The HadGEM3 (HadGEM3-GC3.1 or HadGEM3-GC3.1-N96ORCA1) PI simulation was initialized using the standard CMIP6 protocol using constant 1850 GHGs, ozone, solar, tropospheric aerosol, stratospheric volcanic aerosol and land-use forcing. The PI spin-up was 700 model-years, which allowed the land and oceanic masses to attain approximate steady state. The HadGEM3 LIG (Last Interglacial) simulation was initialized from the end of the spin-up phase of the equivalent pre-industrial (PI) simulation. After initialization, the LIG was run for 350 model-years. This 350 LIG spin-up permits the model to reach atmospheric equilibrium and to achieve an upper-ocean equilibrium. The model was then run for a further 200 model-years of LIG production run. This has been demonstrated to be an adequate run length to appropriately capture the model internal variability. This dataset contains outputs from the 200 years of production run of the period. The HadCM3 PI simulation was run for a period of over 600 years. The HadCM3 LIG simulation was initialized from the end of a previous CMIP5 LIG simulation, which was of length 400 years and initiated from the end of the corresponding PI, and run for further 250 years. The total spin-up phase for the HadCM3 LIG simulation used in this study was thus 600 model-years, and the length of the production (at atmospheric and upper-oceanic equilibrium) LIG HadCM3 simulation is 50 model-years. This work was funded by NERC standard research grant nos. NE/P013279/1 and NE/P009271/1.

Global monthly outputs of orography, surface air temperature and water stable isotopes (d18O) were run by the isotope-enabled atmosphere/ocean coupled model HadCM3 for the last interglacial (128 ka). An ensemble of ten idealised Antarctic Ice Sheet (AIS) simulations were processed, included a pre-industrial and a last interglacial control simulations. The eight other simulations used changed topography of the AIS relative to Dome C to ensure the preservation of the atmospheric pathways. The simulations were run 100 years and the last 50 years were used for the analyses. This work was funding through the European Research Council under the Horizon 2020 research and innovation programme (grant agreement No 742224, WACSWAIN) and NERC grant NE/P009271/1.

The file contains Southern Hemisphere winter (September) sea ice concentration (sic) from a simulation performed using the isotope-enabled HadCM3 climate model forced with early last interglacial boundary conditions, centred approximately 128,000 years ago. The resulting sic represents a reduction in winter sea ice area of approximately 54% relative to pre-industrial and is proposed as the best explanation for the Antarctic ice core data from 128,000 years ago. The spatial pattern of sea ice retreat was determined using a large ensemble of model experiments and a pattern search optimization approach to match the last interglacial ice core isotope peak. Further details can be found in the published manuscript (https://doi.org/10.1002/2017GL074594). This work was funded by NERC grants NE/P009271/1, NE/P013279/1, and NE/K004514/1.

We present steady-state ice thickness, bed elevation, and ice surface elevation output from simulations of the Antarctic Ice Sheet (AIS) on a suite of reconstructed Antarctic palaeotopographies using the DeConto and Pollard (DP16) ice sheet model. Ice surface mass balance inputs were provided using the GENESIS v3.0 global atmosphere general circulation model coupled to a 50 m slab ocean model, which provides boundary meteorology for the RegCM3 regional climate model. Three climate/ocean scenarios were simulated: (1) cold climate orbital parameters, preindustrial CO2 levels (280 ppm) and modern ocean temperatures, (2) a subsequent shift to warm climate orbital parameters, an increase in CO2 levels to 500 ppm, and a 5 deg C ocean temperature rise, and (3) as for (2), but with CO2 levels increased to 840 ppm. The steady-state simulations were performed on a suite of reconstructed Antarctic palaeotopographies pertaining to the following four time slices: (1) the Eocene-Oligocene boundary (EOB; ca. 34 Ma), (2) the Oligocene-Miocene boundary (OMB, ca. 23 Ma), (3) the mid-Miocene (MM; ca. 14 Ma), and (4) the mid-Pliocene (MP; ca. 3.5 Ma). Simulations were performed for minimum, median, and maximum end-member topographies, and equivalent simulations were run on the modern (ice-free) Antarctic bed topography for comparison. Further details are given in the accompanying publication. For more information, please contact G. Paxman. Funding was provided by NERC Ph.D. studentship NE/L002590/1.

