Representing convection in models - How stochastic does it need to be?

The representation of convection in large-scale models remains one of the most difficult problems in atmospheric science. This is due to the need for this representation to be of conceptual nature as convective processes act on scales much smaller than the grid-spacing of global weather and climate models. As a consequence, convective processes need to be parametrized, i.e., the behaviour of variables on the (relatively) small convective scales must be described in relation to that of the variables on the (relatively) large scales resolved by the model. The ideas and concepts that underpin parametrizations of convection date back more than 40 years and were originally developed for descriptions of tropical cyclones (e.g., Ooyama (1969), Kuo (1965)) before being introduced in a more general context (e.g., Arakawa and Schubert (1974), Kuo (1974)). The basic ideas of traditional convection parametrization as well as recent developments are summarized in the review by Arakawa (2004).