Geoscience Reference
In-Depth Information
Also, traditional nowcasts can produce forecasts within a few minutes of data time, but
complex data assimilation methods and numerical integration of the governing atmospheric
equations are more costly and therefore take longer. However, if these techniques produce
improved forecasts at longer lead times, the benefits outweigh the timeliness issue and
reduced accuracy in the first 2 hours.
Another performance issue with NWP-based nowcasts relates to the latency of the boundary
conditions. This arises because domain sizes are usually small and are nested in coarser
resolution forecasts or larger domain forecasts with less frequent analysis cycling and later
data cut-off times. The consequences are that the boundary conditions and synoptic scale
forcing cannot be refreshed as frequently or as recently on the larger domain(s) as they are
on the nowcast inner domain. This limits the skill at longer forecast ranges and possibly
close to the boundaries.
Nonetheless, the advantage of NWP-based nowcasting lies in the fact that model
formulation, dynamical equations and physical parameterizations can predict the non-linear
evolution of weather elements and, in particular, the generation and decay of precipitating
weather systems.
5.3 A status report
To investigate the direct use of NWP for nowcasting, the Met Office in the UK is developing
an hourly cycling 4D-Var high resolution (1.5 km) NWP system to run on a domain covering
southern England (see section 5.4). This is nested within the most recent forecasts for the
whole of the UK (1.5 km resolution forecasts produced every 6 hours from 3 hourly 3D-Var
data assimilation cycles at 3 km resolution) to obtain boundary conditions. The latter may be
up to 6 hours old. Although 4D-Var is more expensive than 3D-Var, the aim is to evaluate
the benefit of assimilating high time-frequency sub-hourly data (Ballard et al., 2011): see
section 5.4 for more details.
Over the past 20 years, NCAR has undertaken many studies to explore the assimilation of
radar data into high resolution cloud and NWP forecast models. These have included using
the Variational Doppler Radar Assimilation System (VDRAS - Sun, 2005a, 2005b; Sun &
Crook, 1994, 1997, 1998, 2001; Sun & Zhang, 2008) with 4D-VAR (Sun et al., 1991, 2012).
These tend to use very short time-windows and have exploited the mesoscale model, MM5
3D-Var (Xiao et al., 2005) and the Weather Research & Forecasting Model (WRF) 3D-Var and
4D-VAR, or ensemble Kalman filter (Caya et al., 2005). These were run using VDRAS as part
of the forecast demonstration project during the Beijing Olympics (Sun et al., 2010).
Meteo-France has a 2.5 km, 3-hourly cycling 3D-Var scheme covering France (the
Application of Research to Operations at Mesoscale - AROME-France). This has been
operational since December 2008 (Seity et al., 2011; Brousseau et al., 2011). Radial Doppler
winds (Montmerle & Faccani, 2009) and humidity profiles derived from radar reflectivity
(Caumont et al., 2010) are assimilated. Meteo-France is also undertaking a project entitled,
“AROME-Nowcasting”, to adapt their 2.5 km grid length model, AROME, to meet the
requirements of nowcasting. The main difference to AROME-France is the production of an
analysis every hour, but without cycling. The potential benefits of a system called AROME-
airport, based at Charles de Gaulle airport near Paris, are also being explored. This model
will provide an input to a Wake-Vortex forecast model. The main goal is to add new,
