Alongshore winds force warm Atlantic Water toward Helheim Glacier in southeast Greenland

,

due to highly variable ocean circulation patterns between the fjord and the continental shelf (Jackson et al., 2014;Straneo et al., 2010).Within Sermilik Fjord, relatively cold (<4°C) and fresh Polar-origin water (PW) resides at the surface above AW, forming a two-layer circulation structure (Straneo et al., 2010;Sutherland et al., 2014).In summertime, fjord circulation is predominately glacier-melt driven with flow in the surface layer away from the glacier and inflow in the AW layer.However, the dominant variability mode in all other seasons is an oscillatory "intermediary" circulation caused by wind-driven coastal geostrophic currents and changing offshore watermass properties (Jackson et al., 2014;Straneo et al., 2010;Svendsen & Thompson, 1978).Under this highly dynamic circulation scheme, alongshore northeasterly (prevailing) winds drive onshore surface Ekman transport   1b) and create a compensating offshore flow at depth (Håvik & Våge, 2018), resulting in coastal downwelling on the continental shelf (Figure 1b).During such downwelling events (blue and red arrows in Figure 1b), the sea surface height on the continental shelf can rise ∼0.15 m (Cowton et al., 2016;Jackson et al., 2014), isopycnals heave downwards (Jackson et al., 2014;Straneo et al., 2010), and the near-surface PW layer thickens.Opposing southwesterly winds drive the opposite set of changes (Figure 1c).Ocean pressure gradients between the coastal waters and fjord, created by coastal downwelling or upwelling, drive rapid  ((ℎ)) current reversals within intermediate layers of Sermilik Fjord on synoptic timescales of 4-10 days (Jackson et al., 2014).These strong intermediary currents can flush the upper 300 m of the fjord within days (as opposed to >30 days from melt-driven circulation) and drive large water and heat exchanges between the fjord and continental shelf (Sciascia et al., 2014;Straneo et al., 2010).The wind-driven circulation can either lead to water property variations within the fjord or result in only oscillations with no net water property change (Jackson et al., 2018).
It is unclear what conditions drive one outcome or the other.
Outside the fjord, a complex circulation system allows offshore AW below about 300-400 m to intrude onto the continental shelf along a bathymetric trough (∼15 km wide, ∼400-900 m deep) in an onshore flow of  (1) Rossby number that leads to Sermilik Fjord and Helheim Glacier (Figure 1a; Harden et al., 2014;Snow et al., 2021;Sutherland et al., 2013).On the continental shelf, the East Greenland Coastal Current (EGCC; nearest shore) and East Greenland Current (EGC; further offshore) flow at the surface along the coast carrying PW equatorward (Sutherland & Pickart, 2008), and AW spreads onto the shelf beneath them.Offshore from the continental break, the Irminger Current (IC) carries AW southward throughout the upper water column (Johannessen et al., 2011;Rudels et al., 2002;Våge et al., 2011), which can be diverted onto the shelf (Harden et al., 2014;Snow et al., 2021;Sutherland et al., 2013).Aside from general inflow at depth, full-depth AW inflow occurs on synoptic timescales (2 days to 2 weeks; hereafter referred to as "rapid") along the trough or as seasonally varying (long timescale) inflow across portions of the continental shelf.Intrusions of AW may be linked with EGCC/EGC variability (Harden et al., 2014;Murray et al., 2010;Sutherland et al., 2013), cyclonic eddies (Bruce, 1995;Sutherland et al., 2013;Sutherland & Pickart, 2008), tidal variability, or fluctuations in alongshore winds (Hampson, 2020;Harden et al., 2014) that stimulate coastal-trapped waves (CTWs) (Fraser et al., 2018) coherent for hundreds of kilometers along the shelf (Gelderloos et al., 2021).However, little work has been done in the Sermilik continental shelf region to link the open ocean and fjord during these events, so the drivers of the intrusions and linkages to water property changes nearshore remain unclear.
Rapid intrusions of offshore waters onto continental shelves have been observed and modeled to occur along bathymetric valleys, canyons, and troughs around the globe (Cottier et al., 2005;Fraser & Inall, 2018;Kämpf, 2006;Lentz et al., 2014;Nilsen et al., 2008).Field observations at some locations (e.g., Hudson Shelf Valley, West Spitsbergen Shelf) have demonstrated that coastal upwelling can induce full-water-column intrusions from offshore onto the continental shelf with velocities of >0.2 m/s (Cottier et al., 2007;Inall et al., 2015;Lentz et al., 2014).The flow speed of these intrusions varies regionally as a result of ocean stratification, Coriolis forcing, wind stress, and other regional differences (e.g., Nilsen et al., 2008;Ramos-Musalem & Allen, 2019).Numerical models suggest these inflows result from onshore pressure gradients (often wind-driven) (Kämpf, 2006;Klinck, 1996;She & Klinck, 2000) and are associated with CTWs generated at the bathymetric depression (e.g., Allen & Madron, 2009;Johnson, 1978;Kämpf, 2012;Killworth, 1978;Zhang & Lentz, 2017).The flow responses to along-shelf winds of opposite directions are asymmetric (i.e., strong offshore flow does not occur during coastal downwelling) (Inall et al., 2015;Lentz et al., 2014;Zhang & Lentz, 2017), and strong stratification can restrict the vertical extent of the onshore flow to the deeper portions of the water column (Ramos-Musalem & Allen, 2019).Observations of these intrusions are sparse, and most hydrographic and mooring measurements lack the combined spatial (horizontal or vertical) and temporal coverage required to contextualize the events.They have also never been observed on the Greenland continental shelf.Previous work in Sermilik Fjord (Jackson et al., 2014(Jackson et al., , 2018;;Straneo et al., 2010) and along Greenland's southeastern coast (Harden et al., 2014;Le Bras et al., 2018;Oltmanns et al., 2014;Sutherland & Pickart, 2008) has shown strong correlations between ocean circulation patterns and various wind drivers, including cyclones, barrier winds (high wind speed flow parallel to the coast), and piteraqs (hurricane-intensity downslope winds in the offshore direction); these relationships support the notion that winds may drive intrusions along continental shelf troughs in this region.
Here, we show the first remotely sensed evidence of rapid intrusions of offshore water along the Sermilik Trough leading to Sermilik Fjord.Temperature contrast between warm offshore AW and cold coastal PW make detection of the intrusion by remote-sensing feasible.We use MODerate Resolution Imaging Spectroradiometer (MODIS) sea surface temperatures (SST) and optical imagery of the area to characterize AW intrusions and infer their effect on the water column based on observed patterns.Unlike previous work on AW variability within Sermilik Fjord and near the fjord mouth, we provide a continental shelf-wide view of the AW intrusion and its variability from 2010 to 2013 with coincident ocean mooring observations.We show that coastal upwelling, often associated with upwelling-favorable winds generated by the passing of a cyclone and subsequent sea surface lowering, likely drives these intrusions and brings warm water into Sermilik Fjord at depth.This work demonstrates that remote sensing and reanalysis data provide an accessible approach for studying these wind-ocean features in the Greenland coastal region where their impact on glacier mass loss may be important, but in situ ocean observations are rare or nonexistent.