This data set comprises sea ice-related biomarkers for three time intervals, corresponding to the pre-MPT (1.53-1.36 Ma), the MPT (1.22-0.8 Ma), and the post-MPT (0.5-0.34 Ma) climate cycles from International Ocean Discovery Program (IODP) Site U1343 in the eastern Bering Sea (57deg33.4''N, 176deg49.0''W, 1950 m water depth). The biomarkers are the Arctic sea ice biomarker IP25, together with HBI III and brassicasterol, indicative of open water in the ice marginal zone and general phytoplankton production, respectively. Funding was provided by the NERC grant NE/L002434/1.

Uncertainties in future sea level projections are dominated by our limited understanding of the dynamical processes that control instabilities of marine ice sheets. A valuable case to examine these processes is the last deglaciation of the British-Irish Ice Sheet. The Minch Ice Stream, which drained a large proportion of ice from the northwest sector of the British-Irish Ice Sheet during the last deglaciation, is well constrained, with abundant empirical data which could be used to inform, validate and analyse numerical ice sheet simulations. We use BISICLES, a higher-order ice sheet model, to examine the dynamical processes that controlled the retreat of the Minch Ice Stream. We simulate retreat from the shelf edge under constant "warm" surface mass balance and subshelf melt, to isolate the role of internal ice dynamics from external forcings. The model simulates a slowdown of retreat as the ice stream becomes laterally confined at a "pinning-point" between mainland Scotland and the Isle of Lewis. At this stage, the presence of ice shelves became a major control on deglaciation, providing buttressing to upstream ice. Subsequently, the presence of a reverse slope inside the Minch Strait produces an acceleration in retreat, leading to a "collapsed" state, even when the climate returns to the initial "cold" conditions. Our simulations demonstrate the importance of the Marine Ice Sheet Instability and ice shelf buttressing during the deglaciation of parts of the British-Irish Ice Sheet. Thus, geological data could be used to constrain these processes in ice sheet models used for projecting the future of our contemporary ice sheets. Funding was provided by the Natural Environment Research Council (NERC) SPHERES Doctoral Training Partnership (NE/L002574/1) with CASE support from the British Geological Survey.

This dataset presents the input and output data from a set of sensitivity experiments to simulate the evolution of the Laurentide ice sheet in the Early Holocene (10-7 thousand years ago). These data are presented in the manuscript "Simulating the Early Holocene demise of the Laurentide Ice Sheet with BISICLES (public trunk revision 3298)". Simulating the demise of the Laurentide Ice Sheet covering the Hudson Bay in the early Holocene is important for understanding the role of accelerated changes in ice sheet topography and melt in the ''8.2 ka event'', a century long cooling of the Northern Hemisphere by several degrees. Freshwater released from the ice sheet through a surface mass balance instability (known as the saddle collapse) has been suggested as a major forcing for the 8.2 ka event, but the temporal evolution of this pulse has not been constrained. Dynamical ice loss and marine interactions could have significantly accelerated the ice sheet demise, but simulating such processes requires computationally expensive models that are difficult to configure and are often impractical for simulating past ice sheets. Here, we developed an ice sheet model setup for studying the Laurentide Ice Sheet''s Hudson Bay saddle collapse and the associated meltwater pulse in unprecedented detail using the BISICLES ice sheet model, an efficient marine ice sheet model of the latest generation, capable of refinement to kilometre-scale resolution and higher-order ice flow physics. The setup draws on previous efforts to model the deglaciation of the North American Ice Sheet for initialising the ice sheet temperature, recent ice sheet reconstructions for developing the topography of the region and ice sheet, and output from a general circulation model for a representation of the climatic forcing. The modelled deglaciation is in agreement with the reconstructed extent of the ice sheet and the associated meltwater pulse has realistic timing. Furthermore, the peak magnitude of the modelled meltwater equivalent (0.07-0.13 Sv) is compatible with geological estimates of freshwater discharge through the Hudson Strait. The results demonstrate that while improved representation of the glacial dynamics and marine interactions are key for correctly simulating the pattern of early Holocene ice sheet retreat, surface mass balance introduces by far the most uncertainty. The new model configuration presented here provides future opportunities to quantify the range of plausible amplitudes and durations of a Hudson Bay ice saddle collapse meltwater pulse and its role in forcing the 8.2 ka event. Ilkka Matero was funded by the Leeds-York Natural Environment Research Council (NERC) Spheres Doctoral Training Partnership (NE/L002574/1). The contribution from Ruza Ivanovic was partly supported by NERC grant NE/K008536/1. Lauren Gregoire is funded by a UKRI Future Leaders Fellowship (MR/S016961/1). The work made use of the N8 HPC facilities, which are provided and funded by the N8 consortium and EPSRC (EP/K000225/1) and co-ordinated by the Universities of Leeds and Manchester.