Two storm forecasts with very different skill

Linus Magnusson
Ivan Tsonevsky
Tim Hewson


During autumn 2017, extreme weather events in Europe included the ongoing drought on the Iberian Peninsula, flash floods in Greece, and the landfall of ex-tropical cyclone Ophelia in Ireland. In this article we focus on two devastating windstorms for which the skill level in ECMWF forecasts was very different.

11 August 2017

On 11 August 2017 severe winds hit northern Poland, causing the deaths of five people, significant damage to trees and power disruptions affecting 340,000 households. This severe storm was caused by a mesoscale convective system. Wind gusts of over 40 m/s were reported in the region. However, even the shortest-range high-resolution forecasts (HRES) failed to predict anything near these values, nor did the Extreme Forecast Index (EFI) or Shift of Tails (SOT) for wind gusts show any significant signal, at any time range. This illustrates that directly capturing fine-scale extremes in convective cases is still out of reach for global models. The finer-scale COSMO-EPS from the German national meteorological service DWD (data archived at ECMWF as part of TIGGE-LAM) did predict extreme wind gusts in the region. However, ECMWF has developed a number of products that make the best use of information contained in the medium-range forecast to identify potential fine-scale weather hazards consistent with the large-scale flow. Examples of this are the EFI for convective indices, and also point rainfall (sub-grid) precipitation probabilities. Indeed in this case the risk of severe convective hazards in the affected region was captured in ECMWF medium-range forecasts by the EFI and SOT product for a composite parameter that combines CAPE and wind shear, albeit somewhat too far east in the 6-day forecast.

Observations and forecasts for the two windstorms
Observations and forecasts for the two windstorms. Reports of severe weather from the European Severe Weather Database (ESWD) for 11 August (top-left) and 29 October (top-right); one- day forecasts of EFI (shading) and SOT (contours) for 24-hour maximum 10-metre wind gusts for the two respective cases (middle); and 6-day forecasts of EFI and SOT for CAPE–shear in the August case (bottom- left) and for 10-metre wind gusts in the October case (bottom-right).

28–29 October 2017

On the night of 28 October a deepening cyclone, named Herwart by the Free University of Berlin, moved southeastwards across southern Sweden and the southern Baltic Sea. Within the large circulation of this cyclone, very strong winds developed, most notably over Germany, the Czech Republic and Poland. There were at least four fatalities, damage to trees and buildings and disruption to infrastructure. According to reports in the European Severe Weather Database (ESWD), the Czech Republic was probably worst affected. The short-range EFI for wind gusts agrees well with those reports. Even at a lead time of six days, the ECMWF EFI and SOT clearly highlighted a greatly elevated risk of a severe wind event over a large area. Indeed, the strongest signal in the EFI was centred on the Czech Republic. Throughout the lead-up to this event, ECMWF ensemble forecasts (ENS) provided a consistent signal for a dangerous windstorm, which grew stronger with time.


These two cases illustrate very different skill for the same variable (wind gusts). In the first case no skilful forecast regarding the extreme wind was provided directly by ECMWF forecasts, while in the other case the forecast showed skill almost a week in advance. As discussed, the meteorological conditions behind the two events were very different. Such differences should be borne in mind when interpreting verification results for wind gusts as both cases contribute to the statistical sample. Indeed verification over a full year shows significantly lower skill during summer, when the contribution of deep moist convection to cases of severe wind gusts is higher in the sample.