Satellite Imagery
We initially identified AW intrusion events along Sermilik Trough using MODIS optical imagery from NASA WorldView (MOD02HKM/MYD02HKM Terra/Aqua Level 1B Calibrated Radiances, 500 m resolution; MODIS Characterization Support Team, 2017) to characterize these intrusions and determine their frequency and variability in time (Table S1 and Figure 2).The key observable in MODIS imagery that indicated an intrusion event was the ice-free ocean surface cross-cutting the EGCC toward shore along the Sermilik Trough.To ensure we consider ocean features solely associated with these along-trough intrusions, we identified the most distinctive (i.e., most obviously ice-free ocean surface) AW intrusion events and selected intrusions that were clearly associated with the circulation path leading from the continental shelf break to Sermilik Fjord and either: 1. Travel inshore along the Sermilik Trough between 38.2°W and 37.2°W to at least 65.3°N (inner extent of the trough), or 2. Cut more than 15 km inshore at ∼65°N (outer extent of the trough) from the sea ice edge along the trough when the EGCC was wide and covered most of the continental shelf.
Less distinctive sea ice-free waters that do not clearly cut across the EGCC may be attributable to other ocean processes (e.g., eddies traveling south along the continental slope) and do not necessarily indicate intrusions of AW.They are not counted in our analysis.We considered clusters of intrusions that have fewer than 3 days of separation to be one event.We analyzed the satellite observations during 2010-2013 when simultaneous moored temperature records existed for the area and only for January through June when EGCC sea ice cover makes the intrusions readily identifiable.We use MODIS SSTs from the Group for High Resolution Sea Surface Temperature (GHRSST) Level 2P Global Sea Surface Skin Temperature v2019.0 products from the NASA Aqua and Terra satellite (JPL/OBPG/RSMAS, 2020a, 2020b; Kilpatrick et al., 2015) to determine the surface ocean temperature during the intrusions.The GHRSST products provide SSTs that have already been masked for cloud and ice.

Atmospheric Conditions
We investigated atmospheric variability influences on the intrusion events using atmospheric reanalysis data from the ECMWF ERA-5 operational reanalysis data (Copernicus Climate Change Service, 2017; Hersbach et al., 2018).The mean sea level pressure (Harden et al., 2011) and large-scale wind field over the Greenland Ice Sheet (not expected to resolve fjord-scale variability; Moore et al., 2015;Oltmanns et al., 2014) and ocean near Sermilik Fjord (Harden et al., 2011;Oltmanns et al., 2014)  We used ERA-5 6 hourly 10 m wind fields, instantaneous turbulent surface stresses, and mean sea level pressure fields (0.5° × 0.5° grid) to determine atmospheric changes before and after the intrusions.The northeasterly alongshore wind velocity and stress are defined as the component of each parameter along the principal axis of the shelf: 242° from north.We recognize that the shoreline and trough bends near the Sermilik Fjord region and a slightly more north/south wind direction may be more suitable for determining alongshore upwelling/downwelling.However, these values provide similar results, so we use 242° to be consistent with previous work (e.g., Straneo et al., 2010).We averaged ERA-5 wind speed and stress over a region of Sermilik Trough we defined as between 64.4°N and 65.0°N, 38.0°W and 35.0°W (Figure 1a, purple box), which represents an area where winds have the strongest correlation with regional EGCC transport changes and the greatest impact on trough transport (Harden et al., 2014;Le Bras et al., 2018).We then rotated the wind velocity and stress vectors to determine the northeasterly alongshore wind components.An alongshore northeasterly wind component (positive) represents coastal downwelling-favorable winds; a southwesterly wind (negative) represents upwelling.
We derived cyclone frequency and tracks using a cyclone detection and tracking algorithm (Crawford et al., 2020;Crawford & Serreze, 2016).To identify barrier wind events, we averaged winds across the southern Denmark Strait region (64.0°N to 65.4°N, 37.5°W to 31.5°W; yellow box in Figure 1a).Any events with >20 m/s winds between 0° and 90° (clockwise) were identified as barrier wind events following (Harden et al., 2011).Piteraqs (or downslope wind events) were similarly identified from winds averaged across the Tasiilaq region (65.5°N to 65.7°N, 37.82°W to 37.42°W; orange box in Figure 1a) if wind speeds reached >10 m/s from between 270° and 20° (clockwise) following (Oltmanns et al., 2014).Similarly to the intrusion events, we considered barrier wind events or piteraqs that occurred in rapid (i.e., fewer than 3 days) succession to be one event.

