Analysis of the Campinas tornado (Brazil) in June 2016: damage track, radar characteristics and lightning observations of the supercell

Shortly after midnight on 05 June 2016, in the city of Campinas with >1.2 million inhabitants, located in the State of São Paulo, intense precipitation, including hail, a large number of electric discharges and very strong wind gusts, causing significant damage, were recorded. No fatalities were documented, probably due to the day and time 15 (Sunday around 00:20 Local Time). The affected areas are middle and even upper-middle class neighborhoods, with solid buildings, confirming the potency of the phenomenon. The destruction pattern indicates an intense perturbation resulting in the twisting of structures and tree branches, as well as large objects having been airborne over a distance of about 50 m, and large trees ripped from the ground, all suggesting that it was a tornado of category EF2-3. Severe damage was also reported from other towns in the region. About three hours before the tornado occurred in Campinas, 20 an even stronger event devastated part of the small town of Jarinu, 40 km southeast of Campinas, possibly an EF3 tornado, which caused one fatality and overturned two semi-trailer trucks. No alerts that a disturbance of this magnitude would impact the region were raised, demonstrating that Campinas, and probably most other Brazilian cities and towns, are not prepared for such an event. During the beginning of June 2016, the synoptic situation over Brazil was characterized by a strong anticyclone 25 centered over the northern half of South America at the 250 hPa level and bounded by a strong zonal Subtropical Jet (STJ) in the south, resulting in moist air being advected from the Amazon and Pacific region, creating favorable conditions for strong convection in the State of São Paulo, even during the night. Images from a Doppler S-band radar, located in Bauru, recorded a supercell storm lasting 8.5 hours, which traversed the eastern half of the State of São Paulo during the night of 04/05 June 2016 and spawned a tornado in the city of 30 Campinas during the early hours of the morning (Local Time). Despite the distance of >200 km, these radar observations confirmed typical tornado signatures, such as a rotational damage pattern, a hook echo and a mesocyclone with a rotational velocity of 12.5 m s-1. The supercell was accompanied by intense lightning activity throughout its life cycle with a “lightning jump” from 0 to 55 ground strokes per minute within 12 min just prior to the tornado touchdown, culminating in a frequency of 238 strokes per minute of Total Lightning. Although some of the Severe Storm 35 https://doi.org/10.5194/wcd-2021-35 Preprint. Discussion started: 10 June 2021 c © Author(s) 2021. CC BY 4.0 License.

Parameters calculated by the TITAN (Thunderstorm Identification, Tracking, Analysis and Nowcasting) Software were slightly lower than found in previous tornado cases in the State of São Paulo, this is most likely due to the fact that this was the first occurrence of a tornado observed by radar during the dry austral winter season in this region of Brazil, as well as a nocturnal event.

