-
The China automatic CG monitoring system began operation in the 1980s. Initially, it mainly imported foreign CG location equipment (Chen et al., 2008). By 1990, a domestic CG location system had been put into operation (Chen et al., 2008). The domestic CG location system is mainly used for local CG monitoring and has not yet covered the whole country. It is used in the fields of meteorology, electric power, telecommunications, civil aviation, and military. From 2009 to 2017, the number of sensors managed by the China Meteorological Administration (CMA) increased from 265 to 406, with the largest increases occurring in Qinghai, Inner Mongolia, Tibet, and Xinjiang, where 24, 29, 24, and 43 stations were added, respectively (Fig. 1) (CMA, 2010, 2018). It should be noted that the Xinjiang stations were installed in 2013, while most of the Qinghai and Inner Mongolia stations were installed in 2016. In addition, as seen in Fig. 2 [a redrawn figure from Zhu (2016)], the sensors cover a large portion of China; however, there are more sensors in the east than the west, and more in the south than the north. As soon as a regional lightning network was established, the CMA began to digitize CG lightning strokes on a daily basis (Xia et al., 2015) and provide access to the data.
Figure 2. The site locations of CG detection sensors in China [the stations established before 2008 are blue and those after 2009 are red. The original figure, which is from Zhu (2016), has been redrawn in this paper].
The ground-based Advanced Time of Arrival and Direction (ADTD) system, used in the CMA CG monitoring network, was developed by the National Space Science Center, the Chinese Academy of Sciences. This system uses the Time of Arrival (TOA) and the Improved Performance and Combined Technology (IMPACT) methods to locate the CG flashes. The CG flashes are identified by the waveform characteristics. Different lightning types can be identified by distinguishing these characteristic parameters, like the rise time of the wave head, zero crossing time, unipolar, peak, and overshoot ratio. Influenced by the network’s settings and layouts, there is a possibility of misclassified flashes, primarily in terms of cloud flashes being called small positive CG strokes. Studies have shown that most (~90%) positive small flashes (<10 kA) are IC flashes and most (~90%) large positive flashes (>10 kA) are CG flashes. The misclassification rate for negative flashes was very small (Diendorfer, 2007). The detection range of a single station is 300 km, its detection efficiency can reach 80%–90%, and the locating accuracy is 500 m (Wang and Chen, 2015). It should be mentioned that the location accuracy and detection efficiency of this network has not been systematically evaluated due to the lack of adequate ground truth information (Xia et al., 2015). The average operation rate for a single station was approximately 89% in 2009 and over 95% in 2017 (CMA, 2010, 2018). Each station provides reports on the time of the original return stroke (RS), locations (longitude and latitude), peak current, polarity, current maximum steepness (current maximum rate of rise), location method and provinces, cities, and counties.
-
Due to the strict data management of the CMA, our data access was limited, especially for recent years. Thus, only the CG flashes occurring from 1 January 2014 to 31 December 2018 in the mainland of China were considered. Data before 2014 were not used, not only because of limited access, but also because there were relatively fewer sites before 2014. RS data were grouped using the criterion that a CG flash, which may consist of a series of separate strokes that occur for 1-s time periods within 10 km of the first stroke detected. The time interval of each RS involved in a flash is less than 0.5 s (Cummins et al., 1998; Zheng et al., 2016). The position of a CG flash is the average value of its RSs’ positions. To eliminate any possible IC contamination, positive RSs with currents less than 10 kA have been removed (Zheng et al., 2016).
The grouped CG data were then binned into grids of 0.1° × 0.1° (longitude × latitude), ranging from 73°E to 137°E and 3°N to 54°N, which includes land and coastal sea regions of China and areas of other countries. However, only return strokes located over land were included in the gridded dataset because sensors are located over land and the detection efficiency of flashes, especially those with moderate or smaller peak currents, falls off rapidly with distance.
