th 5 European Conference on Severe Storms 12 - 16 October 2009 - Landshut - GERMANY

ECSS 2009 Abstracts by session

ECSS 2009 - 5th European Conference on Severe Storms 12-16 October 2009 - Landshut – GERMANY List of the abstract accepted for presentation at the conference: O – Oral presentation P – Poster presentation

Session 05: Forecasting, nowcasting and warning of severe storms

Page Type Abstract Title Author(s)

P. Groenemeijer, J. Dahl, C. Probabilistic severe weather forecasting at the European 111 O Gatzen, T. Púčik, O. Schlenczek, Storm Forecasting Experiment (ESTOFEX) H. Tuschy, O. van der Velde Sounding-derived indices for forecasting hailstorms using 113 O A. Manzato ensembles of artificial neural networks O Severe local storm forecasting in the British Isles P. Knightley Forecasting QPF uncertainty for heavy rainfalls produced D. Rezacova, P. Zacharov 115 O by local convective storms A diagnostic tool based on MSG 6.2/7.3µm channel for the P. Santurette, C. Georgiev, C. 117 O analysis and forecasting of deep convection Piriou Comparison of several Ensemble Prediction Systems M. Vich, R. Romero, V. Homar O applied to Mediterranean high impact cyclones

associated with heavy rainfall and strong winds Convective parameters computed with ALADIN and P. Marquet, P. Santurette 119 O AROME models for the Hautmont (F4) tornado C. Forster, A. Tafferner, T. Nowcasting of thunderstorms within a weather information 121 O Zinner, H. Mannstein, S. Sénési, and management system for flight safety Y. Guillou Objective NearCasts of convective destabilization prior to O isolated summer-time convective events using moisture R. Petersen, R. Aune products from Geostationary satellites O Operational use of satellite and radar products at Portugal P. Leitão Nowcasting thunderstorm activity across the C. Price, M. Kohn, E. Galanti, K. 123 O Mediterranean Lagouvardos, V. Kotroni Temporal Evolution of Total Lightning and Radar V. Meyer, H. Höller, H.-D. Betz, 125 O Parameters of Thunderstorms in Southern Germany and K. Schmidt its Benefit for Nowcasting Analysis of Thunderstorms with the Dynamic State Index T. Schartner, P. Névir, G. C. 127 O (DSI) in a Limited Area High Resolution Model Leckebusch, U. Ulbrich 129 O A waterspout forecasting technique W. Szilagyi 131 O Tornadoes in Germany – Current developments at DWD A. Friedrich Nowcasting and Warning in convective weather situations D. Murer O at MeteoSwiss

109 Page Type Abstract Title Author(s)

A. G. Keul, A. M. Holzer, P. 133 O Are Austrian radio weather warnings user-friendly? Sterzinger, S. Rudolf, A. Reinmueller Convective-scale Data Assimilation and Numerical M. Xue O Weather Prediction at the Center for Analysis and

Prediction of Storms: A Status Update 'À la carte' ensemble perturbations with customizable scale V. Homar, D. J. Stensrud O and amplitude J. S. Kain, S. J. Weiss, M. C. New developments in applied research for severe Coniglio, M. Xue, F. Kong, M. 135 O convection forecasting in the Hazardous Weather Weisman, M. Pyle, R. Sobash, C. Testbed, Norman, OK, U.S.A. Schwartz, D. Bright, J. Levit, G. Carbin. A cloud model study of wind shear effect on the satellite P. K. Wang 137 O observed storm top IR features Recent advances in precipitation nowcasting at the RMI of M. Reyniers, L. Delobbe, P. 139 O Belgium: storm severity product Dierickx, M. Thunus, C. Tricot Predictability of Extreme Storm Events in the State of São G. Held, A. M. Gomes, M. 141 P Paulo, Brazil Teixeira, J. M. Bassan Detection Parameters of Thunderstorms: adaptation of A. C. Nóbile Tomaziello, A. W. P thresholds to Metropolitan Region of São Paulo, BRA Gandu Nowcasting of severe storm at a station by using the soft S. Sharma, D. Dutta, J. Das, R. 143 P computing techniques to the radar imagery M. Gairola Extreme Events in the Amazônia Region during the Rainy J. Saraiva, J. L. M. Lopes, R. H. P Season of 2009 Braga, G. G. Ribeir A proposed air masses classification for the Mediterranean T. la Rocca P to predict severe weather A. Udogwu, J. B. Omotosho, S. Forecasting and Nowcasting of Severe Storms and their 145 P Gbuyiro, I. Ebenebe, G.C Osague, Preferred Tracks across West Africa E. Olaniyan West African weather system in the development of tropical T. Salami, O. S. Idowu, E. E. P cyclones Balogun Nowcasting and assessing thunderstorm risk on Lombardy P. Bonelli, P. Marcacci, E. 147 P Region (Italy) Bertolotti, E. Collino,G. Stella 149 P Triggering of deep convection by low-level boundaries K. J. Rae J. L. Sánchez, L. López, B. Gil- Short-term forecast of hail precipitation parameters 151 P Robles, J. Dessens, C. Bustos, C.

Berthet. Precipitation forecast by the COSMO NWP model using 153 P Z. Sokol, P. Pesice radar and satellite data 155 P Non Mesocyclone Tornadoes in Hungary Z. Polyánszky A statistical study of stability indices as convective weather M. Salvati, D. Berlusconi 157 P predictors in Lombardia P Diagnostic tool convective modes R. Groenland, R. van Westrhenen

110 5h European Conference on Severe Storms 12 - 16 October 2009 - Landshut - GERMANY

PROBABILISTIC SEVERE WEATHER FORECASTING AT THE EUROPEAN STORM FORECAST EXPERIMENT P.Groenemeijer1, J. Dahl2, C. Gatzen3, T. Púčik4, O.Schlenczek5, H. Tuschy6, O. van der Velde7

1 Institute of Meteorology, University of Munich, Germany, [email protected] 2 Institute for Atmospheric Physics, German Aerospace Center DLR, Oberpfaffenhofen, Germany 3 Meteogroup Deutschland, Berlin, Germany 4 Faculty of Natural Sciences, Masaryk University, Brno, Czech Republic 5 Institute for Atmospheric Physics, Johannes Gutenberg-University, Mainz, Germany 6 Institute of Meteorology and Geophysics, Leopold-Franzens-University, Innsbruck, Austria 7 Lightning Research Group, Technical University of Catalonia (UPC), Terrassa, Spain

1. INTRODUCTION a person within an threat level area to experience severe The European Storm Forecast Experiment (ESTOFEX) weather”, which is arguably the most elementary quantity to issues Storm Forecasts that provide an assessment of the forecast, cannot be verified using the dataset. severe convective storm risk, that is the threat of large hail, convective wind gusts, tornadoes and recently, excessive precipitation across Europe (Dahl et al., 2004). These risk estimates are communicated in a forecast text, and more precisely in a map that assigns threat levels to specific areas in Europe. Four threat levels are used, three of which are numbered on the forecast maps: 1, 2 and 3. Level 0 is implied where no level 1, 2 or 3 is indicated. The assigned threat level is determined by a meteorologist who weighs several types of information including data from global and regional numerical models, and observational data from satellites, radiosondes, and surface networks. The threat level system as it was used until 1 May 2009 described the number of severe events to be expected within a 200 km x 200 km area for each threat level area, but this proved to be a difficult criterion for forecasters to work with. To users, it was not easy either to derive the probability of experiencing severe weather within a particular threat level. Still forecast verification could be carried out, for some methods do not require the risk level to be specified, enabling Brooks et al. (2008) to present verification results FIG 1: Severe weather events in the period considered, 1 May 2008 of ESTOFEX forecasts without having to use the – 30 April 2009. Triangles pointing upward: large hail, pointing troublesome criteria. downward: tornadoes, circles: wind gusts. The dark contour denotes The wish to develop an straightforward and usable the area across which the analysis has been performed. definitions of the probability of severe weather in each threat level remained a goal of the team. In what follows, the The obvious solution to this conundrum is to specify an method of deriving those will be explained. arbitrary area of vicinity. We have chosen a circle of 40 km, which is comparable with the 25 nautical miles radius used II. METHOD by the U.S. Storm Prediction Center to facilitate future In making the transition from a qualitative to a quantitative comparisons. The predictand of the forecasts was formally forecast scheme, it was decided not to discard the old threat defined as: levels, but rather to quantitatively define the threat levels that were already used. To find the corresponding probability “The probability that one or more severe weather events values, an a posteriori analysis of the average frequency of occur within a circle with radius 40 km of a point” severe weather in the threat areas was performed. Such an exercise would ideally make use of a dataset that contains all severe weather events that occurred, and the size of the area severe weather extremely severe affected by each event. Naturally, such a dataset does not weather exist. The dataset that comes closest to this is the severe hail ≥ 2 cm ≥ 5 cm weather database ESWD (Dotzek et al., 2009), managed by tornado any ≥ F2 the European Severe Storms Laboratory. This dataset however, contains severe weather events as points in space wind gust ≥ 25 m/s ≥ 32 m/s and time. The size of the area affected by the event is TABLE. 1: Fraction of rectangles contained within each threat level normally not available. This implies that “the probability of area that was labelled “severe” and “extremely severe”, respectively.

111 The procedure was repeated with severe weather replaced by extremely severe weather. For definitions of these concepts, please refer to Table 1. Subsequently, a part of the forecast area of Estofex was selected as the domain for the analysis (see Fig. 1). Within this area, which includes the Benelux, Germany, the Alps, Hungary, the Czech Republic, Slovakia and Poland, active storm spotter networks provided a –subjectively judged– good coverage of severe weather throughout the considered period (1 May 2008 – 30 April 2009). This area was divided into rectangular areas with a surface area corresponding to a circle with radius 40 km. Subjectively, an area across which several active storm spotters delivered data to the ESWD data was selected. Using the dataset, each rectangle could be labelled as being FIG. 2: Fraction of rectangles contained within each threat level area “non-severe”, “severe”, or “extremely severe”. This that was labelled “severe” and “extremely severe”, respectively. procedure to some extent mitigates the fact that the ESWD dataset only contains a part of all severe weather events that IV. CONCLUSIONS AND OUTLOOK occurs in reality: if severe weather strikes a particular With the presented analysis, ESTOFEX has made a first step rectangle, one report from the area suffices to label the towards quantitative forecasting. Since the presented results square as “severe”. It is thus not necessary that an observer have become available, ESTOFEX has also changed its of severe weather be present in every square kilometer. lightning forecasts from deterministic to probabilistic. An important new question to solve is how forecast verification III. RESULTS can be performed using this new data. Threat levels defined It was found that the coverage of severe weather (i.e. the by a range of probabilities may not be the easiest to work fraction of contained rectangles labelled as “severe” within with when developing a forecast verification method. each area) increases strongly with increasing threat level, Perhaps the definitions need to be reviewed in this light. which is what one would hope for. The values that were found are displayed in Fig. 2. IV. ACKNOWLEDGMENTS Using these coverages of severe and extremely The authors want to thank Chuck Doswell, who is always severe weather, the future definitions of the threat level areas eager to discuss probabilistic forecasting and forecast could be redefined. At the same time the definition was to be verification within the ESTOFEX group. We also thank developed, there was a very strong wish of the forecasters to Kirstin Kober for her critical comments that contributed to revise the criterion for severe weather to include excessive this work. Furthermore, we are grateful to the EUCLID precipitation. Such a step would likely lead to higher lightning detection network for supporting the experiment by coverages of severe weather within the risk areas. Moreover, providing lightning detection data, and to ESSL and its a further slight increase of reporting efficiency was partners for providing the severe weather reports. anticipated. As a consequence it was decided to define the threat levels with rather high percentages relative to the values of past severe weather coverage. V. REFERENCES Brooks, H. E., T. E. Thompson, C. M. Shafer, C. Schwartz, The new definitions are: P. Marsh, A. Kolodziej, N. Dahl, and D. Buckey, 2008: Evaluation of ESTOFEX forecasts: Severe thunderstorm • Level 0 Lower than 5% probability of severe forecasts, 24th Conference on Severe Local Storms, 27–31 • Level 1 5 - 15% probability of severe October 2008, Savannah (GA), USA. • Level 2 > 15% probability of severe Dahl, J., C. Gatzen, P. Groenemeijer and O. van der Velde, (where level 3 does not apply) 2004: ESTOFEX the European Storm Forecast • Level 3 > 15% probability of extremely severe Experiment - towards Operational Forecasting of European Severe Thunderstorms, Preprints, 3rd European It can be seen that level 0, 1 and 2 have been Conference on Severe Storms, 9-12 November 2004, defined in terms of the coverage of severe weather, but that León, Spain for level 3, the coverage of extremely severe weather is the Dotzek, N., P. Groenemeijer, B. Feuerstein, and A. M. defining quantity. These thresholds are high relative to the Holzer, 2009: Overview of ESSL's severe convective percentages that were found. For example, in the dataset the storms research using the European Severe Weather observed coverage of severe weather in level 1 was only Database ESWD. Atmos. Res, 93, 575-586. slightly above the 5 % threshold (5.1%). The threat levels Groenemeijer, P., O. van der Velde, H. Tuschy, C. Gatzen, J. have been defined each as a range of probabilities, so that Dahl, and N. Verge, 2007: Verification of dichotomous their borders correspond with lines of equal probability. lightning forecasts at the European Storm Forecast Experiment (ESTOFEX), Preprints, 4th European Conference on Severe Storms, 10-14 September 2007,

112 5th European Conference on Severe Storms 12 - 16 October 2009 - Landshut - GERMANY

SOUNDING–DERIVED INDICESFOR FORECASTING HAILSTORMS USING ENSEMBLESOF ARTIFICIAL NEURAL NETWORKS Agostino Manzato

OSMER – Osservatorio Meteorologico Regionale dell’ARPA Friuli Venezia Giulia Via Oberdan, 18/a, 33040 Visco (UD), Italy, [email protected] (date: 14 September 2009)

I. INTRODUCTION Frequency distribution of the number of hit hailpads for the 964 6h cases of years 1992−2007 A method based on neural networks for

forecasting thunderstorm and rain from N(x)=2100 * t * 0.5^t * x^−(t+1), with t=0.623 sounding–derived indices in Friuli Venezia Giulia has been presented in the ECSS–2004 conference (Manzato 2007). This time, a

similar approach, but using neural network Numb. of cases ensembles, will be applied to the hail foreca- sting problem.

Frequency of the 964 haily cases per 30−days moving average (1992−2007) and mean Sea Surf. Temperature

1 254% 5 10 2095% 50 100 200 500 25 25 1 hailpad 1 < hailpads < 15 0 5 10 1520 3040 5060 7080 100 hailpads >= 15 Hit hailpads in 6h [number]

Mean Mediterranean SST 20 20 fig. 2: The total number of hit hailpads distri- bution in 6–hours. The bin width is 2, that is, 15 15 the first bin shows the 68% of cases with 1 or 2 hit hailpads, the second one shows the ca- 10 10 haily case frequency [%]

Mean Sea Surface Temperature [C] ses with 3 or 4 hit hailpads and so on. The 5 5 dashed line is a Pareto distribution fit, which underestimates the first bin. 0 0

91 121 152 182 213 244 274 Apr May Jun Jul Aug Sep 6h (Fig. 2). Julian Day 52 different sounding-derived indices (8 fig. 1: The distribution of the number of hit out of which newly developed for this work), hailpads clustered in three classes for the dif- plus the Sea Surface Temperature, the Julian ferent Julian days of the hail season (30–days day and the period of the day, are used as moving average). candidate predictors for forecasting hail. Two II. DESCRIPTION problems are studied: the first is to classify In this work, more than 50 different the occurrence of at least 2 hailpads hit in 6h, sounding-derived indices are used for fore- while the second is to estimate the number of casting the occurrence and extension of hail- hit hailpads in 6h. storms in the Friuli Venezia Giulia region (NE Linear bivariate methods (discriminant Italy) during the 6h after the sounding launch. analysis and regression) are applied to the A hailpad network collects data from April whole database, finding a maximum Pier- to September in the plain of this region since ce Skill Score as high as 0.46 (for UpDr 1988, but only data after 1992 have been used, and Hail Diameter, followed by CAPE, DTC because the sounding database starts in 1992. –Manzato 2003–, DT500 and SWISS – A climatology of hail in the 1992-2007 pe- Huntrieser et al. 1997–) for the classification riod is presented, stratifying the 6h cases per and a linear correlation as high as R = −0.32 Julian day (Fig. 1), per period of the day (6h- (for Showalter Index –Showalter 1953–, fol- long, because it is associated to the 4 daily lowed by DT500 and UpDr) for the number soundings) and per number of hit hailpads in of hit hailpads regression problem.

113 5th European Conference on Severe Storms 12 - 16 October 2009 - Landshut - GERMANY2

After that, a neural network non-linear ap- the regression problem. In the first case, proach is chosen, dividing the database in 3 just 4 networks are used (Fig. 3), while in samples: training, validation and independent the second, the first 7 networks chosen by test. The candidate predictors are preproces- the stepwise model selection algorithm are sed transforming them to their empirical po- combined together. sterior probability for the classification pro- blem and to their Z–scores (standardization) III. CONCLUSIONS for the regression problem. In this multiva- The final results using the neural network riate approach a subset of 5 to 9 predictors, ensemble are better than the simpler linear chosen by a stepwise algorithm, can usually methods and than the best neural networks optimize the neural network in terms of vali- used alone. For the classification, the indices dation error (Cross Entropy for classification which seem to be chosen more frequently in and Mean Square for regression). the neural network first input positions are the To avoid overfitting, the training/validation SWISS index, the surface-850 bulk shear and sample division is made in different ways (re- the period of the day. sampling), leading to 17 different lists of se- On the other hand, the indices which seem lected predictors. Varying also the number of to be more important for the neural networks hidden neurons and the sounding-derived in- of the regression ensemble are the SWISS dex database (computed with the 3 thermo- index, the Showalter Index and the mean dynamic schemes described in Manzato and wind between 6 and 12 km of height. Morgan 2003) a total of 238 classification and 204 regression networks are built, respective- IV. REFERENCES ly for the hail occurrence and the hailstorm • Huntrieser, H., Schiesser, H. H., Sch- extension problem. mid, W. and A. Waldvogel, 1997: PSS vs. the number of ensemble ANNs for the hail2 classification Comparison of Traditional and New- ly Developed Thunderstorm Indices for 1992−2001 + 2006−2007 Total PSS Switzerland, Wea. Forecasting, 12, 2002−2005 Test PSS 108–125. • Manzato, A., 2003: A climatology of in- stability indices derived from Friuli Ve- nezia Giulia soundings, using three dif- ferent methods. Atmos. Res., ECSS maximum Peirce Skill Score 2002 special issue, 67–68, 417–454. • ———–, 2007: Sounding–derived indi- ces for neural network based short–term

0.55 0.60 0.65 0.70 0.75 0.80 0.85 0.90 thunderstorm and rainfall forecasts, At- 1 2 3 4 5 6 7 8 9 10 mos. Res., ECSS 2004 special issue, 83, Number of ANNs in the MMMV ensemble 349–365. fig. 3: Modified Mojirsheibani Major Voting • ———–, and G. M. Morgan, 2003: ensemble using different number of ANN mem- Evaluating the sounding instability wi- bers for classifying the occurrence of at least th the Lifted Parcel Theory, Atmos. two hit hailpads in 6h. The independent Test Res., ECSS 2002 special issue, 67-68, dataset start to decrease the Pierce Skill Sco- 455–473. re after that more than 5 neural networks are selected. • Mojirsheibani, M., 1999: Combining Lastly, a subset of these networks are classifiers via discretization. Journal combined together in an ensemble using a of the American Statistical Association, newly proposed variant of the Mojirshei- 94, 600–609. bani Major Voting (Mojirsheibani 1999) • Showalter, A. K., 1953: A stability in- for the classification problem and a linear dex for thunderstorm forecasting, Bull. multiregression of the network outputs for AMS, 34, 250–252.

114 FORECASTING QPF UNCERTAINTY FOR HEAVY RAINFALLS PRODUCED BY CONVECTIVE STORMS Daniela Rezacova1, Petr Zacharov2

1Institute of Atmospheric Physics AS CR, Bocni Str. II/1401, 141 31 Prague 4 - Sporilov, Czech Republic, [email protected] [email protected]

(Dated: 14 September 2009)

I. INTRODUCTION developed on 10 July 2004 and the two last events were Forecast uncertainty is currently considered to be an recorded in May 2005 (23rd and 30th May). The storms, the inherent part of high-resolution quantitative precipitation QPF verification, and a first analysis of the relationship forecast (QPF), and it is particularly pronounced when between ensemble skill and ensemble spread are described predicting heavy convective precipitation. In order to assess in Rezacova et al. (2009). the uncertainty in short–range QPF, several convective storms, which produced heavy local rainfalls in the Czech 30_05_2005 Republic, were studied. The storms differed in precipitation localisation and area extent and in the convective environment as well. The NWP model LM COSMO was run with a horizontal resolution of 2.8 km and a forecast ensemble was created by modifying model initial and boundary conditions. The forecasts were verified by gauge adjusted radar-based rainfalls (Sokol, 2003; Rezacova et al., FIG. 1: An example of the event with heavy convective 2007). We applied so-called “fuzzy” verification techniques precipitation; the comparison between predicted (left) and gauge adjusted radar-based(right) 12h rainfall (10.00UTC – 22.00UTC) (Ebert, 2008), which allow some relaxation of the 30.5.2005. The horizontal resolution of both forecasted and requirement of exact matches between grid point (or area) observed precipitation fields is 2.8 km. forecasts and observations. Our verification is based on the so-called Fractions Skill Score (FSS), which corresponds to The ensemble FSS-skill and FSS-skill spread values a “fuzzy” verification approach (Ebert, 2008; Roberts and were determined as mean values over the ensemble Lean, 2008). The FSS expresses how the area of interest is members. Like the FSS they depended on the scale (a size of covered by a rainfall that exceeds a given threshold. of elementary area), and on a precipitation threshold. The In order to evaluate the ensemble forecast, FSS- evaluation was performed separately for 1, 3, and 6 h based ensemble skill and ensemble spread were determined. rainfalls using various threshold values (TH = 0.1, 1, 2, and The spread represents the forecast uncertainty and follows 5 mm) and scales (square elementary area with sides of 5, from the differences between the control forecast and the 11, 15, 21, 25, 31, 35, 41, 51, and 61 grid points). forecasts provided by ensemble members. The forecast The rainfalls were determined with a time step of 1 h accuracy is characterized by the skill which evaluates the starting after 7 h (13UTC - 14UTC), 5 h (11UTC - 14UTC), differences between the precipitation forecasts and radar- and 4 h (10UTC - 16UTC) of integration time for 1, 3, and 6 based rainfalls. User-oriented information about forecast h rainfalls, respectively. The last considered rainfalls uncertainty should be available at the time of forecast, unlike corresponded to the time periods 21UTC - 22UTC (1h forecast accuracy, which can be expressed by an a posteriori rainfalls), 19UTC - 22UTC (3h rainfalls), and 16-22UTC verification using measurements. The relationship between (6h rainfalls). It means that the FSS related values were ensemble skill and ensemble spread is important information computed for 9 (1h rainfalls), 9 (3h rainfalls), and 7 (6h showing how the ensemble spread reflects the forecast rainfalls) rainfall fields at every event. The whole set of accuracy (e.g., Whitaker and Loughe, 1998; Sherrer et al., {FSS-skill, FSS-spread} couples comprised 360 values (1h 2004; Grimmit and Mass, 2007). and 3h rainfalls) and 280 values (6h rainfalls) for each of This study deals with the estimation of prognostic five convective events. FSS-skill by using the ensemble FSS-spread and the relationship between FSS-spread and FSS-skill. The first III. RESULTS AND CONCLUSIONS numerical experiments included 5 events (Zacharov and The results show that the skill estimation based on Rezacova, 2009). They used the skill and spread values determining the ensemble spread and on a simple statistical related to 4 events to estimate the skill-spread relationship. evaluation of the spread-skill relationship appears to be a The relationship was applied to the fifth event to predict the useful technique. The distribution of differences between ensemble skill given the ensemble spread. prognostic and diagnostic skill shows low bias, and the interquartile range between 0.10 and 0.30. The Percent II. THE PREDICTION OF FSS-SKILL Correct score gave a mean of 0.68. One of five events The ensemble values were determined for five local showed a marked overestimation of the FSS-skill and the convective events that produced heavy local rainfall. Two mean PC of 0.39. The mean PC over the other 4 events gave th th events occurred in July 2002 (13 and 15 July), one storm a value 0.75.