Ocean Observations
We investigated the extent to which intrusions affect ocean properties nearshore and within Sermilik Fjord by examining ocean temperature and sea surface height (SSH) during the intrusion events.We analyzed temperature data within the Sermilik Trough and Fjord (Figure 1a) collected by moorings deployed between 24 August 2009 and 18 August 2013 (Harden et al., 2014;Jackson et al., 2014;Jackson & Straneo, 2016).For the continental shelf mooring (Figure 1a, blue star), we used the temperatures from either a Microcat SBE37SM or XR 420 RBR sensor deployed between 262 and 301 m depth near the mouth of Sermilik Fjord.We also used SSH records from (Harden et al., 2014), which were calculated using bottom pressure measurements at the same shelf mooring.
The fjord mooring (Figure 1a, pink star) was located midfjord (32 km from the shelf mooring; ∼70 km from Helheim Glacier's front) and had sensors deployed between 120 and 131, 250 and 294, 324 and 350, and 390 and 400 m depths (referred to as the 125 m, 250 m, 350 m, and 400 m records, respectively).Midfjord mooring temperatures were recorded using similar instruments as the continental shelf mooring (Jackson et al., 2014) at 7.5-15 min intervals and were averaged to 6-hourly observations.These moorings provide a time varying record of subsurface AW flowing onto the continental shelf and into Sermilik Fjord (Jackson & Straneo, 2016;Straneo et al., 2011).
As fresher currents running along the shelf, the EGCC and EGC may moderate the AW intrusion events by serving as a barrier (Murray et al., 2010) or diluting the intruding waters as they cross the continental shelf (Snow et al., 2021).Presumably, this influence of the EGCC/EGC changes with their width.To investigate this possibility, we estimated the monthly aggregated width of the surface PW layer during intrusion events using MODIS SST-derived observations of PW extent following (Snow et al., 2021).The analysis is based on the MODIS Level 3 SST V2014 products (Minnett et al., 2019).We sampled MODIS SSTs from the same trough region as (Snow et al., 2021) (13 14 km × 14 km sampling boxes), calculated the SST anomaly relative to the Irminger Current temperature, and then generated a monthly mean for each sampling box.We approximated the PW width along the trough by measuring the distance from the fjord mouth (65.6°N, 38.0°W) to the center of the last sampling box along the trough transect that had a SST anomaly <−1.5°C, a previously defined threshold for delineating temperature difference between surface PW within the EGCC and AW from the IC (Snow et al., 2021).The EGC is difficult to distinguish from the EGCC based on temperature and we cannot be sure we are capturing the EGC using our temperature threshold.Therefore, we refer to the width of the PW layer nearshore herein as the "EGCC width" and recognize it may also include the EGC.

Intrusion Event Characteristics
AW intrusions are observable throughout the MODIS record near Sermilik Fjord, and their size and extent appear to vary widely.A rapid AW intrusion near Sermilik Fjord from 27 February 2013 to 4 March 2013 serves to illustrate characteristics of the bathymetrically steered onshore intrusion and demonstrate the types of observations required to assess them (Figures 3 and 4).The intrusion bisected the sea ice covered EGCC, making the intrusion observable with optical imagery.Prior to the intrusion, sea ice hugged the coast and flowed southward under persistent prevailing northeasterly winds (a 17-day pattern recorded by ECMWF reanalysis; not shown).
On 27 and 28 February, cloud cover obscured the surface, but wind shifted to westerly at ∼10 m/s late on 27 February 2013 and again on 1 March after the passing of two cyclones (second cyclone shown in Figure 4).A when winds returned to their prevailing northerly direction, the cross-shelf intrusion was no longer visible on the surface and no evidence of an opposing offshore flow occurred.This phenomenon has not been previously observed using remote sensing, and its influence has not been accounted for in regional ocean models.Intrusions of this kind will rapidly advect AW and heat into Sermilik Fjord, toward Helheim Glacier, increasing the likelihood of melting (Figure 1c), and therefore serving as an important factor in glacier dynamics.
We visually identified 58 intrusion events during winter and spring (January to June) between 2010 and 2013 (Figure 5 and Table S1).All intrusions are marked by ice-free, relatively warm waters (in comparison to the surrounding sea ice covered EGCC and fjord surfaces) cutting through sea ice along Sermilik Trough, which we interpret as surface AW inflows.In several of the farthest traveling intrusions, cross-shelf velocities were between 0.13 and 0.19 m/s based on the distance that the intrusions crossed within a 24 hr period.Within 24-48 hr after the intrusion became visible in optical imagery, sea ice and cool surface water (observed in SST) spread outward from the coast except along the trough between sequential images (Figure 3 and Movie S1).The cold, ice-free surface water that expands outward from Sermilik Fjord mouth along shore to the south (1 March 2013 14:50 and 2 March 2013 15:15 panels in Figure 3) indicates surface outflow from the fjord during intermediary circulation.
Based on the distance that the ice-free water edge moves between images, the surface outflows have approximate velocities of 0.35-0.7 m/s near the fjord mouth, which are consistent with fjord surface velocities measured by Jackson et al. (2014) during intermediary circulation.

