a, The Antarctic Peninsula. b, James Ross Island, location of the ice-core drilling site, and Prince Gustav Ice Shelf in 1988. The red square shows the location of c. c, Ulu Peninsula with published radiocarbon ages (circles)9, 29 and cosmogenic nuclide ages (diamonds)17, 19, Brandy Bay Moraine and boulder train. The plan view along line A–B is shown. Spot heights are in italics. The digital elevation model was produced by the Czech Geological Survey28. Bathymetric data are from the Antarctic and Southern Ocean Data Portal of the Marine Geoscience Data System. d, Cross-section of flowline A–B.
Glaciological input data.
Glaciological input data include ice thickness23, velocity, MAAT, topography28 and bathymetry (Fig. 1). The most recent re-advance was reconstructed from our own geological data17, 18 (Fig. 1) and from published calibrated radiocarbon9, 29 and cosmogenic nuclide ages17, 19 (Supplementary Information).
Numerical model description.
We used a one-dimensional, finite-difference glacier flowline model to investigate glacier–climate interactions on Ulu Peninsula, James Ross Island. The glacier model and its degree-day scheme have previously been described in detail24, 30, so are only summarized here. The model uses a forward explicit numerical scheme, implemented on a 100-m-horizontal-resolution staggered grid that spans the length and foreland of Glacier IJR45 into Prince Gustav Channel (Fig. 1). Horizontal flux is calculated through a cross-sectional plane described by a symmetrical trapezoid, and incorporates a width-dependent shape factor. The model assumes no transfer of ice flux between adjacent, but dynamically independent portions of the glacier. Velocity is determined by both the flow-enhancement coefficient (deformation factor), which accounts for the softening of the ice by impurities or contrasts in crystal orientation, and by basal sliding. Outliers in the velocity field are sensitive to transients in the model.
Modelling strategy.
The flowline model was tuned to present-day conditions to reproduce observed glacier extent, volume and velocity (Supplementary Table 3 and Methods), and was then dynamically calibrated using temperature and accumulation data over the past 160 years from the James Ross Island ice core3, 20 (Fig. 1b). Small adjustments were made to the degree-day factors until the glacier replicated observed recession and thinning rates over the past 30 years23 (Supplementary Information). The glacier stabilized in a position that matched present-day velocity and geometry, thus increasing confidence in model initialization.
Response-time tests performed at 0.1 °C increments from −0.5 °C to +1.0 °C investigated time taken to reach equilibrium following perturbation. Sensitivity tests investigated glacier response to perturbations in MAAT, mean annual precipitation, snow and ice degree-day factors, precipitation seasonality and flow-enhancement coefficient. Further, each incremental change in precipitation was run against each incremental change in temperature. Glacier sensitivity to summer precipitation seasonality under different MAATs was also analysed. Subsequent time-dependent simulations used the tuned parameters to model Holocene and future glacier characteristics. Holocene accumulation and air temperatures were derived from the ice-core record3, 20. Future transient runs were forced by output from a regional atmospheric climate model (RACMO2), described in more detail in ref. 16 and the Supplementary Information.
Experiment advantages and limitations.
Advantages of this model domain are, first, that this is a simple model applied to one of the best observed and instrumented glaciers on the Antarctic Peninsula. Second, Glacier IJR45 is land-terminating and represents a well-constrained system that isolates the controls on surface mass balance. Most notably, we are able to ignore the uncertainties associated with a more complex oceanic and tidewater glacier system. By restricting the number of assumptions and independent variables, we are able to present an entirely new and original analysis of glacier–climate sensitivities in a critical, and rapidly changing, region. Third, Holocene dynamics are well constrained by detailed geomorphological data and the ice core3, 20.
Limitations of the model include the debris cover on the snout of the glacier (Fig. 1c, d); the glacier bed is interpolated underneath the debris cover. The effect of the debris cover on ablation is taken into account by the degree-day factors. However, the debris cover is sparse, is likely to have accumulated only recently, and is not considered an important factor in this study. Measurements of temperature, velocity, accumulation and ablation are short (2–3 years). Glacier IJR45 receives a high volume of wind-blown snow, rendering precipitation lapse rates calculated from accumulation recorded at sea level and at the summit of Mount Haddington inappropriate, as well of low confidence. Given the limited altitudinal range of this glacier and its forefield, the precipitation lapse rate is considered to be 0, and precipitation is distributed evenly across the glacier surface.
The 10,000 year Holocene experiment finishes with a glacier that is larger than that of the present day, but is rapidly receding. This is a limitation in the model; the enlarged modelled glacier is unable to respond fast enough to the rapidly increasing air temperatures.
As the forefield is very flat, adding mass from an adjoining flow unit could force a more rapid re-advance. However, Glacier IJR45 needs to be relatively advanced before it would be affected by adjacent ice. During an advance, adjacent ice may have enhanced expansion, but with limited effect. If it did enhance an earlier advance during lesser cooling, it would logically also have to add to the biggest advance during the late Holocene, so although adjacent ice may affect the absolute length of IJR45, it would not change the pattern of modelled response.