Citizen Science Observation of a Gamma-Ray Glow Associated With the Initiation of a Lightning Flash

Gamma-ray glows are observational evidence of relativistic electron acceleration due to the electric field in thunderclouds. However, it is yet to be understood whether such relativistic electrons contribute to the initiation of lightning discharges. To tackle this question, we started the citizen science “Thundercloud Project,” where we map radiation measurements of glows from winter thunderclouds along Japan's sea coast area. We developed and deployed 58 compact gamma-ray monitors at the end of 2021. On 30 December 2021, five monitors simultaneously detected a glow with its radiation distribution horizontally extending for 2 km. The glow terminated coinciding with a lightning flash at 04:08:34 JST, which was recorded by the two radio-band lightning mapping systems, FALMA and DALMA. The initial discharges during the preliminary breakdown started above the glow, that is, in vicinity of the electron acceleration site. This result provides one example of possible connections between electron acceleration and lightning initiation.

Gamma-ray glows are thought to be bremsstrahlung from relativistic electrons accelerated through the Relativistic Runaway Electron Acceleration (RREA) process by the strong electric field (≥0.28 MV m −1 ) in the thunderclouds (Babich et al., 2004;Dwyer, 2003;Gurevich et al., 1992;Kelley et al., 2015). However, many unanswered questions still remain about the characteristics, meteorological conditions, mechanism of gamma-ray glows, and their relationship with lightning discharges. In particular, it is an interesting question whether the accelerated relativistic electrons in a glow can eventually enhance the chance of lightning initiation.
Gamma rays are attenuated in the atmosphere; the mean free path for, for example, a 10 MeV photon is about 400 m at the ground pressure (Köhn et al., 2017). Hence, ground-based measurements can detect glows from low-altitude clouds with base altitudes of not much greater than the gamma-ray mean free path. During winter in Japan, cold and dry air from the Siberia air mass frequently flows over the Tsushima warm current in the Japan Sea and brings low-altitude thunderclouds with cloud bases of 0.2-0.8 km (Goto & Narita, 1992) in the north coastal area of Japan. The area is ideal for ground observations of gamma-ray glow. We have performed multi-point radiation measurements around this area (Tsuchiya et al., 2007(Tsuchiya et al., , 2013Wada et al., 2018Wada et al., , 2021Yuasa et al., 2020).
Given that the atmospheric gamma-ray mean free path is ∼400 m, it is desirable to place detectors at intervals of a few hundred meters to track in detail the time variation of gamma rays emitted from a moving thundercloud. The observation system with this high-density grid would allow us to potentially answer several unsolved questions of high-energy atmospheric physics, including (a) the conditions for a glow to start and end, (b) the type of cloud where glows occur, for example, size and structure of the electron-acceleration region, and (c) relationship between glows and lightning discharges. Therefore, we launched the new citizen science "Thundercloud Project" at Kanazawa in 2018 for a high-density grid observation system, and here we present our first result from this project.

Citizen Science "Thundercloud Project"
For the citizen science "Thundercloud Project," we newly developed a portable, easy-to-operate radiation detector named "COmpact GAmma-ray MOnitor (Cogamo)." Every winter, our project researchers send the Cogamo detectors to citizen supporters after conducting detector maintenance. Citizen supporters receive and deploy them in their houses along Japan sea coast area and join the team of radiation observations. In the Japanese fiscal year (FY) 2021, we constructed a large-scale and multi-point observation network with 58 Cogamos installed in the Ishikawa, Toyama, and Niigata prefectures. Among them, the region in and around Kanazawa city had the highest detector density with 16 radiation detectors installed in a 5 km square area. Figure 1a shows Cogamo locations during the FY2021 winter campaign from December 2021 to March 2022.
