Constraining the dynamic evolution of past ice sheets is critical for unravelling their responses to external forcing and feedbacks over long timescales. This is particularly true in the context of marine ice sheet collapse, as this is one of the largest sources of uncertainty for future sea-level rise projections. The Iceland Ice Sheet (IIS) provides an empirically constrained case study for investigating such an instability, having retreated from a predominantly marine-based ice sheet to isolated mountain ice caps during the last deglaciation. However, previous reconstructions of the IIS have been limited by either sparse data or a restricted exploration of model parameter space, lacking a robust quantification of uncertainties. Here, we address this gap by performing a truncated history matching of the last glacial cycle of the IIS. We use the Glacial Systems Model (GSM) constrained by a curated set of geochronological data to generate an envelope of not-ruled-out-yet(NROY) ice sheet histories.Our results indicate that numerous asynchronous ice streams effectively drain ice from the interior to the margins, resulting in an extensive yet relatively thin ice sheet. During its local Last Glacial Maximum (23.6-20.9 ka), the IIS reaches the continental shelf edge in most sectors with a total volume of 0.41 to 0.76 metres equivalent sea level (m e.s.l.). In the most extreme NROY glaciation scenarios, our model reveals an ice bridge connecting the Iceland and Greenland ice over Denmark Strait.We find that accelerated ice discharge (at the grounding line) dominates mass loss during deglaciation. This acceleration is primarily driven by atmospheric warming through a cascade of mechanisms: surface meltwater induces hydrofracturing, leading to both ice shelf disintegration and tidewater calving, which in turn reduces buttressing and triggers rapid ice stream acceleration. The critical role of hydrofracturing in enabling model capture of deglacial data constraints is shown by explicit sensitivity experiments. This thereby supports inclusion of hydrofracturing for modelling of ongoing ice sheet response to climate change.

GLAC3 is the first history matching of every last glacial cycle ice sheet. It therefore includes North American, Greenlandic, Icelandic,Eurasian, Tibetan, Patagonian, and Antarctic components. Instead of determining a non-robust "optimal" chronology, history matching aimsto "bound reality" with robust assessment of both proxy data and model (both parametric and structural) uncertainties. For the four major icesheets, this entails Bayesian artificial neural network emulation of the glaciological model predictions to enable adequate Markov ChainMonte Carlo sampling of chronologies. The history matching is against a large set of geophysical (such as relative sea level and marinelimit), geological (cosmogenic exposure and C14 ages), and glaciological (such as present-day ice surface velocity) constraints.Aside from being a product of history matching, GLAC3 has two additional unique features. Firstly, it is the only availabledeglacial, let alone full glacial cycle, global set of chronologies from glaciological modelling, using the Glacial Systems Model(GSM) with hybrid shallow ice and shallow shelf ice dynamic. This enables physical resolution of ice sheets, ice streams, ice shelves,and grounding line migration. As such, GLAC3 is subject to glaciological constraints such as borehole temperature profiles thatnon-glaciological reconstructions can't resolve. Secondly, the glaciologically modelling is self-consistently coupled with fullvisco-elastic glacio-isostatic adjustment enabling joint history matching of ice history and regional earth viscosity.The presentation will focus on the relative phasing of each ice sheet, rates of mass gain and loss, and rates of ice margin migration. Thiswill be compared against both far-field relative sea level records as well as the results of fully coupled ice and climate modelling of thelast glacial cycle with LCice (LOVECLIM + GSM).

