Atmos. Chem. Phys., 8, 921–953, 2008 www.atmos-chem-phys.net/8/921/2008/ © Author(s) 2008. This work is distributed under the Creative Commons Attribution 3.0 License. Atmospheric Chemistry and Physics Lightning activity in Brazilian thunderstorms during TROCCINOX: implications for NO x production H. Huntrieser1, U. Schumann1, H. Schlager1, H. Höller1, A. Giez2, H.-D. Betz3, D. Brunner4,*, C. Forster5,** , O. Pinto Jr.6, and R. Calheiros7 1Institut für Physik der Atmospḧare, Deutsches Zentrum für Luft- und Raumfahrt (DLR), Oberpfaffenhofen, Germany 2Flugabteilung, Deutsches Zentrum für Luft- und Raumfahrt (DLR), Oberpfaffenhofen, Germany 3Physics Department, University of Munich, Germany 4Institute for Atmospheric and Climate Science, ETH Zurich, Switzerland 5Norwegian Institute for Air Research (NILU), Atmosphere and Climate Change Department, Kjeller, Norway 6National Institute for Space Research, INPE, Brazil 7Instituto de Pesquisas Meteorológicas – Universidade Estadual Paulista, IPMet/UNESP, Bauru, Brazil * now at: Laboratory for Air Pollution and Environmental Technology, Empa, Swiss Federal Laboratories for Materials Testing and Research, Dübendorf, Switzerland ** now at: Institut f̈ur Physik der Atmospḧare, Deutsches Zentrum für Luft- und Raumfahrt (DLR), Oberpfaffenhofen, Germany Received: 12 September 2007 – Published in Atmos. Chem. Phys. Discuss.: 16 October 2007 Revised: 18 January 2008 – Accepted: 25 January 2008 – Published: 25 February 2008 Abstract. During the TROCCINOX field experiment in January and February 2005, the contribution of lightning- induced nitrogen oxides (LNOx) from tropical and subtropi- cal thunderstorms in Southern Brazil was investigated. Air- borne trace gas measurements (NO, NOy, CO and O3) were performed up to 12.5 km with the German research aircraft Falcon. During anvil penetrations in selected tropical and subtropical thunderstorms of 4 and 18 February, NOx mix- ing ratios were on average enhanced by 0.7–1.2 and 0.2– 0.8 nmol mol−1 totally, respectively. The relative contribu- tions of boundary layer NOx (BL-NOx) and LNOx to anvil- NOx were derived from the NOx-CO correlations. On aver- age∼80–90% of the anvil-NOx was attributed to LNOx. A Lightning Location Network (LINET) was set up to moni- tor the local distribution of cloud-to-ground (CG) and intra- cloud (IC) radiation sources (here called “strokes”) and com- pared with lightning data from the operational Brazilian net- work RINDAT (Rede Integrada Nacional de Detecção de Descargas Atmosféricas). The horizontal LNOx mass flux out of the anvil was determined from the mean LNOx mix- Correspondence to:H. Huntrieser (heidi.huntrieser@dlr.de) ing ratio, the horizontal outflow velocity and the size of the vertical cross-section of the anvil, and related to the num- ber of strokes contributing to LNOx. The values of these parameters were derived from the airborne measurements, from lightning and radar observations, and from a trajec- tory analysis. The amount of LNOx produced per LINET stroke depending on measured peak current was determined. The results were scaled up with the Lightning Imaging Sen- sor (LIS) flash rate (44 flashes s−1) to obtain an estimate of the global LNOx production rate. The final results gave ∼1 and∼2–3 kg(N) per LIS flash based on measurements in three tropical and one subtropical Brazilian thunderstorms, respectively, suggesting that tropical flashes may be less pro- ductive than subtropical ones. The equivalent mean annual global LNOx nitrogen mass production rate was estimated to be 1.6 and 3.1 Tg a−1, respectively. By use of LINET ob- servations in Germany in July 2005, a comparison with the lightning activity in mid-latitude thunderstorms was also per- formed. In general, the same frequency distribution of stroke peak currents as for tropical thunderstorms over Brazil was found. The different LNOx production rates per stroke in tropical thunderstorms compared with subtropical and mid- latitude thunderstorms seem to be related to the different stroke lengths (inferred from comparison with laboratory Published by Copernicus Publications on behalf of the European Geosciences Union. http://creativecommons.org/licenses/by/3.0/ 922 H. Huntrieser et al.: Lightning activity in Brazilian thunderstorms data and observed lengths). In comparison, the impact of other lightning parameters as stroke peak current and stroke release height was assessed to be minor. The results from TROCCINOX suggest that the different vertical wind shear may be responsible for the different stroke lengths. 1 Introduction A general introduction to the LNOx topic and overviews of past and present measurements of LNOx in thunderstorms are given in accompanying papers by Huntrieser et al. (2007) (HH07) and by Schumann and Huntrieser (2007) (SH07). Observations from local field experiments have been extrap- olated to the global scale to estimate the average amount of LNOx produced annually over the globe which is one crucial, yet highly uncertain, parameter in the global NOx budget. A LNOx nitrogen mass source strength between 2 and 20 Tg a−1 has frequently been given in the literature in the past (WMO, 1995; Bradshaw et al., 2000). More re- cently, lower values between 1 and 14 Tg a−1 have been re- ported based on estimates from airborne and satellite mea- surements (Huntrieser et al., 2002; Beirle et al., 2004, 2006; Ridley et al., 2004; Boersma et al., 2005; Ott et al., 2007). Furthermore, chemical transport models (CTMs) have been used to reduce the LNOx range by a comparison of modelled NOx concentrations, for different LNOx source strengths and vertical distributions, with local field and satellite measure- ments. The results obtained with model fits indicate best- estimate values for the global LNOx nitrogen mass between 2 and 8 Tg a−1 (SH07). Different methods have been used to estimate the amount of LNOx based on ground-based, airborne and laboratory measurements, and theoretical calculations, as reviewed by SH07. Airborne NOx measurements can be combined with lightning observations to estimate the amount of LNOx pro- duced per flash or per metre flash length. These numbers have been scaled up with the mean flash length and the an- nual global flash frequency. A large uncertainty in the es- timate of LNOx still results from the assumed NO produc- tion rates by CG and IC flashes (Martin et al., 2007). Up to now, it has been suggested that most components of a discharge produce NOx with varying, not determined effi- ciencies (Chameides, 1986; Coppens et al., 1998; Dye et al., 2000). In addition, it has been pointed out that the different flash lengths for CG and IC flashes may play an important role in the LNOx production rate (Defer et al., 2003). Re- cently Rahman et al. (2007) presented first direct measure- ments of NOx generated by rocket-triggered lightning in the field. Based on the results from a small data set of three trig- gered flashes, they suggest that it is the longer-lasting and continuous current portions of flashes that are responsible for most of the NO production. In comparison, the production by short-term return strokes was found to be minor. However, these longer and continuous current portions of flashes are currently not measured by operating lightning detection net- works as the National Lightning Detection Network (NLDN) and the very low frequency/low frequency (VLF/LF) light- ning location network LINET used here. NLDN only detects the high-current return stroke of a discharge. Results from the European Lightning Nitrogen Oxides Ex- periment (EULINOX) and Stratosphere-Troposphere Exper- iment: Radiation, Aerosols, and Ozone (STERAO) (DeCaria et al., 2000; Fehr et al., 2004; DeCaria et al., 2005; Ridley et al., 2005; Ott et al., 2007) indicate that IC flashes produce about as much NO per flash as CG flashes (IC/CG produc- tion ratio 0.5–2). In addition, laboratory results from Gal- lardo and Cooray (1996) and model simulations from Zhang et al. (2003) support that IC and CG flashes are similarly en- ergetic. On the other hand, laboratory studies by Wang et al. (1998) showed that LNOx depends less on energy and more on atmospheric pressure and the peak current of the flash. They concluded that “NO production per metre dis- charge length as a function of peak current appears to provide a more appropriate scaling factor for estimates of total global NO production”. The present study makes use of this finding by combining Wang et al. (1998) NOx measurements for lab- oratory flashes with our NOx and lightning peak current mea- surements from the field. First results were briefly presented in Huntrieser et al. (2006), indicating differences for tropical and subtropical thunderstorms in Brazil, which are discussed here in more detail. A further study is in preparation by Ott et al. (2008)1. The authors find that the mean peak currents and the NO production amounts per flash in five different thun- derstorms decrease with increasing latitude: the lowest value of NO production (360 moles/flash) was found for a EULI- NOX storm (48◦ N) and the largest value of NO production (700 moles/flash) was found for a CRYSTAL-FACE storm (26◦ N). Recently, Barthe et al. (2007) incorporated the re- lationship between produced LNOx per m laboratory spark and atmospheric pressure according to Wang et al. (1998) in their simulations with an explicit electrical scheme and a 3-D mesoscale model (Meso-NH). Up to now, only a few airborne experiments have been conducted that are suitable to provide an estimate of the LNOx production rate in the tropics (see SH07). In this pa- per we present measurements from the “Tropical Convection, Cirrus and Nitrogen Oxides Experiment“ (TROCCINOX) carried out in the wet season in January and February 2005 in the State of S̃ao Paulo and its surroundings in southern Brazil (10◦ S to 28◦ S and 38◦ W to 55◦ W). Both tropical and sub- tropical thunderstorms were investigated, since the operation area was located along the South Atlantic convergence zone 1 Ott, L. E., Pickering, K. E., DeCaria, A. J., Stenchikov, G. L., Lin, F.-F., Wang, D., Lang, S., and Tao, W.-K.: Production of light- ning NOx and its vertical distribution calculated from 3-D cloud scale chemical transport simulations, in preparation, J. Geophys. Res., 2008. Atmos. Chem. Phys., 8, 921–953, 2008 www.atmos-chem-phys.net/8/921/2008/ H. Huntrieser et al.: Lightning activity in Brazilian thunderstorms 923 (SACZ) (HH07). The main questions of this study are: 1.) How much LNOx is produced by these tropical and subtrop- ical thunderstorms? 2.) What are the relative contributions from strokes with different peak currents? 3.) How large is the LNOx production rate per stroke or flash? 4.) Is this LNOx production rate different for tropical and subtropical thunderstorms? 5.) What are the possible reasons for the difference? 6.) Can the findings from TROCCINOX help to explain the large LNOx productivity observed in Florida thunderstorms during CRYSTAL-FACE? To answer these questions we analyse airborne measure- ments of NO, NOy, CO, and O3 mixing ratios, the J(NO2) photolysis rate and meteorological parameters performed in the outflow of thunderstorms, trajectory analyses with the FLEXPART model and measurements from LINET, which was set up during TROCCINOX to monitor the local light- ning distribution (Sect. 2). This system registers VLF/LF radiation sources (here called “strokes”) from both CG and IC flashes. LINET data are compared with data from the operational Brazilian lightning detection network RINDAT (Sect. 3) and with LIS data (Sect. 4). Airborne NOx and ground-based lightning measurements are combined to give an estimate of the amount of LNOx produced per LINET stroke, and as a function of peak current according to Wang et al. (1998) (Sect. 4). From the ratio between LIS and LINET during one overpass, the amount of LNOx per LIS flash is determined. Knowing the global and annual LIS flash rate (44±5 flashes s−1, Christian et al., 2003; Chris- tian and Petersen, 2005), the equivalent annual global LNOx production rate based on individual TROCCINOX thunder- storms is estimated (Sect. 4). The different LNOx produc- tion rates estimated in tropical and subtropical thunderstorms are investigated through a comparison of LINET measure- ments, e.g. frequency distributions of stroke peak currents and mean peak currents (Sect. 5). In addition, the light- ning properties are compared with those in mid-latitude thun- derstorms over Germany, where the same lightning location network (LINET) was set up in July 2005. Airborne NOx measurements over Germany are available from previous campaigns (Huntrieser et al., 1998, 2002), but not for July 2005 (Sect. 5). The results are discussed and summarised in Sects. 