Wintertime Polynya Structure and Variability From Thermal Remote Sensing and Seal-Borne Observations at Pine Island Glacier, West Antarctica

Antarctica’s ice shelves play a critical role in modulating ice loss to the ocean by buttressing grounded ice upstream. With the potential to impact ice-shelf stability, persistent polynyas (open-water areas surrounded by sea ice that occur across multiple years at the same location) at the edge of many ice-shelf fronts are maintained by winds and/or ocean heat and are locations of strong ice–ocean–atmosphere interactions. However, in situ observations of polynyas are sparse due to the logistical constraints of collecting Antarctic field measurements. Here, we used wintertime (May–August) temperature and salinity observations derived from seal-borne instruments deployed in 2014, 2019, and 2020, in conjunction with thermal imagery from the Moderate Resolution Imaging Spectroradiometer (MODIS) and the Landsat 8 Thermal Infrared Sensor (TIRS) to investigate spatial, temporal, and thermal structural variability of polynyas near Pine Island Glacier (PIG). Across the three winters considered, there were 176 anomalously warm ( $3\sigma $ from background) seal dives near the PIG ice front, including 26 dives that coincided with MODIS images with minimal cloud cover that also showed a warm surface temperature anomaly. These warm surface temperatures correlated with ocean temperatures down to 150 m depth or deeper, depending on the year, suggesting that MODIS-derived surface thermal anomalies can be used for monitoring polynya presence and structure during polar night. The finer spatial resolution (100 m) of TIRS wintertime thermal imagery captures more detailed thermal structural variability within these polynyas, which may provide year-round insight into subice-shelf processes if this dataset is collected operationally.

2014, 2019, and 2020, in conjunction with thermal imagery from the Moderate Resolution Imaging Spectroradiometer (MODIS) and the Landsat 8 Thermal Infrared Sensor (TIRS) to investigate spatial, temporal, and thermal structural variability of polynyas near Pine Island Glacier (PIG). Across the three winters considered, there were 176 anomalously warm (>3σ from background) seal dives near the PIG ice front, including 26 dives that coincided with MODIS images with minimal cloud cover that also showed a warm surface temperature anomaly. These warm surface temperatures correlated with ocean temperatures down to 150 m depth or deeper, depending on the year, suggesting that MODIS-derived surface thermal anomalies can be used for monitoring polynya presence and structure during polar night. The finer spatial resolution (100 m) of TIRS wintertime thermal imagery captures more detailed thermal structural variability within these polynyas, which may provide year-round insight into subice-shelf processes if this dataset is collected operationally.
Circulation of mCDW beneath PIGIS reaches the grounding zone (the region where grounded ice transitions to ice shelf), and the depth at which it subsequently exits the cavity can vary seasonally. Depending on their buoyancy, plumes may flow out of the subice-shelf cavity and into Pine Island Bay (PIB) at depths that coincide with the base of the ice shelf (∼200-400 m depending on local ice thickness; see [11], [12]) or they may rise to the surface [13]. The depth at which plumes exit the subice-shelf cavity depends on upper ocean (less than ∼450-m depth) density and stratification, which vary This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ seasonally. In wintertime, the homogeneous mixed layer is relatively dense due to surface cooling and brine rejection and can extend from the surface down to as deep as ∼400 m on the Amundsen Sea continental shelf [14], allowing more frequent surfacing of meltwater (MW) plumes than in summertime [13]. In summertime, the upper ocean is strongly stratified, causing plumes to spread along isopycnals rather than rising to the surface [13], [15].
