What are the benefits for convection of increasing the horizontal resolution of convection permitting NWP models?

Many operational centres are now realising the great potential for improvements to forecasts of rainfall from kilometer scale models resolving convection explicitly. Such models already provide big benefits compared to the previous generation of order 10km models with parameterised convection. At kilometer scales deep convection is inevitably under-resolved, hence the often used term "convection permitting" to describe these models. Being under-resolved manifests itself as well documented deficiencies in the representation, too heavy peak rainrates, too large convective cells, delayed initiation etc.

An important question going forwards is to what degree these issues are helped by increasing the horizontal resolution of the models. Increasing horizontal resolution is expensive and there are many competing demands on computer time (more ensemble members, larger domains etc) so, in practice, the resolution of the operational models is only likely to be further increased where there are major improvements to the forecasts.

At the Met Office the operational UK model runs with a gridlength of 1.5km. We have been experimenting with research models with various gridlengths down to 100m in conjunction with observations from the DYMECS and COPE convection field projects. These projects produced observations of convection around the Chilbolton radar facility in southern England and of lines of convection along the UK south west peninsula respectively. With current model configurations many apects of the representation of convection (initiation, cell lifetime, updraft width) improve as the gridlength is reduced. However it is not clear that the representation of the clouds and rain is improved. Clouds, and the surface rain fields, often appear more fragmented. This may be related to the model representation of the downdrafts associated with convection. New observational techniques for estimating vertical velocities shed some light on this. Additionally it is not clear if the usual bias observed with explicit convection of too much heavy rain and too little light is helped by increasing the resolution.

In this paper we will discuss the above likely benefits and problems with explicit convection in models as their gridlength is reduced below ~1km. Many of these aspects have been found to be dependent on the mixing and microphysical formulation of the models. We will describe work to unpick the fundamental causes of some of these issues.