Exploiting Lagrangian particle tracking in the Instructed Glacier Model (IGM) to model coupled debris-covered glacier dynamics

Debris cover on glaciers is expanding worldwide as glaciers are retreating. While the impact of debris cover on local mass balance is relatively well-established, the long-term dynamics of these glaciers are more complex and not fully understood. The dynamics are a key factor when trying to establish relationships between erosion rates, debris fluxes, and debris cover thicknesses while all of these properties are either completely unknown or only known locally in space and time.

Numerical modelling can help us better understand the data scarce debris-covered glacier system. While most recent approaches model englacial debris as an advected concentration, we establish a novel approach that exploits Lagrangian particle tracking in the Instructed Glacier Model (IGM). IGM uses deep learning to solve ice flow equations, greatly reducing computation times and enabling long-term model runs with large amounts of particles. In our implementation, a single particle represents a unit volume of debris and can be assigned any other property. The user can define a particle seeding area through either manual mapping or automatic classification based on conditions. Once particles emerge at the glacier surface in the ablation area, they are evaluated to compute debris cover thickness, which is then tied back to surface mass balance through a user-defined function.

As examples to showcase the capabilities of the model we use Zmuttgletscher, Switzerland, and Satopanth Glacier, India. We explore the sensitivities of the model to the use of different seeding strategies, changes in debris input amounts, mass balance functions, and model parameters such as grid size and seeding frequency.