We present a glacial isostatic adjustment (GIA) analysis for a joint ice and GIA history matching of the Antarctic Ice Sheet (AIS) since the last interglacial. This was achieved using the Glacial Systems Model (GSM) – which includes a glaciological ice sheet model asynchronously coupled to a viscoelastic Earth model. A large ensemble of 9293 simulations was conducted using the GSM. The history matching was against the AntICE2 database, which includes observations of past relative sea level, present-day (PD) vertical land motion, past ice extent, past ice thickness, borehole temperature profiles, PD geometry and surface velocity (Lecavalier et al., 2023). The 38 ensemble parameters of the GSM were history matched using Markov Chain Monte Carlo sampling that in turn employed Bayesian Artificial Neural Network emulators. The implications on the evolution of the AIS are detailed in a companion paper which predominantly focuses on the ice sheet component (Lecavalier and Tarasov, 2025). The history-matching analysis identified simulations from the full ensemble that are Not-Ruled-Out-Yet (NROY) by the data. This yielded a NROY sub-ensemble of simulations consisting of 82-members that approximately bound past and present GIA and sea-level change given uncertainties across the entire glacial system. The NROY Antarctic ice sheet chronologies and associated Earth viscosity models represent the Antarctic component of the “GLAC3-A” set of global ice sheet chronologies over the last glacial cycle. The NROY set of ice sheet histories in combination with a wide range of Earth rheologies is evaluated against available data. Data-model comparisons are shown against a subset of the AntICE2 database which directly constrains relative sea-level (RSL) change and bedrock displacement. This displays significant spatial variability in Antarctic GIA. The limited number of observational constraints contributes to wide inferred RSL bounds with max/min ranges up to 150 m during the Holocene. Finally, estimates of PD rates of bedrock displacement with tolerance intervals are presented and compared against those from previously published inferences. These previous Antarctic GIA studies are key inputs for geodetic studies of the contemporary AIS mass balance. We demonstrate that, by adequately exploring glacial and rheological uncertainties against a comprehensive database, past studies have underestimated Antarctic GIA uncertainties across large sectors, while other sectors are now more narrowly constrained. This history matching presents meaningful Antarctic GIA bounds of the rate of PD bedrock displacement with direct implications on mass balance estimates of the PD AIS.

We quantify the contribution of glacial isostatic adjustment (GIA) to land subsidence and sea-level rise along the Atlantic coast of Europe. Using both geologic reconstructions of relative sea-level (RSL) change and geodetic determinations of vertical land motion, we assess GIA model accuracy and determine model parametric uncertainty. Seven ice history models and 440 spherically symmetric Earth viscosity models were used to explore the model parameter space. Our results indicate that the inferred earth model parameters are highly dependent on the barystatic component of the adopted ice sheet model. On the other hand, the modelled RSL output is relatively insensitive to variations in the ice thickness distribution of the Eurasian component of the ice history model. Of the two global ice models considered (ANU and ICE-6G) our results demonstrate the barystatic component of the ANU model to be more accurate. Model uncertainty was determined using data-model misfit values and a Bayesian approach to define a subset of parameter sets (1 sigma confidence). When considering the model uncertainty, 95 % of the Holocene RSL observations and 93 % of the present-day vertical land motion rates can be explained at this confidence level. Our results support previous work in showing that GIA-related subsidence along the Atlantic coast of Europe is dominated by ice-loading (peripheral bulge) effects and that ocean loading is also important in some areas, such as northwestern France. Using our subset of best-fitting parameter sets, we performed a sea-level budget analysis for the period 1957 to 1997 using data from 10 tide gauges in our study region and find that the sterodynamic signal is the largest contributor at most of the considered tide gauge sites, followed by contributions from GIA and glaciers. The GIA signal dominates the modelled uncertainty at all sites. Of the 10 tide gauge stations considered, the budget is closed at six (to within 2 sigma uncertainty). The largest residual was found for station Dieppe in northern France, with an unexplained signal of 3.8 +/- 0.6 mm/yr, which appears to be related to localized subsidence at the tide gauge site.

We document the glacial system model (GSM), which is designed for large ensemble ice sheet modelling in glacial cycle contexts. A distinguishing feature is the extent to which it addresses relevant forcing and process uncertainties. The GSM has evolved from three decades of effort to constrain the last glacial cycle evolution of each ice sheet that was present (North American, Greenlandic, Icelandic, Eurasian, Patagonian, and Antarctic, and soon Tibetan). The core ice dynamics uses a hybrid shallow-shelf and shallow-ice approximation with full thermo-mechanical coupling. It also includes one of the largest range of relevant processes for the above context of any model to date, ranging from visco-elastic glacial isostatic adjustment with 0-order geoidal deflection to state-of-the-art subglacial sediment production, transport, and deposition. Furthermore, the GSM is to date the only model to have all of the above processes bidirectionally coupled with each other. Other relevant distinguishing features include: permafrost resolving bed-thermodynamics, a fast diagnostic solution of down-slope surface drainage and lake filling, subgrid hypsometric surface mass balance and ice flow, simple thermodynamic lake and sea ice representations, subglacial hydrology with dynamically evolving partitioning between distributed and channelized flow, and surface melt that physically accounts for insolation changes via a novel insolation above freezing scheme. To address the most challenging part of paleo ice sheet modelling, the GSM includes both a 2D energy balance climate model and variants of traditional input time series weighted interpolation (aka “glacial indexing”) of fields from General Circulation Model (GCM) simulations, all under ensemble parametric specification. It also includes options for one and two way scripted coupling with climate models. We demonstrate the significant errors that can ensue in the glacial cycle simulation of a single ice sheet when three aspects of glacial isostatic adjustment are ignored (as is typical). These are geoidal deformation, global ice load input, and correction of initial topography for present-day isostatic disequilibrium. We also draw attention to the relatively high sensitivity of the GSM (and presumably other ice sheet models) to the specification of the temperature dependence for basal sliding activation. The associated code archive includes configuration options for all major last glacial cycle ice sheets as well as idealized geometries and validation test setups.