6–7. The present study is the first to our knowl- edge that investigates whether tropical, subtropical and mid- latitude thunderstorms have different potentials to produce LNOx by combining lightning peak current measurements with airborne NOx and meteorological measurements. 2 Data and model description For general information on the TROCCINOX field exper- iment, see the papers by Schumann et al. (2004), HH07 and SH07. The following subsections describe the air- borne data obtained mainly from the research aircraft Fal- con of the Deutsches Zentrum für Luft- und Raumfahrt (DLR) as well as partly from the Russian M55 Geophys- ica aircraft (Sect. 2.1), lightning data from LINET, LIS and RINDAT (Sect. 2.2) and model simulations from FLEX- PART (Sect. 2.3). In addition, we use data from two S- band Doppler radars in Bauru (22.4◦ S, 49.0◦ W) and in Pres- idente Prudente (22.1◦ S, 51.4◦ W) operated by the Instituto de Pesquisas Meteorológicas (IPMet). Two different radar reflectivity products are presented: surveillance Plan Posi- tion Indicator (PPI, range 450 km) and 3.5 km Constant Al- titude PPI (CAPPI, range 240 km). The meteorological en- vironment of tropical, subtropical and mid-latitude thunder- storms was characterised with analysis data (temperature, water vapour mixing ratio, pressure, wind velocity and direc- tion) from the European Centre for Medium Range Weather Forecasts (ECMWF) with 3 h temporal resolution, 1◦ hori- zontal resolution and 60 vertical levels. The equivalent po- tential temperature is calculated as described in HH07. The separation of tropical and subtropical air masses is based on meteorological data, as already discussed for the two selected flights of 4 and 18 February 2005 in HH07. The 4 and 18 February flights were classified as tropical and subtropical, respectively. 2.1 Airborne instrumentation: Falcon and Geophysica Airborne measurements up to 12.5 km were carried out with the Falcon, which was equipped with DLR instruments to measure NO, NOy, O3, CO and J(NO2). The chemical in- strumentation is the same as that used during several DLR field campaigns in the past (HH07). Position, altitude, tem- perature, humidity, pressure and the 3-dimensional wind vec- tor (u, v andw) were measured with the standard Falcon me- teorological measurement systems (Schumann et al., 1995). Wind and pressure were measured with a Rosemount flow angle sensor (model 858) at the Falcon’s noseboom tip. The aerodynamic measurements were analysed according to an extensive in-flight calibration programme (Bögel and Bau- mann, 1991). In addition, NO and CO measurements were obtained from the high-flying Geophysica aircraft (∼20 km) (Ste- fanutti et al., 2004). The SIOUX instrument, developed and operated by the DLR, measures the NO mixing ratio (chemiluminescence technique) with a time resolution of 1 s, and an accuracy and precision of 10% and 5%, respec- tively. The CO-TDL instrument (cryogenic Tunable Diode Laser technique) operated by the Istituto Nazionale di Ottica Applica/Consiglio Nazionale delle Ricerche (INOA/CNR), measures the CO mixing ratio with an averaging time of 5 s, the accuracy and precision being 5% and 2%, respectively. All flight altitude values refer to pressure height and all times to UTC (Coordinated Universal Time) time (see also HH07). www.atmos-chem-phys.net/8/921/2008/ Atmos. Chem. Phys., 8, 921–953, 2008 924 H. Huntrieser et al.: Lightning activity in Brazilian thunderstorms Table 1. Positions of LINET sensors in Brazil (January–February 2005) and Germany (July 2005). Country/Station Name Latitude Longitude Brazil ◦ S ◦ W Marilia 22.235 49.965 Novo Horizonte 21.466 49.226 Bauru 22.358 49.027 Qurinhos 22.951 49.896 Araquara 21.813 48.199 Botucatu 22.848 48.432 Germany ◦ N ◦ E Ravensburg 47.801 09.696 Regensburg 49.043 12.103 Weissenburg 49.019 10.960 Berchtesgaden 47.634 13.001 Lahr 48.365 07.828 Karlsruhe 49.093 08.426 Basel 47.561 07.969 Bamberg 49.880 10.914 Weiden 49.667 12.184 Stegen 48.076 11.139 Passau 48.572 13.424 Garching 48.269 11.674 Peissenberg 47.801 11.010 Geretsried 47.870 11.476 Buchloe 48.037 10.728 Stadtbergen 48.349 10.850 Lichtenau 47.881 11.080 Lagerlechfeld 48.181 10.840 Oberpfaffenhofen 48.087 11.280 2.2 Lightning measurements: LINET, LIS and RINDAT During the TROCCINOX field campaign from 21 January to 27 February 2005, the VLF/LF (5–300 kHz) lightning detec- tion network LINET was used to monitor the local lightning distribution with high spatial resolution. LINET was set up by DLR in cooperation with IPMet. The network included six sensors from DLR to observe the area 19.5–24.5◦ S and 46.5–51.5◦ W (see Table 1). The average distance to the next closest sensor was∼80 km. For comparison of light- ning characteristics, measurements from southern Germany in summer 2005 with 19 sensors (from both DLR and the University of Munich), monitoring the area 47–51◦ N and 5– 14◦ E, were also included in this study (Table 1). The average distance to the next closest sensor was∼80 km in the outer region and∼20 km in the inner region. The basically similar features of the LINET arrays in Germany and Brazil allows for comparison of the characteristics of thunderstorms sys- tems in both regions (Schmidt et al., 2005). The LINET system has been developed by the Univer- sity of Munich and the sensor technology and measure- ment procedures have been described in detail by Betz et al. (2004), Schmidt et al. (2004, 2005), Betz et al. (2007a) and Schmidt (2007). For an overview of system character- istics see SH07. LINET continuously measures the tran- sient magnetic components of VLF/LF emissions from light- ning discharges. These signals are emitted by certain com- ponents of the flashes, and therefore a direct comparison with published flash statistics (e.g., IC/CG ratio) is not pos- sible. At the current stage, VLF sources are considered sep- arately. A routine algorithm to combine them into flashes is under development. However, a selected set of strokes were combined manually into flash “components” (nearby strokes within <1 s). These analyses indicate that LINET locates few VLF strokes per flash components, on average 3 and up to 9 (not shown). It is known that the amplitude of a measured electromag- netic signal is proportional to the peak current (Uman et al., 1975; Rakov et al., 1992; Cummins et al., 1998; Orville, 1999; Jerauld et al., 2005; Schulz et al., 2005). Thus, the peak current of LINET strokes is estimated from the VLF pulse amplitude. The registered amplitude depends on the distance between the VLF pulse and the measuring LINET sensor. The registered pulse is normalised by the recipro- cal value of the distance between pulse source and sensor, and averaged over all sensors that registered the VLF pulse. Owing to refined antenna techniques, optimised waveform handling and a shorter sensor base line of<100 km, a high detection efficiency of low peak currents is possible. The detection efficiency, stroke-current dependent, is highest in the LINET centre area (2◦×2◦) with >90% and decreases rapidly down to 30% towards the periphery (Betz et al., 2004, 2007a, b; Schmidt et al., 2007). Currents as low as∼1–2 kA can be detected by the system within the LINET centre area (periphery∼5 kA). In comparison, most other VLF/LF light- ning networks report only strokes>5–10 kA (Cummins et al., 1998). In addition to LINET data, spaceborne measurements from LIS onboard the Tropical Rainfall Measurement Mis- sion (TRMM) satellite (Christian et al., 1999; Thomas et al., 2000; Boccippio et al., 2002) were used to estimate the to- tal regional flash density (sum of CG and IC flashes) over the TROCCINOX area. For an overview of system charac- teristics see SH07. The sensor can view any area on its foot- print for a period of 90 s. This is long enough to estimate the flashing rate of most thunderstorms in the field of view during the passage (seehttp://thunder.msfc.nasa.gov/lis/). At noon the detection efficiency is 73±11% and at night 93±4% (Boccippio et al., 2002). Here we used LIS science prod- ucts (total count of flashes) from the “LIS space time domain search” (seehttp://thunder.nsstc.nasa.gov/lightning-cgi-bin/ lis/LISSearch.pl). A recent comparison between LIS and LINET data showed a good agreement between two systems that are based on completely different measurement tech- niques (Schmidt et al., 2005). Here LIS data for one overpass on 4 February 2005 were compared with LINET data (see Atmos. Chem. Phys., 8, 921–953, 2008 www.atmos-chem-phys.net/8/921/2008/ http://thunder.msfc.nasa.gov/lis/ http://thunder.nsstc.nasa.gov/lightning-cgi-bin/lis/LISSearch.pl http://thunder.nsstc.nasa.gov/lightning-cgi-bin/lis/LISSearch.pl H. Huntrieser et al.: Lightning activity in Brazilian thunderstorms 925 Sect. 4.5). LNOx estimates per LINET stroke were scaled up with LIS observations to provide an estimate of the regional and global strength of the LNOx production. LINET data were also compared with data from the oper- ational Brazilian lightning detection network RINDAT (see http://www.rindat.com.br/). Like LINET, RINDAT operates in the VLF/LF range. The detection efficiency for strokes with peak currents above 10 kA is 80–90% and the loca- tion accuracy is 0.5–2.0 km (Pinto and Pinto, 2003). The RINDAT system mainly registers CG flashes. A recent com- parison between LINET and RINDAT strokes indicates rea- sonable agreement for CG strokes when LINET peak cur- rents are above 12 kA (Schmidt et al., 2005); see further com- parisons in Sect. 3. 2.3 Transport modelling: FLEXPART The distribution of LNOx in the vicinity of thunderclouds was simulated with the Lagrangian particle dispersion model FLEXPART. General information on FLEXPART used for TROCCINOX is given in HH07. The model has mainly been used for studying long-range air pollution transport (e.g. Stohl et al., 2003a, b; Huntrieser et al., 2005), but also to investigate LNOx transport (Stohl et al., 2003b; Beirle et al., 2006; Cooper et al., 2006). The LNOx emissions used here as input for FLEXPART were based on lightning stroke data from the LINET sys- tem. LNOx was released uniformly in the vertical between 5 km (freezing level with negative charge centre) and 13 km altitude (cloud top) at the accurate horizontal position of ob- served VLF sources. Previous cloud model simulations for tropical continental thunderstorms (Pickering et al., 1998), indicate that LNOx was released mainly between 5 and 13 km increasing with altitude. However, because of the low resolution of the ECMWF wind fields (0.5◦ horizontally) used as input for FLEXPART, the distribution of lightning sources is assumed to be uniform in the vertical in the present study. A hundred particles were released per stroke, carrying the mass of LNOx produced (here set to 1 kg). The convec- tion scheme, used in these FLEXPART applications, trans- ports the particles upward into the anvil, from where they follow trajectories computed with the ECMWF wind fields. No quantitative estimate of the amount of LNOx is possible from these simulations; they can, however, be used to es- timate the extension of the LNOx field advected out of the anvil region. 3 Observations during the field experiment An overview of the observations on the two selected TROC- CINOX days, 4 and 18 February 2005 with thunderstorms in tropical and subtropical air masses, respectively, is given in HH07. Here we briefly focus on the performance of the lightning detection network LINET (Sect. 3.1) and on the Comparison LINET - RINDAT 04 February 2005 Longitude /°E -49.8 -49.6 -49.4 -49.2 -49.0 -48.8 -48.6 -48.4 -48.2 La tit ud e /° N -22.6 -22.4 -22.2 -22.0 -21.8 -21.6 LINET RINDAT (a) 18 February 2005 Longitude /°E -49.4 -49.2 -49.0 -48.8 -48.6 -48.4 -48.2 -48.0 -47.8 -47.6 La tit ud e /° N -20.4 -20.2 -20.0 -19.8 -19.6 -19.4 -19.2 (b) -21.4 Fig. 1. Horizontal distributions of RINDAT and LINET strokes for the (a) 4 February 2005 in the centre area of the LINET detection network and for the(b) 18 February 2005 along the northern pe- riphery, 00:00 UTC–24:00 UTC. representation of the Falcon measurements in the anvil out- flow (Sect. 3.2). 