Lightning accompanies strong convective weathersystems. As such, not only it is closely related to development of the dynamic and microphysical processesin such systems, but is also an indicator of their initiation and development. Through the interaction ofappropriate dynamic and thermodynamic conditions and terrains, cumulonimbus clouds form in unstableconvective environments, accompanied by strong up and down-drafts, creating a thunderstorm. The chargeseparation and transfer between different hydrometeorparticles with varying speeds result in formation of oppositely charged regions, leading to the occurrence oflightning. Lightning meteorology has mainly focusedon investigating the lightning activities in differenttypes of strong convective weather systems, and the relationship of lightning to the dynamic and microphysical processes in a thunderstorm. The primary application of lightning meteorology is convective weatherwarning and forecasting.

Observational data derived from experimentsbased on advanced lightning location technology arethe basis of underst and ing of lightning activity characteristics in severe convective weather systems, and areused in forecasting of severe convective weather. Thispaper first reviews development of the advanced lightning location technology; then introduces the characteristics of lightning in different convective systems, the relationship of lightning to the dynamic and microphysical processes in a thunderstorm, and the chargestructure of thunderstorms; and finally reviews lightning data assimilation and its application to severeweather forecasting. 2. Development of lightning location technology

With development of the electronic and information technology, many countries such as the UnitedStates(US), France, and China have successively developed advanced lightning location systems since the1970s. For example, the well-developed cloud-toground(CG)lightning detection networks since thelate 1970s, the very high frequency(VHF)lightningradiation source location system that accurately mapsthe lighting discharge processes since the 1990s, theWorld-Wide Lightning Location Network(WWLLN), and the space-based lightning detection sensors aboardthe satellites. These lightning detection technologieshave been improved continuously, and are applied inmany significant international research programs. Forexample, the Severe Thunderstorm Electrification and Precipitation Study(STEPS)conducted in the American Great Plains in 2000, the Thunderstorm Electri-fication and Lightning Experiment(TELEX)in 2004, and a 10-yr(2010-2020)weather observation plan, Hydrological cycle in the Mediterranean Experiment(HyMEX)project launched by France, etc. All ofthese projects used the lightning VHF radiation sourcelocation technology in three dimensions(3D)withhigh temporal and spatial resolutions. Based on thecomprehensive observ ations using advanced lightninglocation technology, the integrated observation datawith high temporal and spatial resolutions can be obtained. Then, the lightning activity in different convection systems and the relationship of lightning tothe dynamic and microphysical processes in a thunderstorm can be investigated, which is an active researchfield in the current atmospheric electricity. This lineof research also promotes the application of lightningdata in the monitoring and forecasting of catastrophicweather. 2. 1 Ground-based lightning lo cation

Lightning discharges emit electromagnetic radiation in broadb and frequency ranges, which is strongin the radio spectrum and even detectable in the highenergy radiation spectrum(X-ray and Gamma-ray). The electromagnetic radiation provides an importantsignal to detect and locate lightning discharges. Asthe most widely used lightning location system, thecrossed-loop magnetic direction finders(DF)workingin the very low frequency/low frequency(VLF/LF)range was first developed by the University of Arizonain the 1970s. The system could recognize the CGlightning flashes by using the feature of electromagnetic field waveforms produced by the return stroke, and then locate the CG flashes effectively. With thedevelopment of the Global Positioning System(GPS), the IMP ACT(improved accuracy from combined technology)algorithm was proposed as a combination ofDF and the time-of-arrival(TOA)technology. Basedon the IMP ACT, the commercial US National Lightning Detection Network(NLDN)was first operated(Cummins et al., 1998), and then similar lightning detection networks were set up in many areas and countries, such as North America, South America, Europe, and China. The detection efficiency of the NLDNreaches about 95%, the corresponding location precision is better than 500 m, and the time precisionis about 1 ms(Biagi et al., 2007). Similar lightningdetection networks were established by the China Meteorological Administration and the Chinese ElectricPower Companies. The CG location system providesinformation of time and location of the return strokes, and the peak current is estimated by using the peakvalue of the measured magnetic field. In the 1980s and 1990s, the CG lightning location systems playedan important role in the study of lightning activitiesin severe convective weather systems(e. g., Reap and MacGorman, 1989; Qie et al., 1993).