The warm, buoyant, MW-rich plumes can incise inverted channels at the base of the ice shelf, called basal channels [16], [17], [18], which provide pathways that concentrate warm buoyant water from the grounding zone toward ice shelf fronts. Depending on basal channel outflow strength and on upper ocean density, buoyant plumes sometimes reach the surface where they can melt sea ice and generate open-water areas (or intermittently thin sea ice areas) surrounded by thicker sea ice, called sensible-heat polynyas. Alternatively, polynyas can also be mechanically opened and maintained by off-shore winds pushing sea ice away from the ice front/coast called latent-heat polynyas; these are linear in shape and typically follow coastline orientation compared to the near-circular sensible-heat polynyas. Polynya formation mechanisms (i.e., ocean heat and wind forcing) are not mutually exclusive, meaning both polynya types can coexist at the same time and place [19]. Sensible-heat polynyas often form in the same location for multiple years (i.e., persistent polynyas) and are collocated with shear margins or subice-shelf channel outlets for ice shelves with cavities flooded by relatively warm water (i.e., warm-cavity ice shelves) [16], [17], suggesting that basal channels are likely intrinsically related to polynya formation processes. Furthermore, basal channels are important ice-shelf features that substantially change basal melt patterns (see [20], [21]) and may influence ice shelf stability through fracture [22], particularly when channels occur in already-weak shear margins [17], as is the case for PIG. However, basal channel, plume, and polynya variability and coevolution remain relatively unexplored.
Persistent polynyas are key sites of consistent, year-round interactions between atmosphere, sea ice, ocean, ice shelves, and subice-shelf ocean cavities that drive Southern Ocean carbon dynamics [23], [24]; however, in situ observations are sparse due to the logistical constraints of collecting field measurements given the remoteness of Antarctica [25]. Instead, visible (see [16], [26]), thermal (see [27], [28]), and microwave (see [29], [30]) remote sensing techniques have been used to identify and quantify polynya processes in coastal Antarctica. Although multiple remote sensing techniques can identify the presence of polynyas within sea ice, thermal remote sensing, with spatial resolution ∼10-25 times finer than passive microwave [31], has the unique potential to fingerprint regions of high sea surface temperatures within PIB, which can be used to infer processes that govern sensible-heat polynya evolution. For example, warm thermal anomalies were consistently identified in summertime at the PIGIS front, where both persistent polynyas and strong basal outflow occur [18], [32]. However, although satellite thermal infrared measurements are spatially extensive, they only measure the temperature of the upper ∼10 µm of the ocean [33].
Here we used thermal infrared satellite imagery from the Moderate Resolution Imaging Spectroradiometer (MODIS) and Landsat 8 Thermal Infrared Sensor (TIRS) combined with seal-borne hydrographic ground-truth measurements to study wintertime polynya processes at PIG. Recently, seal-borne oceanic measurements in the Amundsen Sea have provided a substantial hydrographic dataset [14] that allows for both wintertime (when sea ice is more extensive) and more frequent observations than are available via ship-borne methods (largely limited to sea ice-free conditions). These seal-derived hydrographic data provide some of the first vertical measurements of polynya/plume structure in the Antarctic (see [13]). Therefore, combining these remote sensing and in situ datasets allows for a more comprehensive understanding of how polynya processes evolve year-round. Coupling these datasets, we found that MODIS thermal anomalies can be used for monitoring polynya structure and evolution during polar night and that the finer spatial resolution (100 m) of Landsat 8 wintertime thermal imagery captures detailed thermal structural variability within these polynyas, which may provide insight into subice-shelf processes during polar night if these scenes are collected operationally. Acquiring high resolution wintertime ocean thermal measurements near Antarctica's ice shelves in polar winter is crucial for our year-round understanding of polynya and frontal processes. This effort becomes particularly important when ocean field records are sparse, as is often the case in Antarctica.