The mid-Pleistocene Transition (MPT) from 41 kyr to 100 kyr glacial cycles occurred in the absence of a change in orbital forcing. This presents a challenge for the Milankovitch theory of glacial cycles. A change from a low to high friction bed under the North American Ice Complex through the removal of pre-glacial regolith is hypothesized to play a critical role in crossing the threshold to longer and stronger glaciations. However, testing this Regolith Hypothesis requires constraint on currently unknown pre-glacial regolith cover as well as assessing whether glacial sediment processes remove the appropriate amount of regolith to enable glacial system change consistent with the MPT. Pleistocene regolith removal has not yet been simulated for a realistic, 3D North American ice sheet fully considering basal processes. Constraints on pre-glacial bed elevation and sediment thickness are sparse and the bounds are wide.What limits on pre-glacial regolith thickness in North America can be inferred from our current understanding of glacial processes and the present-day distribution of unconsolidated sediment? How does pre-glacial sediment thickness influence the evolution of Pleistocene glacial cycles? We answer these questions with an ensemble of whole-Pleistocene simulations with high-variance parametrizations and range of pre-glacial regolith thicknesses.We use the 3D Glacial Systems Model which incorporates the relevant glacial processes: 3D thermomechanically coupled hybrid SIA/SSA ice physics, fully coupled sediment production and transport, subglacial linked-cavity and tunnel hydrology, isostatic adjustment from dynamic loading and erosion, and climate from a 2D non-linear energy balance model and glacial index. This fully coupled system is driven only by atmospheric CO2 and insolation. The model captures the Pleistocene evolution of North American glaciation: 41 to 100 kyr glacial cycles shift, similar latitudinal extent in the early and late Pleistocene, LGM ice volume, deglacial ice margin chronology, and the broad present-day sediment distribution within the parametric and observational uncertainty. Constrained by large scale reconstructions of present-day surface sediment distribution, regional sediment distribution estimates, and regional bedrock erosion estimates, these results bound the mean pre-glacial sediment thickness.Our results suggest thin (

In this study we present the evolution of the Antarctic Ice Sheet (AIS) since the Last Interglacial. This is achieved by means of a history-matching analysis where a newly updated observational database (AntICE2) is used to constrain a large ensemble of 9293 model simulations. The Glacial Systems Model (GSM) configured with 38 ensemble parameters was history-matched against observations of past ice extent, past ice thickness, past sea level, ice core borehole temperature profiles, present-day uplift rates, and present-day ice sheet geometry and surface velocity. Successive ensembles were used to train Bayesian artificial neural network emulators. The parameter space was efficiently explored to identify the most relevant portions of the parameter space through Markov chain Monte Carlo sampling with the emulators. The history matching ruled out model simulations which were inconsistent with the observational-constraint database. During the Last Interglacial (LIG), the AIS yielded several metres equivalent sea level (m e.s.l.) of grounded ice volume deficit relative to the present, with sub-surface ocean warming during this period being the key uncertainty. At the global Last Glacial Maximum (LGM), the best-fitting sub-ensemble of AIS simulations reached an excess grounded ice volume relative to the present of 9.2 to 26.5 m e.s.l. Considering the data do not rule out simulations with an LGM grounded ice volume >20 m e.s.l. with respect to the present, the AIS volume at the LGM can partly explain the missing-ice problem and help close the LGM sea-level budget. Moreover, during the deglaciation, the state space estimation of the AIS based on the GSM and near-field observational constraints allows only a negligible Antarctic Meltwater Pulse 1a contribution (−0.2 to 0.3 m e.s.l.).

