A convection-permitting Ensemble Prediction System to study the predictability of Mediterranean Heavy Precipitating Events

Heavy Precipitating Events (HPEs) frequently occur during fall in the Mediterranean region, causing disastrous flash-floods responsible for many casualties and heavy damages in past decades. Some of these HPEs can be attributed to back-building quasi-stationary Mesoscale Convective Systems (MCSs), producing large rainfall totals in a very short time (typically over 200 mm in 6 to 24 h).

This study considers the design of a short-range, convection-permitting Ensemble Prediction System (EPS) to quantify the uncertainties associated with convection-permitting forecasts. Specifically, it is used to assess the predictability of Mediterranean HPEs at short range and fine scale. The non-hydrostatic, high resolution Météo-France AROME model is used over a domain mainly covering France. AROME has a 2,5 km horizontal grid spacing, a bulk microphysics scheme governing six prognostic water variables, and has its own 3D-VAR data assimilation scheme.

The three major sources of uncertainty for a limited-area model are the uncertainty on Lateral Boundary Conditions (LBCs), on Initial Conditions (ICs), and model errors. The relative impact of the uncertainty on LBCs and ICs on probabilistic forecasts was examined over a 31-day period in October-November 2008 and on two specific HPE case-studies. To sample the uncertainty on the LBCs, the members from a global EPS at lower resolution are used to provide different LBCs for each AROME forecast. To sample the uncertainty on the meso-scale ICs, an ensemble data assimilation technique is used. All observations are randomly perturbed according to their observational error before being assimilated by the AROME 3D-VAR scheme, thus producing different ICs for each member.

It was shown that uncertainties on LBCs and ICs have an impact at different forecast ranges, with the effect of different ICs being rapidly overcome by the LBCs. Noticeable differences were also found in the geographical localization and extent of ensemble spread when considering each uncertainty source separately.

Ongoing research assesses the following two key questions regarding the uncertainty on LBCs and ICs.

Global EPSs tend to have more and more members. However, due to the high computing time cost of convection-permitting forecasts, we cannot afford to downscale each global forecast with the AROME model. We have to perform a selection of a few representative members of the global EPS to drive the AROME forecasts. A dedicated clustering technique is described and evaluated against a random selection of the large-scale ensemble members.

Against the random perturbation, a selective perturbation of the observations is evaluated, which focuses on low-level parameters known to have a strong impact on both the location and intensity of the precipitating systems. The sensitivity to ICs is also further studied by the generation and assimilation of virtual observations over the sea.

This work aims at the development of an experimental high resolution Ensemble Prediction System before the special observational phases of the HyMeX project in 2012-2013 in the western Mediterranean.