Clara Burgard

As the Antarctic ice sheet melts in response to climate change, freshwater is released into the ocean over the continental shelf, altering its hydrographic properties, stratification and circulation. This can, in turn, affect melting of Antarctic ice shelves by modifying the delivery of oceanic heat to sub-ice-shelf cavities. To study the response of the ocean to enhanced Antarctic meltwater, we use an equilibrated 1° global ocean-sea ice configuration with relatively realistic continental shelf regimes. We conduct three freshwater perturbation experiments of increasing complexity covering the period 1990-2100. Our results show that meltwater injection reduces wintertime convection around Antarctica, by strengthening vertical stratification. Consequently, Dense Shelf Water (DSW) is no longer renewed while Circumpolar Deep Water (CDW) is no longer ventilated on the shelf. CDW thus permanently intrudes onto the lower part of the continental shelf, causing warming there. However, this response depends strongly on how the meltwater perturbation is implemented: warming and CDW intrusion on the shelf are largely mitigated under a realistic forcing design, in which meltwater is distributed vertically and horizontally and the latent heat of fusion is accounted for. Under this realistic forcing, shelf bottom waters warm only slightly (+0.1°C on average in 2100), whereas coastal surface injection causes strong warming (+0.7°C). Further, DSW formation proves more resilient under more realistic forcing, but ultimately ceases under all meltwater injection scenarios considered. These results indicate that models using simplified meltwater anomaly forcing likely exaggerate both continental shelf warming and DSW decline.

Ocean-driven melting at the base of the floating ice shelves of the Antarctic Ice Sheet is an important driver of current and future ice losses. Ocean models that resolve sub-shelf melt rates at sufficient resolution are too expensive to run in large ensembles or for multi-centennial projections. Instead, ice-sheet models use simple modules to infer sub-shelf melt rates based on far-field ocean properties. However, there is no consensus on the optimal set of parameters for these modules. Here, we present a protocol for ice-ocean interactions in ISMIP7 Antarctic Ice Sheet projections. We describe (1)~how CMIP model datasets are processed to be used in melt modules, (2)~how parameters in melt modules can be selected to be physically constrained by oceanographic knowledge, (3)~the rationales for a range of datasets that span possible evolution of the Southern Ocean between 1950 and 2025, (4)~design for a melt Module Intercomparison Project (meltMIP) that will allow us to better assess uncertainty in sea level projections related to melt modules. Finally, we summarise lessons from ISMIP7 and potential research directions towards the next generation of ice-ocean interaction modelling.

The spatial pattern of melting beneath Antarctic ice shelves influences ice flow and retreat, and the resulting freshwater input into the Southern Ocean influences global carbon storage and primary productivity. It is therefore crucial that interactions at the ice-ocean interface are adequately represented in global climate models. However, due to computational limitations, existing climate models are forced to choose between either high resolution or long period simulations, and struggle to resolve melt rates in ice shelf cavities. Here, we show that a simple multilayer perceptron can be used to emulate sub-shelf melt rates in entire small cavities and in the parts of large cavities closest to the grounding zone. We find that the melt rates produced by applying neural network emulators are a good match for melt rates from high resolution simulations, provided that similar conditions were included in the neural network training. We also find that neural networks are particularly sensitive to temperatures and ice-draft slopes outside those used in the training dataset. However, if we train multiple neural networks on the same input data, we demonstrate that the ensemble spread of the neural networks is a good indicator of the reliability of the emulator in any given conditions. The neural network emulator of sub-shelf melt which is presented here, DeepMelt-GL, can be used to improve the representation of ice shelf cavities in both ocean and climate models.

The spatial pattern of melting beneath Antarctic ice shelves influences ice flow and retreat, and the resulting fresh-water input into the Southern Ocean influences global carbon storage and primary productivity. It is therefore crucial that interactions at the ice-ocean interface are adequately represented in global climate models. However, due to computational limitations, existing climate models are forced to choose between either high resolution or long period simulations, and struggle to resolve melt rates in ice shelf cavities. Here, we show that a simple multilayer perceptron can be used to emulate sub-shelf melt rates in the parts of cavities closest to the grounding zone. We find that the melt rates produced by applying neural network emulators are a good match for melt rates from high resolution simulations, provided that similar conditions were included in the neural network training. We also find that neural networks are particularly sensitive to temperatures and ice-draft slopes outside those used in the training dataset. However, if we train multiple neural networks on the same input data, we demonstrate that the ensemble spread of the neural networks is a good indicator of the reliability of the emulator in any given conditions. The neural network emulator of sub-shelf melt which is presented here, DeepMelt-GL, can be used to improve the representation of ice shelf cavities in both ocean and global climate models.