II. METHODS
A. Datasets 1) MODIS Thermal Data: We used MODIS Level 2 Atmospherically Corrected Surface Reflectance thermal data (MOD/MYD09 Band 31; 10.780-11.280 µm) from the MODIS instrument operating on NASA Aqua (MYD) and Terra (MOD) satellites to investigate the sensible-heat polynyas near the PIG ice front. We use MOD09 brightness temperature in lieu of absolute temperatures because sea surface temperatures are particularly difficult to retrieve in polar regions (see [34]). The temperatures we provide have not been corrected for surface emissivity; however, the thermal infrared emissivity of ocean water is close to that of a blackbody so this correction would not substantially change the temperatures we provide here. MODIS thermal images are taken at high temporal resolution at the Earth's poles (>10 per day), making them ideal for monitoring the temporal evolution of sensible-heat polynyas; however only 1-3 wintertime images per day are processed into the MOD/MYD09 datasets due to errors arising from cloud shadow, high aerosol content, high view angle, or high solar zenith angle [35]. The spatial resolution of MODIS thermal images ranges from 1-km pixels directly at nadir to ∼4 km at the across-track edges of the footprint. As polynyas at PIG can be several kilometers across, we investigated whether MODIS data are adequate to monitor the spatial evolution of polynyas. We selected all of the MODIS thermal images obtained between May and August (constrained to these months to match the time period of wintertime seal-tag data availability near PIGIS) of 2014, 2019, and 2020 that contained minimal dense cloud coverage; due to the difficulty of implementing effective cloud filtering algorithms during polar night (see [36], [37]), we selected only images where surface features were clearly defined in the thermal imagery, which indicated that clouds were at a minimum. We performed this step manually on a scene-byscene basis.
2) Landsat 8 Thermal Data: We used Landsat 8 Collection 2 Level 1 Systematic Terrain Correction (L1GT) Band 10 (10.6-11.19 µm) imagery acquired during the 2019 austral winter to investigate the sensible-heat polynya signal at a finer spatial resolution. L1GT is a brightness temperature product (i.e., not calibrated surface temperature) as there is currently no calibrated sea surface temperature product from Landsat [38]. Landsat 8 does not operationally collect data during polar night, and so this austral winter dataset was collected through a data acquisition special request. Landsat 8 thermal bands have much finer spatial resolution (100 m) than MODIS thermal bands (1 km), but, during operational data collection, have lower temporal resolution (one scene collected every 3-5 days).
3) Seal-Borne Oceanographic Data: We used temperature and salinity profiles measured during austral winter (May to August) by seal-borne conductivity-temperature-depth-satellite relayed data loggers (CTD-SRDLs) deployed in 2014, 2019, and 2020 [39]. Although limited reference data in remote locations may degrade data quality slightly [40], each CTD-SRDL recorded ocean temperature and salinity with an accuracy better than 0.005 • C and 0.02 (using the practical salinity scale), respectively, making seal-tag measurements a reliable means of collecting hydrographic data. CTD-SRDLs were temporarily attached to a seal's head and recorded conductivity, temperature, and pressure at 1 Hz [40] across the Amundsen Sea (a region >100 000 km 2 ). Seal positions estimated to be inland were adjusted using standard repositioning algorithms (see the Supplementary Material for method). Onboard processing reduced each dive profile into 17 or 18 depth levels following the methods of [13]. This data reduction step was meant to maximize data retention where the largest vertical conductivity and temperature gradients occur in the water column and to minimize data volume to transfer via the bandwidth-limited Argos satellite-communication system [40]. Only the deepest dive within every four-hour period was transmitted to ensure the best possible spatial and temporal resolution for the limited battery power available [40]. For each CTD profile, we vertically interpolated the data to 1 m intervals using a piecewise cubic hermite interpolating polynomial [42].
The 2014 hydrographic dataset was collected by 14 seals tagged with CTD-SRDLs during the UK's Ice Sheet Stability Program (iSTAR) JR294/295 cruise in February 2014 onboard the RRS James Clark Ross [14]. Hydrographic datasets in 2019 (12 tagged seals) and 2020 (12 tagged seals) were collected with similar methods during cruises NBP19-02 and NBP20-02 onboard the RVIB Nathaniel B. Palmer [43], as part of the International Thwaites Glacier Collaboration: Thwaites-Amundsen Regional Survey and Network (ITGC: TARSAN).
