Improving predictability of Mesoscale Extreme Precipitation events with Convection-permitting models

Extreme precipitation events are both changing in frequency of occurrence and intensity. Among them, we can identify the Mesoscale Convective Systems (MCSs), which have typical dimensions of the order of ~100 km and consist of large clusters of powerful thunderstorms that produce intense rainfall, large hailstones and, occasionally, dangerous tornadoes. Italy is being affected more and more frequently by such systems.

Here, we carried out sensitivity tests using a limited-area setup at 2km horizontal resolution of the ICON numerical model, combined to a double nesting at 1km and 500m to simulate a strong quasi-linear convective system that triggered a huge flooding in central Italy on 15-16 September 2022, causing 13 deaths and 2 billion euros worth of damage. The convergence of warm, humid southwesterly winds (that produced an atmospheric river with peak values of 1000 kg*m^-1*s^-1) with northwestern atmospheric currents triggered intense convection over the central Apennine regions leading to a destructive V-shaped self-regenerating thunderstorm, despite the absence of very cold air at higher levels. The numerical models underestimated such event and did not reproduce well the spatial extent and intensity of the system. These discrepancies are related to model errors usually due to uncertainties in (i) estimation of moisture in the several layers of the atmosphere, (ii) parameterization schemes for cloud microphysics, (iii) orography representation and (iv) turbulence diffusion. It follows that convection-permitting models are crucial to better solve these phenomena. Thus, we did some sensitivity tests operating on the above-mentioned parameterizations to identify the main physical processes that are the most likely to impact the forecasts’ accuracy.

Results showed that the activation of the Smagorinsky 3D turbulence scheme, compared to the COSMO 1D scheme, noticeably improves the forecast accuracy in terms of  precipitation amounts and spatial distribution. Moreover, we employed an ensemble-based approach to enhance predictability, with a focus on extreme percentiles within the ensemble tail of specific variables such as total precipitation. This method offers more valuable insights compared to the traditional analysis of ensemble means, especially when  it comes to rare and unusual weather patterns that lead to extreme precipitation events.