A Review of Atmospheric Electricity Research in China from 2019 to 2022

Atmospheric electricity is an important branch of atmospheric science research that combines a broad range of electrical phenomena in the atmosphere. In China, there are several lightning research groups (LRGs) devoted to atmospheric electricity, and significant progress has been made by LRGs in recent decades. Qie (2012) and Qie and Zhang (2019) reviewed the research progress in atmospheric electricity achieved in China during the periods 2006–10 and 2011–18, respectively. Qie et al. (2015) reviewed research that focused primarily on ground-based thunderstorm and lightning field experiments conducted in different regions of China during 2005–14. Zhang et al. (2016) reviewed the research advances on the discharge processes of artificially triggered and natural lightning, as well as the advances in lightning protection technology testing conducted during the period 2006–15 in Guangdong, China. Qie et al. (2021b) reviewed the research advances resulting from the “Dynamical-microphysical-electrical Processes in Severe Thunderstorms and Lightning Hazards (STORM973)” project, during which the main focus was the collection of field observations in the Beijing metropolitan region over the period 2014–18.

In the last few years, most advancements resulting from the study of atmospheric electricity in China were related to a project sponsored by the National Key Research and Development Program of China: “Research on Thunderstorm Electrification-discharge Processes and Lightning Effects”. The goals of this project are to: 1) develop a real-time three-dimensional (3D) lightning channel mapping system with a high resolution, 2) observe thunderstorm electrification–discharge processes collaboratively in China, 3) reveal the relationship between multiple characteristics of lightning activities and thunderstorm structure, 4) develop a seamless lightning forecasting and warning operational demonstration platform by leveraging artificial intelligence techniques, 5) analyze the effects of grounded objects on the lightning striking process, multi-dimensional electromagnetic radiation features, and coupling mechanisms of the lightning process, and 6) establish an experimental research platform for lightning damage and protection.

In this paper, the atmospheric electricity research progress achieved in China during 2019–22 is reviewed focusing on the following aspects: (1) lightning detection and location techniques, (2) thunderstorm electricity, (3) lightning forecasting methods and techniques, (4) physical processes of lightning discharge, (5) high energy emissions and effects of thunderstorms on the upper atmosphere, and (6) the effect of aerosol on lightning.

Total lightning positioning based on Very Low Frequency/Low Frequency (VLF/LF) signals has the advantages of long detection distance and strong anti-interference performance. In China, in addition to the cloud-to-ground lightning (CG) location network based on VLF/LF signals being built by the business department (Zhang et al., 2022c), several LRGs have also built their own total lightning location networks and are constantly developing new positioning techniques. Since 2010, different location networks have been continuously operating in the scientific research fields, including the Low-Frequency E-field Detection Array (LFEDA) of the Chinese Academy of Meteorological Sciences (CAMS), Beijing Lightning NETwork (BLNET) of Institute of Atmospheric Physics (IAP), Chinese Academy of Sciences (CAS), and hybrid baseline total flash positioning network of University of Science and Technology of China (USTC).

For thunderstorm and lightning research, starting from 2014, the CAMS LRG has built a lightning detection array based on LF electric field fast antennas, named as LFEDA. Through waveform post-processing positioning, it can obtain the 3D location of LF sources with high efficiency and positioning accuracy (~100 m, evaluated by triggered lightning). In recent years, Zhang et al. (2021f) upgraded the original hardware setup of LFEDA and developed a new generation of the real-time LFEDA (RT_LFEDA) substations based on the FPGA real-time signal processing technique, as shown in Fig. 1. The new setup has the advantages of lower power consumption (10–24 W), higher integration, and faster data processing speed. Each substation is used to digitize and extract VLF/LF signal characteristics in real time and save the original waveforms. The Time of Arrival (TOA) or 3D algorithm developed with the graphical processing unit (GPU) is used for real-time positioning.

Figure 1.  The (a) new, low power consumption substation of RT_LFEDA and (b) positioning results for a lightning flash.

The IAP LRG built a multi-band lightning 3D positioning network in Beijing (BLNET) around 2010. In 2015, the sensitivity of sensors, the operation efficiency of software, and the detection performance of the network were improved by updating the hardware, station network layout, and positioning algorithm of BLNET (Wang et al., 2020a; Yuan et al., 2020). Ma et al. (2021c) then upgraded the location technique, and a 3D total lightning positioning network of eight stations was deployed in Binzhou, Shandong Province. The system synchronously digitized the signals of derivation of electric field change, the fast and slow electric field, through a multi-channel data acquisition card controlled by the industrial computer. The TOA particle swarm optimization (PSO) positioning algorithm is adopted based on waveform post-processing, which has a high positioning accuracy verified by triggered lightning.

The USTC LRG has been building a sferic detection array in the Jianghuai region (JASA). Five stations in the Jianghuai region and a remote station located in Conghua (~950 km away) were deployed before 2019 to investigate narrow bipolar events (NBE) activities in the relatively low latitude region (Liu et al., 2021b). Since 2019, the network has been expanded to 13 stations with multiple baselines ranging from 100–200 km to about 1000 km and is designed to geolocate the large-scale lightning flashes during thunderstorms over the eastern coastal areas of China. The network adopts a GPU parallel processing technique with real-time positioning capability.

How to improve the performance of total lightning location system (LLS) operating at VLF/LF band has always been a research focus. Different techniques have been adopted to improve the precision.

The CAMS LRG developed a positioning algorithm combining time reversal and the TOA method when LFEDA was built in the early stage (Chen et al., 2019b; Zhang et al., 2021f). Compared with the traditional TOA method, this method can obtain higher-quality 3D positioning results with fewer matching stations (as low as four stations), poor signal-to-noise ratio, and lower time accuracy, which is beneficial to the positioning of long-distance weak signals.

Other location techniques have also been applied to the LF total lightning location problem, which improves the matching ability of the pulses from the same discharge source. For example, Fan et al. (2021a) applied the empirical mode decomposition (EMD) technique in LFEDA, which achieves LF signal filtering and high-frequency noise reduction. After waveform correlation matching by gradually reducing the window, the optimal solution was solved by the traditional time of arrival method, and the positioning number was increased by more than five times. Furthermore, based on the waveforms obtained by the 3D lightning radiation source location system established in Datong County, Qinghai Province, Wang et al. (2021g) combined the Pearson correlation method with EMD to match the electric field pulse of lightning discharge, and a better performance (nearly seven times of sources were located) was achieved compared to the pulse-peak feature matching method.

Compared with waveform-based localization, feature-based localization is fast, but the precision is low and needs to be improved. Wang et al. (2021c) proposed a fast and refined positioning method based on deep learning features that can significantly improve the matching ability of multi-stations signals and the location speed. The locating results for a natural lightning flash using this method are shown in Fig. 2. Compared to the pulse-peak feature matching and waveform cross-correlation matching methods (TOA-TR methods), this new method has a higher matching efficiency (increased by more than 50%).

Figure 2.  Positioning results based on encoding features for a natural lightning flash. (a) Source altitude development, where the blue to red colors illustrate the beginning to the end of the flash. (b) North–south vertical projection. (c) Source height distribution. (d) Plan view. (e) East–west vertical projection of lightning radiation sources. [Reprinted from (Wang et al., 2021c)]

Compared with the LF positioning technique, the Very High Frequency (VHF) positioning technique has natural advantages in depicting lightning channels. Zhang et al. (2020f) developed the continuous interferometer (CINTF) system in 2016. It consists of three VHF broadband discone antennas arranged at the vertices of an equilateral triangle of 20 m and a high-speed acquisition system. In the first few years since the system’s inception, the signals of the three antennas were filtered and digitized continuously with a sampling rate of 200 MS s^–1, and a maximum record length of 2 s was achieved. Recently, the system has been upgraded to achieve a 400-MS s^–1 sampling rate and a maximum record length of 4 s. The correlation time delay method was used to calculate the elevation, azimuth, and other information. It can achieve high-precision positioning with sub-microsecond resolution.

Chen et al. (2022c) further developed a calibration method for the source positioning error of CINTF at close range and validated this method with rocket-triggered lightning experiments. During the initial stage of an upward positive leader (UPL), the calibrated altitude positioning errors of CINTF are approximately 11 m, 14 m, and 20 m, when the source elevation angles are 40°, 50°, and 60°, respectively, which suggests an error increasing with the source ascending. Based on the calibrated observation results, Chen et al (2022c) investigated the development characteristics of the initial stage of the UPL and discussed the possible physical mechanisms.

A lightning broadband VHF interferometer has been deployed in Lhasa since 2019 by the IAP LRG. Li et al. (2021b) introduced a new hybrid algorithm in this interferometer to image the two-dimensional (2D) lightning channels. The new algorithm combines the time difference of arrival (TDOA) and the electromagnetic time reversal (EMTR) technique, so it was named the TDOA-EMTR technique. Different from the traditional EMTR technique, FFT spectrum metrics were used to remove the low-power points in each time window. Results show that TDOA-EMTR can improve positioning efficiency by more than a factor of three to four.

