A6 - Representing forecast uncertainty using stochastic physical prarameterizations

Principal investigators: Dr. Kirstin Kober, Prof. Dr. George C. Craig

Other researcher: Stephan Rasp (PhD), Mirjam Hirt (PhD)

Former researcher: Fabian Brundke (PhD)

Disturbances on convective scales, like thunderstorms, have the potential to grow upscale and influence the synoptic-scale weather pattern. For the initiation of convection several conditions are necessary: the large-scale environment, the amount of available instability at low levels, local triggers to overcome possible inversions and the availability of moisture in higher levels. Due to this complexity, the forecast of the exact timing and location of convective initiation with numerical weather prediction (NWP) models is an ongoing challenge and only reasonable with probabilistic methods.

Several physical processes are relevant for the distribution of moisture and temperature in the atmosphere and serve as local triggers for convection: turbulent processes in the boundary layer, the modification of boundary layer by orography or by outflows of already existing deep convective cells. The representation of these processes varies with atmospheric model but also with phenomenon. Even if they are parameterized with deterministic methods only their mean effect but not their variability is represented. Stochastic parameterizations allow to represent the variability of unresolved physical processes and hence have the potential to improve the representation of uncertainty in NWP forecasts.

This project will develop approaches on how to represent the inherent uncertainty of convective initiation relevant processes like the subgrid-scale orography, cold pools or mountain circulations in the COSMO model. Basis will be an an existing stochastic boundary layer parameterization that represents variability in turbulent fluxes due to surface heating. The extended parameterization will be tested in weather situations where these processes contribute to the initiation of convection. The error growth properties and their interaction with the large-scale flow will be investigated as well as ensembles based on the stochastic physics representation. Finally, the potential interaction of the different perturbations representing different phenomena will be analyzed.