Head(s): Prof. Dr. Henning Rust (FU Berlin), PD Dr. Peter Névir (FU Berlin), Dr. Péter Koltai (FU Berlin)
Project member(s): Dr. Annette Müller, Robert Polzin
Participating institution(s): FU Berlin

Project Summary

Aiming at the efficient modelling of large scales for climate change studies, the objectives of this project are (O1) to develop a small scale stochastic model for convective activity and (O2) to describe convective feedback on the large scale atmospheric flow. This leads to a hybrid model with a stochastic component for a conceptual description of convection embedded in a deterministic atmospheric flow model – a so-called stochastic parametrisation.
During the first phase, we implemented and refined a multi-level stochastic precipitation model. These Poisson-cluster-type models consist of a hierarchy of Poisson-processes on different scales. The “rain cell” at the smallest scale bears some resemblance to the stochastic massflow component of the Plant–Craig scheme – a stochastic parametrisation for convection. The next level introduces a relation between individual rain cells in terms of a clustering on the meso-scale. A potential driver to control these small scale models is the Dynamic State Index (DSI), an “adiabaticity indicator”. This concept has been analytically transferred from the atmospheric root model (“primitive equations”) to the quasi-geostrophic and Rossby approximation. The latter is insensitive to small-scale processes and thus well suited as a large-scale driver controlling the small-scale precipitation model.
For the second funding period we plan the following contributions to our objectives.
(O1) Downward coupling: large-scale flow driving small-scale convection.

• To overcome limitations of the Poisson-cluster process, we pursue two approaches: a) Feature identification and tracking from visual data analysis combined with coherent set analysis to obtain a conceptual description of convection using characterisations motivated by the Poisson-cluster model. b) Data-based methods for direct probabilistic modelling of categorised vertical mass flow.
• Development of further DSI variants as indicators for general thermodynamic processes. Together with other parameters related to moisture processes, these build the set of potential control variables for convective activity. 

(O2) Upward coupling: small-scale convective feedback to large-scale flow.

• A model for cloud characteristics relevant for the feedback is learned from available largeeddy simulations with the help of coherent set analysis.
• A conceptual characterisation of convection with associated cloud properties is used in conjunction with purely data-based dynamics-learning techniques to directly assemble a quantitative feedback function driving the large scale evolution.

We pursue these aims by application and simultaneous development of coherent set analysis, further by utilising data-based identification and investigation of non-linear systems. Including Péter Koltai as new PI provides the required expertise in these fields.

Publications A01