Preprints
https://doi.org/10.5194/gmd-2021-281
https://doi.org/10.5194/gmd-2021-281

Submitted as: model description paper 25 Aug 2021

Submitted as: model description paper | 25 Aug 2021

Review status: this preprint is currently under review for the journal GMD.

WIFF1.0: A hybrid machine-learning-based parameterization of Wave-Induced sea-ice Floe Fracture

Christopher Horvat1 and Lettie A. Roach2 Christopher Horvat and Lettie A. Roach
  • 1Institute at Brown for Environment and Society, Brown University, Providence, RI, USA
  • 2Department of Atmospheric Sciences, University of Washington, Seattle, WA, USA

Abstract. Ocean surface waves play an important role in maintaining the marginal ice zone, a heterogenous region occupied by sea ice floes with variable horizontal sizes. The location, width, and evolution of the marginal ice zone is determined by the mutual interaction of ocean waves and floes, as waves propagate into the ice, bend it, and fracture it. In previous work, we developed a one-dimensional “superparameterized” scheme to simulate the interaction between the stochastic ocean surface wave field and sea ice. As this method is computationally expensive and not bitwise reproducible, here we use a pair of neural networks to accelerate this parameterization, delivering an adaptable, computationally-inexpensive, reproducible approach for simulating stochastic wave-ice interactions. Implemented in the sea ice model CICE, this accelerated code reproduces global statistics resulting from the full wave fracture code without increasing computational overheads. The combined model, Wave-Induced Floe Fracture (WIFF v1.0) is publicly available and may be incorporated into climate models that seek to represent the effect of waves fracturing sea ice.

Christopher Horvat and Lettie A. Roach

Status: final response (author comments only)

Comment types: AC – author | RC – referee | CC – community | EC – editor | CEC – chief editor | : Report abuse
  • RC1: 'Referee report on gmd-2021-281', Timothy Williams, 14 Sep 2021
  • RC2: 'Comment on gmd-2021-281', Predrag Popovic, 11 Oct 2021
  • CEC1: 'Comment on gmd-2021-281', Juan Antonio Añel, 12 Oct 2021

Christopher Horvat and Lettie A. Roach

Data sets

Training data used to train WIFF1.0 Christopher Horvat, Lettie Roach https://doi.org/10.5281/zenodo.5108636

Model output from different WIFF implementations Christopher Horvat, Lettie Roach https://doi.org/10.5281/zenodo.5106703

Model code and software

WIFF1.0 Code: Neural networks and code for developing training data and training WIFF neural networks Christopher Horvat, Lettie Roach https://doi.org/10.5281/zenodo.5171498

Christopher Horvat and Lettie A. Roach

Viewed

Total article views: 600 (including HTML, PDF, and XML)
HTML PDF XML Total BibTeX EndNote
503 89 8 600 1 0
  • HTML: 503
  • PDF: 89
  • XML: 8
  • Total: 600
  • BibTeX: 1
  • EndNote: 0
Views and downloads (calculated since 25 Aug 2021)
Cumulative views and downloads (calculated since 25 Aug 2021)

Viewed (geographical distribution)

Total article views: 518 (including HTML, PDF, and XML) Thereof 518 with geography defined and 0 with unknown origin.
Country # Views %
  • 1
1
 
 
 
 
Latest update: 25 Oct 2021
Download
Short summary
Sea ice is a composite of individual pieces, called floes, ranging in horizontal size from meters to kilometers. Variations in sea ice geometry are often forced by ocean waves, a process that is an important target of global climate models as it affects the rate of sea ice melting. Yet directly simulating these interactions is computationally expensive. We present a neural network-based model of wave-ice fracture that allows models to incorporate their effect with no added computational cost.