Exploring the benefits of a Hi-EnKF system to forecast an extreme weather event. The 9th October 2018 catastrophic flash flood in Mallorca.

The afternoon of 9 October 2018 a catastrophic flash flood episode (above 500-y return period) hit the town of Sant Llorenç, in northeastern Mallorca (Spain), leading to 13 casualties and economic losses over 91 M€. Operational models, including HARMONIE-AROME run at the Spanish Meteorological Agency, missed this event and thus no relevant warnings were issued to take preventive actions. Satellite, radar and conventional observations indicate the presence of convective activity along a convergence line, forming during the morning east of Menorca and moving westwards. This convective line is identified as a key factor for the initiation of the moist deep convection responsible for the extreme rain rates associated to this episode. Indeed, the failure of operational numerical models in predicting the flash flood is most likely attributable to their inability to properly simulate the convergence line and the associated convection. To further understand the predictability of this case, we assess multiple different high-resolution ensemble generation strategies, including data assimilation. Results show that among all, only configurations that assimilate reflectivities have the potential to predict significant precipitation amounts over useful lead times in Sant Llorenç basin. In any case, simulated precipitation values are underestimated, and we prove that the predictability of the responsible moist convective systems drops off near the 3–4 h lead times. Numerical predictability challenges of this episode and limitations of the EnKF in these situations are discussed in detail. Overall, the study shows that accurate prediction of this episode is beyond the current prediction capabilities of state-of-the art numerical models and data assimilation techniques. • Potential effect of Hi-EnKF to improve short-range predictability of a non-predicted catastrophic flash flood in Spain. • Assimilation of hourly conventional and 10-min reflectivity radar observations at 900 m grid resolution. • EnKF analysis provide more valuable information than forecasts where no DA is used. • Forecasts initiated more than 3–4 h ahead of the peak of registered intensity do not provide realistic atmospheric evolutions. • Challenges of current DA methods to deal with nonlinearities and non-Gaussian uncertainties. [ABSTRACT FROM AUTHOR]