115 A decrease in the ensemble spread (increasing FSS- IV. AKNOWLEDGMENTS spread) and an increase in the forecast skill (increasing FSS- This work was supported by the Ministry of skill) are event-dependent. This means that there is no fixed Education Sports and Youths CR in the project OC112 scale that can give a threshold FSS-skill value. This is why (COST 731) and by the grant GA AS IAA300420804. We the regression projecting the FSS-spread on the FSS-skill acknowledge the CHMI for the meteorological data support was constructed for all the scale sizes together. However, it and the German Weather Service for providing LM COSMO would be worth testing the stratification of the skill (spread) code for the research. relationship according to the scale after extending the input dataset. V. REFERENCES Ebert E. E., 2008: Fuzzy verification of high resolution gridded forecasts: A review and proposed framework. Meteorol. Apps., 15, 51-64. Grimit E. P., Mass C. F., 2007: Measuring the ensemble spread-error relationship with a probabilistic approach: Stochastic ensemble results. Mon. Wea. Rev., 135, 203- 221. Roberts N. M., Lean H. W., 2008: Scale-selective Verification of rainfall accumulations from high- resolution forecasts of convective events. Mon. Wea. Rev, 136, 78-97. Rezacova D., Sokol Z., Pesice P., 2007: A radar-derived verification of precipitation forecasts for local convective storm. Atmos. Res., 83, 211-224. Rezacova D., Zacharov P., Sokol Z., 2009: Uncertainty in the area-related QPF for heavy convective precipitation. Atmos. Res., 93, 238-246. Scherrer S. C., Appenzeller Ch., Eckert P., Cattani D.: 2004: Analysis of the Spread–Skill Relations Using the ECMWF Ensemble Prediction System over Europe. Wea and Forecasting, 19, 552-565. Sokol Z., 2003: The use of radar and gauge measurements to estimate areal precipitation for several Czech river basins. Stud. Geophys. Geod., 47, 587-604. Whitaker J. S., Loughe A. F., 1998: The Relationship between Ensemble Spread and Ensemble Mean Skill. Mon. Wea. Rev., 126, 3292-3302. Zacharov P., Rezacova D., 2009: Using the fractions skill score to assess the relationship between an ensemble QPF spread and skill. Atmos. Res., doi:10.1016/j.atmosres. FIG. 2: Forecast skill (FSS-skill FIT, vertical axis) against 2009.03.004 measurement-based skill (FSS-skill, horizontal axis) for 3 h rainfall and all the thresholds considered. The numbers placed inside the blocks represent absolute frequency values in corresponding FSS intervals. The values referring to various events are distinguished by colors in the upper panel. The values referring to scale are distinguished by colors in the lower panel.

We show that the so called fuzzy verification measures, like FSS, are applicable also in estimating the regional ensemble spread/skill relationship. Second conclusion deals with the fact that it is difficult if possible to find general threshold scale given forecast accuracy. The scale effect can be event dependent. Searching an effective expression of the spread-skill relationship, it is perhaps more useful to take advantage of the whole scale dependence. The FSS appeared to be a suitable score to overall assess the forecast over the whole verification domain. However, a more extended dataset comprising more heavy precipitation events should allow us to consider smaller sub- areas of the Czech territory. This is why enlarging the case studies from the days with severe convective weather and/or local flash floods is our main aim in future work. Next, a technique of ensemble construction should be improved and we suppose that future application to time series would be useful in order to examine the technique with more general precipitation fields.

116 5th European Conference on Severe Storms 12 - 16 October 2009 - Landshut - GERMANY

A DIAGNOSTIC TOOL BASED ON MSG 7.3µµµ/6.2µµµ CHANNELS FOR THE ANALYSIS AND FORECASTING OF DEEP CONVECTION Patrick Santurette1, Christo Georgiev2, Catherine Piriou1

1Forecast Laboratory, Météo-France, 42, Avenue G. Coriolis, 31057 Toulouse Cedex 01, France, [email protected] 2 National Institute of Meteorology and Hydrology, Bulgarian Academy of Sciences, Tsarigradsko chaussee 66, 1784 Sofia, Bulgaria, [email protected] (15 September 2009)

I. INTRODUCTION • The 6.2 µm imagery gives a view of the upper-level The first step of the forecasting process, the ‘early dynamics and may be used for upper tropospheric warning of convection’, consists in determining the area conditions analysis (jet stream evolution, PV anomalies). where deep convection is possible in the near future (1 to 12 See Weldon and Holmes, 1991; Santurette and hour period). The second step is to forecast precisely the Georgiev, 2005. time and location of convective developments up to 3 hours, • The 7.3 µm channel is able to detect moisture at low- and then the last step is to identify existing convection and mid-level; the images in 7.3 µm channel often exhibit to forecast the evolution of the convective systems. The clear features that can provide relevant information as numerical forecast has improved these last years, but regards to the low-level thermodynamic context remains relatively poor as regard predicting intense (Georgiev and Santurette, 2009). So the study of the 7.3 convection. The job of the forecaster is to expertise all µm imagery features helps to diagnose favourable low- relevant available data in order to detect severe weather mid-level environment for strong convection. systems appearing or very likely in the near future. The main Potential instability is pronounced essentially when information source to perform this task remains upper-levels are cold and dry while low-levels are warm and observations, among which satellite imagery is the most moist; that corresponds to 6.2 µm exhibiting warm important. brightness temperature (BT) while 7.3 µm exhibits cold Water vapour channels 6.2 and 7.3 µm offering brightness temperature. So a relative small WV BT respectively information about the upper-level conditions difference (WVBTdif = 6.2 µm BT – 7.3 µm BT) when 6.2 and low- to mid-level atmospheric situation can be used to µm BT is warm indicates potential instability; decreasing of diagnose the convective potential of the situation. this difference marks development initiation. FIG. 1 shows Interpreting 6.2 and 7.3 µm channels’ information jointly maps of WVBTdif over Europe on 30 June 2006. with relevant fields can provide new diagnostic to help the analysis and nowcasting of deep moist convection. The aim of the present work is to propose an efficient method for diagnosis that can help forecasters in predicting convection, and how best to exploit it. This diagnostic is derived from satellite data and available analysed (or very short range forecast) fields. To perform this task the investigation is accomplished through many convective cases studies in the framework of a joint project between 11 UTC 12 UTC Météo-France and the NIMH of Bulgaria. First results of the preliminary studies are presented as well as future prospects to improve the method for operational purposes.

II. WATER VAPOUR CHANNELS FOR DIAGNOSING PRE-CONVECTIVE ENVIRONMENT 13 UTC Different authors have stressed on the benefit that 15 UTC can be got from water vapour (WV) imagery in predicting the environment of deep severe convection (e.g. Thiao et al., FIG. 1 WVBTdif images on 30 June 2006 over Europe; dark blue 1993; Georgiev, 2003; Krennert and Zwatz-Meise, 2003; for difference from -18 to -10 °C, light blue for -10 to -5 °C, green for -5 to 0 °C, yellow for > 0°C. Grey shades indicate areas of Santurette and Georgiev, 2007). New MSG satellites offer µ potential instability; green and yellow indicate clouds radiances in 7.3 m WV channel that contains information developments. Convection develops after 12 UTC on the west and for low- to mid-level moisture distribution (Georgiev and central part of Balkans; See text for explanations. Santurette, 2009). The two MSG 6.2 and 7.3 µm WV channels are sensitive to water vapour content at different Grey shades indicate areas of potential instability altitudes and allow observing moisture and wind regimes in since these areas correspond to 6.2 µm BT warmer than – different layers of the troposphere. The MSG WV channels 39°C with weak value of WVBTdif; potential instability is exhibit the following information content: more pronounced in light grey areas where in addition 7.3

117 5th European Conference on Severe Storms 12 - 16 October 2009 - Landshut - GERMANY

µm BT is colder than – 19 °C. It can be seen that convection superimposition of strong MOCON and instability as seen develops over the Balkans after 12 UTC inside light grey by WVBTdif image (red arrow, FIG. 3b). areas, which have appeared within the dark grey shades some time before (arrows in FIG. 1). (a) (b)

III. COMBINING WV CHANELS AND

RELEVANT FIELDS TO ANTICIPATE CONVECTIVE DEVELOPMENT During the early warning phase, the forecaster attempts to assess the degree of potential instability as well

as the possible forcing mechanisms. Knowing potential instability is not enough to predict correctly strong FIG. 3: WVBTdif images at (a) 11 UTC, (b) 15 UTC on 24/09/2006 convection, which requires potential instability and a low- over the southwest of France (same colours as in Fig. 1); MOCON level moisture source but also a lifting mechanism. In other (orange, from 3 x 10-7s-1, interval 5 x 10-7s-1). See text. words the motion field plays an important role in the deep convective process, mainly in case of severe long-lived No thunderstorm was expected this day; However convective systems. strong convection developed in the afternoon producing hail FIG. 2 presents 100 m wind field and 950 hPa and the highway near Toulouse has been closed temporarily. convergence field forecast by ARPEGE (06 UTC run) valid Images of WVBTdif combined with convergence field (here on 30 June at 12 and 15 UTC, superimposed on infrared analysed 10 m MOCON) could allow to discriminate the image; It can be seen that convection developments areas where the convection was most likely to develop. occurring over the Balkans are well correlated with convergence wind areas forecast by the model (red and blue IV. CONCLUSIONS arrows on FIG. 2; corresponding with red and blue arrows Water vapour channels 6.2 and 7.3 µm can be used on FIG. 1). to diagnose the convective potential of the situation. We found the difference of BT very useful as an observation of the evolution of the potential instability. Preliminary studies show that combining this information with relevant fields representing crucial elements in the deep convection development (analysis or very short range forecast of MOCON or low-level convergence wind) can be of great help in anticipating strong convective system. This work has to be completed by discriminating the other relevant fields that could improve the efficiency of the diagnostic tool, like observed low-level dew point, upper-

level divergence. We must also propose clear method FIG. 2: IR image with superimposition of the wind at 100 m (blue) regarding interpretation of WVBTdif images probably by and of the convergence field at 950 hPa (red, > 2 x 10-5s-1, interval 2 x 10-5 s-1), forecast by the French ARPEGE model (06 UTC model considering types of situations. Experimentations have also run) on 30 June 2006, valid for 12 UTC (left) and 15 UTC (right). to be completed in order to adjust the BT values as regards the threshold used for the definition of the grey shades. Satellite imagery is not enough to anticipate the pre- convective environment. Assessing the wind field structures V. REFERENCES in addition to the potential instability is crucial in early Georgiev C. G., 2003. Use of data from Meteosat water forecasting convection. That’s why we propose composite vapour channel and surface observations for studying pre- diagnostic tools, combining WVBTdif images with convective environment of a tornado-producing storm. meteorological fields representing crucial process of deep Atmos. Res. 67−68, 231−246. convection outbreak. It is well known that low-level Georgiev C. G., Santurette, P., 2009. Mid-level jet in Intense convergence (especially moisture flux convergence Convective Environment as seen in the 7.3 µm Satellite ‘MOCON’), is required to produce long-lived convective Imagery. Atmos. Res. 93, 277–285. system. Over continental areas, observations of 10 m wind is Krennert, T., Zwatz-Meise, V., 2003. Initiation of available hourly (like 2 m humidity). When not available, convective cells in relation to water vapour boundaries in wind field can be well anticipated by numerical forecast satellite images. Atmos. Res. 67−68, 353−366. thanks to the good skill models as regards to this parameter. Santurette, P., Georgiev, C. G., 2005. Weather Analysis and So combining analysed or forecast low-level wind Forecasting: Applying Satellite Water Vapor Imagery and convergence with WVBTdif images offers a first step for a Potential Vorticity Analysis. Elsevier Academic Press. composite diagnostic tool to predict deep convection. ISBN: 0-12-619262-6, 200 pp. FIG. 3 shows a superimposition of the 10 m Santurette, P., Georgiev, C.G., 2007. Water Vapour Imagery MOCON analysis (got from French observations) and Analysis in 7.3/6.2µm for diagnosing thermo-dynamic WVBTdif image on 24 September 2006, when cold air context of intense convection. 2007 AMS-EUMETSAT overruns the southern part of France in southerly cyclonic Meteorological Satellite Conference, Amsterdam, The flow. FIG. 3a shows that instability set on the south-west of Netherlands, 24-28 September 2007. France (light grey shade on FIG. 3a), while pronounced Thiao W., Scofield, R., Robinson, J., 1993. The Relationship MOCON area is analysed in the Toulouse area (red arrow, Between Water Vapor Plumes and Extreme Rainfall FIG. 3a). A strong convective development occurs near Events During the Summer Season. NOAA Tech. Memor. Toulouse some hours latter well correlated with the NESDIS 67, Sat. Appl. Lab., Washington, 69 pp.

118 5h European Conference on Severe Storms 12 - 16 October 2009 - Landshut - GERMANY

CONVECTIVE PARAMETERS COMPUTED WITH ALADIN AND AROME MODELS FOR THE HAUMONT (F4) TORNADO. P. Marquet1, P. Santurette1

1Météo-France, DPrévi/LABO (Forecast Laboratory), 42 av. G. Coriolis, 31057 Toulouse CEDEX 1, France, [email protected] ; [email protected]

I. INTRODUCTION The two terms SBCAPE and SRH are normalized in A F4-tornado has produced severe damages in the the EHI formula (Davies, 1993) with some relevant values, North of France, during the late afternoon of the 3rd of set to 1000 J/kg and 160 m2/s2, respectively. The aim is to August, 2008 (FIG. 1). This severe convective event has not provide an index varying from 0 to some units, leading to a been explicitly forecasted in the basic fields of the suite of possible increasing scale of risk for Super-cells or Tornado: Météo-France nested models, neither with the hydrostatic ALADIN LAM (9.5 km) nor the AROME meso-scale and non-hydrostatic LAM (2.5 km).

FIG. 2: The EHI scale of risk for Super-cells or Tornado.

The other Convective Parameters (see for instance Thomson et al., 2003) are: the BRN (Bulk Richarson Number) ; the SCP (Super-cell Composite Parameter) and the STP (Significant Tornado Parameter).

II. PRESENTATION OF RESEARCH The motivation for this study is to compute with the ALADIN (9.5 km) and the AROME (2.5 km) models all the normalized USA Convective Parameters (BRN, EHI, SCP, STP). The aim is to determine the regions with higher risk of severe convection, for the now-casting or short-range forecast purposes, i.e. where the indexes reach (or go beyond) the appropriate levels in each of the “scale of risk” (like > 4 for EHI and for F4 tornados on FIG. 2). In the present preliminary study, we have used the same thresholds and “scale of risk” than the ones determined

FIG. 1: The F4 event occurred between 20h25 and 21h UTC, close in the USA from the climatology of RS over the Great- to the Haumont town in the North of France, and close to Belgium. Plains. These values will have to be confirmed (or retuned) to be adapted to the conditions prevailing over France. The aim of the Forecast Laboratory is to test new ideas, new products. Among the available outputs for the III. RESULTS ALADIN model, there already exist the CAPE, the Storm Since the Meso-scale model AROME start to resolve Relative Helicity and a “Composite Map” which the convective scale, the CAPE computed from AROME corresponds to an attempt to draw altogether different outputs can be logically small (even equal to zero) in the indices measuring the “availability in severe convection”, simulated convective cores. But the CAPE is a common with either enhancement or inhibition characteristics, such as factor of all the Convective Parameters and it could be “θ’w>14°C” and moisture convergence at 925 hPa ; ω and interesting to apply some spatial filtering to the outputs of Q.G. forcing within 400-500 hPa ; Saturation-deficit, … AROME, in order to catch (may be) more relevant environ- Differently to this set of moisture and dynamical mental values into the CAPE computations (with unchanged properties, the approaches followed in the U.S.A. for the or recomputed cores). tracking of severe thunderstorms and tornados promote the The “Filtered EHI” is thus computed with the use of special “Convective Parameters”. environment based on 100 km – Filtered data, leading to: These Convective Parameters are all based on a CAPE information (either the Most Unstable, the Mean- Level or the Surface Based one). They also include other terms, added as product terms to the (MU, ML or SB) FIG. 3: The EHI computed with AROME (2h forecast CAPE, like the Helicity, the vertical shear, the Lifting from the 18 UTC analysis). Condensation Level height or the MLCIN. As an example, the “EHI” is the Energy (SBCAPE) Note that the significant and Storm Relative Helicity (SRH) Index. It is defined by: values for the EHI are only located close to the region of Haumont, i.e. with small level of False Alarms.

119 5h European Conference on Severe Storms 12 - 16 October 2009 - Landshut - GERMANY

It results the following EHI maps, shown in FIGS. 3 and 4. The “STP” and the associated “scale of risk” are They are based on a 18 UTC analysis, from 1 to 3 hours. The F-EHI signal appearing in FIG. 4 is almost well localized and with the right timing. Only the intensity - from FIG. 3 - is smaller than the expected value F-EHI > 4 (with possible dark purple colours).

FIG. 8: The formula and the “scale of risk” for the STP (AROME). Like for the SCP values, the STP ones seen by AROME are realistic, with clear and narrow maxima passing through the region of Haumont. The STP values almost reach the relevant ones for AROME (0.8 to 0.9 red spot at 21 UTC), whereas from the FIG. 8 they should reach 1 unit, or more.

FIG. 4: The “movies” of the F-EHI for the F4 event, as simulated by the AROME (2.5 km) NH model / Zoom of FIG. 3. Forecast from the 18 UTC analysis: 1 h and 19 UTC (left) ; 2 h and 20 UTC (center) ; 3 h and 21 UTC (right). The SCP and F-SCP are defined as follow, with the associated “scale of risk” (Marginal Super-Cells -> Non- Tornadic SC -> Weak Tornadic-SC -> Significant Tornadic- SC).

FIG. 9: The STP for AROME. Forecast from the 18 UTC analysis: 2 h and 20 UTC (left) ; 3 h and 21 UTC (right).

IV. CONCLUSIONS It seems that the USA Convective Parameters could provide additional and interesting outputs for the ALADIN FIG. 5: The formula and the “scale of risk” for the SCP (ALADIN) (9.5 Km) and AROME (2.5 km) hydrostatic and non- and the F-SCP (AROME). hydrostatic Météo-France models, respectively. For the special case of this F4 tornado in the region For ALADIN – on the left in FIGS. 6 and 7 - the of Haumont, the EHI, SCP and STP maps do not generate SCP signal clearly exhibits a narrow maximum, but too False Alarm over France, even if False Alarms and Non- North–West from the region of Haumont. The F-SCP values Detections have been more frequent on the 2009 JJA period. as seen by AROME (on the right) are more realistic, passing As a future development, it could be interesting to through the region of Haumont. The SCP and F-SCP values analyse the behaviour of the prognostic turbulent kinetic are smaller than the expected ones (for both ALADIN and energy (with indeed a strong signal in the ALADIN and AROME): from the FIG. 5 they should be above 10 units… AROME models for this Haumont’s case / not shown). It could be also interesting to modify these Convective Parameters, in order to avoid the cancellation of the index as soon as any of the (SB/ML/MU)-CAPEs are equal to 0. An attempt will be made to transform the Composite Map, developed at the Météo-France Forecast Laboratory, into the frame of the USA “Composite Parameter” and with the associated thresholds to be determined.

FIG. 6: The SCP for ALADIN (left) and the F-SCP for AROME VI. REFERENCES (right), for 2 h forecast from the 18 UTC analysis (20 UTC). Davies J. N.., 1993: Hourly helicity, instability, and EHI in forecasting supercell tornadoes. 17th Conf. On Severe Local Storms, St.Louis, MO, Amer. Meteor. Soc. 107- 111. Thomson R. L., Edwards R. and Hart J. A., 2003: Close proximity soundings within supercell environments obtained from the rapid update cycle. Wea. Forecasting., 18, 1243-1261.

FIG. 7: As on FIG. 6, but for 3 h forecast (21 UTC).

120 NOWCASTING OF THUNDERSTORMS WITHIN A WEATHER INFORMATION AND MANAGEMENT SYSTEM FOR FLIGHT SAFETY Caroline Forster1, Arnold Tafferner1, Tobias Zinner1, Hermann Mannstein1, Stéphane Sénési2, Yann Guillou2

1 Deutsches Zentrum für Luft- und Raumfahrt (DLR), Institut für Physik der Atmosphäre, Oberpfaffenhofen, Germany 2 Météo France, Toulouse, France

(Dated: 15 September 2009)

I. INTRODUCTION: THE FLYSAFE PROJECT This information must be compiled into simple messages FLYSAFE (http://www.eu-flysafe.org/) is a and graphics that are easy to interprete and enable quick European Commission funded project aiming at improving decision making. Therefore, the strategy within Cb WIMS is flight safety through the development of a Next Generation not to describe thunderstorms in any observable detail, but to Integrated Surveillance System (NG-ISS). The NG-ISS render hazard areas for aircraft as objects describing the provides information to the pilot on a number of external hazard levels “moderate” and “severe”, where “severe” hazards which address the three types of threats: traffic indicates a no-go volume of air space. Figure 1 renders a collision, ground collision, adverse weather conditions. With schematic depiction of such thunderstorm hazard objects regard to the latter, the NG-ISS is coupled to a ground-based (re-drawn from Tafferner et al., 2008) . The Cb top volumes Weather Information and Management System (WIMS) that represent the turbulent anvil area, while the Cb bottom has been designed to provide the best possible nowcast of volumes represent areas with heavy rain, hail, and the most dangerous meteorological hazards over a defined turbulence. Nested volumes indicate the two severity levels area ranging from high resolution short-range on a local “moderate” and “severe”. Note that the horizontal shape of scale to long-range forecasts on a global scale. Individual the volumes do not have to be cylinders as depicted here, but WIMS have been developed for the weather hazards icing can be polygon shaped. (ICE WIMS), clear air turbulence (CAT WIMS), wake vortex turbulence (VW WIMS) and thunderstorms (CB WIMS; Cb = Cumulonimbus). All WIMS data are sent to a ground-based weather processor (GWP). By request from an aircraft selected information about a weather hazard tailored to the respective flight corridor is passed through the GWP to the on-board NG-ISS. The NG-ISS fuses the WIMS data not only with on-board weather data, but also with data of the threats terrain and traffic in order to achieve a consolidated picture of the hazard situation. Finally, the situation is presented to the pilot by means of simple, easy to read graphics on a FIG. 2: Left: Thunderstorm cells as seen in the METEOSAT high special display together with the possible solution on how to resolution visible channel within the TMA Paris overlaid with CB avoid the hazard. top contours for 4th July 2006 1445 UTC. Mature cells in red, This paper describes the CB WIMS part of the rapidly growing in orange. Also shown are the nowcasts for 5 and system and its evaluation in operational flight tests. 10 minutes ahead in time in white and grey contours. Right: Radar reflectivity in shades of colour over the area of TMA Paris. II. THE CBWIMS APPROACH Identified Cb bottom volumes encircled in orange (severity “moderate”) and red (severity “severe”) contours at 4th July 2006 Thunderstorms are very complex weather features 1455 UTC. appearing in various shapes and sizes and corresponding life times from a few tens of minutes to several hours. The Cb top volumes are detected by satellite with DLR’s cloud tracker Cb-TRAM (Zinner et al., 2008), a fully automated algorithm for the detection and nowcasting of convection using METEOSAT data. The Cb bottom volumes are diagnosed by Météo France’s CONO algorithm (Hering et al., 2005) which detects and nowcasts convective cells using radar data. Fig. 2 shows an example of top and bottom volumes over the terminal manoeuvring area (TMA) Paris.