Wind and SSH Conditions Before Intrusion Events
Intrusions coincided most frequently with cyclones and barrier wind events, but not piteraqs unless the piteraqs were also accompanied by the other wind events.Intrusions most frequently occurred after the passing of a cyclone (67% of selected intrusions) or similar low-pressure system offshore (26%) and less frequently with only a high pressure system over Greenland or the Irminger Sea (i.e., blocking event; 7%).We observed six barrier wind events (or clusters of events) between January and June across the 4 years.These barrier wind events accompanied the strongest northeasterly (i.e., downwelling-favorable) winds across our records and all preceded AW intrusion events (Figure 5).We observed 15 piteraqs, seven of which coincided with AW intrusion events (12% of the total intrusion events), and all but one of the seven were associated with simultaneous westerly to southwesterly (i.e., upwelling-favorable) winds offshore.
Local alongshore wind velocity and SSH shifts preceded the AW intrusion events (Figure 5).Winds during most of the intrusion events took two forms within the 24 hr preceding the intrusions: (a) winds that shifted from the prevailing northeasterly direction to a westerly direction, and (b) winds that weakened following strong northeasterly winds (typically >10 m/s; Figure 5); the latter occurred less frequently (19%).For the events that occurred when winds came from a westerly direction, the duration of those winds was between 6 and 90 hr.
Intrusions often (78%) coincided with a SSH drop (mean change between the 18 hr before and after the intrusion began: 10 ± 2 cm).Only one intrusion occurred with persistent, weak northeasterly winds (i.e., unfavorable downwelling conditions), although the event coincided with a cyclone passing over the Irminger Sea.

Nearshore and Fjord Ocean Temperatures
Records from atmospheric reanalysis products and ocean moorings show that alongshore wind stresses have significant (p < 0.05) positive correlations with SSH and negative correlations with inshore ocean warming (Figure S2 in Supporting Information S1).We also created normalized composites of atmospheric and ocean variability during intrusion events (Figure 6) to help distinguish between lateral warm water advection into the fjord during the intrusion events and isopycnal heaving as winds shift from downwelling-to upwelling-favorable conditions.If isopycnal heaving drives the warming during upwelling, we would expect the warming to be short-lived and to see no measurable warming in the composite records once downwelling-favorable conditions set in.Alternatively, we can attribute ocean warming to advection if significant warming persisted once downwelling has occurred.
Together, the satellite thermal imagery (Figure 3) and mooring observations (Figure 6) show a shift to warmer ocean temperatures on the shelf and in the fjord during the onset of the intrusions, with the new conditions persisting for at least 8 days post-intrusion event (Figure 6).MODIS sea surface temperatures revealed surface water temperatures (Figure 3, Figure S1 in Supporting Information S1, Movie S1) within the intrusion up to ∼4°C warmer than those in the EGCC and fjord, similar to AW temperatures within the IC.Further, we examined mean mooring temperature differences from the 4 days before and the 4 days after the events (ΔT 4,−4 ) as an indication of temperature change (Figure 7a): the intrusion events corresponded to significant warming (p < 0.05) at the continental shelf mooring (0.6 ± 0.13°C) and midfjord moorings at 125 m (0.17 ± 0.10°C), 250 m (0.35 ± 0.11°C), 350 m (0.2 ± 0.07°C), and 400 m (0.07 ± 0.05°C).The mean warming trend held for ΔT 4:8,−4:−8 (difference between mean ocean temperatures 4-8 days before and 4-8 days after the wind events) only at moorings between 250 and 350 m depth (shelf: 0.33 ± 0.18°C; midfjord 250 m: 0.28 ± 0.17°C; midfjord 350 m: 0.15 ± 0.08°C), with a maximum increase of 2-3°C during an event at those moorings (Figure 7b).These results indicate that the intrusion events led to sustained warming in the upper AW layer on the continental shelf and in the fjord (250-350 m deep; Figure 7) through laterally transporting warm AW from offshore to the continental shelf and fjord, rather than merely producing the vertical heaving of isopycnals (Jackson et al., 2014).

Reliable Detection of Intrusion Events Using MODIS Optical and Thermal Imagery
We present a unique, high temporal-resolution MODIS optical and thermal data set that, for the first time, directly captures the spatially contiguous pattern of cross-shelf upwelling flows along a continental shelf trough and provides a new method for investigating this type of flow around Greenland (and potentially elsewhere) when conditions allow (Figure 3, Figure S2 in Supporting Information S1, Movie S1).The large winter and springtime difference between PW and AW in temperature (∼4°C) and sea ice extent (i.e., presence versus absence) make it possible for satellites to detect the surface variability in these water masses.Therefore, satellite imagery is particularly useful for detecting, analyzing, and quantifying the pronounced surface intrusion that cross-cuts the EGCC toward Sermilik Fjord within Sermilik Trough.These observations serve to compliment conductivity-temperature-depth profiling (temporally limited) and mooring time series (spatially limited) to provide the surface evolution of intrusion events with high spatial and temporal coverage.This remote-sensing technique will translate well to trough systems with substantial temperature or sea ice gradients, such as other troughs along southeastern Greenland (e.g., Kangerdlugssuaq and Umiiviip Kangertiva; Figure 8).
Some limitations exist for detecting these intrusions using satellite imagery.Satellite identification of the intrusions restrict observations to AW intrusions that have surface expressions; therefore, intrusions that do not fully penetrate to the surface are not captured (or considered in this study).Further, cloud cover frequently obscures observations of the ocean surface.Thus, observations made by this method and our observations in this study likely include only a fraction of the total intrusion events during the study period.However, the available data sets support a clear characterization of the intrusion processes.

