|Title||Impact of truncation on variable resolution forecasts|
|Year of Publication||2010|
|Secondary Title||Technical Memorandum|
The ECMWF variable resolution ensemble prediction system (VAREPS) uses a higher resolution up to the truncation time and a lower resolution afterwards. For variables with a large variance on the small scales such as total precipitation, ensembles run at higher resolution have larger spread than ensembles run with lower resolution (and with the same initial perturbations). Thus, it is not surprising that the VAREPS spread is higher up to the truncation time and lower afterwards, but besides this effect, results indicate that up to 24 hours due to the truncation the VAREPS spread decreases too excessively, to a level below the one expected for the lower resolution ensemble. Although this excessive reduction does not have any effect on upper-level variables, it is detectable in the time evolution of the ensemble spread for variables such as total precipitation. This work discusses results from a series of investigations performed to identify the main cause of this too large reduction in the VAREPS spread. Results indicate that the main reason of the too excessive spread reduction is the interpolation of the high-resolution fields to define the initial condition of the low-resolution integration. After the interpolation, the VAREPS members have too little variability on the small scales, and as a consequence the VAREPS spread component that projects onto the small-scales is reduced. Experiments run with a realistic or a flat orography show a very small sensitivity to orographic forcing. Experiments performed in aqua-planet mode show that after the truncation the more active convection can more quickly re-generate spread on the small-scales. Considering the operational VAREPS, it is concluded that the current configuration with a 24-hour overlap period should not be modified.