III. CB WIMS EVALUATION During sumer 2008, the real time Cb WIMS FIG. 1: Photography of a thunderstorm with Cb top (bluish production as well as the data flow to the GWP and up to the cylinders) and bottom (reddish cylinders) volumes superimposed. cockpit have been tested involving two research aircraft: the

121 ATR42 from SAFIRE French atmospheric research aircraft of the aircraft (see bottom panel), such an information could unit (http://www.safire.fr/) and a Metro Swearingen II have helped in delaying the turn for safer operations. operated by the dutch NLR (http://www.nlr.nl). Here, we Another situation where this extended spatial coverage could present results from the SAFIRE flights which was devoted significantly help pilots is the take-off and landing to the recording of in-situ and conventional onboard radar maneuvers where strong turns are much more constrained data for off-line evaluation of the products from CB WIMS for aligning with the runways and can occur also without (as well as ICE and CAT WIMS). Setting up a digitized any visibility in case of embedded convection. recording of the full on board radar data on SAFIRE proved to be intractable in the time schedule of the project. IV. FURTHER RESULTS AND CONCLUSIONS Therefore, a video recording of the on-board radar screen Further results of the evaluation showed that CB was settled. The Cb WIMS data were then superimposed on WIMS objects can give valuable additional information to the recorded on board radar images. the pilot in cases where CBs are located in lines one after the other and the on board radar is attenuated by the CB located closest to the aircraft. CB WIMS objects also help to distinguish ground clutter from thunderstorms in the on board signal. Expert pilots were involved in the assessment of the new products. They recognized the potential operational value of WIMS Cb, and pointed out that among WIMS CB objects, severity “moderate” bottom objects and top objects can be seen as areas outside which there is definitely no hazard. They took note that accuracy and details of the on-board radar data is nevertheless of fundamental value at shortest ranges, and that more than two levels of WIMS CB objects severity could be used. Overall, it can be concluded that:  Thunderstorms can be represented by relatively simple bottom and top volumes in a meaningful way for aviation (pilots and controllers)  CB WIMS objects generally agree well with the on board radar depiction  Cb WIMS information is free of ground clutter which is not the case for the on board radar  Cb WIMS objects provide valuable information beyond and outside of the on board radar range  Cb WIMS objects provide valuable information on Cbs which are attenuated by closer Cbs in the on board radar  Cb WIMS top objects especially valuable in radar void areas or in situations when a CB is in its developing phase and can not yet be detected by radar.

V. AKNOWLEDGMENTS This study was supported by the European Community 6th FIG. 3: On board display of radar images (colored shading) and CB WIMS objects (colored contours) during a sharp turn. Cb top Framework Programme under the EC contract AIP4- objects are indicated in orange, Cb bottom objects with severity CT-2005-516167. “moderate” in yellow and with severity “severe” in magenta. The flight path of the aircraft is marked by a red line with the aircraft’s VI. REFERENCES position indicated by a white aircraft symbol. LSE is a waypoint near Lyon. Top image at 13h20'0" on August 19th, bottom image 61 Hering, A, S. Sénési, P. Ambrosetti, and I. Bernard- seconds later. Objects are diagnosed using data from 13h20. Bouissières. Nowcasting thunderstorms in complex cases using radar data. WMO Symposium on Nowcasting and The on-board radar scans ahead of the aircraft over a Very Short Range Forecasting, Toulouse, France, 2005. sector which is usually 90 to 120° degrees wide. When the Tafferner, A., C. Forster, S. Sénési, Y. Guillou, P. Tabary, P. aircraft has to turn sharp, this can cause a temporary Laroche, A. Delannoy, B. Lunnon, D. Turb, T. Hauf, D. blindness which can be detrimental to the safety, or at least Markovic, 2008: Nowcasting Thunderstorm Hazards for to the smoothness of aircraft operations. Fig. 3 shows an Flight Operations: The CB WIMS Approach in example of such a case: on August, 19th, the aircraft reached FLYSAFE. ICAS2008 Conference, International Council Lyon (LSE waypoint) at 13h20'00" (top panel) where it of the Aeronautical Sciences Conf. Proc. (8.6.2), turned left sharply. The WIMS CB objects were depicting Optimage Ltd., Edinburgh, UK, S. 1 - 10, Anchorge, level “severe” bottom objects and a top object close to the AK(USA), ISBN 0-9533991-9-2. aircraft path (red line), at a location not yet covered by the Zinner, T., H. Mannstein, A. Tafferner: Cb-TRAM: on-board radar. Some 61 seconds later, the on-board radar Tracking and monitoring severe convection from onset showed a strong reflectivity pattern which matched very over rapid development to mature phase using multi- well the bottom object, and was also confirmed at later channel Meteosat-8 SEVIRI data. Meteorology and times. In such a case, where the WIMS CB objects also Atmospheric Physics, 101, S. 191 - 210, DOI: showed that convection was scattered at longer ranges ahead 10.1007/s00703-008-0290-y, 2008

122 5th European Conference on Severe Storms 12 - 16 October 2009 - Landshut - GERMANY

NOWCASTING THUNDERSTORM ACTIVITY ACROSS THE MEDITERRANEAN C. Price1, M. Kohn1,E. Galanti1, K. Lagouvardos2, V. Kotroni2

1Tel Aviv University, Ramat Aviv, Israel ([email protected]) 2National Observatory of Athens, Greece (Dated: 15 September 2009)

I. INTRODUCTION 15 minutes Flash floods in the Mediterranean are one of the -Minimum size of cluster size: scale 0: 100 km2, scale 1: most disastrous and damaging natural phenomena, claiming 1000 km2, scale 2: 2000 km2. lives and causing damage to property and agriculture on a large scale. Since flash flood by definition is a flood that hits III. RESULTS AND CONCLUSIONS strong and fast, it is difficult to predict. And therefore, a We assessed the success of the now-casts by need exists for an improved alert system. comparing clusters found in the now-cast with clusters found This research is part of the EU funded FLASH in the actual observations and determining if it was a hit, Project whose goal is to improve forecasting and now- miss or false alarm. This was done for different forecast casting of flash floods in the Mediterranean region. radii around the forecasted cluster, ranging between 0-5 Lightning data is immediately available and can be pixels around the cluster. detected over thousands of kilometers and can assist in improving forecasts in areas without radar coverage. 60 Min Forecast with radius 60 Min Forecast with radius Lightning is known to have a strong correlation with rain 0 5 rate and therefore prediction of lightning storms can eventually assist in prediction of intense rain fall and potential floods. In this research we attempt to now-cast thunderstorms using lightning data from the Zeus lightning detection network and the WDSS-II analysis and forecasting model (Lakshmanan et al., 2003). Using the WDSS-II model, the motion of storm cells is detected and projected

forward in time to give a now-cast of between 30 to 120 Figure 1: Success of thunderstorms nowcast for 60min lead minutes of the location and intensity of the storm cells. time. The numbers 1 (on the cluster) implies a hit, 2 implies a miss, and 3 implies a false alarm. The boxes represent the uncertainty of the nowcast, with the radius 5 implying II. DATA AND METHODOLOGY adding a 50km uncertainty around the original nowcast The data used comes from the Zeus VLF Lightning region (radius 0). The colors represent the different detection network centered at the National Observatory of clusters. Athens and which has 6 sensors throughout Europe. The Warning Decision Support System – Figure 2 represents the skill of our nowcast. The Integrated Information (WDSS-II) model from NOAA top, middle, and lower panels show the percentage of hits, NSSL is a suite of algorithms and displays for severe false alarms, and misses, according to forecast times (x-axis) weather analysis, warnings and forecasting. and as a function of the forecast radius (y-axis). (I) Using a hierarchical k-means clustering method, WDSS-II is able to define storm clusters (lightning data) at different scale sizes. Scale 0 has the smallest clusters and is therefore the most detailed, in scale 1 smaller clusters are combined to form bigger clusters, and scale 2 has the biggest clusters. (II) The motion of the storm clusters is then estimated by comparing two consecutive frames, objectively matching clusters over time. A wind field is created based on the motion estimates, and the future position of every pixel is then interpolated accordingly. ZEUS data from the entire year of 2008 was analyzed using WDSS-II, and now-casts we created for 30, 60, 90, 120 minutes. The main model setup includes: -The area of research: latitude [32 50], longitude [-8 35] -Average lightning density was calculated for a 0.1x0.1 deg grid -A 30 minutes lightning density average was calculated every 15 minutes Figure 2: Statistics of nowcasts over entire year of 2008 -The bottom limit of lightning activity: 1 flash per pixel per

123 5th European Conference on Severe Storms 12 - 16 October 2009 - Landshut - GERMANY

The blue arrow marks an 80% possibility of a VI. SUMMARY AND CONCLUSIONS thunderstorm in a radius of 0-5 pixels around the nowcasted Now-casting of lightning location and intensity is storm as a function of forecast time. This can be useful when studied using the Zeus observing system and the WDSS-II issuing warnings for potential severe weather. model. Location of the lightning clusters is automatically detected by the system on several spatial scales, and now- IV. EXPERIMENTAL NOWCASTS casting is performed for time lags of 30 to 120 minutes. Using the statistics shown in Figure 2, we have Results show the system is able to effectively detect attempted to produce real time experimental nowcasts on our lightning regions of relevance to strong precipitation and FLASH project website (www.flashproject.org). These advect them forward in time. With a good hit percentage up experimental nowcasts determine the center location of the to 120 minutes. thunderstorm in the next 30 min to 120 min. Depending on The movement vector, forecast radius and the slope the initial size of the thunderstorm nowcast (mean radius of created by the percentage of hits, can assist in defining an cluster), we forecast a circular region of activity in the next area of warning according to forecast time. 30 minutes. As we progress in time, we increase the size of The clustering analysis is also useful to derive the region (circle) to allow us to have a 80% probability that statistics on the location and intensity of lightning density. thunderstorms will occur in this region, in the next few hours (Figure 3). VII. AKNOWLEDGMENTS This work is part of the FLASH project, which is supported by the EU 6th Framework Program. More information on FLASH project can be obtained at Observed http://www.flashproject.org Cluster VIII. REFERENCES 30 min Lakshmanan, V., R. Rabin and V. DeBrunner: 2003. Forecast 60 min Multiscale Storm Identification and Forecast, Atmospheric Radius = 1 Forecast 90 min Radius = 3 Forecast 120 min Research, Vol. 67-68, pp. 367-380. Radius = 5 Forecast Radius = 10

Figure 3: Procedure to produce realtime nowcasts.

V. CLUSTER STATISTICS Information from the WDSS-II can also be used to obtain statistics of storm characteristics, such as size, velocity and intensity. The figures below (Figure 4) show the annual variation of number of thunderstorms, average storm size, average storm speed, and the maximum lightning density in these storms. A dramatic difference can be seen in most cases between summer storms and winter storms.

Figure 3: Monthly statistics on thunderstorm clusters derived from ZEUS data from the year of 2008. Number of identified storms (top left), average size (top right), average speed of storm (bottom left), and maximum lightning density (bottom right).

124 5h European Conference on Severe Storms 12 - 16 October 2009 - Landshut - GERMANY

TEMPORAL EVOLUTION OF TOTAL LIGHTNING AND RADAR PARAMETERS OF THUNDERSTORMS IN SOUTHERN GERMANY AND ITS BENEFIT FOR NOWCASTING Vera Meyer1, Hartmut Höller1, Hans-Dieter Betz2, Kersten Schmidt1

1Deutsches Zentrum für Luft und Raumfahrt, Institut für Physik der Atmosphäre, Oberpfaffenhofen, 82234 Wessling, Germany, [email protected], [email protected], [email protected] 2Physics Department, University of Munich, Germany, [email protected] (Dated: 15 September 2009)

I. INTRODUCTION extremely non-linear evolution of thunderstorms. Radar cells Improved nowcasting of thunderstorm related are identified and tracked based on the rad-TRAM hazards like lightning, hail or heavy rain would be of high Algorithm (Kober, 2009) with a reflectivity threshold of 33 benefit for weather related industries. A better understanding dBZ. Spatial nowcasts for cell position and cell shape are of the driving processes and the extreme non-linear calculated for the next 30 minutes by applying a pixel based behaviour of thunderstorms would help to improve the short- displacement vector field. term nowcasting. Lightning cells are identified based on spatially and The new cell tracking algorithm ec-TRAM (tracking temporally clustered lightning frequency maps. Total and monitoring of electrically charged cells) (Meyer, 2009) lightning data are provided by the European lightning was developed to identify, track, and record electrically detection network LINET (Betz, 2008). The network uses a active cells by combining the information of separately three-dimensional time of arrival (TOA) method to tracked lightning cells and high reflectivity fields. Based on triangulate lightning sferics in the VLF/LF regime. Besides three-dimensional lightning information and conventional the discharge time and amplitude the emission height is two-dimensional radar information the algorithm assesses calculated for every discharge event, so that in-cloud (IC) the actual cell state and records the temporal evolution of the and cloud-to-ground (CG) events are consequently cell parameters. Spatial nowcasts for cell position and cell discriminated. To remove local sensitivity variations due to shape are calculated for the next 20 to 30 minutes. non-uniform sensor spacing a minimum value of 2.5 kA for In this study, which is part of the DLR project the discharge amplitude is used for the cell tracking. To RegioExAKT, several selected thunderstorm life cycles obtain a net insensitive emission height statistic an recorded by ec-TRAM were complemented with volumetric additional height threshold of at least 5 km is applied on IC polarimetric radar information and investigated in terms of emission heights. While the amplitude threshold is used for the underlying microphysical processes. the cell tracking and the total lightning analyses the height As demonstrated in the example of a thunderstorm threshold is used only for the emission height statistics and cell from 25 June 2008 the combination of radar and does not affect the other cell parameters such as discharge lightning data has features capable to assess trend nowcasts count or discharge density. Total lightning data of three for thunderstorms. Especially the information about graupel minutes are accumulated and mapped on a discharge map. and hail formation in the mixed phase layer as warning Events are then clustered to a lightning cell if the spatial parameter for future electrical activity and the evolution of distance to the closest event is less than 3 km. Because total lightning for cell trend prognoses are promising electrical discharge is not only a hint but the direct prove of correlations which still have to be tested. electrical cloud activity, lightning cells are identified with a threshold of one event. Lightning cells are tracked using a temporal overlap instead of the ‘first-guess’ method used for II. METHOD the radar cell tracking. Every 2.5 minutes discharge events The new thunderstorm tracking algorithm ec-TRAM of the last three minutes are clustered. So every actual cell was developed at the DLR (German Aerospace Centre) in includes events, which are also used to identify the previous Oberpfaffenhofen to combine the information of cell. Now cells can be tracked via simple spatial overlap of independently tracked radar and lightning cells. In every previous and actual cells. Spatial nowcasts for cell position time step the two cell types are assigned via spatial overlap and cell shape are calculated in the same way as for radar to one single object called ec-cell (electrically charged cell). cells. Ec-cells are tracked in a hierarchical order. This approach The cell clustering and cell identification parameters guarantees the highest information content for the hybrid for both cell types are selected with the focus to achieve on cells. Because every cell type is closely and independently the one hand the most efficient cell assignment between tracked the combined cell parameters complement but do not radar and lightning cells and on the other hand reasonable affect each other. and comparable cell areas. The radar cell tracker is a further development of the Ec-cells emerge from a spatial overlap of actual radar radar cell tracker Rad-TRAM (Kober, 2009). Radar cells are and lightning cells and are tracked in a hierarchical order. identified based on the two dimensional low-level Because of the tracking method the tracks of lightning cells precipitation fields of the DWD C-band Radar in Fürholzen. are considered to be more reliable than the radar cell tracks. The DWD provides the low level scan operationally every 5 So first it is looked for tracked lightning cells to pass on the minutes on a 1 km x 1 km grid. The high spatial and ec-cell number of a previous ec-cell to the actual ec-cell temporal resolution of the data allows a closed cell tracking complex. In case of any ambiguity, the largest area is the which is especially advantageous to investigate the usually determining factor for cell association. If there is no

125 5h European Conference on Severe Storms 12 - 16 October 2009 - Landshut - GERMANY lightning cell track available in the actual time step it is 33 dBZ and lightning activity starts. Graupel and hail were looked for tracked radar cells to pass on an ec-cell number, recorded in the mixed-phase layer at 4 km height five again with the largest area as determining factor. FIG.1 minutes prior to the first obvious electrical activity. The shows a snapshot of an ec-TRAM map detail. The radar electrical activity of the storm starts with IC events during reflectivities are illustrated in blue shades, actual electrical an enhanced cell growth. The first CG event was recorded 5 discharge events, both IC and CG events, are indicated by minutes after the first IC event. The ratio of IC and CG green crosses. White contours enclose actual radar cells, the events ranges between 2 and 40 during the whole lifecycle. red contour an actual lightning cell. White lines are radar At 15:25 UTC hail is observed near the ground. At 16.30 cell tracks. The spatial cell prognoses for 10 and 20 minutes UTC both the ground precipitation field and in the are drawn as black and grey contours, respectively. The volumetric radar data indicate a cell splitting marked as yellow sign indicates Munich airport. yellow line in FIG.2. 15 minutes prior to the cell splitting total lightning frequency increases significantly by a factor of two. After the cell splitting the main electrical activity shifts to the right moving cell and then diminishes while the precipitation rate of the cell increases significantly. At 16:35 UTC the hail at the ground changes into heavy rain. Lightning activity finally stops after 140 minutes. The radar cell persists until it merges at 16:45 UTC. FIG.1: ec-TRAM snapshot of an ec-cell with actual lightning and The synoptic conditions with high CAPE and low radar cells, the radar cell track and radar cell prognoses. directional but high absolute wind shear are considered supportive to real cell splitting processes according to theory For the selected events the C-band dual-polarization (Klemp, 1978). Interpreted in terms of the cell splitting Doppler radar POLDIRAD of the DLR in Oberpfaffenhofen, theory the increase of total lightning just before the cell Germany, provides additional polarimetric radar quantities splits might be due to an updraft intensification. While the such as reflectivity, linear depolarization ratio (LDR) and cell splits, the downdraft increases intensifying the differential reflectivity (ZDR), which allow a hydrometeor precipitation at the ground and indicating cell dissipation as classification (Höller, 1994). Volumetric and cross section the cell core is ‘washed out’. scans give information about the vertical structure and the The example of the thunderstorm of 25 June 2008 hydrometeor distribution of the storm and are used to suggests several warning parameters. Hail and graupel interpret the ec-cell evolution in terms of the underlying formation in the mixed phase layer can be an indicator for dynamic and microphysical processes. ongoing charge separation in the cloud and therefore be used as warning parameter for electrical activity in the near III. RESULTS AND CONCLUSIONS future. This is also supported by the current theory for FIG.2 shows the temporal evolution of radar and charge separation in thunderstorms with ice phase (Dash, lightning parameters of a thunderstorm cell recorded on 25 2001). Beginning IC activity could be used as warning June 2008 near Munich. The upper diagram shows the cell parameter for subsequent CG events. And finally the change areas in km² as a function of time. The black line represents of hail to rain and an increasing rain rate at the ground can the lightning cell area, the grey line the radar cell area. The be an indicator for cell dissipation. lower time diagram shows the evolution of the electrical The next steps will be to identify more life cycle activity in counts per cell cluster of the storm. The black line patterns, which either support the actual warning parameters indicates the total discharge activity, the red line the IC and or can lead to new warning parameters for trend prognoses. the green line the CG discharge activity. The volumetric The warning parameters will then be used to add trend radar information is investigated but not shown in the FIG.2. prognoses to the spatial prognoses which build on the individually recorded cell histories. Finally the quality of the trend prognoses must be tested.

V. REFERENCES Betz H. D., Schmidt K., 2008: LINET-An international lightning detection network in Europe, Atmos. Res., 564- 573 Dash J. G., 2001: Theory of Charge and Mass Transfer in Ice Collisions, J. Geophys. Res., 106 20395–20402 Klemp J. B., 1987: Dynamics of Tornadic Thunderstorms, Annu. Rev. Fluid Mech., 19 369-402 Kober K., Tafferner A., 2009: Tracking and nowcasting of convective cells using remote sensing dta from radar and satellite, Meteorol. Z., 18 75-84(10). Meyer V., Höller H., 2009: Improved Tracking and Nowcasting Techniques for Thunderstorm Hazards using FIG.2: Life cycle of a thunderstorm cell recorded on 25 June 2008 3D Lightning data and Conventional and Polarimetric Radar Data, WSN09 World Meteorological Organization The first reflectivity core with a maximum Symposium on Nowcasting 2009, Whistler, Canada. reflectivity of 33 dBZ was detected at 15:04 UTC in the Höller H., 1994: Life Cycle and Precipitation Formation in a volumetric radar scan at 4 km height. 15 minutes later at Hybrid-Type Hailstorm revealed by Polarimetric and 15:20 UTC ec-TRAM recorded the cell for the first time as Doppler Radar Measurements J Atmos. Sci., 51 2500- the precipitation field at the ground exceeds the threshold of 2522

126 5th European Conference on Severe Storms 12 - 16 October 2009 - Landshut - GERMANY

ANALYSIS OF THUNDERSTORMS WITH THE DYNAMIC STATE INDEX (DSI) IN A LIMITED AREA HIGH RESOLUTION MODEL T. Schartner1, P. Névir1, G. C. Leckebusch1, U. Ulbrich1

1Freie Universität Berlin, Carl-Heinrich-Becker Weg 6-10, 12165 Berlin, Germany, [email protected] (Dated: 15 September 2009)

I. INTRODUCTION Thus, the index (DSI ≠ 0) incorporates the effects of all non- stationary, diabatic, moist processes an friction in the Convective extreme weather events like heavy atmosphere, calculated from temperature, geopotential precipitation, large hail and lightning strikes are related with height and wind velocity data. Weber and Névir (2008) have non-stationary, diabatic and moist processes. Parameters applied the DSI concept to describe synoptic scale generally used for the analyses of the potential of processes. They have shown that the DSI can identify the thunderstorms like CAPE or the KO-Index consider only the location and the strength of extra-tropical storm tracks. thermodynamical aspects. However, the state of the Claußnitzer et al. (2008) have applied the DSI to high atmosphere described by these parameters is used for the resolution numerical models of the atmosphere, like parameterization of convective activity in global and high COSMO-EU and COSMO-DE. They found a high resolution models like COSMO-DE. On the basis of the correlation between local values of DSI and precipitation in energy-vorticity theory a novel parameter is proposed to the data. Moreover, calculating the DSI with the non- diagnose non-stationary, diabatic, moist and friction hydrostatic COSMO-DE convective cells can be detected. processes. This parameter, called Dynamic State Index (DSI), shows the deviation from an adiabatic, dry, inviscid and stationary state of the atmosphere. Therefore, the DSI III. METHODS shows important atmospheric phenomena associated with diabatic, wet, frictional and non-stationary processes. For our comparison we choose the time period from st th In this study, we investigate the DSI and other 1 May until 31 May 2007. The meteorological data sets thunderstorm parameters computed from output of a limited stem from the limited area high resolution model COSMO- area model. We utilize the DSI as an activation parameter, DE of the German Weather Service (DWD). This model has and CAPE as an availability parameter. It is shown that the 50 layers in vertical direction and a spatial resolution of 2.8 combination of these two parameters gives a more reliable km. The temporal resolution of the archived data is one indication for convective activity than a single parameter. hour. The model area of COSMO-DE comprised Germany, Austria and Switzerland. For detection of thunderstorms the nowcast GmbH provided the lightning data from the II. THEORY lightning detection network (LINET). The examined parameters are CAPE and the combination of DSI&CAPE. The Dynamic State Index was innovated on basis of the energy-vorticity theory. In the physical sense, the index Our approach is the combination of an activation links the conservation of energy together with the parameter (DSI) and the availability parameter (CAPE), the conservation of Ertel's potential enstropy. An index value of Thundery Index (TI). zero (DSI=0) means the atmosphere is in an energy vorticity state, which is given by a stationary, inviscid, dry and adiabatic solution of the primitive equation (Weber and Névir, 2008). The index is defined in the following way. In this equation α and β are adjustable exponents. For our comparison we choose α = 0.6 and β = 0.5.

