Predicting surface wind speed (SWS) is crucial to operational weather forecasting. Accurate predictions can guide public activity planning, improve air safety, and forecast severe haze associated with stagnant air (Feng et al., 2018(Feng et al., , 2020b. They can also help improve the efficiency of wind power systems and the maintenance of wind farms, making the predictions relevant to clean energy development (Wang et al., 2020). Traditional models for the numerical weather prediction (NWP) of SWS use physical surface-layer schemes. Commonly used single-layer diagnostic schemes (Benjamin et al., 2004; European Centre for Medium-Range Forecasts, 2013;Skamarock et al., 2019) provide vertical wind profiles from the ground to the first level of the NWP model under various stability conditions Optis et al., 2014). As these schemes assume a homogeneous surface, they lose validity when the surface is a complex combination of buildings, vegetation, and variable topography (