Multi‐Decadal Record of Sensible‐Heat Polynya Variability From Satellite Optical and Thermal Imagery at Pine Island Glacier, West Antarctica

Open ocean areas surrounded by sea ice and maintained by ocean heat, or sensible‐heat polynyas, are linked to key ice‐sheet processes, such as ice‐shelf basal melt and ice‐shelf fracture, when they occur near ice‐shelf fronts. However, the lack of detailed multi‐year records of polynya variability prevent assessing coupling between polynya and frontal dynamics. Here, we present the first multi‐decadal polynya area record (2000–2022) at Pine Island Glacier (PIG), West Antarctica, from thermal and optical satellite imagery. We found substantial interannual variability in polynya area, with consistencies in the timing of polynya opening, maximal extent, and closing. Furthermore, the largest polynya in our record (269 km2) occurred at PIG's western margin just 68 days before iceberg B‐27 calved, suggesting that polynya size and position may influence rifting dynamics. Our new data set provides a pathway to assess coevolving polynya and frontal dynamics, demonstrating the importance of building long‐term, year‐round polynya variability records.

with Pine Island Glacier ice shelf (PIGIS) when inflows of relatively warm and dense modified Circumpolar Deep Water (mCDW; typically >0°C and >34.7 g/kg absolute salinities; e.g., Dutrieux et al., 2014;Jacobs et al., 2011) reach the grounding zone via bathymetric troughs.mCDW interacts with the base of the ice shelf and generates warm, buoyant, meltwater rich plumes that can incise basal channels that concentrate water flow toward the ice-shelf front (Alley et al., 2016(Alley et al., , 2019;;Mankoff et al., 2012).Depending on plume volume and buoyancy, and local upper ocean stratification, these plumes sometimes reach the surface and can deliver enough heat to melt sea ice and form persistent, sensible-heat polynyas.Sensible-heat polynyas, which require a heat supply to form, are therefore surface expressions of subsurface ice-ocean interactions that, when large enough, can be detected from visible (e.g., Alley et al., 2016), thermal (e.g., Bindschadler et al., 2011;Mankoff et al., 2012;Savidge, Snow, et al., 2023), and microwave (e.g., Markus & Burns, 1993;Massom et al., 2001) remote sensors.However, thermal and visible satellite images have spatial resolutions ∼10-200 times finer than passive microwave, making them ideal for monitoring the km-scale sensible-heat polynyas found at PIG.
Although multi-year and intra-annual records of polynya variability exist for large ice-front latent-heat polynyas, such as those in Pine Island Bay (PIB; Arrigo et al., 2012), and the Amundsen Sea (Arrigo et al., 2012;Macdonald et al., 2023), and open ocean sensible-heat polynyas such as the Weddell Polynya (Gordon & Comiso, 1988;Holland, 2001), the evolution of ice-front sensible-heat polynyas on multi-decadal timescales remains understudied.Here, we build a 22-year record  of sensible-heat polynya area at the PIGIS front with visible and thermal imagery from the MODerate resolution Imaging Spectroradiometer (MODIS) instrument on NASA's Aqua and Terra satellites.We then use this time series of polynya area to investigate multi-decadal and intra-annual polynya trends and the coevolution of the polynya with ice-shelf frontal dynamics.

MODIS Visible and Thermal Imagery
We primarily used MODIS Level 2 Atmospherically Corrected Surface Reflectance visible data (MOD/MYD09 Band 1; 620-670 nm) from the MODIS instruments operating on the NASA Aqua (MYD) and Terra (MOD) satellites to delineate sensible-heat polynyas and quantify their spatial extent near the PIGIS front.When visible imagery was not available (i.e., from ∼April to September when the solar zenith angle is high), we used MOD/ MYD09 thermal Band 31 (10.780-11.280 μm).We selected all MODIS thermal and optical scenes from 2000 to 2022 that contained minimal dense cloud coverage, ensuring visibility of surface features.We excluded scenes where a latent-heat polynya was present (i.e., when a polynya crossed the entire calving front) and scenes where little to no sea ice occupies PIB as sensible-heat polynyas cannot be detected in these scenarios.We manually performed this image filtering step on a scene-by-scene basis which resulted in a total of 342 MODIS scenes used (273 visible, 69 thermal), where >90% of scenes in our record captured at least one sensible-heat polynya >1 km 2 in size.

