952), with a resolution of 12.5 km/week51. Both the meteorological and landfast ice data are spatially interpolated to the grid of the SIC data.

Data Record

The daily extent of each polynya in our dataset is recorded in NetCDF files, which are available at https://doi.org/10.5281/zenodo.1137914752. The dataset covers the period 2003–2022, but 2011–2012 is partially missing. The naming convention for each file is “DEEP_sSR_DS_OM_Time_vX.nc”, where SR indicates the spatial resolution; DS indicates the source of the input data; OM is the method to define open water; Time is composed of the four-digit year, two-digit month, and two-digit day; and X is the version number. Each polynya has a unique ID code in the form “YYWWWWSSX”, which is determined by the time and location where the polynya first appears in the dataset., i.e., it is composed of the two-digit year (YY), four-digit longitude (WWWW, rounded to one decimal place, and the range is 0°–360°), two-digit latitude (SS), and one-digit category-code (X) indicating a coastal (even number) or open-ocean (odd number) polynya. To enable users to quickly find the ID of a polynya, we provide an overview map (OverviewMap.mat, Fig. 2a) in which we mark the typical area of each polynya by the ID masks, and the key parameters used to calculate the dataset are provided (Input.txt).

Fig. 2
figure 2

(a) The typical area of each polynya, obtained as detailed in the main text (Section: Methods). To facilitate distinction, adjacent polynyas are marked with different colors. (b) Maps of frequent polynyas. A total of 35 frequent polynyas, defined as described in the main text (Section: Overview of the polynyas in the DEEP-AA dataset), are marked in the figure, with their numbers and names written at the bottom of the figure. Their IDs and interannual variation of size and lifespan are listed in Supplementary Figure 10. In both (a,b), we divide the Antarctic into two regions. The locations of these two regions are marked on the upper left side of the subfigures.

Full size image

Validation data

All the data used to validate our dataset are available online. The visible remote sensing images come from the Moderate Resolution Imaging Spectroradiometer (MODIS) Atmospherically Corrected Surface Reflectance product (https://doi.org/10.5067/MODIS/MOD09.061, https://doi.org/10.5067/MODIS/MYD09.061) with a horizontal resolution of 500 m53,54, and the near-infrared images (band 32) are also from the Level-1B product of MODIS (https://doi.org/10.5067/MODIS/MOD021KM.061) with a horizontal resolution of 1 km51. The sea ice thickness data are from the Soil Moisture and Ocean Salinity (SMOS) mission (https://doi.org/10.1594/PANGAEA.93473255) with a horizontal resolution of 12.5 km56. The hourly ship-based observations in the Ross Sea were obtained by the Polynyas and Ice Production in the Ross Sea (PIPERS) cruise NBP1704 (https://doi.org/10.15784/601183)35,57.

Technical Validation

Overview of the polynyas in the DEEP-AA dataset

In our dataset, we identify 174 polynyas in the Southern Ocean, which triples the number of traditionally identified polynyas, collectively covering an area of 98,040 km2 on average. For each polynya, by integrating over its lifespan, we obtain their respective probability maps. We refer to the area with a probability exceeding 20% as the “typical area” of this polynya and map each polynya’s typical area as shown in Fig. 2a. Among the identified polynyas, 158 are coastal, contributing to 95% of the total area. Following Li et al. (2016), we divide Antarctica into two parts with meridians of 160°E and 20°W, which avoids cutting the main polynyas as much as possible. Interestingly, we observe that the total area of polynyas in the eastern region surpasses that in the western region (east: 64,707 km2; west: 33,333 km2), which may be driven by the west-east asymmetry of katabatic winds9. However, the numbers of polynyas do not differ greatly (east: 89; west: 85).

We define polynyas with an annual mean exposure time longer than 120 days as frequent polynyas. On this basis, our dataset reveals 35 frequent polynyas along the Antarctic coast (Fig. 2b), whose names refer to those given in Arrigo et al.21 and Kern23. These 35 polynyas account for 75% of the total area of Antarctic polynyas. They are prominently distributed along the coast from the Ross Sea to the Prydz Bay in East Antarctica, with sparse occurrences in the Atlantic and Indian Ocean sectors, consistent with findings from previous studies25,36,58. The interannual variations in the area and lifespan of each of these high-frequency polynyas are shown in the Supplementary Figure 10.