B. Data Processing 1) Polynya and Background Profile Partitioning: Hydrographic profile measurements that sample sensible-heat polynyas record warmer near-surface ocean temperatures than those that do not [13], [18]. We therefore statistically partitioned "warm" polynya profiles and "background" profiles to detect polynyas based on the ocean temperature in the upper 20 m of the water column, where we have dense CTD sampling despite the inherent vertical data reduction that occurs for each profile (see Figs. S1 and S2). We estimated an annual wintertime-mean background surface temperature and standard deviation using May-August upper 20 m ocean temperatures in the center of PIB (pink dashed box in Fig. 1), excluding profiles directly adjacent to or west of the ice front, where warm plumes commonly impact surface temperatures [13], [18], [32]. Capturing annual wintertime-means ensures that we account for interannual temperature variability in PIB. We used three standard deviations warmer than mean annual background surface (i.e., upper 20 m) ocean temperatures as a threshold for delineating warm polynya profiles, and labeled the remaining profiles as background. This method therefore defines warm polynya profiles where surface temperatures are warmer than 99% of the background surface temperatures.
2) Meltwater Content: We followed the method presented in [13] to calculate MW content from seal-tag hydrographic profiles. This method infers the fraction of glacial MW at each observed location assuming the ocean in this region is composed of a linear mixture of the three wintertime PIB water masses (mCDW, Winter Water [WW], and MW). We used conservative temperature, , and absolute salinity, S A , as endmembers and assumed they are both conservative. We determined the mCDW endpoints for each year by extracting the maximum seal-derived and S A that fall within an expected range for mCDW in PIB (i.e., S A > 34.7 g/kg and > 0.5 • C) from all May to August seal-tag data. WW is formed in the winter due to strong winds, sea ice formation, and surface cooling and always has a temperature near the in situ freezing point (∼−1.86 • C [13], [44]); therefore, we determined the WW S A endpoints for each year by extracting the highest observed salinity that lies on the freezing line (where = −1.86 • C). We used MW endpoints from [13].
Uncertainty in the endpoints may be as much as 30% of the averaged wintertime near-surface MW content due to seal-tag hydrographic measurement uncertainties [13]. Our seal-tag data captured stable wintertime mCDW properties, so errors originating from the mCDW endpoint selection are negligible. Following [13], we estimated the error in the WW S A endpoint to be ± 1.1 g/kg using a Monte Carlo simulation on a set of 1000 randomly generated hydrographic measurements (with criteria: > −1.9 • C and MW content >25 g/kg) with 2000 different WW endpoints (normally distributed around 34.21 ± 0.05 g/kg). If the above procedure is replicated, these uncertainties may change the values of the calculated MW content we present, but no qualitative change should be expected because the uncertainties calculated are much lower than the difference found between the background and warm profile MW content (see Section III-C). 3) Surface-Subsurface Water-Column Correlations: Following [45], we used seal-derived temperature fields to investigate to what extent surface temperature data could be used for characterizing subsurface ocean temperatures. We quantified the depth at which subsurface temperatures decorrelated from the surface layer for each warm profile by comparing seal-derived mean temperatures at 50 m depth intervals down to 300 m (e.g., T 50-100 corresponds to the mean temperature within the 50-100 m layer) to the upper 50 m mean temperature, T 0-50 . We estimated correlations between near-surface (T 0-50 ) and mixed-layer temperatures at each depth interval and identified all statistically significant (p < 0.05) correlations.

A. MODIS Observations
In all three winters, we observed relatively warm surface temperatures (up to −3.0 • C) near the PIG ice front [see Fig. 1(d)-(f)]. These warm areas appear as either circular features or linear features following coastline orientation. The circular/subcircular surface temperature anomalies (∼1 to >8 km in diameter, 1 • C to 10 • C warmer than mean surface temperature) are collocated with known locations of basal channels [16]. Therefore, we followed existing literature (see [16], [17], [18]) and attributed these surface temperature anomalies to sensible-heat polynyas. Relatively warm surface temperature anomalies that cross the entire PIGIS calving front (∼1 to >20 km in width; see Fig. 2) were attributed to latent-heat polynyas; MODIS observations showed that latent-heat polynyas can open quickly (hours to days) and that overlapping sensible-and latent-heat polynyas can be distinguished based on the magnitude of the thermal anomaly (see Fig. 2). The warmest wintertime surface temperatures were almost always recorded near the western shear margin, and sometimes also appeared near the middle and eastern shear margin of PIGIS where basal channel outflow has previously been identified [16], [18], [32].

