E-WP7 - STEPCLIM: Severe thunderstorm evaluation and predictability in climate models

Project aims

Meteorological hazards associated with severe thunderstorms (e.g., hail, severe wind, tornadoes) pose a major threat to human life and property. Since 1980, severe thunderstorms have caused nearly € 3 billion in damages annually in Europe. While recent studies suggest that the frequency of extreme weather events will likely increase throughout the twenty-first century, the predictability of severe thunderstorms on seasonal-to-decadal time-scales is not well understood. Furthermore, the limited spatial and temporal resolution of present-day climate models prohibits the explicit simulation of meso-scale convection and resultant hazards. The STEPCLIM project has identified three main objectives to address these issues:

  • Implement probabilistic convective hazard models into the MiKlip central evaluation system (CES).

  • Optimise the convective hazard models by improving the European Severe Weather Database (ESWD).

  • Evaluate and correct biases in the frequency of thunderstorms and specific convective hazards as predicted by decadal hindcast simulations.

Deliverables

STEPCLIM will contribute the following deliverables:

  • Scripts that calculate severe weather parameters from model output and estimate the probability of lightning and convective hazards

  • a MiKlip CES plug-in that creates gridded frequency and probability maps for lightning and convective hazards

  • Supporting documentation for the plug-in

Progress so far

First, probabilistic models for lightning, large hail (≥ 2 cm), severe wind (≥ 25 m s-1), and tornadoes were developed and optimised using ERA-Interim reanalysis data, EUCLID lightning data, and ESWD severe weather reports during the 2008–2016 period. These models employ additive logistic regression to estimate the 6-h probability of lightning and individual hazards based on the values of multiple parameters relevant to severe weather (e.g., tropospheric moisture, instability, and wind shear). Next, annual mean probabilities of lightning and individual hazards were computed for ERA-Interim reanalysis, the uninitialised historical MPI-ESM-LR simulations, and the Baseline1 MPI-ESM-LR decadal hindcasts. Lastly, the accuracy and ensemble spread of the decadal hindcasts was assessed using the MurCSS plug- in. Accuracy and ensemble spread metrics, such as the mean square error skill score (MSESS) and continuous ranked probability skill score (SCRPSS), were calculated for each 5° × 5° grid cell within the 22.5°W, 25°N – 52.5°E, 75°N domain. Additionally, the climate forecast recalibration method developed by Pasternack et al. (2018) was applied to the Baseline1 decadal hindcasts to reduce unconditional and conditional biases in the predicted frequency of lightning and convective hazards.

As Figure 1 illustrates, the decadal hindcast projections of lightning probability achieve positive forecast skill over portions of north-western, central, and eastern Europe. Compared to the uninitialised simulations, improvements in forecast skill are largest over the British Isles, the Benelux countries, central Europe, Italy, and the Balkans. Reductions in forecast skill are most pronounced across northern Scandinavia and Finland. Recalibration yields considerable increases in MSESS, particularly over the British Isles, Scandinavia and Finland, and the Balkan Peninsula. The only regions where negative forecast skill remains after recalibration are France and the Iberian Peninsula. One important aspect of the model predictions excluded from this analysis is the tendency of the decadal hindcasts to overestimate the probability of lightning and convective hazards throughout Europe.

Figure 1: Gridded maps showing anomaly correlation (top), conditional bias (middle), and MSESS (bottom) for the historical (left), Baseline1 (middle), and Baseline1 recalibrated (right) predictions of mean annual lightning probability. The climatological reference forecast is represented by ERA-Interim reanalysis. The analysis was conducted for forecast lead years 1–4 and 30 initialisation years spanning the 1984–2013 period.

As Figure 2 illustrates, the decadal hindcasts significantly overestimate the frequency of moist (850–500-hPa relative humidity > 60%), unstable (lifted index < 0) environments compared to ERA-Interim reanalysis. Pattantyús-Ábrahám et al. (2016) previously showed that the Baseline1 decadal hindcasts exhibit large negative temperature biases and positive relative humidity biases in the middle and upper troposphere.

Figure 2: Gridded maps showing the mean annual number of moist, unstable 6-h periods as predicted by Baseline1 (left) and ERA-Interim reanalysis (right). The analysis was conducted for forecast lead years 1–4 and 30 initialisation years spanning the 1984–2013 period.

Figure 3 shows the continuous ranked probability skill score of the ensemble spread (CRPSSES) and the logarithmic ensemble spread score (LESS). The CRPSSES provides a measure of how well the ensemble spread represents the forecast uncertainty [i.e, the mean square error (MSE)], whereas the LESS indicates whether the ensemble variance is under- or over-dispersive. The forecast uncertainty is well-represented by the ensemble spread of the decadal hindcasts in some areas, but in general, the ensemble variance is over-dispersive throughout much of continental Europe (except over the Balkan Peninsula).

Figure 3: Gridded maps showing CRPSSES (top) and LESS (bottom) for the historical (left), Baseline1 (middle), and Baseline1 recalibrated (right) predictions of mean annual lightning probability. The analysis was conducted for forecast lead years 1–4 and 30 initialisation years spanning the 1984–2013 period.

Contact

European Severe Storms Laboratory
Dr. Lars Tijssen
Dr. Pieter Groenemeijer
Dr. Christopher Castellano

Identification of favorable environments for thunderstorms in reanalysis data

2016 - Meteorologische Zeitschrift

Westermayer A.T. | P. Groenemeijer, G. Pistotnik, R. Sausen and E. Faust

Validation of Convective Parameters in MPI-ESM Decadal Hindcasts (1971–2012) against ERA-Interim Reanalyses

2016 - Meteorologische Zeitschrift, Vol. 25 No. 6, pp. 753-766

Pistotnik, G | P. Groenemeijer, and R. Sausen

A Climatology of Tornadoes in Europe: Results from the European Severe Weather Database

2014 - Monthly Weather Review, Vol. 142 (12), pp. 4775-4790

Groenemeijer, P. | T. Kühne