Evaluation of frequent polynyas

Many previous works have mapped frequent polynyas9,21,23,58. Here, we follow Nihashi and Ohshima (2015)36 and compare the 13 most recognized major coastal polynyas (Supplementary Figure 4). All these polynyas can be identified in our dataset. Their shapes are mostly similar to the results obtained from the traditional method, and the spatial correlations between them are significant (Supplementary Figure 4). This confirms the reliability of our new method in identifying well-known polynyas.

In addition, we also highlight the unique capability of our dataset to capture both the spatial structures and temporal variability in a manner consistent with ship-based observations, particularly those provided by the PIPERS project35. During the cruise, the ship visited two polynyas (Fig. 3c): the Terra Nova Bay polynya (TNBP, Apr. 30 to May 12) and the Ross Sea polynya (RSP, May 16 to 17). Using DEEP-AA with the ship’s position, we are able to extract similar time series from their cruise report, thereby demonstrating the capability of DEEP-AA (Fig. 3a). Second, the ship-polynya distance obtained based on DEEP-AA changes to 0 many times (Fig. 3b), indicating that the ship entered and exited the polynya many times, which is consistent with Guest’s report59. Not only that, but the ship-polynya distance also fits significantly with underway ice thickness changes (Fig. 3b, r = 0.73, p < 0.001, within 120 km of the polynya). This phenomenon agrees with previous analyses60. Third, the non-polynya thin-ice areas reported observationally are also correctly identified by DEEP-AA. On Apr. 19 and Jun. 05, the underway thin-ice areas are labeled as the open sea in our dataset (Fig. 3a,b), consistent with the cruise report35. When observations recorded that the ship briefly encountered a wide non-polynya break on Jun. 04 (blue triangle in Fig. 3b)35, our dataset also indicates that the ship was very close (~5 km) to a non-polynya open-water area.

Fig. 3
figure 3

Comparison between our dataset and ship-based observations35. (a) Ice and open-water regimes in the course from DEEP-AA and observations. The band along the top comes from DEEP-AA. The light grey blocks indicate the ship was not in a polynya; black ones indicate that it was in the Terra Nova Bay polynya (TNBP); green ones indicate that it was in the Ross Sea polynya (RSP); and blue ones refer to the open sea. The onboard observed ice and polynya regimes are marked below the band; MIZ indicates the marginal ice zone, and FYI indicates first-year ice. All of them come from the PIPERS report in Ackley et al.35. The time of entry or exit from the polynya is also marked in (b) with dotted lines. In (b), the blue line shows the ice thickness from the ship-based observations; the blue shaded areas highlight where the ice was thinner than 10 cm (i.e., where the blue line is below the grey dashed line), and the underway wide break is marked by the blue inverted triangle. The red line shows the distance from the ship to the polynyas found by our dataset. The ship trajectory is shown in Supplementary Figure 5.

Full size image

Evaluation of open-ocean polynyas

Maud Rise polynya

The Maud Rise polynya (MRP), as a representative case of open-ocean polynyas, is one of the largest polynya37, which can facilitate intense deep ocean ventilation and possibly contribute to AABW modification26,61. In 2017, MODIS’ visible remote sensing fully documented the occurrence of MRP, offering a robust benchmark for evaluation. The visible images depicted MRP opening up on Sep. 2, and then its size expanded rapidly within two weeks. MRP persisted to the summer (left-hand columns in Fig. 4a–k). Our dataset correctly traces the entire development process of MRP during this event (red borders), including the early stages of the event when the MODIS images illustrate that the polynya was very small (~3000 km2, ~5% of the area on Oct. 1). Additionally, we compare our dataset with the ice thickness data from SMOS, which is less susceptible to cloud cover than MODIS (right-hand columns in Fig. 4a–k). The accuracy of SMOS ice thickness data has been verified by Kaleschke et al.’s work56. The polynya (defined as the region with ice thickness < 10 cm) is accurately identified and the whole MRP event is effectively traced. Thus, the reliability of our method is reaffirmed.

