Upcoming Event: PhD Dissertation Defense

Paleo-reconstruction of Greenland Ice Sheet Dynamics over the Holocene via Bayesian Inversion

Shreyas Gaikwad, CSEM Ph.D. Candidate

Abstract

The Greenland Ice Sheet (GrIS) currently contributes around 15% to the Global Mean Sea Level (GMSL) rise budget, and will be a dominant contributor to the GMSL rise budget over the coming century. The meltwater from Greenland also affects important aspects of climate, such as the Atlantic Meridional Overturning Circulation, as well as the economics and livelihoods of the local communities in Greenland. It is thus important to predict with increased confidence the future behavior of the GrIS.

Existing ice sheet models differ substantially in the forecasts they make. This is because these forecasts use non-calibrated parameters and highly uncertain initial conditions. These shortcomings can be partially overcome by model calibration and initialization that leverages recently collected data using state-of-the-art gradient-based optimization methods. We make use of the comprehensive remotely-sensed age-layer data set of the GrIS, along with the present-day surface velocity and ice sheet thickness data. These age layers encapsulate the past surface climates over the ice sheet as well as the dynamical evolution of the ice sheet.

The underlying inverse problem is naturally formulated as PDE-constrained optimization and solved within a Bayesian framework. We present newly developed open-source tangent linear and adjoint models of the ice sheet model SICOPOLIS, developed using the open-source Algorithmic Differentiation tool Tapenade. The special nature of our models and the observables in our datasets necessitates a multi-staged approach to inverse modeling. We use performance-optimized adjoint models for sensitivity analysis and to robustly calibrate sensitive model parameters characterizing the initial state, time-averaged basal and time-varying surface boundary conditions, as well as hyperparameters in the various constitutive models and ice rheology. These calibrated models will help provide better forecasts of the GMSL contributions from the GrIS in the coming century.

Biography

Shreyas Sunil Gaikwad is a Ph.D. candidate in the CSEM program at the Oden Institute, advised by Dr. Patrick Heimbach. His research focuses on PDE-constrained inverse problems and machine learning for ice sheet and ocean modeling, including the development of open-source adjoint models. He received his undergraduate degree in Mechanical Engineering from the Indian Institute of Technology (IIT) Bombay in 2019.