As verification parameters, the Thundery Case Probability – TCP (Haklander and van Delden, 2003) and

scores based on contingency tables such as Heidke Skill Π – Ertel’s Potential Vorticity Score and True Skill Score are used. θ – Potential Temperature

B – Bernoulli Streamfunction ρ – Density

127 5th European Conference on Severe Storms 12 - 16 October 2009 - Landshut - GERMANY

IV. RESULTS AND CONCLUSIONS

This chapter points out that the combination of an activation parameter (DSI) and the availability parameter (CAPE) is useful for the diagnosing of thunderstorms. Answering the question whether a lightning strike occurs or not, we take the Thundery Case Probability (TCP). The TCP for the all individual CAPE values shows a rather small dependence on CAPE for values higher than about 400 J/kg, with a maximum TCP of 50% (Fig. 1).

Fig3: Heidke Skill Score and True Skill Score for the Thundery Index (TI)

The difference between Heidke and TSS is due to the characteristics of this score. While the TSS is linked with a high probability of detection (POD), the Heidke Skill Score penalises every mistake (missing and false). So, the Heidke Skill Score is related with the false alarm rate (FAR). The Heidke skill score and the TSS for the Thundery Index have Fig1: Thundery Case Probability for CAPE maximal values of 0.37 and 0.66 respectively (Fig 3). Considering the random errors in space (2.8 km) and time (hourly data sets) for high resolution models like COSMO- Concerning CAPE alone, a thunderstorm may occur DE, the Thundery Index is in contrast to CAPE a novel or not. By contrast, the TCP has a stronger dependence from approach for the diagnosing of thunderstorm. The advantage the Thundery Index, with a maximum value of 100% of the Thundery Index is the combination of all relevant (Fig.2). dynamical processes and thermodynamical processes.

In conclusion this study reveals the potential of the DSI to forecast convective and lightning events in spatial temporal range of the mesoscale. Down to the present day, the understanding of this scale is a challenge. Therefore, further work will focus on the agreement of TI with local convective cells resolved in the non-hydrostatic COSMO- DE.

V. ACKNOWLEDGMENTS

This work is funded by the Federal Ministry of Education and Research (BMBF) and is part of the klimazwei project. The cooperation with other groups was essential for the success of this work. H.-D. Betz (nowcast GmbH) provided the lightning data from the LINET System and J. Sander (DLR) the CAPE algorithm.

Fig2: Thundery Case Probability for the Thundery Index (TI) VI. REFERENCES

Thus, the TI gives a good estimation and a definite Claußnitzer, A., Névir, P., Langer, I., Reimer, E., Cubasch, value of the occurrence of a thunderstorm event in the U., 2008: Scale-dependent analyses of precipitation COSMO-DE analysis, which is a clear improvement against forecasts and cloud properties using the Dynamic State the estimation from CAPE. Furthermore, the curve shows Index, Meteorol. Zeitschrift, 17, 813-825. that the magnitude of the Thundery Index is related to Haklander, A. J.., Van Delden, A., 2003: Thunderstorm atmospheric conditions generating thunderstorms. Predictors and their Forecast Skill for the Netherlands, If we choose the Thundery Index as a dichotomous Atmos. Res., 67-68, 273-299 thunderstorm predictor, indicating the occurrence of Weber, T., Névir, P., 2008: Storm Tracks and Cyclone thunderstorms when a certain upper threshold value is Development using the Theoretical Concept of the exceeded, we calculate the Heidke Skill Score and the True Dynamic State Index (DSI), Tellus, 60(A), 1-10 Skill Score (TSS) for all values of the Thundery Index (Fig.3).

128 5h European Conference on Severe Storms 12 - 16 October 2009 - Landshut - GERMANY A WATERSPOUT FORECASTING TECHNIQUE Szilagyi, W.

Meteorological Service of Canada, 4905 Dufferin St., Toronto, Canada, [email protected] (Dated: 15 September 2009)

I. INTRODUCTION In recent years, other focused studies on waterspout Waterspouts are a known hazard to both the marine forecasting have been conducted by Sioutas and Keul and aviation communities. Providing accurate and (2007), Kuiper and Van der Haven (2007), and timely forecasts has always been a challenge for Dotzek, Emeis, Lefebvre, and Gerpott (2009). operational meteorologists around the world. More often than not, users would be notified of a waterspout II. PRESENTATION OF PROJECT event after a report was received. No technique existed An essential part in the development of any empirical that allowed meteorologists to predict this technique is the collection of a large data sample. phenomenon in advance. Meteorological data associated with waterspout events over the Great Lakes was collected from the period of Until recently, very little research has been conducted 1988 to the present. During this period, a total of 263 in the field of waterspout forecasting. Work by confirmed waterspouts events occurred. This includes Wegener (1917), Rossmann (1961), and Golden the sighting of an unbelievable 824 individual (1974a,b, 1977) was already carried out, however, waterspouts! Also, numerous photographs and videos these studies concentrated on waterspout physical have been collected. processes and structure, not prediction. The first known attempt at developing a forecast technique was Fourteen parameters were investigated as possible made by Keltieka (1987). His technique was based on correlators to waterspout formation. Out of these, two a very limited sample of 9 waterspout events. The instability parameters (Water-850 mb temperature geographic coverage for which the technique applied difference (∆T) and convective cloud depth (∆Z)) and was also limited, being used only over Lake Erie (one one wind constraint (850 mb wind speed (W850)) of the Great Lakes of North America). were judged to be the strongest correlators.

In 1994, Szilagyi (1994, 2004, 2009) initiated an The data set was then vetted for completeness based intensive investigation on waterspout activity over the on recorded values of these three parameters. This Great Lakes, which continues at present. The reduced the total number of events that could be used technique that he developed, known as the Szilagyi from 263 to 172; meaning that out of the original 263 Waterspout Nomogram (Fig. 1), is based on a large events recorded, only 172 events had complete ∆T, sample of 172 waterspout events spanning over 21 ∆Z, and W850 values. years. ∆T and ∆Z, where then plotted for each of these events. The majority of the plotted data was then enclosed by two curves called “waterspout threshold lines” (lower and upper bound). In the area bounded by these curves, conditions are favourable for the development of waterspouts. Outside this area, waterspouts are not likely to occur (Fig. 1).

Additionally, a waterspout classification scheme was developed and indicated on the nomogram. It was noticed that each “waterspout type” was uniquely located on the nomogram (Fig. 1).

Fig. 1: The Szilagyi Waterspout Nomogram (2005). To quantify the likelihood of waterspout occurrence, the Szilagyi Waterspout Index (SWI) was developed In that same year, Choy and Spratt (1994) proposed a (Fig. 2). The SWI is a stability index and is derived forecast strategy for the east central Florida coast directly from the waterspout nomogram (Fig. 1). The using specific radar products to recognize precursor values of SWI range from −10 to +10. Waterspouts signatures to waterspout formation. are likely to occur when the SWI≥0.

Brown and Rothfuss (1998) also developed an approach which applied to south Florida and Keys.

129 5h European Conference on Severe Storms 12 - 16 October 2009 - Landshut - GERMANY

2009). Expanded use of the nomogram over other temperate climatic regions is now being considered.

Current work on the nomogram and SWI includes a collaborative project with NOAA in developing a diagnostic tool for meterologists to use in real-time. This will be a first step in the development of an operational waterspout forecast model.

Possible future enhancements to the nomogram include incorporating a low-level convergence parameter as well as correlators used by Kuiper and Van der Haven.

Fig. 2: Szilagyi Waterspout Index (SWI) IV. ACKNOWLEDGMENTS

Another advantage of the SWI is that computer The author would like to thank all meteorologists, sailors, pilots, media and members of the public in providing algorithms can now be developed using the observations, photographs and videos which were essential combination of both SWI and W850 to produce in the development of the waterspout nomogram. waterspout forecasts. In 2006, an experimental waterspout forecast model was developed. Preliminary V. REFERENCES results of the computer output look favourable. Brown, D., Rothfuss, J., 1998: An Approach to Waterspout Forecasting for South Florida and the Keys. Internal In addition to developing the nomogram and SWI, a report, National Oceanic and Atmospheric Great Lakes Waterspout climatology was constructed. Administration, Miami, Florida, USA. Also, a significant observation was made during the Dotzek, N., Emeis, S., Lefebvre, C., and Gerpott, J., 2009: study. It was noted that waterspouts were capable of Waterspouts over the North and Baltic Seas: Observations forming during windy conditions. This was contrary to and climatology, predictability and reporting. Meteorol. Z., Manuscript No. 72. the previous belief that waterspouts only occurred Golden, J.H., 1974a: The life-cycle of Florida Keys' during light wind scenarios. Another significant point waterspouts I. J. Appl.Meteor. 13, 676–692. was the sighting and confirmation of the rare “winter Golden, J.H., 1974b: Scale-interaction implications for the waterspout” (Szilagyi,1994). waterspout life cycle II. J. Appl. Meteor. 13, 693–709. Golden, J.H., 1977: An assessment of waterspout III. CONCLUSIONS frequencies along the U.S. East and Gulf Coasts. J. Appl. The waterspout nomogram has proven itself as valid Meteorol. 16, 231–236. prognostic tool and is continuously being refined with Keltieka, F., 1987: Internal report, .National Oceanic and the addition of new data every year. Meteorologists in Atmospheric Administration, Cleveland, Ohio, USA. North America now routinely use the nomogram to Keul, A.G., Sioutas, M., Szilagyi, W., 2009: Prognosis of Central-Eastern Mediterranean waterspouts. Atmos. Res. forecast waterspouts up to 2 days in advance, as was 93, 426-436. the case over Lake Huron in September 9, 1999 (Fig. Kuiper, J., van der Haven, M., 2007. The KHS-Index, a 3). newindex to calculate risk of (water)spout development. Poster, 4th ECSS Trieste, 10–14 September. Rossmann, F., 1961: Ueber Wasserhosen auf dem Mittelmeer [German, On waterspouts at the Mediterranean Sea]. Dt. Hydr. Z. 14, 63–65. Sioutas, M.V., Keul, A.G., 2007: Waterspouts of the Adriatic, Ionian and Aegean Sea and their meteorological environment. Atmos. Res. 83 (2007), 542–557 Choy, B.K., and S.M. Spratt, 1994: A WSR-88D approach to waterspout forecasting. NOAA Tech. Memo. NWS SR- 156, 24 pp. Szilagyi, W., 1994: 1994 Waterspout Report. Internal report, .Meteorological Service of Canada, Toronto. Wegener, A.L., 1917: Wind- und Wasserhosen in Europa [German, Wind- and waterspouts at Europe]. Vieweg, Braunschweig. Fig. 3: Three watersouts over Lake Huron successfully predicted using The Szilagyi Waterspout Nomogram (September 9, 1999) (photograph by the Kincardine News).

Recently, the nomogram was tested over the Mediterranean with success (Keul, Sioutas, Szilagyi,

130 5th European Conference on Severe Storms 12 - 16 October 2009 - Landshut - GERMANY

TORNADOS IN GERMANY – ACTUAL DEVELOPMENTS AT DWD Dipl. Met. Andreas Friedrich

Deutscher Wetterdienst (DWD),Frankfurterstrasse 135,D-63067 Offenbach, Germany, [email protected] (Dated: 15 September 2009)

I. INTRODUCTION Skywarn has started a new digital reporting system, which is starting to use now in DWD. This digital information will be The presentation will treat the topic Tornados from made available in future versions of the DWD the point of view of DWD as the National Meteorological meteorological workstation NinJo. The Skywarn Spotter Service (NMS) of Germany. Some information will be given reports could then be also used in future in a directly manner concerning the statistics of Tornados in Germany in the last in the DWD warning system. years and the cooperation of DWD with ESSL and Skywarn Germany. It will be reflected the different activities at DWD The operational DWD radar network consists of 16 in relation to Tornados, especially the efforts in Doppler radars. A Doppler radar is capable of discovering implementing new methods such as integration of a strong azimuthal shear regions typical for Mesocyclones. Mesocyclone detection algorithm in NinJo (DWD Weather The velocities measured by a Doppler radar within such a Presentation System). An overview of the actual warning vortex range from zero up to a maximum depending on the management for this severe weather phenomenon and an currently measured radial component of the vortex. In the outlook to future activities in this topic will be given. last two years a Mesocyclone detection algorithm (MDA), formerly developed at the Storm Prediction Centre (SPC) in Oklahoma and also in use at the Meteorological Service of II. PRESENTATION OF ACTUAL TORNADO- Canada (MSC), was transformed to the DWD radar network ACTIVITIES IN DWD and evaluated with the different synoptic conditions in central Europe. The goal is to detect Mesocyclones by a Since 2007 DWD has started his full membership in pattern recognition algorithm. The MDA is searching for the the ESSL organization. The presentation will give an typical signature, such vortices leave within a Doppler overview of the participation of DWD in the field of ESWD. velocity field, the so-called RCV (Rankine Combined In the scope of this cooperation DWD has installed a special Vortex) velocity profiles. version of ESWD in DWD. Together with ESSL a new structure was developed to archive the ESWD data in DWD. The MDA is running within the meteorological The ASCII based data set was transformed in the DWD software NinJo. It is completely embedded in the NinJo operational data base MIRAKEL. This offers the architecture and uses its features to process and visualize the opportunity that ESWD information can be used in DWD data. After the processing is done, a warning status is with other operational data sources for further investigations. generated and displayed. The status contains the severity In the DWD Intranet a special user interface is available for index. It ranges from one to five (max. severity) and is based including new reports to ESWD. There is built up a data upon a set of parameters. exchange between the ESWD data server at the ESSL site and MIRAKEL data base at DWD. There are stored all During the convective season 2009 the MDA and severe weather reports for Germany. other related new tools are actually being evaluated in the Forecast and Advisory Center of DWD in Offenbach. First One important area of activities is the topic of results of the preoperational usage of the new MDA tool will eyewitness reports. As well for the realistic climatology of be presented. In Figure 1 an example of recognized Tornados as for a possibility of an effective warning Mesocyclones is shown, in combination with reflectivity and management it is necessary to have enough and trustworthy lightning data in the area of the Rhine valley near the town eyewitness reports. Beside the cooperation with ESWD Baden-Baden. At the valid time of this picture a F2 Tornado DWD has started in 2006 the project “Severe weather was observed in the central town of Baden-Baden. observation”. On the DWD internet warning site (www.wettergefahren.de) an online forum for the public is available. Every year several thousands of reports were received over this platform. But every of these reports has to be manually checked before it will be used e.g. for warning activities in DWD.

Especially for receiving time critical and plausible reports DWD has established a close and fruitful cooperation with Skywarn Germany. Up to the beginning of 2009 Skywarn has disseminated reports from Advanced Spotters with emails in form of voicemails in an mp3 data format. With this technique aroused some problems, like difficulties in understanding of recorded handy calls and in the time critical dissemination of the relevant emails. Since 2009

131 5th European Conference on Severe Storms 12 - 16 October 2009 - Landshut - GERMANY

FIG. 1: Example of the Mesocyclone Detection Algorithm visualized with NinJo

III. RESULTS AND CONCLUSIONS Tornados are also in Germany dangerous weather phenomena. Therefore DWD, as the NMS of Germany is engaged in this topic and active in the climatology and detection of Tornados. In this field DWD is involved in the ESWD, has become a full institutional member of ESSL and built up a close cooperation with Skywarn Germany. State of the art technologies are in use for the risk assessment and warning management of Tornados. Since 2006 Tornado warnings has become an official part of the DWD warning management. All these activities will lead to an optimization of the DWD warning strategy concerning severe thunderstorms and Tornados.

IV. AKNOWLEDGMENTS

The author would like to thank all colleagues in DWD who are involved in the activities concerning Tornados, especially Marco Preiss (Business Area "Research and Development",) and also all external people who assist and cooperate with DWD. Here I have to honour the strong and successful cooperation with Dr. Nikolai Dotzek (DLR, ESSL) and Ansgar Berling (Skywarn Deutschland e.V.).

V. REFERENCES

Preiss Marco, 2008: ERAD 2008 – The 5th European Conference on Radar in Meteorology and Hydrology

132 5h European Conference on Severe Storms 12 - 16 October 2009 - Landshut - GERMANY

ARE AUSTRIAN RADIO WEATHER WARNINGS USER-FRIEDLY? Keul, A.G.1, Holzer, A.M.2, Sterzinger, P.2, Rudolf, S.1, Reinmüller, A.1 & Messerklinger, S.1

1Psychology Department, Salzburg University, Austria, [email protected] 2ORF Austrian Broadcasting Corporation, Vienna, [email protected]

I. INTRODUCTION their own words. 37 replies were tape-recorded for content analysis. Although the experimental situation with historical „Oe3 (Austrian Radio 3) weather warning: today, information did not closely resemble real life, it was tonight, tomorrow morning severe gusts around 130 expected that the arousing severe weather reports would be kmh, on mountains 180... in the afternoon increasing better remembered. wind speed, widespread damage possible, wind speed will After the recall experiment, a questionnaire asked decrease tomorrow afternoon ...“ for additional information on personal weather interest, media use and other topics not reported here. These are parts of a ORF Austrian radio short severe weather warning. The whole message took 32 seconds. Was III. RESULTS AND CONCLUSIONS it understood by listeners? Severe weather forecasting and

warning has to send an intelligible message via mass media Two transcript examples: (NDIS, 200).

Media weather forecasters know that they are major Male, city, >40 yrs: „A Ö3 [Austria radio 3] science communicators (Wilson, 2008), that their reports weather warning ... heavy wind to come, mostly in higher reach a broad audience with different perception habits and regions. Speed 130, it will happen in the night to motivation (Neumann&Russell, 1976; Ayton, 1988; Saturday and Saturday the weather will calm down.“ Berland, 1994; Doswell, 2003) and that their presentation Warning identified, windspeed correct, time error (storm modes makes the difference, even for professional users starting in the night to Friday, not Saturday) (Keul, 1980; Wehry, 1998; O’Hare&Stenhouse, 2009;

Wostal, 2009). Female, rural, >40 yrs: „Oh god, a heavy storm Wagenaar and Visser (1979) criticized a standard warning. Saturday is the peak, Sunday it will calm forecast as too long for effective storage in memory. Out of down.“ Warning identified, time error (peak on Friday, 12-32 items per message, only a maximum of 5-9 could be calming down Saturday) reproduced. Selective listening further reduced recalled

items. Bulliard and Reeder (2001) found that self-reported Of the sample of 62 subjects, 10% remembered no understanding of broadcast UV burn times (96%) was higher information at all, 43% recalled general information and than measured comprehension (65%) which is called 47% also report details (correct: 2 to 4, incorrect: 1). No „overconfidence“. Gigerenzer et al. (2005) pointed out statistical differences existed between remembered fair- and problems with the understanding of probabilistic forecasts severe-weather reports as well as between report lengths. („30% chance of rain“) for untrained people. Gender, age and education had no influence on recollection. Consequently, further research is needed to optimize Rural residents recalled significantly more data than city the format of weather news reports, particularly in the case residents [Chi²=11.49, p<.01]. of severe weather warnings. Existing tests were mostly done For a comparative content analysis of the original by weather professionals, only few by linguists (Sevchenko warning texts versus transcripts of 37 taped respondent texts, & Uglova, 2006) or psychologists. three codes were used: A+ general weather situation identified II. PRESENTATION OF RESEARCH D+ weather detail correctly recalled D- weather detail falsely recalled In 2008, a field experiment was organized by Salzburg University in cooperation with ORF, the Austrian Test run n A+ D+ / D- Broadcasting Corporation. The second and third authors are Vienna senior radio forecasters, the first author a Salzburg Fair-weather, short 9 6 17 / 4 environmental psychologist, the fourth, fifth and sixth fair-weather, long 10 4 14 / 6 authors students of his social research seminar. stormwarning, short 9 7 14 / 3 The experiment used four Vienna ORF radio weather stormwarning, long 9 8 20 / 7 reports spoken by the third author. Two of them dealt with sum 37 25 65 / 20 fair weather, two with severe weather (approach of a major TABLE 1: Results of 37 tape-recorded memory tests. storm), each of the pairs in a long and a short version. 62 (mostly rural) interviews took place in Upper Table 1 shows the cumulative absolute values for the Austria and Salzburg Province with 31 male and 31 female memory tests. The mean weather situation (A+) recall for respondents. The mean age was 38.7 years (range: 17-75 short reports was FW 67% and SW 78%. For long reports, it years). The quota sample was listening to four ORF radio was FW 40% and SW 89% – fair-weather was better tape versions (fair weather 31 sec. „short“/56 sec. „long“, memorized in the short report, stormwarning recall was storm 32 sec. „short“/52 sec. „long“). One of the four equal for short and long reports. weather reports was played at random for every subject who was asked to repeat the message immediately afterwards in

133 5h European Conference on Severe Storms 12 - 16 October 2009 - Landshut - GERMANY

Table 1 gives the absolute numbers for correct Neuman, W.R., 1976: Patterns of recall among television weather details (D+). For short reports, the means are FW news viewers. Public Opinion Quarterly, 40, 115-123. 1,9 versus SW 1,6; for long reports, FW 1,4 versus SW 2,2 – NDIS, 2000: Effective Disaster Warnings. Report by the more fair-weather details were recalled for short reports, Working Group on Natural Disaster Information Systems, more stormwarning details for long reports. Subcommittee on Natural Disaster Reduction. Table 1 also contains the cumulative absolute Washington, DC: NDIS. numbers for falsely remembered weather details (D-). For O'Hare, D., Stenhouse, N., 2009: Under the weather: An short reports, the mean false weather details are FW 0,4 evaluation of different modes of presenting meteorological versus SW 0,3; for long reports they are FW 0,6 versus SW information for pilots. Applied Ergonomics, 40, 4, 688- 0,8 – more false details appear after long reports. 693. Shevchenko, T., Uglova, N, 2006: Timing in News and Test run ORF male female Weather Forecasts: Implications for Perception. Paper for words mean mean Speech Prosody. Dresden, Germany, May 2-5. Fair-weather, short 51 27.2 26.8 Wagenaar, W.A., Visser, J.G., 1979: The weather forecast fair-weather, long 101 32.6 32.4 under the weather. Ergonomics, 22, 8, 909-917. stormwarning, short 35 31.0 36.3 Wehry, W. Ed., 1998: Wetterinformationen für die stormwarning, long 120 52.2 37.3 Oeffentlichkeit - aber wie? [German, Weather information TABLE 2: Results of 37 tape-recorded memory tests. for the public – but how?] Berlin: Deutsche Meteorologische Gesellschaft. Looking at the ORF message number of words Wilson, K., 2008: Television weathercasters as potentially compared to what subjects recounted, the differences for FW prominent science communicators. Public Understanding and SW, for males versus females are not impressive except of Science, 17, 73-87. for males recalling the long stormwarning in more detail. Wostal, T., 2009: Personal communication, Vienna, July 31. The questionnaire after the experiment also asked for the media channels used for every day weather information compared to a severe weather situation. Table 3 gives the ranks of prefered media for weather warnings – high-speed media like TV and radio are on top, followed by text media.