Wang et al. (2020d) used a method similar to Stock and Krehbiel (2014) to improve the performance of the interferometer by using multiple baselines. Seven antennas are used to form an “L” shape interferometer (also known as a continuous VHF imaging array), and a method of applying multisignal classification to estimate the direction of arrival of a lightning VHF radiation source is proposed. This method improves the precision of traditional continuous interferometer positioning.

A dual-band location system operating at VLF-LF/VHF ranges was developed and reported by Liu et al. (2019c). It consists of 14 sensors operating at both VLF-LF and VHF ranges and covers an area of approximately 100 km in diameter in Chongqing, China. They suggest that the link between the sources of different frequency ranges can be investigated and a more complete picture of lightning channels can be described with sources from different frequency bands.

VHF positioning data can provide a variety of information with post processing. Li et al. (2021e, 2022) developed the lightning extension channel construction method and channel morphology characterization method based on 3D radiation source location data from the Lightning Mapping Array (LMA). These methods reduce the number of incorrect connections to obtain a more realistic discharge scale and are used to reveal the characteristics of charge-level distributions.

China’s first geostationary satellite lightning imager, i.e., the Lightning Mapping Imager (LMI) onboard the Fengyun-4A, which was launched in December 2016, provides unprecedented lightning detection over China and its neighboring regions. The LMI can continuously detect total lightning over a large area and provide important information for the early warning of severe storms (Yang et al., 2017).

Cao et al. (2021) compared the optical signatures and spatial distribution characteristics of lightning observed by the LMI and the Lightning Imaging Sensor (LIS) on the International Space Station and found that fewer and shorter lightning flashes were observed by LMI. The number of LMI groups and the peak area of lightning density are consistent with reports from the China lightning locations network. The statistical results for duration, brightness, and coverage area of the LMI in different regions were also given based on the 2019 LMI lightning dataset. For isolated thunderstorm clusters, Liu et al. (2021f) found that the horizontal detection error of the LMI is about 15 km when comparing with the ground-based lightning observations. And LMI detection is relatively less efficient during the daytime and increases to a higher value from 1300 Beijing Time to midnight (Li et al., 2021c).

Hui et al. (2020) studied the influence of lightning light radiation characteristics on the detection capacity of the LMI and found that flashes detected by the LMI exhibit higher optical radiance but shorter durations than flashes detected by the LIS onboard Tropical Rainfall Measuring Mission (TRMM). The LMI has detected fewer lightning events over the Tibetan Plateau (TP) and presents a lower detection efficiency there. However, it has advantages in revealing the higher intra-cloud lightning ratio in intense convection (Zhang et al., 2020d).

The performance of the FY-4A LMI was evaluated during multiple convective events over the Beijing area by Chen et al. (2021). They suggest that the spatial distributions of lightning from the LMI and BLNET are consistent with each other. However, the number of flashes detected by the LMI was roughly one order of magnitude less than that of the BLNET.

The electric field sonde is a key tool used to study the charge distribution of thunderstorms. In recent years, the IAP LRG has constructed a thunderstorm electric–meteorological integrated sounding system based on the developed double-metal sphere electric field sonde in combination with the weather radiosonde. The integrated sounding system performs synchronous measurement of the electric field, temperature, and humidity in thunderstorm clouds. In the summer of 2019, a field experiment was conducted in the North China Plain, and Zhang et al. (2021b) presented the distribution characteristics of the electric field and charge structure in thunderstorm clouds in this area for the first time.

Dynamical and microphysical processes in thunderstorms determine the intensity, type, polarity, and even channel propagation characteristics of lightning activity by influencing the charging process and the structure of charge distribution. The electrical activity may also affect the dynamic and microphysical field in thunderstorms (Lian et al., 2020).

Mesoscale convective systems (MCSs) typically cover a spatial scale from tens to hundreds of kilometers and yield active lightning flashes. Liu et al. (2021a) reported that three types of MCSs accounted for about 73% of linear MCSs occurring in Beijing, including linear leading convective lines with a trailing stratiform region (TS), leading stratiform region with a trailing convective region (LS), and leading convective lines with a parallel stratiform region (PS). They also found that lightning mainly occurred in the linear convective region of TS, LS, and PS MCSs; at the dissipating stage, the lightning increased in the stratiform region of TS MCSs; fewer lightning flashes occurred in the LS MCSs; and the ratio of positive CG (PCG) lightning to CG lightning was high in PS MCSs. PS MCSs generally had the highest ratio of PCG to CG lightning, followed by the LS and TS MCSs. Zhao et al. (2021a) found that the positions where the lightning channels propagate are mainly dominated by the graupel and ice crystal particles in the convective regions of MCSs, by dry snow first and ice crystal particles second in the stratiform regions of MCSs. In typical MCSs, the lightning frequency may be positively correlated with the volume of the regions with reflectivity greater than 30 dBZ or 40 dBZ (Liu et al., 2021a; Yu et al., 2022). Meanwhile, Chen et al. (2020c) noted that, in a rapidly developing MCS, the flash rate increased much faster than storm volume growth. In MCSs, only small regions were associated with more than two flash initiations in six minutes (Wang et al., 2022a). These regions were found to be primarily distributed in a stable altitude ranging from 9 km to 13 km, while only a few of these regions were at a lower altitude near the melting level. Wang et al. (2022a) suggested that the spatial relationship of these lower regions to the reflectivity core was different to that of the main altitude regions.

Cells merging is common in the convective lines of MCSs. Lu et al. (2021) found that, during the merging process of cells, the total flash rate slightly decreased at the beginning, sharply increased later, and peaked when the merging process was completed. The cloud bridge was found to be one of the regions where lightning events increased the most, which benefited from the increase and expansion of ice particles at the middle-upper levels. The simulation performed by Lu et al. (2022) suggested that the charge distribution in the cloud evolved from a staggered charge pocket pattern into a vertically stratified five-layer pattern during the merge stage.

Lightning activity in the stratiform regions of MCSs has recently been an area of focus. It is reported that more than 90% of flashes in the stratiform region of MCSs propagated above bright-band areas (Wang et al., 2019b), and most of the stratiform flashes had a single-layer structure in their channel extension area (Wang et al., 2020b). Usually, most of the stratiform flashes were initiated within a certain distance from the reflectivity core and near/in the melting layer (Wang et al., 2020b). Wang et al. (2021b) further reported that the negative leaders of stratiform flashes mainly propagated within a height range of 9–12 km in most MCSs, especially small or developing MCSs, and propagated more frequently in a low-altitude range (5–7 km) in the MCSs with a large and developed stratiform region.

Hailstorms usually produce strong lightning activity accompanied with hailfall and a high proportion of PCG lightning (Liu et al., 2020a, c; Zheng et al., 2021a). For example, Sun et al. (2020) reported that the ratio of PCG lightning rapidly increased before the hailfall and reached a peak of 58% in a hailstorm, which should be due to the inverted tripolar charge structures (Zheng et al., 2021a).

Lightning data can provide useful information for hailfall warnings. The 2σ lightning jump algorithm can provide a mean lead time of around 30 min for predicting the occurrence of hailfall, but it also exhibits a relatively high false alarm rate exceeding 30% when using total lightning data and 50% when using CG lightning data (Tian et al., 2019, 2021; Sun et al., 2020). A new method was proposed by Tian et al. (2022) that combines the hydrometeor classification results of dual-polarization radar and the 2σ lightning jump algorithm. They suggested that the evolution of large hail, graupel and small hail, rain and hail, and ice crystal grid point numbers in a thunderstorm can be used to effectively identify valid and invalid lightning jumps, thus improving hail warning performance. From a test based on 17 hail cases, they found that the new method decreased the false alarm rate from 58.5% to 29.2% and increased the critical success index from 41.5% to 70.8%.

In terms of number and intensity of tropical cyclones (TCs), the northwest Pacific is the most active basin. TCs contribute approximately 4.9% of all lightning in this area, and the regions with highest contributions are located in the northern South China Sea and the ocean northeast of the Philippines. Meanwhile, the dominant lightning contributors are TCs of tropical storm strength (Zhang et al., 2020e). Furthermore, in the northwest Pacific region, TCs contribute more lightning activity during La Niña periods compared with El Niño periods (Zhang et al., 2020e, 2021e).

There is a strong correlation between lightning frequency and the timing of the TC maximum sustained wind speed. Kong et al. (2021) reported that the maximum lightning frequency occurred about 25–49 hours before the maximum intensity in terms of wind speed. Lightning in the inner-core or eyewall may have a higher ratio of intra-cloud lightning (Zhang et al., 2020d) and present an active phenomenon with a sharp increase in frequency (i.e., lightning outbreak) during the rapid intensification (RI) stage (Zhang et al., 2019b). RI lightning outbreaks occur primarily in the downshear quadrants and inside the radius of maximum wind (RMW) (Zhang et al., 2019b). For landing TCs, the dominant factor for producing lightning and convective asymmetry is the vertical wind shear. Zhang et al. (2022a) analyzed the lightning activities in both weak (< 32.7 m s^–1) and strong (≥ 32.7 m s^–1) TCs. They suggest that the lightning asymmetries were enhanced when the shear magnitude increased. This implies that lightning is an indicator of intense convection and TC intensity. And thus, lightning data may be assimilated into numerical models to understand the deep convection in TCs and to predict their evolution (Zhang et al., 2020c).