3.1 Performance of LINET compared with RINDAT To evaluate the performance of the LINET system in more detail, a comparison with the operational lightning detec- tion network in Brazil (RINDAT) was carried out for 4 and 18 February 2005. Horizontal distributions of RINDAT and LINET strokes were compared for the LINET centre area on 4 February 2005 (Fig. 1a), and for the northern LINET periphery area on 18 February 2005 (Fig. 1b), 00:00 UTC– 24:00 UTC. Overall, a general agreement was found, but with a slight shift of RINDAT strokes to the west compared with LINET strokes, especially in Fig. 1b. In some areas the density of LINET strokes was much larger than of RINDAT strokes (probably because IC strokes and strokes with low peak are not registered by RINDAT). The correlations be- tween LINET and RINDAT peak currents (absolute values) www.atmos-chem-phys.net/8/921/2008/ Atmos. Chem. Phys., 8, 921–953, 2008 http://www.rindat.com.br/ 926 H. Huntrieser et al.: Lightning activity in Brazilian thunderstorms 18 February 2005 LINET peak current /kA 0 20 40 60 80 100 120 140 160 R IN D A T pe ak c ur re nt /k A 0 20 40 60 80 100 120 140 160 peak current LINET vs. RINDAT regression LINET vs. RINDAT PCRINDAT = -0.76 +0.68*PCLINET r2 = 0.95 1 to 1 relationship (b) 04 February 2005 LINET peak current /kA 0 20 40 60 80 100 120 140 160 R IN D A T pe ak c ur re nt /k A 0 20 40 60 80 100 120 140 160 peak current LINET vs. RINDAT regression LINET vs. RINDAT PCRINDAT = -2.58 + 0.83*PCLINET r2 = 0.92 1 to 1 relationship (a) Fig. 2. Correlation between LINET and RINDAT peak currents (PC) for 222 and 173 selected strokes on(a) 4 and(b) 18 February 2005, respectively, in the areas shown in Fig. 1a–b. for 222 and 173 selected strokes of 4 February (21.5–22.5◦ S and 48.5–49.5◦ W) and 18 February (19.4–20.0◦ S and 47.7– 49.2◦ W) are shown in Fig. 2a and b, respectively. On 4 February only negative CG strokes were compared. On 18 February no separation between CG and IC strokes was pos- sible for the LINET data, owing to the location of the selected strokes along the northern periphery area. LINET strokes were therefore compared with both positive and negative CG strokes from RINDAT. About 10% of the selected LINET strokes were positive ones. The strokes shown in Fig. 2 were selected manually to represent peak current values over the entire current range. LINET peak currents above 13–14 kA are in general also detected by RINDAT, occasionally even LINET peak currents down to 7 kA. The slope (0.83) seen in Fig. 2a indicates that a 20 kA LINET stroke is on aver- age registered as 14 kA by RINDAT. In Fig. 2b the slope is slightly lower (0.68) owing to the lower LINET detection ef- ficiency along the northern periphery. In addition, the mean peak current is higher for LINET strokes (35 kA in Fig. 2a and 31 kA in Fig. 2b) than for RINDAT strokes (27 kA in Fig. 2a and 20 kA in Fig. 2b). The high correlation coef- ficient (r2= 0.92 and 0.95, respectively) between the peak currents of the two systems indicates a good agreement in general. Lower RINDAT CG+ peak currents (<30 kA) are frequently registered as IC+ by LINET (∼40%) and stronger RINDAT CG- peak currents (>100 kA) are frequently reg- istered as IC- by LINET (∼40%). This finding can be com- pared with results from EULINOX in Germany where flashes registered with a LPATS system (same technology as used for RINDAT) were compared with the French Office National d’Etudes et de Recherches Aérospatiales (ONERA) VHF in- terferometer measurements. Théry (2001) found that 61% of the positive LPATS flashes (those of low intensity) and 32% of the negative LPATS flashes were in fact IC flashes. A re- cent study by Pinto et al. (2007) also confirmed that a large percentage of the positive CG flashes registered by RINDAT over Brazil are in fact IC flashes. For the analysed dataset we found that weak positive RINDAT peak currents (<10 kA) are occasionally (<10%) registered as negative strokes by LINET. 3.2 NOx in the anvil outflow derived from aircraft measure- ments In HH07 it was briefly discussed whether the outflow altitude where LNOx maximises was reached with the Falcon aircraft (important question for comparison with results from other field campaigns and for further calculations in Sect. 4). It was concluded that this altitude was reached with certainty on 18 February, but on 4 February the Falcon measured the largest mixing ratios in the uppermost flight levels so that larger mixing ratios at higher altitudes inside the anvil cannot be excluded. Therefore, for the latter day measurements from the high-flying Geophysica in the upper part of this thunder- storm were briefly analysed as discussed below. The Falcon measurements in two of the anvils of 4 Febru- ary (anvil 1a and 5a, listed in Table 2a and described in Sect. 4.1) can be compared with coincident measurements with the high-flying Geophysica. The Geophysica pene- trated anvil 1a during ascent between 15.9 and 16.6 km (pen- etration at flight time: 67 070–67 298 s, at position: 21.3– 21.5◦ S and 49.1–49.3◦ W) and anvil 5a during descent be- tween 17.2 and 16.5 km (penetration at flight time: 66 569– 66 696 s, at position: 21.8–21.9◦ S and 48.5–48.7◦ W). The mean NO mixing ratios in anvil 1a and anvil 5a were 0.30 and 0.35 nmol mol−1, respectively. The mean anvil-NO mix- ing ratio is the mean value of all NO 1s-values measured be- tween the entrance and exit of the anvil (determined from the distinct increase and decrease in the NO mixing ra- tio). The mean mixing ratios measured by the Geophys- ica between∼16–17 km altitudes are distinctly lower than the NO mixing ratios measured by the Falcon at lower alti- tudes (10.6–10.7 km): 0.80 and 1.16 nmol mol−1 in anvil 1a and anvil 5a, respectively. The Geophysica measurements in Atmos. Chem. Phys., 8, 921–953, 2008 www.atmos-chem-phys.net/8/921/2008/ H. Huntrieser et al.: Lightning activity in Brazilian thunderstorms 927 Table 2a.Estimates of horizontal LNOx mass fluxFLNOx, LINET stroke rateRLINET , LNOx production rate per LINET stroke and per LIS flashPLNOx, and global LNOx production rate per yearGLNOx. Flight Entry and Exit Pressure Mean, |Va − Vs | 1, ρa , kg 1x, 1z,km F2 LNOx, R3 LINET , PLNOx, g(N) PLNOx, g(N) GLNOx, and Anvil Time (UTC), s Altitude, χLNOx m s−1 m−3 km g(N) s−1 (LINET (LINET (LIS flash)−1 Tg(N) a−1 Penetration / km nmol strokes)s−1 stroke)−1 tropical (t) mol−1 or subtropical (s) 0402051a (t) 66 199–66 433 10.6 0.76 6.5 0.36 35 4 120 0.055 2205 1103 1.5 0402055a (t) 67 682–67 833 10.7 1.10 5.9 0.36 25 4 113 0.054 2082 1041 1.4 0402052b (t) 69 905–70 169 10.1 0.57 9.2 0.39 45 4 178 0.061 2914 1457 2.0 mean tropical4 2400 1200 1.6 180205bI (s) 74 056–74 209 10.6 0.42 17.7 0.36 28 3 109 0.025 4258 2129 3.0 180205bII (s) 74 453–74 623 10.7 0.18 20.0 0.36 33 3 62 0.025 2430 1215 1.7 180205bIII (s) 75 013–75 186 10.1 0.65 12.2 0.39 32 3 143 0.025 5623 2811 3.9 180205bIV (s) 75 601–75 761 10.1 0.21 20.0 0.39 30 3 71 0.025 2792 1396 1.9 180205bV (s) 76 102–76 280 9.4 0.39 11.9 0.41 33 3 91 0.025 3568 1784 2.5 180205bVI (s) 76 584–76 757 9.4 0.13 17.7 0.41 35 3 48 0.025 1876 938 1.3 mean subtropical4 4483 2241 3.1 relative max. error ∼50% ∼50% ∼40% ∼50% ∼190% ∼90% ∼280% ∼310% ∼320% 1 Horizontal anvil outflow velocity, calculated from values in Table 2b. 2 The horizontal LNOx mass flux out of the anvil, see Eq. (4). 3 Only LINET strokes with peak currents≥10 kA were considered for an equivalent comparison between 040205 (strokes mainly inside the LINET centre) and 180205b (strokes along LINET periphery). 4 The mean value for the tropical anvil penetrations 1a, 5a and 2b of 4 February 2005 is given. The mean value for the subtropical anvil penetrations I, III, and V of 18 February 2005 (penetrations closest to the maximum anvil outflow) is given. the anvils on 4 February indicate an increase in NO mixing ratios with decreasing altitude, opposite to the Falcon mea- surements. Hence the outflow level where NO mixing ratios maximise was likely to be located between the altitudes at which the Falcon and Geophysica penetrated the anvils. CO measurements from the Geophysica (personal communica- tion P. Mazzinghi, INOA/CNR) can be used to determine this outflow level more precisely (on the assumtion that LNOx maximises where CO maximises). The vertical CO profile (ascent and descent in the vicinity of the selected anvils) shows enhanced mixing ratios mainly between∼10–14 km altitude. The mixing ratios were rather constant throughout this layer,∼130–140 nmol mol−1. The outflow level, where the CO mixing ratio maximises (132–138 nmol mol−1), was located between∼12.0–12.5 km, about 1.5–2 km above the Falcon penetration. At the levels where the Falcon pene- trated the anvils (10.6–10.7 km), however, the CO mixing ratio (132 nmol mol−1) was similar to the lowest Geophys- ica CO mixing ratios in the outflow level. The Falcon data may therefore underestimate the mean NO mixing ratios in the selected anvils to a degree which cannot be quantified from the available dataset. These mean NO mixing ratios are needed for further calculations in the next section. Prelimi- nary results from cloud-resolved modelling for the 4 Febru- ary thunderstorms by Pickering et al. (2007) suggest that the anvil outflow NO maximum is located between 12 and 13 km, which supports our estimates derived from the verti- cal CO profile. Table 2b. Measured wind velocity and direction in the anvil outflow and at the steering level1. Flight and Measured Measured Wind Wind Anvil Wind Wind Direction Velocity Penetration Direction Velocity in at Steering at in Anvil Anvil Level1ds , ◦ Steering Outflow Outflow Level1Vs , da , ◦ Va , m s−1 m s−1 0402051a 71±31 5.2±1.7 160 3.9 0402055a 182±50 3.9±2.7 280 3.9 0402052b 107±23 6.9±2.0 350 3.7 180205bI 283±12 16.6±4.1 185 4.3 180205bII 277±4 19.4±2.3 185 4.3 180205bIII 274±11 11.5±5.3 185 4.3 180205bIV 279±5 19.2±1.3 185 4.3 180205bV 268±10 11.6±5.7 185 4.3 180205bVI 278±5 16.9±1.4 185 4.3 1 The wind at the steering level∼3 km (∼700 hPa) determines the mean motion of a thunderstorm cell. 4 Estimate of the LNOx production rate per flash and per year In this section the measurements in selected tropical and sub- tropical thunderstorms of 4 and 18 February 2005 are dis- cussed in more detail. The spatial and temporal distribu- tions of LINET strokes are presented (Sect. 4.1). The con- tribution from observed LINET strokes to measured anvil- NOx mass and the resulting LINET stroke rates are estimated (Sect. 4.2). Furthermore, the contribution of BL-NOx and www.atmos-chem-phys.net/8/921/2008/ Atmos. Chem. Phys., 8, 921–953, 2008 928 H. Huntrieser et al.: Lightning activity in Brazilian thunderstorms Method to estimate the global annual LNOx production based on field measurements mean anvil-NOx Falcon LINET strokes mean BL- NOx subtracted mean χLNOx |Va-Vs| ρa T, p Δx, Δz u, v Radar hor. LNOx mass flux F(LNOx) out of anvil LNOx production rate per LINET stroke P(LNOx)LINET LNOx production rate per LIS flash P(LNOx)LIS global LNOx production rate per year G(LNOx) global LIS flash rate 44 s-1 LINET scaled with LIS stroke rate in selected TS strokes contributing to anvil-NOx selected TS FLEXPART u, v LIS Geo- physica Fig. 3. Flow chart showing the introduced method to estimate the annual global LNOx production rate from TROCCINOX field measure- ments (Falcon, Geophysica, LINET, Radar) combined with LIS data and model simulations (FLEXPART) of the selected thunderstorms (TS) all indicated with grey background. Different line colours are used to avoid misunderstanding of the flow direction at line intersections. LNOx to measured anvil-NOx is estimated (Sect. 4.3). The horizontal LNOx mass flux rate out of the anvils is calculated by means of estimated LNOx mixing ratios and horizontal outflow wind velocities from the flights combined with the size of the vertical cross-section of the anvils (Sect. 4.4). LNOx nitrogen mass flux rates (g s−1) and LINET stroke rates (strokes s−1) are combined to estimate the production rate of LNOx (in g of nitrogen mass or number of NOx molecules) per LINET stroke and per LIS flash (Sect. 4.5). Finally, the annual global LNOx nitrogen mass production rate is estimated (in Tg a−1). Figure 3 gives an overview of these different steps described in detail in the following sub- sections, starting with the selection of a thunderstorm (TS) and ending with an estimate of the annual global LNOx pro- duction rate G(LNOx). 