40
Tornadoes are the most intense vortices in the atmosphere over land, formed in environments with intense wind shear.
They are associated with conditions of great thermodynamic instability, reinforced by specific surface parameters, such as relief, vegetation, urbanization and the presence of water bodies. There are several scales to classify tornadoes by damage, such as F (Fujita, 1981), later officially modified to EF (Enhanced Fujita; National Weather Service, 2020), Torro, developed in England by H. E. Brooks (TORRO, 2020), and EBRAV, emerging in Brazil (Candido, 2012). All 45 these scales are based on the degree of destruction brought about by tornadoes, but since the characteristics of buildings and building materials are different across countries (for example, in the USA motor homes are commonly used, while in the plains of North America timber buildings prevail, less common in much of Brazil), the comparison among nations is challenging.
Tornadic cells can develop in atmospheric environments that produce severe weather conditions such as transient frontal 50 systems, mesoscale convective complexes, cyclones, local convective systems and supercell storms. They are not associated with a single weather condition and thus, occur in different climates and during different periods of the year.
However, the most severe tornadoes are associated with supercells, which have an extremely intense upward rotating current, reaching speeds of 50 m s-1 or more (Houze, 1993).
The rising number of detected tornadoes could be associated with the increase and spread of the world's population, as 55 well as greater awareness of such phenomena (Doswell et al., 1999), because the rapid dissemination of news about severe events is increasingly facilitated. No association between the frequency of tornadoes and global warming has so far been confirmed and requires further studies. Rosenfeld et al. (2008) claim that the higher concentration of aerosols tends to intensify convection in the atmosphere, generating deeper clouds and thus, more suitable conditions to produce severe events, such as tornadoes. This process is related to reducing the average size of water droplets within the cloud; 60 being lighter, the droplets tend to reach higher parts of the cloud, where they freeze and, in the process, release more latent heat, boosting convection at medium and high levels. Due to the higher concentration of anthropogenic pollutants, large urban and industrial centers would be conducive environments for this type of intensification and, therefore, would contribute to the formation of more severe storms. This was also confirmed indirectly by Naccarato et al. (2003), who found significantly increased lightning flash densities over large cities in the State of São Paulo. more common in the fall and spring, which are seasonal transition periods with more atmospheric instability. Other authors have studied the occurrence of tornadoes in different parts of Brazil with radar information (Bertoni, 2013;Vilar, 2016) while some others were developed without radar information (Marcelino, 2003;Marcelino et al., 2003;Nunes et al., 2008;Insee et al., 2016).
According to historic media coverage, Campinas and the surrounding region experienced devastating tornadoes in 1995 75 and 2001, but these events had not been scientifically evaluated. At the end of November 1995, a tornado with winds of about 180 km h-1 hit the campus of the State University of Campinas (UNICAMP), destroyed the 250 ton. roof of the sports gym, broke windows of several buildings and uprooted 12 trees on the campus, as well as caused considerable damage in other parts of the city (Folha de São Paulo, 1995). During late afternoon of 4 May 2001, a tornado with winds of up to 300 km h-1 hit nine towns in the Campinas region, caused general destruction, and left several persons 80 injured, as well as at least one fatality (Folha de São Paulo, 2001).
The first tornadic cell fully documented by radar in Brazil occurred in 1994 within the range of the Bauru S-band Doppler radar (Gomes et al., 2000). Several more events could be tracked in the State of São Paulo from 2004 onwards, when already lightning observations were available, with the objective to identify characteristic tornado signatures in the radar images and lightning data which could be used for Nowcasting (Held et al., , 2006a(Held et al., , b, c, 2009(Held et al., , 2010a(Held et al., , b, 85 2011Gomes et al., 2009;Nascimento et al., 2014). Some of the studies also attempted to investigate numerical forecast models to identify parameters which could indicate the possible occurrence of severe storms likely to generate a tornado up to three hours in advance (Held et al., , 2006b(Held et al., , 2009Nascimento et al., 2014).
Since 1994, several events of supercell storms traversing the State of São Paulo while spawning tornadoes, although relatively rare, had been observed and tracked by the Doppler S-band radars located in Bauru and Presidente Prudente, 90 respectively, (Held et al., 2010a). The majority of these storms, which were analyzed in detail, had spawned severe tornadoes and occurred during the atmospheric transition periods (May and September). The timing of the transition periods varies slightly from year to year, depending on the large-scale circulation in the troposphere. However, mostly less severe occurrences, with several ones only generating funnels that did not reach the ground, were also observed during the subtropical "Rain Season" (October -March), e.g., near Lins (a small town in the central State of São Paulo) 95 on 25 December 2008 (Gomes et al., 2009). Mean monthly rainfall totals in the central interior of the State of São Paulo remain well below 50 mm from June to August (INMET, 2019;IPMet, 2019;Held and Nachtigall, 2002), thus this period can be classified as the dry austral winter period. Severe storms associated with strong winds are quite uncommon during this period, although severe hailstorms, mostly associated with baroclinic events, had been recorded occasionally in the interior of the state (Gomes and Held, 2009).

