This paper presents an efficient, physics-based procedural model for the real-time animation and visualization of cumulus clouds at landscape size. We couple a coarse Lagrangian model of air parcels with a procedural amplification using volumetric noise. Our Lagrangian model draws an aerology i.e., the atmospheric physics of hydrostatic atmosphere with thermodynamics transforms, augmented by a model of mixing between parcels and environment. In addition to the particle-particle interactions, we introduce particle-implicit environment interactions. In contrast to the usual fluid simulation, we thus do not need to sample the transparent environment, a key property for real-time efficiency and scalability to large domains. Inheriting from the highlevel physics of aerology, we also validate our simulation by comparing it to predictive diagrams, and we show how the user can easily control key aspects of the result such as the cloud base and top altitude. Our model is thus fast, physical and controllable.