4.1 Spatial and temporal LINET stroke distributions The spatial distributions of LINET strokes of 4 and 18 Febru- ary 2005 are shown in Fig. 4a and b, respectively. For the selected thunderstorms, strokes occurring before the penetra- tions by the Falcon are highlighted in colour. Superimposed is the Falcon track showing the successful, repeated penetra- tions of the subtropical thunderstorm system of 18 February, and the zigzag pattern between the tropical thunderstorms (labelled 1a, 5a and 2b) of 4 February. The time periods of the anvil penetrations are listed in Table 2a. The direction of the thunderstorm movement (red arrows in Fig. 4) is in- ferred from lightning data. The main wind direction in the anvil outflow (green arrows), as inferred from Falcon wind measurements, controls the transport of LNOx out of the anvils. On 4 February the prevailing wind direction in the flight level (influenced by the Bolivian High, see Fig. 5c in HH07) varied between north-east and south-east in vicinity of anvil 1a and 2b, and was from the south-west in the vicin- ity of anvil 5a. The NOx mixing ratio along the flight track is also superimposed in Fig. 4. Elevated mixing ratios ex- ceeding 0.6 nmol mol−1 NOx were frequently measured in the anvil outflow downwind of nearby lightning strokes. The selected thunderstorms of 4 and 18 February occurred in the centre and at the northern border line of the LINET network, respectively. Because of a higher sensitivity in the network centre, the fraction of strokes with low currents (<10 kA) was much higher on 4 February (87%) than on 18 February (45%). For the latter thunderstorm system no sep- aration between IC and CG strokes was possible because of the large distance from the centre. For an adequate compar- ison of the stroke rates in these storms, it was necessary to restrict comparisons to higher stroke peak currents (≥10 kA) which were observed with about the same detection effi- ciency, independently of their location within the LINET net- work. On 4 February strokes were widespread with some at the LINET periphery. LINET strokes were therefore com- pared with LIS flashes and RINDAT strokes to determine the Atmos. Chem. Phys., 8, 921–953, 2008 www.atmos-chem-phys.net/8/921/2008/ H. Huntrieser et al.: Lightning activity in Brazilian thunderstorms 929 Fig. 4. Horizontal distributions of LINET strokes on(a) 4 and(b) 18 February 2005. All strokes registered before the Falcon penetration within the selected thunderstorm systems on 4 February (tropical: labelled 1a, 5a and 2b) and 18 February (subtropical) are coloured. Falcon flight paths and NOx mixing ratios are superimposed (colour/grey scale). The red arrows indicate the direction of the storm motion and the green arrows the main wind direction in the anvil outflow. In addition, the positions of the 6 LINET sensors listed in Table 1 are indicated in (a). detection efficiency of the LINET system relative to the other two systems. The change in detection efficiency for these se- lected LINET strokes towards the LINET periphery was only minor (<10%) compared with the other two systems and not considered further. The temporal distributions of LINET stroke rates in the se- lected thunderstorms for peak currents≥10 kA are presented in Fig. 5. The storms of 4 February were mainly in a mature stage during the aircraft passage. In comparison, the long- lived storm system of 18 February was in a decaying stage and probed long after the peak lightning activity (first light- ning was registered already 6 h before the first penetration). www.atmos-chem-phys.net/8/921/2008/ Atmos. Chem. Phys., 8, 921–953, 2008 930 H. Huntrieser et al.: Lightning activity in Brazilian thunderstorms LINET stroke distributions UTC 14 15 16 17 18 19 20 21 nu m be r o f L IN E T st ro ke s (5 m in .)-1 0 20 40 60 80 100 180205: anvil I-VI 040205: anvil 1a 040205: anvil 2b 040205: anvil 5a I-VI1a 2b5a Fig. 5. Time series of LINET stroke rates for the selected thun- derstorms (only strokes with peak currents≥10 kA considered). On 4 February 2005, stroke rates in the investigated thunderstorms (active) are shown from storm initiation until penetration (tropi- cal: labelled 1a, 5a and 2b). On 18 February 2005, the stroke rate within the selected thunderstorm system (subtropical) is shown from storm initiation until decay. The repeated penetration started first at 20:34 UTC (labelled I-VI) when the lightning activity de- cayed. 4.2 Contribution of LNOx to anvil-NOx and determining LINET stroke rates For evaluation of the LNOx production rate per stroke, it is necessary to estimate which of the LINET strokes during the storm lifetime contributed to the measured anvil-NOx en- hancement and its horizontal and vertical extension. This is a very difficult task which might be best performed by using cloud-scale modelling. First cloud-resolving model simulations have been performed for selected TROCCINOX storms of 4 February 2005 by Chaboureau et al. (2007) and by Pickering et al. (2007) but cloud-resolving simulations are presently not available for the thunderstorm system of 18 February 2005. Instead we make use of FLEXPART light- ning tracer simulations, as explained in Sect. 2.3. The simu- lations follow lightning tracers from the horizontal LINET stroke distributions (Fig. 4b) using ECMWF wind fields. For the long-lived thunderstorm case of 18 February, am- bient wind velocities were strong and LNOx was advected far downwind. The ECMWF wind agrees well with Falcon measurements of wind velocity and direction, except in the core of the anvil penetrations (Fig. 6). As shown later in this section, comparison of the ECMWF wind fields and FLEX- PART results with radar and airborne wind and anvil-NOx observations, as indicated in Fig. 3, supports the validity of the FLEXPART simulations of the 18 February thunderstorm system, in spite of the coarse horizontal resolution (0.5◦) of the ECMWF wind velocity fields used. For the thunderstorm system of 18 February, a time se- quence of FLEXPART lightning tracer simulations (Fig. 7) indicates a rather fast development of an elongated area with enhanced LNOx downwind of the storm system, following the wind in the upper troposphere (UT). Tracer distributions for six different simulations are shown in this figure (out- put resolution: 30 min and 0.08 degrees, horizontal cross- sections at 10 km altitude corresponding to the flight level) considering transport of emissions from strokes in various time intervals. The simulated tracer distributions may be compared with the anvil-NOx observations from the Falcon (Fig. 4b). Only the last four simulations (Fig. 7c–f) indicate distinctly enhanced LNOx along the right hand side anvil transect, as observed by the Falcon. Furthermore, mixing ratios in the left hand side transects, closer to the core, were twice as high as in the right hand side transects. Given the measured UT wind velocity of 15 to 20 m s−1, it is clear that strokes that occurred between 19:00–19:30 UTC (along the left hand side anvil transect, 49.7–49.8◦ W) or earlier do not contribute to the anvil-NOx enhancement observed along the right hand side anvil transect. The air with enhanced LNOx is advected further downwind to the right in Fig. 7d–f. Only strokes after 19:30 and before 20:55 UTC (when the storm decayed) (Fig. 7c) were therefore considered to have con- tributed to the observed anvil-NOx. During this 85 min pe- riod about 130 strokes with peak currents≥10 kA were de- tected, corresponding to a stroke rate of 0.025 strokes s−1 (Table 2a). For the 18 February 2005 thunderstorm system, the hor- izontal extension (1x) of FLEXPART lightning tracer in Fig. 7c, perpendicular to the wind direction (see Fig. 4b), was estimated to be∼30–35 km. This width agrees well with the extension of the flight path segment with enhanced NOx ob- served during the single anvil transects (28–35 km); see the grey scale along the flight track in Fig. 4b and Table 2. This parameter (1x) will be used to estimate the horizontal LNOx mass flux out of the anvil in Sect. 4.4. Finally, a radar image of the 18 February thunderstorm system (Bauru radar, elevation angle 0◦), indicates a pro- nounced, elongated structure of the storm system (Fig. 8), similar to the FLEXPART result at 10 km altitude. The 18 February thunderstorm system is located in the upper, north- ern domain of the radar range, about 240 km from the radar site. Unfortunately, the radar information is sparse in this re- gion and no more detailed data are available since the domain is out of the quantification range where volumetric data are collected. For the thunderstorms of 4 February, no FLEXPART simu- lations were performed since the storms just developed about one hour before the penetrations and this time was consid- ered too short for realistic simulations. In addition, the am- bient UT wind velocities were low (4–7 m s−1) and LNOx remained in the vicinity of the storms. Instead, as indicated in Fig. 3, the LNOx production rate per stroke and the width 1x were estimated from a combination of horizontal LINET Atmos. Chem. Phys., 8, 921–953, 2008 www.atmos-chem-phys.net/8/921/2008/ H. Huntrieser et al.: Lightning activity in Brazilian thunderstorms 931 Fig. 6. Comparison between Falcon and ECMWF wind velocity (in black and blue, respectively) and wind direction (in red and orange, respectively) for the flight on 18 February 2005. stroke distributions, radar images, Falcon wind and anvil- NOx observations. The average altitude (arithmetical mean) of all IC strokes (Table 4a) in anvil 1a (10.0 km) and anvil 5a (11.6 km) was below or just above the flight level (10.6 km and 10.7 km, respectively; see Table 2a), indicating that the majority of LNOx, produced by the observed strokes left the anvil at about the flight level. It is assumed that all LINET strokes observed in the vicinity of these storms between storm initiation and Fal- con penetration (coloured in Fig. 4a) contributed to the ob- served anvil-NOx (Fig. 4a). (This is a working hypothesis with large uncertainties, which cannot be quantified with- out cloud-model simulations.) For comparison with the 18 February case, only the number of strokes with peak cur- rents≥10 kA is counted. In anvil 1a about 278 strokes were registered between 16:55 and 18:20 UTC, in anvil 5a about 130 strokes were registered between 18:05 and 18:45 UTC and in anvil 2b about 311 strokes were registered between 17:55 and 19:20 UTC; see Fig. 5, which corresponds to the following stroke rates: 0.055, 0.054 and 0.061 strokes s−1, respectively (see Table 2a). For each of the three anvil penetrations, the width (1x) of the LNOx plume perpendicular to the wind direction was es- timated from the horizontal LINET stroke distribution, from the anvil-NOx observations (Fig. 4a) and from the radar im- ages at the time of the penetrations (shown only hourly in Fig. 9). The1x values are∼35, ∼25 and∼45 km for anvils 1a, 5a and 2b, respectively (see Table 2a). 