Ice sheets are an active component of Earth's climate system. Their topography influences atmospheric circulation and changes in their volume alters freshwater fluxes to the oceans, affecting ocean water masses, atmospheric carbon uptake, and global sea level. Sea-level rise has a marked societal impact, and thus ice sheet models are indispensable tools to predict it. To increase confidence on sea-level rise projections, it is necessary that ice sheet models accurately represent the relevant processes governing ice sheet dynamics. Given the fact that ice sheets respond to geological-scale changes in Earth's system, it is necessary that their performance is compared with in-situ data of past geological periods, which are discrete in space and time. One useful constraint used for validating model results is past ice surface elevation, which is reconstructed based on rock samples taken from nunataks (mountain summits that pierce through the ice sheet surface). However, two main problems prevent reliable comparisons of past ice surface elevations between model and empirical results. First, data-model comparisons are hindered by the fact that most large-scale ice sheet models capture neither the timing nor the magnitude of ice thinning reconstructed for the last deglaciation. Second, the complex subglacial topography of regions where nunataks are present is also reflected on the ice sheet surface, through pronounced elevation gradients. As a result, the choice of a reference point on the present-day ice sheet, which can be subjective, is a significant source of uncertainty when computing thickness-change estimates.               

In this thesis, I aim to reconstruct changes in ice sheet geometry over Dronning Maud Land (DML, East Antarctica) during periods that were warmer and colder than present, and the climate drivers behind such changes. I assess whether the comparison between empirical and model results can be improved by resolving local features in ice sheet models, and by using data and models in an iterative way (using data to constrain the model, and models to interpret the data). The results of this thesis demonstrate that ice flow in areas of complex topography is poorly resolved in continental-scale ice sheet models and requires modelling in high resolution to match results from empirical constraints. High-resolution ice-sheet models, in turn, show that accurate ice sheet surface elevation reconstructions from empirical data require systematic sampling and definition of reference points over the modern ice sheet surface. Moreover, a consistent reconstruction of regional ice-thickness changes needs both empirical and ice sheet model results. Based on constrained models and empirical datasets, the ice sheet in DML responds to an interplay between sea level, ocean warming, surface mass balance, and subglacial topography. Samples from nunataks mainly reflect local ice surface elevation changes, potentially missing catchment-scale (regional) changes. Accurately determining regional changes using high-resolution modelling plays a significant role when interpreting the evolution of ice streams. Hence, the work presented here highlights that accurately reconstructing past ice sheet geometry is an effort that can only be truly successful if field scientists and ice sheet modellers work in tandem, at experiment-design, sampling, and result-interpretation stages.