This paper explores the use of model based ensemble forecasting and Gaussian process (GP) modelling of ice drift as a risk reducing and situational awareness tool for supporting under ice operations with autonomous underwater vehicles (AUVs). Ice tethered navigation buoys will be used to guide the vehicle under the ice and back to the ice-relative starting position. As an operational tool, we have developed a forecast model, where we attempt to predict the final destination of the vehicle, the total AUV path length, the drift of the buoys, and the properties of the underlying velocity field of drifting ice. The operation is a part of the Nansen Legacy project, and is planned for the Barents sea in May 2020. Ice and ocean in the Barents sea are in constant flux, and forecasting the motion of the ice gives insight into how the vehicle will move, and if it gets lost, where to start the search. In order to test our model, we have developed a simulator in which we can simulate the trajectories of the ice at a higher spatiotemporal resolution than most ice models. We present the simulator along with the method of forecasting where a model for each buoy is estimated and a GP is used to estimate the vector field. The divergence of the vector field is then used as an indicator of the probability of nonlinear events such as lead or pressure ridge formation. The forecast accuracy is analysed, and we can observe a dependence on spatiotemporal covariance.