4.3 Contribution of BL-NOx to anvil-NOx The boundary layer (BL) contribution (χBL−NOx) to the NOx mixing ratio in the anvil (χAnvil−NOx) is derived from the cor- relation between NOx and CO mixing ratios in the BL and in the anvil. It is assumed, that BL air is transported upwards rapidly within strong, well-developed updrafts with little am- bient mixing and without chemical loss of NOx and CO. Hence, about the same CO mixing ratio is observed in the main anvil outflow (χAnvil−CO) as in the BL layer (χBL−CO): χAnvil−CO = χBL−CO (1) and LNOx (χLNOx) is the difference between anvil-NOx and BL-NOx : χLNOx = χAnvil−NOx − χBL−NOx (2) These assumptions are supported by cloud-model simula- tions (Pickering et al., 1992; Thompson et al., 1997; Ott et al., 2007) and airborne thunderstorm observations (Dicker- son et al., 1987; Hauf et al., 1995; Huntrieser et al., 1998, 2002; Ḧoller et al., 1999; Lopez et al., 2006; Bertram et al., 2007; Koike et al., 2007). The ratio of NOx to CO in the BL (<2 km) is conserved during the rapid upward transport into the anvil: χBL−NOx/χBL−CO = (χAnvil−NOx − χLNOx)/χAnvil−CO (3) Vertical NOx, CO, and O3-profiles from the 18 February flight are shown in Fig. 10a. The CO mixing ratios, mea- sured during the anvil penetrations, are in a similar range as those measured at∼2 km altitude (see red box), supporting the assumption of rapid upward transport from the top of the BL into the anvils. Unfortunately, no NOx measurements are available below 3 km for this flight. Instead, NOx measure- ments in the BL were only available for ten TROCCINOX “fair weather” flights without active thunderstorms (Fig. 1a in HH07). NOx and CO data from all available flights in the BL (<2 km) were therefore used to estimate the average BL NOx-CO correlation. It can be justified that this rela- tionship is representative, since CO mixing ratios in the BL were in the same range both for “thunderstorm” and for “fair weather” flights. The BL data were sampled mainly during take-off and landing near the campaign base. Hence, it was assumed that these values are representative for the entire BL covered by the selected flights. In Fig. 10b the correlation between measured NOx and CO for the Falcon flight of 18 February is shown (black dots). Different types of air mass origin (Pacific, Amazon basin, anvil and background), as discussed in HH07, are www.atmos-chem-phys.net/8/921/2008/ Atmos. Chem. Phys., 8, 921–953, 2008 932 H. Huntrieser et al.: Lightning activity in Brazilian thunderstorms a) b) c) d) e) f) 20:30-21:00 20:00-21:00 19:30-21:00 19:00-21:00 18:30-21:00 18:00-21:00 pmol mol -1 Fig. 7. FLEXPART lightning tracer simulations (NOx at 10 km in pmol mol−1) for the 18 February 2005 subtropical thunderstorm system considering all LINET strokes (black dots) after(a)20:30 UTC,(b) 20:00 UTC,(c)19:30 UTC,(d) 19:00 UTC,(e)18:30 UTC,(f) 18:00 UTC and until 21:00 UTC. The Falcon track is superimposed in red. marked. The measured NOx mixing ratios were mainly be- low 0.2 nmol mol−1, except during the anvil penetrations. The average NOx-CO correlation in the BL for all TROC- CINOX flights (data from Fig. 1 in HH07) is also shown in Fig. 10b (red-yellow dots). Average CO mixing ratios dur- ing the anvil penetrations of the 18 and 4 February flights were 95–105 and 105–115 nmol mol−1, respectively. From the measured BL-CO (90–120 nmol mol−1) and the correla- tion, the average BL-NOx mixing ratio and its standard de- viation (std) were estimated to be 0.11±0.07 nmol mol−1. For the anvil penetrations of 4 and 18 February listed in Table 2a, average LNOx volume mixing ratios (χLNOx) were determined by subtraction of the mean BL-NOx con- tribution (0.11 nmol mol−1) from the mean anvil-NOx val- ues. The mean values for anvil-NOx range between 0.2– 0.8 nmol mol−1 in the subtropical thunderstorm of 18 Febru- ary and between 0.7–1.2 nmol mol−1 in the tropical thun- derstorms of 4 February (Table 2a in HH07). As a result, χLNOx values in the range from 0.1 to 1.1 nmol mol−1 were obtained, as listed in Table 2a in the present paper. Overall, the contribution of BL-NOx to anvil-NOx in the selected thunderstorms of 4 February (anvil 1a, 5a, and 2b) and 18 February (only anvil penetrations I, III and V clos- est to the core considered here) was∼10–20%. This range Atmos. Chem. Phys., 8, 921–953, 2008 www.atmos-chem-phys.net/8/921/2008/ H. Huntrieser et al.: Lightning activity in Brazilian thunderstorms 933 a) b) c) d) 19.0°S 20.0°S 49.5°W 48.0°W 18:00 UTC 19:00 UTC 20:00 UTC 21:00 UTC 21:00 UTC Fig. 8. Radar reflectivity as PPI scan (PPISURVEILLANCE operational product) in dBZ units measured at 0.0◦ elevation by the Bauru radar (22.4◦ S, 49.0◦ W) for the 18 February 2005 subtropical thunderstorm system (marked in red) at(a) 18:00 UTC,(b) 19:00 UTC,(c) 20:00 UTC and(d) 21:00 UTC. is slightly lower than the average found in European thun- derstorms with 25 to 40% (Huntrieser et al., 1998, 2002). In the investigated TROCCINOX thunderstorms, the contri- bution from LNOx clearly dominated the anvil-NOx budget with ∼80–90%. This contribution is higher than observed during the TRACE-A experiment at the end of the dry (burn- ing) season, where only 30–40% of anvil-NOx was attributed to LNOx (Pickering et al., 1996). 4.4 Estimate of the horizontal LNOx mass flux Cloud-model simulations indicate that most LNOx produced in a thunderstorm is transported into the anvil (Skamarock et al., 2003). If the total LNOx mass in the anvil region (depen- dent on the LNOx mixing ratio and the volume covered by this LNOx) and the total number of flashes in the thunder- storm that contributed to this LNOx were known, the LNOx production rate per flash could be estimated, assuming a con- stant LNOx production per flash. Up to now, however, no method exists which can determine the required parameters exactly. Model approaches have e.g. estimated the horizontal NOx flux out of the anvil through a vertical control surface (Skamarock et al., 2003; Barthe et al., 2007). A combination of in situ aircraft observations and cloud-model simulations was used to separate the outflow flux into a LNOx flux and an environmental NOx flux. This approach was originally introduced by Chameides et al. (1987) for airborne measure- ments in thunderstorms during GTE/CITE and has also been applied by us for measurements in LINOX and EULINOX thunderstorms (Huntrieser et al., 1998, 2002). Alternatively, the NO content in the thunderstorm is estimated from the product of airborne in situ measurements of NO at certain levels in the anvil and the estimated volume of the appro- priate cloud segments (Ridley et al., 2004). The total vol- ume is derived from the sum of the vertically staggered flight segments. The two methods are described in more detail in SH07. The TROCCINOX thunderstorm penetrations listed in Ta- ble 2a provide only snapshots of the conditions at a certain level of the cloud at a certain time. It is not known how rep- resentative these anvil penetrations are for the average anvil conditions (see also discussion in Sect. 3.2). These are, how- ever, the only measurements that are available. Time se- ries of trace gas measurements (NOx, CO, and O3) during the penetrations listed in Table 2a have already been pre- sented and discussed in HH07. On the 18 February flight, the anvil outflow from the selected thunderstorm system was successfully penetrated 6 times (Fig. 4b). In addition to the mentioned trace gases, NOy was measured and mixing ratios during the 6 penetrations are shown together with the ver- tical velocity (absolute values) in Fig. 11. The 1 s absolute www.atmos-chem-phys.net/8/921/2008/ Atmos. Chem. Phys., 8, 921–953, 2008 934 H. Huntrieser et al.: Lightning activity in Brazilian thunderstorms * * (b) (a) 48.0°W50.0°W 20.0°S 22.0°S * * Fig. 9. Vertical maximum of the radar reflectivity (max CAPPI frame, unit dBZ). Composite from the Bauru (22.4◦ S, 49.0◦ W) and Presidente Prudente (22.1◦ S, 51.4◦ W) radars for the 4 February 2005 thunderstorms at(a) 17:59:47 UTC, and(b) 19:00:14 UTC. Selected thunderstorms are marked (tropical: yellow circle anvil 1a, white circle anvil 5a and red circle anvil 2b). The change in loca- tion of each of these circles between (a) and (b) indicates the storm motion, which can be compared with the storm motion indicated by arrows in Fig. 4a. The Bauru and Presidente Prudente radar sites are indicated with a yellow and white *, respectively. The left and upper axes give the distance in km, and the latitude and longitude are indicated in (a). (TITAN Software, installed at IPMet in collab- oration with NCAR.) velocity values mainly varied between 0.1 and 1.0 m s−1 in- dicating that the measurements were carried out outside the core region of the thunderstorm cell, where far higher verti- cal velocities are to be expected. The highest NOy mixing ratios were measured during the anvil penetrations with the strongest vertical velocities, which is closer to the core re- gion (∼10–30 km) where most lightning occurs (penetration I, III and V). The closest penetration to the maximum anvil outflow level was penetration III, where the mean updraft ve- locity (0.8 m s−1) was distinctly higher than the mean down- draft velocity (0.2 m s−1), and the highest mean NOy mixing ratio (1.1 nmol mol−1) was measured. About 30 km farther downwind (penetration II, IV, VI), the measurements indi- cate that a large part of the outflow already mixed with the ambient air (similar mean updraft and downdraft velocities). Moreover, on 4 February the selected thunderstorms were penetrated only once, but rather close to the core. Hence, too few repeated anvil penetrations and limited radar reflec- tivity data are available to apply the method introduced by TROCCINOX - F#11 180205b CO /nmol mol-1 40 60 80 100 120 pr es su re a lti tu de /m 0 2000 4000 6000 8000 10000 12000 O3 /nmol mol-1 10 20 30 40 50 60 70 80 90 NOx /nmol mol-1 0.0 0.1 0.2 0.3 0.4 CO O3 NOx (a) TROCCINOX - F#11 F180205b CO /nmol mol-1 40 60 80 100 120 140 N O x / nm ol m ol -1 0.01 0.1 1 180205b all TROCCINOX Falcon-flights <2 km altitude anvil Amazon basin Pacific background (b) Fig. 10. (a)Vertical profiles for CO, O3, and NOx mixing ratios from the Falcon flight on 18 February 2005. The red box at 2 km altitude indicates the top of the mixed layer.(b) Correlation plot for NOx and CO for the same flight (black dots), and superimposed data from all TROCCINOX Falcon-flights in the boundary layer (<2 km) (red-yellow dots). Ridley et al. (2004). We therefore use a modified version of the method introduced by Chameides et al. (1987) and as- sume that the measurements during each anvil penetration (snapshots) are representative average anvil conditions. We consider the horizontal mass flux of LNOx through a ver- tical control surface. The vertical surface dimensions can Atmos. Chem. Phys., 8, 921–953, 2008 www.atmos-chem-phys.net/8/921/2008/ H. Huntrieser et al.: Lightning activity in Brazilian thunderstorms 935 be estimated from the combination of e.g. airborne measure- ments and FLEXPART simulations as explained before in Sect. 4.2 and as indicated in Fig. 3. Repeated penetrations of the 18 February thunderstorm system indicated that1z was >1.3 km (10.7–9.4 km, Table 2a). The entire vertical extent of the anvil outflow can be most clearly seen in ver- tical profiles of the CO mixing ratio measured by the high- flying Geophysica (personal communication, P. Mazzinghi, INOA/CNR). On 18 February the most distinct enhancement in the CO mixing ratio was observed between∼9–12 km altitudes and1z was set to∼3 km (see Table 2a). On 4 February the enhancement in CO was less clear owing to elevated background mixing ratios: enhanced mixing ratios were mainly observed between∼10–14 km altitudes and1z was set to∼4 km (see Table 2a). The horizontal LNOx mass fluxFLNOx (in nitrogen mass per time, g s−1) was calculated for each thunderstorm pene- tration listed in Table 2a according to: FLNOx = χLNOx · MN Mair · ρa(Va − Vs) · 1x · 1z (4) whereχLNOx is the mean NOx volume mixing ratio produced by lightning (mol mol−1), MN andMair are the molar masses of nitrogen (14 g mole−1) and air (29 g mole−1), respectively, ρa is the air density (g m−3) calculated from measured tem- perature and pressure in the anvil, andVa − Vs is the differ- ence between the wind vectors in the anvil outflow and at the steering level (see Table 2b). The last term1x·1z is the area (m2) of the vertical cross-section perpendicular to the wind direction in the anvil outflow. In general, the wind at the steering level (∼700 hPa) determines the mean motion of a thunderstorm cell (e.g. Keenan and Carbone, 1992), but this parameter is not available from the airborne measurements. Instead, horizontal LINET stroke distributions, as shown in Fig. 4, were plotted with a higher temporal resolution (10 min) and the storm motion (Vs) was determined from the temporal stroke evolution. The parameters in Eq. 4, except 1x (Sect. 4.2) and1z, were calculated directly from Falcon measurements by averaging the measured data over the time period when the thundercloud was penetrated (between entry and exit of anvil), see Table 2a. FLNOx values were calculated for the selected thunder- storms by insertion of the parameters listed in Table 2a into Eq. 4, which give nitrogen mass flux values between 48 and 178 g s−1 (Table 2a). The flux values for subtropical thun- derstorms (only anvil I, II, and V considered) and tropical thunderstorms are within a similar range. The flux values in Table 2a can be divided by the molar mass for nitrogen and the area of the vertical cross-section (1x·1z) to estimate the flux in the unit mol m−2 s−1. The range of these fluxes, 3.3– 7.1×10−8 mol m−2 s−1, is well comparable to nitrogen mass flux values simulated by Barth et al. (2007) who ran different cloud-scale models (range 2.7–13.0×10−8 mol m−2 s−1) and to Barthe et al. (2007), who simulated 6×10−8 mol m−2 s−1 on average in the anvil outflow of a STERAO storm. elapsed UTC time in seconds since midnight 73000 74000 75000 76000 77000 78000 ve rti ca l v el oc ity /m s -1 0.01 0.1 1 N O y m ix in g ra tio /n m ol m ol -1 0.01 0.1 1 |wpos| |wneg| NOy |wpos| mean |wneg| mean TROCCINOX - F#11 180205b I II III IV V VI Fig. 11. Time series of NOy mixing ratio and absolute vertical ve- locity for the Falcon flight on 18 February 2005. The anvil penetra- tions are labelled I–VI. The parameters listed in Table 2a have large uncertainties. The relative maximal error of theFLNOx estimate was there- fore calculated. The uncertainty forχLNOx is given by the standard deviation (on average∼50% of the mean value); for Va−Vs the standard deviations listed in Table 2b indicate an uncertainty of up to∼50%; for 1x the uncertainty was ∼5–10 km corresponding to∼40%; and for1z the vertical anvil extension on 4 February varied between 3.5–6 km and on 18 February between 2–4 km indicating an uncertainty up to ∼50%. Summing up these uncertainties, the relative max- imal error of theFLNOx estimate is∼190%. 