Geoscience Reference
In-Depth Information
with UK4 or UKV precipitation forecasts. An 8 member ensemble and control member
(unperturbed) nowcast to T+7 h are produced every 15 minutes. Recent Root Mean Squared
Factor error statistics for STEPS control member advection nowcasts and UK4 and UKV
forecasts of precipitation have shown that STEPS nowcasts are superior in the first 2.5 hours
(see Figure 3).
Fig. 3. Root Mean Squared Factor errors for November 2009 based on hourly accumulations
greater than 1mm, smoothed to a scale of 6km, and measured using radar derived
accumulations as the reference observation. The performance of 1.5 km (green) and 4 km
(red) grid length, UK configurations of the Met Office's Unified Model are compared with
control member STEPS nowcasts blending radar extrapolation with UM: 1.5 km (purple)
and UM: 4km (blue) model forecasts. The performance of the STEPS nowcast blending
extrapolation with the UM: 1.5 km forecast does not asymptote to that of UM: 1.5 km model
because this was an experimental configuration run without prior calibration.
The implementation of an NWP-based nowcast system in the Met Office is focused on
improving the prediction of convective storms for flood forecasting. The ultimate aim is to
replace the existing extrapolation-based precipitation nowcasts and site specific forecasting
techniques. Boundary conditions will be provided by the 6 hourly 1.5 km UKV system.
An hourly analysis and forecast system for southern England has been run experimentally
for a limited number of cases of summer rain and convection, using conventional data and
3D-Var or 4D-Var, plus latent heat nudging of radar derived rain rates and humidity
nudging based on analysed 3D cloud cover nudging (Macpherson et al
.,
1996; Jones &
Macpherson, 1997; Dixon et al., 2009). The direct variational assimilation of cloud cover has
not yet been tested in the hourly cycling system. This has used a fixed 1.5 km resolution
configuration of the Unified Model and a 3 km resolution 4D-Var grid or 1.5 km and 3 km
resolution 3D-Var grid.