Continental Shelf and Fjord Ocean Warming
We found that AW intrusion events produced significant inshore surface and subsurface warming indicating enhanced upwelling and shoreward transport of AW along Sermilik Trough.MODIS thermal infrared imagery detected surface waters that become 4°C warmer nearshore along the Sermilik Trough during intrusion events.As captured by mooring data (Figure 6), the intrusions, on average, led to sustained warming on the continental shelf (e.g., at 290 m: 0.33 ± 0.18°C) and at some subsurface sites in the midfjord (e.g., at 250 m: 0.28 ± 0.17°C; at 350 m: 0.15 ± 0.08°C).Subsurface warming at the fjord mouth and midfjord are consistent with our satellite-based findings that the intrusions drive warm surface water inshore toward Sermilik Fjord.The subsurface warming within the fjord shows that this water is also transported into the fjord at depths between 250 and 350 m and potentially the entire AW layer in the fjord as an intrusion event progresses.These rapid events can increase subsurface AW temperatures flowing toward Helheim Glacier by as much as 2-3°C (Figure 5) and therefore present a potentially important heat transport mechanism.

Continental Shelf Circulation During Intrusion Events
Combining MODIS imagery with data from ocean moorings and atmospheric reanalysis products, we provide new insight on the dynamics of AW intrusion events linking wind-driven processes in the fjord and those on the neighboring continental shelf (Figures 1b and 1c).As observed by MODIS imagery and inferred from mooring ocean temperature records, when prevailing downwelling-favorable winds transition to upwelling-favorable winds (Figure 1c), they produce AW upwelling and a rapid (∼0.2 m/s) surface and subsurface AW inflow along the Sermilik Trough.This inflow develops over only a few hours.The velocity of inflowing water is consistent with similar onshore intruding flows observed around the world (Cottier et al., 2007;Harden et al., 2014;Inall et al., 2015;Lentz et al., 2014).The onshore intrusion flow in the trough during this upwelling regime opposes surface offshore flow outside the trough.However, an opposing flow response in the trough is not manifested when winds are in the opposite direction in the downwelling regime (Figures 3, 4 and Figure S2 in Supporting Information S1).That is, the flow responses in the trough to along-shelf winds of opposite directions are asymmetric.
The intrusions in Sermilik Trough are mostly driven by cyclones and less frequently by anticyclones.Barrier wind events, which are commonly associated with the passage of cyclones through the Denmark Strait also caused intrusions.Other factors such as tides and cyclonic eddies propagating along the continental slope (Brearley et al., 2012;Bruce, 1995;Magaldi et al., 2011) may moderate intrusions, but we rule them out as primary forcing factors because of the frequency mismatch (tides and eddies: superinertial to 1-2 days; intrusions: >2 days).Meanwhile, coincident onshore intrusions can occur over a Greenland coastal region much larger than a mesoscale eddy (Figure 8; Bruce, 1995).We also rule out piteraqs, which rarely coincide with AW intrusions (12%; Table S1), although piteraqs may contribute to fjord surface outflow and potentially subsurface outflow (Spall et al., 2017) when they cooccur with upwelling-favorable winds on the continental shelf.
Our observations of the appearance of warm water at the surface (upwelling and inflow) and at depth (warming from inflow at depth) in Sermilik Trough, as well as the timing of these events (<1 day lag with upwelling-favorable winds) are consistent with bathymetrically induced onshore intrusion of warm AW.The observed AW onshore intrusion along the Sermilik Trough's northeastern flank is consistent with localized upwelling flow on the upstream (in the sense of CTW propagation) slope of a continental shelf/slope canyon (Allen & Hickey, 2010;She & Klinck, 2000;Zhang & Lentz, 2017).This interpretation is also supported by previous work showing that ocean densities along the onshore side of Sermilik Trough (the northwestern flank) correlate more closely with wind forcing than those on the offshore side (Harden et al., 2014).Therefore, during upwelling-favorable winds or a sudden reduction in downwelling-favorable winds, water along the trough's northeastern flank likely upwells and flows shoreward throughout the water column, lifting dense AW from the continental slope toward Sermilik Fjord (Figure 1b).
Our proposed process of localized upwelling driving intrusion events is consistent with previous studies showing onshore flows in continental shelf/slope canyons associated with arrested (i.e., standing) CTWs (Kämpf, 2007;Lentz et al., 2014;She & Klinck, 2000;Zhang & Lentz, 2017).The onshore flow of the intrusion at the trough is likely either a part of a standing CTW that is excited at the trough with a southwestward propagation tendency and then arrested by a sustained northeastward shelf flow, or a transient consequence of the excessive onshore pressure gradient force during a rapid reduction of the southwestward shelf flow.We term it an "excessive" onshore pressure gradient force because it exceeds the pressure gradient force required for the cross-shelf geostrophic balance.This excessive force is caused by the greater water depth in the trough breaking the cross-shelf geostrophic balance (with cross-shelf sea level tilt) achieved over the shallower depth outside of the trough (Allen & Hickey, 2010).
In both scenarios, the canyon flow is ageostrophic with an  (1) Rossby number, where Rossby number Ro = U/ (fL) (Hampson, 2020;Lentz et al., 2014;Zhang & Lentz, 2017).Here, U is a scale of the along-shelf flow, f is the Coriolis parameter, and L is a length scale of the canyon/trough.Both the observed asymmetric responses of the trough flow under coastal upwelling versus downwelling and the intensification of onshore flow along the right (northeastern) side of Sermilik Trough match the modeled surface expression of arrested CTWs over a canyon (Zhang & Lentz, 2017).
Since the Ro of the intrusion flow in Sermilik Trough is  (1) , we hypothesize that the observed onshore intrusions in Sermilik Trough resulted from CTWs that were excited and arrested at the trough and represent an asymmetric response of the trough circulation to ambient along-shelf flows of opposite directions.During downwelling-favorable winds, enhancement of the offshore flow in a trough is minimal (Allen & Madron, 2009;Lentz et al., 2014) as topographically generated CTWs propagate freely downstream (to the southwest in this case) away from the trough.Therefore, downwelling-favorable and upwelling-favorable winds do not drive equivalent opposing flows along the trough, and oscillatory along-shelf winds can generate localized net onshore inflow in the trough, which is apparent in the MODIS imagery (Figure 3).
The localized intrusion transports offshore warm water toward the fjord and alters water properties along the way (Håvik & Våge, 2018).This impact is undiminished during the subsequent return of the winds to downwelling-favorable (Figures 6d and 6f) and thus cumulative.For this reason, even if the intruding AW is not advected into the fjord immediately during an upwelling event, the AW brought onto the shelf can be delivered into the fjord and enhance warming there during subsequent upwelling events (Fraser et al., 2018;Kämpf, 2006).During weak wind periods, the oscillatory shelf flows are also weak with a low Ro (Kämpf, 2009;Lentz et al., 2014) and would not produce net onshore intrusion of the offshore warm AW.This process could explain why only some wind-driven intermediary circulation within Sermilik Fjord results in advection of warm waters into the fjord (Jackson et al., 2014(Jackson et al., , 2018)).
Although our findings qualitatively agree with modeling studies of arrested-CTW-induced onshore flows in canyons, differences in the flows may exist.In particular, previous studies (e.g., Hampson, 2020;Zhang & Lentz, 2017) used relatively simple coastlines and bathymetries, which are not representative of the complex study area.The Sermilik system has a bending coastline and a trough carved into an undulating continental shelf that leads to a fjord.Moreover, during upwelling-favorable winds, the EGCC does not necessarily reverse, but may merely slow down at times (Foukal et al., 2020;Sutherland & Pickart, 2008).This difference suggests that at times there could be no northeastward shelf flow to arrest CTWs.In the case that the flow reversal did not occur, the intrusions may be a transient response of the trough flow to a sudden weakening of the EGCC: onshore intrusion flow could result from an excessive onshore pressure gradient induced by the cross-shelf sea level tilt over the deep trough.However, the timescale associated with this process remains unknown.To understand the details of the onshore intrusion flow and confirm its formation mechanism, further modeling work of the Sermilik circulation system is needed.

