On the use of radar and aircraft data in Ensemble Data Assimilation of convection for non-hydrostatic numerical weather prediction

The application of Ensemble Data Assimilation (DA) is a method to produce initial states (so-called analyses) for numerical forecasts of thunderstorms. By the means of radar observations of reflectivity and radial winds, the complex inner structure of a convective cell or a mesoscale convective system can be analyzed. At the same time, thunderstorms are embedded in their surrounding air mass and interact dynamically with this convective environment. This thesis aims to answer some questions which link the observational aspect of convective dynamics to the forecasting aspects of the numerical model, and to the practical and intrinsic predictability when applying convective scale DA. Since the atmosphere is a chaotic system, errors in the initial states grow during the forecast and thus limit the predictability. Errors that are small in amplitude and spatial scale intensify with a faster rate than large errors, so that an initial state with relatively large errors may produce, after a few hours, an equivalently good thunderstorm forecast compared to an initial state with smaller errors. To test this hypothesis, simulated radar observations of wind and reflectivity were assimilated into the convection-permitting COSMO model with a resolution of 2~km, by the means of an Ensemble Kalman Filter (EnKF), and under the assumption of a perfect forecast model. The spatial and temporal resolution of the observations was coarsened from 2 to 8~km and from 5 to 20~minutes, and their spatial influence was increased from 8 to 32~km in order to produce analyses that contained less detailed information about the observed storms. After three hours, the resulting forecasts were comparable to such experiments in which the observations had been assimilated with the highest resolution and a close fit in the analysis had been reached. Thus, a recommendable setting for the assimilation of real radar data was found that respects the limited predictability of convection. Due to the input of high resolution data, noise was introduced into the model that disturbed the dynamics and triggered spurious convection in wrong locations. The study showed that this noise consisted of internal gravity waves which caused an over-abundance of variability in the vertical velocity field, both in the storm vicinity and in the environment that was at least 32~km distant to the storms. This noise disturbed the internal dynamics of the convective systems in such a way that the emergence of new updrafts became decoupled from the underlying pools of cold air. This was quantified by applying a spatio-temporal correlation method. In addition to the DA-aspects of mainly storm-interior dynamics, aircraft observations in clear air regions were assimilated into the pre-operational Kilometre-scale Ensemble Data Assimilation (KENDA) system of the German Weather Service (DWD) that couples an EnKF to the COSMO-DE model over Germany with a resolution of 2.8~km. These observations are collected by Mode-S air traffic control radars and contain measurements of wind and temperature along the flight tracks of commercial aircrafts. Their data density is ten times larger than the previously assimilated Aircraft Meteorological Data Relay (AMDAR) observations. Their assimilation was successfully implemented and it was shown that forecasts of three hours lead time benefit from the additional data by exhibiting a smaller error in the tropospheric wind and temperature profiles which, together with humidity, define the convective environment. Conclusions were drawn on how KENDA reacts to the large increase of observation data in terms of ensemble spread and other EnKF parameters, and recommendations on the operational use of Mode-S data were derived. The point-like updraft of a convective storm is coupled to the environment by a divergent circulation that has a wider scale than the core. By assimilating real radar observations in the newly implemented framework of the nested COSMO-MUC-KENDA system, the influence of reflectivity and radial wind observations was investigated. A conceptual model was introduced that relates the scales that are influenced by the assimilation of these two observation types. By applying the noise measures that were developed earlier in the thesis, it was shown that the assimilation of radial wind observations has a damping effect on the noise that is caused by the DA. Reflectivity observations on the other hand are helpful to improve the storm positions for the first tens of minutes into the forecast. It was shown that the EnKF assimilation of radar data produces better forecasts and less noise than the previously used nudging scheme. Combining the assimilation Mode-S data and radar observations with the methods that were developed in this thesis is a promising way to improve analyses and forecasts of severe and damaging convection with the KENDA system.