Dynamical Impacts of Warm-Starting Operational Weather Models over Africa

Weather models which allow explicit convection can add value to weather forecasting by improving the intensity and timing of precipitating systems and their dynamics, which is particularly valuable in the tropics where moist diurnal convection dominates. In West Africa, convection can become organised to form mesoscale convective systems which are crucial for supplying water but may have devastating impacts, and while convection-permitting models improve forecasts, issues remain in the implementation of operational convection-permitting models. A major problem is initialising weather models in the tropics, where measurements are sparse and weather systems are dominated by nonlinear diabatic processes, which makes data assimilation challenging. Currently, the UK Met Office runs a regional operational convection-permitting model in Africa, the Tropical Africa Model, which is initialised using lower-resolution global models. However, the global models have extremely limited convective-scale structures and as a result, there is a spin-up time of around 12-18 hours before the model begins to accurately reflect precipitation.

To counteract this problem, a “warm-starting” method has been trialled. The warm-starting technique blends large-scale features from the global model and fine-scale fields below a certain length scale from previous runs of the high-resolution model to use as an initial state in a new model run. It combines the more realistic convective structures of the regional model and the more accurate synoptic conditions in the global model. Not only is the approach cheaper and quicker than traditional data assimilation, both in terms of development and computational cost, but it also shows demonstrable improvements in the representation of precipitation for the first 12 hours of the model and beyond relative to simulations where the model has simply been initialised with the global model (a “cold-start”). The cold-start simulations appear to consistently predict rainfall that is too intense even beyond the first 12 hours.

We investigate why warm-starting models produces more realistic rainfall distributions by examining the dynamical structures: producing statistics of rainfall objects as forecasts evolve and examining their connection to the dynamics. We examine the energetics of convection in the convection-permitting models, aiming to provide a justification for the scale length at which we include structure from previous model runs using this technique.