The Antarctic Ice Sheet (AIS) mass loss is a significant source of uncertainty in future sea-level rise projections due to various factors like ocean and atmospheric warming, ice dynamics, and boundary conditions. We explore the use of perturbed-parameter ensemble and variable-resolution ice sheet modeling for the AIS catchment through the Amery Ice Shelf Basin.

Ice sheet models provide projections of sea-level rise subject to many uncertainties and are computationally expensive, making uncertainty quantification challenging. Statistical emulation offers an alternative approach by creating surrogate models allowing faster exploration of the parameter space and capturing essential features of ice sheet models. In order to quantify uncertainties in future sea-level rise projections, we build Gaussian process (GP) emulators of a variable-resolution (4-20km mesh grid) AIS model, MPAS-Albany Land Ice (MALI) applied to Amery Ice Shelf Basin. We consider six input parameters controlling basal friction, bed rheology, ice stiffness, calving, and ice-shelf basal melting to produce the simulated data. First, we generate 200 perturbed input parameter initializations using space-filling Sobol sampling and pass them through the MALI model with 50 years of historical spinup to collect the simulation ensemble. The GP emulators are then trained on this simulation ensemble and calibrated using observations of the total ice area, grounded ice area, and grounded ice volume available in the year 2015 by assigning expert priors to the input parameters. Next, we filter the runs using the plausible ranges of the previous 3 quantities and branch out the 97 filtered runs up to the year 2175. Preliminary results indicate a likely range of 14 mm sea-level fall to 97 mm sea-level rise (mean = 38mm sea-level rise and SD = 22mm) can be expected from the Amery Basin through 2175.

Currently, we are building a PCA emulator for the sea-level rise time series data. Once trained, it will take samples from posterior distributions of the input parameters from the historical calibration and pass them through the PCA emulator. As a proof of concept, we will propagate the historically-calibrated input parameter uncertainties through the MALI model projections to produce probabilistic sea-level rise estimates through 2175.