1. ORF television 2. ORF radio Ö3 (Austria3) 3. ORF teletext 4. Internet www.wetter.at 5. ORF regional radio TABLE 3: Information ranking for severe weather

This emphasizes the importance of readability and intelligible wording of the message for lay people. Without paying enough attention to communication quality, weather information can stimulate rumours, false comfort or false alarms. More qualitative and experimental research, also on TV weather, seems justified.

V. REFERENCES

Ayton, P., 1988: Perceptions of broadcast weather forecasts. Weather 43, 5, 193-197. Berland, J., 1994: On reading "the weather". Cultural Studies 8, 1, 99-114. Bulliard, J.L., Reeder, A., 2001: Getting the message across: Sun protection information in media weather reports in New Zealand. New Zealand Medical Journal, 114, 67-70. Doswell, C.A.III, 2003: Societal impacts of severe thunderstorms and tornadoes: lessons learned and implications for Europe. Atmospheric Research 67–68, 135–152. Gigerenzer, G., Hertwig, R., van den Broek, E., Fasolo, B., Katsikopoulos, K.V., 2005: “A 30% chance of rain tomorrow”: How does the public understand probabilistic weather forecasts? Risk Analysis, 25, 3, 623-629 Keul, A.G., 1980: Der Wetterbericht im Fernsehen. [German, The television weather report]. Vienna: ORF Austrian Broadcasting Corporation.

134 5th European Conference on Severe Storms 12 - 16 October 2009 - Landshut - GERMANY

NEW DEVELOPMENTS IN APPLIED RESEARCH FOR SEVERE CONVECTION FORECASTING IN THE HAZARDOUS WEATHER TESTBED, NORMAN, OK, U.S.A.

John Kain1, Steven Weiss2, Michael Coniglio1, Ming Xue3,4, Fanyou Kong3,4, Morris Weisman5, Matthew Pyle6, Ryan Sobash4, Craig Schwartz5, David Bright2, Jason Levit2, Gregory Carbin2

1NOAA/National Severe Storms Laboratory, Norman, OK, U.S.A 2NOAA/NWS/NCEP/Storm Prediction Center, Norman, OK, U.S.A 3University of Oklahoma School of Meteorology, Norman, OK, U.S.A 4University of Oklahoma Center for Analysis and Prediction of Storms, Norman, OK, U.S.A 5National Center for Atmospheric Research, Boulder, CO, U.S.A 6NOAA/NWS/NCEP/Environmental Prediction Center,U.S.A 15 September 2009

I. INTRODUCTION potential have also been examined. For example, from 2007-2009 SEs included initiatives to examine CAM In the United States, convection-allowing numerical ensembles (e.g., Kong et al. 2008; Schwartz et al. 2009) and weather prediction models (hereafter CAMs) emerged as advanced data assimilation techniques (Xue et al. 2008; potentially valuable guidance tools for next-day (i.e., 12-36 Weisman et al. 2009) that show promise but are still years h integration period) weather forecasts during the 2003 away from routine operational use. BAMEX field program (Davis et al. 2004; Done et al. 2004). This paper focuses on aspects of CAM forecasts that Researchers and forecasters gained further confidence in have already had at least some impact in routine forecasting CAMs during 2004 and 2005 when they were used in operations. Forecasters at the SPC have been using CAM- separate experiments to provide guidance for severe based guidance on a routine basis since 2004, when the convection (Kain et al. 2006, 2008) and winter weather NOAA Environmental Modeling Center (EMC) first started (Bernardet et al. 2008), respectively. This confidence was generating daily CAM forecasts for the SPC (M. Pyle, EMC, largely inspired by displays of the models’ simulated personal communication; Weiss et al. 2006). The EMC reflectivity field (SRF), a field that was widely produced for forecasts were supplemented by alternative CAM forecasts the first time in these innovative tests of CAMs. The SRF from NSSL beginning in 2006. Thus, SPC forecasters have can be used to infer important details about mesoscale had several years to examine CAM output and assess its circulations (Koch et al. 2005) and the mesoscale potential operational utility, during both daily operations and organizational structures of convective systems (Done et al. annual SEs. Collaboration between SPC forecasters and 2004; Kain et al. 2006; Weisman et al. 2008). Also, it hints NSSL scientists has led to the development and refinement at the presence of an entirely new set of phenomena that are of unique CAM output fields and innovations in the way that generated during the integration of CAMs, but are absent in CAMs can be used in the operational forecasting output from traditional operational models because of environment. A few of these advances are highlighted here. relatively limited resolution and the parameterization of convective processes. II. RESULTS Since 2004, numerous aspects of CAM output have been examined each spring during intensive examination a. Generating probabilistic guidance for severe weather periods known as Spring Experiments (SEs – Kain et al. from simulated reports of severe phenomena. 2006). These annual experiments are organized by Data mining of the WRF-NSSL4 model output, along forecasters from the NOAA Storm Prediction Center (SPC) with various statistical techniques, has allowed us to identify and research scientists from the NOAA National Severe several useful “surrogates” for severe weather in 4 km Storms Laboratory (NSSL) as part of a thriving model output. For example, one of these surrogates is low collaboration between researchers and practitioners at the to mid level mesocyclones, as characterized by high values Hazardous Weather Testbed in Norman, OK. In general, the of updraft helicity (UH - see Kain et al. 2008). UH maxima annual SEs are designed to “beta test” research tools or have proven to be useful proxies for supercells, which are concepts that could be implemented in forecast operations associated with a variety of severe weather at the ground. within 1-2 years. In addition, some topics with longer-range Using a technique described by Brooks et al. (1998) and WRF-EMC4 WRF-NSSL4 WRF-CAPS4 WRF-CAPS2 Sobash et al. (2009), density Dynamic Core NMM ARW ARW ARW plots of these surrogates can be Horiz. Grid (km) 4 4 4 2 generated and used to produce Vertical Levels 35 35 51 51 probabilistic forecasts of severe PBL/Turb. Param MYJ MYJ MYJ MYJ weather. This strategy can be applied to output from a single Radiation (SW/LW) GFDL/GFDL Dudhia/RRTM Dudhia/RRTM Dudhia/RRTM model, or multiple models (i.e., Initial Conds. 32 km NAM 40 km NAM 12 km NAM 12 km NAM an ensemble). For example, Fig. Initial Time 0000 UTC 0000 UTC 0000 UTC 0000 UTC 1 demonstrates that this Table 1. Model configurations referenced in this paper. approach can help highlight

135 5th European Conference on Severe Storms 12 - 16 October 2009 - Landshut - GERMANY regions of severe convective activity quite effectively and thresholds. For example, the WRF-NSSL4 produced a concisely, even when only a single deterministic forecast is frequency bias for next-day QPF of close to 1 for used for input. accumulations up to about 50 mm over three-hour time intervals (Fig. 2). Furthermore, for precipitation rates above a a few mm/3h it performed significantly better than the NAM as measured by threat scores and “neighborhood” metrics such as fraction-skill scores (not shown). These results and other related studies (e.g., Weisman et al. 2008; Clark et al. 2009) suggest that there is a significant benefit to be gained in numerical prediction by eliminating convective parameterization and decreasing grid spacing to the point where this action can be justified. There seems to be general agreement that approximately 4 km spacing is adequate for this purpose. Yet, most convective clouds are very poorly resolved on a 4 km grid, so it is natural to ask how much additional benefit would be gained by further increases in resolution. Separate studies of SE model output by Kain et al. (2008) and Schwartz et al. (2009), coupled with subjective assessments during SE2005 and SE2007, show that an additional doubling of resolution b (decreasing grid length from 4 to 2 km) results in little, if any, added value for guidance related to the timing, location, and mesoscale evolution of convective activity (Fig. 3).

Fig. 1. a) observed reports of severe storms and b) surrogate reports of severe convection from the WRF-NSSL4 forecast, both plotted in terms of report density (see Sobash et al. 2009), valid for the period 1200 UTC 8 May – 1200 UTC 9 May 2008. b. Quantitative Precipitation Forecasts (QPFs) and sensitivity to horizontal resolution In the U.S., heavy rainfall does not officially fall under the category of severe convective weather, but convective rainfall prediction remains one of the most challenging forecast problems in the U.S. and elsewhere. Part of the reason for this ongoing difficulty is that NWP guidance for Fig. 3. Fractions skill score (FSS) as a function of radius of QPF has been deficient, especially in the warm season when influence for the NAM (12km), WRF-CAPS4 (4km), and WRF- convection predominates. For example, EMC’s operational CAPS2 (2km) model configurations, aggregated during 1800- North American Model (NAM – see Black 1994), the 0600 UTC (f.21-f.33) over all days of SE2007 using primary 1-3 day forecast model in the U.S., over-predicts the accumulation thresholds of (a) 0.2 mm/hr, (b) 0.5 mm/hr, (c) 1.0 coverage of lighter precipitation and under-predicts coverage mm/hr, (d) 2.0 mm/hr, (e) 5.0 mm/hr, and (f) 10.0 mm/hr. of heavy accumulations (Fig. 2). But preliminary results III. SUMMARY from daily forecasts of the WRF-NSSL4 over a two year period suggest that properly configured CAMs can provide In the United States National Weather Center in much better guidance, especially for higher precipitation Norman, OK, there is a thriving Hazardous Weather Testbed built upon the mutual collaboration between research scientists and operational severe weather forecasters. This collaboration has led to unique and innovative advances in both forecasting and research in recent years. Advances related to CAMs have had a significant impact on the development and application of high-resolution models as guidance tools for the prediction of severe convective weather. Fig. 2. Bias scores for an aggregate of 3-hourly QPFs from daily 18-36h WRF-NSSL4 and NAM forecasts over the central IV. REFERENCES (Available at U.S., covering the period from April 2007 – April 2009. http://www.nssl.noaa.gov/users/kain/public_html/ECS S-ext-abs_w_refs.pdf)

136

A CLOUD MODEL STUDY OF WIND SHEAR EFFECT ON THE SATELLITE OBSERVED STORM TOP IR FEATURES

Pao K. Wang

Department of Atmospheric and Oceanic Sciences, University of Wisconsin-Madison, 1225 W. Dayton Street, Madison, WI 53706, USA. [email protected] (Dated: 15 September 2009)

I. INTRODUCTION cold-V/U appears when the wind shear is stronger. In this paper, we will use a cloud resolving model Satellite observations are very useful for WISCDYMM to perform simulations of storms under thunderstorm studies and forecasting or nowcasting. They different wind shear conditions and analyze the cloud top can provide information about storms that occur in locations temperature field to see if this field exhibit the observed IR where no ground observations are available. They can cover features. We will compare these temperature fields to study a very wide area over the earth, even global and be the effect of wind shear. continuous in time if necessary. The model we will use is the same as that used in But because satellite observation is a remote sensing some of our previous studies (Wang, 2003, 2004, 2007a, b) technique, it must be correctly interpreted in order to be and the details can be found in these references. effective in its usage. Because a satellite views a storm system from the top, it mainly sees the top layer of the III. RESULTS AND CONCLUSIONS storm. While it is possible to use some features on the storm top purely as indicators of the storm behaviour (through, for We have obtained preliminary results that example, purely correlation studies), it is much better if we basically confirm positively the hypothesis made in can understand the physical mechanisms responsible for the the previous section, namely, the cold ring feature is generation of these features. Such understandings will associated with weak wind shear conditions whereas eventually lead to much better and more reliable forecasts of the cold-V/U is associated with the higher wind shear storms. One way to understand such physical mechanisms is condition. Fig. 1 and 2 show the examples of the storm to perform numerical storm model studies to simulate certain top temperature field corresponding to these two wind storms to see if the model can reproduce the features shear scenarios. observed. If this is successful, then one can use the physics built in the model to track the responsible mechanisms. This is the approach of the present paper. The present paper will focus on the infrared features on top of storms as observed by satellites. Infrared images are available day and night, and thus can provide information when visible data are not available. Important major IR features atop thunderstorms have been studied previously (see, for example, Heymsfield and Blackmer, 1988; Wang et al., 2001; Setvak et al., 2007).

II. SATELLITE OBSERVED STROM TOP IR FEATURES

The storm top IR features observed by meteorological satellites include the enhanced-V/U or cold- V/U, close-in warm area (CWA), distant warm area (DWA), warm-cold couplet, and cold area (CA) (see heymsfield and Blackmer, 1988). A new feature, called the cold ring (CR), was reported by Setvak et al. (2008). In some storms, the CR eventually evolved into the cold-V. It Fig. 1 Cloud top temperature field of a simulated is this behaviour that motivated the present study. thunderstorm at t = 1800 sec. The storm is initialized by an idealized sounding with the assumption that the wind increases from zero at We propose that this behaviour is mainly due to the surface to 20 m/s at z >= 13 km. The cloud top temperature field the wind shear. The different wind shear that is present at reveals an elongated cold ring along the outer edge of the anvil with different level and different time causes the storm top IR warmer temperature inside. The coldest point is at the upstream “apex” whereas the warmest point is located in the downstream part features to change their characteristics. We hypothesize that of the anvil but close to the coldest point. The temperature inside the the cold ring feature occurs when the storm top is located at ring is not uniform but is apparently associated with the cloud top a level where the wind shear is relatively weak, whereas the wave motions.

137

Fig. 2 Cloud top temperature field of the simulated CCOPE supercell at thunderstorm at t = 7200 sec. The storm is initialized by a sounding as reported by Wang (2003). The cloud top temperature field reveals an enhanced-V/U (cold-V/U) at the upstream and a general warm area in the downstream part of the anvil. The CA, CWA, DWA and the warm-cold couplet features as described in Heymsfield and Blackmer (1988) are also reproduced well.

Currently, we are making more detailed analysis of the model results and the comparison of the modelled cloud top temperature fields with that observed by satellites. These results will be reported in the conference.

IV. AKNOWLEDGMENTS

This work is partially supported by NSF Grant ATM- 0729898.

V. REFERENCES

Heymsfield, G.M., Blackmer, R.H., Jr., 1988. Satellite- observed characteristics of Midwest severe thunderstorm anvils. Mon. Wea. Rev. 116, 2200-2224. Setvák, M., Rabin, R.M., Wang, P.K., 2007. Contribution of the MODIS instrument to observations of deep convective storms and stratospheric moisture detection in GOES and MSG imagery. Atmos. Res. 83, 505-518. Wang, P.K., Lin, H.M., Natali, S., Bachmeier, S., Rabin, R., 2001. A cloud model interpretation of the enhanced V and other signatures atop severe thunderstorms. Preprints 11th Conf. Satellite Meteorology and Oceanography, 15-18, Oct., 2001, Madison, Wisconsin, 402-403. Wang, P.K., 2003. Moisture Plumes above Thunderstorm Anvils and Their Contributions to Cross Tropopause Transport of Water Vapor in Midlatitudes. J. Geophys. Res., 108(D6), 4194, doi: 10.1029/2003JD002581. Wang, P.K., 2004. A cloud model interpretation of jumping cirrus above storm top. Geophys. Res. Lett., 31, L18106, doi:10.1029/2004GL020787. Wang, P.K., 2007a. The Wisconsin Dynamical/Microphysical Model (WISCDYMM) and the use of it to interpret satellite-observed storm dynamics, in Measuring Precipitation from Space EURAINSAT and the Future. Edited by V. Levizzani, P. Bauer, and F.J. Turk, Springer, 435-446. Wang, P.K., 2007b. The thermodynamic structure atop a penetrating convective thunderstorm. Atmos. Res. 83, 254-262.

138 5th European Conference on Severe Storms 12 - 16 October 2009 - Landshut - GERMANY

Recent advances in precipitation nowcasting at the RMI of Belgium: storm severity product

M. Reyniers1, L. Delobbe1, P. Dierickx2, M. Thunus2, C. Tricot1 1Royal Meteorological Institute (RMI) of Belgium, [email protected] and 2Hydrological service of the Walloon administration (Service public de Wallonie), Belgium (Dated: September 15, 2009)

I. INTRODUCTION

We report on the development at the RMI of an operational storm severity product, in which radar data are combined in real-time with Intensity-Duration-Frequency (IDF) informa- tion, in order to get an overview of the return period of an on- going event. The product is designed to be useful especially in situations with extreme local rainfalls causing flash floods. A stationary precipitating storm cell is a typical example of such a situation. This can happen for example if the cell movement is opposite to the direction of the global flow. Such situations can be dangerous since large amounts of rain are accumulated in the same basin in a short period of time. These situations are, however, hard to recognise on single images, and are only discovered when studying closely the radar animations. An example of such a situation is shown in Fig. 1. FIG. 2: Scheme of the dataflow of the storm severity product.

II. IDF MAPS AND RADAR DATA not sooner than 1993, and in which not more than 10% of the values were missing. With these criteria, the final number of IDF curves give the relation between rainfall intensity (I), daily stations used in the project decreased to 184, while the the duration (D) of the accumulation and the return period (F). number of 10-min stations decreased to 22. IDF maps for Belgium were recently determined by the RMI The storm severity product was implemented for two (Mohymont & Demaree,´ 2006). The RMI has gauge data of radars. The first one is the weather radar of Wideumont roughly 375 stations. 345 of these stations are part of the cli- (2001), in the south of Belgium, which is owned and operated matological network, reporting daily accumulation values; the by the RMI. The second one is the radar de l’Avesnois (2005), remaining 30 stations have an update frequency of 10 min- installed by Met´ eo-France,´ with a financial participation of the utes. In order to achieve a homogeneous data set to calculate Walloon Region. Both radars are C-band radars and generate the IDF curves, only those stations were selected with a time a pseudo-CAPPI every 5 minutes, which is used here as in- series of minimum 25 years, starting before 1968 and ending put for the product. The relation Z = aRb with a = 200 and b = 1.6 is applied to convert radar reflectivities into rainrates.

III. METHOD

The preparatory work consisted of the interpolation of the IDF maps to the radar grid of the accumulation images. In the operational context, the following steps are executed every time a new radar image (pseudo-CAPPI) becomes available: • Calculate the precipitation accumulations for different time windows in a computational cheap way. The accu- mulations are real-time, so the time windows for these accumulations are “running”; • Combination of the calculated accumulations with the IDF grid to real-time “return-period images”;

FIG. 1: An example of an event with stationary cells (Wideumont • Combination of “return-period images” to one single radar, 11 June 2007, 16h30-19h00 UT). The green arrow indicates return-period image as the final output of the product. such a cell. For every pixel on the map, the maximum of the return

139 2

periods for that pixel is taken. This maximum is then a measure for the “severity” of the event as it develops. The method is schematically illustrated in Fig. 2. The real- time accumulations are calculated for the following eight du- rations: 10 min, 20 min, 30 min, 1 h, 2 h, 6 h, 12 h and 24 h. Every time a new radar image arrives (in normal operation, every five minutes for both radars), the product is updated us- ing this latest image.

IV. RESULTS AND DISCUSSION

In Fig. 3 an example of the final product is shown. It is the same case that was shown in Fig. 1. The map on top shows the maximum return period of the rainfall for the past dura- tions mentioned above, and calculated following the scheme in Fig. 2. Since the IDF curves were only calculated for the Belgian territory, the return period map is limited to Bel- gium as well. The region that was marked in Fig. 1 with a green arrow as the location of a stationary cell producing very large local rainfalls, is indeed characterised by a very long re- turn period in Fig. 3. The second map (bottom) specifies for which duration this maximum return period is reached, so it expresses at which timescale the most severe rainfall occured. The real-time return period map can be used to quickly es- timate the “severeness” of a given event. However, the end- users should certainly be aware of the limitations of the prod- uct. For example, it is well known that for high radar reflec- tivities, the calculation of a reliable rainrate is hampered by FIG. 3: Result of the storm severity product for the same situation the contamination of hail. Therefore the highest return peri- that was shown in Fig. 1 (radar de l’Avesnois). The numbers between ods (say >30y) are the least reliable, and should only be used brackets in the legend of the bottom map denote the number of radar qualitatively. In general, the return periods produced by the files that were used for that particular accumulation duration. product should never be used as real and validated climato- logical values, but only as indicative values. Another source of error is the fact that the real-time accu- severity. Nevertheless, the product will be a valuable now- mulated rainrates are calculated with radar data, while the IDF casting tool for the real-time evaluation of the severity of an curves are based on historical gauge data. A possible solu- ongoing event, and it will allow fast reaction by the hydrolog- tion to decrease this uncertainty could be the merging of the ical service in case of potential flash floods, without running a radar data with real-time gauge data. Our group (Gouden- time-consuming hydrological model. hoofdt & Delobbe, 2009) recently studied different methods to merge radar data and rain gauge data. However, since the storm severity product must be available as soon as possible VI. ACKNOWLEDGEMENTS after the receipt of a new radar file, this merging cannot be applied here: the gauges in our network do not have the same This research is supported by the Walloon Region (Service update frequency as the radar images. Moreover, the product Public de Wallonie). We are indebted to Met´ eo-France´ for also focusses on events with very local rainfall; in these sit- providing us the radar data of l’Avesnois. uations, the radar-gauge merging is less efficient due to the high spatial variation of the rainfield that cannot be accurately captured by a gauge network, even if it is very dense. VII. REFERENCES

Goudenhoofdt E., Delobbe L., 2009: Evaluation of radar- V. CONCLUSIONS gauge merging methods for quantitative precipitation esti- mates. Hydrology and Earth System Sciences, 13 195-203 We have developed a new product at the RMI for the real- Mohymont B., Demaree´ G.R., 2006: final report time detection of heavy local rainfall. Due to the large uncer- IDF curves for the Walloon region. Ministere` Wal- tainties in radar-based rainfall accumulations and the derived lon de l’Equipement´ et des Transports, available on return periods, it offers only a qualitative view on the storm http://voies-hydrauliques.wallonie.be/

140 5h European Conference on Severe Storms 12 - 16 October 2009 - Landshut - GERMANY

PREDICTABILITY OF EXTREME STORM EVENTS IN THE STATE OF SÃO PAULO, BRAZIL Gerhard Held , Ana Maria Gomes, Mateus Teixeira, José Marcio Bassan and Geórgia Pellegrina

Instituto de Pesquisas Meteorológicas-IPMet, São Paulo State University-UNESP, Bauru, Brazil, [email protected] (Dated: 15 September 2009)

