SPHERA, a new convection‐permitting regional reanalysis over Italy: Improving the description of heavy rainfall

Regional reanalyses allow us to better describe weather patterns related to rapidly evolving high-impact events thanks to substantially finer detailing than global datasets. However, most regional datasets still do not permit the explicit representation of deep convection. SPHERA (High rEsolution ReAnalysis over Italy) is a new high-resolution convection-permitting reanalysis centred over Italy. It covers 26 years (1995-2020), is based on the non-hydrostatic limited-area model COSMO, and is produced by dynamically downscaling the global reanalysis ERA5. A nudging data assimilation scheme steers the model toward observations. The fine horizontal grid spacing of 2.2 km allows us to switch off deep-convection parametrization. This study reports the added value of SPHERA over ERA5 in representing rainfall over Italy, particularly for severe precipitation, using rain-gauge observations during 2003-2017 as reference. Concerning the 95th percentile of spatial rainfall distributions, ERA5 presents dry estimates with biases reaching -12 mm & BULL;day(-1) over mountainous regions. At the same time, the enhanced locally driven effects of SPHERA produce seasonal biases ranging from wet in JJA (up to +12 mm & BULL;day(-1)) to dry in DJF (down to -9 mm & BULL;day(-1)). For daily maximum rates, the regional reanalysis shows better skill in detecting occurred events (with hit rates higher than ERA5 by roughly 0.4 points in the range of 15-80 mm & BULL;day(-1)) and frequency biases closer to 0 at all intensities when coming to daily averages. Similarly, for hourly maximum accumulations, improved adherence to observations is detected for SPHERA at all intensities, conversely to the underprediction of the global driver (with frequency biases < 1 starting from 1.5 mm & BULL;hr(-1)). Additionally, the analyses of two specific events reveal the enhancements of SPHERA in simulating extreme precipitation, with a maximum intensity underestimation on the order of 24% versus the 73% detected for ERA5. Further improvements include the spatial detailing, timing, and temporal evolution of the events.