Predicting the amount of wind power generation with high accuracy represents a priority for any market participant or power balancing authority. One of the main ingredients when deriving these predictions are the wind speed data. The current research in progress aims at refining our understanding of wind speed measurements and developing a method to find those which are most relevant to the wind power generation. In practice, we perform this study on the wind farms in Southern California market (also known as CAISO SP15). For this region, we build random fields by time series of gridded historical forecasts and actual wind speed measurements provided by the aviation reports (METAR). As none of these values are taken from where the wind farms are, we use kriging techniques in order to direct our field in space and time to the exact farm locations. In this process, an important challenge is brought by the fact that the actual measurements are instantaneous values in time and they can very often be zero-valued. The main benefits of this work is that (1) it will highlight when the wind speeds will not allow the mills to produce electricity and (2) it will provide a way to forecast wind speeds at wind farm locations.