Fig. 4
figure 4

The Maud Rise polynya obtained by our dataset in 2017, overlaid on visible light remote sensing images and the SMOS seaice thickness. The red hatched areas denote the extent of the polynya from our dataset, while the yellow areas denote other open waters. In (a–k), the base maps on the left are the visible light remote sensing images from MODIS/Terra, and those on the right indicate the sea ice thickness from SMOS data. Panels (l,m) show the Maud Rise polynya in 2004 and 2016, whose base maps are also visible light images from MODIS/Terra. The location of the study area is marked with a black box in the lower right of the figure.

Full size image

Moreover, in 2004 and 2016, when previous studies reported the opening up of MRP37,39,62, our dataset also successfully identifies the polynya’s signal, which is verified by limited visible images (Fig. 4l,m). Although there are large temporal gaps between these polynya events, the largest of which is even more than 10 years, our method still traces all of them effectively.

Unlike our method, however, the traditional multi-year-integral method23,36 finds it difficult to identify MRP. Figure 5a shows the frequency of open water from 2003 to 2022. Due to the few exposure times, MRP’s occurrence frequency is not significantly higher than that of the surrounding area. This renders the traditional multi-year-integration method incapable of effectively identifying MRP, while our new method can accomplish the task easily.

Fig. 5
figure 5

Comparison between the traditional integration-based polynya frequency map and the result from our dataset for open-ocean polynyas: (a,b) Maud Rise polynya; (c,d) Cosmonaut polynya; (e,f) Cooperation polynya; (g,h) North Ross Sea polynya; (i,j) halos in the Maud Rise region in 2005 and 2018. All of their locations are marked on the map of Antarctica in the center of the figure. Panels (a), (c), (e), and (g) are the traditional open-water frequency maps integrated from 2003 to 2022. The red borders in (b), (d), (f), and (h) indicate the daily polynya edges obtained by our tracing method in our dataset; the yellow hatched areas indicate the open sea, and the yellow borders without hatching delineate other open waters. The base maps come from MODIS visible light images (b,d,f from MODIS/Terra; h,i,j from MODIS/Aqua). The time is indicated at the bottom of each subfigure.

Full size image

Cosmonaut polynya

The difficulty in identifying open-ocean polynyas through traditional methods exists not only with MRP. Almost all open-ocean polynyas have limited occurrences and short durations, making them unrecognizable via traditional methods. Many studies have pointed out that there is an open-ocean polynya located in the Cosmonaut Sea (the Cosmonaut polynya, CP)1,63. However, existing methods employing open-water frequency maps find it hard to identify CP (Fig. 5c). In contrast, our DEEP-AA dataset successfully delineates the daily edges of CP in a manner consistent with visible remote sensing images (Fig. 5d). This further reiterates that, compared to the traditional approach and existing maps, our novel method and dataset can detect open-ocean polynyas more effectively.

More exhaustive and accurate identification of Antarctic open-ocean polynyas promotes our understanding of them. While many previous studies have rightfully emphasized the significance of MRP as the largest polynya in the Southern Ocean37,64, our work contends that CP deserves equal attention. Our dataset reveals that the size of CP is only 29% smaller than MRP (CP: 2.6 × 104 km2; MRP: 3.6 × 104 km2), but its cumulative number of exposure days is nearly twice that of MRP (CP: 342 days; MRP: 135 days). This is owed to the significantly higher frequency of CP, which appeared in 10 out of 17 years, while MRP appeared only in 3 years. The longer cumulative persistence time and more frequent occurrence of CP together imply a potential significance of this polynya in the formation of the Antarctic Bottom Water.

Cooperation polynya

Recently, Qing et al.65 reported an open-ocean polynya in the Cooperation Sea (60°–90°E). Whilst this polynya cannot be clearly captured in the previous method, it is identified in our dataset (Fig. 5e,f). However, different from Qing’s work, we identify more than one open-ocean polynya in this region (Supplementary Figure 6), and the result is not sensitive to the input data or algorithm parameters. Gyres formed by the southern Antarctic Circumpolar Current front and Antarctic Slop Current are the key to opening up this polynya65. The presence of multiple polynyas in the Cooperation Sea suggests a complex circulation structure there.

