The Ice Model and the Climate Model Won't Merge

Projecting future sea level rise requires coupling climate models (which simulate atmosphere, ocean circulation, and heat transport) with ice sheet models (which simulate ice flow, calving, and mass loss from Greenland and Antarctica). These two model families have been developed independently for decades by separate scientific communities using different numerical methods, spatial grids, and timescales. Climate models typically run on 50-100 km grids with 30-minute timesteps; ice sheet models require 1-5 km resolution to capture outlet glaciers and use annual-to-decadal timesteps for ice flow dynamics. Coupling them requires passing ocean heat fluxes to ice sheet model boundaries and returning freshwater fluxes back to the climate model — an exchange that must occur at the interface between ocean and ice, in the most poorly observed and difficult-to-model region of either system: the grounding line where ice sheets meet the ocean. IPCC AR6 acknowledged that the lack of coupled ice-ocean models is the primary reason that sea level projections include "low confidence" caveats for the Antarctic contribution beyond 2100.

Sea level rise is among the most consequential impacts of climate change: 680 million people live in low-lying coastal zones, and ~$1 trillion in global coastal infrastructure is at risk per meter of rise. The uncertainty in sea level projections is dominated by the Antarctic ice sheet contribution — IPCC AR6 projected 0.28-1.01 m of global mean sea level rise by 2100 under a high-emissions scenario, but noted that "a rise approaching 2 m by 2100 cannot be ruled out." This factor-of-7 spread exists primarily because climate models cannot reliably predict how much warm ocean water will reach ice sheet grounding lines and how fast ice sheets will respond. Coastal adaptation planning — where to build sea walls, when to retreat from coastlines, how to design infrastructure — requires narrowing this uncertainty, which requires coupled models that don't yet exist in production form.

"Offline" coupling — running a climate model, extracting ocean temperature/salinity at ice sheet boundaries, and using those as boundary conditions for a standalone ice sheet model — is the current standard approach. This method ignores feedbacks: ice sheet meltwater freshens the ocean surface, affecting ocean circulation, which in turn affects heat delivery to the ice sheet. The few attempts at "online" coupling (e.g., in CESM2, E3SM, UKESM) have demonstrated the concept but face persistent problems: the ocean model's coastal resolution is too coarse to resolve the fjords and ice shelf cavities where ocean-ice interaction occurs; the ice sheet model's calving laws are empirical and poorly constrained; the coupling interface introduces numerical artifacts where the grids don't align; and coupled runs are so computationally expensive that ensembles (needed for uncertainty quantification) are infeasible. Marine ice cliff instability (MICI) — the hypothesis that tall ice cliffs could collapse under their own weight, triggering rapid retreat — could potentially add meters of sea level rise on century timescales, but the physics is debated and no coupled model has resolved it.

Variable-resolution ocean models that provide kilometer-scale resolution in ice shelf cavities and fjords while maintaining global coverage at coarser resolution — approaches like MPAS-Ocean (unstructured mesh) show promise but are computationally demanding. Physically-based calving laws derived from fracture mechanics and validated against observed calving events, replacing the empirical parameterizations in current ice sheet models. Machine learning emulators of ice sheet behavior that can be embedded in climate models, providing fast approximations of ice sheet response without running a full ice sheet model at every coupling timestep. Observational validation datasets from ice-ocean interfaces — sub-ice-shelf ocean measurements (from autonomous underwater vehicles, borehole access, or instrumented seals) that constrain the heat fluxes driving ice sheet retreat.

A student team could implement and benchmark a simple ice-ocean coupling framework using a 1D melt parameterization (e.g., quadratic melt-rate dependency on ocean thermal forcing) within an open-source ice sheet model (ISSM, Elmer/Ice, or PISM), testing how different coupling frequencies and melt parameterizations affect modeled retreat rates for a simplified glacier geometry. Alternatively, a team could train a neural network emulator of an ice sheet model's response to varying ocean forcing scenarios, assessing whether ML surrogates can reproduce multi-century ice dynamics at reduced computational cost. Relevant disciplines: computational fluid dynamics, glaciology, physical oceanography, machine learning, numerical methods.