-
Theoretically, the amount of LNO can be estimated if the number of NO molecules per lightning flash is known. However, the data used in this paper only includes CG flashes, without IC flashes. Therefore, to obtain the total LNO yield, we need the number of NO molecules for each CG flash and average per IC flash, and the assumed IC frequencies. In Fig. 3, the flowchart is used to show major steps, substeps, and key assumptions. It should be noted that the amount of LNO was estimated according to the method described in section 3.2. The process to estimate LNOx from LNO can be seen in section 4.3. Details are as follows.
Figure 3. A flowchart for calculating the total NOx of total lightning flashes (
${I}_{\mathrm{r}}$ is the peak current of the RS. QF, QS are the charge amount of a first and subsequent RS, respectively. QSj is the charge amount of the$ j $ -th subsequent RS. QIC is the average charge amount per IC flash. NCG, NIC are the number of NCG, IC lightning flashes, respectively. LNOCG, LNOIC are the total amounts of NO for NCG, IC lightning flashes, respectively. LNOx is the total amount of NOx for total lightning flashes).The number of NO molecules for each CG flash, or average per IC flash, is determined by
$ \rho $ no and the energy per lightning flash (EF). For$ \rho $ no, the theoretical calculation value is (3–16) × 1016 J–1, the laboratory measurement is (2–17) × 1016 J–1, and the field measurement is (20–30) × 1016 J–1 (Zhang et al., 2002). In this study,$ \rho $ no was taken as the average value of theoretical calculation and laboratory measurement (10 × 1016 J–1), which is the same as that used in other studies (Price et al., 1997; Zhang et al., 2002). The EF of a single CG flash can be estimated from the CG location data. In the RS process of a CG flash, the air in the whole channel is punctured and ionized, and the air breakdown potential (V) is taken as approximately the potential of the whole channel. Assuming that the air molecules in the channel are completely ionized, the EF can be calculated by multiplying the potential V and the deposited charge Q (Zhang et al., 2002).Researchers have adopted different breakdown potential values determined using different methods. To compare the impact of different breakdown potential values on the estimation results, three groups of breakdown potential values were adopted in this study, hereafter referred to as groups V1, V2, and V3. These three groups were estimated based on the following three groups of breakdown electric fields (E).
Griffiths and Phelps (1976a, b) obtained the relationship between the air breakdown electric field threshold and the height by laboratory pulse tests and numerical simulations of lightning initiation in a positive corona streaming process. Stolzenburg et al. (1998a, b, c) then corrected the relationship based on electric field sounding observations. Finally, in this study, the first group of E was calculated by
where z is the altitude (km) and Einit+, Einit− are the breakdown thresholds of the positive and negative electric field (V m–1), respectively.
Considering the theory that lightning may be initiated by high-energy electrons (~1 MeV) continuously produced by the thunderstorm electric field, the runaway breakdown electric field threshold can be represented as (Marshall et al., 1995):
where z is altitude (km),
$ \rho $ is the air density related to height (kg m–3), and Ebe is the runaway breakdown electric field threshold varying with height (kV m–1). Here, we used Eq. (12) to calculate the second group of E.Winn et al. (1974) used a rocket-borne aerial electric field instrument to conduct electric field penetration sounding of thunderstorm clouds in the active phase, and the average value of electric field in thunderstorms was found to be approximately 5 × 104 V m–1. Here, we took this value as the third group of E. It should be noted that thunderstorm clouds are cumulonimbus clouds that produce thunderstorms after reaching a certain intensity.
The height of the negative charge center in a thunderstorm cloud is approximately 5–7 km. So, and 6 km was taken by Price et al. (1997) and also used in this study. The PCG usually originates from the anvil of the thunderstorm cloud, which is approximately 10 km in height (Price et al., 1997); therefore, a value of 10 km was used in this study. Assuming that the CG breakdown voltage is calculated by multiplying the threshold of the breakdown electric field and the initial height of the channel, we obtain three groups of CG breakdown voltage (Table 1).