North ross sea polynya

We also discovered some open-ocean polynyas that have rarely been recorded in the past, such as the North Ross Sea polynya (Fig. 5g), which is also confirmed by visible remote sensing (Fig. 5h).

Maud Rise halos

Lastly, we carried out a verification of DEEP-AA’s capability to prevent false positives (i.e., distinguish between non-polynya open waters and polynyas). Apart from the polynya itself, at the Maud Rise polynya, there were occurrences of transient and unstable non-polynya open waters in some years. Previous researchers have referred to these as “halos” and differentiated them from polynyas based on hydrological preconditions, mechanisms, and effects28,66,67,68. Here, we present two halos in 2005 and 2018 (Fig. 5i,j). Benefiting from the tracing algorithm and the newly introduced temporal definition, our dataset effectively distinguishes these halos from polynyas, aligning with prior knowledge.

Evaluation of infrequent coastal polynyas

In addition to open-ocean polynyas, DEEP-AA also newly identifies numerous infrequent coastal polynyas, which we were able to verify via several cases.

South drygalski ice tongue polynya

In the southwestern Ross Sea, our new tracing method reveals a previously unrecognized polynya south of the Drygalski Ice Tongue (SDITP, highlighted in red in Fig. 6d). This region is typically covered by fast ice, and the polynya opens up infrequently, especially compared to the nearby TNBP and McMurdo polynya (McP) (Fig. 6c). A few previous studies considered it to be part of McP (Fig. 6b)23, but our dataset shows it to be an independent polynya. The wind drivers of this polynya which are different from those of McP (Fig. 6d), as well as the pattern of spatiotemporal evolution (r = 0.01, p = 0.78, non-significant), confirm our conclusion.

Fig. 6
figure 6

Newly found infrequent coastal polynyas. Panel (b) shows the climatic potential position of polynyas in the Ross Sea in Kern’s work23. The newly found South Drygalski Ice Tongue polynya and the adjacent polynyas (the Terra Nova Bay polynya and McMurdo polynya) are shown in (c,d). Panels (e,f) illustrate the Peter I Island polynya, and (g,h) the Balleny Islands polynya, which are also marked in (a). Panels (c), (e), and (g) indicate the traditional open-water frequency maps integrated from 2003 to 2022, while the red lines in (d), (f), and (h) indicate the newly found polynyas’ daily edges in our dataset, and the white (yellow) lines indicate the edges of other polynyas (open waters). The base maps come from MODIS/Aqua visible light images at the corresponding times.

Full size image

Peter I Island polynya and balleny islands polynya

Our dataset also identifies polynyas near islands far from the Antarctic mainland. The polynyas by Peter I Island in the Bellingshausen Sea and the Balleny Islands west of the Ross Sea are some examples of these. These polynyas around islands are driven by winds and are considered key components of the local ecosystem by biologists69,70, despite not being marked on previous maps. These polynyas are identified in our dataset and verified by MODIS images (Fig. 6e–g).

Extreme iceberg event and potential issues

We also evaluated DEEP-AA in terms of a special and extreme event in which iceberg B15A entered into the TNBP in 200571.

The overall good agreement between our dataset and observations, in this case, demonstrates the reliability of our method and DEEP-AA even under extreme conditions. The MODIS images in Fig. 7a–g show the distribution of the TNBP and iceberg B15A during this event, with the polynya extent identified by our dataset (red hatched areas). Normally, the whole TNBP is simultaneously controlled by westerly katabatic winds (Fig. 7h). However, in this event, with the iceberg’s residence, the TNBP was split into two parts that developed asymmetrically: the west part shrank since the iceberg inhibited the newly formed sea- ice outflow, while the east part expanded72. Our dataset (red hatched areas in Fig. 7) successfully identified both parts of the TNBP segmented by B15A and traced the whole process of asymmetric development.

Fig. 7
figure 7

Results of our dataset with infrared remote sensing images during the intrusions of iceberg B15A into the Terra Nova Bay polynya: (a–g) monochrome mosaics of infrared remote sensing images (band 32) from MODIS/Terra, in which the red-hatched regions indicate the extent of the Terra Nova Bay polynya from our dataset, the yellow regions are other open waters, green arrows are the winds, and the edge of iceberg B15A is highlighted by the black border lines; (h) blue mosaic of polynya frequency from our dataset, with the mean wind field from 2003 to 2022 shown by the green arrows.