I. INTRODUCTION convection parameterization of Betts & Miller and Kain- The State of São Paulo is situated in the south-eastern Fritsch, respectively (Betts and Miller, 1993; Kain, 2004). region of Brazil and is characterized by summer rain and In this paper we shall show which of these new relatively dry winter months. However, severe storm events outputs are the most suitable for predicting a severe event up can occur any time of the year, but during summer they have to 48 hours ahead of its possible occurrence, which is then a more tropical character, while from about mid-April to used to alert regional emergency services for the possible September they are usually associated with baroclinic need of a standby (especially important before weekends). synoptic situations and cold fronts advancing from southern On the day of severe thunderstorm development, Brazil in a north-easterly direction into the western and NCAR´s Software TITAN is deployed (Dixon and Wiener, central State of São Paulo. 1993), utilizing observations from the two S-band Doppler The Meteorological Research Institute (IPMet) of radars, to issue alerts of possible heavy rainfall, probability UNESP operates two S-band Doppler radars, located at of hail or extremely strong winds on the time scale of 60 – Bauru (Lat: 22°21.5’ S, Lon: 49°1.7’ W, 624 m amsl) and 30 min. Presidente Prudente (Lat: 22°10.5’ S, Lon: 51°22.5’W, 420 m amsl), respectively (Figure 1). The main III. RESULTS AND CONCLUSIONS characteristics of both radars are: 2° beam width, 450 km The following cases had been selected on the basis of range for surveillance mode and 240 km in volume scan severe damage that had been reported in News Papers, on mode, with 11 (15) elevations from 0.3° to 35°, 1 km (250 TV, on Websites or documented by the Civil Defence m) radial and 1° azimuthal resolution, and a temporal Organization of the State of São Paulo: resolution of 15 minutes or less, recording and archiving 27 February 2004, when heavy rain was accompanied by an reflectivity, radial velocity and spectral width (numbers in extremely severe wind storm, resulting in the destruction of brackets are applicable from 2007). about 30 high-voltage transmission towers, as well as

damage to more than 50 houses in several small towns. 25 May 2004, when a supercell, lasting 8,5 hours, and two tornado-spawning cells were tracked by IPMet´s radars, causing the death of 2 and severe injuries to 51 people, as well as great damage to sugar plantations. The tornadoes had a strength of F2-3 and F2, respectively, on the Fujita scale. 24 May 2005 was another day, when an F3 tornado was observed, killing one person and totally destroying an industrial suburb and damaging many buildings and structures in other parts of the town, resulting in a total loss

of USD 42 million. FIG. 1: Doppler radar network of IPMet (BRU = Bauru; PPR = 29 March 2006, when a strong frontal event provoked the Presidente Prudente), showing the 240 and 450 km range rings. formation of a bow-echo, which generated extremely strong wind gusts at its leading edge (almost tornado-like), causing II. PRESENTATION OF RESEARCH enormous damage in two towns in the centre of the State. The most severe 7 events during the last five years 19 January 2007 marked the occurrence of a small tornado (2004 – 2008) had been selected to verify their predictability in the western State during the early hours of the morning. It from about 48 hours before the occurrence up to the caused damage to some houses, uprooted large trees and nowcasting time scale of 30 – 60 min before reaching a destroyed an electric transmission line. town, city or industrial installations, like high-voltage power 24 July 2007 was characterized not only by the unseasonal lines, etc. heavy rainfalls over the State, but also by a severe hail IPMet, in collaboration with CPTEC, runs the Meso- swath, destroying the orange crop in the NW regions. Eta model, centered over the Bauru radar, twice daily 29 October 2008, when a microburst near Bauru destroyed 6 (initiated at 00 and 12 UT) with a temporal resolution of 3 high-voltage transmission towers and caused other damage. hours up to 72 hours. Its domain amply covers the State of Detailed case studies of some of these storms have São Paulo at a resolution of 10x10 km horizontally with 30 already been presented, analyzing the synoptic situations and levels from 1000 to 100 hPa. This model has been radar and lightning observations, using TITAN in Archive specifically fine-tuned to compute additional convective mode (Held et al., 2006, 2007b; Gomes and Held, 2006, parameters (SR helicity, BRN Shear, supercell index, etc), as 2009) and simulating a forecast between 30 and 60 minutes, well as the generation of vertical profiles (Skew T – Log P) which can then be compared to the actual situation. There- at any specified grid point (Held et al., 2007). Furthermore, fore, the findings from the synoptic and radar analysis will each run of the model is executed twice, using the only be briefly dealt with in this paper, while the emphasis

141 5h European Conference on Severe Storms 12 - 16 October 2009 - Landshut - GERMANY will rather be on the medium-term Meso-Eta forecasts, viz. Figure 3 shows the TITAN analysis of the storm track mostly between 24 and 48 hours for most reliable results. at 14:45 UT (11:45 LT, light blue) with the past (yellow) The severe windstorm, which occurred on 29 March and the future (forecast for 30 min). The predicted cell track 2006 at the locations P and SB in the central State, as a at ±90 km.h-1 is exactly across the towns P and SB. Figure 4 result of the strong bow echo traversing the State of São demonstrates the SSS (Storm-Structure-Severity) Index at Paulo, is used to demonstrate the Meso-Eta medium-term 14:22 UT (11:22 LT) with 30 min forecast matching exactly and TITAN nowcasts, as they are being exploited in IPMet´s the area where the wind storm occurred. Operational Center. Figure 2 shows the 39-hour forecast of The onset time of a severe event and the region where the CAPE for the Betts-Miller and Kain-Fritsch convection it could be expected, is generally well predicted by the algorithms, respectively. The model was initiated on 27 Meso-Eta model 24-48 hours ahead, while TITAN captures March, 00 UT and the forecast is valid for 29 March, 15 UT the nowcast range very well. (12 LT), about the time when the wind storm peaked.

FIG. 4: TITAN SSS Index on 29 March 2006 at 11:22 LT with a 30 min forecast (red arrows). Index 4 (blue) indicates very strong wind.

IV. AKNOWLEDGMENTS The authors would like to thank J. M. Kokitsu for the implement- ation of TITAN routines used in the analysis and Drs J. Wilson and M. Dixon of NCAR for facilitating the implementation of TITAN at IPMet/UNESP. J.L. Gomes and G. Sueiro of CPTEC are acknowledged for providing the historic input data sets and for advice on the model installation and occasional upgrading

V. REFERENCES Betts, A.K. and Miller, M.J. 1993 The Betts-Miller scheme. Chapter in The representation of cumulus convection in

FIG. 2: Meso-Eta run on 27 March 2006, 00 UT with 39 hour numerical models of the atmosphere. Eds. K.A. Emanuel and forecast for the CAPE, using Betts-Miller (top) and Kain-Fritsch D.J. Raymond. American Meteorological Society. (bottom) convection parameterization Dixon M. and Wiener G., 1993: TITAN: Thunderstorm Identification, Tracking, Analysis and Nowcasting - A radar- based methodology. J. Atmos. Ocean. Technol., 10, 785-797. Gomes A.M. and Held G, 2009: Severe Winter Storms over the Western and Central State of São Paulo, Brazil. Proceedings, 5th European Conference on Severe Storms, Landshut, Germany, 12-16 October 2009, 2pp. Gomes A.M. and Held G, 2006: Identificação, rastreamento e previsão de tempestades severas, Parte II: Evento de ventos intensos. Proceedings, XIV Congresso Brasileiro de Meteorologia, (CD ROM), Florianópolis, 27 de novembro – 01 de dezembro de 2006, SBMET, 6pp. Held G., Gomes A.M., Naccarato K.P. and Pinto O. Jr., Nascimento E., 2006: The Structure of Three Tornado-Generating Storms Based on Doppler Radar and Lightning Observations in the th State of São Paulo, Brazil. Proceedings, 8 International Conference on Southern Hemisphere Meteorology and FIG. 3: TITAN storm track on 29 March 2006 at 11:45 LT just west of P, with a 30 min forecast traversing SB. The annotation is cell Oceanography, Foz do Iguaçu, 24-28 April 2006, 1787-1797. speed, varying between 80–95km.h-1. Cell envelopes every 15min. Held G., Gomes J.L. and Nascimento E.L., 2007a: Forecasting Severe Weather Occurrences in the State of São Paulo, Brazil, th The Betts-Miller output indicates high CAPE values Using the Meso-Eta Model. Proceedings, 4 European Conference on Severe Storms, Trieste, Italy, 10-14 Sept. 2007. (3000 – 3500 J.kg-1) for the western part of the State with a Held G., Gomes A.M., Naccarato K.P. and Pinto O. Jr., 2007b: slightly isolated small high in the region where the wind Signatures of Severe Thunderstorms for Nowcasting in the State storm will occur. In contrast, Kain-Fritsch predicts in the of São Paulo, Brazil.l. Proceedings, 4th European Conference on -1 same region only up to 1500 J.kg , clearly under-estimating Severe Storms, Trieste, Italy, 10-14 September 2007, 2pp. the severe storm system that would traverse the State during Kain J.S., 2004: The Kain-Fritsch convective parameterization: An update. J. Appl. Metor., 43, 170-181. the morning of 29 March 2006.

142 5th European Conference on Severe Storms, 12-16 October 2009-Landshut-GERMANY

NOWCASTING OF SEVERE STORMS AT A STATION BY USING THE SOFT COMPUTING TECHNIQUES TO THE RADAR IMAGERY

Sanjay Sharma1, Devajyoti Dutta2, J. Das3, R.M. Gairola4

1 , Kohima Science Collage, Kohima, Nagaland, 797002, India, [email protected] 2 Kohima Science Collage, Kohima, Nagaland, 797002, India, [email protected] 3. Ex-Indian Statistical Institute, Kolkata, India, das. jyotirmay @gmail.com 4 MOG, Space Application Centre, Ahmedabad, India, [email protected]

(Dated: 14 September, 2009)

I. INTRODUCTION successive imageries at t and t + ∆t time interval (c) tracking The severe storms pose a great threat to lives of the precipitating structures and (d) training of the and property of man kind. An accurate rain nowcast extracted input/output image features for the nowcasting from weather radar can identify the potential for heavy of rain by using the ANN technique. rain and possibility for flash flooding (Vieux and Bedient ,1998). By virtue of higher temporal sampling Morpholo Matching of Tracking Training of the precipitating systems, the radar based gical the two of the of the observations pose a good candidate to deal with the analysis successive precipitate input and nowcasting problems. The nowcasting of rainfall from of DWR precipitatin- -ng images output radar involves the identification of precipitating reflectivit- g structure parameter y images. -s with the systems, its evolution and movement. The help of conventional short term forecast of both position and ANN size is based on a weighted linear fit to the storm history data (Rinehart, and Garvey 1978, Dixon and Winner 1993). The use of artificial neural network (ANN) has been recognized as a promising way of making prediction on time series data. This method is Fig 1: Block diagram of methodology of nowcasting of non parametric where it does not required the severe rain assumption of any form of model equations. The main features of ANN is its ability to map input data to The inputs from the radar imageries are provided to the ANN output data to any degree of non linearity. Many at T(0), T(0+1/2) and T(0+1) hrs and the nowcasting is carried researchers have utilized this technique to predict the out with 2 hrs lead time at T (0+3) hrs. The architecture of rainfall by using the data from various platforms such the ANN is shown in figure 2. which consists of one input as model output (Kuligowski and Barros, 1998), layer (15 nodes), two hidden layer (with 35 and 25 nodes satellite (Rivolta et al. 2006), and radar (Denceux and and one output layer ( with 01 node). Rizand, 1995)

II. PRESENTATION OF RESEARCH The main objective of the present work is to develop a soft computing based methodology to nowcast the severe rain situations over a station with the help of Doppler Weather Radar (DWR) imageries. The main advantage of using the DWR observations is that the ambiguous reflectivity images due to anomalous propagation can be discarded with the help of radial velocity measurements. For the present study the DWR facility at Satish Dhavan Space Center, Shriharikota (13.66 °N & 80.23 °E), India is utilized. For this purpose, the reflectivity (dBZ) imageries of DWR are utilized. The storm Figure2: Architecture of ANN structures are identified with the threshold values of reflectivity factor ≥ 40 dBZ. The block diagram of the The following 15 input parameters are selected for the ANN proposed methodology is shown in Figure (1). The network: 2 main components of the proposed methodology are (a) (1). Area (km ) of precipitating system for reflectivity ≥20 2 the morphological analysis of precipitating structures dBZ at T0, (2). Area (km )of precipitating system of in the DWR imageries (b) matching of the two reflectivity ≥ 40dBZ at T0, (3). Line of Sight Distance (km)

1 143 5th European Conference on Severe Storms, 12-16 October 2009-Landshut-GERMANY

between system and target station at T0 for the same input/output data set is utilized to estimate the system ≥ 20 dBZ, (4). Ratio of the change of area for coefficients for the linear multiple regression in the following th th ≥ 20 dBZ during T0 to T0+1/2 (5). Ratio of the form i.e. Y = ∑ci xi, where xi is the i input and ci is the i change of area for ≥ 40dBZ during T0 to T0+1/2, (6). coefficient. The index i is from 1 to 15. Similar types of Ratio of the change of distance of the system with statistics are also obtained from the conventional linear respect to target station during T0 to T0+1/2 for the multiple regression analysis and is presented in Table 1. system ≥ 20 dBZ (7). Deviation of the system from Overall nature of the statistics is same as observed from the line of sight during T0 to T0+1/2 for the system ≥ 20 ANN. But the accuracy is better for ANN methodology dBZ (8). Velocity of the system during T0 to T0+1/2 compared to linear multiple regression analysis. for the system ≥ 20 dBZ (9). Correlation between the Comparatively, reasonably good results are obtained by soft images during T0 to T0+1/2 for the system ≥ 20 dBZ computing method and there is significant improvement by (10). Ratio of the change of area of ≥ 20dBZ during the proposed methodology compared to linear multivariable T0+1/2 to T0+1 (11). Ratio of the change of area of ≥ regression method. 40dBZ during T0+1/2 to T0+1 (12). Ratio of the change of distance of the system with respect to target station during T0+1/2 to T0+1 for the system ≥ 20 IV. AKNOWLEDGMENTS dBZ (13). Deviation of the system from line of sight Financial grant from the department of space, govt. during T0+1/2 to T0+1 for the system ≥ 20 dBZ (14). of India, to carry out this work [No. 10/4/ 524, under Velocity of the system during T0+1/2 to T0+1 for the RESPOND programme] is thankfully acknowledged. The system ≥ 20 dBZ (15). Correlation between the authors are thankful to Indian Meteorological Dept. for images during T0+1/2 to T0+1 = for the system ≥ 20 providing the DWR data. Approval of the financial support , dBZ The output parameter is rain intensity as observed to the first author, from the Office of Navel Research-Global from the radar imageries ( for reflectivity ≥ 40 dBZ) (ONRG) to participate in ECSS-2009, is thankfully acknowledged The image analysis and training of the ANN is carried out on MATLAB platform. The results V. REFERENCES obtained from the soft computing approach are also Vieux B. E. , Bedient, P.B. 1998, Estimation of rainfall for compared with the results obtained from linear flood prediction from WSR-88 D reflectivity: A case multiple regression methods. study 17-18 October 1994, Weather and Forecasting, 13 407-415, . III . RESULTS AND CONCLUSIONS Dixon M., Winner G., 1993, TITAN, “Thunderstorm The results from the training of ANN in Identification, Tracking, Analysis and Nowcasting - A terms of matching and mismatching of the ANN Radar-based Methodology,” J. Atmos. Oceanic output with the observed values are presented in Table Technol., 6 785-797. -1. Rinehart R.E., Garvey, E.T., 1978, Three dimensional storm detection by conventional weather radar, Nature, 273 Metho- Training/ Case Match Mismatch 287-289. dology Validation Kuligowski R. J., Barros A.P., 1998, “Localized ANN Training Yes Rain 89% 11% precipitation forecast from a numerical weather Validation Yes Rain 77% 23 % prediction model using artificial neural networks,” Linear Training Yes Rain 62% 38% Wea. and Forecasting, 13 1194-1204. Regres Validation Yes Rain 55% 45% Rivolta, G., Marzano F.S., Coppola, E., Verdecchia, M., sion 2006, Artificial neural network technique for precipitation nowcasting from satellite imagery, Table1: Statistics of the Training and Validation Adv. Geosci., 7 97-103. experiments Denceux T., Rizand P., 1995, Analysis of radar images for rainfall forecasting using neural networks, Neural The statistics of matching/mismatching is Comput. & Applic, 3 50-61. found out for YES-RAIN situations . Overall it is Gonzalez C. R., Woods R. E., Eddins, S. L., 2007, Digital observed that the results are better for training data set Image Processing Using MATLAB”, Pearson compared to validation data set (Table -1). Further the Education Inc., 348-391.

2 144 5th European Conference on Severe Storms 12 16 October 2009 Landshut GERMANY

Forecasting and Nowcasting of Severe Storms and their preferred tracks across West Africa A.O.B Udogwu 1, J.B Omotosho 2, S.O Gbuyiro 2, I. Ebenebe 1, G.C Osague 1, E. Olaniyan1

1Nigerian Meteorological Agency,Plot 507 Pope John Paul Street, Maitama, Abuja,Nigeria, [email protected] 2The Federal University of Technology, Akure,P.M.B 704, Akure,Nigeria, [email protected] (Dated: 11 September 2009)

I. INTRODUCTION This paper attempts to use these two storms in 2009 as case There are two distinct seasons in West Africa – the rainy studies: season in northern summer and the dry season in northern
To highlight the role of the progressive northward winter. The two seasons are clearly defined by the advance of the ITD (ITCZ) with the onset of the movement of the Inter Tropical Discontinuity (henceforth rainy season in West Africa by comparing two ITD) which usually oscillates from the West African coast events. northwards with the northern summer to about 22ْ N and
To use surface thermodynamic properties to infer back again with northern winter. As the ITD moves important midtropospheric conditions necessary northwards, moisturebearing winds from southern Atlantic for storm development and intensity in the face of penetrate West Africa. The most destructive storms in dearth of upper air sounding. Nigeria occur just before the onset of the rainy season
To see whether analysis of surface conditions can (FebruaryMay) and just before the cessation of the rains be used to infer storm severity. (SeptemberOctober). Omotosho (1990) noted that the rains
To see to what extent relief features contribute to cannot start until the deep convective systems violent storms usually experienced during onset of (Thunderstorms and Squall lines) that deliver most of the rains. precipitation has started to develop and organize. Therefore
To ascertain how surface conditions contribute in squall lines and thunderstorms are the major contributors to determining the tracks of storms. West African and in particular Nigerian rainfall. A number
To compare morning and afternoon conditions to of very high impact weather have been observed in the sub see how much they contribute in the overall storm region. A severe storm occurred on the 22nd of September formation scenario. 1952 producing very high rainfall amounts for the three rainfall stations in Lagos Metropolis at the time (Lagos II. DATA AND COMPUTATIONS Roof, 159mm; Oshodi, 159mm and Ikeja, 134mm). This is Meteosat 9, February 24, 2009 at 1800UTC (fig. 1) over 94% of the monthly normal of 160mm in just one day. Another severe storm developed over the south west coast of Nigeria, around Lagos in the early hours of September 20, 2000; just becoming visible on the Meteosat 7 satellite imagery at 0530UTC. The progressive development of the storm was accompanied by very intense rainfall not observed in Lagos since the 1952 rains, producing very high rainfall amounts in the four rainfall stations in the Lagos Metropolis within just 6 hours of its occurrence (Lagos Roof, 280mm; Lagos Marine, 177mm; Ikeja 140mm; and Oshodi, 81mm). This is an average of about 106% of the monthly normal within just 6 hours. October 18, 2003 was the last day of the Commonwealth Games being hosted by Nigeria in Abuja. A powerful storm of 80kts hit the city Meteosat 9, May 2, 2009 at 1800UTC (fig.2) between 1700 and 1745hours local time ripping apart the Velodrome and other facilities being used for the games with unconfirmed loss of three lives. A similar storm had occurred at Abuja airport on March 27, 2002 with maximum gust over 80kts and causing severe damage to parked aircrafts particularly light aircrafts. An unusual dust storm gusting 60kts brought total darkness (zero visibility) for nearly 2hrs to Maiduguri and its neighbouring cities of north east Nigeria at 1600UTC (1700LT) on 30th May 1988. This phenomenon is a regular occurrence to that part of the country (being in the arid region) during the onset of the rainy season in May/June (Omotosho, 1995) most of which are not accompanied by rainfall. The list is endless and the story is similar in many West African cities and communities at these periods of the year.

145 5th European Conference on Severe Storms 12 16 October 2009 Landshut GERMANY

Equivalent Potential Temperature (Өe) last 24 hours. Its analysis is referred to as isallobaric This is the temperature a parcel of air would reach if it is Analysis of Dew Point Temperature at 0900Z on May 2, 2009 lifted dry adiabatically until all the water vapour in the 14.00 parcel condenses out releasing its latent heat and the parcel Sok Kat Kan is brought down to a pressure of 1000hpa. It is usually 12.00

Kad approximated to the static energy of the atmosphere. Bau Jos 10.00 Min Abj Yol

Ilo

Өe = Ө exp ( Lv q / Cp Tv ) where Tv = virtual 8.00 Mak Aku Deg. Lat N Lat Deg. Ikj Temperature, q = Specific humidity Ben Enu 6.00 Owe 6 1 Phc Cal Lv = the Latent heat of vapourization ( 2.253 x 10 J kg ) 4.00

1 1 Cp = 1004 J kg deg. 2.00

2.00 4.00 6.00 8.00 10.00 12.00 14.00 analysis Deg. Long E T = T ( 1 + 0.608 q ) v

k III. RESULTS AND CONCLUSIONS Ө = T ( 1000 / p ) where k = R/ Cp , R = Rd = 287 J kg 1deg.1 , k = 0.286 , Two storm events of February and May 2009 were studied in this work to highlight the role of the ITD (ITCZ) during T = Temperature, p = pressure the onset of the rains and the predisposing thermodynamic conditions at the surface. The analysis of the winds at (10m AMSL, 850 and 700hpa) q = qs RH / 100 where RH = relative humidity in percent, on the affected days was carried out in order to trace moisture transport into the subregion. Surface analysis of some thermodynamic variables (thetae and dew point qs = Saturation Specific Humidity temperature) were carried out in addition to isallobaric analysis for both morning (0900UTC) and afternoon qs = Є1 es / ( p – Є2 es ) where Є1 = 0.622 and Є2 = 0.378 Analysis of Equivalent Potential Temperature at 0900 GMT on May 2, 2009 (1500UTC) conditions to ascertain which periods present more telltale signals for the formation of these storms and 14.00 Sok Kat perhaps their severity as well.

