Quantifying Ensemble Divergence in Large-Domain Convective-Scale Simulations over Africa

Evidence indicates that since convective-scale simulations can explicitly resolve motion around deep convection, they may improve representation of coupling between small-scale moist convection and upscale modes of atmospheric variability.  Prior studies indicate that there can be a shift in the mean state of large-scale tropical circulations in convective-scale simulations relative to models with parameterised deep convection. However, it is uncertain whether this shift is systematic in convective-scale simulations or simply the response in a single model realisation.

To resolve this uncertainty, we run a 9-member ensemble of simulations over tropical southern and eastern Africa, using the Met Office Unified Model on a 2.2km grid with no deep convection parameterisation. ERA5 forces the lateral boundaries and simulations use FLake, a lake scheme to reduce over-lake biases in precipitation. The ensemble will be compared to a similar configuration which uses a deep convection parameterisation and a 12km grid.

Our ensemble experiments quantify the internal variability associated with varying initial conditions in the tropics and subtropics, relative to the variability induced by lateral boundary forcings. The ensemble divergence will be compared for the simulations with and without convection parameterisations to explore implications of ensemble design for high-resolution simulations of large domains. Furthermore, the hypothesis that there is a systematic mean-state shift in large-scale tropical circulations in kilometre-scale simulations relative to coarser GCMs will be evaluated using the two ensembles.

Effects of incorporating FLake and the role of soil moisture in initialisations will also be discussed, as well as their implications for predictability in kilometre-scale simulations. Model outputs will be compared to in-situ observations over northwest Zambia obtained during the 2022 DRYCAB field campaign, and we will outline how these results inform the design of planned further simulations to investigate monsoon onset predictability on subseasonal-to-seasonal timescales.