An artificial neural network correlation for the wave-induced melt rate of floating icebergs in the Barents Sea

Iceberg deterioration models are commonly coupled to iceberg drift models in the literature to improve the accuracy of predicting iceberg drift trajectories when exposed to open water. Wave-induced erosion is the major mechanism of iceberg deterioration, typically estimated by simple parametrized equations in the literature. Using linear wave theory, this paper develops a novel model to determine the wave-induced melt rate on the submerged lateral surface of a floating cylindrical iceberg with two degrees of freedom, including surge and heave motions. Existing solutions to the boundary value problems due to wave diffraction and radiation are adopted in this study to yield the water particle and iceberg velocities and, subsequently, the wave convective coefficient. A parametric study is conducted to investigate the sensitivity of the modeled melt rate to input parameters, including wave period, wave height, iceberg radius, iceberg width to height ratio, circumferential angle, and the vertical position on the iceberg draft. This model for the melting rate is then used to produce a dataset within the range of input parameters. The main novelty of this paper is to use this dataset to train an Artificial Neural Network (ANN) model and present a new correlation for the wave-induced melt rate. This correlation, which is easier to implement than the original analytical solution, is a function of iceberg radius, iceberg height, wave period, and wave height. Results indicate a peak in the wave-induced melt rate diagrams along with the wave period for different values of input parameters. This peak occurs at a wave period close to the iceberg's natural heave period because the iceberg in this condition undergoes a significant movement in the vertical direction due to heave resonance. The Reynolds number is found to be the key parameter on model generalization. In addition, predicted melt rates by the ANN correlation agree well with model results.