Ice shelves, the floating extensions of grounded glaciers, subdue the discharge of grounded ice into the global ocean (Rignot et al., 2004; Scambos et al., 2004; Gudmundsson, 2013; Borstad et al., 2016). Re-examination of past ice shelf collapse events can help to shed light on the mechanisms of collapse and improve the understanding of ice shelf dynamics for future projections of sea-level rise (Rignot et al., 2004; Gudmundsson, 2013; Borstad et al., 2016). The final collapses of the Larsen A (LAIS) in 1995 and the Larsen B (LBIS) ice shelves in 2002 have been attributed to decreased structural integrity brought on by a combination of factors: most notably, regional atmospheric warming (Scambos et al., 2000; Mulvaney et al., 2012), extended melt seasons (Scambos et al., 2003), multi-year firn pore space depletion (Kuipers Munneke et al., 2014; Trusel et al., 2015), melt pond flooding (Glasser and Scambos, 2008; Trusel et al., 2013; Leeson et al., 2020), crevasse expansion through hydrofracture (Scambos et al., 2003; Banwell et al., 2013; Pollard et al., 2015; Alley et al., 2018; Banwell et al., 2019; Robel and Banwell, 2019), glacier structural discontinuities (Glasser and Scambos, 2008​​​​​​​), basal melt (Pritchard et al., 2012; Rignot et al., 2013; Depoorter et al., 2013; Schodlok et al., 2016; Adusumilli et al., 2018), warm meltwater intrusion (Braun and Humbert, 2009​​​​​​​), melting of the ice melange within rifts conducive to rift propagation (Larour et al., 2021), and regional sea ice loss allowing ocean swell flexure stress on the calving front (Banwell et al., 2017; Massom et al., 2018).

While the list of mechanisms that can destabilize ice shelves is extensive, a conceptual model for rapid ice shelf collapse proposed by Massom et al. (2018) identifies four essential prerequisites for sudden collapse: (1) extensive surface flooding and hydrofracture, (2) reduced sea ice or fast ice at the ice shelf front, (3) outer margin or terminus fracturing and rifting, and (4) initial calving trigger at the ice shelf margin. They theorize waves led to calving front damage and small calving events that breached the “compressive arch” of stability of both ice shelves proposed by Doake et al. (1988). At the same time the ice shelves were covered in extensive surface melt lakes that were unlikely to drain horizontally because of the relatively flat surface (Banwell et al., 2014). Satellite observations and ice shelf stability model studies determined the LBIS was covered with >2750 melt lakes that were on average 1 m deep before collapse (Glasser and Scambos, 2008; Banwell et al., 2013). Ice shelves inundated with surface melt lakes are susceptible to disintegration through a process known as hydrofracture, where meltwater applies outward and downward pressure to the walls and tip of crevasses that can propagate through the ice shelf (Scambos et al., 2003; Banwell et al., 2013; Bell et al., 2018; Lhermitte et al., 2020). Furthermore, melt lakes that rapidly drain by hydrofracture can create fracture patterns that split ice shelves into sections with aspect ratios that support unstable rollover and hydrofracture cascades that begin when melt lakes drain and/or calving occurs at the ice shelf terminus (Scambos et al., 2003; Banwell et al., 2013; Burton et al., 2013; Robel and Banwell, 2019). The combination of ocean swell stress on the calving front and extensive melt ponds led to large-scale hydrofracture cascades that Massom et al. (2018) proposed ultimately caused the rapid collapse of the LBIS and possibly the LAIS.