100
This study presents the analysis of the life cycle of a supercell which spawned a tornado at around 00:20 Local Time (LT) on 05 June 2016, causing severe damage in parts of Campinas, a Brazilian city in the State of São Paulo that houses more than 1 million inhabitants. The day (Sunday) and time of the tornado occurrence probably contributed to the fact that only a small number of persons suffered injuries, without any fatalities having been reported. However, based on the damage pattern, it could be classified as EF2-3 on the Enhanced Fujita scale, which since 2007 replaced the original Fujita Scale (Fujita, 1981). Both scales classify tornadoes from 0 to 5 based on the damage observed after the passage of a tornado, but the EF scale incorporates more damage details (National Weather Service, 2020). Nearby municipalities, located within a 50 km radius from Campinas (marked in Fig. 6), also recorded heavy rains and strong winds during the night, e.g., Morungaba (28 km east of Campinas; in the extended path of the tornadic supercell), but others were affected by different storm systems, such as Itupeva, Atibaia and Jarinu. In the latter town an earlier 110 supercell peaked at around 21:50 LT on the previous day and based on damage and eyewitness reports also spawned a tornado estimated to have been an EF3. However, being just outside the Bauru 240 km quantitative radar range, only a qualitative description of observations is included.

Data and methods
The analysis of this event is primarily based on a field survey of the affected areas, performed during the days 115 immediately following the occurrence of the tornado, in order to determine damage patterns in loco, on the interpretation of data from the Doppler S-band radar in Bauru (BRU; 22.36º S, 49.03º W) operated by the Centro de Meteorologia de Bauru (CMB) of the Universidade Estadual Paulista (UNESP), as well as on lightning stroke data obtained with the Brazilian Lightning Detection Network (Rede Brasileira de Detecção de Descargas Atmosféricas BrasilDAT) operated by the Divisão de Impactos, Adaptações e Vulnerabilidades (DIIAV) of the Coordenação-Geral 120 de Ciências da Terra (CGCT) of the Instituto Nacional de Pesquisas Espaciais (INPE). Additional radar data from the Doppler S-band radar in Presidente Prudente (PPR), also operated by the CMB, as well as from the S-band radar at São Roque (SRO), operated by the Departamento de Controle do Espaço Aéreo (DECEA), are supplementing the study of convective activity for this case. It must be pointed out that the SRO is an operational radar, primarily deployed for Air Traffic Control, and thus not very suited for research, resulting in unreliable data quality (many VOL-scans are 125 incomplete, due to missing higher elevations, as well as individual rays sometimes missing). Furthermore, no radial velocities were available for this case. Also, a difference of 10-15 dBZ between BRU and SRO was noted, and therefore the authors decided to only use these data in a qualitative manner.
Volume scans of the original radar data (BRU and PPR), recorded and archived in the propriety format Interactive Radar Information System (IRIS) of SIGMET Inc. (SIGMET, 2005), were converted to the MDV format 130 (Meteorological Data Volume), and subsequently processed with the software "TITAN" (Thunderstorm Identification, Tracking, Analysis and Nowcasting; Dixon and Wiener, 1993), available from NCAR (National Center for Atmospheric Research, Boulder, Colorado, USA). The CMB radars have a 2° beam width and each volume scan with 16 elevations is completed in 7.5 min within the 240 km quantitative radar range. The range extends to 450 km in surveillance mode (PPI only at 0.3° elevation). The SRO radar also has a 2° beam width, but each volume scan with 15 elevations is 135 initiated every 10 minutes. The conversion of the raw data (HDF5 format) to the MDV format was performed up to 250 km. It is important to point out that TITAN not only identifies potentially severe cells, based on a pre-defined reflectivity threshold and volume, but also tracks them along their life cycle, and additionally provides forecasts of their future positions up to 60 min. The latter facility assists the meteorologist with the issuing of short-term severe storm warnings (nowcasting).

150
Operational Doppler radars do not record the occurrence of a tornado directly, due to its relatively small horizontal diameter of generally <500 m. However, good indications are being provided by certain signatures in the radial velocity field, such as regions where opposing radial velocities along the azimuth are observed, which would identify a rotating circulation associated with a mesocyclone. A tornado will be spawned when the cloud funnel extends to the ground, where it will create a characteristic damage pattern, which in turn will reveal the magnitude of the phenomenon. The 155 analysis of the severe storm studied in this paper confirmed the presence of cyclonic rotation (velocity couplet) in the radar echo core, as well as in the surveyed damage track, which would identify it as a tornadic cell. However, due to the nocturnal occurrence shortly after midnight (LT), no funnel was reported.