The rapid transformation of the ocean–atmosphere–cryosphere system during the last deglaciation holds clues that may be useful to understand climate change during this century and beyond. Within this context, sea-level change connects the climate system components, but the limited temporal resolution and spatial distribution of relative sea-level records has hindered progress towards closing the ice sheet–sea-level budget since the Last Glacial Maximum. Here we present a relative sea-level record, compiled using radiocarbon-dated basal peat, from the Mississippi Delta stretching back to ~10,000 years ago and combine it with the best available relative sea-level data worldwide for the final episode of the last deglaciation (9,000–7,000 years ago). Geophysical modelling shows that these precise data constraints favour approximately 14 m of ice melt in North America through this interval, 4–10 m greater than previously estimated, and at least three times greater than the Antarctic contribution. Our results call for a major revision of the deglacial ice history with implications for, among others, collapse of the saddle connecting two ice domes over Hudson Bay, the associated abrupt cooling ~8,200 years ago and the sensitivity of the Atlantic Meridional Overturning Circulation to freshwater forcing. The melting of the last remnants of the North American ice sheets in the early Holocene led to 14 m of global sea-level rise, higher than prior estimates, according to proxy constraints from the Mississippi Delta and other localities.

So far, melting of the Greenland Ice Sheet (GrIS) has been the biggest contributor from the Earth’s cryosphere to global sea-level rise. Major uncertainties remain about how oceanic heat is transported across the shelf and through the fjords to the faces of marine-terminating glaciers, and how this affects rates of ice melt and calving. In turn, the increasing supply of meltwater and nutrients to the ocean around Greenland is impacting marine ecosystems as primary productivity rises, subsequently increasing the potential for carbon to be buried as “blue carbon” in Greenland’s fjords as warming continues. In July-August 2024, the UK-funded KANG-GLAC project completed a 40-day multidisciplinary research cruise to SE Greenland where the 40-strong scientific party made a suite of integrated geological, ocean and biological observations. The main aims of the project are two-fold. First, it aims to better understand how marine-terminating glaciers respond to oceanic heat on longer timescales (decades to centuries) by reconstructing glacier and ice-sheet behaviour during the Holocene and in particular during the climatic warm period of the Holocene Thermal Maximum. Second, the project will quantify nutrient cycling in the water column and uppermost seafloor sediments in order to improve our knowledge of marine ecosystem response to meltwater supply from the GrIS. The cruise on the UK’s premier polar research vessel, the RRS Sir David Attenborough, is the start of a 3.5 year project. Here, we will present an overview of our field observations in this past-to-future project and outline the plans for future data-driven modelling of the Greenland Ice Sheet.

Deglacial abrupt warming event is a ubiquitous feature of deglaciations during the Late Pleistocene. Nevertheless, during the last deglaciation an unusually early onset of abrupt Northern Hemisphere warming event, known as Bølling/Allerød (B/A) warming, complicates our understanding of their underlying dynamics, especially due to the large uncertainty in histories of ice sheet retreat and meltwater distributions. Here applying the latest reconstruction of ice sheet and meltwater flux, we conduct a set of transient climate experiments to investigate the triggering mechanism of the B/A warming. We find that the realistic spatial distribution of meltwater flux can stimulate the warming even under a persistent meltwater background. Our sensitivity experiments further show that its occurrence is associated with an orbitally induced climate self-oscillation under the very deglacial climate background related to atmospheric CO2 level and ice sheet configurations. Furthermore, the continuous atmospheric CO2 rising and ice sheet retreating appear to mute the oscillation by freshening the North Atlantic via modulating moisture transport by the Westerly.

To date, the Icelandic Ice Sheet (IIS) and Patagonian Ice Sheet (PIS) have been poorly understood with regard to their configuration, dynamics, and evolution during the last glacial cycle. The few glaciological modelling studies of the IIS and PIS to date have placed minimal attention on addressing model uncertainties. As such, their inferential value is poorly interpretable. To address this, we present the results of history matchings of the 3D Glacial Systems Model (GSM) against curated sets of paleo constraints for the last glacial cycle IIS and PIS. History matching identifies a set of model simulations that are not ruled out given available data constraints and robust uncertainty analysis (including both model and data uncertainties). As such, it aims to “bracket reality” as opposed to the much more difficult task of determining a meaningful most likely chronology.The GSM is a thermo-mechanically coupled glaciological model with hybrid shallow ice and shallow shelf/stream physics. The climate forcing consists of a fully coupled energy balance climate model and glacial indexed climate forcing using the results of PMIP3 (Paleo Model Intercomparison Project). Approximate 30 GSM ensemble parameters partially account for uncertainties in climate, basal drag, and marine ice processes. The GSM configuration includes fully coupled visco-elastic glacio-isostatic adjustment enabling physically self-consistent relative sealevel predictions. Our presentation focuses on bracketing chronologies for the last glacial cycle IIS and PIS as well as disentangling the relative contribution of atmospheric and marine forcings on mass loss during the deglaciation.