Lightning type Breakdown voltage (V) CG V1+: 6.85×108 V1-: −9.74×108 V2+: 6.13×108 V2−: −5.93×108 V3+: 5.00×108 V3−: −3.00×108 IC VC: 1.75×108 Notes: Groups V1, V2, and V3 respond three groups of breakdown voltages , respectively. “+” and “−” repond for the polar of breakdown voltages, respectively. Table 1. Breakdown voltage values of CG and IC.
Because the charge deposited by a RS is difficult to measure, one can calculate the charge transferred to the Earth by integrating over the current pulse:
where
$ I\left(t\right) $ is the RS current intensity varying with time. RS currents can be estimated by RS current mode. The analytic expression that best matches the observed values of lightning currents is (Price et al., 1997):where
$ {I}_{\mathrm{r}} $ is the peak current of the RS, e is the natural constant, the first term is the rising part of the current, the second term is the falling part of the current, and the third term is the current tail (Zhang et al., 2002). For the first RS, A = B = 1, C = 0.25, α = 3.3 × 104 s–1, β = 4.5 × 105 s–1, and γ= 8.8 × 102 s–1. For the subsequent RS, A = B = 1, C = 0.25, α= 2.5 × 104 s–1, β = 3.8 × 106 s–1, and γ = 8.8 × 102 s–1 (Price et al., 1997). The RS current can therefore be represented as follows:First RS
where
$ {I}_{0} $ is the peak current of the first RS. Subsequent RSwhere
$ {I}_{i} $ is the peak current of the$ i $ -th subsequent RS. Integrating the time of Eqs. (15) and (16), the charge amount of a first and subsequent RS is approximately as follows:and
For a CG flash with N RSs, its total charge is
To estimate the total LNO, it is necessary to know the parameters of IC. Because the CG location network cannot detect the IC, the IC frequencies were estimated from the CG frequencies and the IC to CG ratio. In this study, the IC/CG ratio was assumed to equal 3, corresponding to Rakov and Uman (2003), Ma et al. (2005), and Boccippio et al. (2001). However, it should be noted that the IC/CG flash rate ratio is influenced by many factors (Soriano and de Pablo, 2007; de Souza et al., 2009; Medici et al., 2017; Bandholnopparat et al., 2020), and it is not constant over a large area.
The channel length of an IC flash is generally 1–6 km (Ogawa and Brook, 1964), and a value of 3.5 km was used in this study. The average electric field inside a thunderstorm used in this study was 5 × 104 V m–1, which was obtained by Winn et al. (1974) via an airborne electric field instrument carried by rockets into thunderstorms. Thus, the breakdown voltage of the IC used in this study was 1.75 × 108 V (Table 1). The charge of a single IC flash was taken as 8.4 C (Brook and Ogawa, 1977). It is the charge transferred during 6 K–changes, not the whole process. So, the average LNO amount per IC flash was assumed to be 14.7 × 1025 molecules flash–1.
-
Overall, lightning density and range distribution were both largest in summer (Fig. 4). While there were more flashes in spring than in autumn, the former covered a smaller area than the latter. There were few lightning flashes in winter. Lightning in the mainland of China mainly occurred in east, central, and south China, followed by parts of northeast, north, and southwest China. The lowest densities were in parts of northwest and southwest China, which is consistent with satellite-based lightning data (Yuan and Qie, 2004; Ma et al., 2005; Cecil et al., 2014). Lightning density is mainly affected by topography, latitude, and land–sea distance (Yuan and Qie, 2004; Ma et al., 2005). Based on the ADTD system, the average annual flash total is approximately 1.19 × 107 flashes in themainland of China. However, based on gridded OTD/LIS data, the average annual flash total is approximately 4.33 × 107 flashes in the mainland of China, which is about 3.6 times the ADTD value. The reason is that OTD and LIS obtain CG and IC flashes, while ADTD only obtains CG flashes.