160
Campinas is the third largest town in the State of São Paulo (Fig. 1) and is among the 15 cities with the largest population in Brazil, comprising >1.2 million inhabitants, with a demographic density of 1,527.6 inhabitants per km 2 (IBGE, 2020).
Campinas and neighboring towns have a diverse and dynamic economy. During the last decades the region became one of the most sophisticated centers for research and development of science, technology and innovation in the country, 165 concentrating more than 50 branches of the 500 largest companies in the world and accounting for 10% of industrial production in Brazil in various sectors, such as automotive, textile, metallurgy, food, pharmaceutical, petrochemical, telecommunications and electronics (Cavalcanti et al., 2017). Campinas is one of the 15 National Brazilian Metropolis (IBGE, 2018), with 34 cities under its influence; in other words, Campinas is an attraction pole for health services, high education and technology.

195
According to the Civil Defense Department, 74 mm of rain were recorded in just 45 minutes -an extraordinary value since the monthly average of June for the period 1990-2019 is 42.8 mm, while in June 2016 the total accumulated rainfall recorded was an exceptional 180 mm (CEPAGRI, 2020).

Synoptic Situation
The synoptic situation during the first few days of June 2016 was rather typical for occasionally occurring severe storms    ranging from 50-70 km h -1 . Figure 4b shows the situation with several severe storm complexes at 00:00 LT, shortly before the tornadic cell impacted Campinas, coinciding with the GOES image at 03:00 UT (Fig. 3b). The severe convective cells (≥35 dBZ) correspond well with cloud tops colder than -60°C, as observed by GOES.  storms associated with supercells observed previously in the State of São Paulo (Held et al., 2010a). The cell which spawned a tornado over the Campinas region can certainly be classified as a supercell, based on several of its characteristics, such as velocity, echo tops penetrating the tropopause, as well as "severe storm parameters" generated by the TITAN analysis, which will be discussed later. It should be noted that the BRU radar has a 2° beam width (at 240 km range the aperture already has a diameter of 8 km) and the reflectivity is not range-corrected, resulting in a slight  In order to support the finding that the tornado had been spawned by a supercell, raw data (HDF5 format) from the S-

270
this supercell now moved at ±60 km h -1 towards east-northeast, it increased its VIL by >160% during 40 min until reaching a peak at 00:50 UT. This would imply a very strong updraught which was probably responsible for spawning an EF3 tornado in Jarinu between 00:50-01:10 UT, as reported by eyewitnesses and confirmed by a subsequent ground survey, revealing enormous damage in the town, including two overturned semi-truck trailers, as well as causing one fatality. Due to the lack of radial velocity data, no rotation inside the cell could be documented. Thereafter, this 275 supercell continued parallel to the one that had split off at 00:30 UT (Fig. 6)