Different studies report lightning spatiotemporal size parameters for different locations (You et al., 2019a, b; Zheng and Zhang, 2021) or for different-type thunderstorms (Zheng et al., 2018, 2019b). In most studies, a negative correlation between flash spatial extent and flash rate for various types of thunderstorms has been suggested, that is, the thunderstorms with relatively weak convection tend to have lower flash frequency and greater flash-length compared to those with relatively strong convection. Zheng and Zhang (2021) reported a more complicated relationship between flash rate and flash size through the comparison of the thunderstorm and lightning features among the TP, Central and Eastern China (CEC), and the Himalayas (SHF). Combined with the concept suggested by Bruning and MacGorman (2013), Zheng and Zhang (2021) put forward a new perspective (Fig. 3). Below a certain convection intensity threshold, the enhancement of convection will reinforce the charging rate and amplify the area of the high-density charge region, which causes the common increase in flash rate and spatial extent; if the convection exceeds the threshold, the strong airflow may break up the charge regions and help form small, staggered charge regions (like the pattern of charge pockets suggested by Burning and MacGorman), which contributes to the high-frequency but small-spatial-extent flashes, i.e., negative correlation between flash rate and length. However, the relationship between flash duration and thunderstorm structure is still obscure. Flash duration also shows weak correlation with the flash length and radiance (You et al., 2019b).

Figure 3.  A conceptual diagram suggested by Zheng and Zhang (2021) to describe the variation of the flash rate and size with the enhancement of thunderstorm dynamical process. Vertical grey dashed lines mark the possible situations of the thunderstorms over the Tibetan Plateau (TP), Central and Eastern China (CEC), and the southern foothills of the Himalayas (SHF). The suggested dynamic intensity threshold (purple dotted dashed line) splits the upward and downward variation tendency of flash size.

Zheng et al. (2018, 2019a) suggested that flash initial leader (IL) properties may be related to thunderstorm structure. They found that the ILs of flashes in supercells were characterized by shorter propagation distance, more tilted direction, and slower average speed, relative to those in normal thunderstorms. These differences were attributed to the fact that supercells tend to be predominated by the pattern of charge pockets, and normal thunderstorms tend to be predominated by a horizontally broad and vertically layered charge distribution pattern. They further found that the IL properties of the flashes initiated in strong and weak dynamic regions may also be different. In supercells, the ILs initiated in the regions with relatively weak dynamic processes tended to have a greater average propagation speed than those initiated in the regions with relatively strong dynamic process, which was opposite of the situation in normal thunderstorms. A proposed explanation is that, in supercells, the large, horizontally broad charge regions in the weak dynamic areas provide stronger ambient electric fields than the small, staggered charge regions in the strong dynamic areas. Meanwhile, in a normal thunderstorm, the strong and weak dynamic areas may both be predominated by horizontally broad charge regions. Also, strong dynamic areas had larger charging rates, and therefore, greater ambient electric fields.

Charge structure is the bridge connecting the dynamic and microphysical processes of thunderstorms and lightning activity, and new knowledge about charge structure has recently been gained. In situ electric field sonde experiments have been conducted, and they suggest an inhomogeneous and complex charge distribution pattern within the clouds (Yu et al., 2021; Zhang et al., 2021b, d). Xu et al. (2020) suggested that the charged ice phase particles transported by the backward flow behind the convective line into the stratiform region were the main charge source; meanwhile, the in situ charging of noninductive collisional and melting processes also contributed to the formation of charge structure. By using a new explicit lightning model implemented within WRF-ARW and a cloud model featuring explicit inductive and noninductive charging processes, Gan et al. (2020), Zhang et al. (2021c), and Zou et al. (2023) explored the charge structures and the formation mechanisms of thunderstorms in South and North China. As for the formation of inverted charge structure, which typically exists in severe thunderstorms, Xu et al. (2019) suggested that a strong updraft (>16 m s^–1), large liquid water content (>2 g m^–3), and high graupel rime accretion rate (>4.5 g m^–2 s^–1) above the –20°C layer might cause positive charging of graupel and negative charging of ice crystals.

Some studies have further explored charge structures of distinctive thunderstorms. Zheng et al. (2019b) first documented the inverted dipolar and inverted tripolar charge structures in winter thunderstorms. Wang et al. (2021a) indicated that the high PCG percentage in Hokuriku winter thunderstorms was caused by the inverted charge structure rather than the various deformations. In a rainstorm dominated by warm precipitation, Liu et al. (2020c) noted that the charge accumulation regions indicated by the relatively active lightning deviated from the charging region indicated by the relatively strong convection. They suggested that, with weak convection in the rainstorm, the charges were transferred from where they originated (i.e., convection core) faster than the charging rate, causing a relatively low charge density in the convection core; meanwhile, the transferred charged particles might aggregate in regions away from the convection core and form relatively high charge density locally, therefore producing relatively frequent flashes. TP thunderstorms have been found to be prone to a larger-than-usual lower positive charge center (LPCC) (Qie et al., 2005, 2009). Wang et al. (2019a), by means of numerical modeling, suggested that weak convection and a low freezing level are essential for LPCC formation; meanwhile, the prominence of the LPCC benefits from weakened charge density in the upper positive charge region, but the charge density in the LPCC of a TP thunderstorm is not significantly different from that in the LPCC of a typical plain thunderstorm.

The association of the charge structure with the lightning discharges is important. By coupling an improved stochastic lightning parameterization for upward lightning into a thunderstorm model, Tan et al. (2019) investigated the relationship between the electrical environment characteristics of thunderstorms and the initiation of upward lightning. It was found that when self-initiated upward lightning started, the corresponding bottom charge region was closer to the ground and its charge was comparable to that of the upper charge region. The model suggested that with a greater concentration of the bottom charge region came a higher weighted height of the bottom charge region. In addition, Zheng et al. (2021b) quantitatively evaluated the effect of upward lightning discharges on space charges and potential and electrostatic fields using the thunderstorm model. They also analyzed the neutralization process of the deposited opposite-polarity charge in the lightning channel and its effect on the space charge structure and subsequent lightning discharges. Xu et al. (2021) analyzed the effects of the horizontal distribution characteristics of thunderstorm charge structure on lightning type as well as discharge behavior. They proposed that a compact upper positive charge region and a uniform main negative charge region favored the initiation of positive cloud-to-ground flashes, while the horizontal offset of the high concentration charge center in the upper positive charge region was conductive to the occurrence of bolt-from-the-blue flashes.

Thunderstorms and lightning flashes are coupled weather phenomena that exist within larger weather systems. They are affected by and have an impact on climate change. Some studies have focused on their climatological characteristics.

Using the FY-2E black body temperature (TBB) and cloud classification products, as well as the lightning data provided by the World Wide Lightning Location Network (WWLLN), Ma et al. (2021a) created a thunderstorm feature dataset (TFD) covering the full-disc observation range of the FY-2E satellite and analyzed the thunderstorm activity over the Chinese mainland and the adjacent seas. They reported that the most frequent thunderstorm activity occurred in South China, Southwest China, the Eastern and Central TP, and the South China Sea and the weakest thunderstorm activity occurred in Northwest China. They also reported the differences in monthly and diurnal variations and cloud area between the thunderstorms occurring over the Chinese mainland and the adjacent seas.

Thunderstorms occurring over the TP have been investigated in several studies. Generally, TP thunderstorms feature weak convection, small horizontal extent, and low flash frequency and flash density, and they predominantly occur during May–September and in the afternoon (Zheng et al., 2020; Zheng and Zhang, 2021). Based on data from the Precipitation Radar (PR) aboard the TRMM satellite, Qie et al. (2022b) reported that TP thunderstorms increased in size and intensity from west to east, the thunderstorm frequency was the highest in the central TP, and the thunderstorm lightning-yielding ability was the strongest over the east TP in June. Using the TFD (Ma et al., 2021a), Du et al. (2022) revealed three high-frequency thunderstorm activity centers in the southeast, south-central, and southwest regions of the TP, respectively.

Furthermore, thundersnow is a special kind of thunderstorm that combines snowfall and lightning. A 10-year climatology of thundersnow in China was developed based on observations during 2008–2017 (Xu et al., 2022b). They found thundersnow events to be widely distributed across China, while the regions of high-frequency thundersnow events are located in the TP and northeastern China.

Lightning activity across most of China’s mainland was investigated by Xu et al. (2022c). Over China’s mainland area, lightning activity generally decreases from south to north and from east to west, with a mean CG density of 0.9 fl km⁻² yr⁻¹ and maximum CG densities of more than 10 fl km⁻² yr⁻¹. However, differences in the spatiotemporal distributions may be revealed by using different observations, such as WWLLN and LIS, as suggested by Ma et al. (2021b). Both WWLLN and LIS show intense lightning activity over the central and southeastern TP during the period from May to September. LIS indicated very intense lightning activity over the northeastern TP, while WWLLN indicated relatively weak lightning activity. And WWLLN indicated a high-density center of lightning activity over the southwestern TP, which is not suggested by LIS. Li et al. (2020) suggested a new parameter that is a product of rainfall, the Bowen ratio, and surface specific humidity, and found it showed a better correlation with lightning activity than rainfall alone or other parameters over the TP.

