Cloud properties, including droplet number (n_c) and mass (q_c) concentrations, vary on a wide range of spatial scales many of which cannot be resolved by regional and global atmospheric models due to their coarse horizontal resolutions. This necessitates treating the effects of subgrid horizontal variability of cloud properties on grid-mean process rates in a parameterized way. Several approaches of accounting for the subgrid horizontal variability in q_c have been proposed and used, but models typically neglect variability in n_c. In this study, we analyze structure of low-level warm clouds using an ensemble of large-eddy simulations (LES) of stratocumulus and shallow cumulus cases in continental and marine environments. The n_c and q_c, computed from a bin microphysics scheme that predicts cloud droplet spectra in LES, are found to be highly correlated. Using a parameterization of autoconversion rate (Au) as an example, we demonstrate the effects of the variability on a horizontally average rate of a process that has a non-linearly dependency on n_c and q_c. To account for the horizontal variability of the cloud properties across the model grid, the Energy Exascale Earth System Model (E3SM) version 2 climate model applies an enhancement factor to Au estimates obtained from grid-mean n_c and q_c. However, this enhancement factor is overestimated because it considers only variability of q_c and neglects the variability of n_c and its co-variability with q_c. Reasonable estimates of grid-mean Au values are still obtained by tuning the local Au rate to have an offsetting low bias. The tuned Au, however, has a stronger dependency on q_c and a weaker dependency on n_c, thereby potentially modifying the model sensitivity to aerosol perturbations. To eliminate compensating biases in the Au and its enhancement factor, the variability in n_c must be considered, and the presentation will demonstrate how the presented analysis provides a basis for developing such an approach.