Abstract. Little is known about the evolution of continental ice sheets through the last two glacial inceptions (Marine isotope stages, MIS 7d and MIS 5d). Here, we present the results of a perturbed parameter ensemble of transient simulations of the last two glacial inceptions and subsequent interstadials (MIS 7e-7c, 240–215 ka and MIS 5e-5c, 122–98 ka) with the fully coupled ice/climate model LCice. LCice includes all critical direct feedbacks between climate and ice. As shown herein, it can capture the inferred sea level change (of up to 80 m) of the last two glacial inceptions within proxy uncertainty. One key underlying question of paleoclimate dynamics is the non-linear state dependence of the climate system. Concretely, in a model-centric context, to what extent does the capture of one climate interval in an Earth systems model guarantee capture of another interval? For LCice, the capture of present-day climate is insufficient to predict capture of glacial inception climate, as only a small fraction of ensemble members that performed "well" for present-day captured inception. Furthermore, the capture of inferred sea level change in one inception has weak correlation with the same outcome for the other. After partial history matching against present-day and past sea level constraints, the resultant NROY (not ruled out yet) ensemble of simulations have a number of features of potential interest to various paleo communities, including the following. (i) In correspondence with the inferred last glacial maximum configuration, the simulated North American ice sheets are substantially larger than the Eurasian ice sheet throughout MIS 5d-MIS 5c and MIS 7d-MIS 7c. (ii) Hudson Bay can transition from an ice-free state to full ice cover (grounded ice) within 2000 years. (iii) The North American and Eurasian ice sheets advanced southward with rates well above 100 m/yr during the penultimate glacial inception and over 70 (Eurasia) and 90 (North America) m/kyr during last glacial inception. (iv) the Laurentide and Cordilleran ice sheets merge in their northern sectors in 13 out of 14 NROY simulations for MIS 7d, contrary to what is assumed from limited geological data. (v) larger ice sheets display a larger lag in the timing of stadial maximum ice volume compared to that of the insolation minimum; the North American ice sheet maximum lags 5.3 ± 0.5 kyrs behind the MIS 7d insolation minimum. Supplemental resources include a dynamic display of ice advance and subsequent retreat for a sub-ensemble of 14 NROY simulations from MIS 5d-5c and MIS 7d-7c.

Ice sheet evolution profoundly influences the climate system through changes in orography, surface albedo, freshwater fluxes to the ocean, and ocean gateways. The changes to the climate system will, in turn, affect the ice sheets, leading to complex feedback loops. To date, the relative roles of these feedback loops have not been examined over a full glacial cycle. To address this, we employ the glacial earth system model of intermediate complexity LCice in transient simulations of the complete last glacial cycle. Through ensemble-based sensitivity experiments, we isolate the impact of individual ice sheet orography, albedo, meltwater input, Bering Strait opening/closure, and glacio-isostatic adjustment on the climate system and back onto the ice sheet evolution itself. To assess possible state dependencies, we compare the individual impact of ice-climate feedbacks on both the ice sheet growth and decay phase around MIS 5d (Last Glacial Inception) and MIS 2 (Last Glacial Maximum). The sensitivity of the North American and Eurasian ice sheets to some feedbacks changes from MIS 5d to MIS 2, suggesting a potential threshold behaviour and complex non-linear dynamics. Our analysis also examines which characteristics of last glacial cycle ice sheet evolution are relatively robust and which are more likely to be highly sensitive to incompletely resolved feedback loops. This work thereby not only improves our understanding of paleo ice/climate coupled dynamics but also identifies feedback pathways that are likely to generate the largest uncertainties in coupled ice and paleoclimate modelling.

