A key component of robotic path planning for monitoring dynamic events is reliable navigation to the right place at the right time. For persistent monitoring applications (e.g., over months), marine robots are beginning to make use of the environment for propulsion, instead of depending on traditional motors and propellers. These vehicles are able to realize dramatically higher endurance by exploiting wave and wind energy, however the path planning problem becomes difficult as the vehicle speed is no longer directly controllable. In this paper, we examine Gaussian process models to predict the speed of the Wave Glider autonomous surface vehicle from observable environmental parameters. Using training data from an on-board sensor, and wave parameter forecasts from the WAVEWATCH III model, our probabilistic regression models create an effective method for predicting Wave Glider speed for use in a variety of path planning applications.