A Driver of Fjord Circulation
Our observations suggest that after the onset of the onshore intrusion, AW flowed into Sermilik Fjord toward Helheim Glacier, resulting from concurrent wind-driven intermediary circulation within the fjord (Figure 1c).During an intrusion event, we observed in imagery a large surface outflow (∼0.35-0.7 m/s) from the fjord, which is presumably coincident with AW inflow at depth based on the physics of wind-driven intermediary circulation in Sermilik Fjord (Jackson et al., 2014;Straneo et al., 2010).The subsurface warming at multiple depths within the fjord (ΔT 4:8,−4:8 = 0.28 ± 0.17°C at 250 m; ΔT 4:8,−4:8 = 0.15 ± 0.08°C at 350 m) supports this interpretation.Sustained subsurface warming likely stems from a combination of the following during intermediary current reversals within the fjord: (a) subsurface influx into the fjord of offshore-derived AW that directly stems from the intrusion, and (b) mixing of subsurface AW within the water column during velocity reversals.Glacier-meltwater-driven circulation is weak in wintertime (Sciascia et al., 2014;Straneo et al., 2010) and relatively weak in comparison to the MODIS-observed flooding of surface water out of the fjord; therefore, we conclude that the meltwater has a minimal contribution to the dynamics.

Intrusion Suppression by the EGCC
The EGCC likely serves as a barrier to the wind-driven AW intrusions by increasing the physical distance that AW must travel to reach Sermilik Fjord (Murray et al., 2010).Our results indicate that warming occurred on the continental shelf over 77% of the time when the EGCC was narrow, substantially more common than when the EGCC was wide (55% of the time; Figure 7).Narrowing of the EGCC allows a more efficient AW intrusion onto the shelf by reducing the distance AW must travel to reach the fjord and the extent to which the water dilutes along the way (Snow et al., 2021).
Enhanced ambient continental shelf stratification and greater sea ice concentrations caused by the EGCC may also suppress inshore intrusion flow in the surface layer.Consistent with enhanced temperature variability observed at the subsurface shelf mooring (r 2 = 0.40; Figure S3 in Supporting Information S1), a seasonal widening of the EGCC coincides with EGCC greater transport (Harden et al., 2014) and increased stratification of the deeper water layers along the inner shelf (Harden et al., 2014;Jackson et al., 2014).Modeling studies have suggested that strong stratification can create a lid over upwelling within a canyon and cause isopycnals to squeeze together between the base of the surface layer and the rim of the canyon (Ramos-Musalem & Allen, 2019).That is, a deeper pycnocline suppresses the vertical extent of the bathymetry's influence and thus reduces the chance of upwelling flow reaching the surface.This vertical suppression of the intrusion would limit our ability to observe the intrusion from space and restrict the depth range in which AW is transported toward the fjord.Further, because the EGCC is often choked with sea ice in wintertime, a wide wintertime EGCC often results in greater sea ice coverage over the trough.Such conditions would decouple wind and surface ocean stresses (Martin et al., 2014) and reduce the intrusion event or prevent it from occurring.Meanwhile, we stress that, although a stronger EGCC likely suppresses intrusion events, the lack of a surface expression does not preclude the intrusion events from still occurring at depth (Figure 1c).
Together, wind characteristics (duration and magnitude) and EGCC width likely regulate when the wind events can effectively transport AW across the continental shelf and into Sermilik Fjord.Modeling of a simple canyon system suggest that intrusions resulting from upwelling-favorable winds intensify throughout the first 24 hr and remain at that strength as long as the winds persist (Zhang & Lentz, 2017).This result suggests that the strength and duration of an upwelling alongshore wind event could greatly affect overall AW transport during an upwelling event.Further, when the EGCC widens and deepens, greater wind forcing is likely required to produce intrusions and enhanced heat transport.Vertical turbulence during upwelling tends to diffuse stratification (Ramos-Musalem & Allen, 2019), making the intrusion more likely to extend upwards in the water column with time.Offshore AW temperature and its vertical distribution also may moderate whether warming or cooling is observed inshore and within the fjord (Fraser et al., 2018).Other factors that may impact the recorded warming signals inshore include source water temperature variability-potentially linked with EGCC width (Figure S1b in Supporting Information S1)-or a rapid subsequent temperature fluctuation.For instance, a coastal downwelling occurring 4-8 days after an intrusion would be considered by our analysis as a cooling event (Figure S1c in Supporting Information S1).

