Master thesis topics

External projects


Topics for master theses in the theory group

Assimilation of measurements from surface stations with KENDA-COSMO

Surface observations represent an important source of information for numerical weather forecasts but their use has proven to be difficult and their full potential is currently not exploited.

A dense network of surface stations provide measurements of surface pressure, 10 m wind, 2 m temperature and 2 m humidity over Germany. Using a kilometer-scale ensemble data assimilation (KENDA) system, these data can be assimilated into the COSMO-DE limited area model. However, currently only temperature and humidity are assimilated from all stations in Germany, while wind observations are only used in Northern Germany where the terrain is sufficiently flat, and surface pressure is not used at all.

The aim of the project is to exploit the potential of these observations for convective-scale short-range forecasts using the KENDA-COSMO system. Error statistics of surface observations shall be investigated using already available COSMO-DE forecasts with a focus on the representativeness error of wind observations and height adjustment of pressure observations. These findings shall provide the basis for a series of data assimilation experiments computed with the local linux cluster.

The project requires a basic knowledge of programming and numerical weather prediction. The project will be performed in the framework of the HErZ Data Assimilation Branch at MIM.

Contact: Florian Harnisch, Martin Weissmann

Assimilation of humidity observations in COSMO-KENDA

In the framework of the HErZ Data Assimilation Branch , a forward operator for visible MSG SEVIRI observations has been developed with the goal to assimilate these observations.

This MSc project shall investigate the effect of assimilating complementary humidity observations (e.g. GPS delay or humidity-sensitive radiance channels) in the experimental KENDA esnsemble data assimilation system of DWD.

Research question: - What is the effect of humidity- versus cloud-related observations?

Contacts: Martin Weissmann

Assimilation of Radar observations in COSMO-KENDA

Using the COSMO model of the German Weather Service (DWD) several data assimilation experiments will be performed using radar observations. The KENDA-BACY System (Kilometre-scale ensemble data assimilation, Basic Cycling) provides a framework in which the LETKF can be used.

First, the general behaviour of the system will be analysed. Then the impact of various parameters like localization radius on the predictability of the system. Various fields can be analyzed and plotted to show the quality of the data assimilation and forecast.

Contact: Michael Würsch

Interactive 3D Visualization of High-Resolution Numerical Weather Forecasts

In collaboration between the Computer Graphics and Visualization Group at TU Munich and the Meteorological Institute at LMU Munich this interdisciplinary thesis project aims at visualizing numerical weather prediction data from the high-resolution COSMO model ( in an interactive 3D system.

The Computer Graphics and Visualization Group develops “Met.3D”, a 3D visualization system until now targeted at the visualization of weather predictions from the European Centre of Medium Range Weather Forecasts (ECMWF, Your task will be the integration of COSMO forecasts into the system and the visualization and analysis of a case study. This will require the adaptation of the Met.3D visualization pipeline (data management, visualization algorithms) to the COSMO grid. Challenges will include the adaptation of visualization algorithms (e.g. raycaster) to the special vertical grid structure of COSMO, handling data fields larger than the available GPU memory, and handling the rotated coordinate system of the COSMO grid. Due to the high resolution of the COSMO model, you will create visualizations that will allow fascinating insights into the atmospheric structure over Europe.

The thesis is suited for computer science students interested in meteorology as well as meteorology students interested in computer graphics. However, requirements for the thesis are good programming skills in C++, good knowledge of a graphics API (preferably OpenGL), and interest in the meteorological application. Knowledge of Linux and meteorological data formats (NetCDF and GRIB) would be a plus. In return, we offer an interesting thesis that will enable you to gain insight into the exciting world of computer science meeting atmospheric research.

If you are interested, please contact:

At TU Munich: Marc Rautenhaus, Email:
At LMU Munich: Tobias Selz, Email:

Calculation and interpretation of the convective adjustment time-scale

Many properties of convection and precipitation (like forecast quality, sources of uncertainty, predictability etc.) depend in some way on the prevailing meteorological weather regime. Thus it is desirable to objectively determine the dominant process in a given meteorological situation. Such a measure represents the convective time-scale τ that comprises a physically-based measure to distinguish equilibrium and non-equilibrium convection. Since it has been first suggested by Done et al. (2006) the time-scale has been applied in various applications.

The aim of the project is to investigate several characteristics of the areal and temporal averaging as well as the optimal usage in the context of ensembles. Here, the convection-permitting ensemble prediction system COSMO-DE-EPS will be used.

The project requires basic knowledge of programming (F90, python) and numerical weather prediction models.