Kan It was found that areas of high thetae values (above 1, 12.00 340K) that coincide with areas of high dew point (≥23.2ْC Kad Bau Jos Min for the south and ≥21.1ْC for the north) were highly probable 10.00 Abj Yol

Ilo zones for afternoon storm formation especially near high

8.00 Mak Aku elevation areas. It is important to note that morning Deg. Lat N Lat Deg. Ikj Ben Enu conditions of these variables tend to be more significant for 6.00 Owe the afternoon storm conditions hence it is therefore possible Cal Phc to determine where afternoon storms would form well ahead 4.00 of time knowing the morning situation. Results also showed that more stations have higher values of thetae during the 2.00 morning hours than the afternoon which may be due to a 2.00 4.00 6.00 8.00 10.00 12.00 14.00 Deg Long E convectively unstable atmosphere. While the storms formed Dew point Temperature (Td) near areas of tight pressure gradients in the north with the storms being driven by 700hpa winds, storms appear to This is the temperature to which air must be cooled at propagate along negative morning tendency in conjunction constant pressure until saturation is reached, at which point with midtropospheric flow in the south. The storms move liquid water condenses out. Dew point temperature is an along West Africa in an East –West direction as reflected by absolute measure of how much water vapour is in the air (i.e the cold clouds on the satellite images (Figures 1 and 2). how humid it is). Hence highest dew point temperatures correspond to highest humidity. Analysis of Dew Point Temperature at 0900Z on May 2, 2009 IV. AKNOWLEDGMENTS 14.00 Sok Kat The authors are indebted to Mr Odule, Mrs M.O Iso, Mrs

Kan Oyeshile and Mrs Afuye for providing the vital data for this 12.00 work. Kad Bau Jos 10.00 Min V. REFERENCES Abj Yol Ilo Omotosho J.B (1985): The Separate contributions of Squall

8.00 Mak Aku Lines, thunderstorms and the Monsoon Deg. Lat N Lat Deg. Ikj Ben Enu to the total Rainfall in Nigeria. J.Climatol.5: 476488 6.00 Owe Omotosho et al (2000): Predicting Monthly and Seasonal Cal Phc Rainfall, Onset and Cessation of the 4.00 Rainy Season in WesAfrica using only Surface Data. Int. J. Climatol. 20: 865880. 2.00 Eumetsat: www.eumetsat.com 2.00 4.00 6.00 8.00 10.00 12.00 14.00 Deg. Long E Pressure Tendency(P24) Refers to the net change in station level pressure over the

146 5h European Conference on Severe Storms 12 - 16 October 2009 - Landshut - GERMANY

NOWCASTING AND ASSESSING THUNDERSTORM RISK ON LOMBARDY REGION (ITALY) Bonelli P.1, Marcacci P. 1, Bertolotti E. 2, Collino E. 1, Stella G. 1

1 ERSE S.p.A., ENEA Ricerca sul Sistema Elettrico (former CESI RICERCA), V Rubattino,54 Milan, ITALY, [email protected], [email protected], [email protected], [email protected]

2 Lombardy Region - Civil Protection Office Via Rosellini, 17 - 20124 Milan, ITALY, [email protected] (Dated: 15 September 2009)

I. INTRODUCTION described in chapter II. Severe thunderstorms in Lombardy (Italy) are a serious risk during warm seasons. The region between the II. PRESENTATION OF RESEARCH Po valley and the Alps experiments one of the highest It is well known that severe weather caused by a TS, thunderstorms frequency in Italy. Most events are caused by as gust, hail, tornado, depends on the intensity of convection high intensity convective cells that born and dissolve in few inside the cell and on the content of liquid or solid water hours and that produce their effects in narrow area of about inside it. For this reason two variables, detected by radar and 10 km. The associated deep convection causes severe Meteosat, could be well correlated with damages caused weather as: big hailstone, gust, tornado and heavy rain. under the cell: the vertical profile of reflectivity inside the Due to the high urbanization of the Lombardy, cumulonimbus cloud and its top temperature. almost every severe thunderstorm causes strong wind that In STAF, radar reflectivity vertical profile is damages house roofs, crashes scaffolds and old trees; often analysed on every pixel in order to calculate the number of road signs flight away breaking down electric overhead lines vertical levels with reflectivity larger than 44 dBz, among (Bonelli et al., 2003). Furthermore short but heavy rain the 12 ones. This is very important to evaluate the presence causes flooding of underpasses, flash flood or landslides. of deep convection. In 2003, about 90 thunderstorms (TS) affected In order to formulate the cause/effect relation a people’s health, houses, cars, agriculture and electrical lines. statistical regression between the probability of damage and Some tornadoes have been detected too. The 2003 summer radar/satellite variables, has been developed. For this goal a was not particularly anomalous compared with others, storm archive ST-AR has been built up. It contains both concerning TS activity. In the 2007 summer a couple of radar and satellite data together with damage reports, noted tornadoes destroyed some houses in Guidizzolo, near by local press. The database contains data for more than 300 Mantova. In these situations insurance companies and TS cells, severe or not. (Collino et al., 2009). sometimes also local administration often appeal to the The logistic regression has been used. The equation "exceptional phenomena" in order not to compensate injured obtained has this formulation: people. They also hope the Central Government will declare ++= cXbXaY "the state of natural disaster" in order to obtain public 21 where X are the radar and satellite predictors, a,b,c are support. In this context an objective analysis would be i constant coefficients and Y is the logit of p: requested to evaluate if the phenomena was really extraordinary. ≈ p ’ pY == ln)(logit ∆ ÷ In Lombardy the spare ground stations do not allow «1− p ◊ to evaluate severe weather associated to thunderstorms. where p is the probability of the severe event. Storms are however well monitored by some C-band radar, So, it is possible to calculate the probability like this: located at the border of the region, and by the Meteosat 1 satellite. In particular, Mount Lema radar, operated by p = cXbXa Switzerland Meteorological Service, provides useful data on [ exp1 ( ( ++−+ 21 ))] quite the whole Lombardy region (Hering et al., 2007), (Joss et al., 1997). In our case X1 and X2 are the normalized variables: Since 2006, ERSE (former CESI RICERCA), in the frame of its activity on interaction between severe weather N Ttop −− )75( 44 X = and electrical grid, collaborates with Lombardy Authority of X 1 = 2 Civil Protection Office, developing and testing a severe 12 −− ))75(3.20( thunderstorms detection and nowcasting system (Bonelli et al., 2008). In fact, despite the difficulty to forecast TS N44 is the number of radar vertical layers with intensity and position less than one hour in advance (Wilson reflectivity higher than 44 dBz and Ttop the brightness et al, 1998), a well designed alert system can be useful in temperature of the cloud from Meteosat in °C. preventing damages to people and to reduce the time of a, b, and c are the coefficients of the regression, operations for the rescue teams. computed by means of the ST-AR data-set and with Recently ERSE developed STAF (Storm Track Alert SYSTAT software: and Forecast), a nowcasting system based on Radar and a = -2.507; b = 7.388; c = -9.502 2 MSG (Meteosat Second Generation) data that selects only with R = 0.66 severe TS, tracks them and sends alert messages to users, as The SYSTAT can also compute the improvement of

147 5h European Conference on Severe Storms 12 - 16 October 2009 - Landshut - GERMANY the regression due to the introduction of a new independent variable. In this case the algorithm shows that Ttop temperature adds a valid contribute to the regression compared to the one obtained using only radar data, despite of an evident correlation between the two variables. The system STAF analyses every 5 minutes the full 3D radar reflectivity matrix, with a horizontal resolution of 2 km and 12 layers of 1 km in the vertical. STAF software processes radar data in order to: − identify only the most intense cells, using a threshold of 36 dBz on the vertical maximum reflectivity (Hering et al., 2004) and compute their position; − compute cell velocity by means of its two previous positions; FIG 1: Example of the WEB-GIF interface developed for STAF. − attribute to every selected cell a damage probability based on radar reflectivity profile and top temperature IV. AKNOWLEDGMENTS from Meteosat data, as explained above; This work has been financed by the Regione Lombardia − forecast cell position for 30 minutes ahead by means of Civil Protection and by the Research Fund for the Italian its velocity; Electrical System under the Contract Agreement between − generate alert SMS messages, when a severe cell crosses ERSE and the Ministry of Economic Development. a pre-defined alert area, and send them to selected users; − display on-line and past situations on a WEB-GIS site. V. REFERENCES Details about some features of the tracking algorithm Bertolotti, E., Minardi, G.P., 2007: Badweather: protection can be found in Bonelli et al., 2008. and information in Lombardy Region. National During the 2008 and 2009 summers about 15 alert Conference of Climate Change - Poster session. September areas of 10 km in radius over the Lombardy region were 2007, Rome (I). defined. In these areas the system STAF has been tested. Bonelli P., P. Marcacci, 2008: Thunderstorm nowcasting by Selected volunteers and Civil Protection personnel received means of lightning and radar data: algorithms and the SMS alert from STAF and, after each received alert, applications in Northern Italy. Natural Hazards and Earth they registered local observations about TS effects and the System Sciences, 8, 1187-1198. effectiveness of the message. Bonelli P., M. Del Frari, G. Mariani, 2003: Impact of severe An important function of STAF is the possibility for weather on electric energy delivering: actual regulation in the users to write down and upload a WEB form with own the Italian electricity market and methods for treating reports about the TS event. The information registered is: meteorological data. European Conference on Applied presence of strong gust, big hailstone, damages to houses, Meteorology (ECAM) 2003 – Roma (Italy), 15-19 tree falling, electric blackout and the reception time of SMS. September 2003. A new important feature of 2009 STAF version, Collino, E., P. Bonelli, L. Gilli, 2009: ST-AR (STorm - with respect to the 2008 one, is the WEB-GIS interface that ARchive): a project developed to assess the ground effects allows the user to have a look on the TS cells with their of severe convective storms in the Po Valley”, Atmos. velocity, intensity and probability of damage, over a Res., 93, 483-489. standard geographic map with roads and cities, as shown in Hering, A.M., P. Ambrosetti, U. Germann, and S. Sénési, FIG 1. Optionally one can also display the overhead electric 2007: Operational nowcasting of thunderstorms in the grid. Alpine area, 4th European Conference on Severe Storms 10-14 September 2007 - Trieste - Italy. III. RESULTS AND CONCLUSION Hering, A. M., Morel, C., Galli, G., Sénési, S., Ambrosetti, The experimental activity concerning forecast and P., and Boscacci M., 2004: Nowcasting thunderstorms in observations of severe thunderstorms in Lombardy, during the Alpine region using a radar based adaptive the 2008 and 2009 summers allowed the collection of many thresholding scheme, Proceedings of Third European data. These data will be used to better set up the STAF Conference of Radar in Meteorology and Hydrology software. (ERAD), September 2004, available on An important goal of this effort is also to sensitize http://www.copernicus.org/ local Authority for Civil Protection. In fact these nowcasting Mariani, L., Bertolotti, E., Iorio, R: 1997 Lightning flashes practices will appear useful only if users are correctly monitoring and meteorology: the experience of Lombardy trained on the thunderstorm risk and about the available Agrometeorological Service with CESI SIRF System. monitoring instruments. Lightning and Mountains ’97. Chamonix, (F). The challenge is due to the very short forecast time, Joss, J., Schadler, B., Galli, G., Cavalli, R., Boscacci, M., in contrast with the common day to day weather forecast, Held, E., Della Bruna, G., Kappenberger, G., Nespor ,V., Operational Use of Radar for that implicates a different management of the prevention and and Spiess, R. 1997: rescue activities. Precipitation Measurements in Switzerland, NFP31: An increase in the number of users and alert areas is Projekt NOWRAD, 1997, Meteo Swiss, Locarno. expected in the future. Also new alert systems, as local radio Wilson, J. W., Crook, N. A., Mueller, C. K., Sun, J., and broadcast, will be tested. The user feedback will be always Dixon, M.: Nowcasting Thunderstorms: A Status Report, Bull. of American Meteorological Soc. very important to test the real effectiveness of the alert , Vol. 79, No. 10, system. October 1998.

148 5h European Conference on Severe Storms 12 - 16 October 2009 - Landshut - GERMANY

TRIGGERING OF DEEP CONVECTION

BY LOW-LEVEL BOUNDARIES

Kevin Julian Rae1

1South African Weather Service, Bolepi House, Rigelpark, Rigel Ave, Erasmuskloof, Pretoria, South Africa, [email protected]

(Dated: 15 September 2009)

I. INTRODUCTION supporting literature study is complete, with attention now shifting to the collection of suitable field case studies during Southern Africa is a region that frequently the southern hemisphere summer of 2009/2010. experiences thunderstorms, particularly in the summer This study aims to synchronise and integrate months over the central and eastern parts of the region. established literature (regarding low-level boundary Many of these storms can be severe (as per NSSL interactions with thunderstorms) with local (southern definition), posing a distinct threat to life and property (Pyle, African) operational nowcasting practise. 2006). Rural communities are often at risk from flash- Standard Meteosat 9 channels as well as flooding, lightning, hail and wind damage (be it tornadic or EUMETSAT RGB combinations (such as ‘FOG A’, ‘Day non-tornadic in nature). Microphysical’ and ‘Dust’) are to be used as diagnostic tools Accurate, appropriate and timeous nowcasts (lead- to identify incipient thunderstorm outflows (with respect to time on the scale of a few minutes to no more than a few both diurnal and nocturnal cases), whilst Doppler S and X- hours) are desirable for any weather agency. The pursuit of band RADAR data will be utilised to monitor and quantify any methodology or knowledge which could be applied in subsequent convective development deemed to have been practice, to mitigate the destructive and adverse effects of triggered or influenced by such outflows. As an ancilliary such storms, should be an ongoing, high-priority activity of remote-sensing assessment tool (in terms of assessing a modern weather service. possible storm severity in a southern African domain (Rae & In South Africa in recent years, the role of remotely- Clark, 2005)), the EUMETSAT ‘Deep convection/ sensed meteorological data (particularly from satellite and overshooting tops’ RGB will be used to supplement radar sources) in the diagnosis and monitoring of severe RADAR data. thunderstorms has been significantly elevated. This is A multi-sensor approach, utilising a variety of especially true in the case of Meteosat 9 (previously MSG) meteorological sensing equipment will be employed. data and in particular, RGB composite images. In a Multiple overlays of satellite, RADAR as well as Vaisala (meteorologically) data-sparse domain such as South Africa, LDN lightning data will be viewed and manipulated with the the judicious and effective use and interpretation of radar use of SUMO visualisation software as well as via a NinJo and satellite data is critical if forecasters are to perform meteorological workstation. Direct measurement of surface robust and sustainable storm monitoring activities meteorological variables (by surface AWS and Vaisala countrywide. AWOS) at a high temporal rate (5min or less) of particularly Low-level boundaries are a well-documented and air temperature and pressure are critical in order to common triggering mechanism for deep convection in discriminate between a gravity wave disturbance and that of general, Weckwerth & Wakimoto (1992), Wilson & Roberts a density current flow (Kingsmill, 2003), Haertel, et al., (2006), Wilson & Schreiber (1986), Wilson, et al.(1992), (2000). Schreiber-Abshire (2003). The landmark southern Vertical measurement of the lower layers (surface hemisphere case study by Sills, et al. (2004), regarding to about 3km AGL) of the troposphere will be performed tornadic storms during the Sydney 2000 Olympic Games, using a combination of Vaisala RS92-SGP radiosonde confirms the role that boundaries (including thunderstorm balloon soundings as well as in situ aircraft measurements in outflow interaction with sea-breeze fronts) played in the form of GPS-referenced AMDAR reports from triggering severe storms. Rasmussen, et al. (2000) commercial airliners. In particular, aircraft takeoff and highlights and investigates relationships between tornadic approach profiles for Oliver Tambo International Airport storms and boundaries. Dial & Racy (2004) discuss (ORTIA), near Johannesburg, occur at short intervals with forecasting aspects, in the context of the challenge only 5 to 10 minute separation (per flight) and are a valuable forecasters face when trying to anticipate possibly severe nowcast and research-quality data source. The accuracy of interactions between storms and boundaries. AMDAR reports has been found to be comparable to the high performance and accuracy of traditional radiosonde instruments (Drue et al., 2008). Radiosonde and AMDAR II. PRESENTATION OF RESEARCH data, being GPS-referenced, as well as containing altitude data, are well-suited to 3-D spatial representation (eg. The author’s study represents a body of work in Google Earth) and it is hoped that new and innovative ways progress and is being approached in phases, as partial of presenting data spatially may well lead to additional requirements for MSc study at Pretoria University. The insights being gained during the course of this study.

149 5h European Conference on Severe Storms 12 - 16 October 2009 - Landshut - GERMANY

III. RESULTS AND CONCLUSIONS Kingsmill D.E., 1995: Convection Initiation Associated with a Sea-Breeze Front, a Gust Front, and Their Collision. During 2008/2009, preliminary assessment of the Mon. Wea. Rev., 123 2913–2933. viability of Meteosat 9 (MSG) data to identify and track low-level boundaries has been quite successful. It has been Kingsmill D.E. & Crook N.A., 2003: An Observational very encouraging to establish that the EUMETSAT “FOG Study of Atmospheric Bore Formation from Colliding A” RGB (which incorporates IR 3.9
m information) lucidly Density Currents. Mon. Wea. Rev., 131 2985-3002. picks up fine-scale structure, clearly demarcating the leading edge of outflow boundaries. Similarly, for diurnal Pyle D.M., 2006: Severe convective storm risk in the applications, the EUMETSAT “Day Microphysical” RGB is Eastern Cape province of South Africa. Unpublished PhD equally useful, as is the “Dust” RGB. In last-mentioned case, thesis. Rhodes University, South Africa, 1-2. a useful (if indirect) means of establishing the presence of an outflow boundary (given arid, dusty environments, typical of Rae K.J. & Clark L-A., 2005: Meteosat Second Generation: central and western South Africa) is to identify billowing Severe thunderstorm nowcasting applications. A case study dust plumes rising up on the leading edge of the boundary of KwaZulu-Natal, 3rd January 2005, conference (Kerkmann et al, 2004). Naturally, the high-resolution VIS presentation, SASAS Annual Conference, Richard’s Bay, channel is also invaluable for diurnal monitoring of the Republic of South Africa. evolution of outflow and potential subsequent triggering of deep convection. Rasmussen E. N., Richardson S., Straka J.M., Markowski Further cautionary insight has been the realisation P.M. & Blanchard D.O., 2000: The association of that gravity wave disturbances often manifest themselves in significant tornadoes with a baroclinic boundary on 2 June superficially similar ways to that of density currents. The 1995. Mon. Wea. Rev., 128 174-191. value and role of traditional surface meteorological data (especially pressure and temperature) at a high temporal Schreiber-Abshire W., 2003: Convection initiation by resolution is thus confirmed as a critical component of the boundaries. COMAP 2003 course presentation, COMET, author’s study design. The selection and screening of University Corporation for Atmospheric Research (UCAR). potential case studies will almost certainly require confirmatory surface AWS and/or AWOS data. Kingsmill Sills D.M., Wilson J.W., Joe P., Burgess D.W., Webb (2003, 2985-2986) gives a clear description of the R.M. & Fox N. I., 2004: The 3 November Tornadic Event meteorologically measurable discriminatory (surface) during Sydney 2000: Storm Evolution and the Role of Low- features between the two phenomena, as does Haertel et al., Level Boundaries. Wea. Forecasting, 19 22-42 (2001). Weckwerth T.M. & Wakimoto R.M., 1992: The Initiation IV. ACKNOWLEDGMENTS and Organization of Convective Cells atop a Cold-Air Outflow Boundary. Mon. Wea. Rev., 120 2169–2187.

The author would like to extend his appreciation to Wilson J. W. & Roberts R. D., 2006: Summary of his MSc supervisor, Ms. Liesl Dyson (Pretoria University) convective storm initiation and evolution during IHOP: for her ongoing guidance and support. Furthermore, he Observational and modeling perspective. Mon. Wea. Rev., would also like to acknowledge SAWS librarians, Ms. Karin 134 23–47. Marais and Ms. Anastasia Demertzis for their tireless Wilson J. W. & Schreiber W.E., 1986: Initiation of dedication and assistance. convective storms at radar-observed boundary-layer convergence lines. Mon. Wea. Rev., 114 2516–2536. V. REFERENCES Wilson J. W., Foote G.B., Crook N.A., Fankhauser J.C., Dial G.L. & Racy J.P., 2004: Forecasting Short Term Wade C.G., Tuttle J.D., Mueller C.K. & Krueger S.K., Convective Mode And Evolution For Severe Storms nd 1992: The role of boundary-layer convergence zones and Initiated Along Synoptic Boundaries. Preprints, 22 Conf. horizontal rolls in the initiation of thunderstorms: A case on Severe Local Storms, Hyannis, MA. study. Mon. Wea. Rev., 120 1785–1815.

Drue C., Frey W., Hoff A. & Hauf T., 2008: Aircraft type- specific errors in AMDAR weather reports from commercial aircraft. Q. J. R. Meteorol. Soc., 134 229-239.

Haertel P.T, Johnson R.H. & Tulich S.N., 2001: Some Simple Simulations of Thunderstorm Outflows. J. Atmos. Sci., 58 504-516.

Kerkmann J., Lutz H. J., Pylkkö P., Roesli H. P., Rosenfeld D., 2004: Applications of Meteosat Second Generation (MSG), ESAC course presentation, Pretoria, Republic of South Africa.

150 5th European Conference on Severe Storms 12 - 16 October 2009 - Landshut - GERMANY

SHORT-TERM FORECAST OF HAIL PRECIPITATION PARAMETERS

J. L. Sanchez1, L. López1, B. Gil-Robles1, J. Dessens2, C. Bustos3 and C. Berthet2

1 Group for Atmospheric Physics, IMA, University of León, [email protected] 2 ANELFA, Toulouse, France. 3 Gobierno de Mendoza, Argentina.

