The Representation of the Extratropical Tropopause in the ECMWF Model Evaluated by Lidar and Radiosonde Observations

Kurzfassung

Accurate initial conditions in numerical weather prediction (NWP) models are a fundamental prerequisite for reliable weather forecasts. A key to further improving weather forecasts is to remove systematic errors (i.e., biases) in the initial conditions, which requires a robust identification of these biases and their causes in NWP models. This work focuses on biases at the extratropical tropopause and its surrounding layer, the Upper Troposphere and Lower Stratosphere (UTLS). This region is characterised by sharp vertical gradients in temperature, wind and trace species such as water vapor. whose distribution at the tropopause impact the weather evolution in the midlatitudes. Meteorological profile observations with high precision and vertical resolution are required that sufficiently capture the sharp temperature, wind and humidity gradients at the tropopause. However, the global observing network largely lacks such observations. In this thesis, campaign observations that fulfil these requirements are used to investigate systematic biases in initial conditions in the leading global NWP model, the Integrated Forecast System (IFS) of the European Centre for Medium-Range Weather Forecasts (ECMWF), from a new perspective. In the first part, highly-resolved and accurate airborne water vapor profile observations measured with a differential absorption lidar (DIAL) are used to study the lower-stratospheric (LS) moist bias in the ECMWF’s global reanalysis ERA5. 41 flights provide 33,000 water vapor profiles over more than 200,000 km flight distance with high data coverage across the UTLS covering different seasons and synoptic situations representative of midlatitude weather. This unique dataset is suitable to address open questions regarding the LS moist bias in ERA5, in particular its vertical structure, its seasonal and synoptic variability and potential processes leading to this bias. The statistical comparison of the DIAL with the ERA5 humidity data shows a strong LS moist bias in the model with a maximum of 55 % at 1.3 km altitude above the tropopause. In addition, the analysis reveals the evidence for a decrease of this bias towards 4 km altitude in the LS. The LS moist bias is also found to be stronger in summer and covers a deeper layer at low tropopause heights associated with trough situations. Another novelty is that, for the first time, a connection of LS moist bias to mixing processes is investigated. Collocated ozone and water vapor DIAL data are used to classify the UTLS into tropospheric, stratospheric, and mixed air. The moist bias turned out to be especially strong in the mixing layer while it is weak in the tropospheric and stratospheric air. This highlights that overestimated transport processes across the tropopause in the model play a crucial role for the formation of the bias. One particular mixing process that transports large amounts of water vapor into the LS is overshooting convection. To test the sensitivity of the LS moist bias to the representation of convection in the IFS, DIAL data of a research flight with observed high convective activity are used to evaluate model output of two IFS versions with different convection parameterization. It is found that the IFS version with an overshooting convection limitation exhibits a reduced LS moist bias. In the second part, data assimilation output is used to evaluate UTLS temperature and wind biases in the IFS and their impact on the tropopause representation in this model. Ideally, data assimilation should act to reduce biases in the background forecast and improve the analysis, however, the influence of data assimilation on the sharp vertical gradients near tropopause was an open question. This study investigates the influence of assimilated radiosondes on tropopause sharpness and altitude based on 9,729 profiles observed over North America, the Atlantic, and Europe in a one-month autumn period during the North Atlantic Waveguide Downstream Impact Experiment (NAWDEX) campaign. The observed temperature and wind profiles are compared with their model equivalents, i.e. the background and the analysis at the position and time of the observations (observation-space). The background overestimates (underestimates) temperature at the tropopause (in the LS) leading to a smoother change in static stability (N²), and thus to a less sharp tropopause. Additionally, the background underestimates wind speed across the entire UTLS, especially near the UT wind maximum. Data assimilation improves temperature and wind in the analysis and acts to sharpen temperature and wind gradients near the tropopause. This influence on temperature (wind) is stronger for sharper tropopauses (strong wind situations) which are also characterised by high background errors. In addition, a positive influence on the tropopause altitude is found. 500 non-operational NAWDEX radiosondes are used in a special observing system experiment (OSE) that comprises two independent forecast runs, one with and one without assimilation of the 500 NAWDEX radiosondes. The OSE evaluation provides evidence that the tropopause sharpening and the improved wind profile can be mainly attributed to the assimilated radiosondes. An additional evaluation in model-space shows a similar average influence of data assimilation on temperature and wind beyond the observation location. Furthermore, a stronger influence on temperature and wind is indicated in regions associated with enhanced diabatic activity, such as the warm conveyor belt outflow. This dissertation demonstrates how independent and assimilated campaign data can be effectively used to approach model biases at the tropopause from a novel perspective supporting progressive development of current NWP models. Findings on the LS moist bias in the IFS, obtained with the unique DIAL data, may contribute to reduce this error (and the associated temperature bias) in future IFS versions for instance by adjusting the representation of mixing processes, such as overshooting convection. The tropopause-based evaluation of data assimilation output not only provides new insights into the influence of assimilated radiosondes on temperature and wind at the tropopause in the IFS, but also paves the way for the application of these diagnostics to study observation impact and biases in other atmospheric phenomena relevant to NWP in the future.