Due to the lack of sounding data for the ocean area, especially the high sea, and the errors of the Bevis empirical model that is commonly used for the sea area, the marine area E-T_m regression model was built for the sea area based on ERA-Interim reanalysis data. Using the ERA-Interim data and offshore sounding data at different times, the fitting effect and forecasting ability of the E-T_m regression model were tested and compared with the conventional T_m acquisition method. The results show that the E-T_m regression model fits well and demonstrates higher prediction stability and accuracy compared with the Bevis empirical model. The E-T_m regression model was also compared with the localization model built based on sounding data in the typical sea area in terms of the prediction accuracy. The result shows that the E-T_m regression model with a resolution of 2.5°×2.5° has the equivalent accuracy to the localization model built based on the sounding data, with only 1% of average conversion error. Thus, the E-T_m regression model can be used in the sea area where the sounding station cannot be constructed.