Focusing on lightning activity in the South Asia Region over the past two decades, Qie et al. (2021a) identified a clear increase in lightning density with an increasing rate of 0.096 fl km^–2 yr^–1. They suggest that the surface latent heat flux along the west coast of the Indian subcontinent can explain 52% of the lightning variance and may contribute to a 0.025 fl km^–2 yr^–1 increase. Furthermore, Qie et al. (2022a) suggested that in the TP region, lightning activity has exhibited a significant increase in the last two decades in both the OTD/LIS and WWLLN observations, with an average rate of 0.072 ± 0.069 fl km^–2 yr^–1 during 1996–2013.

Lightning nowcasting and warning play vital roles in lightning disaster reduction. Accurate and timely lightning forecasting is still a big challenge. Different from the traditional lightning nowcasting methods that mainly refer to the precipitation structural parameters of the thunderstorm, Zhao et al. (2021b) analyzed the differences in the characteristics of vertical turbulence between thunderstorms and non-thunderstorms. They found that, at the first flash stage, the eddy dissipation rate of thunderstorms can reach 0.19 m² s^–3, and the achievable height of turbulence can exceed the –30°C layer; in contrast, during the whole non-thunderstorm lifetime, the corresponding maximums are 0.12 m² s^–3 and about the –10°C layer (hard to exceed this layer), respectively. The first radar echoes of thunderstorms and non-thunderstorms were investigated by Zhao et al. (2022). They found that thunderstorms and non-thunderstorms show different characteristics of liquid/ice particles in different height layers in their first radar echoes, and non-thunderstorms tend to feature larger echo intensities below the –10°C level. These results indicate that we can warn of the occurrence of lightning earlier, even when the storm first appears.

Based on BLNET total lightning data and radar observations, Srivastava et al. (2022) proposed a thunderstorm tracking algorithm to identify and validate convective cells. By applying a neighborhood and smoothing technique, the boundaries of convective cells were distinguished. Then the identified cells were extrapolated to forecast the lightning activity area. Their results show that the algorithm has accuracies of 63%, 80%, and 91% for 30 min, 15 min, and 5 min lead times, respectively.

Quantitative and diagnostic relationships have also been established based on tracking entire thunderstorm cells. By using 3D total lightning location data and continuous-waver radar data (vertical detection), Cui et al. (2022) investigated the relationship between lightning occurrence and the precipitation cloud column (PCC) structure at a fixed position. They found the prominent differences in the structure between the PCC with and without lightning and indicated that the hydrometeor size and diversity at a specific spatial grid box should be the key parameter determining the lightning occurrence in PCCs.

Very short-range lightning forecasting has more lead time. Existing prediction methods rely on numerical weather prediction (NWP) models, and the forecast performance fundamentally depends on the physical parameterization schemes.

Using the Weather Research and Forecasting (WRF) model coupled with electrification and discharge schemes (WRF-Electric), Xu et al. (2022a) carried out numerical experiments for lightning prediction in North China. Their results indicate that lightning activity regions can be predicted well by WRF-Electric, particularly for the 6–12-hour period. However, compared with the observations, the predicted lightning activity region was relatively more concentrated and the flash density was higher. They suggest that the lightning initial threshold should be modified according to the resolution of the model in the discharge parameterization process, and the neutralization charge magnitude in a single discharge should refer to the observations.

In recent years, artificial intelligence has been applied in many fields, and lots of machine learning techniques have been developed. Some machine learning techniques were also introduced into lightning forecasting.

Zhou et al. (2020) developed a new CG nowcasting algorithm using a semantic segmentation deep learning network they named LightningNet. In this technique, both Himawari-8 satellite and radar echo data were used, and the 2D convolution in the semantic segmentation network was replaced with 3D convolution to obtain the lightning occurrence probability. For 0–1-h lightning nowcasting, the evaluation results show that the probability of detection, the false alarm ratio, and the threat score of LightningNet reached 0.633, 0.386, and 0.453, respectively.

Cui et al. (2022) established quantitative and diagnostic relationships based on grid box correspondence by using 3D total lightning location data and continuous-waver radar data (vertical detection). By using the Light Gradient Boosting Machine algorithm, they developed a lightning diagnosis program integrating multiple radar parameters. Evaluation results show that the algorithm has a lightning occurrence hit rate of 93.5% and a threat score of 0.421.

Geng et al. (2019) proposed a data-driven model for lightning prediction based on neural networks named LightNet. The dual encoders in LightNet were designed to extract the simulation spatiotemporal features of WRF data and recent observation of lightning data to attempt to calibrate the simulation products and assist the prediction. Experiments show that LightNet can significantly improve 6-hour predictions and achieve an equitable threat score three times better than the previous designs. Geng et al. (2021) further established a multi-source data-driven forecasting framework based on deep neural networks including a lightning scenario. This scenario may be heterogeneous in spatial and temporal domains. Evaluation results show that with more data sources, a higher forecasting score is achieved, and a better performance compared to the formerly established 6-hour lightning forecasting schemes is obtained.

For 12-hour lightning forecasts, Lin et al. (2019) proposed an attention-based dual-source spatiotemporal neural network (ADSNet). In the forecasting procedure, this model draws on the advantage of the RNN and deploys a channel-wise attention mechanism to adaptively enhance the valuable information carried in the simulation data. The attention mechanism can not only improve forecasting performance, it can also endow the model with interpretability for the contributions of various inputted parameters. A bidirectional spatiotemporal propagator for encoding the forward and backward trend information of WRF data, LightNet+, was designed by Zhou et al. (2022). Compared with previous single-direction encoders, LightNet+ can fully analyze and utilize the temporal dependencies in the simulation data. Evaluation results show that LightNet+ can improve equitable threat scores by 10% compared to ADSNet.

By using a semantic segmentation deep learning network, Zhou et al. (2021) merged multi-source observation data and high-resolution NWP data and found that it can yield a good 2–6-h lightning forecast. A multiple input and output lightning nowcasting model, Convolutional Long- and Short-Term Memory Lightning Forecast Net (CLSTM-LFN), was constructed by Guo et al. (2022) to improve 0–3-h lightning nowcasting performance.

Lightning data can be introduced into weather models to improve forecasts of severe weather because of the close relationship between lightning and strong convection. Since lightning cannot be directly modeled, it needs to be converted into a variable used in the model.

Some studies have used lightning data to modulate the dynamic parameters of models. Chen et al. (2020d) did experiments to assimilate lightning data from the LMI aboard the FY-4A geostationary satellite in combination with radar data. The LMI data was found to add benefits to the assimilation of radar data by reducing wind errors and promoting updraft development at locations where lightning was observed. Wang et al. (2020c) designed a lightning data dynamic nudging method to adjust the dynamic field in convective clouds based on the relationship between lightning and vertical velocity. They found that the lightning data dynamic nudging did not significantly change the intensity of the updraft of the investigated squall line relative to the simulation without assimilation; but it did expand the spatial distribution of positive vertical velocity at 700 hPa by approximately 2% and therefore extend the spatial distribution of rainfall greater than 40 mm h^–1. The positive effect could last for 2–3 h. Xiao et al. (2021a, b) assimilated total lightning data in a four-dimensional variational (4DVAR) system for the purpose of very short-term convective forecasting. They used the total lightning rate to modulate the vertical velocity and found that the assimilation improved the model’s dynamical states by enhancing the convergence and updraft in and near the convective system. Furthermore, it was proposed that the simultaneous application of lightning data and radar data in the assimilation could produce better results. Gan et al. (2021) developed a total lightning data assimilation (LDA) scheme at the cloud-resolving scale that assimilates total lightning data through the ensemble square root filter (EnSRF) method based on the relationship of the maximum vertical velocity with the flash rate. The LDA scheme improves forecasting by improving the water vapor field, providing a warm and moist environment, and increasing convergence at low levels and divergence at upper levels.

Lightning data also can be used to modulate the microphysical parameters of models. Chen et al. (2019a) suggested an LDA scheme for nudging water contents modulated by the total lightning rate in the WRF model and tested it for two severe squall line cases in Beijing. The LDA scheme caused the simulated surface cold pool to agree well with the observations, therefore resulting in significantly-improved quantitative precipitation forecasts. Liu et al. (2021e) improved upon forecasting precipitation in leeward slope areas by using lightning data to yield pseudo-water vapor. Zhang et al. (2020c) used the lightning data from the WWLLN to modulate the WRF (Weather Research and Forecasting) model’s relative humidity for Super Typhoon Haiyan in 2013 and found significant improvement in the intensity forecast of the typhoon by assimilating inner-core lightning but only faint improvement in the rainband by assimilating the lightning there. Sun et al. (2022) introduced LMI lightning data to retrieve graupel mixing ratio fields. Their assimilation experiments for two oceanic mesoscale convective systems based on the WRF model via nudging terms exhibited that most of the convective cells missed by the control experiments were recovered.