310
The rotational velocity is defined as modulus of the radial velocities divided by 2, viz., (|Vin| + |Vout|)/2, which then permits the calculation of the relative rotational velocity of the tornadic cell affecting Campinas during this event. The radial velocities found in this analysis ranged between -25 m s -1 (towards the radar) and +15 m s -1 (away from the radar). Based on these radial velocities recorded by the Doppler radar in Bauru, a value of 12.5 m s -1 is derived for the rotational velocity Vr, where the centers of maximum radial velocity were separated by a distance of 5 km. It should be 315 pointed out that this velocity couplet was observed despite its radial distance of just over 200 km from the origin of the radar in Bauru. Considering the beam width of the Bauru radar, a significant degradation of the vortex at that distance should be expected (Brown and Wood, 1999). The rotational vorticity can be calculated from the ratio between the speed and the distance between the pair of opposing radial velocities, yielding a value of 2.5·10 -3 s -1 for this case. From studies of tornadic storms in the USA, threshold values of Vr ≥ 12.5 m s -1 within a radius of 150km, and ≥8.5 m s -1 for 320 distances further than 150km were defined (NSSL, 1985). Thus, the values associated with the Campinas cell can be considered to characterize an intense tornadic event. The harmful consequences associated with the event, such as the uprooting of many dozens of large trees, total or partial destruction of a large number of residences and impacts on the urban infrastructure also attest the severity of the phenomenon.  from SRO were around 15 dBZ higher and consequently yielding unrealistically high VIL values. Therefore, only track data from BRU will be presented in this study, despite the great radial distance of the tornadic cell during its most intense stage, resulting in an underestimation of especially maximum reflectivities and their heights (volumes) and subsequently VIL data, as explained above.
From eyewitness reports it is concluded that the tornado occurred during the 03:14-03:29 UT Volume Scans, with 340 maximum intensity at around 03:22 UT. As can be seen from Fig. 8, the maximum reflectivity and VIL reached a peak during the 02:59 UT Volume Scan, well before the tornado had spawned and reached the ground. Their observed values of 49.5 dBZ and 15.8 kg m -2 , respectively, are certainly an underestimate, but considerably lower than those recorded during the Indaiatuba EF3 tornado (about 25 km south-southwest of Campinas, also at a range of ≥200 km from BRU), viz., 57.5 dBZ and 40.5 kg m -2 , respectively (Held et al., 2010a;Nascimento et al., 2014). Although both cases must be 345 regarded as underestimates due to the large distance of >200 km, there may also be other reasons to be considered. The

365
It is noteworthy that the Campinas tornado was spawned directly from the mother cell, while all earlier case studies (Held et al., 2010a) showed the tornado-spawning cell to be an isolated cell trailing the storm complex.

Damage
A field recognition conducted along the storm path shortly after the occurrence of the event in Campinas revealed typical signatures of an EF2-3 tornado. It is noteworthy that the affected neighborhoods are middle to upper-middle There are no photographic records or videos of the probable tornado funnel available, most likely due to the nocturnal 380 occurrence of the storm, as well as the power outage, which made it impossible for surveillance cameras scattered around various parts of the city to provide any kind of registration. Residents of the affected areas reported a loud noise similar to an explosion or to an airplane turbine.
About three hours before the tornado in Campinas, severe wind damage occurred in Jarinu, a small town about 40 km southeast of Campinas. It can be attributed to an earlier supercell traversing the region (Fig. 6). An on-site survey and 385 informal interviews with residents and civil defense officials leave no doubt that a severe event took place in the town between 21:50-22:19 LT on 04 June. Although this area is outside the Bauru radar coverage, the evidence strongly suggests that it also was a tornadic cell, as can be seen in Fig. 10, with metal structures twisted around light poles, concrete poles twisted, large trees uprooted, as well as widespread damage to homes and commercial buildings. The local press reported public and private losses above R$ 18 million, with 275 homes damaged and 83 totally destroyed, 390 two overturned semi-truck trailers, as well as injured people and one fatality. Eyewitnesses also mentioned that at least three cars were thrown several meters away. Based on the massive destruction, the phenomenon could be rated as an EF3 tornado.

Lightning Observations
Lightning strokes recorded by BrasilDAT (Naccarato et al., 2012;Naccarato and Pinto, 2012)  The TITAN storm tracking software was run with reflectivity thresholds of 30 and 35 dBZ to generate ellipses for each 7.5 min volume scan. Figure 11a shows these ellipses for the 30 dBZ threshold from 23:00 UT (20:00 LT) until 04:00 UT (01:00 LT). Figure 11b highlights the vector track of the supercell (marked as "TRK 5/"), which finally spawned the tornado in Campinas, from its origin at 23:06 UT until it reached the 240 km quantitative range of the Bauru radar 410 (BRU) as TRK 5/73. These tracks are based on the center point of the ellipsoids around the 30 dBZ reflectivity contour (Fig. 11a). Since TITAN allows merging or splitting of cells, the sub-number of the complex (TRK 5/…) changes along its path, as can be seen in Fig. 11b. The ellipses from the tornadic cell "TRK 5/" were consolidated into a continuous polygon, which delineated the area for which stroke data were extracted for the period 23:00 UT until 04:00 UT.
Subsequently, they were separated into positive and negative cloud to ground (CG) strokes, as well as intra-cloud stokes 415 (IC) for further analysis.
A first glance at the lightning stroke data reveals an exceptionally large number of negative CG and IC strokes within the 30 dBZ polygon, viz., 2145 negative CG strokes and 8689 discharge events classified as IC, during the period 23:00-04:00 UT. It is also noteworthy, that 52.5% of the ground strokes recorded within the 30 dBZ contour were confined within the 35 dBZ envelope, while 58.1% of IC activity occurred within the 35 dBZ contour. Maximum Peak

