See attachment.

This manuscript introduces a 3D dynamics model for mountain glaciers based on the Community Ice Sheet Model, and its application to study the evolution of mountain glaciers using protocols from the third phase of the Glacier Model Intercomparison Project (GlacierMIP3). While the paper is overall well written, I have several minor concerns that should be addressed before the manuscript is considered suitable for publication.

I would like the authors to better motivate, in the Introduction, the need to use high-fidelity models for mountain glaciers, particularly considering their increased sensitivity to data errors. The authors touch on this in the Discussion, but this should be featured in the Introduction as well.

The initialization of the model is done with ad hoc tuning methods (e.g., Pollard and DeConto, 2012). However, I think the reader would benefit from a discussion of more advanced initialization techniques based on (transient) PDE-constrained optimization methods, which have become increasingly standard in the literature and offer a more rigorous approach to model initialization.

Detailed comments:

Equation (1): For clarity and completeness, please include the y-component of the DIVA model alongside the x-component.

Equation (2): Please clarify whether the cap applies to the absolute values of the separate x and y derivatives of the surface elevation, or to the magnitude of the surface elevation gradient. Either way, this seems a bit of a crude fix. How frequently is this correction active in simulations performed in this study? 

Equation (3): The scalar representation of the sliding law is potentially confusing, as basal shear stress and basal velocity are vector fields.

        Please rewrite this equation in vector form, for example: tau_b = C_p |u_b|^{1/m-1} u_b.

Section 5.1: Data Blocks and Repartitioning: The term "data block" needs to be defined. Are data blocks the fundamental units used for distributing computational workload across processor cores? How is the repartitioning of the data blocks performed, especially considering they are now unstructured?