Figure 4. Seasonal spatial distribution of average CG densities (flashes km–2 month–1) in China from 2014 to 2018 (this study considers March to May as spring, June to August as summer, September to November as autumn, and December to February as winter).
The average peak current of PCG in the mainland of China was 61.6 ± 4.9 kA (Fig. 5), and that of NCG was 41.2 ± 4.7 kA (mean of the absolute values) (Fig. 6). Combined with Fig. 4, the area with the largest PCG and NCG current is not located in the southeastern coastal area with the most frequent CG, but in north, northeast, and southwest China with lower flash rates.
Lightning occurs mainly from May to September, which corresponds to the period of late spring to early autumn (Fig. 7) (Ma et al., 2005; Yadava et al., 2020). This is mainly because the summer monsoon begins to affect China in April and May, and reaches its strongest in August; therefore, August is the strongest month for lightning activities in the mainland of China. From September, the winter monsoon begins to affect the mainland of China from the north, leading to a sharp decrease in lightning activity, which reaches the lowest level in November (Ma et al., 2005). In CG flashes, NCG flashes are dominant, whereas PCG flashes are rare. Moreover, the PCG percentages are higher in cold months (Antonescu and Burcea, 2010; Xia et al., 2015), which is contrary to NCG percentages. In some situations, high PCG flash rates tend to originate outside the largest radar reflectivity and in the regions of the downshear anvil of the thunderstorm (Weiss et al., 2012). Some researchers consider that tilted charge structures account for this phenomenon, that is, the positively charged upper region of the thunderstorm is displaced downshear to the anvil from the negatively charged region due to the strong wind shear, so the negative charge region does not shield the positive charge region overhead from the ground anymore, and PCG flashes can originate from this displaced positive charge region (Wang et al., 2016).
Figure 7. Monthly variation of average lightning densities and average lightning density percentage in China from 2014 to 2018.
The peak current of NCG flashes was smallest in May, and largest in January and December (Fig. 8). On the whole, the average peak current of NCG flashes was largest when the frequency was low. The average peak current of PCG flashes was similar to that of NCG flashes except in July, August, and December. The average peak current of PCG flashes was significantly larger than that of NCG flashes, which is consistent with the results presented in Figs. 5 and 6.
-
In this study, three groups of different breakdown voltages were used to estimate the output of LNO. Due to the similarity in spatial and temporal characteristics of the three groups, only one group is shown. The group of breakdown voltage V3 (V3+, V3-), the same as in Zhou and Qie (2002), was taken as an example to analyze the spatial and temporal characteristics of the estimated LNO. It is convenient for us to compare our results with Zhou and Qie (2002).
In short, the seasonal spatial distribution of LNO (Fig. 9) was basically consistent with that of lightning frequencies (Fig. 4); however, CG, which had a lower frequency than IC, produced more NO than IC.
Figure 9. Seasonal spatial distribution of average LNO production (1014 molecules cm–2 month–1) in China from 2014 to 2018.
Figure 10 shows that LNO is mainly produced from May to September, which is consistent with the main months of lightning occurrence shown in Fig. 7. The NO generated by PCG flashes was only 2.6 × 1014 molecules cm–2 month–1, and the NO generated by NCG flashes was 12.2 × 1014 molecules cm–2 month–1. Even though the peak current of PCG was obviously higher than that of NCG, the frequencies of PCG flashes were small, and most PCG flashes only had one RS. The CG frequencies were smaller than IC frequencies (Fig. 7), whereas the amount of NO produced by CG flashes was larger than that produced by IC flashes (Fig. 10). The average NO produced by CG flashes was 2.2 × 1014 molecules cm–2 month–1 more than IC flashes. This indicates that the energy of a single CG flash is much larger than that of a single IC flash in group V3. In other words, the energy of lightning flashes is more important than lightning frequencies in the estimation of LNOx. If the total amount of LNOx is only the assumed average LNOx of a single lightning flash multiplied by the lightning frequencies, the estimation can cause considerable errors (Beirle et al., 2014; Mecikalski and Carey, 2018).