Polynya Area Time Series
To quantify polynya area we first separated polynya from non-polynya pixels by creating polynya masks for each of the MODIS scenes using pixel selection threshold criteria.For the visible imagery, we selected a digital number (DN) threshold of 3,000 based on a DN distribution analysis on characteristic sample scenes (e.g., Figure S1 in Supporting Information S1); we also tested our thresholding method with ±1,000 DN to determine how sensitive our calculated polynya area was to our selected threshold (Figures S2 and S3 in Supporting Information S1).For the thermal imagery, we used a scene-specific brightness temperature threshold due to the thermal variability induced by thin clouds and other atmospheric effects, which can vary greatly on short temporal scales (hours to days).For example, thermal scene means, standard deviations, and ranges differ by 7.7, 1.5, and 9.5°C, respectively, from 20 to 22 April 2018 for the region of interest (Figure S4 in Supporting Information S1).This variability makes systematic thresholding challenging (e.g., Figure S5 in Supporting Information S1).We therefore opted for manual threshold selection rather than a systematic/automated approach.Scene-specific brightness temperature thresholds were determined visually by selecting the widest and warmest temperature range that captured the entire polynya signal.From the 69 thermal scenes used, this manually selected threshold categorized the warmest ∼5% to <0.1% of pixels in each thermal scene (i.e., 2 to >4 standard deviations from each scene-specific mean) as polynya pixels.
10.1029/2023GL106178 3 of 9 We calculated the total polynya area for each scene by multiplying the total number of pixels in each polynya mask by the area of a single MODIS pixel to generate a time series from our 342 images.Although documented MODIS pixel areas are 0.0625 km 2 for visible imagery and 1 km 2 for thermal (Vermote et al., 2018), pixel dimension increases with zenith view angles (e.g., Campagnolo & Montaño, 2014) causing effective pixel dimensions to deviate from these values depending on where our study area appeared in the 2,330 km wide MODIS swath.Therefore, here we present our polynya area time series calculated from effective pixel dimension; we compare area calculations using nominal and effective pixel dimensions in Supporting Information S1 (Figures S6 and S7).
We subsetted our total polynya area time series by frontal location (i.e., west, mid-shelf ["mid"], east, or other; Figure S8 in Supporting Information S1) to capture polynya spatial variability at the PIGIS front.To calculate each individual polynya's area, we attributed polynya pixels in each scene to a polynya location depending on how close (up to 14, 6.5, and 12 km radius for the west, mid, and east polynyas based on observed polynya sizes) these pixels were to a representative polynya coordinate along the time-evolving ice front (Table S1 in Supporting Information S1).The other polynya category most often captured polynyas that opened offshore after calving events (e.g., Figure S8 in Supporting Information S1), or otherwise captured polynyas westward and detached from the calving front.This polynya category was calculated by taking the difference between the total area and the combined east, mid, and west polynya area for each scene.Merging of polynyas occurs in <1% of our record (e.g., west and other merge; Figure 3d inset), resulting in reduced accuracy in calculating the other, east, and west categories.Merging, however, does not change the calculated total area, and has negligible impact on our analysis.Lastly, we manually verified and if needed, corrected each polynya mask to ensure our attribution of polynya pixels was successful (e.g., removed misidentified pixels when appropriate, which typically occurred when part of PIB was sea ice free, as seen e.g., in Figure 1c).

Multi-Decadal Polynya Variability
We used 342 MODIS images between 2000 and 2022 along the PIGIS front to construct the first multi-decadal record of sensible-heat polynya variability (Figure 2).We observed substantial year-to-year area variability, with the largest total polynya area in 2007 (322 km 2 ).Aside from July to September of 2007, when the total polynya area exceeded 200 km 2 for 43 days, only six other days in the 22-year record had a "total" area greater than 150 km 2 : 27 January 2006 (151 km 2 ), 9 February 2006 (155 km 2 ), 28 February 2006 (162 km 2 ), 1 March 2006 (153 km 2 ), 2 March 2006 (157 km 2 ), and 14 September 2019 (157 km 2 ).Mean total polynya area between 2000 and 2022 was 49 km 2 , much smaller than the >150 km 2 cases.The gap in observations from 2008 to 2017 (Figure 2a) resulted from a lack of landfast sea ice in PIB, which prevented polynyas from existing.
By location, the west polynya was largest on average (mean: 33 km 2 ; range: 269 km 2 ), followed by east (mean: 7 km 2 ; range: 110 km 2 ), other (mean: 6 km 2 ; range: 111 km 2 ), and mid (mean: 2 km 2 ; range: 19 km 2 ).Our observation that the western polynya had the largest mean areal extent is consistent with previous polynya observations covering shorter time periods (e.g., Bindschadler et al., 2011;Mankoff et al., 2012;Savidge, Snow, et al., 2023) and with both modeled and observed intensified outflow from the PIGIS cavity along the western shear margin (e.g., Nakayama et al., 2021;Thurnherr et al., 2014).We note that the other polynya category differs from the other three frontal polynyas (west, mid, east) in that it occurs offshore, not at the calving front edge (e.g., Figure S8 in Supporting Information S1).The other polynyas observed adjacent to icebergs may have different formation mechanisms (e.g., wind stress and/or thermal forcing), but we suspect that basal channels formed prior to calving likely still concentrate relatively warm, buoyant water and can produce sensible-heat polynyas.

