*4.1. Weather Models*

WRF model v3.8.1 [43] was used for the numerical simulations. It is a next-generation mesoscale numerical weather prediction system designed to serve both operational forecasting and atmospheric research needs that was developed at the National Center for Atmospheric Research (NCAR) in collaboration with several institutes and universities for operational weather forecasting and atmospheric science research. The model represents the atmosphere with a fully compressible non-hydrostatic set of equations, which is discretized over a staggered Arakawa C-grid. The WRF simulations are produced in two phases: the first configures the model domains, ingests the input data, and prepares the initial conditions, and the second runs the forecast model, which is performed by the forecast component that contains the dynamics solver and physics packages for atmospheric processes (e.g., microphysics, radiation, and the planetary boundary layer). The forecast model components operate within the WRF's software framework V3.8.1 from National

Center for Atmospheric Research (Boulder, Colorado), which handles input/output (I/O) and parallel-computing communications. WRF is written primarily in Fortran, can be built with several compilers, and runs predominately on platforms with UNIX-like operating systems, from laptops to supercomputers. The WRF model is applied extensively under both real-data and idealized configurations for research activity, but it is also used operationally at governmental centers around the world, as well as by private companies. Three domains with a grid spacing of 22.5 km, 7.5 km, and 2.5 km and parametrization schemes used for the WRF operational chain at the CIMA Foundation [44–46] were adopted for the numerical experiments of this work.

PhaSt is a spectral-based nowcasting procedure based on the precipitation fields provided by radar measurements and the stochastic evolution of the transformed fields in spectral space [47]. In the framework of the SINOPTICA project, the PhaSt algorithm has been applied to Vertically Integrated Liquid (VIL) fields, a precursor of convective activity [48]. The algorithm takes an empirical nonlinear transformation of the two precipitation fields used as initial conditions. The method uses the Fourier transform of the two Gaussianized initial fields and their Fourier spectra to obtain the Fourier phase for each wavenumber. The latter is then evolved in time by a stochastic process, while Fourier amplitudes are kept fixed. A Langevin-type model is used to evolve the Fourier phases and to generate a nowcasted Gaussian field. For the SINOPTICA project, the PhaSt algorithm is applied in a sort of Rapid Update Cycle (RUC). For this purpose, three different approaches are evaluated:


The statistical analysis performed in terms of correlation coefficients and Continuous Ranked Probability Score (CRPS), as well as through an object comparison between observed and forecasted cell clusters, proves that the restart every 10 min approach provides the best results; thus, it was decided to apply this approach to all case studies.

RaNDeVIL is a new method for identifying 2D convective cells with the potential to produce nowcasts for severe weather. It is based on previous techniques but considers the Density of the Vertically Integrated Liquid (DVIL) instead of reflectivity fields. By considering only one level instead of several in volumetric data, the new method has the advantage of greater speed, an important capability when air traffic controllers have to make decisions in a very short period of less than a few minutes. Considering that thunderstorms can develop close to airports and the TMA, the algorithm is a valuable tool that can support ATCs in organizing approaching traffic under severe weather conditions [26].

The algorithm can effectively approximate future thunderstorm dynamics [49], with particularly good performance for predictions up to 30 min. Distance differences of less than 4 km between the centres of the observed and predicted areas were observed for more than 85% of cells in the first 5 min nowcasts and between 2 km and 16 km for most 30 min nowcasts. Beyond this time, the results are more unreliable but can still provide valuable information to end users regarding storm propagation direction and possible impacts in future scenarios.