Based on the long-baseline TOA lightning location technology, the Lightning Mapping Array(LMA)was designed by the New Mexico Institute of Mining and Technology(Rison et al., 1999), and has become the most precise location system in 3D across theworld. The center frequency was originally 63 MHzwith a b and width of 6 MHz, and now the center frequency is adjustable in the VHF b and to avoid thebackground radio frequency interference. Total lightning, including both the CG lightning(less than 1/3of the total) and the intra-cloud(IC)lightning(more than 2/3 of the total), can be imaged in 3D, and the progression of lightning discharges can also be depicted in detail. The nonlinear least-square regressionalgorithm is applied in the 3D location of the radiation sources. The LMA determines hundreds to thous and s of radiation events per lightning flash, and accurately maps the progression structure of lightningdischarge in 3D. The LMA has been widely utilizedas the major detection technology in many researchprograms related to severe convection storms in theUS and Europe, such as STEPS, TELEX, DC3(theDeep Convective Clouds and Chemistry Project), and HyMEX. The US Lightning Detection and Ranging(LDAR II)system adopts the similar location principle as the LMA with two adjustable frequency b and s(60-66 and 222-228 MHz), and is another lightningVHF radiation location system with high detection efficiency and location precision in 3D.

In China, the Lightning Mapping System(LMS)has also been designed based on the long-baseline TOAtechnique, similar to the LMA in the US. The centerfrequency of the system is 270 MHz. The horizontalerror is less than 11 m and the altitude error is 2-3times greater, when lightning occurs inside or about10 km near the network with an altitude between 4 and 15 km. The altitude error increases with the reduction of the radiation source height(Zhang et al., 2010). Figure 1 shows the 3D location result of an IClightning detected by the LMS. It is found that thelightning was initiated from the lower negative chargeregion, and propagated upward into the upper positive charge region. Subsequently, the lightning channels spread horizontally in both charge regions. Atpresent, the LMS is only used in scientific research, while has not yet supported the real-time and continuous location over the duration of an entire thunderstorm. Therefore, it cannot be applied to operationalmonitoring of the lightning activity in severe convective storms.

Fig. 1. The 3D location of an IC lightning detected by LMS. (a)The height of lightning radiation versus time, (b)east-west(W-E)vertical projection, (c)the number of lightning radiation sources versus height, (d)plan view, and (e)south-north(S-N)vertical projection. Di®erent colors st and for time evolution. The discharge started in blue. [Adaptedfrom Zhang et al., 2010]

Figure options

SAFIR(System d'Alerte Fondre par Interferometrie Radio electrique)initiated in France is a lightning VHF detection and location system based onthe interferometry technique. The center frequencyof the system is adjustable within the range of 110-118 MHz and a b and width of 1 MHz. The SAFIRdetermines the lightning radiation sources in 3D, and reproduces the progression of lightning discharge channels. Meanwhile, the LF electric field antenna is integrated to distinguish the IC and CG lightning according to the electric field signals from 300 Hz to 3 MHz(Richard et al., 1986). Using data from the BeijingSAFIR3000 lightning detection system and Dopplerradar, Liu et al. (2011)analyzed the temporal evolution of lightning radiation sources in a leading-line and trailing stratiform mesoscale convective system(LL TS-MCS)over Beijing, and found that the radiation sources over the central detection network had agood agreement with the radar echo, while the altitudeerror of many radiation sources appeared relativelylarge.

In addition, there are other lightning VHF location systems, including the narrowb and interferometer(Zhang et al., 2008) and broadb and interferometer(Dong et al., 2002)based on the interferometry technique, and the lightning VHF radiation location system based on short-baseline time-difference of arrivaltechnique(Cao et al., 2012; Sun et al., 2013). Mostof the systems have merely realized the 2D location orfinite 3D location(Wu et al., 2012), and the coverageareas are relatively small. Therefore, these systemsare only applied in the study of lightning physics.

The WWLLN is operated in the VLF range(3-30kHz), with the center station established in the University of Washington. The network could effectivelydetect the electromagnetic radiation emitted by thelightning discharge at thous and s of kilometers away. With about 60 ground-based observ ations around theworld, the WWLLN detects global lightning activitiescontinuously. With a GPS antenna for accurate arrivaltime of the lightning impulses, the WWLLN locatesthe lightning based on the time-of-arrival technique(Dowden et al., 2008). Due to the lower operating frequency and fewer numbers of stations, the detectionprecision and efficiency of WWLLN were very low, and the detection precision was about 10 km. With increasing WWLLN station numbers, the detection efficiency of the WWLLN has increased from 3. 88% in2006-2007 to 10. 30% in 2008-2009. Besides, there are also other VLF/LF lightning location networks, for example, the Los Alamos Sferic Arrays(LASA)in America(Smith et al., 2002), Lightning DetectionNetwork in Europe(LINET; Betz et al., 2009), and Beijing Lightning Network in China(BLNet; Wang et al., 2009). 2. 2 Space-b ased optical lightning lo cation