Chen et al. (2020b) used LMI lightning data to yield a proxy of radar reflectivity and then assimilated the proxy reflectivity in the Rapid-refresh Multi-scale Analysis and Prediction System-Short Term system. They reported that the positive effect of the cycling assimilation of proxy reflectivity on rainfall forecasts is similar to the parallel experiments that directly assimilated radar reflectivity. The application of this method in mountainous cases proved that it can be used as a substitute in the areas where radar data are missing.

Qie et al. (2019) provided a detailed comparison of the development of positive, negative, and recoil leaders. They summarized the propagating characteristics of rare positive recoil leaders through pre-conditioned channels in both tower lightning and rocket-triggered lightning and clarified the intermittent propagation of the positive leader during its the initial and sustained development. Jiang et al. (2020b) investigated the intermittent propagation of UPL and upward negative leaders in rocket-triggered lightning and found an abrupt luminous crown blooming due to a step of UPL. The clustered space leaders resulted in negative channel branching, while the residual structure in positive leaders just resulted in an individual step. In laboratory lightning-like discharges, Huang et al. (2022) observed and analyzed the step of positive leaders led by a separate luminous structure. They found that a streamer-like common zone connects the primary channel with a separate luminous structure, which exhibits bidirectional development.

Yuan et al. (2019) documented the positive leader branching feature that occurs during the propagation of a UPL. Nearby bidirectional leaders with clear asymmetrical channel extensions at opposite ends were observed (Fig. 4). By connecting with the lateral side of the progressing main positive channel, new branches of positive leaders were formed. It is suggested this kind of side discharge, together with head-splitting, are considered as two important branching mechanisms of positive leaders.

Figure 4.  The development of a bidirectional leader: (a) inception, (b–d) bidirectional propagation, (e) connection of the bidirectional leader to the existing positive channel, and (f) after connection. [Reprinted from (Yuan et al., 2019)]

Using synchronous multi-frequency radio sensors, Yuan et al. (2020) observed and analyzed the origin of an uncommon three-stroke PCG flash with different terminations. Their results revealed that the downward positive leader preceding the positive return strokes possibly developed either from different decayed leaders or from the opposite end of a developing in-cloud negative leader. For active horizontal negative channels, Yuan et al. (2021a) found two kinds of side breakdowns from these channels during a PCG flash. They suggested that the inception of the positive leader is independent of the disconnection or current cutoff on the negatively charged channel.

Wu et al. (2021) presented a detailed analysis of the formation of an in-cloud channel branch that extended toward the ground and finally developed into a PCG return stroke. The initiation and propagation characteristics of four vertical bidirectional leaders that made connection to the previously formed aloft horizontal channel were analyzed. It was found that the connection of the upper negative end of the bidirectional leaders to the aloft horizontal channel led to the abrupt elongation of the lower positive end.

In a recent study, Wu et al. (2022) reported the optical characteristic of needles in a PCG flash for the first time (Fig. 5). It was found that all the needles in this case were observed during the later stage and continuing current period of the return stroke. These needles initiated near the previously formed predominantly horizontal channel and extended almost perpendicular to the horizontal channel. Flickering events, which are recoil-type streamers (or leaders) retracing the needle-created channels show repetitive features. Wu et al. (2022) suggested that the needles in this PCG flash are induced by the radial motion of negative charge from the hot core into the surrounding corona sheath of the positive leader channel, when the core is rapidly recharged by the processes of return-stroke and continuing current with its radial electric field reversed.

Figure 5.  (a) Composite image of 30 selected frames obtained by a high-speed video camera operating at 20 000 frames per second. (b) Composite image of 400 selected frames obtained by a high-speed video camera operating at 50 000 frames per second showing needles. [Adapted from (Wu et al., 2022)]

Jiang et al. (2022) documented a unique intracloud lightning flash with abundant positive leader branches extending from one position. The recoil leaders’ activation and the promotion effect on the breakdown of negative-end were discussed. They found that the positive leaders, which were accompanied by frequent recoil leaders, exhibited a noticeable transition signature from the stable and smooth extension stage to the active developing stage. And the back-and-forth interaction from the opposite ends was associated with the activation and propagation of long recoil leaders with regular electromagnetic pulse trains.

Considering the limitations of the stochastic lightning parameterization scheme, Zheng et al. (2022) established a new self-sustained electrical neutrality lightning discharge parameterization scheme, which simulated the dynamic aspect of channel development for the first time, including conducting, extinguishing, and reactivation. The new scheme can also simulate the evolution of the channel nonlinear electrical parameters, maintain the complete electrical neutrality of the entire discharge channel, etc., which agrees well with the current knowledge of the lightning discharge process. Therefore, it may become an effective tool for subsequent exploration of discharge phenomena associated with the dynamic development of lightning channels.

Compared with the ground or low objects, the top of ground-based tall objects (such as tall buildings, tall trees, tall towers, windmills, and power transmission towers) is more likely to reach the initiation threshold of the upward leader due to the enhancement distortion of electric fields, which makes tall objects not only easy to be struck by downward lightning, but also able to initiate upward lightning. Tall objects are an important platform for lightning observation and research because of their relatively high probability of lightning occurrence.

In recent years, Chinese researchers have carried out some observation campaigns and studies on tall-object lightning: the LRG of the CAMS and the Guangzhou Institute of Tropical and Marine Meteorology, China Meteorological Administration have established the Tall-Object Lightning Observatory in Guangzhou (TOLOG) and began conducting a comprehensive tall-object lightning observation experiment from 2009 (e.g., Lü et al., 2020); the IAP LRG began observing lightning associated with the 325-m-high meteorological tower in Beijing from 2012 (e.g., Jiang et al., 2021b); and the Shenzhen Meteorological Bureau and Hongkong Polytechnic University have conducted observation of lightning associated with the 356-m-high Shenzhen Meteorological Gradient Tower in the southern coastal area of China (e.g., Gao et al., 2020).

Generally, the lightning attachment process can be separated into two phases: the initiation and propagation of one or more upward leaders from grounded objects toward the downward propagating leaders and the so-called breakthrough phase (BTP, also known as the final jump) (Rakov and Tran, 2019). Because the attachment process determines the location of the lightning grounding point (lightning return stroke position) and the area impacted by the strike, it is a crucial aspect of lightning physics and lightning protection to understand. For the lightning attachment process, tall objects can play the role of “Magnifying Glass” (Lü et al., 2020).

By using high-speed video cameras operating at 20 kilo frames per second (kfps) and 525 kfps, Qi et al. (2019) documented video recordings with the highest framing rate to-date of the attachment process in a natural CG flash. Several important parameters, including the average inter-step interval, step length, and 2D propagation speed of the downward negative leader (DNL) and that of the UPL were statistically analyzed. It is interesting to note that the last step of the DNL contacting the UCL was captured by the high-speed video camera operating at 525 kfps. As shown in Fig. 6, the 2D length of the final imaged gap between the DNL tip and UCL tip analyzed in this case was estimated to be about 13 m.

Figure 6.  Sequential high-speed video camera (framing rate: 525 kfps) images of a natural negative CG flash [Reprinted from (Qi et al., 2019)]. (UL: Upward Leader; PDL: Primary branch of Downward Leader; SDL: Secondary branch of Downward Leader).

Jiang et al. (2021b) investigated the fine structure of the breakthrough phase in a natural lightning flash striking the Beijing 325-m meteorology tower. The high-speed video images with a framing rate of 380 kfps recorded two consecutive frames for the BTP (Fig. 7). From these high-rate frames, they found a new route scenario during the BTP, which is the conversion from the previously high-impedance common streamer zone (CSZ) to a hot plasma channel. The route appeared when the bright tips of the DNL and UCL were at least 23 m apart. A space leader-like luminous segment was captured in the route.

Figure 7.  High-speed (380 kfps) video recordings of the attachment process in a natural CG flash show the formation of the route and the leader-like luminous segment in the route. (a)–(e) show the full view and the lower panel shows an expanded view of the red rectangle area in (a)–(c). (ax), (bx), and (cx) are grey and (ay), (by), and (cy) are pseudo-color.[Adapted from (Jiang et al., 2021b)]

Srivastava et al. (2019) reported the intermittent propagation of a UCL connecting a DNL in a negative CG strike to the Beijing 325-m meteorology tower. The DNL and UCL exhibited different intermittent features with temporally inconsistent steps, suggesting independent stepping development. However, they also suggest that when the DNL and UCL are very close in space, the DNL can induce and support the development of the UCL.

Qi et al. (2020, 2021) analyzed the high-speed video recordings of CG flashes striking a cluster of tall buildings (height of 100 m to 600 m) in the TOLOG field of view. The results show that the height and the top geometry of the tall objects are two essential factors that affect the 2D striking distances (SD) of the first return stroke. Combining the results with the peak current data of return strokes provided by LLS, they also indicated that the correlation between the SD and the LLS-inferred peak current appeared to be poor.