Intra-Annual Polynya Variability
To evaluate short timescale polynya variability, we highlight four austral summers from our record where the persistence of landfast sea ice made it possible to observe the full seasonal polynya lifecycle (Figure 3).Even during three consecutive years, we observed important differences in each intra-annual record (Figures 3b-3d), where different frontal polynyas (e.g., west, east, other) dominate the total polynya area.In all 4 years, the peak total polynya area occurred in February (2 February 2001;15 February 2004;3 February 2005;28 February 2006; Figures 3a-3d, respectively).Additionally, the observed polynyas consistently took longer to open (145 days, 161 days, 135 days, and 165 days for 2000-2001, 2003-2004, 2004-2005, and 2005-2006, respectively) than to close (53 days, 38 days, 62 days, and 21 days, respectively) each year, and therefore display a similar trend when visualized with both axes (area and time) normalized (Figure 3e), suggesting a characteristic seasonal growth and decay curve in these four cases.Here, we show four intervals of temporally dense data (i.e., N ≥ 35) that display a clear seasonality, but we include all relevant intervals (labeled I-X in Figure 2) in Supporting Information S1 (Figure S9).

Small-Scale Drivers of Polynya Variability
Polynya area ranges across all locations were large (10-100s of km 2 ) during the 22-year record, especially in the marginal east and west polynyas.These polynyas appeared seaward of shear margins on either side of the PIGIS trunk, locations where basal channels form disproportionally often (Alley et al., 2019(Alley et al., , 2023)).Basal channels can evolve very rapidly; for example, incision rates on the order of 10s of meters per year have been documented on Getz Ice Shelf in West Antarctica (Chartrand & Howat, 2020).Therefore, the observed frontal polynya area variability may arise from uneven channel growth rates across the glacier-with larger channels having the potential to channelize greater volumes of relatively warm, meltwater-rich water toward the PIGIS front.This channelized seaward flow at the ice-shelf base is a precondition to sensible-heat polynya formation, and so basal channel and polynya variability are likely closely linked (e.g., Alley et al., 2019).In addition to basal channel evolution, factors governed by local ice-ocean interaction such as plume volume and buoyancy and local upper ocean stratification may have important controls on polynya variability at the PIGIS front.Furthermore, local atmospheric forcing modulating surface heat fluxes can also drive oceanic variability near PIGIS through changes in near-surface temperature and consequently ocean mixing and circulation (e.g., St-Laurent et al., 2015;Webber et al., 2017), which could impact polynya variability.The observed asymmetry in polynya opening versus closing rates (Figure 3e) may reflect a complex interplay between the competing importance of these processes (e.g., plume volume and buoyancy vs. upper ocean stratification) and potential local ice-ocean feedbacks that may abruptly weaken or periodically terminate such driving processes.

Large-Scale Drivers of Polynya Variability
More broadly, the observed multi-decadal variability may arise from both local ocean/sub-ice-shelf processes (i.e., plume buoyancy, basal channel outflow, upper ocean stratification; e.g., Zheng et al., 2021), local atmospheric forcing (e.g., St-Laurent et al., 2015;Webber et al., 2017), and/or regional atmospheric circulation (Steig et al., 2012;Thoma et al., 2008).Circumpolar Deep Water (CDW) intrusions on the Amundsen Sea continental shelf are strongly related to Amundsen Sea Low (ASL) variability-a Southern Pacific climatological low pressure system (e.g., Turner et al., 2017) that governs wind forcing and drives seasonal on-shelf flow (Thoma et al., 2008).ASL variability is in turn influenced by global climate phenomena such as variability in phases of El Niño-Southern Oscillation (ENSO) and Southern Annular Mode (SAM; Kim et al., 2021;Walker & Gardner, 2017).All of these dynamic and interconnected climatic processes together may govern the volume and properties of mCDW (the polynya's ocean-heat source) delivered to West Antarctica's sub-ice-shelf cavities and grounding zones.Whereas finer spatiotemporal scale differences in polynya area (e.g., differences in west vs. east polynya evolution) are likely more closely related to local processes, broader changes in total area over longer timescales are likely more strongly influenced by regional/global forcing.