The Lightning Imaging Sensor(LIS)aboard theTropical Rainfall Measuring Mission(TRMM)is designed to monitor the characteristics of total lightningactivity in severe convective weather. The LIS observes the lightning activity between 35°S and 35°Nwith a view of 600 × 600 km²area on the earth, and the spatial resolution is about 3-6 km. The view timefor an individual storm or storm system is about 90s. As a similar optical sensor before LIS, the OpticalTransient Detector(OTD)on the Mictolab-1 satelliteobserves an area of 1300 × 1300 km²in 70°inclination orbit. The spatial resolution is 10 km, and thetemporal resolution is 2 ms. The detection efficiencyof the OTD is between 50% and 66%. For LIS, the IC and CG lightning cannot be distinguished, while continuous observation is available day and night. Boccippio et al. (2002)compared LIS and LMA observ ation data, and found that the lighting location resultsfrom the two datasets were well consistent in space and time. The IC lightning occurs at middle-upperlevels of the storm, which is easier to be detected bythe LIS than the CG lightning, and the LIS tends todetect the latter period of the CG lightning. The LISdetection efficiencies near local noon and near localovernight are about(73ffi11)% and (93ffi4)%, and location errors for IC and CG lightning are about 4 and 12 km, respectively.

Although there are significant differences betweenOTD and LIS in the spatial resolution and observationregions, NASA researchers have combined and averaged the two sensors' data to create a global lightningactivity database spanning from May 1995 up to now, which is the first lightning database with long time and high precision in the world. For the database, not onlythe temporal-spatial changes in instrument detectionefficiency are considered, the different view time of thesensors in different latitudes is also corrected. Figure 2shows the lightning distribution of the world from May1995 to February 2012 by LIS/OTD. It is obvious thatlightning activity over l and is far more than that overocean, and the lightning density ratio of l and to oceanis about 10:1(Zhu et al., 2013). The LIS, the precipitation radar(PR), TRMM Microwave Imager(TMI) and other sensors on the TRMM can be combined together to provide information on the characteristicsof lightning activity, precipitation, and microphysical processes in different severe convective weathersystems.

Fig. 2. Lightning(fl yr-1km-2)distribution across the globe from May 1995 to February 2012 by LIS/OTD. [From Qie et al., 2013]

Figure options

The development of lightning detection technology has greatly improved the underst and ing of lightning activity in thunderstorms. Combined data fromthe lightning detection network, Doppler weatherradar, and satellites have been particularly useful. Thegradual development of the CG lightning detectionnetwork in the US since the late 1970s has enabled thecharacterization of CG lightning activities of convective weather systems, based on the lightning locationdata including both IC and CG lightning observed bylightning detection systems such as SAFIR, LDAR and LMA, WWLLN, and satellite(LIS/OTD). The lightning activities in convective weather with different precipitation characteristics have been investigated, suchas squall line, supercell, hailstorm, typhoon, etc. (Qie et al., 2005, 2006; Zhang et al., 2006a; Qie, 2012). Wenext separately consider in detail the different types ofprecipitation systems that may accompany lightning. 3. 1 Lightning char acteristics of hailstorm

Generally, the frequency of -CG(negative CG)lightning is higher than +CG lightning during the lifecycle of a thunderstorm, but hailstorms are usually accompanied by a higher ratio of +CG lightning(Carey and Rutledge, 1998; Feng et al., 2006; 2007; Liu et al., 2009). Based on CG lightning location, radar, and TRMM/LIS data, Feng et al. (2007)observed thatthe +CG lightning accounted for a high percentage oftotal lightning over the lifetime of a hailstorm, withthe hail-falling stage coinciding with the most rapidincrease of +CG lightning. Using total lightning detected by SAFIR3000 and radar data, Zheng et al. (2010)found that +CG lightning frequency and hailfall were correlated. Therefore, the +CG lightning frequency may indicate the occurrence of hail fall. However, Soula et al. (2004)analyzed the lightning activity of hailstorms in the south of France, and demonstrated that the CG lightning frequency was no morethan 2 fl min^-1(flashes per minute): far less than thatexpected in a rain-generated storm, where the lightning frequency was 12 fl min^-1. This result showedthat +CG lightning frequency alone may not reliablyidentity hailstorms.