B. Landsat Observations
There were 58 TIRS scenes that covered the PIGIS front from May 2019 to August 2019. We show two of these scenes that overlapped with MODIS imagery (where cloud cover was minimal in both TIRS and MODIS scenes) and that coincided with seal-borne measurements (see Fig. 3). Consistent with MODIS observations, we observed warm TIRS brightness temperature anomalies near the PIG ice front with the warmest signal occurring near the western shear margin. The warm signal near the western shear margin often extended west of the shear margin and paralleled the coast (see Fig. 3, red dotted box), consistent with surface current flow estimates (see [47]). The finer spatial resolution of TIRS imagery (100 m) revealed a more detailed picture of ocean-surface thermal variability, including a cluster of warmer pixels near the middle of the ice front on May 27, 2019 (see Fig. 3, white arrow), located near where a persistent polynya has previously been observed (see [18], [32]). Darker linear features and textured circular areas are visible on the May 27, 2019 imagery (see Fig. 3, orange arrows) and are consistent with types of sea ice that occur early in the ice-formation process (e.g., grease ice, pancake ice; see [48]).
The difference in the time of acquisition between MODIS and TIRS scenes (9 h 10 min difference on May 25; 7 h 33 min difference on May 27) resulted in spatial differences in the thermal data. On May 25 for example, the width of the latent-heat polynya is ∼8 km wider in the MODIS scene than in the TIRS scene taken ∼9 h later [see Fig. 3 (top row)], indicating that the polynya opened at a rate of ∼0.9 km/h. These spatial differences arise because the TIRS image was taken earlier on (7:15 A.M. UTC) in the opening of the latent heat polynya (i.e., sooner after the onset of offshore winds), relative to the MODIS image (4:25 P.M. UTC). Cloud presence also creates thermal variability between scenes.

C. Seal-Derived Observations
We used a total of 1229 CTD profiles collected during seal dives in wintertime near the front of PIG [see Fig. 1(a), orange box]. Applying our profile-partitioning statistical method, we found 462, 325, and 261 background profiles in 2014, 2019, and 2020, respectively. From these profiles, the calculated temperature thresholds that partition warm profiles from background profiles were found to be −1.54 • C, −1.69 • C, and −1.74 • C for each respective year. This thresholding resulted in 66, 72, and 38 warm profiles in 2014, 2019, and 2020, respectively.
We determined and S A endpoints for mCDW and WW for each year following our MW content calculation method (see Table I). We were unable to calculate the 2020 WW endpoint because no hydrographic data captured conditions near the PIGIS front with temperatures at the freezing point that year; we used the 2019 WW S A endpoint instead. Using these water mass endmembers, we calculated the wintertime MW content in each year (see Fig. 4).
In all months where we had multiple warm profiles near the front (N warm = 7-50 per month), there was a consistent relationship between temperature and calculated MW: warm profiles corresponded to greater MW content relative to the background profiles in the upper ∼300 m (see Figs. 5 and 6). This temperature-MW relationship is consistent with the observations in [13], which were made for 2014 only. The difference in temperature and in MW content between background and warm profiles was usually greatest near the surface Corresponding seal-derived surface temperatures are overlaid for each day. Mismatch between MODIS-and seal-tag-derived absolute surface temperature retrievals is discussed further in Sections III-D and IV-C. Each panel shows the area within the white dashed box on each inset on a narrower temperature scale to highlight the thermal structure within the latent-and sensible-heat polynyas. The red dotted box highlights relatively warm ocean surface temperatures that often extend west of the western shear margin parallel to the coast and are consistent with surface current flow estimates (see [47]). The white arrow indicates warmer surface temperatures near the middle of the ice front, where strong basal channel outflow has been previously identified [16], [18], [32]. The orange arrows indicate linear features and circular textured areas, consistent with types of sea ice present early in the ice-formation process.
