Obtaining mesoscale singular vectors reflecting synoptic-scale uncertainty by projection in phase space

The choice of initial perturbation technique in regional ensemble prediction systems is important for forecasts of heavy rain events. Although the uncertainty at synoptic scale has a strong impact on the forecast performance of mesoscale phenomena, there is no established method for explicitly providing synoptic-scale uncertainty when generating mesoscale initial perturbations. In this study, a new method of calculating mesoscale singular vectors (SVs) that reflect synoptic-scale uncertainty by applying a regional targeting technique to a subspace of the model phase space spanned by global ensemble perturbations was developed. Comparison of the results of this new method with those of the original SVs showed that the new SVs with high growth rate corresponded to the original SVs with high growth rate, whereas the new SVs with relatively lower growth rate had a larger contribution from the global ensemble perturbations. The new SVs mitigated the tendency of the original SVs to localize over the southern sea of the computational domain, and the forecast errors were better captured by the new SVs than by the original SVs. Furthermore, the effect on forecast performance of the new SVs as initial perturbations was confirmed by using the operational regional ensemble prediction system at the Japan Meteorological Agency. It was found that the precipitation probability forecast performance was superior in the first half of the forecast period at all rainfall thresholds.