Vol. 110, No. 3-4 * Pages 193–445 * July - December 2006

This overview presents a short account on how the theory and practice of the numerical weather prediction developed at the early stage of this discipline. Following a concise review of the work carried out abroad, investigations of the Hungarian meteorologists are described.

The development of the ARPEGE/ALADIN modeling system was initiated in 1990 (by Météo-France). Recently the project encounters 15 partners from Europe and Northern Africa. The main original objective of the cooperation was to develop a numerical weather prediction model for dynamical adaptation, which takes into account all the advantages and constraints coming from its “mother” system ARPEGE/IFS. In a later stage it was natural – based on the inspiration from the ARPEGE/IFS modeling family – to consider the development of all numerical weather prediction related configurations in a single computer code beside the initially established pre-processing (interpolation) and model integration modules. Sophisticated post-processing algorithms were added and then data assimilation procedures were developed (first optimal interpolation and then three-dimensional variational data assimilation). The code has been extended to the tangent linear and adjoint versions, which make possible to apply configurations for sensitivity studies and the computation of singular vectors. The non-hydrostatic version of the model is an essential part of the software: this is the heart of the new AROME model, which is under intensive development for the meso-gamma spatial scales. The article briefly summarizes the most important configurations of the ALADIN model together with some illustration of their practical use at the Hungarian Meteorological Service.

In this paper the results of the sensitivity experiments concerning the impact of using different target domains and target times during the global singular vector computation are presented. The system used is made up of 10+1 member ensembles generated with the global model ARPEGE and downscaled with the limited area model ALADIN. The target domain and target time dependency is studied by using 5 different target domains and 2 different target times. Verification shows that the proper choice of the singular vector target domain and target time can increase the spread and (on average) improves the skill of the ensemble for the Central European area. On the other hand, the studied limited area ensemble system was found not to provide significant additional information with respect to the global one, therefore, the computation of mesoscale initial perturbations for the limited area model might be desirable for a more efficient short-range ensemble system.

The ECMWF/ALADIN system is a limited area ensemble prediction system, which has been developed at the Hungarian Meteorological Service (HMS). The main objective of this limited area ensemble system was to dynamically downscale the ensemble forecasts of the ECMWF/IFS model with the ALADIN limited area model. For the reduction of the computational cost a cluster analysis is performed on the ECMWF ensemble members, and the representative members of the clusters were chosen for providing initial and lateral boundary conditions for the limited area runs. The downscaling system was tested using four different clustering configurations. Preliminary results were obtained by the investigation of four case studies. The subjective evaluation – using stamp diagrams and probability maps – showed that the downscaling system improved the precipitation forecasts of the global EPS system. Objective verification was performed on the basis of Talagrand and ROC diagrams. The Talagrand diagrams showed that the ensemble spread of the downscaled forecasts is not satisfactory, which is a consequence of the loss of information due to the reduced ensemble population. Investigation of the precipitation ROC diagrams confirmed that the best ECMWF/ALADIN EPS configuration improved the forecasts provided by the original ECMWF EPS.

Thunderstorms often cause serious damages due to the strong surface outflow or heavy precipitation. There are some weather patterns, which especially promote breaking out of severe thunderstorms. Radar and visual observations show that some of these thunderstorms can develop into supercell. One of these typical weather patterns is the prefrontal squall line moving from southwest direction (so called Slovenian Squall Lines). In this paper the results about the formation and development of a thunderstorm associated by this type of squall line is presented. Severe thunderstorms formed on May 18, 2005 over the eastern part of Hungary were investigated using radar observations and a mesoscale numerical model (MM5). The time and position of the most intensive thunderstorm coincide well with radar observations.
The case study shows that there is a competition between thunderstorms for the wet and warm air. A thunderstorm which can collect the wet and warm air from larger area will have longer lifetime and more intensive updraft. The case study shows that in the case of absence of directional wind shear, the merging of the updraft cores can result in a supercell. Although the formation of the new cells frequently occurs by splitting, this process did not happen in this case.
Analysis of the numerical simulation indicates the presence of three different types of downdraft regions in an intensive thunderstorm. The low level downdraft was generated by the precipitation loading. The intensity of this downdraft is also affected by melting and evaporation of the precipitation elements. The midlevel downdraft does not reach the surface, and it is driven by negative thermal buoyancy and set by an interaction of the updraft with the vertical wind shear. Downdraft cores at high level could be associated with the overshooting.

Laboratory simulations in water tanks provide an attractive alternative to full-scale field experiments, moreover, they can be utilized to benchmark analytical and numerical calculations. Here we discuss the possibilities and limitations of modeling large scale atmospheric flow in laboratory. As a case study, we describe experiments on quasi two-dimensional mountain wave formation behind obstacles towed through a linearly stratified fluid. Differences between measured wave fields and predictions of linear theories indicate that nonlinear effects are significant in our parameter range. Experiments with a double bell-shaped obstacle revealed that average wave amplitudes at high enough flow velocities are systematically lower than those produced by an isolated obstacle. We attribute this anomaly to the dominance of essential nonlinearities such as strong wave dispersion and resonance effects.

