We present a compartmentalization of ocean surface wind in coastal seas by analyzing statistical features of high-resolution wind measurements derived from Synthetic Aperture Radar (SAR) onboard European Remote sensing Satellite (ERS) -1/2. Probability distributions of wind speeds over land and sea, to which two-parameter Weibull distribution gives a good fit, are classified in a feature space of two Weibull parameters. As a result, we identify three distinguished clusters, which feature variability of wind over land, open ocean, and coastal sea. The typical wind speed distribution assigned to coastal sea has both characteristics of those assigned to land and open ocean, i.e., peak frequency in the lower wind and higher frequencies in higher wind speeds. Moreover, it can be reproduced by mixing the typical wind-speed distributions of land and open ocean at a given rate.