The ocean surface wind vector field is a key element for short term weather forecasting. Those forecasts are needed to issue timely and accurate ocean weather warnings and avoid major catastrophes. Many recent research attempts to measure and forecast ocean surface wind speed using polarimetric microwave radiometry provided by satellites. However, the amount of data provided by a satellite is very large and has many redundant data. Consequently, any type of analysis of this data should be done on a subset.

In this research we use support vector machines to extract a subset that is composed of support vectors. The size of the subset is fewer than ten percent of the total data for this data type. Moreover, to deal with the size of the satellite data, we first apply Voronoi tessellation. The results obtained show that the support vectors allow reconstruction of the ocean surface wind speed vectors with high accuracy regardless of the initial data set size.