The influence of emissions scenarios on future Antarctic ice loss is unlikely to emerge this century

Human-induced global warming has reached about 1.0 °C above preindustrial levels and is increasing at roughly 0.2 °C per decade. Projected global mean sea-level rise by 2100 ranges from about 0.25–0.59 m under low emissions (RCP2.6) to 0.61–1.10 m under high emissions (RCP8.5). The Antarctic Ice Sheet (AIS) is the largest source of uncertainty in these projections, as highlighted by ISMIP6, which reported a wide spread of AIS contributions under RCP8.5 (−0.08 to +0.30 m by 2100). Uncertainties arise from climate forcing differences, model initialization, and especially poorly constrained ice-physics parameters, including sub-ice-shelf basal melt, surface mass balance, basal friction, and ice rheology. Consequently, AIS projections show substantial overlap across scenarios in the 21st century, and the time at which emissions scenario dependence will clearly emerge is unknown. This study asks: when will emissions scenario dependence in AIS contributions to sea level emerge, and how do ice-flow parameter uncertainties affect this timing? The authors use historically constrained process-based simulations and a Gaussian Process emulator to quantify time of emergence across scenarios and relate AIS sea-level contributions to global warming levels through 2300.

Prior work (e.g., IPCC reports and ISMIP6) indicates a large spread in AIS contributions due to uncertainties in basal melting parameterizations, surface mass balance, and ice dynamics. ISMIP6 showed broad ranges even under a single scenario, reflecting model and forcing diversity. Studies emphasize the sensitivity of regions like the Amundsen Sea Embayment (ASE) and Wilkes Land, where retrograde beds predispose to marine ice sheet instability. Some projections show divergence in loss rates between scenarios by around 2060–2080, particularly linked to Thwaites Glacier response, and others have explored higher-end outcomes when including processes like marine ice-cliff instability. However, the precise timescale for clear emissions-scenario emergence and its dependence on poorly constrained sliding and rheology parameters remained unresolved, motivating the present analysis.

Process-based ice sheet modeling: The authors use PISM v1.1, a hybrid SIA/SSA thermo-mechanical model with freely migrating grounding lines and a sub-grid grounding-line parameterization. Initialization follows the VUW ISMIP6 approach: iterative nudging and multi-stage spin-up preserving thermal evolution while matching present-day geometry, velocities, and grounding-line positions. A 65-year historical run (1950–2015) precedes projections. Basal melt rates are computed with a thermodynamic ice–ocean boundary layer model (after Holland & Jenkins) tuned to reproduce contemporary shelf melt patterns and totals (~1443 Gt/yr), acknowledging uncertainty in the melt factor’s temporal stationarity. Forcings and scenarios: Primary ensembles use CMIP5 NorESM1-M anomalies for RCP2.6 and RCP8.5, extended to 2300 with anomalies held constant after 2100 (minimum commitment assumption). To sample Southern Ocean uncertainty, additional forcings include CCSM4 RCP8.5 (cooler ocean, RCP8.5-CO), HadGEM2-ES RCP8.5 (warmer ocean, RCP8.5-WO), and CMIP6 CNRM-CM6-1 SSP126 and SSP585. A constant-climate control is also run. Parameter ensembles: A full-factorial 3×3×3×3 design (81 runs) varies four poorly constrained ice-flow parameters within literature ranges: SIA enhancement (E_SIA = 1.2, 2.4, 4.8), SSA enhancement (E_SSA = 0.4, 0.6, 0.8), sliding law exponent (q = 0.25, 0.50, 0.75), and minimum till friction angle (phi_min = 5°, 10°, 15°). Grid resolution is 16 km and simulations run 2015–2300. Statistical emulator: A Gaussian Process regression model is trained on the process-based simulations (45,846 time-labeled data points across parameter sets and two RCPs, using 5% for training). Inputs (features) include the four ice-physics parameters and three climate descriptors: global mean temperature (GMT), cumulative GMT, and time since last GMT change (to capture committed responses). The emulator reproduces simulation behavior with low error and enables efficient exploration of intermediate scenarios (RCP4.5, RCP6.0) and idealized global warming levels (1–5 °C). Historical constraints: Emulator hindcasts of AIS mass change are compared against observational estimates (IMBIE 2018; Rignot et al. 2019; Frederikse et al. 2020). Parameter combinations whose hindcasts over- or under-shoot historical AIS contributions (bounds set at ~3 SD) are discarded. Using the most conservative observational constraint (IMBIE), 7 of 81 parameter sets remain for historically constrained projections.