The spatial pattern of melting beneath Antarctic ice shelves influences ice flow and retreat, and the resulting fresh-water input into the Southern Ocean influences global carbon storage and primary productivity. It is therefore crucial that interactions at the ice-ocean interface are adequately represented in global climate models. However, due to computational limitations, existing climate models are forced to choose between either high resolution or long period simulations, and struggle to resolve melt rates in ice shelf cavities. Here, we show that a simple multilayer perceptron can be used to emulate sub-shelf melt rates in the parts of cavities closest to the grounding zone. We find that the melt rates produced by applying neural network emulators are a good match for melt rates from high resolution simulations, provided that similar conditions were included in the neural network training. We also find that neural networks are particularly sensitive to temperatures and ice-draft slopes outside those used in the training dataset. However, if we train multiple neural networks on the same input data, we demonstrate that the ensemble spread of the neural networks is a good indicator of the reliability of the emulator in any given conditions. The neural network emulator of sub-shelf melt which is presented here, DeepMelt-GL, can be used to improve the representation of ice shelf cavities in both ocean and global climate models.

The spatial pattern of melting beneath Antarctic ice shelves influences ice flow and retreat, and the resulting fresh-water input into the Southern Ocean influences global carbon storage and primary productivity. It is therefore crucial that interactions at the ice-ocean interface are adequately represented in global climate models. However, due to computational limitations, existing climate models are forced to choose between either high resolution or long period simulations, and struggle to resolve melt rates in ice shelf cavities. Here, we show that a simple multilayer perceptron can be used to emulate sub-shelf melt rates in the parts of cavities closest to the grounding zone. We find that the melt rates produced by applying neural network emulators are a good match for melt rates from high resolution simulations, provided that similar conditions were included in the neural network training. We also find that neural networks are particularly sensitive to temperatures and ice-draft slopes outside those used in the training dataset. However, if we train multiple neural networks on the same input data, we demonstrate that the ensemble spread of the neural networks is a good indicator of the reliability of the emulator in any given conditions. The neural network emulator of sub-shelf melt which is presented here, DeepMelt-GL, can be used to improve the representation of ice shelf cavities in both ocean and global climate models.

In past decades, substantial advances on understanding ocean–ice-shelf interactions have been made, and a number of parameterisations that provide sub-shelf melting for ice sheet modelling studies have been developed. Through ISMIP6, it was found that the choice of parameter values in melt parameterisations can influence the order of magnitude of melt rate changes in projections. Moreover, it has been shown that constraining those parameters with present-day observations is not sufficient to constrain melt rate changes under future warming. For ISMIP7, we hence propose a “come-as-you-are” approach for the choice of the sub-shelf melt parameterisation, but suggest a protocol for calibrating parameters using ocean model simulations and observations that show large changes in cavity temperatures as additional constraints. This is embedded into an updated and revised protocol for processing CMIP model data for the ISMIP7 Antarctic ice-ocean forcing protocol. Come to the presentation if you are interested to learn (more) about the protocol, discuss your testing experiences, or provide feedback.

The future contribution of the Antarctic Ice Sheet to global sea-level rise remains the largest source of uncertainty in climate projections. This uncertainty is primarily driven by the complex interaction between the ocean and the ice shelf cavities. Most ice sheet models still rely on simplified melt parameterizations that fail to capture the complex oceanographic processes within sub-ice-shelf cavities, while fully coupled ice-ocean models remain too computationally expensive for large-scale sensitivity studies. In this study, we present ADMIRE (Antarctic Deep MELT and Ice REpresentation), a new ongoing-work intermediate-complexity framework. ADMIRE couples the ice sheet model Kori-ULB with DeepMELT, a deep learning emulator trained on high-resolution NEMO-SI3 simulations. This coupling allows for a more physically consistent representation of the ice-ocean interface at a fraction of the computational cost of a coupled ice-sheet-ocean model. We compare the sensitivity of the Antarctic grounding line migration and overall mass balance when using the DeepMELT emulator versus traditional melt parameterizations. Furthermore, we investigate the impact of temporal coupling steps and interpolation methods on the projections. Our preliminary results highlight the potential of machine learning-based emulators to bridge the gap between simple parameterizations and complex coupled models, providing more robust projections of Antarctica’s future but at a low computational cost, allowing for comprehensive and multi-century studies.

Anomalous freshwater fluxes from the Greenland and Antarctic ice sheets and ice shelves are impacting the surrounding oceans, and we need to be able to account for these effects in climate model simulations over the historical period and in future projections. In previous phases of the Coupled Model Intercomparison Project (CMIP), models mostly either assumed that the ice sheets were in mass balance, or that discharge from the ice sheets was constant, but in neither case was the observed increasing discharge over the historical period properly represented. In this paper, we present data products of absolute and anomalous freshwater mass fluxes from both major ice sheets, and recommendations for their use in historical simulations. These fluxes can be implemented in climate simulations as a forcing for models that do not (yet) include interactive ice sheets, or used to evaluate models that do. We also make recommendations for how climatological and anomalous fluxes can be implemented in climate models that may have different approaches to interactions with the ice sheets. These forcings are available for CMIP7 simulations and should lead to more robust and coherent simulation of sea surface temperature, sea ice and regional sea level trends in the recent historical period and, as these data are extended, improve the credibility of projections.