Palaeodata is essential for out-of-sample testing of climate models of all kinds. Reconstructions of the Antarctic contribution to sea level in the warm mid-Pliocene (3.2Ma BP) have been used to evaluate ensembles of ice sheet model simulations in several studies. Some have suggested that rapid cliff failure (Marine Ice Cliff Instability: MICI) is required, or at least preferred, to simulate very large sea level contributions. In general, sea level reconstructions have not provided strong constraints, unless assuming a fairly restricted range. We present new perturbed parameter ensemble simulations of Antarctica in the mid-Pliocene using the BISICLES ice sheet model, and evaluate these with reconstructions of sea level contribution and grounding line retreat in the Wilkes Basin. Varying five model parameters relating to basal friction, basal melting, surface mass balance, and glacial isostatic adjustment, and using four climate model forcings from PlioMIP2, in a Latin Hypercube design (N=120), we obtain a much wider range of sea level contributions than previous studies: -15.9 to +28.3m SLE after 10,000 years, where the range is mostly driven by basal friction. Smaller ensemble tests (N=30) show that sea level contributions increase to more positive values if initialising simulations from a Pliocene ice sheet, rather than modern, or subtracting a full perturbed parameter ensemble of control simulations, rather than a single default, suggesting that both approaches are needed until better understood. Wilkes retreat is more effective at adiscriminating between ensemble members than sea level reconstruction, suggesting that spatial data are more promising for reducing uncertainties. The two are uncorrelated: all but two of the simulations that show Wilkes retreat contribute less than ~10m SLE. Calibration of the main ensemble with both metrics strongly reduces the range: -3.1m to 8.3m SLE, though these values may increase under Pliocene initialisation and full control ensemble.
King’s College London, United Kingdom