Coastal offshore observation stations are rare, but coastal land observation stations are relatively dense. Carrying out research on the characteristics of land-sea wind speed difference caused by land-sea distribution and realizing the estimation of sea surface wind speed from land observation wind speed is helpful to improve the service ability of sea surface wind forecast. In this paper, the hourly mean wind speed and gust speed data of two groups of buoys and their adjacent land observation stations in the north coast of China from 2016 to 2020 are used to statistically analyze the characteristics and regularity of the difference between sea surface wind speed and land wind speed. And the support vector machine method is used to build the sea surface wind speed estimation model based on the land mean wind speed, land gust speed, distance between land-sea stations, month and hour. The estimation model is tested by using the observation data of the other two groups of land-sea observation stations in 2021. The results show that for the mean wind speed ≥ scale 6 and gust wind speed ≥ scale 7, the model has a high estimation accuracy. The root mean square errors of the mean wind speed (gust speed) of the two groups estimated by the model are 2.40 m·s-1 (3.20 m·s-1) and 2.35 m·s-1 (2.57 m·s-1), respectively. Compared with ERA5, the errors decreased by 24% (14%) and 23% (20%) respectively. In a strong wind process jointly affected by an extratropical cyclone and cold air, the mean absolute errors of the mean wind speed (gust speed) estimated by the model for the two test groups are 1.6 m·s-1 (2.3 m·s-1) and 1.1 m·s-1 (1.5 m·s-1) respectively, and the mean wind speed (gust speed) errors at the extreme moment are -1.3 m·s-1 (-0.6 m·s-1) and -1.2 m·s-1 (-3.1 m·s-1) respectively, which are better than those from ERA5. The sea surface wind speed estimation model based on support vector machine can estimate accurate heavy sea surface wind speed using the land observation wind speed, which can reduce the impact of insufficient sea observation data, and has a certain application prospect.