Summary and Outlook
Satellite-derived visible and thermal imagery have tremendous potential for detecting local to regional ocean circulation processes across the globe.Using this new satellite-based observational record of intrusion events along Sermilik Trough in conjunction with wind reanalysis data and in situ ocean measurements has allowed us to create more temporally and spatially holistic observations of intrusion events than has been previously achieved in studies of similar oceanic events around the world.With this data set, we document the existence of the intrusion events, which are caused by an asymmetrical response of trough flows to opposing winds and drive onshore intrusion of warm offshore (AW) water toward Sermilik Fjord in southeast Greenland.These intrusions can substantially impact heat transport toward Sermilik Fjord and Helheim Glacier and likely other large glacier systems around Greenland.These findings provide insight into entire continental shelf circulation that are nearly impossible to capture with field-based measurements alone, helping us to fill in gaps in current knowledge of the processes that connect the broader ocean to southeast Greenland fjords.
With this broader view of the southeast Greenland continental shelf, we have identified important atmospheric and ocean feedbacks that may moderate ocean heat transport toward southeast Greenland glaciers.If more frequent upwelling-favorable wind events were to occur, this shift would lead to higher volumes of AW (and its heat) flushing onto the shelf and into Sermilik Fjord.Greater cyclone and barrier wind activity, specific high pressure blocking patterns and a high NAO index (which relates to storm variability in southeastern Greenland; Harden et al., 2011;Straneo & Heimbach, 2013) could all increase intrusion frequency based on our results.Further, EGCC weakening as a result of reduced freshwater and sea ice transport out of the Arctic (Harden et al., 2014), enhanced PW ejection off the continental shelf, and reduced Greenland Ice Sheet runoff/iceberg calving (Sutherland & Pickart, 2008) will permit more intrusion events, reduce the AW dilution as it crosses the continental shelf, and make southeast Greenland fjords more susceptible to warm water intrusion (Murray et al., 2010;Snow et al., 2021).These intrusion events can result in 2-3°C of warming in the fjord, and models may be able to predict years of high onshore heat transport using this improved understanding of the linkage between subsurface water temperatures, EGCC width and transport, and intrusion-favorable winds.A better understanding of the deep-water onshore transport processes that may directly feed into Sermilik Fjord and bring heat to Helheim Glacier has substantial implications for pinpointing the role of the ocean in glacier change in the past and future.Intrusion events of this nature may also occur to varying extents at other glacier systems around Greenland, such as Kangerdlugssuaq Glacier (Fraser et al., 2018), which could make it an important mechanism for regulating the mass loss of a large-scale ice sheet.

Figure 1 .
Figure 1.Sermilik Trough study region and wind-driven ocean circulation along the trough.(a) Continental shelf bathymetry near Sermilik Fjord and Trough.Purple, yellow, and orange boxes show the sampling regions for the ERA-5 reanalysis alongshore winds, barrier wind events, and piteraqs (see Section 2), respectively.Stars indicate the continental shelf (blue) and midfjord (pink) mooring locations.Inset shows study location in southeast Greenland.Bathymetry is from BedMachine v3 (Morlighem et al., 2017) with the thin black lines representing contours at 300, 400, 500 m, and every 500 m thereafter.Schematics of the ocean circulation along the axis of Sermilik Trough and Fjord during (b) downwelling-favorable and (c) upwelling-favorable winds.Black arrows indicate current direction and strength within the East Greenland Coastal Current (Polar Water) and Irminger Current (Atlantic Water) on the continental shelf.Red (Atlantic Water) and blue (Polar Water) arrows indicate flow direction in the fjord associated with intermediary circulation, and on the continental shelf from coastal upwelling/downwelling and intrusion along the trough.Clockwise-rotating purple arrows indicate the vorticity created by coastal-trapped waves, which stimulates Atlantic Water intrusion along the trough.White and brown features indicate ice and land, respectively.The presence of a mooring is indicated in the trough with its corresponding observations of bottom pressure (BP) and temperature (T) in each scenario.

