In this study, we compute glacial-interglacial sea-level changes across the mid-Pliocene Warm Period (MPWP; 3.264 – 3.025 Ma) produced from ice mass loss of different ice sheets. Our results quantify the relationship between local and global mean sea-level (GMSL) change and highlight the level of consistency in this mapping across different ice melt scenarios. These predictions can help to guide site selection in any effort to constrain the sources and magnitude of MPWP GMSL change.

Sea-level information from the global ocean is sparse in time and space with comprehensive data being limited to the period since 2005. Here we provide a novel reconstruction of sea level and its contributing causes as determined by a Kalman Smoother approach applied to tide gauge records over the period 1900 to 2021. The new reconstruction shows a continuing acceleration in global mean sea level rise since 1970 dominated by melting land ice. Regionally contributors vary significantly by region.

Ice sheets present a thickness of a few kilometres, leading to a vertical deformation of the crust of up to a kilometre. This process depends on properties of the solid Earth, which can be regionally very different. We propose a model that accounts for this often ignored heterogeneity and runs 100,000 simulation years in only minutes. Thus, the evolution of ice sheets is modelled with better accuracy, which is critical for a good mitigation of climate change and, in particular, sea-level rise.

We use advanced statistical techniques and a simple ice-sheet model to produce an ensemble of plausible 3D shapes of the ice sheet that once stretched across northern Europe during the previous glacial maximum (140,000 years ago). This new reconstruction, equivalent in volume to 48 ± 8 m of global mean sea-level rise, will improve the interpretation of high sea levels recorded from the Last Interglacial period (120 000 years ago) that provide a useful perspective on the future.

A relatively recent advance in glacial isostatic adjustment modelling has been the development of models that include 3D Earth structure, as opposed to 1D structure. However, a major limitation is the computational expense. We have developed a method using artificial neural networks to emulate the influence of 3D Earth models to affordably constrain the viscosity parameter space. Our results indicate that the misfits are of a scale such that useful predictions of relative sea level can be made.