Competing climate feedbacks of ice sheet freshwater discharge in a warming world

The study addresses how freshwater flux from melting ice sheets feeds back on climate and, in turn, modulates the future retreat of the Antarctic and Greenland ice sheets under anthropogenic warming. Freshwater input stratifies the upper ocean, potentially cooling the atmosphere and sea surface while warming subsurface waters. These processes create competing feedbacks: surface cooling tends to reduce surface melt and hydrofracturing (negative feedback), whereas subsurface warming enhances ice-shelf basal melt and buttressing loss (positive feedback). The relative strength and net effect of these feedbacks remain uncertain, particularly given varying future warming pathways and climate sensitivity. Because past deglacial meltwater events and geometries are not direct analogs for modern configurations, the authors develop a coupled ice sheet–climate modeling framework to quantify these feedbacks and determine under which conditions one dominates the other, and how large-scale circulation changes (notably AMOC) mediate interhemispheric ice-sheet interactions.

Prior hosing experiments show that imposing meltwater freshening in the North Atlantic or Southern Ocean stratifies the upper ocean, suppresses deep water formation, and alters the AMOC, with global climate repercussions. Geological evidence from the Last Deglaciation (e.g., Meltwater Pulse 1A, Younger Dryas, Antarctic Cold Reversal) highlights the climate impacts of large meltwater inputs. Modeling studies consistently find that ice-sheet-derived freshwater freshens the surface, enhances stratification, reduces vertical mixing, cools the atmosphere and surface ocean, and warms the subsurface. However, offline or asynchronously coupled studies have yielded contradictory net feedbacks for the AIS: some find positive feedbacks accelerating loss via subsurface warming, while others find negative feedbacks delaying loss due to surface cooling. Earth system models with fully interactive ice sheets (e.g., UKESM, E3SM) are under development but have not yet been widely applied to centennial-scale feedback assessments. This motivates a near-synchronous coupled approach to resolve time-evolving two-way interactions and reconcile prior discrepancies by exploring dependence on warming scenarios and model sensitivity.

The authors couple a continental-scale 3-D dynamic–thermodynamic ice sheet model (PSUICE3D) quasi-synchronously to an Earth system model of intermediate complexity (UVic-ESCM v2.8) using a 1-year coupling interval. PSUICE3D uses hybrid SIA/SSA dynamics with grounding-line mass flux, ELRA bedrock adjustment, parameterized hydrofracturing and ice-cliff failure (MICI), and a positive-degree-day surface melt scheme (AIS TPDD=0 °C; GIS TPDD tuned to −4.0 °C). AIS grid: 20 km; GIS grid: 10 km. Antarctic ice-shelf basal melt rates depend quadratically on 400 m ocean temperature above the depth-dependent freezing point with a tuned OMF to match modern observational basal melt totals. Basal sliding (AIS) is tuned by inverse methods to match present thickness under quasi-preindustrial climate. UVic-ESCM components include: a 2-D energy–moisture balance atmosphere, 3-D ocean (MOM2.2) at 3.6°×1.8° with 19 vertical levels, dynamic–thermodynamic sea ice, and a land model (biogeochemistry off). Winds are prescribed (with empirical dynamical feedback). A bias-correction approach applies model anomalies to an observational baseline (CERA20C for atmosphere and WOA2018 for ocean). Freshwater discharge from ice sheets is imposed as a surface salinity flux into top-layer ocean cells; solid discharge (icebergs) is represented as equivalent sea-ice growth at discharge cells (no explicit iceberg drift/decay). The ocean uses a rigid lid with constant volume; sea level is fixed, and changes in land–sea mask, gravitational fingerprints, and latent heat of surface/basal melt are not passed to the climate model. Scenarios: Historical (1850–2014) plus six SSPs (2015–2500). To emulate a range of climate sensitivities, atmospheric CO2 pathways are scaled to target ECS of 3.0, 4.0, and 5.6 °C using a logarithmic scaling consistent with OLR–log(CO2) relationships. The model’s native ECS is 3.4 °C, with weaker polar amplification than typical CMIP6 models; ECS=5.6 °C with SSP5-8.5 is used as a worst-case proxy for polar warming. Four coupling configurations assess freshwater feedbacks: (1) constant preindustrial ice-sheet FWF (FWC), (2) interactive GIS FWF only, (3) interactive AIS FWF only, (4) interactive FWF from both ice sheets (FWAG). Ensemble design: 10-member ensembles for moderate warming (SSP2-4.5 with ECS=4.0 °C) and intensive warming (SSP5-8.5 with ECS=5.6 °C), initialized from different phases of a long PI control (50-year offsets) to reduce internal variability. A feedback factor γ is defined as 1 minus the ratio of AIS mass-loss rate with fixed FWF to that with interactive FWF, matched by GMSAT anomaly bins, to quantify sign and magnitude of feedbacks as a function of ice-loss rate. Model evaluation indicates reasonable present-day ice-sheet states and recent trends, with biases noted for AIS (e.g., overly extensive shelves, too little surface melt).