The Cogamo detector is a small (23 cm × 28 cm × 10 cm) and lightweight (3 kg) radiation monitor, using a CsI (Tl) scintillator (5 cm × 5 cm × 15 cm) coupled with a Silicon Photomultipliers (SiPMs) MPPC (Multi-Pixel Photon Counter) as a photo sensor (Figure 1b). The energy range for gamma-ray spectroscopy is the ∼0.2-10 MeV band. The detector acquires the energy deposit and arrival time of each radiation event and records them into a microSD card. The time tagging is performed using GPS signals. In addition, 20-s bin count rates in six energy bands for 0.2-0.5, 0.5-1, 1-2, 2-3, 3-8, and >8 MeV, GPS status, ambient temperature, humidity, and optical luminosity are recorded on the microSD card and are also sent to the web server for a quick-look purpose. An observation is started simply by connecting a GPS cable and a power cable and then turning on the power switch. Energy calibration of the Cogamo detector was performed for each file of one-hour data when analyzing, using environmental background radiation lines of 40 K (1.46 MeV) and 208 Tl (2.61 MeV).
In FY2021, we also prepared an automatic gamma-ray glow real-time alert system on our web server to remotely monitor observations. With this system, we calculate the moving average of the count rate of the specified energy bands among the six bands (e.g., >1 MeV). Assuming the Gaussian distribution with statistical fluctuations from the latest 20-s count rate values, we send an alert by means of posting it on Twitter when the latest count rate exceeds the predefined threshold. The project summary will be reported in subsequent papers. In FY2021 only, we also conducted optical camera and gamma-ray Compton camera imaging (Kataoka et al., 2013;Kuriyama et al., 2022;Omata et al., 2020) from a high-rise window of a hotel in Kanazawa.

XRAIN Meteorological Radar
To determine the movements of thunderclouds above the Cogamo detectors, we use the eXtended RAdar Information Network (XRAIN) operated by the Ministry of Land, Infrastructure, Transport and Tourism (MLIT) of Japan. We used the data obtained with the XRAIN Nomi radar (Figure 1a), which is located 15 km away from Kanazawa in the southeast direction. The observation area is within 80 km of the radar, which covers around Kanazawa. The resolution of the radar is 150 m (radial) × 1.2° (azimuthal). The radar obtains three-dimensional volume scans with 12 elevation angles in 5-min interval. The radar data used in this paper is composited using scans with elevation angles of the bottom scan angles, 1.7 and 3.6°, corresponding to altitudes of 440 and 940 m at Kanazawa City, respectively. The original data set of XRAIN can be obtained from the DIAS service (https:// diasjp.net, which was supported by JAMSTEC).

Radio Observations of Lightning With the FALMA and DALMA
The Fast Antenna Lightning Mapping Array (FALMA) is a lightning mapping system working in the low-frequency band (0.5-500 kHz). As described by Wu et al. (2018), the FALMA is capable of three-dimensional (3-D) highly-accurate mapping of lightning channels. However, due to the fact that altitudes of lightning discharges in winter thunderstorms are much lower than in summer ones, the FALMA can only determine 2-D locations in most cases in winter observations (Wu et al., 2020). In order to perform 3-D lightning mapping in winter, we developed a new system called Discone Antenna Lightning Mapping Array (DALMA), working in the median-frequency and high-frequency bands (1.0-12.5 MHz) . During the winter campaign of 2021, we deployed 10 FALMA sites and 12 DALMA sites ( Figure 1). Thus, the FALMA provides 2-D location results of lightning discharges and also electric field change waveforms from which we can determine the discharge types. The DALMA provides 3-D location results of lightning discharges. The localization accuracies of FALMA and DALMA are estimated to be about 200 and 50 m Wu et al., 2020), respectively.