(on average 0.4 • C and 5.4 g/kg, respectively) and decreased through the water column until the water column becomes more homogeneous at depth, dominated by mCDW.
The depth interval at which ocean surface temperatures were significantly (p < 0.05) correlated with temperatures at depth varied between years (see Fig. 7). Surface temperatures were correlated with subsurface temperatures down to 150 m depth in 2014 and to 200 m depth in 2019 and 2020 (see Table II) Although these surface-subsurface correlations were statistically significant (p < 0.05), they were sometimes weak (R 2 < 0.3), suggesting that only the 50-100 m layer was both significantly and strongly (e.g., R 2 > 0.55) correlated with the surface layer in 2014 and 2019, for example.

D. Overlapping Seal-Borne and Remote Sensing Observations
Of the 176 anomalously warm profiles near the PIGIS front during the three winters, 26 profiles (across 11 days) We compared the seal-tag-derived surface temperatures to the warmest MODIS surface temperature within 6 km of each seal dive. We selected a 6 km distance threshold so that the threshold was greater than one MODIS pixel and extracted the warmest temperature within the threshold. This process  ensured that we extracted the closest temperature to that of the ocean given that each pixel may represent a mixture of (relatively warm) sea surface temperature and (cold) cloud or sea ice temperature due to the surface heterogeneity at polynya edges; we also tested thresholds between 5 and 7 km, which did not substantially change our results. However, we found no clear correlation between the extracted MODIS-and seal-tagderived surface ocean temperatures when performing this comparison between datasets (see Fig. 8). The extracted MODIS temperatures have a much larger range (42 • C) than that of the seal-tag-derived surface temperatures (1.1 • C) for the same days and sampling locations (±6 km). Additionally, across all 26 data pairs, the warmest extracted MODIS temperature is −3.0 • C, whereas seal-tag data record surface ocean temperatures ranging from −1.9 • C to 1.1 • C in this region, indicating discrepancies between the two datasets.

A. Remote Sensing Identification of Polynyas
By combining seal-tag-derived anomalously warm surface ocean temperatures and remotely-sensed surface temperature anomalies, we showed that MODIS and TIRS can be used to identify and monitor sensible-heat polynyas near the PIGIS in wintertime. We found that both MODIS and TIRS have high enough spatial and temporal coverage to monitor persistent polynyas at PIGIS and can distinguish between latent-and sensible-heat polynyas based on the spatial extent, shape, and magnitude of the thermal anomaly. Furthermore, for all months considered, locations where we observed a relatively warm temperature signal in the thermal imagery coincided with warm seal-tag-derived near-surface temperature measurements. Seal-tag data indicated that these anomalously warm wintertime surface ocean temperatures can be used as a proxy for subsurface ocean temperatures, down to 150-200 m depth, rather than solely being representative of ocean-surface processes. These findings suggest that remotely-sensed thermal anomalies near polynyas can be used for monitoring sensible-heat polynya spatial (horizontal and vertical) and thermal structure during polar night. Our results are likely applicable to all warm-cavity ice shelves.
Furthermore, we found that regions of relatively high surface temperature near PIGIS have a high MW content (see Fig. 6), consistent with the ocean properties found in the PIG persistent polynyas in 2014 [13]. The buoyant MW plumes beneath PIG are turbulent and entrain mCDW as they rise; they carry enough heat to drive basal melting and sometimes advect the residual heat (i.e., heat not consumed during ice melt) to the ocean surface and melt openings in the sea ice that we observe from satellites (see [18]). These polynyas therefore provide windows into processes occurring in the subice-shelf environment and directly represent the integrated interactions of mCDW with the ice-shelf base from the grounding zone to the ice front. The potential for continuous high-resolution monitoring of sensible-heat polynya thermal structure and variability provides a unique opportunity to identify and track the thermal signature of these polynyas in wintertime at PIGIS and elsewhere on the Antarctic coast in regions with similar hydrographic conditions.

