Röpnack, Andreas: Bayesian model verification : Predictability of convective conditions based on EPS forecasts and observations. - Bonn, 2014. - Dissertation, Rheinische Friedrich-Wilhelms-Universität Bonn.
Online-Ausgabe in bonndoc: https://nbn-resolving.org/urn:nbn:de:hbz:5n-35687
@phdthesis{handle:20.500.11811/6074,
urn: https://nbn-resolving.org/urn:nbn:de:hbz:5n-35687,
author = {{Andreas Röpnack}},
title = {Bayesian model verification : Predictability of convective conditions based on EPS forecasts and observations},
school = {Rheinische Friedrich-Wilhelms-Universität Bonn},
year = 2014,
month = may,

volume = 65,
note = {Forecasts of convective precipitation have significant uncertainties. Among the main reasons for these uncertainties are the non-linear dynamics of the atmosphere and approximations within the equations of the numerical weather prediction (NWP) models by unresolved physical processes, which have to be parametrized, and imperfect simulation of resolved physical processes. To account for the forecast uncertainties of convection permitting models, a convection permitting ensemble prediction system (EPS) based on the consortium for small-scale modeling (COSMO) model with a horizontal resolution of 2.8 km covering whole Germany is being developed by the Deutscher Wetterdienst (DWD). The deterministic model is named COSMO-DE.
The potential of convective instability is affected by the vertical structures of temperature and humidity. These vertical profiles of the COSMO-DE-EPS and further ensembles will be investigated in this work. For verification of the vertical model profiles radiosonde observations are used. However, the observations are uncertain by themselves due to the well-known limits in observing the atmosphere.
The focus is to present a probabilistic method to verify and compare ensembles. The approach considers explicitly the observation error as well as the model uncertainty to validate multidimensional state vectors of temperature and equivalent potential temperature profiles of the COSMO-DE-EPS and of two meso-scale ensembles with horizontal resolution of 10 km and parametrized convection. The meso-scale ensembles are the COSMO short-range EPS (COSMO-SREPS) and the COSMO limited-area EPS (COSMO-LEPS).
The approach is based on Bayesian statistics and allows for both verification and comparison of ensembles. Both investigated variables define the dry and moist static stability of the atmosphere, and therefore they determine the necessary conditions of convection. The equivalent potential temperature contains the effect of the humidity, which cannot directly investigated, because the humidity is non-Gaussian. Since the temperature and equivalent potential temperature can be assumed to be Gaussian distributed, the Bayesian approach is solved analytically. Finally, the probabilistic approach gives an "evidence" for the ensemble under investigation in relation to a reference ensemble. This evidence is classified depending on the application either comparison or verification of ensembles.
The investigation period comprises the August 2007 for a comparison of the COSMO-DE-EPS with the COSMO-SREPS and the entire convective and orographically-induced precipitation study (COPS) period 2007 for a verification of the COSMO-SREPS and COSMO-LEPS against COSMO-EU analyses. It is shown that the temperature profiles modeled by the COSMO-DE-EPS have a higher evidence in view of the observations than those of the COSMO-SREPS. Furthermore, the evidence for the equivalent potential temperature is weaker due to the larger uncertainty of this variable in the model as well as in the observed state. This shows the importance of the observation uncertainty. Nevertheless, it seems that the COSMO-DE-EPS as a short range convection permitting ensemble is a suitable approach to consider the uncertainties in forecasting convection. The verification of two meso-scale ensembles COSMO-SREPS and COSMO-LEPS show a linear decrease of the probability (evidence) of the vertical temperature and equivalent potential temperature structure with increased forecast lead time. Furthermore, it is shown that the predictability of the convective conditions are up to 5 days. However, it is to consider that the typical time scale of convection is about hours, and beyond it is difficult to predict convection.
As a general result, the statistical model described in this study is appropriate to compare ensemble systems with each other. The score proposed in this work is a generalization of the Ignorance score taking additionally into account the uncertainty of the observations as well as the spatial correlation structure of the verified forecasts. The approach based on Bayesian statistics allows for a comprehensive evaluation of the forecast quality of three-dimensional samples by using just one score.},

url = {https://hdl.handle.net/20.500.11811/6074}
}

The following license files are associated with this item:

InCopyright