High climate model dependency of Pliocene Antarctic ice-sheet predictions

The mid-Pliocene warm period provides a natural laboratory to investigate the long-term response of the Earth’s ice-sheets and sea level in a warmer-than-present-day world. Proxy data suggest that during the warm Pliocene, portions of the Antarctic ice-sheets, including West Antarctica could have been lost. Ice-sheet modelling forced by Pliocene climate model outputs is an essential way to improve our understanding of ice-sheets during the Pliocene. However, uncertainty exists regarding the degree to which results are model-dependent. Using climatological forcing from an international climate modelling intercomparison project, we demonstrate the high dependency of Antarctic ice-sheet volume predictions on the climate model-based forcing used. In addition, the collapse of the vulnerable marine basins of Antarctica is dependent on the ice-sheet model used. These results demonstrate that great caution is required in order to avoid making unsound statements about the nature of the Pliocene Antarctic ice-sheet based on model results that do not account for structural uncertainty in both the climate and ice sheet models.

The Root-mean square error (RMSE) in terms of predicted ice sheet thickness for EAIS and WAIS shows that MRI-CGCM2.3 is over predicting ice sheet thickness for East Antarctica, which is consistent with the high EAIS volumes predicted for this model ( Supplementary Fig.  9). Both ISMs show high errors in predicting the land ice for West Antarctica using the MIROC4m climate model, which is reflected in the collapse of the WAIS in the control simulations (see also Fig. 2). For MRI-CGCM2.3, SICOPOLIS predicts a reasonable ice sheet thickness, but the ANICE ISM predicts a WAIS and associated ice shelves that are too thick when compared to Bedmap2.
Using the performance criteria outlined in the Methods section of the main paper (Supplementary Tables 2 and S3) and the comparison of predicted modern volume, area and ice-sheet thickness, two GCM-ISM combinations consistently perform poorly against the metrics we have chosen (MRI-CGCM2.3 and MIROC4m). Therefore, these models have been excluded from the Pliocene Scenarios presented in this study.

Supplementary Note 2: Understanding the reasons that certain GCM-ISM combinations lead to a poor representation of the modern Antarctic Ice Sheet
In all of the ISMs, the MRI-CGCM2.3 climate forcing results in an unrealistically large AIS, which is extensively thicker than the Bedmap2 reconstruction, especially over areas of continental East Antarctica ( Fig. 2; Supplementary Fig. 7). Annual average Antarctic precipitation in the MRI-CGCM2.3 is almost double that of other models for the preindustrial control (0.8 mm day -1 ) and it is the coldest model with an annual average AIS temperature of -38.3°C (Fig. 1). Combined, these factors contribute to the most positive surface mass balance (SMB) predictions over both East and West Antarctica when forcing the ISMs (approximately 2000 Gt yr -1 ; Supplementary Fig. 4). When compared to observations of climate, it has been shown that MRI-CGCM2.3 exhibits a strong Austral summer cold bias over Antarctica (of over 8°C), which is attributed to cloud concentration reduction in the model (when compared with a previous version of the model -MRI-CGCM2.0) 2 . MRI-CGCM2.3 also has very different snow albedo values (α = up to 0.9) to the other PlioMIP models (α = 0.55-0.7), which may also contribute to the colder temperatures allowing for a larger modern-day AIS to be predicted. MRI-CGCM2.3 largely overestimates precipitation in comparison to observations when compared zonally between 70°S and 90°S (and this is strongest in the Austral summer months) 2 . We suggest therefore, that the regional model biases exhibited by MRI-CGCM2.3 are likely to be partly the cause of such large volumetric predictions for AIS when using all three ISMs.
Present-day AIS reconstructions using MIROC4m also do not give a reasonable representation of the modern ice-sheet. The simulations using ANICE and SICOPOLIS both exhibit a low volume associated with the collapse of the WAIS in both ISMs. In part, this may be driven by the fact that MIROC4m exhibits the second highest temperatures and lowest precipitation rates over West Antarctica of all of the PlioMIP models ( Supplementary  Fig. 1). Additionally, the removal of ice in this area is likely a response to warm sub-shelf temperatures (up to 1.6°C) in the Marie Byrd area of the Ross Ice Shelf, which are associated with a negative basal mass balance ( Supplementary Fig. 4). On average, the area of the Ross and Ronne Ice shelves is greater in the simulations using ANICE, than those using SICOPOLIS, with only the simulation using the MIROC4m forcing showing a loss of both ice-shelves for present day (Fig 2). This difference could be attributed to the different exposed shelf melt rates prescribed in the models for ANICE and SICOPOLIS, and the applied calving rate in SICOPOLIS (see Methods). In the SICOPOLIS model, this is also a result of a negative SMB on land, leading to an insufficient ice flux from the ice-sheet to sustain the ice-shelves.
The performance of MIROC4m over the Antarctic region has not been addressed in previous model description papers 3 , however, we note that the different treatment of the land-sea mask in MIROC4m relative to the other PlioMIP models (even for the control simulation) may be one reason that the sub-shelf temperatures are particularly high in this model. MIROC4m treats the ice-shelves as ocean grid points allowing sea-ice to grow over them (rather than as land in the other models).  Table 1). The bottom row shows Antarctic topography (m) from each climate model.  Figure 7 in the main text, however it includes the sea level contributions from the two GCM-ISM combinations that were not included in the main Pliocene analysis (MIROC4m and MRI-CGCM2.3). As such the middle is the 7th ranking sea level contribution from the list of 14 SIA-SSA model results. Supplementary Table 1. Details of the PlioMIP climate models used to force the ice-sheet models. This includes the resolution of the atmosphere components and ocean components of the models and the main reference for each model. The Land-Sea Mask (LSM) scheme implemented by each model is also detailed (see Haywood et al., 2010). Regarding the LSM, "preferred" refers to a LSM that has been entirely altered to meet the PlioMIP boundary conditions (e.g. the creation of a West Antarctic seaway within the models). "Alternate" is where modelling groups have had to use a more similar to modern LSM. More comprehensive details of each model, and their implementation of the LSM, can be found in Haywood et al. (2013)