420
Currents recorded for CG-neg and CG-pos strokes were -131 kA and +133 kA, respectively, with the most frequent intervals being 10-19 kA and 20-29 kA, respectively. However, only 0.8% of the CG-neg strokes were ≥50 kA. The  The time sequence of these lightning strokes per minute is presented in Fig. 12. Three distinct peaks can be seen, of which the first two are associated with rapid cell development and marked with x in Fig  The large number of strokes produced by the event (Fig. 12) has to be highlighted, particularly the lightning jump effect,

445
which is a very rapid intensification of the electrical activity inside the thundercloud (Goodman et al., 1988(Goodman et al., , 2005Williams, 1989;Gatlin and Goodman, 2010;Schultz et al., 2011). These authors presented studies in which storms demonstrated a sudden increase in lightning activity just before the occurrence of a severe event, such as hail, tornado, wind gusts, and downbursts or microbursts. This relation of the electrical activity with the severe weather occurrence can be explained by the relationship between the strong updrafts within the cloud and the electrification of the 450 hydrometeors. The updrafts together with the gravitational forces distribute the hydrometeors within the cloud according to their size, which are then quickly electrified by non-inductive electrification processes. The result is a fast formation of positive and negative charge centers within the cloud that leads to an enhancement of the lightning production (Deierling et al., 2008). In general, the intensification speed of the lightning jump phenomena (curve slope) is directly related to the increase of the updrafts (or convection invigoration), which produces much more ice within the 455 cloud, but with smaller size and density, leading to a higher number of collisions, higher charge transfers and consequently more intense and larger charge centers. CG-neg strokes per 7.5min during the 03:14-03:22 UT Radar Volume Scan (Fig. 8), includes the maximum of the lightning jump in Fig. 12 Fig. 12 and Fig. 13, 03:22 UT. At this stage, no definitive reasons for this behavior can be suggested, but it might be attributed to the fact that this tornadic storm was the first of its kind documented by radar during the dry winter season.

Conclusions
The synoptic situation during the first few days of June 2016 was rather typical for occasionally occurring severe storms were calculated throughout the lifetime of storms. Some of these values were found to be lower than observed from previous supercell and tornado cases, e.g., maximum reflectivity = 49.5 dBZ and VIL = 15.8 kg m -2 , respectively, which 515 are certainly an underestimate due to the long radial distance, but possibly also because this was the first tornadic event ever recorded by radar during the austral dry winter period and furthermore, during the night, when convective activity is less intense than during daytime. The peaks in radar reflectivity and VIL aloft occurred shortly before the onset of the most intense phase of the storm, accompanied by a heavy downpour of rain and hail, while the lightning activity dropped to a minimum, before a significant "lightning jump" was recorded. Other typical tornado signatures, such as a 520 rotational damage pattern of uprooted and broken trees, as well as a hook echo and a mesocyclone with a rotational velocity of 12.5 m s -1 , both observed in the radar data at the time of the assumed spawning of the tornado (Volume Scan 03:14-03:22 UT). A vertical cross-section through this Volume Scan clearly pictured a pair of opposite radial velocities ("couplet") at a distance of 5 km, which identifies the presence of a mesocyclone, indicative of the existence of rotation within the storm, which can spawn a tornado.