4.5 Estimate of the LNOx production rate per stroke and per year For the estimate of the LNOx production ratePLNOx (nitro- gen mass per stroke, in g stroke−1), the horizontal LNOx mass fluxFLNOx (g s−1) is divided by the LINET stroke rate RLINET (strokes s−1): PLNOx = FLNOx RLINET (5) PLNOx estimates for the selected anvil penetrations resulted in values between 1.9 and 5.6 kg stroke−1, see Table 2a. Un- fortunately, the dataset in Table 2a is very sparse. Never- theless, meanPLNOx values for three tropical and one sub- tropical thunderstorms (only anvil penetrations I, III and V considered) are estimated to 2.4 and 4.5 kg stroke−1, re- spectively, which corresponds to 4.8×1025 and 9.0×1025 molecules NO stroke−1. These results suggest that a sub- tropical thunderstorm may produce more LNOx per LINET stroke than a tropical thunderstorm (factor∼2). Possible rea- sons for this difference will be discussed in Sects. 5 and 6. www.atmos-chem-phys.net/8/921/2008/ Atmos. Chem. Phys., 8, 921–953, 2008 936 H. Huntrieser et al.: Lightning activity in Brazilian thunderstorms LIS flash all LINET strokes Bauru Radar LINET 10 kA� Fig. 12. Vertical maximum of the radar reflectivity (max CAPPI frame, unit dBZ) on 4 February 2005 at 21:30 UTC measured by the Bauru (22.4◦ S, 49.0◦ W) and Presidente Prudente (22.1◦ S, 51.4◦ W) radars. Superimposed are the horizontal distributions of LIS flashes (in red) and LINET strokes (black and yellow) for the time period 21:23:45 UTC–21:25:21 UTC. For comparison with other published results, thePLNOx estimates per LINET stroke were scaled toPLNOx estimates per LIS flash. During the TROCCINOX field period from 21 January to 27 February 2005, only one overpass of 4 Febru- ary at 21:23:45–21:25:21 UTC provided a sufficient large set of coincident LINET and LIS measurements. On this day, lightning activity in the LINET centre area (21.5–22.5◦ S and 48.5–49.5◦ W) and close-by (covering totally 20.0–23.0◦ S and 48.5–50.5◦ W) was suitable for comparison. Overall 82 LIS flashes and 481 LINET strokes were registered in the studied area during the∼90 s measurement. LINET strokes with peak currents down to at least 4 kA (absolute value) were sensed by LIS. For the selected time period, Fig. 12 shows the horizontal distributions of all available LINET strokes (black dots) and LIS flashes (red dots) for the area where most lightning occurred (21.4–22.4◦ S and 48.5–50.0◦ W) together with radar reflectivity (grey). For thePLNOx estimate, only stronger LINET strokes with peak currents≥10 kA are considered (in total 41 strokes, yellow dots) as mentioned before in Sect. 4.1. The LIS detection efficiency at night (0.93) was taken into account (21:23– 21:25 UTC = 19:23–19:25 Brazilian Summer Time). This implies a LINET/LIS ratio of about (41/82)×0.93=0.5, con- sidering only LINET strokes with peak currents≥10 kA. By means of this ratio, the mean values forPLNOx per LIS flash for tropical and subtropical thunderstorms (only anvil penetrations I, III, and V considered) are 1.2 and 2.2 (range 0.9–2.8) kg, respectively, corresponding to 2.4 and 4.5 (range 1.9–5.6)×1025 molecules NO. These estimates for TROC- CINOX are well within the range of more recent estimates. From a review of previous investigations, SH07 derive a best- estimate of 3.5 (range 0.5–10) kg of nitrogen per flash. The estimates forPLNOx per LIS flash were multi- plied with the number of LIS flashes occurring globally, 44 flashes s−1. If the selected tropical and subtropical TROC- CINOX thunderstorms were representative for the globe, the implied mean annual global LNOx production rateGLNOx would be∼1.6 and 3.1 Tg a−1, respectively (factor∼2 differ- ence). These values are close to previous best estimates for mid-latitude thunderstorms, see introduction. The individual estimates for the single thunderstorm penetrations listed in Table 2a, however, range from 1.3 to 3.9 Tg a−1, indicating a wide range of values and large uncertainties depending on where (horizontally and vertically) the anvil was penetrated. Finally, the relative maximal errors of thePLNOx and GLNOx estimates (Table 2a) were calculated. The uncertainty for RLINET was estimated from the standard deviations of the time series of the LINET stroke rates (Fig. 5). The standard deviations varied between 50–90% of the mean values. From the estimates forFLNOx ∼190% andRLINET ∼90%, the rel- ative maximal error of thePLNOx estimate for LINET strokes was∼280%. For thePLNOx estimate for LIS flashes, it was assumed that the uncertainty in the conversion of LINET strokes (≥10 kA) to LIS flashes was∼30% (depending on which LIS detection efficiency was used: day or night). This gives a relative maximal error of∼310%. For theGLNOx es- timate, the uncertainty in the global LIS flash rate was given with ∼10%, which gives a final relative maximal error of Atmos. Chem. Phys., 8, 921–953, 2008 www.atmos-chem-phys.net/8/921/2008/ H. Huntrieser et al.: Lightning activity in Brazilian thunderstorms 937 040205_anvil 1a 0 20 40 60 80 nu m be r o f L IN E T st ro ke s (k A )-1 to ta l N O p ro d. /k gN (k A )-1 1 10 100 1000 0 10 20 30 40 50 60 number of strokes total LNOx mass LNOx mass per stroke (a) CLINET = 30 km 040205_anvil 5a 0 20 40 60 80 nu m be r o f L IN E T st ro ke s (k A )-1 to ta l N O p ro d. /k gN (k A )-1 1 10 100 1000 0 10 20 30 40 50 60 (b) CLINET = 25 km 040205_anvil 2b peak current /kA 0 20 40 60 80 nu m be r o f L IN E T st ro ke s (k A )-1 to ta l N O p ro d. /k gN (k A )-1 1 10 100 1000 0 10 20 30 40 50 60 (c) 180205b_anvil III 0 20 40 60 80 1 10 100 1000 N O p ro d. /k gN s tro ke -1 0 10 20 30 40 50 60 (d) 040205_LINET centre (00-24 UTC) 0 20 40 60 80 100 1 10 100 1000 10000 N O p ro d. /k gN s tro ke -1 0 20 40 60 80 180205b_anvil I-VI (14-21 UTC) peak current /kA 0 20 40 60 80 100 1 10 100 1000 10000 N O p ro d. /k gN s tro ke -1 0 20 40 60 80 (e) (f) CLINET = 41 km CLINET = 126 km CLINET = 44 km CLINET = 160 km Fig. 13. Frequency distribution of LINET strokes (vertical grey bars) as a function of peak current for the selected tropical and subtropical thunderstorm systems on 4 and 18 February 2005, respectively (see Table 3). Superimposed is the laboratory result by Wang et al. (1998) modified for LINET strokes (blue dashed line) according to Table 3 (CLINET ) and the estimated total amount of LNOx mass produced per 1 kA LINET stroke interval for the selected thunderstorm systems (red line), see mass estimates in Table 3. ∼320%. Given this relative maximum error, the final range for the GLNOx values listed in Table 2a is between 0.4 and 12 Tg a−1. This range is comparable to other ranges given for GLNOx in previous publications (see Sect. 1 and SH07). 5 Possible explanations for different LNOx production rates in tropical, subtropical and mid-latitude thun- derstorms The results in the previous section lead us to hypothesise that tropical thunderstorms over Brazil may produce less LNOx per stroke than subtropical thunderstorms. In this section we investigate whether these differences in the LNOx production rate may be related to differences in the stroke peak currents (Sect. 5.1), stroke lengths (Sect. 5.1) or stroke release heights (Sect. 5.2) (relationships investigated by Wang et al., 1998, in the laboratory). Furthermore, mean stroke peak currents observed by LINET are compared for several tropical and mid-latitude thunderstorms and for one subtropical thunder- storm (Sects. 5.3–5.4). 5.1 LNOx production rate as a function of stroke peak cur- rent In this subsection we combine the result of laboratory mea- surements by Wang et al. (1998) with our field measurements to determine the LNOx production rate as a function of peak current. Wang et al. (1998) determined the NO production rate per unit laboratory spark, whereas our analysis provides the NO production rate per LINET stroke in the field. We as- sume that both follow the same dependency on peak current www.atmos-chem-phys.net/8/921/2008/ Atmos. Chem. Phys., 8, 921–953, 2008 938 H. Huntrieser et al.: Lightning activity in Brazilian thunderstorms Table 3. Lightning-produced NO mass per LINET strokePLNOx, for tropical and subtropical thunderstorms considering different stroke peak currents. Flight and Anvil Registered/ Duration, Number RLINET , F(LNOx)3, Integral4, CLINET 5, m Total PLNOx 7, S,g(N) Pressure, Penetration/ Considered of Stroke of LINET (LINET g(N) s−1 strokes s−1 laboratory Nitrogen g(N) (kA)−1 hPa tropical (t) or Stroke Peak Activity2, Strokes2 strokes) ×10−3g m−1 spark Mass (LINET (×103 subtropical (s) Currents1, min s−1 laboratory stroke−1 Produced6, stroke)−1 laboratory kA spark ×103g spark)−1 0402051a (t) ≥ 2 85 1287 0.252 120 4.04 29732 613 476 2.0 240 0402055a (t) ≥2 40 400 0.167 113 4.52 24 946 271 677 2.3 235 0402052b (t) ≥2 85 1492 0.293 178 4.31 41 239 906 607 1.8 260 180205bIII (s) ≥ 6 85 236 0.046 143 1.13 126 393 731 3097 1.9 260 0402051a (t) ≥ 10 85 278 0.055 120 2.71 44 390 613 2205 2.8 240 0402055a (t) ≥ 10 40 130 0.054 113 3.72 30 285 271 2082 2.6 235 0402052b (t) ≥ 10 85 311 0.061 178 2.48 71 516 906 2914 2.1 260 180205bIII (s) ≥ 10 85 130 0.025 143 0.89 160 324 731 5623 2.1 260 1 On 4 February 2005 strokes with peak currents down to 2 kA were registered (mainly within the LINET centre). On 18 February 2005 only strokes with peak currents≥6 kA were registered (mainly along the LINET periphery). Thus, on 18 February the detection efficiency for low peak currents was lower than on 4 February and the stroke characteristics are not comparable (upper half of the table). For a more equivalent comparison between the 4 and 18 February only strokes with peak currents≥10 kA were considered (lower half of the table). 2 On 4 February 2005 the registered/considered strokes until penetration were active from 16:55 to 18:20 UTC within anvil 1a (85 min.), from 17:55 to 19:20 UTC within anvil 2b (85 min) and from 18:05 to 18:45 UTC within anvil 5a (40 min) On 18 February 2005 the regis- tered/considered strokes until storm decay were active between 19:30 and 20:55 UTC (85 min). 3 The horizontal LNOx mass flux out of the anvil (see Eq. 4). Values from Table 2a are given. 4 For every 1 kA, the total number of strokesNLINET , within the duration of the stroke activity, are summed up and divided by the duration of the stroke activity which givesRLINET(I ) (see Eq. 9).RLINET(I ) is then multiplied with the Wang et al. 1998 laboratory relationship MNOlab(I ) (see below6 and Eq. 7) and summed up over all 1 kA intervals (part of Eq. 8 and Eq. 9): ∞∫ I=1 RLINET(I ) × MNOlab(I ) dI where RLINET(I )=1 t t∫ t=0 NLINET(I, t) dt 5 Conversion of the Wang et al. 1998 laboratory relationship to LINET strokes (see Eq. 8): CLINET = FLNOx/ ∞∫ I=1 RLINET(I ) × MNOlab(I ) dI 6 The total nitrogen mass produced by the thunderstorm within the duration of the stroke activity. For every 1 kA interval and the duration of the stroke activity,MNOlab(I ) (Eq. 7) modified for LINET (Eq. 10) is multiplied with the total number of strokesNLINET(I ) and summed up over all peak currents. 