A physical subgrid‐scale information exchange (PSIE) system for parametrization schemes in numerical weather prediction models.

The choice and interaction of the parametrization schemes in a numerical weather prediction (NWP) model are crucial for the forecast quality. Because, classically, the parametrization schemes can only communicate to each other via the tendencies of the averaged grid‐scale prognostic variables, a considerable amount of subgrid‐scale information available within the schemes is lost. To overcome this problem, subgrid‐scale information should be directly exchanged between the parametrization schemes. In the present paper already existing subgrid‐scale communication paths, which have been successfully used in operational forecasts in the past, are extended to a physical subgrid‐scale information exchange (PSIE) system including all atmospheric parametrization schemes. For our studies we have chosen the parametrization schemes of microphysics, convection, turbulence, fractional cloud cover and radiation. The COSMO model of Deutscher Wetterdienst (DWD) with a grid size of Δx = 7 km serves as the hosting model for the implementation of the PSIE system. However, the couplings are formulated in a way that also allows the PSIE system to be applied to other NWP models. Simulations of real case‐studies of different weather situations are performed and compared to DWD observational precipitation data from RADOLAN (radar measurements calibrated by station data). Visual inspection of precipitation maps reveals that precipitation patterns look more realistic with the full PSIE system, especially during air‐mass convection situations. This is confirmed by fractional skill score (FSS) analyses, which also give an insight into the temporal development of the PSIE effects. The relative impact of the PSIE paths and their interaction is also studied in conjunction with a factor separation analysis. Usually parametrization schemes in numerical forecast models exchange their subgrid‐scale information via the grid scale. To avoid information loss as a result of Reynolds averaging, our PSIE system provides direct exchange paths as shown in the sketch. As demonstrated with precipitation forecasts of real cases (comparisons by eye, a statistical fraction skill score analysis and a factor separation method), direct communication of subgrid‐scale information by means of a PSIE system can significantly improve the forecast. [ABSTRACT FROM AUTHOR]