A clustering method is applied to high resolution simulations of shallow continental convection to investigate the size dependence of coherent structures in the convective boundary layer. The study analyses the geometry of the clusters, along with their profiles of vertical velocity and total water. The main science goal is to assess various assumptions often used in spectral mass-flux convection schemes. Novel aspects of the study methodology include i) a newly developed clustering algorithm, and ii) an unprecedentedly large number of simulations being analysed. In total 26 days of LASSO simulations at the ARM-SGP site are analyzed, yielding roughly one million individual clusters. Plume-like surface-rooted coherent convective clusters are found to be omnipresent, the depth of which is strongly dependent on cluster size. The largest clusters carry vertical structures that are roughly consistent with the classic buoyancy-driven rising plume model, while
smaller clusters feature considerable variation in top height.
The cluster area is found to strongly vary with height and size, with small clusters losing mass and large clusters gaining mass below cloud base.
Similar size dependence is detected in kinematic and thermodynamic properties, being strongest above cloud base but much weaker below.
Finally the efficiency of the top-hat approach in flux parameterization is investigated, found to be 80-85 \% including a weak but well-defined dependence on cluster size. Implications of the results for spectral convection scheme development are briefly discussed.