7 Nitrogen mass produced per considered stroke. as given by Wang et al. (1998). Hence, both differ only by a constant factor, which has the dimension of laboratory spark length per LINET stroke. This factor will be determined be- low. The relationship between the peak current and NO pro- duced per spark as found by Wang et al. (1998) from mea- surements in the laboratory (at 1.01×105 Pa) is given by: nNOlab(I ) = a + b × I + c × I2 (6) where nNOlab(I ) is the NO production normalised to 1 m spark length (1021 molecules NO m−1), a=0.14, b=0.026, and c=0.0025 andI is the peak current of the spark (kA). The number of NO molecules can be converted to the mass of nitrogen according to: MNOlab(I ) = MW × nNOlab(I ) (7) whereMNOlab(I ) is the nitrogen production per to 1 m spark length (10−3 g m−1) andMW is a constant (molecular weight of N, unit g molecule−1). According to Eq. (7) a laboratory spark with 10 kA would produce 0.015 g nitrogen m−1. Furthermore, Eq. (7) was multiplied with a constant fac- tor CLINET (m laboratory spark per LINET stroke) to con- vert the production per laboratory sparks and metre to the total number of LINET strokes. Here we assume that all LINET strokes in one specific anvil have the same length, independent of peak current and flash component, since no further information is available from our dataset. The factor CLINET was estimated from Eq. (8). Values from Table 3 for anvil 1a (lower half of table, here only strokes≥10 kA con- sidered) were inserted in Eq. (8); the mean LNOx mass flux, FLNOx, (120 g s−1), and the LINET stroke rate (strokes s−1), RLINET(I ), for a given peak currentI : FLNOx = CLINET ∞∫ I=10 RLINET(I ) × MNOlab(I ) dI (8) Atmos. Chem. Phys., 8, 921–953, 2008 www.atmos-chem-phys.net/8/921/2008/ H. Huntrieser et al.: Lightning activity in Brazilian thunderstorms 939 whereRLINET(I ) was estimated from Eq. (9): RLINET(I ) = 1 t t∫ t=0 NLINET(I, t) dt (9) andNLINET(I, t) is the number of LINET strokes for a given peak current and time. The value of RLINET in Eq. (9), integrated over all peak currents (here≥10 kA), is 0.055 strokes s−1 (Ta- ble 3 lower half, anvil 1a). The total integral in Eq. (8) (peak currents from anvil 1a inserted) is estimated to 2.71×10−3 strokes s−1 g m−1laboratory spark. The corre- sponding LINET factor,CLINET , was estimated to 44.4×103 (=120/(2.71×10−3)) m laboratory spark stroke−1 (see Ta- ble 3). We expect a LINET stroke in the atmosphere to be shorter than this calculated length (44 km), but probably broader than a laboratory spark. One metre LINET stroke is therefore probably more efficient in producing LNOx than a laboratory spark. Applied to LINET strokes, Eq. (7) changes to: MNOlinet(I ) = CLINET × MNOlab(I ) (10) where MNOlinet(I ) is the nitrogen mass production per LINET stroke (g stroke−1). A LINET stroke with a peak current of 10 kA (anvil 1a) would produce 0.7 kg nitrogen ac- cording to Eq. (10). This value is rather large because peak currents<10 kA were neglected (Table 3, lower half) and the total LNOx mass was distributed only over strokes≥10 kA in Eq. (8). The same calculations were performed for other selected thunderstorm penetrations of 4 and 18 February 2005, as shown in Figs. 13a–d and as listed in Table 3. In Table 3 ev- ery thunderstorm penetration is listed twice. For the first cal- culation (upper half in the table) all registered strokes were considered (peak currents down to 2 and 6 kA, depending on the detection efficiency in that area). For a comparison be- tween the 4 and 18 February selected penetrations, however, only strokes with peak currents≥10 kA were considered, as listed in the lower half of the table. In Fig. 13a–d the frequency distributions of LINET strokes (grey bars) per 1 kA peak current interval are shown for the selected thunderstorms of 4 and 18 February 2005. The stroke frequency rapidly decreases with increasing peak cur- rent. Superimposed are the laboratory results by Wang et al. (1998) concerning the NO dependency on peak current modified for LINET strokes (blue dashed line) according to Table 3 (differentCLINET values considered), and in addi- tion the estimated total amount of nitrogen mass produced per 1 kA LINET stroke interval for the selected thunderstorm systems (red line). The total mass estimates are listed in Ta- ble 3. In Fig. 13e–f the same type of calculations were performed for datasets with a larger number of LINET strokes to point out more clearly the differences between the stroke peak current frequency distributions of 4 and 18 February. All strokes (≥10 kA) in the LINET centre area on 4 February between 00:00 and 24:00 UTC were considered in Fig. 13e. In Fig. 13f the same calculations were performed for the se- lected thunderstorm system of 18 February for all strokes (≥10 kA) between 14:00 UTC and 21:00 UTC (see Fig. 4b). The integral over all peak currents gives a total nitrogen mass of 8.8×103 kg produced by 4359 tropical strokes on 4 Febru- ary (Fig. 13e), and a larger value of 11.2×103 kg produced by 2034 subtropical strokes on 18 February (Fig. 13f). This example also suggests that a subtropical stroke may produce a larger amount of nitrogen mass than a tropical stroke (here by a factor 2.7), mainly owing to differences in the stroke length (160 and 44 km, respectively). The higher production rate of LNOx by subtropical strokes was not caused by the stroke peak currents, since this frequency distribution was shifted to lower peak currents on 18 February (mean 31 kA, calculated from data in Fig. 2) compared with 4 February (mean 35 kA), as indicated in Fig. 13e–f. From the stroke frequency distributions of 4 February it was estimated that strokes with peak currents≥5 kA (only 30% of all strokes) produce the bulk amount (70%) of the total nitrogen mass. This result indicates that the numerous weak strokes with peak currents<5 kA are less important for the LNOx production. In Fig. 14 the same stroke frequency distribution separated, however, into IC and CG strokes, in- dicates that these weak strokes are mainly IC strokes. Fur- thermore, Fig. 13a–b indicates that there was a large fraction of these strokes with low peak currents in anvil 1a compared with anvil 5a. Yet, the high stroke rate in anvil 1a (0.252 s−1) produces a similar mean LNOx mass flux, FLNOx, value as in anvil 5a with a much lower stroke rate (0.167 s−1) (Ta- ble 3, upper half). (The calculations forFLNOx are based on similar penetration levels: 10.6 and 10.7 km.) Furthermore, the calculated stroke length was slightly shorter in anvil 5a (∼25 km) compared with anvil 1a (∼30 km), and the IC stroke release height (Table 4a) was slightly higher in anvil 5a (11.6 km) compared with anvil 1a (10.0 km). This result indicates that the lower stroke rate, shorter stroke length and higher stroke release height (see Sect. 5.2) in anvil 5a can- not explain the similarFLNOx values determined for anvil 1a and 5a. Only if the higher stroke peak currents in anvil 5a (mean 12 kA, Table 4a) compared with anvil 1a (mean 8 kA) are considered, these may give an explanation in this case. 5.2 LNOx production rate as a function of atmospheric pressure A further explanation for the different LNOx production rates of tropical and subtropical strokes in the selected Brazilian thunderstorms may be related to the release height of the strokes. Laboratory measurements by Wang et al. (1998) indicate that the LNOx production rate increases with increasing atmospheric pressure: nNOlab(p) = a + b × p (11) www.atmos-chem-phys.net/8/921/2008/ Atmos. Chem. Phys., 8, 921–953, 2008 940 H. Huntrieser et al.: Lightning activity in Brazilian thunderstorms Table 4a. LINET statistics of positive and negative CG and IC stroke fractions (here VLF pulses) and mean peak currents estimated for strokes with peak currents≥1 kA. Date Type of Thunderstorms/ Area1,2 Number Mean Peak CG IC Height Ratio Mean Peak Mean Peak Mean Peak Mean Peak Peak of Current, Strokes Strokes IC Positive/ Current Current Current Current Current Strokes3 kA (Fraction), (Fraction), Strokes, Negative (Fraction) (Fraction) (Fraction) (Fraction)≥10 kA % % km Strokes for CG-, for CG+, for IC-, for IC+, (Fraction), kA (%) kA (%) kA (%) kA (%) % Tropical (Brazil) 230105 LINET centre 11 324 6 <44 ≥56 10.3±2.9 1.0 −9 (23) +5 (21) −5 (28) +4 (28) 10 240105 LINET centre 419 6 37 63 10.5±3.3 1.1 −11 (21) +5 (16) −5(26) +4 (37) 11 250105 LINET centre 848 7 47 53 9.6±2.8 1.6 −11 (21) +5 (26) −8 (17) +5 (36) 15 040205 LINET centre 36 234 6 43 57 9.7±3.1 1.2 −10 (24) +5 (19) −5 (22) +4 (35) 12 040205 LINET centre: anvil1a* 1278 8 24 76 10.0±3.4 0.7 −15 (20) +5 (4) −8 (40) +5 (36) 22 040205 LINET centre: anvil5a** 439 12 88 12 11.6±3.2 0.7 −17 (55) +7 (33) −4 (5) +4 (7) 34 040205 outside LINET centre: anvil2b*** 1466 8 (55) (45) (14.0±2.5) 0.7 −11 (34) +6 (21) −7 (25) +6 (20) 21 250205 LINET centre 31 221 5 42 58 10.3±3.6 1.5 −7 (19) +5 (23) −5 (20) +4 (38) 8 Transition Trop.-Subtrop. (Brazil) 290105 LINET centre 419 6 40 60 8.6±3.0 1.3 −12 (25) +5 (15) −6 (19) +4 (41) 16 050205 LINET centre 1608 6 66 34 9.1±3.3 1.8 −9 (25) +6 (41) −5 (11) +5 (23) 13 190205 LINET centre 17 228 5 47 53 9.6±3.3 1.6 −9 (21) +5 (26) −5 (17) +4 (36) 9 Subtropical (Brazil) 180205 LINET periphery**** 3368 (13) – – – 0.5 - – – – (57) 180205 LINET periphery: anvil I-VI***** 236 (13) – – – 0.6 - – – – (55) Mid-latitude or Subtr. (Germany) 290605 LINET centre: isolated TS 4232 9 54 46 9.3±2.4 0.6 −11 (35) +6 (19) −11 (29) +5 (17) 25 040705 LINET centre 6337 5 72 28 10.8±3.3 0.8 −6 (39) +4 (33) −6 (16) +4 (12) 10 100705 LINET centre 15 174 6 65 35 9.6±3.3 0.9 −7 (36) +4 (29) −6 (17) +4 (18) 12 150705 LINET centre 8607 6 81 19 10.2±3.3 0.8 −8 (47) +4 (34) −5 (10) +3 (9) 15 290705 LINET centre: isolated TS 2254 9 76 24 9.1±3.4 0.5 −12 (52) +6 (24) −8 (14) +4 (10) 29 2907054 LINET centre: isolated TS, red. 1761 11 81 19 9.0±3.5 0.4 −13 (58) +7 (23) −11 (12) +6 (7) 37 1 The LINET centre area over Brazil covers 21.5–22.5◦ S and 48.5–49.5◦ W (area with highest detection efficiency). 2 The LINET centre area over Germany covers 48.5–49.5◦ N and 11.0–12.0◦ E (area with highest detection efficiency). 3 For statistical reasons only days with at least 400 strokes (≥1 kA) in the LINET centre area were considered and strokes that were defined as IC or CG strokes (undefined strokes were neglected). The numbers given are the total number of strokes registered between 00:00 and 24:00 UTC or for selected anvils. 4 Reduced dataset (sensor configuration similar to Brazilian configuration). ∗Anvil 1a (21.2–21.7◦ S and 48.9–49.2◦ W) is located at the edge of the LINET centre area and partly outside. ∗∗Anvil 5a (21.7–21.9◦ S and 48.4–48.7◦ W) is located mainly inside the LINET centre area and comparable to other estimates. ∗∗∗Anvil 2b (21.0–21.5◦ S and 49.7–50.1◦ W) is located just outside the LINET centre area where the fraction of IC strokes in general decreases, so estimates for this anvil penetration (especially IC height) are not directly comparable to the other estimates. ∗∗∗∗Subtropical thunderstorm system (19.4–20.0◦ S and 47.7–49.2◦ W, 14:00–21:00 UTC) is located along the LINET periphery (detection efficiency decreases) and therefore not well comparable to other estimates in this table (where peak currents≥1 kA are considered). ∗∗∗∗∗Anvil I-VI (19.3–19.8◦ S and 48.9–49.2◦ W, 19:30–21:00 UTC) is located along the LINET periphery (detection efficiency decreases) and therefore not well comparable to other estimates in this table (where peak currents≥1 kA are considered). wherenNOlab(p) is the NO production normalised to 1 m spark length (1021 molecules NO m−1), a=0.34, andb=1.30, andp is the pressure (105 Pa). A laboratory spark at 1000 hPa (ground level) would produce 0.038 gN m−1; at 500 hPa (300 hPa) about 0.023 (0.017) gN m−1 would be produced. The average height of IC strokes at mid-latitudes (Table 4a) is ∼10.0 km (270 hPa) and in the tropics∼10.5 km (250 hPa) (Table 4a). The calculated difference in LNOx produc- tion rate (factor 1.1) between these two altitudes (0.016 and 0.015 gN m−1, respectively) is only minor and cannot ex- plain the distinctly higher LNOx production rate of subtrop- ical strokes. Only if we make the unrealistic assumption that all subtropical strokes are CG strokes (mean release height ∼700 hPa) and all tropical strokes are IC strokes (mean re- lease height 250 hPa) can a factor of∼2 (=0.029/0.015) dif- ference be achieved. In the last subsection it was concluded that weak strokes with peak currents<5 kA are less important for the LNOx production. The majority of strokes with peak currents <5 kA are IC strokes according to the frequency distribu- tions of IC and CG strokes in Fig. 14. These IC strokes are released in the UT at low pressure. Taking this further rela- tionship into account (decreasing LNOx production rate with decreasing pressure), we find that the large number of very weak strokes with peak currents<5 kA only have a minor contribution to the LNOx budget. 