Italian floods in May 2019 proved hard to predict

Ervin Zsoter
Christel Prudhomme
Corentin Carton De Wiart


According to ECMWF’s ERA5 reanalysis, in May 2019 parts of Italy experienced at least two to three times the average rainfall for that month. There were severe flash floods and riverine floods in many areas, associated with an increased landslide risk in some places. One of the worst-hit areas was the Emilia-Romagna region, which experienced severe floods around 13 May. In the upstream areas of the Savio and Marecchia rivers near San Marino, in two days up to 200 mm of rain was observed. Forecasts issued by the European Flood Awareness System (EFAS) for this short-lived event involving relatively small catchments proved challenging in the medium range, and even short-range forecasts were of uneven quality. EFAS is a component of the Copernicus Emergency Management Service for which ECMWF is the computational centre. As requested by the EFAS partners, work is under way to improve flood forecast performance for similar events.

Forecasts of the event

EFAS forecasts showed little sign of a risk of substantial floods near San Marino in the medium range. This began to change a few days ahead of the event, and by one day before there was a strong signal in three of the four hydro-meteorological forecast chains used in EFAS. As shown in the figure, forecasts driven by the COSMO-LEPS limited-area model, the German DWD model and ECMWF’s deterministic high-resolution forecast (HRES) showed a strong hydrological signal for the worst-impacted area west and northwest of San Marino. This can be attributed to good precipitation forecasts and the associated river response in EFAS. Higher resolution seems to have helped in this case as forecasts driven by ECMWF’s lower-resolution ensemble forecast (ENS) did not show the same severity in the flood signal despite indicating a probability of up to 40% of 10‑day precipitation exceeding 150 mm in the area.

Redefining reporting points

When EFAS detects a consistently high flood risk based on a combination of the information provided by the different forecast chains, it defines reporting points where flood forecasters can investigate in detail the shape of the flood hydrograph. One of the issues with this event was that no reporting point was shown in the worst-hit area. There are several reasons for this: high uncertainty in the meteorological forcing; a large variation in forecasts from one forecast run to the next; and the short duration of the event. In addition, rules to identify reporting points are currently slightly less restrictive for river points where partners provide data, which is the case in some of the areas where reporting points are shown in the figure, but not in the San Marino area. The introduction of fixed reporting points for certain river basins, where flood hydrographs are always produced with available hydrological data, should help in the future with cases such as the San Marino floods. Such an upgrade was tested in EFAS during the summer of 2019 and was implemented on 8 October. The way reporting points are identified and shown could also be revised, for example with more flexible rules regarding forecast uncertainties. However, the multiplication of reporting points associated with lower probabilities of flooding might make the interpretation of the forecasts more difficult.

EFAS forecasts starting at 00 UTC on 12 May 2019. The top‑left panel shows the COSMO-LEPS forecast of the probability of total precipitation over 5 days exceeding 150 mm and the probability of river discharge exceeding the 20-year return period level. The top‑right panel shows the corresponding ENS forecasts but for 10‑day precipitation totals. The bottom left panel shows the DWD total precipitation forecast (7 days) and predicted return periods. The bottom-right panel shows the corresponding HRES forecasts but for 10‑day precipitation totals. Red dots and yellow triangles denote five- and two‑year return period river flood reporting points, respectively. Larger triangles show flash flood reporting points.

Improved time stepping and resolution

The fact that ENS-based flood forecasts use a daily time step with daily meteorological forcing leads to a smoothing of flood peaks in small, fast-responding catchments, such as in this event. Furthermore, due to the spatial resolution of the hydrological model in EFAS (25 km2 per pixel), forecast skill tends to be lower for small catchments, such as the Savio and Marecchia catchments (approximately 1,000 km2). There are plans to switch the ENS‑driven EFAS forecasts to a 6‑hourly time step. To this end, a hydrological model calibration using a 6‑hourly time step is currently being developed and will be released in spring 2020. This is expected to improve forecast skill for fast-responding catchments. Finally, preparations have begun to increase the spatial resolution of the hydrological model to enable the production of higher-resolution EFAS forecasts in the coming years.