The statistical verification has been carried out at the Department of Physics of the University of Genoa (DIFI) in collaboration with the Meteo-Hydrological Centre of Liguria Region (CMIRL), in cooperation with the co-authors.

The outputs of the following models, whose products were made available to the MAP participants during the field phase, were considered: BOLAM, Swiss Model, MC2, ALADIN, and Lokal Modell.

BOLAM: it was developed at ISAO-CNR in Bologna. Two different runs were considered, performed on a daily basis during the SOP at DIFI: a 21 km resolution version using the ECMWF analysis/forecast as initial/boundary conditions, and a self nested 6.5 km resolution version.

Swiss Model: it is the operational model developed by MeteoSwiss in collaboration with DWD. During the SOP it computed twice a day a weather forecast up to 48 hours with a resolution of 14 km, and 31 vertical layers. It was driven by the Europa Model, which was nested into the GM of DWD.

MC2: it is a Canadian non-hydrostatic model which was run at two different resolutions (14 and 3 km) to provide guidance and fine tuning of the flight missions during the SOP. The MC2-14 km made a forecast of 30 hours daily (started at 18 UTC) with the initial and lateral boundary conditions from SM. A self-nested MC2-3km run was then executed for a 27-hour forecast (started at 21 UTC). The topography in MC2-3km was time-dependent (relaxed gradually from a resolution of 14 km to 3 km) within the first integration hour, in order to limit the fictitious vertical motion associated with the movement of vertical coordinate system to 10 cm/s. Both runs had 30 vertical levels, and we considered the high resolution run.

Lokal Modell: it is a non-hydrostatic model based on the full compressible basic equations. It has an horizontal resolution of 7 km and 35 vertical layers, and it is nested within GME. The LM daily performed two runs, initialised at 00 and at 12 UTC.

ALADIN: the "Aire Limitée Adaptation dynamique Développement InterNational" is a hydrostatic model, driven by the GCM ARPEGE. The version we considered was daily run at 00 UTC, with a mesh size of 12 km and 31 vertical levels, giving a forecast up to 48 hours.

We evaluated some statistical indices derived from the contingency table, like bias, false alarm rate and threat score, and other scores, like the mean error, the rmse and the standard deviation, for some surface parameters: temperature, wind and humidity, but particular attention was paid to the precipitation.

Detailed information about the performances of the models is obtained, by means of the most recent data set available at MDC for North Italy (ITAMAP v.2.0), with an increased number of data, and which was subject to a simple validation.