Air-Sea Interaction
Persistent polynyas are important for regional and global heat budgets as they represent key sites of consistent interaction between ice-ocean-atmosphere systems.By exposing large ocean surfaces to the atmosphere, polynyas can drive both regional (e.g., ice-front polynyas) and even global (e.g., open-ocean polynyas) heat exchange processes, depending on their spatiotemporal extents (e.g., Gordon & Comiso, 1988).Our intra-annual analysis of polynya variability shows that across four summers (Figure 3), the observed sensible-heat polynyas open at a mean rate of 0.55 km 2 /day (based on total area).Assuming an average sea ice thickness of 1 m (e.g., Xie et al., 2013), roughly 169 TJ/day, or ∼0.002 TW, are required to open and maintain these polynyas (see Supporting Information S1).Although this value is three orders of magnitude smaller than the modeled heat coming into PIB (e.g., ∼6.5 TW; Kimura et al., 2017), the energy required to locally melt sea ice when opening sensible-heat polynyas near PIGIS (∼0.002TW) may be locally important to the energy balance and, when combined with the ocean-atmosphere heat fluxes when these polynyas are open (Morales Maqueda et al., 2004;Smith et al., 1990) may be responsible for a considerable amount of energy flux.Sensible-heat polynyas represent important locations of heat exchange, particularly when scaled to polynya areas across all of Antarctica (Alley et al., 2019), and extended time series of their area can provide a key constraint for ice-ocean-atmosphere heat budget calculations.

Ice-Ocean Interaction
We highlight polynya variability around one calving event that occurred in late September of 2007, where we have 21 cloud-free thermal scenes available prior to the event and 21 cloud-free visible scenes available following calving (Figure 4).The largest recorded individual polynya (western polynya measuring 269 km 2 ) in our 22-year record was captured on 22 July 2007 (Figure 4c), 68 days before iceberg B-27 (714 km 2 ) calved from PIGIS (Figure 4b).This correspondence may suggest that polynya dynamics can influence rift initiation and propagation.We hypothesize that polynya size and the associated relatively warm plume flow, among other factors, may impact calving processes via reduced sea ice buttressing and enhanced ice-shelf melting.We observed that the large sensible-heat polynya in 2007 (Figure 4b) caused the western margin of PIGIS to lose contact with the relatively stagnant and stabilizing marginal ice for >2 months, resulting in sustained and localized reduced buttressing near one margin and not the other, potentially leading to uneven backstress and structural heterogeneity (e.g., Walker et al., 2013) across the PIGIS front.That is, we suspect that large sensible-heat polynyas locally reduce both ice-shelf buttressing (via reduced landfast ice) and shear margin friction (via reduced contact with slower marginal ice), which may lead to instability and eventually contribute to calving.Previous work suggested a similar mechanism that contributed to the 2015 calving event, where in this case, mélange disintegration led to reduced backstress and eventually to complete detachment of PIGIS from the northern margin (e.g., Jeong et al., 2016).Therefore, through their potential to enhance basal melt and reduce resistive stresses imparted on PIGIS (i.e., by removing stabilizing ice), the presence of large persistent sensible-heat polynyas (and associated polynya processes) may be important indicators of PIGIS's calving frequency in the future.However, many factors (or combinations thereof) can cause glacier ice to break-which ones matter (or will matter) most remains uncertain.Building long-term records of coevolving ocean and ice-shelf dynamics is therefore crucial to better constrain the drivers of mass loss.Finally, we also observed a sharp drop in polynya area (Figure 4a) shortly after the 2007 calving event (Figure 4c), which may suggest that calving dynamics can disrupt basal channel outflow by shifting or blocking the channel path and/or upper ocean stratification by mixing the water column (e.g., Meredith et al., 2022).To ensure this area change was not substantially influenced by the sensor change following the 2007 event, we compared areas from thermal and visible images and found they were comparable (Figure S10 in Supporting Information S1).These observations suggest not only that polynya variability may influence ice dynamics, but also that large calving events can influence polynya dynamics (e.g., Cape et al., 2014).Therefore, polynya variability may be indicative of both: (a) ocean-driven basal melt (e.g., Savidge, Snow, et al., 2023;Zheng et al., 2021), and potentially (b) rift initiation processes (i.e., precursor to calving)-two processes that are roughly equally responsible for all of the mass loss observed in Antarctica (Greene et al., 2022;Rignot et al., 2013;Walker et al., 2013).

