Improving forecast skill of a convective storm using ICON hectometric-scale ensembles

Extreme precipitation events in the Mediterranean basin represent a growing societal challenge, as operational numerical weather prediction systems still struggle to accurately forecast their intensity and location. This study investigates the potential of hectometric-scale ensemble modelling to improve the forecast skill of a devastating convective storm that struck the Marche region of central Italy on 15 September 2022, causing 13 fatalities and over 2 billion euros in damage. Using the ICON model with double nesting down to 550 m horizontal resolution, we perform sensitivity tests and ensemble simulations initialized from ECMWF IFS data, comparing two turbulence closure schemes: the operational one-dimensional COSMO-type TKE scheme and a three-dimensional Smagorinsky closure. Our results show that hectometric-scale resolution, combined with the Smagorinsky turbulence scheme, substantially improves the spatial distribution and intensity of simulated precipitation compared to coarser operational configurations. Ensemble simulations further reveal a strong sensitivity to initial and boundary conditions, with individual members showing markedly different precipitation patterns despite identical model setup. Verification using Overlap Ratio and SAL scores confirms that high-resolution ensembles, particularly those employing the Smagorinsky closure, achieve significantly better spatial agreement with observations, especially for extreme precipitation thresholds. These findings suggest that hectometric-scale ensemble modelling, coupled with three-dimensional turbulence representation, offers a promising pathway toward improved forecasting of high-impact convective events in complex orographic environments.