B. Unique Potential of TIRS
The finer spatial resolution (100 m) of Landsat 8 wintertime thermal imagery captures detailed structural variability within these polynyas [see Fig. 3 (left column)], which may provide deeper insight into subice-shelf processes during polar night if this dataset is collected operationally. Additionally, this fine resolution imagery may result in more reliable polynya Fig. 5. Seal-tag-derived temperature profiles near PIGIS front. In all months where there are multiple warm profiles near the ice front, the warm surface temperature anomaly extended down to ∼400 m depth. Blue and red lines represent background (bgd) and warm temperature profiles, respectively, with the corresponding bold lines representing the monthly means for each. N warm and N bgd indicate the monthly warm and background profile count near PIGIS (see Fig. 1, orange box).
absolute surface temperature retrievals than currently available with MODIS. The thermal structure of the polynya signal at the surface can help identify where plume thermal advection and/or basal channel outflow may be strongest, among other influencing factors [16], [18]. In May 2019, and consistent with published literature, we observed a concentration of warm water near the western shear margin of PIG, where strong basal channel outflow has previously been identified [16], [18] from both Landsat and MODIS thermal imagery. However, temperature gradients and plume edges are more distinct in Landsat TIRS, illustrating its potential use in long-term investigations of polynya thermal structure.

C. MODIS and TIRS Limitations
Although satellite thermal detectors provide near-continuous polynya detection capabilities in wintertime at PIG, some TIRS, MODIS, and seal-tag CTD instrumental constraints provide barriers to building a direct proxy between remotely sensed temperatures and surface ocean temperatures. We found that MODIS-derived wintertime surface ocean temperatures near PIGIS did not strongly correlate with seal-derived surface temperatures (see Fig. 8). The absence of a correlation between both surface temperature measurements may be largely attributed to the coarse spatial resolution of MODIS (1 km), meaning that surface temperatures are integrated over too large an area in PIG's complex coastal margin to provide a  meaningful estimate of ocean thermal properties. Instead these observations likely reflect an average temperature observation from a mixed pixel (e.g., any combination of cloud, grease ice, pancake ice, thick sea ice, part of the calving front, open ocean), resulting in a "cold bias." Such "cold bias" has been demonstrated in the Arctic (see [34]), but until now had not been shown in the southern high latitudes. This issue may be overcome by applying subpixel retrieval methods, such as spectral unmixing or multisensor data fusion (see [49], [50]), which may be able to separate the thermal contributions of each constituent to the infrared emissions of the larger pixel. Although TIRS measurements have a much higher spatial resolution, there is no calibrated Landsat sea surface temperature product for Antarctica [38]. Hence, we cannot assess TIRS measurements of polynya thermal structure and instead can only discuss relative differences in brightness temperature. Additionally, some of the observed inconsistencies between datasets may arise in part due to seal-tag positioning errors, which have been estimated to be ∼4 km (see [51]). Furthermore, even when measurements are taken at the same location in the ocean, discrepancies between surface temperature measurements from seal-tag and MODIS datasets may occur due to large temporal variability on subdaily timescales [see Fig. 3 (top row)] caused by rapid wintertime surface cooling, frazil sea ice formation, fast surface currents, advection and obscuration by nearly transparent clouds, and intense wind stirring.

