1. A deep learning-based scheme for parameterizing sea surface roughness was proposed and successfully applied to WindWave 1.0, a coupled ocean-atmosphere-wave model for the Northwest Pacific region developed by my group. In five typhoon cases in August 2020, the RMSEs of the sea surface winds forecasted using the new scheme were reduced by 6.02% to 14.75%, 11.17% to 18.30%, and 11.91% to 19.46% at 24-, 48-, and 72-hour forecast lead time, respectively, compared with the four traditional schemes, which suggests that the new sea-surface roughness scheme can successfully improve the NWP sea surface wind forecast.