Estimating wind velocity fields from geostationary satellite images

The GOES satellites operate in geostationary orbit, recording images of the continental United States every 5 minutes at 16 separate wavelengths. We consider the problem of using these images to estimate wind vectors in the upper atmosphere. The problem is complicated by the fact the images do not contain discrete tracers moving through the atmosphere. Instead, we observe scalar fields with more amorphous features that evolve over time. To address these problems, we make use of spatial-temporal Gaussian process models that have parameters controlling how the field moves over time.