Data Analysis and Results
On 30 December 2021, the typical winter pressure pattern appeared over Japan, and severe thunderstorms passed over Kanazawa from west to east. At 04:07:37 JST (GMT+9 hr), the automatic alert system of gamma-ray glows was triggered by the Cogamo Identification (ID) 53. Figure 2a shows 8-s-binned count-rate histories in the 3-10 MeV band recorded by Cogamo IDs 62, 23, 33, 53, and 15 for a period encompassing the timing of the glow detection, whose locations are shown in Figure 3. The gamma-ray glow was the first simultaneous detection from a thundercloud with five Cogamo sensors of our citizen science campaign. Two of the five detectors recorded the glow termination coinciding with a lightning flash at 4:08:34.85 JST (Figure 2a) identified by FALMA and DALMA ( Figure 2b). In the discharge, Cogamo ID 33 also recorded a short spiky burst (five counts within 1 ms), which corresponds to a low chance-occurrence probability of 1.66 × 10 −10 for the background statistical fluctuations. The independent Compton camera at the same location as Cogamo ID 33 also recorded the spiky burst within 4 ms. We speculate that this burst is a TGF or leader X-ray emission associated with lightning discharges.
We evaluated the significant count-rate increases over background fluctuations, using the energy band above 3 MeV, which is not affected by the environmental background fluctuations. We fitted the histogram of the 8-s-binned and 1 hr count-rate histories with the Gaussian function for each detector. As an example, the background fluctuation of the data of Cogamo ID 53 was successfully fitted with an average count rate of 3.38 count sec −1 and standard deviation σ of 2.08 count sec −1 . Here we set the glow detection threshold at the count rate bins exceeding the 3.5σ significance (e.g., 3.38 + 2.08 × 3.5 = 10.6 count sec −1 for ID 53), at which a chance occurrence probability is 0.05% corresponding an expected false detection of 0.11 bins during the 1-hr observation. The glow period exceeding this 3.5σ threshold is shown in red in Figure 2a. The duration (T 3.5σ ) of the gamma-ray glow calculated from numbers of bins above 3.5σ were 24, 24, 80, 112, and 56 s, for Cogamo IDs 62, 23, 33, 53, and 15, respectively. In this calculation of the duration, we exclude the TGF events from Cogamo ID 33. Figure 3a shows the rainfall intensity maps every minute for a period of 04:06-04:09 JST, calculated from the radar reflectivity data. During the period, the high-intensity region was moving toward the east and passing over the Cogamo detectors. We estimated the on-ground horizontal size of the gamma-ray glow (hereafter "glow region") by multiplying the wind speed and the T 3.5σ duration (see also Wada et al. (2019)). Assuming that the glow region was a circle for the horizontal direction and using the method described in Wada et al. (2019), we estimated the glow region at each detector and plot all of them in Figure 3b. Here we adopted the wind speed and direction at this glow of 21.2 ± 0.8 m s −1 and 266.5° (clockwise with respect to the north), respectively, estimated from the XRAIN data. For example, the diameter of the glow region of Cogamo ID 53 is 2.3 km. The size of the glow region perpendicular to the wind direction is estimated from the distance from the north end to the south end of the glow regions of five detectors to be 2.5 km (Figure 3b).
The FALMA and DALMA identified the lightning flash starting at 04:08:34.8565 JST at the location (36.57816°N, 136.66101°E) in the vincity of the glow event. The lightning flash first developed northeastward and then southeastward (Figure 3c). In Figure 3d, we compared the locations between the Cogamo-detected glows and the initial signals of the discharges, in which we take into account the move of the glow with the wind. The first distinct four and three discharge signals of FALMA and DALMA, respectively, occurred roughly 420 and 540 m to the east-northeast of Cogamo ID 62. Taking into account the horizontal errors (∼50 m for DALMA and ∼200 m for FALMA, Section 2.3), the lightning flash started inside or in the vicinity of the glow region. The low-frequency radio waveform and the altitude of lightning discharge (Figure 2b) indicate that the flash started with a downward negative leader with a large speed of about 3 × 10 6 m s −1 . The downward leader, corresponding to the preliminary breakdown pulses, is followed by a negative return stroke in less than 3 ms. Interestingly, the peak amplitude of the largest preliminary breakdown pulse is larger than that of the return stroke pulse as can be seen in Figure 2c. Figure 4a shows the rainfall intensity maps at 04:08 JST for a wide area and Figure 4b shows a vertical cross-section of it along the axis of interest indicated in Figure 4a. The area of high rainfall-intensity was extended from the northwest to southeast for a distance of 11 km, while the glow was observed only in the northwest area across 2 km. In the glow region, high reflectivity exceeding 40 dBZ was found to develop beyond 2 km in height, while in the southeast no glow region with a reflectivity of more than 40 dBZ was found in the lower altitude (<2 km). Figure 4c is the vertical distribution of hydrometeor in the clouds. We applied the hydrometeor classification (HC) method for X-band polarimetric radar data to classify the radar data into eight types of primary precipitation particles: drizzle, rain, wet snow, dry snow, ice crystals, dry graupel, wet graupel, and rain-hail mixture (Kouketsu et al., 2015). We transformed the coordinate system from a polar one to Cartesian grids, using weighted interpolation according to the distance (Cressman, 1959). Wet graupels appeared up to the altitude 2 km above the glow region, whereas they are seen up to the altitudes only 1.5 km above the non-glow area.