- Freshwater flux magnitude and timing depend strongly on warming scenario and climate sensitivity. In the warmest scenario (SSP5-8.5, ECS=5.6 °C), peak GIS FWF reaches ~0.37 Sv and AIS FWF ~−1.1 Sv; GIS is nearly lost by 2500; WAIS collapse initiates ~2100, peaks ~2300, and is nearly concluded by ~2400, contributing >4 m SLR over three centuries; total AIS SLR contribution exceeds 10 m by 2500. - AMOC weakens in all scenarios until mid-21st century; in less extreme warming it recovers and overshoots, while in the warmest case it continues to decline. Interactive Greenland FWF produces stronger and more prolonged AMOC weakening; in the intensive scenario it suppresses AMOC recovery entirely. - Global-mean surface air temperature reductions due to interactive FWF are scenario- and ice-sheet dependent: Greenland FWF yields modest global cooling (~−0.1 °C in the warmest scenario), concentrated south of Greenland; Antarctic FWF produces larger transient global cooling (~−0.5 °C) around WAIS collapse. - Greenland Ice Sheet response: With interactive FWF, GIS generally retreats more slowly (overall negative feedback) in most scenarios, linked to AMOC weakening and North Atlantic cooling; exceptions in largest warmings likely reflect Antarctic-driven teleconnections. - Antarctic Ice Sheet response: Competing feedbacks operate. Interactive AIS FWF accelerates ice loss (net positive feedback) under moderate warming, doubling AIS-contributed SLR by 2500 relative to fixed-FWF runs. Under intensive warming, interactive AIS FWF accelerates early retreat before ~2100, but thereafter the negative feedback dominates, substantially delaying and slowing WAIS collapse; during mid-2200s runaway retreat, interactive AIS FWF reduces WAIS SLR contribution by ~30% compared to fixed-FWF runs, and after ~2400 it continues to reduce East Antarctic loss rates. - Physical mechanism: Elevated AIS FWF increases upper-ocean stratification, causing surface cooling (including latent heat uptake from iceberg melt proxy and enhanced sea-ice cover via ice–albedo feedback) and subsurface warming that intensifies basal melt. As shelves shrink under strong warming, the positive (basal melt) feedback weakens due to reduced shelf area, while negative surface cooling persists, driving a net sign change. - Feedback factor γ as a function of AIS ice-loss rate (Sv): γ is positive at low loss rates (positive feedback) and transitions to negative around ~0.2 Sv, with peak magnitudes exceeding 0.5 in both regimes. Negative feedback strength declines for loss rates beyond ~−0.4 Sv (WAIS collapse stage), reflecting dominance of internal ice-dynamic instabilities and increased irregularity. - Interhemispheric coupling via AMOC (bipolar seesaw): Greenland FWF freshens the NADW formation region, weakens AMOC, cools the North Atlantic, and warms the Southern Hemisphere, increasing AIS loss; Antarctic FWF freshens AAIW, strengthens AMOC, warms the North Atlantic, and increases GIS loss. In ensembles, interactive Greenland FWF increases Antarctic SLR contribution by ~0.5 m by 2500; Antarctic FWF increases GIS loss by ~0.1 m through the 2300s, despite fixed global sea level in the model setup. - Overall, AIS freshwater–climate feedbacks are strongly scenario dependent: positive feedback dominates under moderate warming and at early stages, while negative feedback prevails under intensive warming as WAIS collapse proceeds.