The Younger Dryas (YD; 12.9 to 11.7 thousand years before present [ka]) was an abrupt cooling event in the North Atlantic region that interrupted the last deglaciation. This interval is an important analogue for ice inception and provides key insights into the impacts of millennial-scale climatic excursions on the broad Earth System. However, the effect of YD cooling on North American ice sheets - by far the largest ice masses in the Northern Hemisphere at that time - is poorly understood. Here, we document our current level of knowledge in ice margin behavior (stabilization, advance or continued retreat) that occurred to North American ice sheets during the YD interval. We assign a quality score for each segment of the YD ice margin that takes into account (i) our degree of confidence in the YD ice position (e.g. whether prominent moraines were formed) and (ii) the quality of chronological control on the ice position (e.g. assumed vs. directly dated). We divide the North American ice sheets into 11 regions and discuss changes in behavior during the YD interval. In some cases, our level of knowledge on the local YD ice margin behavior and chronological control is good (e.g. Ten Mile Lake moraine [Newfoundland], Collins Pond Phase [Atlantic Canada], Grand Marais I moraine, Dog Lake moraine, Hartman-Lac Seul moraine [all in the region of Lake Superior], and Cree Lake moraine [northern Saskatchewan]). In other cases, our knowledge is poor, and the ice margin behavior and age assignment are inferred or interpolated across the broad region. We also bring together paleoclimate data from 155 sites (largely lakes, ponds and peatlands) situated near the YD ice margin to provide additional context for any local or regional climate variations during that interval and to explore potential linkages between cooling and nearby ice margin behavior. We finish by discussing YD ice sheet response in other areas (ie. Svalbard and Fennoscandia); outlining the possible mechanisms for YD cooling; delving into numerical modelling through the YD interval and suggesting future improvements in empirical and numerical approaches.

Past sea level change informs our predictions of future sea level rise [1] and, more broadly, the large scale climate changes that drove ice sheet growth and decay [2]. Approximately one million years ago, the relationship between orbitally-driven insolation and glacial cycle sea level change shifted – small, 40 thousand year cycles ceded to large, 100 thousand year cycles[3]. A leading hypothesis for this Mid-Pleistocene Transition is that the North American ice sheet grew resilient to deglaciation by removing its bed of slippery regolith, exposing the high friction bedrock underneath[4]. Our data constrained numerical experiments show that the North American ice sheet exposed bedrock over an area similar to present day by 1.5 million years ago, well before the Mid-Pleistocene Transition began. Thicker pre-glacial regolith does not simply delay its removal and thus the transition – it limits ice-ocean interaction by infilling shallow marine areas while cooling via the elevation-lapse rate feedback. Topographic changes over previously glaciated domains play a critical role in early Pleistocene glaciation – enabling larger ice sheets during warmer periods. This highlights the important and under-appreciated role topographic change plays in past sea

Despite their recognized significance on global climate and extensive research efforts, the mechanism(s) driving Heinrich events remain(s) a subject of debate. Here, we use the 3D thermomechanically coupled glacial systems model (GSM) to examine the Hudson Strait ice stream surge cycling and the role of three factors previously hypothesized to play a critical role in Heinrich events: ice shelves, glacial isostatic adjustment, and sub-surface ocean temperature forcings. In contrast to all previous modeling studies examining HEs, the GSM uses a transient last glacial cycle climate forcing, global viscoelastic glacial isostatic adjustment model, and sub-glacial hydrology model. The results presented here are based on a high-variance sub-ensemble retrieved from North American history matching for the last glacial cycle.Over our comparatively wide sampling of the potential parameter space (52 ensemble parameters for climate forcing and process uncertainties), we find two modes of Hudson Strait ice streaming: classic binge-purge versus near-continuous ice streaming with occasional shutdowns and subsequent surge onset overshoot. Our model results indicate that large ice shelves covering the Labrador Sea during the last glacial cycle only occur when extreme calving restrictions are applied. The otherwise minor ice shelves provide insignificant buttressing for the Hudson Strait ice stream. While sub-surface ocean temperature forcing leads to minor differences regarding surge characteristics, glacial isostatic adjustment does have a significant impact. Given input uncertainties, the strongest controls on ice stream surge cycling are the poorly constrained deep geothermal heat flux under Hudson Bay and Hudson Strait and the basal drag law. Decreasing the geothermal heat flux within available constraints and/or using a Coulomb sliding law instead of a Weertman-type power law leads to a shift from the near-continuous streaming mode to the binge-purge mode.