In addition to a lack of sea ice and extensive melt ponds, meteorological and satellite observations identify clear skies and warm west-northwest föhn wind at the time of collapse (Fig. 1c–f) (Rott et al., 1998; Rack and Rott, 2004; Cape et al., 2015; Massom et al., 2018). Föhn winds form when relatively cool moist air is forced over a mountain barrier, often leading to precipitation on the windward side of the barrier that dries the air mass (Grosvenor et al., 2014; Elvidge et al., 2015). As the now drier air descends the leeward slope it warms adiabatically and promotes melt directly through sensible heat exchange, and indirectly by the associated clear skies that allow additional shortwave radiation to reach the surface in non-winter months (Turton et al., 2017, 2018; Kuipers Munneke et al., 2018; Elvidge et al., 2020; Laffin et al., 2021). Föhn winds and their capacity to cause surface melt have been studied extensively on the AP. Observations and model studies on the LCIS confirm the föhn mechanism that enhances sensible heat and shortwave radiation and alters local albedo, which can increase surface melt rates upwards of 50 % compared to non-föhn conditions (Cape et al., 2015; Elvidge et al., 2015; King et al., 2015, 2017; Kuipers Munneke et al., 2012, 2018; Bevan et al., 2017; Lenaerts et al., 2017; Datta et al., 2019; Kirchgaessner, et al., 2021; Laffin et al., 2021; W. Wang et al., 2021​​​​​​​). Late-season föhn melt reduces firn pore space and thus pre-conditions ice shelves to form melt ponds that are responsible for the increased firn density pattern east of the AP mountains on the LCIS (Holland et al., 2011; Kuipers Munneke et al., 2014; Datta et al., 2019). Föhn melt climatology studies have aimed to identify how much melt is caused by föhn and the locations most affected and found föhn winds account for up to 17 % of melt and are concentrated in the LCIS inlets (Turton et al., 2017; Datta et al., 2019; Laffin et al., 2021). Pressure gradient differences across the AP range lead to föhn winds that funnel through mountain gaps as highly concentrated föhn jets, particularly in inlets east of the AP range (Luckman et al., 2014; Elvidge et al., 2015; Kuipers Munneke et al., 2012; Grosvenor et al., 2014). In addition to enhancing surface melt rates, föhn winds exert force on sea/fast ice and drag it away from the calving front, thereby exposing the front to ocean waves (Bozkurt et al., 2018). Climatic studies of the Larsen B embayment indicate that föhn winds were coincident with collapse (Rack and Rott, 2004; Leeson et al., 2017). However, it is unknown if concentrated föhn jets spilled onto the former LAIS and LBIS and, if so, whether those föhn winds contributed to their collapse. The following questions, therefore, arise. (1) To what extent did föhn-induced melt contribute to the surface melt budget on each eastern AP ice shelf? (2) Did föhn winds and associated melt play a role in triggering the collapses of the LAIS and LBIS? (3) What are the implications of föhn-induced melt for the remaining eastern AP ice shelves?

To address these questions we consider three metrics: Sect. 3.1 explores the total annual surface melt quantity induced by föhn winds and how melt is spatially distributed across each ice shelf; Sect. 3.2 identifies the coincidence of föhn-induced melt preceding and during the collapse events, and the estimated melt-lake depth in response to melt events.; Sect. 3.3 identifies the contribution of föhn melt to the climatological surface liquid water budget, comparing collapsed and extant ice shelves on the eastern AP. By constructing a timeline of melt and melt mechanisms and comparing melt metrics with collapsed and extant ice shelves, we can identify the factors contributing to collapse.

The north–south temperature gradient present on the eastern AP ice shelves contributes to the differences in the ice shelf melt regime (Fig. 6). Warmer ice shelves can be more vulnerable to long-term thinning and retreat that accelerate disintegration (Scambos et al., 2003; Morris and Vaughan, 2003). However, the temperature gradient alone does not explain the substantial increase in surface melt on the LAIS and LBIS relative to more southerly ice shelves. Only with the addition of föhn-induced surface melt (Fig. 6c) do the LAIS and LBIS stand out significantly from the other eastern AP ice shelves (Fig. 6a, b). Temperature gradient, however, could explain why föhn wind events cause less melt on more southern ice shelves and may cause super melt events on collapsed ice shelves because temperature is already elevated on more northern ice shelves prior to the effect föhn has on temperature. With that in mind, we have examined liquid water processes on the spatiotemporal scales pertinent to AP ice shelf stability. For instance, the structural flow discontinuities or suture zones, where tributary glaciers merge together to form an ice shelf, are mechanically weak points that impact stability (Sandhäger et al., 2005; Glasser and Scambos, 2008; Glasser et al., 2009). These suture zones are further weakened through lateral shear depending on the difference in tributary glacier flow. All ice shelves in the region are composed of numerous outflow glaciers sutured together, and while some studies suggest this is a major contributor to ice shelf instability, only two of the ice shelves have collapsed (Borstad et al., 2016; Glasser and Scambos, 2008). Further research suggests that marine accretion of ice on the bottom of the ice shelves, specifically LCIS, may stabilize these suture zones, which may be why SCAR inlet has remained intact despite major rift formation (McGrath et al., 2014; Borstad et al., 2016).

The timing of surface melt and melt enhanced by föhn winds within the melt season may also provide insight into the fate of LAIS and LBIS, including why neither ice shelf collapsed in the anomalously strong 1992/1993 melt season (Fig. 3d). Pore space within the upper snow and firn layers buffers surface melt before lakes begin to form (Polashenski et al., 2017). Late-season melt is more likely to form surface melt lakes because meltwater from the preceding fall, winter, and spring has partially or completely filled available pore space. On both the LAIS and LBIS, 92 % of surface melt during the 1992/1993 melt season occurred before 9 January when there was more pore space to buffer the anomalous surface melt than at the onsets of their collapses in late January 1995 and early February 2002, respectively (Fig. 4a, c). Melt lakes were present on both ice shelves throughout the 1992/1993 melt season, though melt production slowed dramatically after mid-January 1993 (Scambos et al., 2000). The high melt rates in late November and early December 1992 on the LAIS were perhaps too early in the melt season, and after too many years of nominal melt, to form substantial melt lakes and trigger hydrofracture that season. Nevertheless, the 1992/1993 melt could have preconditioned the shelf for collapse in January 1995. The LBIS collapse began in February 2002 after the surface melt had returned to nominal 1980s levels for 6 years. How much pore space had recovered during those 6 years is unknown, and an important question for future research. Satellite images of surface melt lakes indicate 11 % of the ice shelf was covered in melt lakes prior to collapse (Glasser and Scambos, 2008). However, the preceding melt year (2000/2001) had low melt and high precipitation, which added additional snow and water mass to the unstable ice shelf (Leeson et al., 2017).