Full size image

Although such a special case is very rare (having happened only once in 19 years in the TNBP), and our method identifies and traces the polynya correctly overall, the potential limitations need to be noted. Firstly, small parts of the iceberg may be misidentified as being part of the polynya (e.g., Fig. 7a,d,e). This issue derives from the input SIC data based on 89 GHz, which sometimes cannot distinguish icebergs and polynyas effectively41,73. A suitable iceberg mask is needed to further correct it. Secondly, our tracing algorithm assumes that the locations of polynyas are fixed. However, in this case, the drifting iceberg drives the west part of the TNBP to move quickly, causing it to be missed on May 15 and 23. Similarly, the advancement of ice shelves may also result in the misidentification of one polynya as multiple polynyas (e.g., the Ronne Ice Shelf polynya, Supplementary Figure 7). Fortunately, the impact of these flaws on the total area of Antarctic polynyas is less than 1% and does not affect our main conclusion.

Sensitivity analysis

In this section, we report results from a sensitivity analysis for the method and threshold used to define open waters, the spatial resolution of the input data, the threshold of the maintenance time of polynyas, the tracing parameters, and the landfast ice mask.

In terms of the method for defining open waters, in previous studies, two definitions have commonly been employed: (a) setting a threshold on the SIC map, or (b) applying the Polynya Signature Simulation Method (PSSM) to the brightness temperature directly to obtain the open water21,22,74. Here, we compare the polynyas identified based on the two definition ways. The brightness temperature used in PSSM comes from AMSR-E/2 (https://seaice.uni-bremen.de/data/amsr2/tb_daygrid_swath/s12500 and https://seaice.uni-bremen.de/data/amsre/tb_daygrid_swath/s12500). Similar to previous findings41,75, we find that the total area of polynyas derived from both methods is similar when using a SIC threshold of 60% and a polarization ratio threshold of 0.08 (equivalent to an ice thickness of ~10 cm) for PSSM (Fig. 8a). Additionally, we checked the 12 major Antarctic polynyas, and the results from the two sources also correlated (Supplementary Figure 8).

Fig. 8
figure 8

Sensitivity of the new method to (a) the threshold for defining open water, (b) the spatial resolution of the input data, (c) the threshold for the number of maintenance days, (d) the parameters for daily-scale tracing, (e) the parameters for yearly-scale tracing, and (f) the masking of landfast ice. The filled bars indicate the numbers of coastal polynyas; the shaded bars indicate the numbers of open-ocean polynyas; the error bars indicate the total area of polynyas plus or minus the standard deviation. Red bars indicate that the input open-water maps come from the SIC, and blue bars that they were obtained by PSSM. The parameters used for the dataset in this paper are formatted in bold typeface.

Full size image

Regarding the threshold used to define open water, we found that it can impact the polynya area significantly (Fig. 8a). Increasing the SIC threshold by 10% or decreasing the threshold of polarization difference in PSSM by 0.05 reduced the total area of polynyas by ~40%, but this has a less significant impact on the polynya quantity.

In terms of the spatial resolution of the input SIC data, we compared four resolutions: 3.125 and 6.25 km provided by the University of Bremen, and 12.5 and 25 km obtained by down-sampling. We found that when the resolution drops to 25 km, which is too close to the scale of many polynyas to identify them, both the area and quantity of polynyas drop rapidly (Fig. 8b). Therefore, the resolution of the input SIC data should be no coarser than 12.5 km.

For the threshold of the maintenance time of polynyas, we tested three values: 10 days, 14 days (2 weeks), and 20 days (Fig. 8c). In our method, when the number of exposure days of an open water is below this threshold, it is not identified as a polynya. We found that as the value of this threshold decreases, more polynyas are identified, especially open-ocean polynyas. However, the total area of polynyas is not sensitive to the threshold.

Regarding the tracing parameters, we tested the daily-tracing parameters (#13, #15, #16, and #18 in Supplementary Table 1) and yearly-tracing parameters (#14, #17, and #19 in Supplementary Table 1) separately, and found that the effects of the two are similar (Fig. 8d,e): for the total polynya area, changes in tracking parameters will not have a particularly significant impact, while for quantity, tracking parameters will be much more important.

