Nowcasting Using a Simplified Model of Atmospheric Convection

Forecasting deep moist convection is challenging due to the complex, multiphase processes that occur across a wide range of spatial and temporal scales. In particular, short-term forecasting (or nowcasting) remains difficult, especially the prediction of new convective cells. Idealised modelling can help simplify this complexity by isolating the mechanisms responsible for triggering new convection.

This thesis presents advances in nowcasting by using a simplified model of the atmosphere as an alternative to typical observational data, with the aim of extending the fundamental understanding of short-term prediction of convective initiation. First, a simplified framework focused on secondary initiations by gravity waves is introduced, and the resulting convective regimes are explored. A relaxation-burst attractor is identified, and a relaxation timescale is shown to depend on the model diffusivity. Distinct regimes of convective initiation are identified and linked to different dynamical processes.

The predictability of the relaxation-burst attractor is then examined using Echo State Networks, demonstrating that short-term predictions can be achieved using only large-scale model variables. Finally, forecasts are produced for the full simplified model, with particular focus on predicting the timing and location of new convective initiations. Using reduced-order representations of the dominant dynamics, skilful predictions of plume development are obtained.

Overall, the results presented in this thesis demonstrate the value of simplified models as a tool for incorporating physical understanding into nowcasting and for improving the prediction of new convective initiations.