Probabilistic forecasts of (severe) thunderstorms for the purpose of issuing a weather alarm Maurice Schmeits, Kees Kok, Daan Vogelezang and Rudolf van Westrhenen KNMI

IUGG 2007 Outline Introduction: Weather alarm for severe thunderstorms Method: Model output statistics (MOS) Data used in MOS system for (severe) thunderstorms Illustration of statistical method Definitions of predictands Case (10 June 2007) Verification results Conclusions and outlook

IUGG 2007 Weather alarm for severe thunderstorms (I) Weather alarm: if probability of ≥ 500 discharges/5 min./(50x50 km 2 ) ≥ 90% in next 12 hours One of the least predictable phenomena History (note: other criterion): many misses, only a few hits and no false alarms Goal: decrease number of misses and increase number of hits, while keeping number of false alarms low Means: new objective probabilistic forecasting system

IUGG 2007 Model output statistics (MOS) Aim: Features: To determine a statistical relationship (mostly via regression) between a predictand (i.e. the occurrence of a thunderstorm in this case) and predictors from NWP model forecasts (and possibly from observations)  forecasts possible for predictands that are not available from direct model output  (reliable) probabilistic forecasts possible, even while using output from a single model run  separate regression equation for each forecast projection (correction of systematic model errors)

IUGG 2007 MOS system for (severe) thunderstorms 3/2 years of data: 1 July 2002 until 1 July 2005 (warm half years only, i.e. 16 April – 15 October) 2/3 part for development and 1/3 part for verification predictands: reprocessed lightning data (Saskia Noteboom) potential predictor set 1: radar data (0 to +6 h only) potential predictor set 2: lightning data (0 to +6 h only) potential predictor set 3: 17 thunderstorm indices, computed from weather model # 1 potential predictor set 4: (derived) DMO (forecasts) from model # 2 potential predictor set 5: (co)sine [day of the year] regression equations contain at least 2 and at most 5 predictors severe thunderstorms: all 12 regions pooled run frequency: 8 times per day (every 3 hours) forecast projections: 0 to +12 h (6-h periods)

IUGG 2007 MOS system for (severe) thunderstorms Ensemble of advected radar data (0 to +6 h) (Ensemble of advected) lightning data (0 to +6 h) Probability of thunderstorms (0 to +6 h/ +6 to +12 h) In developing the LR model you need a 3/2-year long data archive Archive: 2/3 part for development 1/3 part for verification NWP model forecasts (0 to +12 h) Logistic regression (LR) model

IUGG 2007 Example of advection vectors and lightning data: 17 July 2004 (1140 UTC)  RVHV

Example of logistic regression equation using only the first predictor (region M-MS; period: UTC) Probability of thunderstorms Fraction of ensemble with no. of flashes ≥ 4 [SAFIR ] binary predictand logistic curve

IUGG 2007 Where the are we? Introduction: Weather alarm for severe thunderstorms Method: Model output statistics (MOS) Data used in MOS system for (severe) thunderstorms Illustration of statistical method Definitions of predictands Case (10 June 2007) Verification results Conclusions and outlook

IUGG 2007 Weather alarm for severe thunderstorms (II) Weather alarm: if probability of ≥ 500 discharges/5 min./(50x50 km 2 ) ≥ 90% in next 12 hours ‘climatology’ on the basis of this criterion: only twice a year (between 30 April and 15 September) Statistical methods are not capable of handling such rare events. Therefore, other predictand definitions have been used.

IUGG 2007 Predictand definitions Predictand for thunderstorms: Probability of > 1 lightning discharge in a 6h period (00-06, 03-09, 06-12, 09-15, 12-18, 15-21, or UTC) in a 90x80 km 2 region. Predictands for severe thunderstorms: Conditional probability of ≥ X, ≥ Y or ≥ Z discharges/ 5 min. in a 6h period in a 90x80 km 2 region with condition > 1 discharge in the same 6h period in the same region. Here X =50 (all 6-h periods); Y = 100 and Z =200 (12-18, and UTC).

IUGG 2007 Case 17 July 2004 (12-18 UTC; 0 to +6 h) 1150 UTC run (based on SAFIR , H and EC ) ‘Clim.’ prob. of thunderstorms: 6-22 % ‘Clim.’ cond. prob. of severe thunderstorms ( ≥ 200 discharges/5 min.): 4 % (abs. prob.: < 1 %) Probability of thunderstorms Cond. prob. of severe thunderstorms (≥ 50 discharges/ 5 min.)(≥ 200 discharges/ 5 min.) Maximum 5-min. lightning intensity

IUGG 2007 Case 25 June 2006 (15-21 UTC; +6 to +12 h) 08:50 UTC run ‘Clim.’ prob. of thunderstorms: 5-19 % ‘Clim.’ cond. prob. of severe thunderstorms ( ≥ 200 discharges/5 min.): 5 % (abs. prob.: < 1 %) Probability of thunderstorms Cond. prob. of severe thunderstorms (≥ 50 discharges/ 5 min.)(≥ 200 discharges/ 5 min.) Maximum 5-min. lightning intensity

IUGG 2007 Case 25 June 2006 (15-21 UTC; 0 to +6 h) 1450 UTC run (based on H and EC ) ‘Clim.’ prob. of thunderstorms: 5-19 % ‘Clim.’ cond. prob. of severe thunderstorms ( ≥ 200 discharges/5 min.): 5 % (abs. prob.: < 1 %) Probability of thunderstorms Cond. prob. of severe thunderstorms (≥ 50 discharges/ 5 min.)(≥ 200 discharges/ 5 min.) Maximum 5-min. lightning intensity