This paper aims to give a detailed description of the three dimensional variational (3DVAR) data assimilation system developed for the Hungarian version of the ALADIN model (ALADIN/HU). The evaluation of the system’s performance will be given through different kind of verification results, and the most important developments related to the design of the assimilation cycle and the background error covariance modeling will be presented. Recently, after a long period of preliminary testing, the ALADIN/HU 3DVAR system has become an operational application at the Hungarian Meteorological Service, which makes possible to take the benefit of local high resolution observations while providing the initial conditions for the production forecast. The evaluation of the system is based on comparisons with a former operational version of the ALADIN/HU model, in which the production forecast was simply initialized by an appropriate interpolation of the analysis provided by the ARPEGE global model.

The Hungarian Meteorological Service (HMS) has contributed to the development of a high resolution limited area model (LAM) in the frame of the ALADIN project since its beginning (1990). The development of the data assimilation system started at the HMS in the year 2000 with the implementation of the three-dimensional variational (3D-Var) analysis scheme. Our research aims to design an optimal assimilation system suitable for LAM application, including various high resolution observations. This paper describes the configuration of the analysis and forecast systems used in our studies. Results of the incorporation of the AMSU-A and AMSU-B data in full resolution – one-by-one field of view (FOV) – are presented. Studies regarding the efficient radiance-bias correction were necessary to get improvement from the Advanced Microwave Sounder Units A and B (AMSU-A and AMSU-B) of the Advanced TIROS Operational Vertical Sounder (ATOVS) radiances in our LAM analysis and forecast systems. Small, but positive impact of the high resolution ATOVS radiances on our analysis and short-range forecasts was obtained, which leaded to their operational implementation.

An Eulerian photochemical reaction–transport model and a detailed dry deposition model have been coupled to describe both continuous air pollution and accidental release over Central Europe. Up to now, model applications have been carried out for estimating ozone flux over Hungary and transport of passive tracers from a point source. Simulating the chemical reactions, the simple GRS (Generic Reaction Set) chemical scheme was used, although, the model allows the utilization of any other, more comprehensive reaction scheme. During the transmission processes of radioactive tracers, only radioactive decay has been considered. Because of detailed parameterization of deposition processes, not only the concentration, but the effective ozone load can also be estimated by the model. Meteorological data utilized in the model have been obtained by the ALADIN meso-scale limited area numerical weather prediction model used by the Hungarian Meteorological Service. Detailed model description is presented in this study. Model sensitivity tests and some results will be presented in a companion paper.

A detailed description of a coupled transport–deposition model has been given in the accompanying paper in this issue (Lagzi et al., 2006). Sensitivity analysis of this model and some applications are presented in this study. Within the framework of sensitivity analysis, the effects of input data on model results have been examined. Some case studies of model applications are also presented here. Using our model, the impact of both short term accidental releases and continuous emissions of air pollutants can be estimated. An example of long-range transport processes resulting from an accidental release from a single concentrated emission source (nuclear power plant (NPP) at Paks, Hungary) is discussed. Another application of the model is the prediction of secondary pollutant loading resulting from the continuous release of pollutant precursors. Estimations of photochemical air pollution and ozone fluxes were performed on a regular grid over Hungary for the first time. Time and space resolutions of the transport–deposition model correspond to the ALADIN meso-scale limited area numerical weather prediction model used by the Hungarian Meteorological Service. Accordingly, the meteorological data utilized in the model were generated by the ALADIN model, which allows further routine model applications. The model simulations show that the predicted regions of high stomatal ozone flux (the effective ozone load) can be very different to predicted regions of high AOT 40 (accumulated ozone exposure over a threshold of 40 ppb) values depending on the weather and soil conditions. The predicted ozone deposition velocities over various vegetation types are shown to be highly sensitive to a range of meteorological parameters for summer, sunny conditions which affects the flux of ozone from the atmosphere to the surface.

In the modeling of real-life complex time-dependent phenomena, the simultaneous effect of several different sub-processes has to be described. The operators describing the sub-processes are as a rule simpler than the whole spatial differential operator.
Operator splitting is a widely used procedure in the numerical solution of such problems. The point in operator splitting is the replacement of the original model with one in which appropriately chosen groups of the sub-processes, described by the model, take place successively in time. This de-coupling procedure allows us to solve a few simpler problems instead of the whole one.
In this paper several splitting methods are constructed (sequential splitting, Strang-Marchuk splitting, weighted splitting, additive splitting, iterated splitting) and analyzed. Application of the operator splitting method to real-life problems is investigated, with great emphasis on long-range air pollution transport. The accuracy (local splitting error) of the methods is discussed, and the main advantages and drawbacks of this approach are listed.

Operator splitting procedure is a widely used approach for modeling physical processes. Of the numerical solving process both accuracy and fast integration are required. These requirements, however, usually contradict. In the present paper our investigations are presented regarding the optimization of the combined effect of the splitting and the numerical method. Their interaction is examined both analytically and numerically in the total error of the solution, and an idea is presented how to control the accuracy while taking reasonable computing time. Furthermore, an example is shown how to optimize the application of splitting procedure in air pollution transport models.

Operator semigroups are widely used for proving well-posedness of partial differential equations and for investigating qualitative properties of the solutions. Here we give a short overview for the reader to get familiar with these objects and to get an insight into their applications.

In this paper the discontinuous Galerkin method is presented, which can be used for the numerical approximation of the solution of partial differential equations arising in the atmospheric modeling. An overview of this recent approach is presented comparing it with other numerical methods, especially with finite element techniques. The implementation of the appropriate numerical procedure is discussed for a diffusion operator of second order in details, which includes also the case of the advection operator.