Using high-speed video observations and BLNET data from the summer seasons of 2012–20 in Beijing, Yuan et al. (2021b) analyzed the detailed characteristics of a total of 25 tower-initiated upward flashes. Twenty-one of the upward flashes were triggered by nearby PCG. They suggested that the positive change of the local electric field or continuing current period during the return stroke may favor the inception of the UPL.

Wu et al. (2019a, b) investigated the upward lightning flashes triggered from tall objects by two single-stroke PCG flashes with high peak current (+310 kA and +141 kA, respectively), which were usually followed by long continuing currents. The results show that the initiation of upward flashes triggered from tall objects by the high peak-current PCGs could be caused by the combined effects of the PCG return stroke, its associated continuing current, and the in-cloud K processes.

Many studies show that upward lightning is usually triggered by a downward PCG. However, Jiang et al. (2021a) reported the first documented downward PCG initiated by upward negative lightning, which demonstrated a new scenario for the initiation of PCG. The optical and electric field change data measured at TOLOG show that an upward lightning (negative) was first triggered from the tip of the 600-m-high Canton Tower by a distant PCG, and then the upward lightning served to facilitate the intracloud discharges that supplied positive charges for a downward PCG.

Using a highly sensitive magnetic field (B-field) antenna, Fan et al. (2021b) examined the electromagnetic characteristics of a UPL that ascended from the Canton Tower. It was found that before the inception of sustained UPL, electric field (E-field) and B-field pulses with small amplitudes are superposed on the E-field and B-field changes. The time scale of pulses and the inter-pulse interval are similar to those associated with the precursors of rocket-triggered lightning. However, the UPL of the upward lightning initiated from the Canton Tower extended significantly in the first several milliseconds, with the initial average 2D velocity being one order of magnitude faster than that of rocket-triggered lightning, indicating that the initiation and development of a UPL triggered from a tall object benefits from the substantial E-field enhancement caused by nearby discharges.

The upward leaders in upward flashes are often followed by an initial continuous current (ICC), which typically lasts for a few hundred milliseconds. When the channels of a UPL or return stroke decay, recoil leaders (RLs) may initiate in or near the decayed channel. RLs usually develop inside the cloud, which makes the optical observation of RLs challenging, and thus the initiation and development of RL are still unclear. Wu et al. (2020) presented a detailed study of RLs in an upward flash. The initial position and propagation of RLs were analyzed, and it was found that each dart/dart-stepped leader initiated near the positive extremity of the preceding bidirectional attempted leader. Based on 20-kfps high-speed video recordings, Wu et al. (2021) further analyzed the abrupt elongation phenomenon of the positive end of the bidirectional leader. It was found that the positive end of the dart/dart-stepped leader extended abruptly via connection between the floating channel and the positive channel tip.

Tall objects play an “Amplifier” role on the lightning electromagnetic field (Lü et al., 2020): the higher the tall object, the larger the LLS-inferred peak current of strokes recorded in the vicinity of the tall object; numerical simulation of the tall object on the electromagnetic field of the lightning return stroke also shows that the higher the tall object is, the more significant the enhancement effect is.

Huang et al. (2021b) analyzed the waveform characteristics of optical pulses from a total of 66 subsequent return strokes in 28 upward lightning flashes. They found that the optical pulse of the subsequent return stroke in upward lightning has the shortest values, and the half-peak width of the optical pulse of the subsequent return stroke in upward lightning is lower than that of the subsequent return stroke in downward lightning.

Using a Rogowski coil installed at a height of 492 m on the 600-m Canton Tower, Chen et al. (2022a) collected lightning current direct measurements of tower-initiated flashes on the Canton Tower for the first time (Fig. 8). They analyzed the current waveforms of three single-stroke downward flashes and two multiple-stroke upward flashes, as well as the luminosity of the corresponding return strokes. For subsequent strokes in the same upward flash, the maximum peak current was found to be proportional to the square of the initial peak luminosity. Note that all five flashes analyzed by Chen et al. (2022a) were negative polarity. Based on a comparison of the directly measured and LLS-inferred peak currents of the strokes, it is indicated that LLS-inferred peak currents of subsequent return strokes are overestimated compared to the first return stroke.

Figure 8.  Diagram for the placement of the Rogowski coil installed on the Canton Tower (a) and current waveforms of the return strokes in an upward lightning flash (b). [Reprinted from (Chen et al., 2022a), with permission from Elsevier]

Comparing to the lower objects, tall objects could be struck by lightning (either downward and upward lightning) with higher probability, so tall objects are “Hot Spots” of lightning flashes (Lü et al., 2020). Based on TOLOG optical data from 2009 to 2014, Wu et al. (2019c) analyzed the distribution characteristics of 119 downward CG flashes in the northwest of Canton Tower within 3 km. The results show that 44% (52/119) of downward CG flashes occurred on the four tallest buildings in the area. Except for the Canton Tower flashes, the attraction that downward CG flashes experience toward Canton Tower makes it that downward CG flashes rarely occur within the vicinity of l km.

The observation area of TOLOG can be used as a “Calibration Field” for LLS (Lü et al., 2020). By using data from Guangdong–Hongkong–Macao LLS (GHMLLS) for the period 2014–18, Chen et al. (2020a) investigated the lightning “Hot Spot” caused by the 600-m-high Canton Tower, the 390-m-high CITIC Plaza, and the 308-m-high GF Securities Head-quarters, which are all relatively isolated in Guangzhou city. The number of strokes within a 200-m radius of the Canton Tower lightning “Hot Spot” was found to be about four times higher than that of the other two tall objects. It is speculated that upward lightning flashes account for the majority of Canton Tower flashes, while those flashes hitting the other two tall objects are mainly downward type.

Recently, the lightning location records of GHMLLS near several supertall structures in Shenzhen were analyzed by Zhang et al. (2022b). It was found that the higher the tall-object height is, the more lightning flashes that are recorded nearby, and the higher the proportion of records classified as in-cloud lightning is, which implies that higher objects are more likely to suffer more lightning strokes, especially subsequent strokes contained in upward lightning flashes.

Modeling the electrification and discharge process is another effective way to examine the effects of tall objects on lightning processes. Jiang et al. (2020a) presented an improved stochastic model to simulate the development of lightning leaders based on optical observations. The propagation of DNL and UCL, their attachment process, and the distribution of lightning strike points under different tall-object situations were analyzed. The results show that as the tall-object height increases, the relative strike frequency of lightning to the object increases with a decreasing rate. They also suggest that tall objects may cause lightning flashes to strike themselves and may attract nearby flashes to strike the ground close to the tall objects.

Employing comprehensive measurements collected from rocket-triggered lightning, subsequent leaders with intensive chaotic pulse trains (CPTs) were analyzed in detail and compared with normal dart leaders and dart-stepped leaders occurring in one flash by Pu et al. (2019). They found that the CPT leaders were accompanied by energetic radiation and emitted much stronger radiation at broadband radio frequencies and optical bands than normal leaders. The results suggested that the large product of return stroke charge and the leader velocity could be a reliable proxy/indicator for CPT.

Li et al. (2019) examined the underground magnetic field data near the channel of triggered lightning. By comparing the microsecond-scale magnetic pulses during the upward leader as simultaneously detected in the subsurface space and at 2-m soil depth, the magnetic signal was found to be modified by the soil medium, with a typical attenuation of more than 55% and a pulse peak delay of about 0.6 μs. The component of the magnetic field with relatively high frequency would attenuate faster than that with relatively low frequency.

From an altitude-triggered lightning flash striking a 30-m tower during the SHandong Triggering Lightning Experiment (SHATLE), Liu et al. (2020b) observed the attachment process with a circuitous “S” shaped route between the lower extremity of the triggering wire and the upper tip of the tower. They found that the DNL and the tower tip “missed” each other, but then both turned horizontal and eventually connected during the break through phase, with a total channel length of 25.3 m, close to a factor of two times the direct distance (13 m) between the wire tip and tower tip. It is interesting to note that the first eight return strokes of the lightning flash developed through the “S” shaped channel, while the last one altered the attachment route by directly bridging the gap between the wire tip and the tower tip, which may be caused by the long inter-stroke interval facilitating the channel cooling and the conductivity decreasing (Fig. 9).

Figure 9.  (left) Still photograph of the circuitously attached discharge to the tower, (a)–(h) high-speed video frames of the leader behaviors for the attachment, and (i)–(k) simulated electrostatic field at three instants for initiation and attachment of leaders. [Adapted from (Liu et al., 2020b), with permission from Elsevier]

Ma et al. (2020) investigated the characteristics and mechanisms of ICC pulses and M-components in rocket-triggered lightning. In two special cases, they found that multiple current surges were superimposed on the extremely long-lasting ICC and continuing current (CC) process. These current surges can maintain the grounding channel and extend the branches of the in-cloud channel, thus lengthening the time scale of the ICC/CC process, which may reduce the possibility of a return stroke thereafter. By checking the in-cloud discharge behaviors based on the VHF radiation source mapping, ground E-field changes, and the simultaneous current data, they also identified a new scenario leading to an M-component that involves a short-term interruption (or weakening of breakdown) of the upward return stroke wave and then reactivation promoted by the residual charge.

