Constraining models of glacial isostatic adjustment in eastern North America

We constrain the GIA signal and its uncertainty for southeastern Canada and northeastern USA using 1013 relative sea-level (RSL) data points distributed over 38 sites from two regional RSL data compilations containing 544 sea-level index points, 232 marine limiting data points, and 237 terrestrial limiting data points. This state-of-the-art dataset was compared to output from 14,960 model runs based on a 1D (spherically-symmetric) Earth model and 34 different North American ice sheet reconstructions. By considering different data subsets (regional partitioning), we found that significantly improved fits can be obtained by separating the data into two larger subregions: the first one close to the central section of the Laurentide Ice Sheet (including Hudson Bay, Ungava Peninsula, and Labrador), and the second one along the Atlantic coast of Canada and the USA (encompassing Newfoundland, St. Lawrence Corridor, New Brunswick and Nova Scotia, and Maine and Massachusetts). For the first subregion, relatively low lower mantle viscosity (1–2 × 1021 Pas; from the considered range of 1–90 × 1021 Pas) and relatively high upper mantle viscosity (greater than 0.5 × 1021 Pas from the considered range of 0.05–5 × 1021 Pas) for a given ice loading history can provide good quality fits. On the other hand, quality fits for the second subregion require intermediate values for these parameters (specifically, upper mantle viscosity of 0.5–1 × 1021 Pas and lower mantle viscosity of 20–30 × 1021 Pas). When considering model uncertainty bounds, ∼91.6% of the observations (928 out of 1013 RSL data points) can be captured, while the remaining residuals (∼8.4%) reflect some observational errors (outliers) and systematic error in the model. Given the 1D analysis suggests the existence of significant lateral Earth structure, we explored this aspect via the use of a 3D finite volume Earth model. Considering two realizations of lateral Earth structure based on two shear-wave tomographic models, the 3D results support the spatial partitioning of the whole study region into two larger subregions. However, the 3D model was unable to improve upon the data-model fits obtained with the 1D model, indicating limitations in the models of lateral structure and/or the adopted ice models, which are reconstructed based on 1D Earth models. Finally, our 1D Earth modelling results indicate that, within the first subregion, ice history models with relatively thick ice during the early phase of deglaciation (prior to ∼13 ka) and thinner ice in the early Holocene are preferred by the RSL data.