A physical subgrid scale information exchange (PSIE) systemfor parameterization schemes in numerical weather prediction models
Zusammenfassung der Projektergebnisse
The quality of forecasts from numerical weather prediction models is largely determined by the physical parameterization schemes which are needed to describe the subgrid scale processes. These parameterization schemes closing the Reynolds-averaged system of underlying prognostic equations derive a significant amount of additional information of which only the tendencies of the grid scale prognostic variables is passed to the hosting model. So, the parameterizations can only communicate to each other by means of a limited set of averaged variables, but the most part of subgrid scale information is confined in the respective parameterization and is consequently lost. To overcome this conceptual shortcoming and to let the parameterization schemes benefit from each other we have developed a physical subgrid scale information exchange (PSIE) system. This system consists of a set of communication paths with send and receive modules evaluating the subgrid scale information of one scheme and passing it to another. E.g., the convective entrainment and detrainment mass fluxes cause subgrid scale buoyancy and wind shear initiating turbulence. Ten such paths have been identified all of which are based on physical processes connecting the parameterization schemes (i.e. microphysics, convection, turbulence, fractional cloud cover radiation scheme). The PSIE system has been implemented into the COSMO model of Deutscher Wetterdienst (DWD) with a grid size of ∆x = 7 km (which is comparable to its successor ICON-EU nest of DWD). Taking the physical processes as a basis for the PSIE paths (rather than specific quantities provided by a scheme) simplifies the application of the PSIE system also to other hosting models. Three model versions have been validated by means of RADOLAN data of DWD (radar data calibrated as precipitation rates by means of station data): COSMO-REF (no PSIE paths at all), COSMO-OP (only with the few operationally parameterization couplings) and COSMO-ALL (full PSIE scheme). Visual inspection of the results (simulated precipitation rates against RADOLAN observations) reveal that the spatial structures are often more realistic with PSIE than without. Especially too localized precipitation from airmass convection is ”broken up” and scattered in a way more similar to observations. As a more quantitative and objective validation method the validation by fraction skill scores (FSS) confirm these first findings. FSS analyses also reveal that the forecast skill usually increases from COSMO-REF via COSMO-OP to COSMO-ALL for at least two days lead time. The significance of the single PSIE paths increases from the fractional cloud cover scheme via convection to turbulence and microphysics. The scale selective FSS also reveals that a grid size of ∆x = 7 km structures down to scales of 60 km can benefit from PSIE. By means of the factor separation method (based on FSS) interactions between the single PSIE paths have been studied. As the most interesting result the convection scheme has a stronger impact on the precipitation from the microphysics scheme (via shear and buoyancy initiating turbulence, which in turn modulates the accretion of cloud droplets) than the more direct seeder feeder effect. In summary, the PSIE system improves the forecast quality, especially the full PSIE scheme is usually superior to the operational parameterization couplings. Its effect is most striking in situations where precipitation is not synoptically prescribed (e.g. along a front) but is expected to be statistically scattered as in airmass convection.