D. Future Use of Thermal Imagery for Sensible-Heat Polynya Monitoring
By linking MODIS and TIRS thermal data to seal-tag measurements we show that we can monitor thermal variations from space, year-round, at high temporal resolution. However, only MODIS currently captures images year-round in polar regions and few of these images are processed into the MOD/MYD09 datasets (due to errors arising from cloud shadow, high aerosol content, high view angle, or high solar zenith angle; see [35]). Further, as a result of its coarse spatial resolution, MODIS cannot resolve the persistent polynyas around Antarctica that are smaller than those at PIG. Landsat 8 and the recently launched Landsat 9 have the spatial resolution to detect these small polynyas, but only acquire images during polar winter if special acquisitions are requested for specific image tiles, leaving most of the rest of Antarctica-including small polynyas-unobserved during wintertime. Consistent high spatial and temporal resolution wintertime thermal imagery of the poles would provide invaluable insight for tracking ice-ocean interactions and changes. Such data would reduce how heavily we currently rely on the scarce ocean record in the critical Amundsen Sea sector of Antarctica (see [52]). Therefore, we recommend that Landsat 8/9 and future missions with high spatial resolution and high radiometric fidelity thermal instruments begin operational wintertime thermal data collection at the high latitudes, particularly in the complex and dynamic coastal margins.
Access to high-resolution wintertime thermal imagery would allow not only continuous monitoring of sensibleheat polynyas, but also characterization of subsurface mixed layer temperature in regions where we observe sensible-heat polynyas. The correlation depth between surface (polynya) and subsurface (plume) anomalously warm temperatures (found here to be down to 150 m depth) is strongest in the winter, where a homogenous mixed layer allows plumes to more frequently rise to the surface, relative to summertime [13]. These seasonally controlled dynamic processes hence allow us to fill the gap of the year-round sensible-heat polynya monitoring by inferring subsurface temperature conditions from remote sensing observations in winter.

V. SUMMARY
Sensible-heat polynyas appear near many ice shelves in the Amundsen Sea, but we have limited wintertime observations of how these polynyas evolve and interact with ice shelves. Here, we linked wintertime thermal remote sensing from MODIS and TIRS with unique in situ seal-tag hydrographic measurements to investigate the spatiotemporal, vertical, and thermal structural variability of sensible-heat polynyas near PIGIS. Pairing these independent observations, we presented the first study that combines wintertime remote sensing and seal-borne observations to investigate polynya processes in Antarctica.
Our work demonstrates that MODIS thermal anomalies can be used to monitor polynya thermal structure and polynya processes during polar night; this finding becomes particularly important when we have limited access to ocean records, as is often the case in Antarctica. Both MODIS and TIRS provide high enough spatial and temporal coverage of polynyas to be used to distinguish between latent-and sensible-heat polynyas based on the spatial extent, shape, and magnitude of the thermal anomaly. Furthermore, we show that the finer spatial resolution of Landsat 8 wintertime thermal imagery captures structural thermal variability within these polynyas in greater detail than MODIS; this fine-scale mapping may provide insight into plume dynamics and subice-shelf processes during polar night if this dataset is collected operationally and can be developed into a calibrated sea surface temperature product. Access to calibrated surface temperatures from Landsat 8 would likely yield a better correlation between seal-borne and space-borne surface temperatures at PIG; this could allow for wintertime characterization of subsurface ocean temperatures down to 150-200 m depth (above which depth near-surface ocean temperatures remain correlated through the subsurface) near PIGIS, from remote sensing observations alone. This MODIS and TIRS pilot study therefore shows promise that high spatial resolution ocean thermal measurements can be used for year-round monitoring of polynya, ice-front, and subice-shelf processes in Antarctica. Our analysis broadens our limited understanding of wintertime ice-ocean processes in this region, where most of our current observations are from summertime, and where ice shelves are particularly vulnerable to oceandriven changes.

ACKNOWLEDGMENT
Logistical support for data collection was provided by NSF-U.S. Antarctic Program and NERC-British Antarctic Survey. ITGC Contribution No. ITGC-089. The data (https://doi.org/10.5281/zenodo.7696898) and code (https://doi.org/10.5281/zenodo.7843547) that support the findings of this study are openly available. We would also like to thank the editor and anonymous reviewers, whose comments improved the manuscript.