(Dated: 15 September 2009)

I. INTRODUCTION world most affected by severe convective phenomena. In the zone, the provincial government has a network of 130 An important number of studies have developed hailpads covering a representative area of 5x5 km each one. short-term statistical models that make use of data on pre- The characterization of the meteorological conditions was convective conditions to determine the risk of storms with also carried out by means of radiosondes launched from the hail precipitation in a particular study zone. The experience city of Cruz Negra. The radiosondes were launched at 15 of the Group for Atmospheric Physics at the University of UTC. A total of 44 days were analysed. León, Spain, has shown that it is more adequate to begin the process by selecting those variables that are the best Finally, the third study zone is located in France, forecasters and then combine them using discriminatory or where in the period between 1999 and 2003, insurance logistical functions (López et al., 2007; Sánchez et al., 2008, companies paid an average of 170 million € per year in 2009). compensation for crops affected by hail. One third of the damage to crops due to hail is concentrated in the south-west Based on previous experience, the aim of this (region of Aquitaine et Midi-Pyrénées), another third in the study is the selection of the most adequate meteorological south-east, and the rest, all over the other regions of the variables to be included in statistical short-term forecast country. In these regions, the ANELFA (Association models of the characteristics of hail, namely the values of Nationale d’Etude et de Lutte contre les Fléaux the following variables per m-2: number of hailstones, Atmosphériques) has installed a dense network with over kinetic energy, ice mass and maximum diameter have been 1,000 hailpads. Sounding data were available for the study determined in relation to the pre-convective conditions. from 46 daily soundings launched at 12 UTC in the city of Bordeaux. In these days, hail precipitation was registered in The first step was to analyse the relations between the French hailpad network between the years 1997 and the values of the 4 variables that characterize the maximum 2005. hail precipitation expected and the meteorological variables obtained from the closest sounding stations. III. DATA BASES AND METHODOLOGY

II. STUDY ZONE For each of the zones, integrated databases were created including, on the one hand, the characteristic Three study zones were selected, characterised by parameters of the hailstorms taken from the hailpads having a high frequency of hailstorms, the availability of (number of hailstones, kinetic energy, ice mass and networks of hailpads in these areas, and the existence of data maximum diameter) and, on the other, a series of from radiosondes close to these areas. meteorological variables and indices (36 in total) based on the data from the radiosondes. It should be noted that The first study zone is located in the Ebro Valley previous studies existed for the three areas, as different in the northeast of the Iberian Peninsula, which covers an discriminating models had already been designed for area of approximately 50,000 km2. The Group for predicting hailstorms in them (López et al., 2007; Sánchez et Atmospheric Physics at the University of León has been al., 2009). In terms of selecting the indices, this experience carrying out summer campaigns in this area gathering data was of great help. As a result, the indices used were the for detailed studies on the characteristics of hailstorms. The same used in previous studies. detection and calculation of the parameters related to hail precipitation is carried out using an extensive network of For each of the days, individual data were hailpads covering an area of 2,700 km2. In this zone, the available (per m2) from hailpads affected by hail: number of characterization of the atmospheric conditions was carried hailstones (N), kinetic energy (E), ice mass (M) and out using data from a sounding balloon launched at 00 UTC maximum diameter (D). However, and with the aim of or 12 UTC in the city of Zaragoza, in the middle of the study correlating these data with the daily data from the zone. To set up the study, 38 days were used with complete radiosondes, different methods for the daily integration of databases. These days corresponded to the months of June to these data were calculated. As a result, we obtained the September from 2003 to 2007. maximum daily values for each of these variables, the accumulated daily values (except for the size), and also the The second study area comprises the north and values recorded in the hailpad that presented the maximum centre of the Argentinean province of Mendoza, at the foot kinetic energy. The aim was to discover which of these of the Andes. This study zone is one of the areas in the parameters (N, E, M or D) and through which daily

151 5th European Conference on Severe Storms 12 - 16 October 2009 - Landshut - GERMANY integration (maximum values, accumulated, or the hailpad correlations, although the total accumulated energy does, with maximum energy) had the best correlations with the which is correlated with the wind at 500 hPa. This data indices that characterised the pre-convective conditions. shows that attempting to construct a prediction model for the This way, the results revealed which of these parameters are maximum energy in the zone would not provide us with the most suitable, a priori, for developing prediction models satisfactory results, and we would instead have to turn to the for the characteristics of hail precipitation on the ground. daily accumulated energy, or preferably for the prediction of the maximum diameter. Subsequently, with the same objective in mind, a Principal Components Analysis was carried out, together In France, the maximum diameter has significant with a Varimax rotation with Kaiser normalisation of all of correlations with up to 7 different parameters, as shown in the variables for each of the zones. The relations obtained Table 1. It is especially interesting to note that in the zone between the different variables were analysed in detail, and neither the VGP index nor the EHI have significant interpreted in meteorological terms. correlations. Also, on analysing the energy data (both maximum and total), we now see the greater importance of IV. RESULTS AND DISCUSSION parameters related to the vertical atmospheric stratification, such as the tropopause height, the isozero height or the We should not forget that the three zones do not altitude of the CCL. have exactly the same characteristics in relation to hail precipitation, especially in terms of the maximum values and In terms of the mass and number of impacts, a frequency histograms. For example, the parameters obtained very low number of correlations were found, especially in in Argentina, particularly with regard to energy and France, meaning that their prediction is not easy in any of maximum diameter, are higher than in the others. The pre- the areas. However, these parameters are not so closely convective conditions that exist in situations with hailstorms connected with the damage produced by hail as the energy are also different (Sánchez et al., 2009). For this reason we or maximum diameter, meaning that their prediction is of decided to analyse the results individually for each of the secondary importance. areas. Also, the Principal Components Analysis (PCA) shows once again that Zaragoza and Argentina seem to show Zaragoza Mendoza Bordeaux a similar type of behaviour. The maximum diameter in the (Spain) (Argentina) (France) two zones is connected in the first components with the EHI EHI 850 hPa dew point indices VGP and EHI, which have proved to be essential in VGP VGP Lifted Index predicting the maximum hail size. However, in France, the 850 hPa dew point Wind 500 hPa WBZ maximum diameter is in the fourth principal component, Showalter Index 0 ºC altitude connected with parameters that essentially measure the vertical atmospheric stratification (CCL height, LFC height). Tropopause height As a result, we may see that the results require a specific Convective temperature meteorological interpretation for each of the areas (Zaragoza CAP and Mendoza on the one hand, and France on the other), in TABLE I. Summary of the significant correlations found for the terms of the mechanisms behind the formation of hailstorms. maximum diameter. Finally, we believe that the results obtained will Firstly, Pearson’s correlation coefficients were make it possible to develop discriminatory prediction calculated, based on the daily data from the hailpads and the models for the hail characteristics in each of the zones. As a daily data from the radiosondes. Table I shows a summary result, in Zaragoza and France, the aim will be to develop of the significant correlations (at a significance level of 0.01) prediction models for both the diameter and the energy found for the maximum diameter for each of the zones. In (maximum or accumulated), while in Argentina the studies the case of Zaragoza, the results reveal that the maximum will focus exclusively on the prediction of maximum diameter recorded by a hailpad is correlated with the dew diameters. point at 850 hPa, the Showalter index, the VGP and the EHI. This is coherent with the results found by Sánchez et V. ACKNOWLEDGMENTS al. (2008), who constructed a logistical equation in order to integrate the prediction models for storms based on This study was supported by the Spanish Ministry of Education and radiosonde variables for four different areas. They found Science through grant CGL2006-13372-C02-01/CLI, the Regional that the Showalter index and the dew point at 850 hPa are Government of Aragón, and the Provincial Government of the variables that best characterise pre-convective situations, Mendoza. regardless of the study area. These two indices have now also been found to be important in order to predict the VI. REFERENCES maximum diameter. The correlations found for the López L., García-Ortega E., and J.L. Sánchez, 2007: A maximum energy in the zone show that this is significantly short-term forecast model for hail. Atmos. Res., 83 176- correlated with the dew point at 850hPa and the VGP. 184. Sánchez J.L., López L., Bustos C., Marcos J.L. and García- In Argentina, the results once again reveal that the Ortega E., 2008: Short-term forecast of thunderstorms in maximum diameter has the highest number of significant Argentina. Atmos. Res., 88 36-45. relations. Once again, the EHI and VGP parameters are Sánchez J.L., Marcos J.L., Dessens J., López L., Bustos C significant, together with the wind at 500hPa. The maximum and García-Ortega E., 2009: Assessing sounding-derived energy in the zone does not have any significant parameters as storm predictors in diferent latitudes. Atmos. Res., 93 446-456.

152

Precipitation forecast by the COSMO NWP model using radar and satellite data Z. Sokol, P. Pešice

Institute of Atmospheric Physics AS CR, Boční II, 1401, Prague, Czech Republic, [email protected] (Dated: 15 September 2009)

I. INTRODUCTION Radar reflectivity is measured by two radars whose Forecasting local heavy convective precipitation is a positions are indicated in Fig. 1. Radar reflectivity is first difficult and complex task. At present, numerical weather transformed into rain rate R (mm/h) using standard Z-R prediction (NWP) models with a resolution of the order of 1 relationship and then the rain rate is expressed as water km are capable of generating fine-scale precipitation fields, vapour mixing ratio q (kg/kg) making use of an empirical the structure of which is similar to observed reflectivity by relationship meteorological radars. However, observed and predicted 2026.70 )R(q = R 9143.0 , precipitation cores differ in rainfall amounts, positions, and ρ temporal evolutions. The errors in the forecast can be where ρ is the air density (kg/m3). attributed to two main sources: the model's ability to Satellite data are considered complementary to radar correctly simulate dynamical and physical processes, and reflectivity. The assimilation procedure uses brightness initial and boundary conditions supplied to the model (e.g. temperatures from two channels: 10.8 (T10.8) and 6.2 (T6.2) Stensrud, 2007). μm measured by Meteosat 8. The difference T10.8 - T6.2 is It is known that improvement of the model initial applied to identify vertically developed convective conditions, especially humidity parameters, including cloud cloudiness and to estimate rain rate which is again water content, rain water, and ice, is important for more transformed into q (Sokol, 2009). Satellite data are accurate precipitation forecasts for the first several hours of assimilated into the model in a grid point only if radar does model integration (Ducrocq et al., 2002). Another frequently not observe rain, the model does not forecast rain and used approach is the assimilation of radar data, which not satellite data indicate significant precipitation. only improves the model initial conditions but also initiates Hourly precipitation obtained by merging radar a model state using detailed information on the development reflectivity measurements and gauge observations is the of convective storms before the forecast starts. Numerous third type of assimilated data. Beside the observed data also studies have shown that assimilating radar data of both nowcasting forecasts of hourly precipitation are assimilated types, Doppler wind velocity, and reflectivity (or derived in the same way as observed data. The nowcasting forecast radar-based rain rates), can improve the quality of is obtained by simple advection of radar echo using precipitation forecasts (e.g. Jones and Macpherson, 1997; COTREC technique (Novak, 2007). The motivation for Snyder and Zhang, 2003; Tong and Xue, 2005; Zhang et al., making use of nowcasting data is the experience that the 2004). In this paper we study the impact of assimilation of COTREC forecast is quite accurate in case of severe radar reflectivity and satellite data on the precipitation convection for the first hour. forecast for next 1-3 hours. Most studied cases include events when NWP model forecasts without assimilation are not able to develop precipitation in corresponding regions and times.

II. PRESENTATION OF RESEARCH In this study we use COSMO NWP model (version 4.6), which is integrated with the horizontal resolution of 2.8 km over the territory of the Czech Republic (Fig. 1). Cumulus parameterization is switched off, but parameterization of shallow convection is included. Prognostic fields of LME (Local model - Europe of German Weather Service) model are used as initial and boundary conditions. The assimilation method is based on corrections of model water vapour (WVC) (Sokol, 2009) and consists of adding/removing water vapour into/from the model water vapour mixing ratio which are performed using nudging technique. Oversaturation or undersaturation, which can follow the FIG. 1: The model domain with topography (in m above MSL) and corrections, result in releasing or absorbing heat, and with marked positions of two radars as well as areas covered by the radar data (dashed circles). consequent changes in the model temperature. In this way, the WVC method is similar to latent heat nudging (Jones and An example of forecasts with and without assimilation Macpherson, 1997). of hourly precipitation is shown in Fig. 2. The corrections of model water vapour are based on the observed radar and satellite data and three types of variables are assimilated: (i) observed radar reflectivity; (ii) satellite data and (iii) measured and forecasted hourly precipitation.

153

III. RESULTS AND CONCLUSIONS Our preliminary results show that the assimilation of radar and satellite data improves the precipitation forecasts in comparison with the model runs without assimilation. In all cases, fairly good forecasts were obtained for lead times two or three hours minimum. If the precipitation forecast is reasonably accurate for the model without the assimilation, then useful forecasts are obtained for longer lead times. The inclusion of the COTREC forecast into the assimilated data improved precipitation forecasts in all tested cases. This is due to the fact that in all tested cases we predicted development of well organized convection and the COTREC prediction was reasonable. However, further investigations are needed.

IV. AKNOWLEDGMENTS The work was supported by grants GACR 205/07/0905 and OC112 (COST731). The COSMO code, provided by the German Weather Service, is highly appreciated. The radar and gauge data for the use in the work was made available by the Czech Hydrometeorological Institute.

V. REFERENCES Ducrocq, V., Ricard, D., Lafore, J.P., Orain, F., 2002: Storm-scale numerical rainfall prediction for five precipitating events over France; on the importance of the initial humidity field. Weather Forecast. 17, 1236–1256. Jones, C.D., Macpherson, B., 1997: A latent heat nudging scheme for the assimilation of precipitation data into an operational mesoscale model. Meteorol. Appl. 4, 269–277. Novák P. 2007: The Czech Hydrometeorological Institute’s Severe Storm Nowcasting System. Atmos. Res. 83, 450– 457. Snyder, C., Zhang, F., 2003: Assimilation of simulated Doppler radar observations with an ensemble Kalman filter. Mon.Weather Rev. 131, 1663–1677. Sokol, Z., 2009: Effects of an assimilation of radar and satellite data on a very-short range forecast of heavy convective rainfalls. Atmospheric Research, 93, 188–206. Stensrud, D.J., 2007: Parameterization schemes. Keys to Understanding Numerical Weather Prediction Models. Cambridge University Press. 459 pp. Tong, M., Xue, M., 2005: Ensemble Kalman filter assimilation of Doppler radar data with a compressible nonhydrostatic Model: OSS Experiments. Mon. Weather Rev. 133, 1789–1807. Zhang, F., Snyder, C., Sun, J., 2004: Impacts of initial estimate and observation availability on convective-scale data assimilation with an ensemble Kalman filter. Mon. Weather Rev. 132, 1238–1253.

FIG. 2: Observed and forecasted precipitation (3 June 2008, 12-15 UTC) by COSMO without assimilation (b), with assimilation of observed hourly precipitation (c) and as (c) but using also the forecast by COTREC (d).

154 5th European Conference on Severe Storms 12 - 16 October 2009 - Landshut - GERMANY

Non Mesocyclone Tornadoes in Hungary

Zoltán Polyánszky1

1Hungarian Meteorological Service, P.O. Box 38, H-1024 Budapest, Hungary, [email protected]

(Dated: September 15, 2009)

I. INTRODUCTION

Most violent tornadoes develop usually from supercells (Fujita, 1981), but in Hungary there were also observed damaging tornadoes in the past years developed in environment not favorable (low vertical wind shear, small SRH) for supercells (Wakimoto et. all., 1989; Brady et. all., 1989; Caruso et. all., 2005; Davies et. all., 2006). A total of 41 visual vortices were studied in 2009, 2008 and 2006 using data obtained from reports of storm spotters and ESWD database. Those cases were not associated with supercell thunderstorms. 11 of these vortices can be classified as tornadoes (eight F0, two F1, one F2) and the others were funnels.

II. PRESENTATION OF RESEARCH

To investigate these cases, several analysis fields (MSL, moisture convergence, 0-2 km vertical temperature gradient, 0-3 km section of Sbcape, LCL, 0-3 km SRH) and vertical profiles FIG. 1: F 1 tornado in Ipolytarnóc (2009.08.24. 12 UTC): T . The typical were obtained from GFS and from hydrostatic run of WRF ARW case of tornado behind the cold front on the lee side of Carpathians 3.0 with horizontal resolution of 10 km and the assimilation of surface information (provided by an automatic measurement IV. AKNOWLEDGMENTS network including approximately 100 instruments operating in HMS) (FIG. 1). Special thanks to Gyula Bondor and Gábor Durbász.

III. RESULTS AND CONCLUSIONS V. References

All tornadoes developed directly for hours on stationary wind shift boundary, that generate strong convergence and pre- Fujita, T. T., 1981: Tornadoes and downburst in the context of existing vertical vorticity circulations. The boundary could be an generalized planetary scales. J. Atmos. Sci., 38, 1511-1534. occluded fronts developed ahead or behind of surface cold front, along the weak quasi-stationary front or in a flat pressure area. Wakimoto, R. M., and J. W. Wilson, 1989: Non-supercell The generally flat and dark parent clouds of these tornadoes were tornadoes. Mon. Wea. Rev., 117, 1113-1140. usually developing cumulunimbus. The intensive precipitation started usually for a short time after the observation of the intense Brady, R.H., and E.J. Szoke, 1989: A case study of vortices. So it may be assumed, that the smaller scale nonmesocyclone tornado development in northeast Colorado: thunderstorm outflow boundaries can enhance the pre-existing similarities to waterspout formation. Mon.Wea. Rev., 117, 843- vertical circulations along wind shift boundaries. By event the 856. low-level lapse rates within the lowest 2 km was greater than 7-7.5 C/km, the 0-3 km Sbcape was greater than 60-70 J/kg, the LCL Caruso, J. M., and J. M. Davies, 2005: Tornadoes in non- height was 600-1750 m, the 0-1 and 0-6 km shear was usually less mesocyclone environments with pre-exisitng vertical vorticity than 5-10 m/sec. These conditions of the environment are not so along convergence boundaries. NWA Electronic Journal of significant, so it may be assumed, that these types of tornadoes Operational Meteorology, June 2005. can give the majority of relatively weak tornadoes in our area. Davies, J. M., 2006: Tornadoes in environments with small helicity and/or high LCL heights. Wea. Forecasting, 21, 579–594.

155

156 5th European Conference on Severe Storms 12 - 16 October 2009 - Landshut - GERMANY

A statistical study of stability indexes as convective weather predictors in Lombardia

M. R. Salvati1, D. Berlusconi2 1Servizio Meteorologico Regionale, ARPA Lombardia, Via Restelli 3/1 Milano, Italy, [email protected] and 2Corso di Laurea in Fisica, Universita` di Milano, Via Celoria 16 Milano, Italy, [email protected] (Dated: September 15, 2009)

I. INTRODUCTION (1999-2008). The presence of at least one lightning strike in each 12 hour period between 00-12UTC and 12-00UTC was Forecasting thunderstorm occurrence is still a difficult task used to discriminate between thundery mornings/afternoons for both NWP models and weather forecasters due to the small (active case) and stable mornings/ afternoons. No other ob- spatial and temporal scales involved. The low predictability servation of convective activity has been used, as none of of convective phenomena still justifies the operational prac- those available to the SMR and used by the authors cited tise of using empirically developed stability indexes to eval- above could cover the same ten year observation period with uate the atmospheric potential for storm development. The sampling characteristics comparable to the lightning detection value of such indexes is in their capacity to summarise char- system (high spatial and temporal resolution, high precision acteristics of the convective environment in a single number of reports, ease of data processing). As expected convection which can be easily computed from operational sounding data is significant only in summer months (Fig. 1); hence only the and used for very short range forecasts of thunderstorm occur- period from May to September has been considered for the rence, generally associated to a threshold. analysis of stability indexes. To be used effectively indexes and reference thresholds Nine common indexes (K index, Total Totals index, Lifted should be verified and tuned using local storm climatology. index, Showalter index, CAPE, CIN, SWEAT, U index, Po- This is even more important when, as in northern Italy, mor- tential Instability index) were computed from Milano Linate phological characteristics that contribute to initiation and de- 00 UTC and 12 UTC soundings. The distribution of stability velopment of convective phenomena are very different from indexes’ values was studied in active and non-active morn- those of the areas for which indexes have been developed, and ings/afternoons in order to evaluate wich indexes are best for which the threshold have been tested. suited for identifying the potential for convective activity in In latest years different studies have been performed to in- the area. Optimal threshold values for thunderstorm occur- vestigate and/or adapt stability indexes to European regions rence were estimated for all of the stability indexes under (among others: Hacklander and Van Delden, 2003; Huntrieser study (maximising True Skill Statistic and ROC curves), for et al., 1997; Dalla Fontana, 2008; Manzato, 2003). This work 00 UTC and 12 UTC soundings separately. The uncertainty in follows their tracks applying some of the same methods to the estimated optimal threshold values and in the forecasting the study of convection in Lombardia. Thanks to the length skill of stability index above the threshold was evaluated by and completeness of the data set used, is has been possible to re-sampling techniques (Fig. 2). The technique was also used stratify cases with respect to time of day, location, intensity to rule out the possibility of difference in results being due to and synoptic configuration, with an emphasis on sampling of changes in the data set (overall number of cases, proportion of phenomena and on statistical significance of results. YES/NO cases). To study the characteristics of widespread and/or intense convection, and to discriminate it from isolated or weak thun- km× km II. PRESENTATION OF RESEARCH derstorm cases, the region was divided in a 29 29 grid (64 cells), and the occurrence and number of strikes in each grid cell was examined. The number of grid cells with at least km× km A study of thunderstorm occurrence in the 300 300 one lightning and the maximum number of lightnings per cell area including Lombardia (Northern Italy) was performed were used to identify days of severe/widespread convection, analysing cloud-to-ground lightnings detected by the CESI- and to study the difference in the distribution of stability in- SIRF system (http://www.fulmini.it/) over a ten year period dexes’ values in this case. TSS and ROC curves were used to identify optimal thresholds for severe convection, as in the preceding analysis of all active cases, and difference in the re- sults was checked for statistical significance by re-sampling the data and changing the proportion of events. The same analysis were also performed for thunderstorms occurring in different areas of Lombardia: north, south and in a 50km square centered on Linate sounding (to check for represen- tativity dependence). Lastly, thunderstorms occurring in dif- ferent synoptic configurations were analysed by subjectively FIG. 1: Monthly distribution of lightning number in each convective classifying flow type in four categories (trough, ridge, west- event of the 00-12UTC period. erly flow and weak flow, following Cacciamani et al., 1995).

157 2

for the 12-00UTC period. Good skill was found also for Showalter and K index (0.48 and 0.45 respectively). CAPE, CIN and SWEAT are, as expected, more difficult to interpret. Thresholds are comparable to those in literature, but optimal- ity seems to be gained for lower threshold values. For some in- dexes (K, LI, CAPE) the threshold differences in the 12 UTC and 00 UTC soundings are also significant. In general parcel indexes and CAPE exhibit higher skill in the northern, hilly, area of the region, whereas TT, K, U, PI are perform better when forecasting convection in the southern area. For intense/widespread convection cases, K index increases its forecasting skill and its optimal threshold values signifi- cantly (TSS from 0.45 to 0.54 and threshold from 24.0 C to 27.0 C, see figure 3); also CAPE’s and LI’s optimal thresh- olds move towards more unstable values, and the differences are significant even taking into account the reduced number of intense/widespread convection cases with respect to active cases (804 active cases, 175 intense/widespread cases on a to- FIG. 2: Distribution of TSS maximum position (above) and value tal of 1418). (below) for 1000 resamlpings of the data vs. fraction of total. The analysis of instability indexes for thunderstorms occur- ring in different synoptic configurations shows a significantly improved forecasting skill of some indexes indexes in sta- ble (ridge) or weakly unstable (westerly flow) configurations (CAPE, LI, PI, SH), with optimal threshold moving towards more unstable values. U and SWEAT indexes, which were not found to have good forecasting skill in general, appear in- stead to be useful in forecasting thunderstorm occurrence in the case of westerly flow. This work has been performed with the final aim of giving a set of quantitatively verified forecasting thresholds for the FIG. 3: Left: distribution of KI values in active/non active cases stability indexes most used in operational activity at Lombar- (red/blue). Right: distribution of KI values in intense/ weak or absent dia’s weather service. To the authors’ best knowledge, this convection (red/blue). is the first systematic study on stability indexes in Lombar- dia, and the first making the effort to give quantitative indi- cations on their use in forecasting thunderstorms for different Statistical significance of the difference in the results was con- areas/intensity/configurations, with indication on significance sidered. and uncertainty of results.

III. RESULTS AND CONCLUSIONS IV. REFERENCES The highest frequency of thundery days and the most se- vere/widespread convection was found to be in July, as ex- Cacciamani C., Battaglia F., Patruno P., Pomi L., Selvini pected (see Fig. 1). The characteristics of convection in the A., Tibaldi S., 1995: A climatological study of thunderstorm two periods (00-12 UTC and 12-00 UTC) as detected by activity in the Po Valley. Theoretical Applied Climatology, 50 lightning strikes was found to be in line with findings by 185-203. other authors for the same area using other observation meth- Dalla Fontana A., 2008: Tuning of a thunderstorm index for ods for convective activity, as hail or heavy rain observation, north-eastern Italy. Meteorological Applications, 15 475-482. SYNOP reports, damage reports, radar echos (Cacciamani et Hacklander A. J., Van Delden A., 2003: Thunderstorm pre- al., 1995). Thunderstorms are more frequent in the northern dictors and their forecast skill for the Netherlands. Atmo- area of Lombardia and in the afternoons. Early morning con- spheric Research, 67-68 273-299. vection and thunderstorms in the low lying southern areas are Huntrieser H., Schiesser H., Schmid W., Waldvogel A., less frequent in general, and their peak frequency is later on 1997: Comparison of traditional and newly developed thun- in the season (August-September vs. June-July). derstorm indices for Switzerland. Wether and Forecasting, 12 Of the stability indexes considered, TT index calculated on 108-125. 12UTC soundings has the highest skill in forecasting occur- Manzato A., 2003: A climatology of instability indices de- rence of thunderstorms in the following 12 hours in Lom- rived from Friuli Venezia Giulia soundings, using three differ- bardia (TSS = 0.51); in general forecasting skill is higher ent methods. Atmospheric Research, 67-68 417-454.

158