The ice loads occurring in ship-ice interaction are uncertain for many reasons. This article addresses some of those uncertainties by studying adjacent frames and identifying cases where the peak ice load occurs on a certain frame. In these cases, we looked at photos of ice to identify what type of an ice response caused the load to peak. We further selected the cases where the peak load was caused by a rotation of an already broken ice cusp. This rotation allowed us to measure the cusp dimensions from the photos, which enabled us to study two research questions. First, how the cusp area distribution is affected by the ice thickness and ship speed. Second, how the peak load distribution is affected by the cusp dimensions and ship speed. Both of these questions were modeled with hierarchical Bayesian models, where the Gaussian process priors compel the distribution parameters to smoothly change in varying conditions. The models predict that the cusp area increases with increasing ice thickness and deceasing ship speed, whereas the peak load increases with increasing cusp volume and decreasing ship speed. The models were tested with posterior predictive checks, which demonstrated good agreement with the models and the measurements.