Using Ice Cores and Gaussian Process Emulation to Recover Changes in the Greenland Ice Sheet During the Last Interglacial

Abstract

The shape and extent of the Greenland Ice Sheet (GIS) during the Last Interglacial (LIG) is a matter of controversy, with different studies proposing a wide range of reconstructions. Here, for the first time, we combine stable water isotopic information from ice cores with isotope-enabled climate model outputs to investigate the problem. Exploring the space of possible ice sheet geometries by simulation is prohibitively expensive. We address this problem by using a Gaussian process emulator as a statistical surrogate of the full climate model. The emulator is calibrated using the results of a small number of carefully chosen simulations and then permits fast, probabilistic predictions of the simulator outputs at untried inputs. The inputs are GIS morphologies, parameterized through a dimension-reduction technique adapted to the spherical geometry of the setting. Based on the emulator predictions, the characteristics of morphologies compatible with the available ice core measurements are explored, leading to a reduction in uncertainty on the LIG GIS morphology. Moreover, a scenario-based approach allows to assess the gains in uncertainty reduction which would result from the availability of better dated LIG measurements in Greenland ice cores.

Sea ice feedbacks influence the isotopic signature of Greenland ice sheet elevation changes: last interglacial HadCM3 simulations

“…While the LIG global average warming is in line with projections for the end of the century (Clark and Huybers, 2009;Hoffman et al, 2017), there is considerable uncertainty on the timing of the sea level high stand during the LIG (Kopp et al, 2009;Düsterhus et al, 2016;Barlow et al, 2018) and the magnitude considerably surpasses near-future (2100-2200) projections (e.g. Fischer et al, 2018;Irvali et al, 2020).…”

Ice Cores and Emulation: Learning More About Past Ice Sheet Shapes

“…The work in Domingo et al (2020) gives even more motivation for our work. Using a similar approach to us, they used PCA to reduce a set of reconstructions to components describing the Greenland Ice Sheet during the Last Interglacial.…”

“…A detailed description of GP emulators is given in Chapter 5. By focusing on the LGM, a more recent time period, we build on the work of Domingo et al (2020) by demonstrating this data-model comparison method when more ice core records are available. The breadth of time that the LGM took place in means the dating errors in these ice cores are relatively low and hopefully will result in a reduced uncertainty around the AIS shape.…”

“…Extensive work has already been done reconstructing past ice sheet shapes or other climate variables with the use of climate models (Holloway et al, 2016). The methods of both Werner et al (2018) and Domingo et al (2020) are similar to our own, using previous reconstructions, GCMs and proxy data to reduce uncertainty around ice sheets at two different time periods. These papers highlight the importance of better understanding past changes in global ice sheets, and demonstrate the role that statisticians can play in this research area; we believe we can build on these methods by incorporating expert judgement, and taking a more detailed approach to building a prior model.…”

“…Before we performed an exploratory data analysis, we transformed our HadCM3 output into anomalies, the difference between our output and a pre-industrial control simulation. This was to remove any underlying climate biases HadCM3 may have (Domingo et al, 2020). We used a pre-industrial simulation to create δ 18 O anomalies from our HadCM3 output.…”