The 1994 Piedmont flood revisited

Enrico Ferrero (University of Eastern Piedmont), Gianpaolo Balsamo (ECMWF)


In the 1994 Piedmont flood in northwest Italy, 77 people lost their lives and the Piedmont region suffered 14.5 bn dollars in economic losses, making this the second costliest European extreme weather event between 1970 and 2012 as documented by the World Meteorological Organization. Twenty‐ five years on, experts meeting in Italy used modern forecasting systems to reanalyse and re‐forecast the rainfall that caused the event.

Forecasts then...

A third of the precipitation falling in one year in Piedmont was observed in 72 hours between 4 and 6 November 1994. The large‐scale circulation saw an Atlantic trough extending from the British Isles to the Iberian Peninsula and a blocking high over central Europe. The northward flux from the Mediterranean Sea provided moist air that sustained the precipitation over northern Italy.

At the time, ECMWF predicted the event three to four days in advance. In 1994, the Centre’s operational high‐resolution forecast used a grid spacing of about 100 km and 31 vertical levels to describe the atmosphere, while an experimental 32‐member ensemble forecast had a grid spacing of about 300 km.

Flood impact. The Piedmont flood of 1994 caused dozens of casualties and extensive damage. (Photo: Associazione Spazioidea - Alessandria)

...and now

On 6 November 2019, experts in weather forecasting, hydrology and civil protection came together for a workshop in Alessandria, organised by the University of Eastern Piedmont (UPO) and AISAM (Italian Association of Atmospheric Sciences and Meteorology), to analyse the Piedmont flood and the progress made in forecasting extreme events. Thanks to the availability of ECMWF’s new ERA5 climate reanalysis, a set of re‐forecasts for this major flood was produced using different forecasting systems, including:

  • ECMWF’s operational Integrated Forecasting System (IFS Cycle 46r1) at grid spacings of 18 km (ensemble forecasts, ENS) and 9 km (high‐resolution forecasts, HRES)
  • COSMO‐2I‐EPS (Ensemble Prediction System over Italy) at 2.2 km, with boundary conditions provided by ECMWF ENS.

It was found that all modern‐day forecasts made more accurate predictions, and provided better guidance at longer lead times, than ECMWF’s forecasts at the time. For example, a 48‐hour forecast for 4 and 5 November 1994, produced using the COSMO‐2I‐EPS system, predicted accumulated rainfall very close to the observed 200 mm in the southern part of Piedmont and close to the 400 mm observed in the western Alpine and pre‐Alpine areas.

A comparison of global reanalyses (ERA‐Interim and ERA5) and regional reanalyses (such as MERIDA) highlighted the value of increased spatial resolution for a detailed reconstruction of events of this kind. A classification of the most extreme precipitation events leading to destructive flooding in northern Italy identified anomalies in integrated water vapour transport from the surface up to 300 hPa and three associated categories of large‐scale flow as clear percursors of such events.

Total precipitation forecasts. The charts show modern re-forecasts, starting at 00 UTC on 3 November 1994, of the probability that total precipitation on 5 November 1994 will exceed the thresholds indicated, for an ECMWF re-forecast at a grid spacing of 18 km resolution (top row), and for a COSMO-2I-EPS re-forecast at a grid spacing of 2.2 km (bottom row). (Plots courtesy of I. Cerenzia and G. Pincini, Arpae Emilia-Romagna)


In the final discussion, it was concluded that today’s modelling capabilities make it possible to predict cases such as the 1994 Piedmont flood with satisfactory accuracy. Similar events that occurred on 21 October and 23–24 November 2019 caused flooding, but the impact on the population was much reduced thanks to progress in forecasting and in the alert chain involving national civil protection and regional environmental protection agencies. However, especially in the context of global warming, the fact that such events can be extremely localised and can develop over a very short period of time represents new challenges for numerical weather prediction.

A special issue of the Bulletin of Atmospheric Science and Technology will collect scientific contributions from the workshop until September 2020.