Figure 2 .
Figure 2. Comparison of ERA-5 (black) near-surface wind speeds with ground measurements made by the Program for Monitoring of the Greenland Ice Sheet (PROMICE) MIT weather station (green) from 2010 to 2013 as a (a) time series and (b) scatterplot.The dashed line in (b) represents a one-to-one agreement.Pearson's r represents the correlation between time series with seasonal signal and trends removed.
have been shown to be well-captured by ERA-Interim (previous version to ERA-5) reanalysis data in the region, though ERA-Interim underestimates peak wind speeds by ∼1-2 m/s over the ocean.Ground-based wind speed measurements captured by the Program for Monitoring of the Greenland Ice Sheet (PROMICE) MIT weather station (65.6922°N, 37.82.80°W) are also well-represented by the 10 m wind speeds from ERA-5 (mean bias: +1.9 m/s; Pearson's r: 0.785, p: <0.001; Figure2).Therefore, similar to ERA-Interim, we assumed the large-scale wind field surrounding Sermilik Fjord remains well-captured in the more recent data product.

Figure 3 .
Figure 3. Evolution of an example Atlantic Water (AW) intrusion event near Sermilik Fjord.Selected MODerate Resolution Imaging Spectroradiometer (MODIS) optical imagery during an AW intrusion event from 27 February 2013 to 4 March 2013 overlain with MODIS-derived sea surface temperatures (SST); black arrow indicates wind stress direction averaged over Sermilik Trough.The cyclone (white arrow) was identified separately as a low in ECMWF mean sea level pressure that corresponds to the swirling cloud cover apparent in regional MODIS imagery.Optical imagery is not available for the 05:10 and 06:45 passes on 1 March 2013, as both of these scenes were before sunrise; gray indicates land instead for these images.SSTs not shown in areas with sea ice or cloud cover.The orange box highlights the intrusion initiation after the shifting of alongshore wind stress over Sermilik Trough.Dark gray contours indicate the 250 and 500 m isobaths, which outline the location of Sermilik Trough.Blue arrows indicate the right-hand side of the trough and the AW intrusion path.

Figure 4 .
Figure 4. Mean sea level pressure snapshots overlain by wind speed vectors from 18:00 on 28 February 2013 to 18:00 on 1 March 2013.The red box outlines the location of the Sermilik Trough and Fjord.Orange box highlights the time of the shifting of alongshore wind stress over Sermilik Trough and the intrusion initiation.Thick black contour indicates the cyclone.

Figure 5 .
Figure 5. Timing of MODerate Resolution Imaging Spectroradiometer (MODIS)-identified Atlantic Water intrusion events (vertical black lines) with alongshore wind stress (black lines; positive is northeasterly), continental shelf sea surface height (SSH; orange lines), mooring ocean temperature from the continental shelf at 290 m depth (blue), and mooring ocean temperature from within the fjord at 250 m (light pink) and 350 m (dark pink) depth.Piteraq and barrier wind events are highlighted with light and dark gray bars, respectively, near the top of each panel.

Figure 6 .
Figure 6.Composites of atmospheric and ocean parameters during the 58 visually identified AW intrusion events: (a) ERA-5 wind direction, (b) alongshore wind stress (positive means northeasterly), (b) sea surface height (SSH), and mooring temperatures from (d) the continental shelf near Sermilik Fjord mouth (290 m) and midfjord at (e) 250 m and (f) 350 m depth.Orange lines indicate the onset of negative alongshore wind forcing, and hours are the same as in (a).Error bars indicate the standard error of the mean across the events.We note that the standard deviation representing the dispersion of the data around the mean is 7.6 times larger than the shown standard error.

Figure 7 .
Figure 7. Mooring temperature differences averaged over (a) 4 days before and 4 days after (ΔT 4,−4 ) and (b) 4-8 days before and 4-8 days after (ΔT 4:8,−4:8) the wind events.The means of all intrusions (black) and of only the intrusions when the East Greenland Coastal Current (EGCC) was narrow (<61 km; blue) and wide (>61 km; red) are shown with 1-standard error (bars; 68% confidence interval) for each mooring location and depth.The one-sample t statistic and p value test the null hypothesis that the mean for all intrusions (black) equals 0 with the alternative hypothesis that the mean is >0.

Figure 8 .
Figure 8. Coincident onshore intrusion events at multiple troughs along the southeastern Greenland continental shelf.MODerate Resolution Imaging Spectroradiometer (MODIS)-derived SST (colorscale) captures coincident AW intrusion events along the troughs leading to the three labeled glaciers: Umiiviip Kangertiva (UK), Helheim (HG), and Kangerdlugssuaq (KG) glaciers.The background imagery is from MODIS visible and thermal infrared on 2 March 2013.Hand-drawn open water edges are shown for 27 February 2013 (yellow), 1 March 2013 (light green), and 2 March 2013 (dark green).White arrows indicate troughs with intrusions and inset provides location of area in Greenland (orange).