-
The amount of LNO was estimated according to the method described in section 3.2. To estimate LNOx, the proportion of LNO in LNOx must be obtained. Experiments by Wang et al. (1998) showed that the ratio of NO/NOx produced by spark discharge exceeds 0.9, and the value of 0.9 is used in this study.
In Fig. 11, the maximum breakdown voltages are seen in group V1 and the minimum is seen in group V3, and the NOx production of CG and total lightning varies consistently (maximum in group V1 and minimum in group V3). The NOx of group V1 is 104% more than group V3, while NOx of group V2 is just 46% more than group V3. In group V1, NOx production of CG (77.5%) was more than three times that of IC (22.5%), while in group V3, NOx production of CG was 8.4% more than that of IC. In addition, less NOx comes from PCG flashes, which contribute no more than 10% in all three groups.
The comparison of LNOx estimations for the mainland of China from this study with those from previous studies is shown in Table 2. It should be noted that in the extrapolation method of Guo et al. (2017), the estimate is influenced by the values of the correction factor CF, the ratio of NO2 to NOx in the troposphere fNO2, and the lifetime τ of NO2. CF is used to correct retrieved NO2 VCDs, since the visibility of trace gases in the troposphere depends on the vertical profile, surface albedo, and cloud cover. They all have an uncertainty range, so an uncertainty of one order of magnitude remains in the LNOx estimate. The estimate is 0.07 Tg(N) yr–1 when CF, fNO2, and τ are 1.5, 0.6, and 4 d, respectively. Meanwhile, the minimum estimate is 0.02 Tg(N) yr–1 when the three parameters are 1, 0.8, and 6 d, respectively, and the maximum is 0.27 Tg(N) yr–1 when they are 2, 0.4, and 2 d, respectively.
Studies Annual production
of LNOx in China
[Tg(N) yr–1]The data used The simplified formula for estimating The average LNOx
per lightning flash
(molecules flash–1)Zhou and Qie (2002) 0.38 CG location data T=PCG(FCG+0.1FIC) 67 × 1025 (CG),
6.7 × 1025(IC)Zhou et al. (2003) – NOx analyzer data – 0.12–0.13 × 1025 Sun et al. (2004) 0.016 satellite total lightning data T= PTFT 0.45 × 1025 Zhang et al. (2014) – VHF lightning locating system data – 1.89 × 1025 (CG),
0.42 × 1025(IC)Ju et al. (2015) 0.15 (0.03–0.38) satellite lightning data and satellite NO2VCD data T= PTFT 16.7 (3.4–41.8) × 1025 Guo et al. (2016) – satellite lightning data and satellite NO2VCD data T= PTFT 19.9 × 1025 Guo et al. (2017) 0.07 (0.02–0.27) satellite lightning data and satellite NO2VCD data T= PTFT 20.87 × 1025 This study 0.157–0.321 CG location data T=${\sum }_{{j} }^{ {F}_{\mathrm{C}\mathrm{G} } }{P}_{\mathrm{j} }$+ PIC FIC 85.3 × 1025 (PCG
in group V3)
47.8 × 1025 (NCG
in group V3)
14.7 × 1025 (IC)Notes: T is the total estimated LNOx. PCG, PIC, and PT are the average single LNOx amount of NCG flashes, PCG flashes, and total lightning flashes, respectively. $ {P}_{j} $ is the NOx produced by each CG flash. FCG, FIC, and FT are the number of NCG flashes, PCG flashes, and total lightning flashes, respectively. $ j $ is the $ j $-th CG flash. Table 2. Comparison of the estimates of this study with other studies for the mainland of China.
