This talk focuses on spatio-temporal models for short-term probabilistic prediction of wind data. Traditionally, wind speed and wind direction have been addressed independently, without taking dependencies into account. However, in many situations it is important to have the full information about the bivariate structure of wind. We compare the performance of a spatio-temporal model for wind speed directly with a coregionalization model for the wind vector. In both cases, the linear predictor is a function of covariates, a smooth function to capture the daily seasonality in wind and a latent Gaussian field to model the spatial and temporal dependencies. The proposed methods are tested on a dataset of hourly wind speed at 28 stations in Saudi Arabia. To meet the computational requirements, we take a Bayesian framework and obtain fast and accurate forecasts not only at locations where recent data are available but also at stations without observations. We validate spatially out-of-sample forecasts on simulated data from a computer model. Based on this case study, we provide a detailed analysis on how increasing the number of locations can improve the forecast performance.