During most of May 2018, northern Europe experienced a heat wave. The intensity and spatial extent of the event are evident in the monthly mean air temperature summary maps provided by the Copernius Climate Change Service implemeted by ECMWF (https://climate.copernicus.eu/). Many records for May average temperature were broken. For Stockholm, where temperature records go back to 1759, the monthly average temperature reached 16.1°C, which is 2.2°C higher than the previous record of 13.9°C, the Swedish national meteorological service reported. The hottest days were at the end of May and continued into the first days of June. In the warm and humid air and with generally weak synoptic-scale forcing over north-western Europe, severe convective systems developed during these days. Global forecasting systems can struggle to capture such relatively small-scale systems. Here we look at the usefulness of different ECMWF products for this type of event. We will focus on 29 May, when extreme rain and flash floods affected parts of Germany, the Netherlands, Belgium and France, where Paris was hit by intense rainfall. Severe thunderstorms were accompanied by other convective hazards, including large hail, strong winds and lightning.
In global forecasting systems, heavy convective rain events are usually associated with low forecast probabilities due to the high spatial variability of precipitation and uncertainties in convective initiation. As a result, the total precipitation Extreme Forecast Index (EFI) generally provides weak signals of extreme rain even in short-range forecasts in such situations. This was the case with the forecasts for 29 May over Germany. The predictability of heavy rain events in Belgium and the Netherlands was higher, with positive Shift of Tails (SOT) in the medium range, indicating that at least 10% of ensemble members were predicting extreme rainfall. On the other hand, the model is usually quite good at predicting the favourable environment for deep moist convection well in advance. In the case presented here, the CAPE EFI, for example, gave a much stronger signal for convective hazards throughout the short and medium range than the EFI for total precipitation.
ECMWF’s recently developed ecPoint-Rainfall product uses an innovative post-processing method to account for sub-grid variability and weather-dependent biases in rainfall totals (Newsletter No. 153, autumn 2017). For cases of severe convection, this product should increase probabilities for extreme rainfall and also for no rainfall, compared to the grid-box average probabilities provided by raw model output. In the case of Paris on 29 May, observations of 24-hour rainfall ranged from less than 5 mm to more than 30 mm within the metropolitan area, most of which fell during the afternoon. The raw ensemble from 5 days before (25 May) indicated a maximum possible value of 13 mm/12 hours (as a grid-box average), while the post-processed ecPoint product indicated that point rainfall above 30 mm/12 hours was possible (forecasts valid for 12 UTC on 29 May to 00 UTC on 30 May).
The convection over north-western Europe on 29 May was also associated with intense lightning activity. ECMWF has recently implemented a new parametrization of lightning density (Newsletter No. 155, spring 2018). For the case of 29 May, probabilities of intense lightning in the 5-day forecast from 25 May highlighted the risk over Belgium, Germany and the Netherlands. Note that ATDnet (used here for verification) detection efficiency is much higher for cloud-to-ground (CG) flashes than for intracloud (IC) flashes, whilst the forecast lightning density accounts for both CG and IC discharges, hence we do not expect a perfect match between predicted and observed quantities. Summarising all lightning forecasts valid for 29 May over western Europe (45°N–55°N, 0°E–10°E) in one plot, we find that as early as 10 days before the event the ensemble had a clear signal, with the ensemble median above the 90th percentile of the model climate, and the signal was consistent in all subsequent ensembles.
In summary, ECMWF forecasts captured the risk of thunderstorms in western Europe more than a week in advance. The high predictability was linked to the ability to predict the large-scale environment in which the convective storms developed. Indices such as CAPE and lightning density forecasts are expected to give good guidance on regions likely to be affected by convective hazards. By contrast, there is low predictability for the location and timing of individual convective cells and associated precipitation, although the point rainfall product can better reflect the range of probabilities at particular points than the raw ensemble.