In situ measurements indicate the complexity and nonunique character of radar reflectivity-liquid water content (Z-LWC) relationships in stratocumulus and cumulus clouds. Parameters of empirical (statistical) Z-LWC dependences vary within a wide range. Respectively, the accuracy of retrieval algorithms remains low. This situation is partially related to the fact that empirical algorithms and parameters are often derived without a corresponding understanding of physical mechanisms responsible for the formation of the Z-LWC diagrams. In this study, the authors investigate the processes of formation of the Z-LWC relationships using a new trajectory ensemble model of the cloud-topped boundary layer (BL). In the model, the entire volume of the BL is covered by Lagrangian parcels advected by a turbulent-like velocity field. The time-dependent velocity field is generated by a turbulent model and obeys the correlation turbulent laws. Each Lagrangian parcel represents the "cloud parcel model" with an accurate description of processes of diffusion growth-evaporation of aerosols and droplets and droplet collisions. The fact that parcels are adjacent to each other allows one to calculate sedimentation of droplets and precipitation (drizzle) formation. The characteristic parcel size is 50 m; the number of parcels is 1840. The model calculates droplet size distributions (DSDs), as well as their moments (e.g., aerosol and drop concentration, mass content, radar reflectivity) in each parcel. In the course of the model integration, Z-LWC relationships are calculated for each parcel, as well as the scattering diagram including all parcels. The model reproduces in situ observed types of the Z-LWC relationships. It is shown that different regimes represent different stages of cloud evolution: diffusion growth, beginning of drizzle formation, and stage of heavy drizzle, respectively. The large scattering of the Z-LWC relationships is found to be an inherent property of any drizzling cloud. Different zones on the Z-LWC diagram are related to cloud volumes located at different levels within a cloud and having different DSD. This finding allows for improvement of retrieval algorithms.