During the SHATLE field experiments in 2014 and 2015, and the Guangdong Comprehensive Observation Experiment on Lightning Discharge in 2018, Fan et al. (2019) documented a kind of magnetic pulse burst (MPB) during the ICC in rocket-triggered negative lightning. MPB sources could originate from the breakdown near the leader tip and have an average peak current on the order of kA. B-field measurements at the nearby site exhibit slow variations with small MPBs superposing on them. They suggest that the appearance of MPBs may indicate the enhancement of both the charge transfer and the channel brightness.

An altitude-triggered lightning flash was observed during the Guangdong Comprehensive Observation Experiment on Lightning Discharge in 2021. Fan et al. (2022) documented the low-medium frequency B-field measurements associated with an altitude-triggered lightning flash for the first time. The propagation of UPL during the inception of attempted DNL is revealed by B-field signals with microsecond resolution. The results show that the bidirectional leaders develop in an uncoordinated manner, and the measurements indicate that the stepping of DNL was much more frequent than that of UPL during the stage of sustained propagation.

Tang et al. (2020) analyzed the characteristics of a triggered bipolar lightning flash with two current polarity reversals. Measurements suggest that the polarity of charge transferred to the ground varies from negative to positive and then reverse to negative (negative-positive-negative). Another example of current polarity reversal during the initial stage of a rocket-triggered lightning flash is documented by Li et al. (2021a).

Using observations from CINTF, Chen et al. (2022c) investigated the development of UPL in its initial stage and discussed its possible mechanisms. A kind of precursor current pulse (PCP), which is generated by a weak positive upward breakdown and a subsequent strong negative downward breakdown, was analyzed in detail. They found that the self-sustaining development of the UPL usually associated with the PCP cluster and the initial PCP cluster, which will stop immediately when the PCP cluster disappears. Li et al. (2021d) examined the characteristics of upward negative precursors in positive rocket-triggered lightning. They suggest that the precursors were actually the un-sustained development of upward leaders, which produced the initial channel segment at the wire tip and eventually extinguished.

Spectral observations can record the emission spectrum of the lightning channel and can be used to retrieve physical information from inside the channel. Lightning spectral measurements with high spatial and temporal resolution have been used to reveal the details of lightning channels (Zhang et al., 2019a, 2020b; An et al., 2019, 2021a, b; Wang et al., 2019c, 2021d, 2022b, c; Liu et al., 2019a, b, 2021d, 2022; Wan et al., 2021; Chen et al., 2022b).

Zhang et al. (2019a) investigated the variation characteristics of return stroke current along the discharge channel, and three types of variation trends with height were found. The attenuation factor is roughly positively correlated with the intensity of discharge. They also suggest that the current tends to decay exponentially during the strong discharge process.

Using a high-speed slitless spectrograph, Liu et al. (2019a) analyzed the channel core of a natural negative CG, confirmed the lightning channel corona sheath model, and suggested that ionic lines are mainly radiated from the channel core. An et al. (2021b) calculated the corona sheath radius of the lightning return stroke channel and investigated the distribution of electrical conductivity along the channel radial direction. The relationship between the radius and electrical conductivity of the channel was investigated. It was found that the electrical conductivity decays steeply to about 500 S m^–1 and then more slowly as the radius increases. Positive correlations between the radius of the corona sheath and the quantity of leader charge and the diameter of the leader channel were also found. In other studies, An et al. (2019) and Wang et al. (2022c) analyzed the temperature distribution along the channel radial direction and the relationship between the initial radius and the discharge parameters.

An accurate spectral diagnosis is important in revealing the physical mechanisms of the discharge processes. The atmospheric attenuation, grating efficiency, and optical response of the camera can affect the lightning spectrum observation. Instead of the traditional Boltzmann plot method for calculating the lightning channel temperature, Liu et al. (2019a, b) used the Saha-Boltzmann plot method to determine the temperature of the return stroke channel. Wan et al. (2021) investigated the effects of atmospheric attenuation on the measured spectrum of lightning channels and calculated the corrected temperature of the channel. The result showed that the corrected channel temperature of the return stroke (calculated by the ionic lines) could be 10 000 K higher than the uncorrected value. They suggest that the temperature of the lightning channel calculated by the atomic lines was barely affected by the atmospheric attenuation. It should be noted that the total intensity of ionic lines and the electrical conductivity could increase with height, as indicated by An et al. (2021a) from the investigation of two downward PCG flashes. This also suggests that the current intensity of PCG increases with height.

Based on the TOLOG spectra data of a lightning flash striking the 600-m-high Canton Tower, Wang et al. (2022b) presented an experimental verification of the opacity thickness of lightning in the near-infrared spectrum for the first time. The results show that singly ionized radiation in the visible wavelength range apparently persists for approximately 320–400 µs, and it coexists with neutral radiation for approximately 240–320 µs, which means that the lightning channel exhibits a hot part radiating ionized lines and a cold part radiating neutral lines at the same time.

The properties of the lightning leader could influence the physical processes of the return stroke. Chen et al. (2022b) analyzed the effects of leader charge on the current intensity and spectral property of return strokes. They found that the return stroke current intensity mainly depends on the total charges deposited during the previous leader process. The radius of channel core is closely related to the current intensity, which can be reflected by the intensity of the singly ionized lines in the spectrum. Zhang et al. (2020b) also documented the electrical conductivity and the transmission properties of lightning leaders. They found that the 2D speed of the leader is proportionate to the corresponding conductivity. Meanwhile, the value of the 2D speed could be significantly affected by the propagation direction of the leader tip.

Terrestrial Gamma-ray Flashes (TGFs) are high-energy emissions with photon energy up to tens of MeV that were discovered firstly by satellite-based and then by ground-based gamma-ray detectors. Since their discovery around the 1990s, TGFs have attracted substantial attention in the lightning community, although studies from China are still limited. Details on spaceborne and ground-based observation platforms and other worldwide TGF-related studies can be found in the recent review papers by Lu et al. (2020) and Lyu et al. (2023). It is suggested that TGFs are a natural high-energy phenomenon associated with lightning discharges that occur frequently during thunderstorms. However, TGF production mechanisms, their associated processes, and their effects on human activity are still unclear. It is found that gamma rays in thunderstorms propagate in either an upward (usually detected by satellite-based detectors) or downward (usually detected by ground-based detectors) direction.

Upward TGFs are usually generated within milliseconds of the initiation of upward negative leaders, and they may produce a kind of distinct radio emission because of the generation and propagation of a large number of high-energy electrons. Satellite-based gamma-ray detectors play a substantial role in the detection and observation of upward TGFs. The Insight-Hard X-ray Modulation Telescope (Insight-HXMT) (Zhang et al., 2020a) and Gravitational wave high-energy Electromagnetic Counterpart All-sky Monitor (GECAM) (Lv et al., 2018) are two platforms currently operating in China that can monitor high energy photons. Xian et al. (2021) analyzed the thermal structure of the environmental atmosphere associated with TGFs. A higher tropopause altitude was found in the TGF-related storms compared with the climatology values, and the land–ocean difference in the thermal structure of TGF-producing storms for subtropical and tropical thunderstorms is slightly different. Zhang et al. (2020a, 2021a) examined the low-frequency (LF) emissions associated with TGFs detected by Fermi and Insight-HXMT. They found that most TGFs were produced during the mature stage of the strong convection, but not in the strongest convection regions (Zhang et al., 2021a). The occurrence contexts of TGFs relative to the production of narrow bipolar events (NBEs) were analyzed, and TGFs that occurred both before and after lightning initiation were found, which may support different TGF production mechanisms (Zhang et al., 2021a).

By investigating five years of satellite-based gamma-ray observations and ground-based radio measurements, Lyu et al. (2021) demonstrated the close connection between upward TGFs and a distinct type of in-cloud discharge, which they named energetic in-cloud pulses (EIPs). The identification of TGFs from the independent search of positive EIPS (+EIPs) implies a high-to-certain probability of almost all EIPs being TGFs. The work by Lyu et al. (2021) not only provides strong evidence of the connection between +EIPs and TGFs, but also experimentally demonstrates the detection of a subset of TGFs from radio signals alone.

Downward TGFs observed by ground-based gamma-ray detectors are found to be associated with several different types of lightning processes, such as DNLs or UPLs, the initial continuing current stage of rocket-triggered lightning flashes, and return stroke processes. Li et al. (2019) used an NaI (T1) detector and found that photons with low energy (X-ray) were associated with the development of stepped leaders of triggered lightning. Zhang et al. (2023) designed a distributed Thunderstorm Energetic Radiation Observation System (TEROS) to detect high-energy photons up to tens of MeV and found that energetic radiations were observed during the leader phases of five rocket-triggered lightning experiments. Compared to upward TGFs, fewer downward TGFs have been observed all over the world. With the recently designed TEROS, more details about the production of downward TGFs will become known as research in China progresses.