The study resolves the question of which freshwater–climate feedback dominates AIS retreat by demonstrating that the net sign depends on the pace and magnitude of warming and the stage of ice-shelf evolution. Early in warming or under moderate scenarios, subsurface-driven basal melt and reduced buttressing dominate, accelerating AIS mass loss (positive feedback). Under intense warming as shelves thin and disappear, surface cooling and enhanced sea ice lead to an emergent negative feedback that slows WAIS collapse. These results reconcile previously conflicting studies by highlighting differences in coupling strategies, time horizons, climate sensitivities, and representation of MICI processes that can shift the balance. The AMOC mediates strong interhemispheric teleconnections: freshwater forcing in one hemisphere exacerbates retreat in the other via the bipolar seesaw. Findings underscore that freshwater–climate interactions materially affect projections of ice-sheet-driven sea-level rise and must be included in coupled modeling frameworks to credibly assess future trajectories.

A near-synchronously coupled ice sheet–climate framework reveals that freshwater discharge from ice sheets yields competing feedbacks whose net effect on AIS retreat is strongly scenario dependent: a net positive feedback under moderate warming and in early retreat phases, but a dominant negative feedback under extreme warming during WAIS collapse. The AMOC acts as an interhemispheric bridge, so freshwater forcing from one ice sheet tends to accelerate mass loss from the other. These mechanisms imply faster AIS retreat and higher sea-level rise unless or until a WAIS collapse is triggered, after which negative feedbacks partly mitigate further loss. Future work should deploy higher-resolution, fully coupled Earth system models with explicit iceberg dynamics and drift, realistic depth of freshwater injection, resolved ice-shelf cavity circulation, improved Antarctic Slope Current representation, and dynamic land–sea and sea-level feedbacks, to refine feedback strengths and tipping thresholds.

- Reduced-complexity climate model (UVic-ESCM) with coarse spatial resolution, simplified atmosphere, prescribed winds, and weaker polar amplification than CMIP6 means results may be model dependent and conservative for polar warming. - No explicit iceberg drift/decay; solid flux treated as local sea-ice growth, likely shifting freshwater input poleward and enhancing local albedo-driven cooling biases. - Freshwater injected at the ocean surface top layer; real basal melt and iceberg melt can inject at depth, potentially altering stratification and subsurface warming magnitude. - No explicit ice-shelf cavity circulation; basal melt parameterized from 400 m temperature may miss key processes (e.g., CDW intrusions, cavity overturning) affecting shelf melt and sea-ice distribution. - Fixed ocean volume (rigid lid), fixed sea level and land–sea mask; no dynamic feedbacks from sea-level change, GIA coupling to the climate model, or gravitational fingerprints. - Latent heat of surface and basal ice melting is not passed to the climate model (only iceberg-as-sea-ice latent heat), potentially underestimating local energy budget effects. - Biases in present-day AIS (overly extensive shelves, too little surface melt) and uncertainties in parameterizations (e.g., MICI, PDD scheme without explicit radiation) affect quantitative projections. - Radiative forcings from non-CO2 greenhouse gases are not included, and biogeochemistry is disabled, possibly underestimating total warming relative to CMIP6 for given ECS.