Another possible reason collapse of the LAIS and LBIS did not occur in the 1992/1993 melt season or other years prior to collapse was a possible misalignment of the four prerequisites for rapid collapse theorized by Massom et al. (2018). An AVHRR image of the LAIS taken on 8 December 1992, just after a series of major föhn wind events that lead to 252 mm w.e. of surface melt in the 8 d prior to the image (Fig. 4a), shows significant melt lakes across the LAIS, which make hydrofracture cascades possible. However, in the same image, sea ice and melange are shown to be at the calving front, protecting the front from long-period ocean swells that could trigger collapse. It may have been too early in the melt season to have substantial gaps in sea ice, the ocean temperature may have been too cold, ocean circulation could have help stabilize the sea ice at the front, the föhn winds speed could have been too weak to push the ice away or may have been in the wrong direction, all of which could have not allowed a proper trigger for collapse even though substantial melt ponds were present. Even if there were years or instances when sea ice extent was low and substantial melt lakes were present, there could have been a lack of long-period ocean swells that are thought to trigger collapse.

Regardless of other possible contributors to ice shelf instability not considered here (e.g., basal melting), föhn-induced surface melt and associated melt lakes, and the off-coast wind direction likely, played an important role in pushing the LAIS and LBIS past a structural tipping point. The estimated surface melt lake depth caused by the 9 d föhn melt event on the LAIS surpassed a melt lake depth identified by modeled and satellite-derived lake depths before the collapse of the LBIS (Banwell et al., 2013, 2014). The LAIS was likely the same thickness (200 m) or thinner at the time of collapse, so the estimate of critical surface lake depth for the LBIS that is applied to the LAIS may reflect an upper limit of melt lake depth of stability for the LAIS. Melt lake depth is likely underestimated because our estimation only accounts for melt during the 9 d melt event. Melt before this time period already exceeded an average melt year by 23 % (118 mm w.e.), so melt lakes probably already existed.

The converging lines of evidence in these results show that observed and inferred föhn-driven melt is present in sufficient amounts, and at the right locations and times, to cause extensive surface melt lakes, while the off-coast föhn wind direction pushed sea ice away from the calving front. The fact that the LAIS and LBIS collapsed catastrophically within weeks and not through long-term thinning and retreat like other ice shelves (Prince Gustav, Wordie) suggests sudden disintegration is anomalous and requires forcings to match vulnerabilities (Scambos et al., 2003). We conclude that föhn winds and the associated surface melt played a significant role in the collapses of the LAIS and LBIS, while extant eastern AP ice shelves are not likely to collapse from föhn-induced melt and hydrofracture in today's current climate. We have come to these conclusions with the following forms of evidence.

First, both the LAIS and LBIS are impacted by powerful melt-inducing föhn jets that affect a large spatial portion of each ice shelf and reach the ice shelf terminus. Surface melt and melt lakes near the ice shelf terminus can lead to calving front collapse and structural instability for the remaining portion of the ice shelves (Depoorter et al., 2013; Pollard et al., 2015). SCAR inlet and the LCIS are not directly affected by a föhn jet, are too vast to have any significant effect near the terminus, or are too far south to experience major melt events.

This research clarifies the roles of föhn-induced melt for collapsed and extant ice shelves on the eastern AP. Future analyses of these ice shelf collapse events using advanced firn density models coupled with ice–ocean–atmosphere coupled simulations may be useful to better understand the role of surface melt in ice shelf instability. Further, the AP föhn wind regime has remained stable over the past half-century (Laffin et al., 2021), which points to enhanced surface temperatures and increased liquid phase precipitation as more important contributors to the future surface liquid budget on remaining ice shelves and is an important area of future research (Bozkurt et al., 2020, 2021). However, changes in climate drivers such as the Southern Annular Mode (SAM), which influences the north–south movement of the westerlies in the region, may alter the temperature and föhn occurrence that will likely enhance surface melt in locations farther south and therefore make more southern ice shelves more vulnerable (Abram et al., 2014; Zheng et al., 2013; Lim et al., 2016). Nevertheless, this research highlights a new understanding behind föhn melt mechanisms for ice shelf collapse and suggests that SCAR inlet and the LCIS may remain stable so long as surface liquid water from melt and precipitation remains within historical bounds.