Except for Zhou et al. (2003) and Zhang et al. (2014), all the studies estimated the total NOx in China. Estimates of 0.016–0.38 Tg(N) yr–1 are shown in Table 2 and are made by considering the average LNOx per lightning flash and the number of the lightning flashes. Compared to previous results, our NOx estimate in China is larger because of our higher NOx estimate of a single lighting flash.
In Sun et al. (2004), the average LNOx per lightning flash was estimated as 0.45 × 1025 molecules flash–1 based on the average energy per lightning flash being 4.5 ×107 J. However, Sun et al. (2004) neglected the reduction effects of the optical thickness on the optical energy and the inversion from the optical energy to the total energy, thus underestimating the energy of a single flash.
The estimations by Ju et al. (2015) and Guo et al. (2017) are done by combining both lightning and NO2 data observed by satellites. After establishing a fitting relationship between the NO2VCD and lightning on the Qinghai–Tibetan Plateau, Ju et al. (2015) obtained the single LNOx amount of 16.7×1025 molecules flash–1 and Guo et al. (2017) obtained the single LNOx amount of 20.87×1025 molecules flash–1 , which was smaller than the minimum mean single LNOx amount of 23.9 × 1025 molecules flash–1 used in this study. The height of thunderstorm clouds in the Qinghai–Tibet Plateau is relatively low, and the lightning channel length may be relatively short, which leads to the smaller LNOx amount of a flash (Zhang et al., 2014). So, the LNOx amount per flash on the Qinghai–Tibetan Plateau may be smaller than that for all of China. In addition, Zhou et al. (2003) and Zhang et al. (2014) both determined that the value of LNOx amount per flash based on the Tibetan Plateau is small. This also shows that the energy of lightning on the Qinghai–Tibet Plateau is relatively small.
Compared with the estimation of Zhou and Qie (2002), our estimate under group V3 is smaller. One reason is that their average LNOx per NCG flash is larger than ours, and LNOx generated by NCG flashes are dominant in the total LNOx amount. Since the breakdown voltage of NCG flashes (3×108 V) and the amount of NO per lightning flash (10 × 1016 molecules flash–1) are the same between Zhou and Qie (2002) and this study, it is the amount of charge in Zhou and Qie (2002) which is larger than ours that causes their result to be larger than ours. However, it is not known whether they had more lightning flashes because their lightning flash number data was not published.
In this study, the average single lightning flash energy of PCG flashes, NCG flashes, and IC flashes is 8.53 × 109 J, 4.78 × 109 J, and 1.47 × 109 J under the minimum breakdown voltage group V3, respectively. These values are larger than the estimate of Sun et al. (2004) (4.5 ×107 J), but on the same order of magnitude as the estimations of Zhou and Qie (2002) (6.7 × 109 J and 0.67×109 J for CG flashes and IC flashes, respectively). The energy of a single lightning flash estimated in this study is relatively large for NCG flashes and is similar to the upper limits given by Wang et al. (1998) and Maggio et al. (2009), who suggest that the energy upper limit generated by a single lightning flash is 6 × 109 J and 7 × 109 J, respectively.
The energy of a single lightning flash estimated in this paper is relatively large, but may be acceptable. According to our estimation of LNOx from the mainland of China [0.157–0.321 Tg(N) yr–1] and the ratio of land area of China to land area of the globe (6.44%), and assuming the value from the mainland of China is representative of global values, the global LNOx would be between 2.44–4.98 Tg(N) yr–1, which corresponds relatively well with the range [2–8 Tg(N) yr–1] from Schumann and Huntrieser (2007).
Lightning type | Breakdown voltage (V) | |
CG | V1+: 6.85×108 | V1-: −9.74×108 |
V2+: 6.13×108 | V2−: −5.93×108 | |
V3+: 5.00×108 | V3−: −3.00×108 | |
IC | VC: 1.75×108 | |
Notes: Groups V1, V2, and V3 respond three groups of breakdown voltages , respectively. “+” and “−” repond for the polar of breakdown voltages, respectively. |