Discussion and Conclusion
The present citizen science campaign at Kanazawa allowed us to perform high-density and multipoint measurements of a gamma-ray glow (Figure 2). We estimated the horizontal size of the glow to be 2 km in the east-west direction (parallel to the wind direction) and 2.5 km in the north-south direction (perpendicular to the wind direction). The radar-monitored moving  clouds of the high reflectivity (>40 dBZ) reached 2 km in altitude ( Figure 4b) and wet graupels found in the lower layers (Figure 4c) in the glow region. The characteristics of lightning discharges in Figure 2b suggest a negatively-charged middle layer appearing at 2 km altitude, and the present thundercloud case is likely to have a tripolar charge structure as expected in developed winter thunderclouds formed primarily via upward winds. According to the riming electrification model (Takahashi, 1978), wet graupels are positively charged at the lower layer with a temperature of around −10°C in winter thunderstorms. A downward gamma-ray glow is expected to be radiated from relativistic electrons accelerated in the electric field between this positively-charged lower layer and the negatively-charged middle layer. The radar data (Figure 4b), combined with altitudes of discharges (Figure 2b, see the discussion below), suggests that electrons were accelerated downward in the strong electric field below the negatively-charged middle layer located lower than ∼2 km in altitude in our case. A similar analysis of radar observations was previously performed but toward the opposite direction (upward) for a TGF (Mailyan et al., 2018).
The FALMA and DALMA data allowed us to further compare the initiation of lightning discharges and the gamma-ray glow. The discharges started about 420 m (FALMA) and 540 m (DALMA) away from Cogamo ID 62 ( Figure 3d). Taking into account the glow movement with the wind flow (846 m shift from Figure 3b), the initial lightning signals started inside or near the glow region, within the radio measurement error (∼100 m) and estimated horizontal size of the glow (∼510 m). This result suggests the electric-field regions initiating the lightning discharges and relativistic electron acceleration for the gamma-ray glow existed in the same space. As the vertical information, Figure 2b shows that the preliminary breakdown started at an altitude of 1.6 km and propagated downward, ending at 0.5 km in altitude. Assuming that a strong electric field is formed between these two altitudes at 0.5-1.6 km with a strength close to the RREA threshold of 0.28 MV m −1 (Babich et al., 2004;Dwyer, 2003), bremsstrahlung from accelerated electrons are detectable on the ground; for example, gamma-ray flux decreases to 28% to the downward direction from the 0.5 km altitude.
