Exploring extreme floods in two Italian watersheds through unseen ensemble scenarios

Panàro and Reno are two rivers (165 and 212km long, respectively) having adjacent mid-sized watersheds (covering areas of 1775 and 4828km², respectively) in N-Italy. Given their spatial proximity they are expected to behave similarly, particularly in response to meteorological forcing leading to extreme floods. Three different river discharge (RD) datasets are analyzed: 1) the historical daily-mean RD observed timeseries by ARPAE in the stations of Bomporto (Panàro) and Casalecchio (Reno); 2) the EFAS5 historical simulations of daily-mean RD, computed using the LISFLOOD hydrological model forced with the EMO1 dataset (1992-2023); and 3) the EFAS seasonal-reforecast daily-mean RD dataset, computed by LISFLOOD forced by the 25-member SEAS5 seasonal-reforecast ensemble (2000-2023).  

The UNSEEN approach (UNprecedented Simulated Extremes using ENsembles) uses many simulations to identify “plausible” extremes. Building upon this approach, an ensemble of 100 “surrogate” RD timeseries is built by concatenating 3-month blocks from the original 25-SEAS5-EFAS members, but starting from four different initial dates (beginning of April, May, June, and July). To ensure statistical independence, the first four months of each timeseries are discarded.  
Watershed are compared by inspecting  quantiles, interannual trends, and annual cycles. It is found that -without any bias correction- the 100 surrogate timeseries at the Bomporto can produce extremes RD higher than those seen in the historical EFAS dataset: conversely, this is not observed for the Casalecchio. The surrogate series show a mostly bell-shaped distribution centered around zero for the slope of the interannual trend, while both the historical EFAS timeseries and the ARPAE observations show a positive trend in recent years. Finally, with respect to the annual cycle, autumnal extremes appear to be underestimated in the surrogate time series, particularly at Casalecchio. 

Possible interpretations for such behaviors are explored by examining the meteorological conditions associated with the most extreme floods, derived from ERA5 for the historical EFAS and from the corresponding simulations of SEAS5 for the surrogate. We find that the meteorological conditions leading to floods in Bomporto and Casalecchio stations (only 33 km apart) are different. For instance, the most extreme events at Bomporto are typically associated with deep trough over the Mediterranean or a cyclone over Central-Italy, while those at Casalecchio seems to be linked to a more zonal flow and a stronger Atlantic depression. Additionally, the frequency of synoptic configurations associated with extreme RD is studied in both the historical and surrogate datasets, to assess whether the surrogate simulations reveal unseen patterns not present in the historical record. We suggest the UNSEEN approach can be a powerful tool to compare the amplitude of climate signals to internal climate variability and highlight the differences in the characteristics of floods in nearby catchments.