Ice-stream stability on a reverse bed slope

Marine-based ice streams whose beds deepen inland are thought to be inherently unstable1–3. This instability is of particular concern because significant portions of the marinebased West Antarctic and Greenland ice sheets are losing mass and their retreat could contribute significantly to future sea-level rise4–7. However, the present understanding of icestream stability is limited by observational records that are too short to resolve multi-decadal to millennial-scale behaviour or to validate numerical models8. Here we present a dynamic numerical simulation of Antarctic ice-stream retreat since the Last Glacial Maximum (LGM), constrained by geophysical data, whose behaviour is consistent with the geomorphological record. We find that retreat of Marguerite Bay Ice Stream following the LGM was highly nonlinear and was interrupted by stabilizations on a reverse-sloping bed, where theory predicts rapid unstable retreat. We demonstrate that these transient stabilizations were caused by enhanced lateral drag as the ice stream narrowed. We conclude that, as well as bed topography, ice-stream width and long-term retreat history are crucial for understanding decadalto centennial-scale ice-stream behaviour and marine ice-sheet vulnerability. Ice streams are fast-flowing arteries of ice sheets that dominate ice discharge into the oceans, impacting directly on sea level. Many ice streams possess beds that are below sea level and typically deepen inland on a reverse slope9. Theory suggests that ice discharge increases rapidly with water depth3, and in the absence of lateral-drag-induced buttressing from a floating ice shelf, grounding lines (marking the transition from grounded to floating ice) on reverse bed slopes may be unstable1,2. Bed topography is therefore cited as a strong control on ice-stream retreat rate3,10 and modern satellite observations of rapid ice-stream thinning and recession seem consistent with this theory4–6. However, with just two decades of data, these records are too short to identify the longer-term centennialto millennial-scale trends crucial for constraining future sea-level projections. Major uncertainties in predictions of ice-sheet vulnerability11 relate to limitations in understanding processes controlling grounding-line motion and, importantly, to deficiencies in grounding-line treatment in icesheet models12. In recent years, significant advances in model development have beenmade3,12–17, but tests have been applied only to simplified bed geometries or to steady-state conditions and lack validation against data over timescales longer than a few decades. We aim to understand the long-term controls and stability of marine ice streams and integrate a fully dynamic ice-stream model with the detailed geomorphological record of palaeo-ice-stream retreat imprinted on the sea floor of Marguerite Bay, western Antarctic Peninsula (Fig. 1). High-resolution mapping from swath bathymetry and analysis of subglacial landforms and sediments18–21 (Fig. 1 and