- Time of emergence: Despite median divergence after ~2050, substantial overlap persists through the 21st century due to ice-flow parameter uncertainties. Defining emergence as when the RCP8.5–RCP2.6 median difference exceeds the 95% CI of RCP2.6, the time of emergence occurs in 2189 (very likely), with a likely emergence by 2116 for the 68% CI. - Historically constrained AIS contributions relative to 2000: by 2100, 0.12–0.44 m (RCP2.6) and 0.21–0.56 m (RCP8.5) at 95% CI; by 2300, 0.45–1.57 m (RCP2.6) and 1.96–3.79 m (RCP8.5). - Scenario divergence is controlled primarily by ocean thermal forcing driving sub-ice-shelf basal melt. Surface mass balance increases under high emissions partly compensate, prolonging overlap (especially evident with CMIP6 SSP585 vs SSP126). - Spatial patterns: Largest scenario contrasts appear at the Ross and Ronne–Filchner ice shelves. Under high emissions (especially warm-ocean cases), substantial thinning and partial collapse occur on decadal to centennial scales, fostering widespread WAIS retreat. ASE and Wilkes Land dominate parameter sensitivity due to retrograde beds and MISI susceptibility. - Rates of loss: For fixed parameters, loss rates in RCP2.6 and RCP8.5 begin to diverge around 2060–2080, first in the ASE (Thwaites/Pine Island), with RCP8.5 accelerating thereafter. Parameter combinations with higher dynamical sensitivity show stronger acceleration. - Global warming levels (GWL): At 2100, AIS contributions overlap across GWLs from 1–5 °C. By 2300, median AIS contribution under 2 °C is entirely below that under 4 °C but still overlaps with 3 °C. Exceeding 2 °C by 1 °C increases the 2300 median AIS contribution by ~0.69 m (~50% increase). The long-term commitment depends on total 21st-century warming, not on the rate of warming: reaching 2 °C by 2020 vs by 2100 yields only ~2 cm difference in 2300 median AIS contribution. - Sensitivity to basal melt: For a given forcing, sea-level equivalent sensitivity increases with cumulative basal melt; at low melt totals, uncertainty from ice-flow parameters alone can exceed 1 m. Warm-ocean forcings (e.g., HadGEM2-ES) lead to earlier shelf collapse and larger commitments, while cool-ocean forcings delay emergence and reduce commitments. - Earliest warning signs: Widespread thinning and partial collapse of Ross and Ronne–Filchner shelves, accruing from decades of ocean warming, likely precede and signal multi-meter AIS contributions under high emissions.

The study demonstrates that emissions-scenario dependence in AIS sea-level contribution is unlikely to emerge unambiguously within the 21st century because uncertainties in ice rheology and basal sliding obscure differences among scenarios. Nonetheless, the physical mechanism of divergence is clear: ocean-driven basal melting of major ice shelves (Ross, Ronne–Filchner) and sustained ASE retreat under higher warming produce a markedly larger long-term commitment. The findings reconcile observed early divergence in loss rates (circa 2060–2080) with continued overlap in cumulative contributions through 2100 by highlighting the dominant role of dynamical parameter uncertainty on decadal–centennial timescales. Importantly, the total warming realized during the 21st century sets the magnitude of committed AIS contributions—even if temperatures plateau post-2100—implying strong benefits from limiting warming to 2 °C or less. The results exceed some ISMIP6 ranges in 2100 largely because the committed response is included and because historical constraints remove implausible parameter sets; this aligns with observations that recent ice losses have outpaced many model projections. Given Southern Ocean heat uptake dominates oceanic warming and is not readily reversible, observed large-scale shelf thinning/collapse would indicate a higher-end AIS commitment may already be locked in. Overall, the work emphasizes monitoring and constraining oceanic forcing and key ice-dynamical parameters to reduce projection uncertainty and clarify scenario emergence timelines.

Main contributions: (1) Even with significant climate differences between scenarios, a clear high-emissions fingerprint in AIS sea-level contribution is unlikely to emerge before the late 22nd century due to ice-flow parameter uncertainties; (2) Long-term AIS sea-level commitment depends primarily on the total 21st-century warming, not the rate at which it is reached; (3) Under higher warming, thinning and partial collapse of the Ross and Ronne–Filchner ice shelves and accelerated ASE retreat drive multi-meter commitments by 2300. Implications: Limiting global warming to well below 2 °C substantially lowers AIS long-term contributions; exceeding 2 °C by even 1 °C raises median 2300 AIS contribution by roughly 0.69 m. Early warning indicators include widespread thinning and destabilization of major ice shelves. Future research: Priorities include better constraints on basal conditions and ice rheology (especially in MISI-prone ASE and Wilkes Land), improved sub-ice-shelf melt parameterizations and Southern Ocean projections (including meltwater feedbacks), higher-resolution ice dynamics, and extended coupled climate projections beyond 2100 to refine sea-level commitments and times of emergence.

- Emulator training is based primarily on a single AOGCM forcing (NorESM1-M) for RCP2.6/8.5; different Southern Ocean warming patterns (cool- or warm-ocean scenarios) shift emergence timing and commitments. - Post-2100 climate is held constant in projections, yielding minimum long-term commitments; continued warming would increase commitments and accelerate divergence. - Model resolution is 16 km; grounding-line dynamics and localized processes may be under-resolved. - Basal melt is tuned via a melt factor assumed temporally invariant; sub-grid basal melt at partially floating cells is not included, which can affect grounding-line retreat rates. - Sliding law and till parameterizations (e.g., phi_min, b_min, hydrology) are uncertain; historical constraints narrow parameters to 7 of 81 sets, but residual uncertainty remains large. - Meltwater feedbacks and some ocean-ice interactions are not represented in the driving climate models; Southern Ocean observational coverage is limited. - Inclusion/exclusion of processes like marine ice-cliff instability can substantially alter high-end outcomes, but was not employed here.