5.3 Comparison of mean stroke peak currents in several tropical and one subtropical Brazilian thunderstorms The results in the previous subsections lead us to hypothe- sise that the different stroke lengths (calculated) may mainly contribute to the different LNOx production rates determined for several tropical and one subtropical Brazilian thunder- storms. The contribution from the different stroke peak current frequency distributions was found to be minor, but Atmos. Chem. Phys., 8, 921–953, 2008 www.atmos-chem-phys.net/8/921/2008/ H. Huntrieser et al.: Lightning activity in Brazilian thunderstorms 941 Table 4b. Same as Table 4a, but for strokes with peak currents≥10 kA. Date Type of Thunderstorms/ Area1,2 Number Mean Peak CG IC Height Ratio Mean Peak Mean Peak Mean Peak Mean Peak Peak of Current, Strokes Strokes IC Positive/ Current Current Current Current Current Strokes3 kA (Fraction), (Fraction), Strokes, Negative (Fraction) (Fraction) (Fraction) (Fraction)≥10 kA % % km Strokes for CG-, for CG+, for IC-, for IC+, (Fraction), kA (%) kA (%) kA (%) kA (%) % Tropical (Brazil) 230105 LINET centre 1144 19 <70 ≥30 10.8±2.9 0.3 −23 (57) +15 (13) −13 (18) +16 (12) 30 240105 LINET centre 47 19 77 23 10.0±3.7 0.3 −20 (66) +13 (11) −20 (11) +13 (12) 28 250105 LINET centre 130 21 61 39 9.3±2.3 0.6 −25 (42) +16 (19) −26 (19) +14 (20) 31 040205 LINET centre 4379 20 73 27 9.5±3.2 0.3 −22 (62) +19 (11) −17 (13) +16 (14) 34 040205 LINET centre: anvil1a* 129 18 65 35 10.2±2.7 0.2 −20 (62) +16 (3) −13 (24) +16 (11) 26 040205 LINET centre: anvil5a** 123 26 98 2 (8.5±0.8) 0.1 −27 (86) +19 (12) −14 (1) +14 (1) 50 040205 outside LINET centre: anvil2b*** 308 19 70 30 (14.2±2.4) 0.3 −22 (58) +14 (12) −16 (19) +16 (11) 28 250205 LINET centre 2406 18 65 35 11.0±4.0 0.4 −20 (50) +17 (15) −16 (20) +14 (15) 24 Transition Trop.-Subtrop. (Brazil) 290105 LINET centre 68 18 80 20 8.3±3.5 0.1 −20 (74) +13 (6) −13 (16) +12 (4) 31 050205 LINET centre 212 22 80 20 8.0±2.6 0.9 −23 (45) +24 (35) −16 (8) +17 (12) 33 190205 LINET centre 1499 18 77 23 10.2±3.9 0.4 −19 (62) +15 (15) −16 (12) +15 (11) 25 Subtropical (Brazil) 180205 LINET periphery**** 1914 17 – – – 0.4 – – – – 21 180205 LINET periphery: anvil I-VI***** 130 17 – – – 0.4 – – – – 22 Mid-latitude or Subtr. (Germany) 290605 LINET centre: isolated TS 1065 23 56 44 9.2±2.3 0.2 −24 (48) +24 (8) −24 (38) +16 (6) 44 040705 LINET centre 607 20 71 29 9.7±3.0 0.2 −19 (56) +26 (15) −18 (24) +23 (5) 30 100705 LINET centre 1845 20 71 29 9.3±3.4 0.2 −21 (60) +17 (11) −20 (23) +19 (6) 32 150705 LINET centre 1253 19 90 10 10.3±3.3 0.1 −20 (80) +19 (10) −19 (9) +16 (1) 31 290 705 LINET centre: isolated TS 659 21 87 13 7.4±3.4 0.1 −22 (76) +17 (11) −24 (11) +18 (2) 43 2907054 LINET centre: isolated TS, red. 659 21 87 13 7.4±3.4 0.1 −22 (76) +17 (11) −24 (11) +18 (2) 43 1 The LINET centre area over Brazil covers 21.5–22.5◦ S and 48.5–49.5◦ W (area with highest detection efficiency). 2 The LINET centre area over Germany covers 48.5–49.5◦ N and 11.0–12.0◦ E (area with highest detection efficiency). 3 For statistical reasons only days with at least 40 strokes (≥10 kA) in the LINET centre area were considered and strokes that were defined as IC or CG strokes (undefined strokes were neglected). The numbers given are the total number of strokes registered between 00:00 UTC and 24:00 UTC or for selected anvils. 4 Reduced dataset (sensor configuration similar to Brazilian configuration). ∗Anvil 1a (21.2–21.7◦ S and 48.9–49.2◦ W) is located at the edge of the LINET centre area and partly outside. ∗∗Anvil 5a (21.7–21.9◦ S and 48.4–48.7◦ W) is located mainly inside the LINET centre area and comparable to other estimates. ∗∗∗Anvil 2b (21.0–21.5◦ S and 49.7–50.1◦ W) is located just outside the LINET centre area where the fraction of IC strokes in general decreases, so estimates for this anvil penetration (especially IC height) are not directly comparable to the other estimates. ∗∗∗∗Subtropical thunderstorm system (19.4–20.0◦ S and 47.7–49.2◦ W, 14:00 UTC–21:00 UTC) is located along the LINET periphery (de- tection efficiency decreases) but can be compared with the other estimates in this table, since only peak currents≥10 kA are considered here. ∗∗∗∗∗Anvil I-VI (19.3–19.8◦ S and 48.9–49.2◦ W, 19:30–21:00 UTC) is located along the LINET periphery (detection efficiency decreases) but can be compared with the other estimates in this table, since only peak currents≥10 kA are considered here. maybe important to explain differences between single trop- ical thunderstorms as mentioned in Sect. 5.1. In this subsec- tion, values of the mean peak current (also separated for CG and IC strokes) are analysed in detail for a larger number of tropical Brazilian thunderstorms in the period with available LINET measurements (21 January–27 February 2005) to in- vestigate the differences between a number of tropical thun- derstorms and the subtropical thunderstorm of 18 February. For an equivalent comparison only strokes in the centre of the LINET detection network (from 21.5◦ S to 22.5◦ S and 48.5◦ W to 49.5◦ W, 00:00–24:00 UTC) were considered to avoid changes in detection efficiency and in the IC/CG VLF source ratio towards the border line. Selected days with a large number of LINET strokes suitable for statistical calcu- lations are listed in Table 4a (4b) for peak currents≥1 kA (≥10 kA). The selected days were classified according to HH07 into different categories: tropical, transition tropical-subtropical and subtropical cases, by use of the meteorological parame- ters listed in Table 4c (daily mean values) and as indicated in Fig. 15 (3 h values). In Table 4c the equivalent potential tem- perature (2e) at 850 and 500 hPa, and the wind velocity and direction at 200 hPa are listed for the selected LINET days in Table 4a–b. As suggested in HH07,2e in tropical air masses exceeded 345 K at 850 hPa and 332 K at 500 hPa and the UT wind velocity was in general low∼5–10 m s−1, influenced by the Bolivian High. The 4 February 2005 was selected as a case representative for tropical thunderstorm activity in general. In the selected region∼36 000 strokes were registered during the whole day (Table 4a). As expected for tropical thunderstorms, the frac- tion of IC strokes dominates over CG strokes and amounts to at least 57%. The≥ symbol indicates that the fraction of IC strokes may be even larger. The 3-D procedure applied to discriminate between IC and CG strokes categorises some strokes as uncertain. In most cases, this stroke is a CG stroke www.atmos-chem-phys.net/8/921/2008/ Atmos. Chem. Phys., 8, 921–953, 2008 942 H. Huntrieser et al.: Lightning activity in Brazilian thunderstorms Table 4c. Equivalent potential temperature (2e) and wind velocity and direction (calculated from ECMWF analyses) in the LINET centre area1 for selected days with LINET strokes as listed in Table 4a–b. Date Mean and Mean and Mean and Std Wind Std2 2e at Std2e at Wind Velocity at Direction at 850 hPa, K 500 hPa, K 200 hPa, m s−1 200 hPa Tropical (Brazil) 230105 349±2 341±1 11±3 SE 240105 349±2 341±1 8±3 SE–SW 250105 349±2 343±1 6±1 SW–NW 040205 344±3 340±1 5±2 SE–SW 250205 342±4 337±1 17±2 SW Transition Tropical-Subtropical (Brazil) 290105 344±2 342±2 16±1 W 050205 343±2 339±3 7±2 SW 190205 347±3 330±2 34±2 W Subtropical (Brazil) 180205 343±4 330±1 27±1 W Mid-latitude (Germany) 290605 331±3 326±2 21±2 NW–SW 040705 323±6 323±1 14±2 W–SW 100705 321±3 319±1 8±2 NE 150 705 323±3 320±1 20±7 NW Subtropical (Germany) 290705 340±5 325±2 19±3 SW 1 The LINET centre area over Brazil covers 21.5–22.5◦ S and 48.5–49.5◦ W, and over Germany 48.5–49.5◦ N and 11.0–12.0◦ E. 2 Std = Standard deviation. (and was here defined as CG), but for unfavourable positions of the measuring network stations it cannot be excluded that it is an IC stroke. This uncertainty may lead to significant biases in the results presented below in Table 4a–b. Note that we do not deal here with flash counts but with flash com- ponent (stroke) counts. In Table 4a the fraction of positive and negative IC and CG strokes for 4 February (all data) was determined to≥35% (IC+),≥22% (IC-),<24% (CG-) and <19% (CG+). The overall mean peak current (magnitude) was 6 kA. The mean peak currents for the different types of strokes (as mentioned above) were +4,−5, −10 and +5 kA, respectively. The ratio of positive to negative strokes was 1.2. The mean height of IC strokes was 9.7 km. Furthermore, the last column in Table 4a indicates that the fraction of peak currents≥10 kA was 12%. These results from the 4 February tropical thunderstorms can be compared with other tropical thunderstorms (see Ta- ble 4a–b) and the 18 February subtropical thunderstorm. In Table 4a a high mean peak current of 13 kA is given for the 18 February which, however, is not comparable to the rest of the data in Table 4a, since the storm was located along the northern periphery of the LINET network (detection ef- ficiency lower). For an equivalent comparison, only strokes ≥10 kA, as listed in Table 4b, were considered and an area along the northern periphery of the LINET detection network (19.4–20.0◦ S and 47.7–49.2◦ W, see Fig. 4b), where the sub- tropical thunderstorm of 18 February 2005 developed, was selected. The calculated mean peak currents for this area indicate a slightly lower mean peak value, 17 kA, for the subtropical thunderstorm of 18 February compared with the mean peak value for tropical thunderstorms of 4 February for the same area, 20 kA (same value as found for the LINET centre area listed in Table 4b, indicating that the detection efficiency for higher peak currents is about the same in the LINET centre and along the northern periphery, as also dis- cussed in Sect. 4.1). Overall, the mean peak currents in different tropical thun- derstorms of 4 February (and other tropical thunderstorms listed in Table 4b) were highly variable between 18 and 26 kA (probably depending on thunderstorm intensity). In the next section these values are compared with mean peak currents in mid-latitude thunderstorms over Germany to in- vestigate if any major differences exist. 5.4 Comparison of mean stroke peak currents in several tropical and mid-latitude thunderstorms The LINET network was also operated in southern Germany in summer 2005 (Sect. 2.2). LINET measurements cov- ered an area reaching from 47◦ N to 51◦ N and from 5◦ E to 14◦ E. 29 July 2005 was one of the days in summer 2005 with the highest lightning activity over Germany. In the LINET area ∼500 000 strokes were registered during the whole day. In Fig. 16a–b the cloud distribution over Europe on this day is shown together with the horizontal distribution Atmos. Chem. Phys., 8, 921–953, 2008 www.atmos-chem-phys.net/8/921/2008/ H. Huntrieser et al.: Lightning activity in Brazilian thunderstorms 943 040205_LINET centre (00-24 UTC) peak current /kA 0 20 40 60 80 100 nu m be r o f L IN ET s tro ke s (k A) -1 1 10 100 1000 10000 number of IC strokes number of CG strokes Fig. 14. Frequency distributions of LINET strokes as a function of peak current separated for IC (blue) and CG (green) strokes on 4 February 2005. of LINET strokes over southern Germany and the positions of the LINET sensors. For further estimates with LINET data, only data in the LINET centre region were considered (here 48.5◦ N to 49.5◦ N and 11◦ E to 12◦ E), as mentioned previously. Other days during the German field campaign in June and July 2005 with a high LINET stroke activity in this area were 29 June and 4, 10 and 15 July.