IUGG 2007 Case 8 June 2007 (15-21 UTC; +6 to +12 h) 09 UTC run (based on H 0806 and EC 0712) ‘Clim.’ prob. of thunderstorms: 5-19 % ‘Clim.’ cond. prob. of severe thunderstorms ( ≥ 200 discharges/5 min.): 5 % (abs. prob.: < 1 %) Probability of thunderstorms Cond. prob. of severe thunderstorms (≥ 50 discharges/ 5 min.)(≥ 200 discharges/ 5 min.) Maximum 5-min. lightning intensity

IUGG 2007 Case 8 June 2007 (15-21 UTC; 0 to +6 h) 15 UTC run (based on H 0812 and EC 0712) ‘Clim.’ prob. of thunderstorms: 5-19 % ‘Clim.’ cond. prob. of severe thunderstorms ( ≥ 200 discharges/5 min.): 5 % (abs. prob.: < 1 %) Cond. prob. of severe thunderstorms (≥ 50 discharges/ 5 min.)(≥ 200 discharges/ 5 min.) Probability of thunderstorms Maximum 5-min. lightning intensity

IUGG 2007 Case 10 June 2007 (15-21 UTC; +6 to +12 h) 09 UTC run (based on H 1006 and EC 0912) ‘Clim.’ prob. of thunderstorms: 5-19 % ‘Clim.’ cond. prob. of severe thunderstorms ( ≥ 200 discharges/5 min.): 5 % (abs. prob.: < 1 %) Probability of thunderstorms Cond. prob. of severe thunderstorms (≥ 50 discharges/ 5 min.)(≥ 200 discharges/ 5 min.) Maximum 5-min. lightning intensity

IUGG 2007 Verification results 2006 (Probability of > 1 discharge) Brier skill score (%) 0 to +6 h +6 to +12 h Brier skill score (%) Time (UTC) Brier skill score (%)

IUGG 2007 Reliability diagrams (’05-’06;12-18 UTC; 0 to +6h) ≥ 50 discharges/ 5 min.≥ 100 discharges/ 5 min. Observed frequency Forecast probability Observed frequency Forecast probability

IUGG 2007 Reliability diagrams (’05-’06;15-21 UTC; 0 to +6h) Observed frequency Forecast probability Observed frequency Forecast probability ≥ 50 discharges/ 5 min.≥ 100 discharges/ 5 min.

IUGG 2007 Reliability diagrams (’05-’06;18-00 UTC; 0 to +6h) ≥ 50 discharges/ 5 min.≥ 100 discharges/ 5 min. Observed frequency Forecast probability

IUGG 2007 Reliability diagram 1 (’05-’06): ≥ 50 discharges/ 5 min. (15-21 UTC; 0 to +6h) Observed frequency Forecast probability

IUGG 2007 Reliability diagram 2 (’05-’06): ≥ 100 discharges/ 5 min. (15-21 UTC; 0 to +6h) Observed frequency Forecast probability

IUGG 2007 Reliability diagram 3 (’05-’06): ≥ 50 discharges/ 5 min. (00-06 UTC; 0 to +6h) Forecast probability Observed frequency

IUGG 2007 Reliability diagram 4 (’05-’06): ≥ 50 discharges/ 5 min. (06-12 UTC; 0 to +6h) Observed frequency Forecast probability

IUGG 2007 Conclusions and outlook Probabilistic forecasts for thunderstorms (> 1 discharge) are skilful with respect to the climatology. Probabilistic forecasts for severe thunderstorms ( ≥ 50/ ≥ 100 discharges per 5 min.) are reasonably skilful with respect to the climatology. The system has been pre-operational at KNMI since Spring of 2006 and will be fully operational later this year. It is expected that this system will help the forecasters to decide whether a weather alarm for severe thunderstorms should be issued.

IUGG 2007 Verification results (Cond. prob. of ≥ 50/100/200 discharges/ 5 min.) Brier skill score (%) Time (UTC) 0 to +6 h+6 to +12 h Brier skill score (%)

IUGG 2007 Reliability diagram 1: ≥ 50 discharges/ 5 min. (12-18 UTC; 0 to +6h) BSS = 30 % Bias = 0.2 % N = 235 Observed frequency Forecast probability

IUGG 2007 Reliability diagram 2: ≥ 100 discharges/ 5 min. (12-18 UTC; 0 to +6h) BSS = 32 % Bias = 4.7 % N = 235 Observed frequency Forecast probability

IUGG 2007 Reliability diagram 3: ≥ 200 discharges/ 5 min. (12-18 UTC; 0 to +6h) BSS = 62 % Bias = 1.3 % N = 235 Observed frequency Forecast probability

IUGG 2007 Reliability diagram 4: ≥ 300 discharges/ 5 min. (12-18 UTC; 0 to + 6h) Observed frequency Forecast probability BSS = 38 % Bias = 0.7 % N = 235

IUGG 2007 Reliability diagram 5: ≥ 400 discharges/ 5 min. (12-18 UTC; 0 to + 6h) Forecast probability Observed frequency BSS = 26 % Bias = 0.1 % N = 235

IUGG 2007 Reliability diagram 6: ≥ 500 discharges/ 5 min. (12-18 UTC; 0 to + 6h) Forecast probability Observed frequency BSS = 13 % Bias = 1.0 % N = 235