Based on ground-based radio measurements as well as satellite-based and ground-based gamma-ray detector observations, the studies focusing on TGF production, the relationship between gamma ray production and the occurrence of lightning discharges as well as thunderstorm circumstances have recently become areas of focus in the fields of atmospheric electricity and high energy atmospheric physics. We expect that a benefit of the development of state-of-art instruments of high temporal and spatial resolutions combined with satellite-based and ground-based measurements will be new insights into the processes and mechanisms of TGFs.

Yang et al. (2018a) analyzed the parent CGs of positive and negative sprites observed in different thunderstorm regions with various lightning activity and thunderstorm structures and discussed the different conditions needed for inducing different polarity sprites. They found that negative sprites are associated with intense and deep convection, while positive sprites are mostly associated with stratiform regions without strong convection. Pan et al. (2021) documented in detail the characteristics and lightning activity of an MCS that produced only one sprite event and found that sprites tend to be produced early in the maturity-to-dissipation stage of MCSs, with an increasing percentage of +CG to total CG. By investigating more than 1000 halo/sprite producing lightning strokes during hurricane Matthew, Huang et al. (2021a) found that the halo/sprite production during Matthew could be separated into two regions: the inner core, which is favorable for negative halos; and the outer rainbands, which are productive for both positive and negative halo/sprite strokes. Their results also suggested a comparable feature between the outer rainbands of Matthew and the trailing stratiform region of continental MCSs.

Wang et al. (2020e) analyzed the occurrence of sprites over the southern foot of the TP by comparing the lightning detection data from WWLLN and the observations from Imager of Sprites and Upper Atmospheric Lightning (ISUAL) aboard the FORMOSAT-2 satellite. The location accuracy of ISUAL for sprites and that of WWLLN for lightning discharges in this region are examined for a total of 17 sprites. Based on the characteristics of parent lightning strokes and the meteorological context of thunderstorms derived from the FY-2 cloud-top brightness temperature data, they found that not only MCSs, but also small-scale convective systems, can produce sprites over the southern TP area.

Using the ISUAL and WWLLN data, Wang et al. (2021f) documented an extremely rare case of a negative sprite observed near the northern border of Bogotá, Colombia. A distinct “sprite current” feature was identified during the observed sprite-producing negative CG stroke. From the charge transfer analysis, they found that the sprite and the intense sprite current might have been produced by and linked to the extraordinarily long charge transfer time after the parent negative CG stroke and the plasma irregularities in the mesosphere.

Lu et al. (2018) investigated the possible relationships between halos and sprites and confirmed the association of positive halos and sprites, as well as the connection between negative sprites and halos. Statistical analysis shows that the majority of ISUAL-observed halos are pure events without any discernible streamer development. They suggest that since streamer development may depend on the impulse charge transfer of a particular time scale, many negative CG strokes may only produce halos even though the associated impulse charge moment changes exceed the threshold for sprite production.

In general, gigantic jets (GJs) are usually produced over strong tropical or tropical-like thunderstorms with convective surges or overshooting tops. In previous studies, it has been suggested that few non-GJ type TLEs were observed during the lifetime of GJ-producing summer thunderstorms. However, Yang et al. (2018b) found that MCSs provide favorable conditions not only for GJs but also for sprites. He et al. (2019) documented the first observed positive polarity GJ, which was recorded near the Yellow Sea over the Chinese mainland during a storm dominated by negative CG flashes. It was produced in a thunderstorm context consistent with a typical summer thunderstorm, and during a CG-increasing period when overshooting appears. Yang et al. (2020) analyzed a GJ in southern China in terms of its morphology, meteorology, storm evolution, lightning, and narrow bipolar events. It was found that the GJ initiated in the region with the coldest cloud top brightness temperature and near the strong convection region. Three NBEs were generated within 30 s before and after the GJ. Based on the observations from ISUAL, magnetic field radio measurements, and lightning locations from WWLLN, Wang et al. (2021e) documented six negative red sprites produced in a tropical thunderstorm that later evolved into Hurricane Harvey (2017), and analyzed the parent stroke characteristics. They found that tropical marine meteorological systems are more likely to be the main production convection systems of negative sprites. It is interesting to note that another 18 GJs were also observed during the sprite-producing storm. They also indicated that the thundercloud charge structures of oceanic sprite-producing thunderstorms could be different from those of continental storms, as revealed from the meteorological context of storms producing both GJs and sprites.

The luminous features of NBEs remain unclear, although radio emissions reflecting their existence have been reported for decades. NBEs have been thought to be relatively “dark” (nonluminous) compared to other normal lightning discharges. Based on the spectral measurements of the Atmosphere-Space Interactions Monitor (ASIM) on the International Space Station and simultaneous radio observations from ground-based radio sensors, the luminous signatures of nine negative and three positive NBEs were documented by Liu et al. (2021b). They found that NBEs of both positive and negative polarities are associated with 337-nm optical emissions, but with weak or no detectable emissions at 777.4 nm, which suggests that NBEs are associated with streamer breakdown (Fig. 10a). For negative NBEs, the optical strength is almost linearly correlated with the peak current of the associated NBEs (Fig. 10b). These findings provide new insights into the means of measuring the occurrences and strength of cloud-top discharges near the tropopause.

Figure 10.  (a) Comparison between the ASIM blue emissions and the VLF/LF sferic signal for a negative NBE. (b) Correlations between the 337-nm peak brightness of blue emission and peak current of NBEs. [Adapted from (Liu et al., 2021b)].

Liu et al. (2021c) investigated the meteorological and electrical conditions of two midlatitude thunderstorms producing thirteen blue discharges observed simultaneously by ISUAL and a ground-based lightning detection array. They found that blue discharges tend to cluster within a bounded area near the convective surge area with low cloud top temperatures (~195 K). Based on the outbreak of negative NBEs at altitudes between 16 km and 18 km, they suggest the existence of a strong and high upper positive charge layer due to the intense convective updraft and the production of upward positive blue discharges because of the favorable charge structures.

Studies show that lighting activity during thunderstorms enables electric coupling effects with the lower ionosphere. By investigating the total electron content (TEC) and its direction of propagation associated with thunderstorms using the method of polynomial filtering, Ogunsua et al. (2020) documented evidence of deviations in the ionospheric. They found that the internal dynamics of the equatorial ionosphere were suppressed by large thunderstorm effects during the daytime. Furthermore, single a lightning stroke can have an impact on both the daytime and nighttime lower ionosphere. By investigating the skywave features of lightning sferics versus the lightning source peak currents, Qin et al. (2020) found a direct and fast nonlinear coupling between lightning-generated electromagnetic fields and the lower ionosphere. The signature of radio sferics can also be used to retrieve the properties of the ionosphere. Qin et al. (2019) introduced a modified ray-theory based model and its comparison with a finite difference time domain model for lighting sferic transmission in the earth–ionosphere waveguide. With this new model, a quantitatively better performance was achieved, and the lightning sferics could be modeled in frequencies down to 3 kHz, 5 kHz, and 7 kHz for distances of up to 500 km, 800 km, and 1000 km, respectively.

Combining the lightning data from BLNET and PM2.5 data from 35 automatic air-monitoring stations in Beijing, Sun et al. (2020) investigated the potential effects of aerosol on lightning activity in this metropolitan region. It was found that the peak lightning activity under relatively polluted conditions occurred four hours later than that under relatively clean conditions, and the total flashes during the peak lightning activity period was doubled. The detailed relationship between the lightning flashes and the content of PM2.5 was analyzed. A significant positive correlation when PM2.5 was lower than 130 μg m^–3, and a negative correlation when PM2.5 exceeded 150 μg m^–3 was found. However, the relationship was very weak when PM2.5 ranged between 130 μg m^–3 and 150 μg m^–3.

To further investigate the effects of aerosols on electrification and lightning activity in the metropolitan Beijing area, Sun et al. (2021) conducted a WRF Model simulation of a multicell thunderstorm with a two-moment bulk microphysical scheme and bulk lightning model. The results show that fewer lightning discharges are produced in the continental regions where aerosol concentrations are low because of reduced latent heat in the upper levels being released, weaker updraft speed, smaller ice particle concentrations, and lower charging rates. However, lightning activity is significantly enhanced during the developing and mature stages under polluted conditions.

Although significant advancements in the study of atmospheric electricity in China have been achieved in recent years, some interesting and important questions still require further research: (1) the mechanism of electrification and charge structure formation of natural thunderstorms; (2) the initiation mechanism of natural lightning discharge; (3) the application of precise lightning location information in lightning parameterization and severe weather forecasting; (4) the detection, inversion, and application techniques of discharge parameters for total lightning; (5) the mechanism of TGFs; (6) the mechanism of leader propagation and attachment to elevated objects on the ground and to the flat ground; (7) the application of artificial intelligence or other new techniques to lightning detection, analysis, and forecasting.

Acknowledgements. This research was supported by the National Key Research and Development Program of China (Grant No. 2017YFC1501500). The authors would like to express special thanks to Prof. Xiushu QIE, Dr. Bin WU, Dr. Qi QI, Dr. Wenjuan ZHANG, Dr. Fei WANG, Dr. Liangtao XU, and Dr. Tianxue ZHENG for preparing this paper.