Models of glacial isostatic adjustment (GIA) play a central role in the interpretation of various geologic and geodetic data to understand and simulate past and future changes in ice sheets and sea level, as well as to infer rheological properties of the deep Earth. During the past few decades, a major advance has been the development of models that include 3D Earth structure, as opposed to 1D spherically symmetric (SS) structure. However, a major limitation in employing 3D GIA models is their high computational expense. As such, we have developed a method using artificial neural networks (ANNs) and the Tensorflow library to predict the influence of 3D Earth models with the goal of more affordably exploring the parameter space of these models, specifically the radial (1D) viscosity profile to which the lateral variations are added.Our goal is to test whether the use of an ANN to produce a fast surrogate model can accurately predict the difference in GIA model outputs (i.e., relative sea level (RSL) and uplift rates) for the 3D case relative to the SS case. If so, the surrogate model can be used with a computationally efficient SS (Earth) GIA model to generate output that replicates that from a 3D (Earth) GIA model. Evaluation of the surrogate model performance for deglacial RSL indicates that it is able to provide useful estimates of this field throughout the parameter space when trained on only approximate to 15% (approximate to 50) of the parameter vectors considered (330 in total).We applied the surrogate model in a model-data comparison exercise using RSL data distributed along the North American coasts from the Canadian Arctic to the US Gulf Coast. We found that the surrogate model is able to successfully reproduce the model-data misfit values such that the region of minimum misfit either generally overlaps the 3D GIA model results or is within two increments of the radial viscosity model parameter space (defined here as lithosphere thickness, upper-mantle viscosity, and lower-mantle viscosity). The surrogate model can, therefore, be used to accurately explore this aspect of the 3D Earth model parameter space. In summary, this work demonstrates the utility of machine learning in 3D Earth GIA modelling, and so future work to expand on this initial proof-of-concept analysis is warranted.

Abstract The rapid transformation of the ocean-atmosphere-cryosphere system during the last deglaciation serves as a potential model for climate change during this century and beyond. Within this context, sea-level change can be viewed as the connecting tissue, but the limited resolution of relative sea-level (RSL) records has hindered progress toward closing the ice sheet–sea level budget since the Last Glacial Maximum, the partitioning of ice melt from different sources, and assessing the role of freshwater forcing in abrupt climate change. Here we present a new RSL record from the Mississippi Delta stretching back to 11 ka and combine it with the best available published RSL data worldwide for the final episode of the last deglaciation (9-7 ka). Glacial isostatic adjustment (GIA) modelling shows that these precise data constraints demand a North American ice melt of about 14 m sea-level equivalent (SLE) during this time, 4-10 m greater than previously estimated. Our results call for a major revision of the North American deglacial ice history and our findings demonstrate the utility of high-resolution RSL observations as a pathway towards closing the ice budget of the last deglaciation and improving our understanding of the ocean-atmosphere-cryosphere system during rapid climate change.

The Marine Isotope Stage 11c (MIS-11c) interglacial is an enigmatic period characterized by a long duration of relatively weak insolation forcing, but it is thought to have been coincident with a large global sea-level rise of 6–13 m. The configuration of the Greenland Ice Sheet during the MIS-11c interglacial highstand is therefore of great interest. Given the constraints of limited data, model-based analysis may be of use but only if model uncertainties are adequately accounted for. A particularly under-addressed issue in coupled climate and ice-sheet modeling is the coupling of surface air temperatures to the ice model. Many studies apply a uniform “lapse rate” accounting for the temperature differences at different altitudes over the ice surface, but this uniformity neglects both regional and seasonal differences in near-surface temperature dependencies on altitude. Herein we provide the first such analysis for MIS-11c Greenland that addresses these uncertainties by comparing one-way coupled Community Earth System Model (CESM) and ice-sheet model results from several different downscaling methodologies. In our study, a spatially and temporally varying temperature downscaling method produced the greatest success rate in matching the constraints of limited paleodata, and it suggests a peak ice volume loss from Greenland during MIS-11c of approximately 50 % compared to present day (∼ 3.9 m contribution to sea-level rise). This result is on the lower bound of existing data- and model-based studies, partly as a consequence of the applied one-way coupling methodology that neglects some feedbacks. Additional uncertainties are examined by comparing two different present-day regional climate analyses for bias correction of temperatures and precipitation, a spread of initialization states and times, and different spatial configurations of precipitation bias corrections. No other factor exhibited greater influence over the simulated Greenland ice sheet than the choice of temperature downscaling scheme.