The radio signals (Figure 2b) indicated the leader speed 3 × 10 6 m s −1 at the preliminary breakdown stage, which is about one order of magnitude higher than those in the previously reported lightning flashes (Saba et al., 2009(Saba et al., , 2014Shi et al., 2019;Wu et al., 2015). It is widely believed that a high leader speed is usually associated with a strong ambient electric field (Nag & Rakov, 2009;Shi et al., 2019;Wu et al., 2022). The existence of such a strong vertical electric field is also supported by the strong preliminary breakdown signal (Figure 2b). An interesting possibility is that the flow of relativistic electrons, which works as the source of the gamma-ray glow, may enhance the chance of the initiation of this lightning flash at the ∼1.6 km altitude. The ambient electric field around the RREA threshold could increase the free-electron density involved in creating a leader plasma channel to promote the onset of lightning discharges (Francisco et al., 2021). Although the RREA-accelerated electron population alone would not be able to trigger lightning (Coleman & Dwyer, 2006), the augmentation of the free electron population around sharp ice particles could enhance the electric field (Dubinova et al., 2015), which may play a significant role in lightning initiation. This hypothesis is supported by some past balloon experiments; they reported the occurrences of lightning discharges at the vicinity environment of the electric field strength of 0.2 MVm −1 (Nicoll, 2012) closer to the RREA threshold and an order of magnitude lower than the classical breakdown threshold in the air of ∼3 MVm −1 (Diniz et al., 2019).
The lightning flash terminated the glow, as detected by two Cogamo detectors (IDs 53 and 62), similar to previously reported events (Hisadomi et al., 2021;Tsuchiya et al., 2013;Wada et al., 2018Wada et al., , 2019. This glow termination means the disappearance of the electric field in the RREA region and the neutralization of the charged layers. Interestingly, Figure 3c suggested that the lightning discharges did not enter the original glow region but propagated along the edge or outside the glow region. The fact implies that the charge cancellation happened quickly after the termination of the glow once the lightning is initiated, and as a result, the lightning channel propagates to the places where the charge separation still remains. We conjecture that the reason why gamma rays were detected in the northwestern region but not in the southeastern region would be due to some differences in their meteorological conditions. A working hypothesis is that at the late stage of the life cycle of a thundercloud associated with the weak updrafts, a pocket in the positively-charged lower layer with wet graupels falls to the ground, dragging the charge-separated region toward the ground and providing a detectable condition of the glow. In our case, a potential glow in the southeastern region with falling graupels had likely already ended before the cloud reached our Cogamo detectors; this hypothesis is consistent with the relatively low (<1 km) high-reflectivity (>40 dBZ) region and observed wet graupel distribution. An alternative scenario is that the electric field condition in the thunderstorm did not develop sufficiently in the southeast region. In the former case, the gamma-ray glow may have occurred before the time 04:08 JST as of Figure 4, but we cannot verify it because there were no detectors in the area. In the coming years, our Thundercloud Project will continue radiation-mapping observations to increase the sample number of gamma-ray glows, extending the first result as reported in this Letter to advance our understanding of thunderstorm physics.

Data Availability Statement
The materials presented in this study are available online (on Mendeley Data; https://data.mendeley.com/datasets/ xy75zjym8f/1). The radar data shown in Figures 1, 3 and 4 is provided by an X-band polarimetric RAdar information network (XRAIN) which is available at: https://search.diasjp.net/en/dataset/MLIT_XRAIN. NHK for the installation of the detectors, and K. Takemoto (Nagoya University) for analysis of the hydrometeor classification algorithm, and T. Sakamoto (Aoayma Gakuin University) for the support for this project. This research is supported by JSPS/MEXT KAKENHI Grants 16H06006, 19H00683, 20K21843, 22H00145, 22F21323, 21K03681, 20K14114, 21H00166 and 22KF0190 by the Hakubi projects of Kyoto University and RIKEN, by JST Grants JPMJFR202O (Sohatsu), JPMJER2102 (ERATO), and by crowdfunding "Thundercloud Project" operated on the academic crowdfunding platform "academist." The XRAIN data were obtained by the Japanese Ministry of Land, Infrastructure, Transport and Tourism and retrieved from the Data Integration and Analysis System (https://diasjp. net). The background image in Figures 1, 3 and 4 was provided by the Geospatial Information Authority of Japan.