525
Lightning stroke data obtained with the Brazilian Lightning Detection Network (Rede Brasileira de Detecção de Descargas Atmosféricas Totais, BrasilDAT), were analyzed in order to characterize the electric properties of the supercell throughout its 8.5-hour lifetime. A first glance at the lightning stroke data reveals an exceptionally large number of negative CG (Cloud to Ground) and IC (Intra-Cloud) strokes within the 30 dBZ polygon, viz., 2145 negative CG strokes and 8689 discharge events classified as IC, during the period 23:00-04:00 UT (96% of total lightning 530 occurred during the period from 01:00-04:00 UT). Besides two earlier distinct peaks in lightning frequency, associated with rapid cell development, the third peak evidences a typical lightning jump, starting just after 03:00 UT and reaching a maximum of 238 strokes min -1 (total lightning) 12 minutes later, preceding the estimated touch-down time of the tornado (around 03:21 UT) in Campinas. This is also supported by the observed sharp drop of the lightning frequency at 03:25 UT (from 185 to 72 IC strokes min -1 ), followed by a very rapid increase again after about five minutes. During 535 earlier studies of supercells and tornadoes in the State of São Paulo, an absence of lightning strokes to ground (CG; the then available lightning data did not include IC activity) during the tornado touch down, was documented. In the current https://doi.org/10.5194/wcd-2021-35 Preprint. Discussion started: 10 June 2021 c Author(s) 2021. CC BY 4.0 License.
case it was also found that very few CG strokes were recorded shortly before the lightning jump. In fact, during the Volume Scan 02:59-03:07 UT only seven CG-neg strokes were recorded ahead of the tornadic echo core (≥40 dBZ), as revealed from CAPPIs with overlaid strokes, generated by TITAN/CIDD, but thereafter the frequency increased rapidly 540 until 03:14-03:22 UT with up to 55 CG strokes per minute striking the ground anywhere within the 30 dBZ reflectivity contour.
According to eyewitnesses, another tornado spawned in the same region from a different supercell some three hours before the one that impacted Campinas, traveling on a parallel track. Based on the damage pattern in the small town of Jarinu, it could be considered as an EF3 tornado.

545
Considering the relatively rare occurrence, small-scale features and short duration of tornadic cells in the central region of the State of São Paulo, it is almost impossible to predict a possible formation of tornadoes timely for issuing warnings to the population. Not even Nowcasts based on continuous radar monitoring would facilitate experienced radar meteorologists to emit a tornado alert, because supercells in this region are generally difficult to identify during their early stages.

550
The occurrence of the phenomenon described in this paper has clearly shown that the municipal government and citizens are relatively unprepared to face this kind of severe event, not only in the Campinas region, but also in other parts of Brazil where tornadoes occasionally occur. Furthermore, the currently available Doppler radars in Brazil are insufficient for a countrywide coverage to detect severe storms in real time for issuing timely warnings. The fact that Campinas was the first Brazilian city certified as a resilient city, crowning its efforts to fight floods, could give a false 555 impression of resilience to other phenomena that cause great impacts, such as tornadoes or devastating hailstorms, for which Campinas and other Brazilian cities are not prepared.
As there is no way to eliminate or even prevent damages caused by a tornado, measures need to be taken to minimize the impact and especially, to reduce the number of injuries and to prevent deaths. This would require an all-out effort to educate the population countrywide how to react in such an emergency at short notice, like evacuation of precarious 560 buildings, sheltering in a safe room, avoid being near trees and power lines, etc. It should be acknowledged that the local Civil Defense Authorities have already initiated such activities, but it will probably take a whole generation before having reached a major portion of the population in order to become routine.
However, a positive highlight was that post-event measures, such as removal of debris and fallen trees, road clearance and the re-establishment of energy were well organized, prompt and effective. This is probably due to the fact that 565 Campinas and even Jarinu routinely experiences flooding, so that there is a preparedness of authorities for post-disaster emergency action. The big challenge, however, is pre-disaster warning and the preparedness of people how to prevent further damage and injuries.  uol.com.br/fsp/1995uol.com.br/fsp/ /11/29/cotidiano/35.html, last access: 8 March 2021uol.com.br/fsp/ , 1995