Finally, we compared the results with and without the masking of landfast ice (Fig. 8f). Removing the mask, both the number and area of polynyas increase. Nihashi and Ohshima36 pointed out that the location of polynyas along the Antarctic coast is closely related to landfast ice, and Kern41 found that 89-GHz SIC production cannot distinguish well between low-SIC polynyas and landfast ice. Therefore, if the landfast ice is not masked, it may lead to an overestimation of the area and number of polynyas.

Differences in seasonal cycles between frequent and infrequent polynyas

At last, we offer an example of how DEEP-AA could be useful for advancing the study of polynyas. In the DEEP-AA dataset, numerous polynyas with various occurrence frequencies are recorded. The histogram in Fig. 9a illustrates the frequency distribution of the average annual number of maintenance days of circumpolar coastal polynyas. Establishing a boundary of 120 d yr−1, we divided the polynyas into infrequent and frequent polynyas (dashed line). In the Antarctic, ~80% of coastal polynyas are infrequent, and about half are open less than 30 days per year. Despite the numerical advantage of these infrequent polynyas, they contribute only a small portion (~20%) of the Antarctic polynya area (black line in Fig. 9a). Nevertheless, the frequent polynyas occupy ~20% of the number and ~80% of the area. Furthermore, we found that there is a significant positive correlation between the polynya maintenance time and area (r = 0.78, p < 0.001).

Fig. 9
figure 9

Comparison between infrequent and frequent coastal polynyas. (a) Frequency distribution (FD) of annual maintenance days of coastal polynyas (bars), and the corresponding cumulative area (black line). Taking 120 d yr−1 as the boundary (grey dashed line), we classified coastal polynyas into low and high frequency, shown in blue and red respectively in all four subfigures. Their basic characteristics are also written in the figure. (b) Weekly total area of infrequent and frequent polynyas. (c) Weekly numbers of polynyas in the two categories. (d) Weekly mean area per polynya.

Full size image

Although both frequent and infrequent polynyas have a significant seasonal area cycle, they differ in their patterns. The seasonal area recovery of infrequent polynyas occurs earlier than for frequent polynyas (Fig. 9b). From August on, infrequent coastal polynyas gradually expand in size, while the area of frequent polynyas does not increase until early September.

The mean area per polynya rather than the quantity could be the primary reason for the seasonal cycle differences in the total polynya area. Quantitatively, the seasonal variations of frequent and infrequent polynyas are similar (Fig. 9c): both briefly increase in early April, followed by a decrease until July, and a gradual recovery in late winter. However, in terms of mean area per polynya, a significant distinction emerges (Fig. 9d). Infrequent coastal polynyas shrink slowly until June (−2 km2 d−1) and then expand rapidly in August (17 km2 d−1), directly promoting the faster recovery of their total area. In contrast, frequent polynyas show a seasonal variation similar to that of the total area (i.e., the red line in Fig. 9b). It is challenging to pinpoint the exact physical mechanisms behind these differences. We examined the mean air temperature and wind speed but found no such pattern (Supplementary Figure 9). We speculate that this difference may come from the frequency of synoptic events or the hydrological precondition, such as the stratification stability.

Usage Notes

Our dataset can be used in many aspects of polynya research. In the technical verification, we preliminary show an example of how DEEP-AA can be exploited: through statistical analysis based on DEEP-AA, we found that there are differences between the seasonal cycles of the area of frequent and infrequent polynyas. Similarly, other differences and variations in polynya characteristics can also be statistically studied based on our dataset. Furthermore, combined with other data, DEEP-AA will help us to better understand the role of polynyas in the earth system. For example, with cyclone track data, DEEP-AA can be used to analyze the response of polynyas to atmospheric events; or with chlorophyll data, we can further discuss the ecological effects of different types of polynyas and their controlling factors. For the convenience of readers with different preferences, in addition to the results based on SIC data with a resolution of 6.25 km/1 d, we also provide the output data for all the different input data and parameters shown in the sensitivity analysis. We hope DEEP-AA, a dataset providing a more comprehensive identification of polynyas than achieved prior to now, can advance our understanding of polynyas.