It has been suggested that the coastal Southern Ocean subsurface may warm over the 21st century in response to strengthening and poleward shifting winds, with potential adverse effects on West Antarctic glaciers. However, using a 1/12° ocean regional model that includes ice‐shelf cavities, we find a more complex response to changing winds in the Amundsen Sea. Simulated offshore subsurface waters get colder under strengthened and poleward shifted winds representative of the SAM projected trend. The buoyancy‐driven circulation induced by ice‐shelf melt transports this cold offshore anomaly onto the continental shelf, leading to cooling and decreased melt below 450 m. In the vicinity of ice‐shelf fronts, Ekman pumping contributes to raise the isotherms in response to changing winds. This effect overwhelms the horizontal transport of colder offshore waters at intermediate depths (between 200 and 450 m), and therefore increases melt rates in the upper part of the ice‐shelf cavities, which reinforces the buoyancy‐driven circulation and further contributes to raise the isotherms. Then, prescribing an extreme grounding line retreat projected for 2100, the total melt rates simulated underneath Thwaites and Pine Island are multiplied by 2.5. Such increase is explained by a larger ocean/ice interface exposed to CDW, which is then amplified by a stronger melt‐induced circulation along the ice draft. Our main conclusions are that (1) outputs from ocean models that do not represent ice shelf cavities (e.g., CMIP5 models) should not be directly used to predict the thermal forcing of future ice shelf cavities; (2) coupled ocean/ice sheet models with a velocity‐dependent melt formulation are needed for future projections of glaciers experiencing a significant grounding line retreat.

Understanding the interannual variability of surface mass balance (SMB) and surface melting in Antarctica is key to quantify the signal-to-noise ratio in climate trends, identify opportunities for multi-year climate predictions and assess the ability of climate models to respond to climate variability. Here we simulate summer SMB and surface melting from 1979 to 2017 using the Regional Atmosphere Model (MAR) at 10 km resolution over the drainage basins of the Amundsen Sea glaciers in West Antarctica. Our simulations reproduce the mean present-day climate in terms of near-surface temperature (mean overestimation of 0.10 • C), near-surface wind speed (mean underestimation of 0.42 m s −1 ), and SMB (relative bias < 20 % over Thwaites glacier). The simulated interannual variability of SMB and melting is also close to observation-based estimates.For all the Amundsen glacial drainage basins, the interannual variability of summer SMB and surface melting is driven by two distinct mechanisms: high summer SMB tends to occur when the Amundsen Sea Low (ASL) is shifted southward and westward, while high summer melt rates tend to occur when ASL is shallower (i.e. anticyclonic anomaly). Both mechanisms create a northerly flow anomaly that increases moisture convergence and cloud cover over the Amundsen Sea and therefore favors snowfall and downward longwave radiation over the ice sheet. The part of interannual summer SMB variance explained by the ASL longitudinal migrations increases westward and reaches 40 % for Getz. Interannual variation in the ASL relative central pressure is the largest driver of melt rate variability, with 11 % to 21 % of explained variance (increasing westward). While high summer SMB and melt rates are both favored by positive phases of El Niño-Southern Oscillation (ENSO), the Southern Oscillation Index (SOI) only explains 5 % to 16 % of SMB or melt rate interannual variance in our simulations, with moderate statistical significance. However, the part explained by SOI in the previous austral winter is greater, suggesting that at least a part of the ENSO-SMB and ENSO-melt relationships in summer is inherited from the previous austral winter. Possible mechanisms involve sea ice advection from the Ross Sea and intrusions of circumpolar deep water combined with melt-induced ocean overturning circulation in ice shelf cavities. Finally, we do not find any correlation with the Southern Annular Mode (SAM) in summer.