Summary
Sensible-heat polynyas persist near many ice-shelves in Antarctica and hence have the potential to directly interact with ice-shelves and sub-ice-shelf ocean cavities.However, our understanding of how polynyas evolve on multi-year timescales is limited due to the scarcity of records documenting polynya change.We generated the first multi-decadal record of sensible-heat polynya variability in Antarctica by building a 22-year data set of polynya area at PIGIS using thermal and visible satellite imagery.Although polynya extent was highly variable from 2000 to 2022 (total area ranging from 0 to 322 km 2 ; Figure 2), the polynyas consistently reached their maximum extent in February in 4 years (2000-2001; 2003-2004; 2004-2005; 2005-2006; Figure 3), and on average took more than three times longer to open (152 days) than to close (44 days).
The largest individual polynya (269 km 2 ) in our record occurred on 22 July 2007 at PIG's western margin, just polynyas reach a certain size threshold, their influence on ice-shelf processes becomes significant.That is, not only do large persistent polynyas (e.g., 100s of km 2 ; Figure 4b) indicate enhanced basal melting, but they also reduce ice-shelf buttressing and shear margin friction, which may lead to instability.Although we expect that polynya variability and ice-shelf calving are interconnected processes, future work (e.g., strain analysis surrounding persistent polynyas, rift propagation in relation to polynya persistence) is required to more confidently quantify this potential relationship.Nevertheless, our PIG polynya area record provides a new pathway to explore a panoply of ice-ocean-atmosphere research questions, from polynya influence in heat flux/loss considerations to ice-shelf buttressing and frontal fracture mechanics.

Figure 1 .
Figure 1.Polynyas at Pine Island Glacier (PIG) from MODIS visible imagery on (a) 13 October 2007, (b) 22 September 2017, and (c) 30 December 2022.In panels (a) and (b), polynyas are visible following calving events that resulted in icebergs B-27 and B-44 (labeled) in 2007 and 2017, respectively.The grounding line (Gerrish et al., 2021) is marked by a thin black line.PIB indicates Pine Island Bay.

Figure 2 .
Figure 2. (a) MODIS-derived 22-year record of total polynya area at Pine Island Glacier.Inset shows total polynya area partitioned by total (red), east (orange), west (yellow), mid-shelf (light blue), and other (blue) polynya area.Calving events are marked by vertical gray lines.Panels highlight three intervals of dense data, shown in (b) light pink, (c) magenta, and (d) purple.Triangles and circles (in b-d) denote area calculated from MODIS thermal and visible imagery, respectively.Dotted outlines of the data used in Figures 3 and 4 are shown in (b) and (c).Labels I through X identify the subsetted time intervals in Figure S9 of the Supporting Information S1.

Figure 3 .
Figure 3. Seasonal variability of polynya area at Pine Island Glacier ice shelf in (a) 2000-2001, (b) 2003-2004, (c) 2004-2005, and (d) 2005-2006, illustrating the polynya lifecycle: opening, reaching maximum extent, and partially or fully closing, and (e) all 4 years with both axes normalized (navy, purple, pink, and light blue line), including a smoothed mean (dotted black line).The insets in (a)-(d) show MODIS visible images of the polynyas at their maximum total extent in each year (also shown on the time series by the gray arrow).Colors represent polynya frontal locations (Figure S8 in Supporting Information S1).Note that panels have different vertical scales.

Figure 4 .
Figure 4. Evolution of a polynya through ice-shelf calving.(a) Polynya area time series capturing area before and after the 2007 calving event (vertical gray line) calculated from MODIS thermal (triangles) and visible (circles) imagery.Pink and green vertical lines show when images (b) and (c) are taken.(b) Thermal image on 22 July 2007 shows the greatest total polynya area (269 km 2 ) in our record, 68 days before iceberg B-27 calves from PIGIS.(c) Visible image on 28 November 2007 shows the polynyas following calving.