Trials of high resolution versions of the Unified Model for short range forecasting of convective events

With the advent of the non-hydrostatic version of the Unified Model and the increases in available computer power the Met Office is moving towards higher resolution models for short range forecasts. A 4km gridlength model has recently been implemented operationally over the UK area and it is anticipated that a 1km model will be feasible in the next few years.

This paper will describe trials which have been carried out on a suite of models with 4km and 1km gridlengths (and also 12km for comparison). The models have been run on domains over southern England for a number of convective cases from the summer 2003 and summer 2004 periods. For each case several forecasts of 6 hour length were run at 3 hour intervals. The 1km model was run without a convection parameterisation. In contrast it has been found necessary to include the convection scheme in the 4km model but with the mass flux restricted to encourage the model to represent deep convection explicitly. Other aspects of the model configuration such as the microphysical formulation will be described in the paper. The suite included assimilation in the 4km and 1km models to enable relatively high resolution information to be assimilated from cloud and radar data. It also enabled high resolution structure to be be carried forward from one forecast to the next. For comparison purposes the models were also rerun without assimilation simply starting each forecast from the relevant 12km analysis.

Results from the trials will be presented. Fields from several of the cases will be shown in order to highlight benefits, and problems currently being seen with the models. Statistics from a scale selective precipitation verification system will also be discussed. It will be concluded that, despite some problems, the 4km and 1km models demonstrate the clear potential to provide improved forecasts particularly for higher rainfall thresholds and for shorter accumulation times. In addition there is evidence that the 1km model can provide significantly better forecasts than the 4km. As a result of the large gridlength relative to the shower size the 4km model has a tendency to replace small scattered showers with a few large, very heavy and widely spaced showers. The 1km model does not have this undesirable characteristic and and also improves the problem of delayed initiation of convection (or no initiation at all in weakly forced situations) seen in the 4km model. Finally it has been found that assimilation will be essential in these models in order to avoid spin up/overshoot problems in the first few hours of the forecast.