Using large-eddy simulations (LESs) of six observed boundary layer cases as benchmark data sets, we diagnosed vertical velocity as the sum of the grid-mean ($overline{w}$) and the subgrid-scale w using an empirical method we have developed. We have found that the subgrid-scale w is best characterized by the root-mean square of w'2. The vertical velocity is expressed in the form of w = $overline{w}$ + c$sqrt{overline{w^{prime\,2}}}$, where c is diagnosed from LES data to be 0.24. The w is then used to diagnose the cloud droplet numbers using a simple predictive equation of cloud nucleation following previous works in the literature. Good agreement is found between diagnosed cloud droplet concentrations and predicted cloud droplet concentrations using an LES model.