Contact: Kirstin Kober, Christian Keil, George Craig

Investigation of precipitation amounts and further development of the Plant-Craig Convection Scheme

In global or coarse grid limited area models convective processes have to be parametrized in order to avoid the onset of unrealistic grid-scale convection. Numerous convection schemes have been developed in the past 40 years. Most of them are deterministic. The recently proposed Plant-Craig scheme is a stochastic parametrization scheme of the convection. Clouds are drawn randomly from a distribution and the closure assumption is only fulfilled on average. The scheme has been implemented into the COSMO model and tested in several case studies at a 7km resolution. These studies have revealed that the amount of precipitation produced by the scheme is in general too low.

The goal of this project is to find out reasons and possible solutions to this problem. An other part of the project will be to further develop the Plant-Craig-scheme, for example to account for a displacement of the clouds by the flow.

The project requires basic knowledge in programming (F90, python) and numerical weather prediction. There is also a related HiWi position.

Contact: Tobias Selz, George Craig

Examination of the performance of a stochastic convective parametrization during different flow conditions

In numerical weather prediction (NWP) models atmospheric processes not resolved by the model grid need to be parameterized. At a horizontal resolution of about 10 km convection comprises such a process. The recently proposed stochastic convective parametrization of Plant and Craig (PCCS, 2008) is implemented in the research version of the COSMO-EU model of DWD. One question is the performance of PCCS during different flow conditions, i.e. compare precipitation epsisodes characterized by either strong or weak synoptic-scale forcing.

The convective timescale can serve as a predictor of the flow regime and has already been computed for the summer 2009. Based on this available classification a handful of case studies ought to be performed with COSMO-EU including PCCS and the model quality in forecasting summertime precipitation shall be examined. The computation will be executed on the local linux-cluster at MIM.

The project requires basic knowledge of programming (F90, python) and numerical weather prediction models.

Contact: Christian Keil, George Craig

Hybrid data assimilation methods

Data assimilation algorithms primarily in use today are based either on variational or ensemble techniques. Recently the algorithms that combine these two methods have been developed and applied for operational use in numerical weather prediction. In this project, the benefits of the combined approach on toy problems will be investigated. Three toy problems will be chosen to mimic different dynamical regimes. Different variants of hybrid algorithms will be evaluated on accuracy as well as ability to preserve physical properties of the system.

The project requires basic knowledge of programming. The project is within HErZ Data Assimilation Branch at MIM.

Contact: Tijana Janjic Pfander

Model vs. observation space localization for ensemble Kalman filter

In order to apply ensemble based Kalman filter methods in practice, localization is additionally applied as part of ensemble Kalaman filter algorithms. Two methods for localization, one in observation space and one in model space are in use. In the work by Campbell et al. 2009 it was shown that the radiance space localization is inferior to the model space localization. Considering only the localization in the vertical, we will explore localization techniques and weighthing functions for satellite data assimilation on similar test problem as described in Campbell et al. 2009.

The project requires basic knowledge of programming (F90). The project is within HErZ Data Assimilation Branch at MIM.

Contact: Tijana Janjic Pfander

Ensemble Generation

In order to improve the forecast of numerical weather prediction models, the proper descriptions of different sources of forecast error is necessary. The goal of this master project is to investigate the impact of combining models with different numerical cores using toy problems. For example, the impact of different advection schemes of each ensemble member on the properties of statistics calculated from ensemble are of interest. Further, the impact of a specific scheme's numerical dissipation, both for space and time discretization on spread will be investigated. For this master thesis, already existing programmes in F90 will need to be modified (moderately).

The project requires basic knowledge of programming (F90). The project is within HErZ Data Assimilation Branch at MIM.

Contact: Tijana Janjic Pfander, Matthias Sommer, Christian Keil

Evaluating observation impact calculations

The knowledge of the impact of different observations on the accuracy of NWP analyses and forecasts is crucial to optimize observing and data assimilation systems. Traditional methods as data denial experiments (i.e. rerunning the model for every possible configuration) can only be conducted occasionally due to their computational costs. For this reason, a new ensemble-based method to estimate the impact of observations following Liu and Kalnay (2008) has been implemented in the pre-operational DWD KENDA-COSMO ensemble data assimilation system in the framework of the HErZ Data Assimilation Branch at MIM.

The task of the master thesis is to evaluate the accuracy of the method for different configurations (different lead time, localization, ensemble size) by conducting experiments with the KENDA-COSMO system. The goal is to find the optimal configuration for such a calculation in the future operational convectional data assimilation system of DWD.

The project requires basic knowledge of programming.

Contact: Martin Weissmann and Matthias Sommer

Bei Interesse an weiteren Masterarbeiten im Lehrstuhl für Theoretische Meteorologie bitte bei Prof. G. Craig, Dr. C. Keil, Dr. K. Kober oder Dr. M. Weissmann nachfragen.

These projects would involve analyses of gridded analysis data and/or observational data and their interpretation.

Contact: Prof. Roger Smith (Room 230)