Severe weather events, including hailstorms, areoften accompanied by frequent lightning activities. Arapid increase in total lightning frequency indicatesthe occurrence of severe weather events. Schultz etal. (2011)analyzed 711 thunderstorms that occurredin four different regions of the US. Their large sample size resulted in good representation for all severethunderstorm types. They included tornado-producedstorms, those where the hail diameter exceeded 1. 9 cm, and those where the wind speed exceeded 26 m s^-1, and also weak thunderstorms. They tested the ability of monitoring the rapid increase of total lightning and CG lightning to forecast severe weather events, using the method of 2σ lightning jumping(the rate ofchange of the total flashes rate; DFRDT). Both the total and CG lightning were found to increase before theoccurrence of severe weather events, but the forecasting ability of total lightning flashes was more effectivethan CG lightning, with the total lightning of 20. 65min and CG lightning of 13. 54 min. Yao et al. (2013)studied the hailstorms in the Beijing area, and foundthat the 2σ lightning jumping method(+CG lightning and total lightning)could also be successfully appliedto hail forecasting in that region. 3. 2 Lightning activity in line ar mesoscale convective systems

Linear mesoscale convective systems(MCSs)aresevere convective weather systems occurring frequently in summer, and are also known as squall lines. Using radar morphology, Parker et al. (2000)dividedlinear MCSs into three categories: leading stratiform(LS-MCS), trailing stratiform(TS-MCS), and parallel stratiform(PS-MCS). They found that LS-MCSsproduced more +CG lightning than the other twotypes. Further observ ational studies found that the-CG lightning dominated at the mature stage of alinear MCS, and +CG lightning occurred relativelyfrequently in the stratiform region at the dissipatingstage(Parker et al., 2000; Feng et al., 2009). Casestudies of the lightning VHF source location showedthat the total lightning frequency of a linear MCS increased from 130 to 600 fl min^-1(Lang and Rutledge, 2008; Liu et al., 2013a, b). Yuan and Qie(2010a)studied a squall line that occurred in South China usingTRMM/LIS data, and found that the instant lightning frequency reached a maximum of 567 fl min^-1during the satellite pass over. This result suggestedthat the lightning activities of squall line systems are more active than normal thunderstorms(Yuan and Qie, 2010b)because of a wide range of strong convection. Using an SAFIR3000 lightning location system, Liu et al. (2013a)analyzed a linear MCS over Beijing. Their results showed that lightning activity was correlated with radar data and precipitation. Figure 3, reproduced from that work, indicates that lightning wasconcentrated over the strong convection area, whilefewer lightning flashes occurred in the stratiform region, and most of them were +CG lightning.

Fig. 3. Composite radar reflectivity and the lightning location within six minutes at two times of a leading line MCS. Black dots represent lightning flashes. [From Liu et al., 2013a]

Figure options

Carey et al. (2005)used LDARII data to examine lightning activity in a linear MCS and obtained thefine structure of IC lightning extension. They foundthat the lightning radiation sources had a two-layerdistribution, sloping from the convective layer to thestratiform region as shown in Fig. 4. To a certainextent, the results reflected the charge distribution inthe lightning discharge of linear MCSs.

Fig. 4. Lightning radiation sources detected by VHF(shading) and radar reflectivity(contours)of a typical linearMCS. [Reproduced from Carey et al., 2005]

Figure options

A positive charge in the stratiform region of alinear MCS may form via either the charge advectionfrom the convective region or the local electrificationmechanism. The charge advection mechanism happens when ice crystals in the stratiform region becomepositively charged through the advection of upper air-flow from the convective region. Figure 4 shows thelightning radiation source distribution, which correlates well with the tra jectory of ice particles sinkingin the MCS shown in Fig. 5. Since ice particles arethe primary positive charge carriers, the charge movesfrom the convective region to the stratiform region bythe advection of upper airflow of the thunderstorm. The local electrification mechanism occurs when insitu charge separation happens within the stratiformregion, through ice particle collisions. Dotzek et al. (2005)found that both the horizontal propagation ofdischarge sources and a radar reflectivity factor abovethe 0℃ bright b and in the mixed phase area were increased, confirming a local electrification mechanismin the stratiform region.

Fig. 5. An MCS conceptual model. [From Carey et al., 2005]

Figure options

Transient luminous events(TLE)usually occurover the MCSs(Pasko et al., 2002; Yang et al., 2008, 2013a, b). Red sprites are thought to be induced by+CG lightning in the stratiform region of an MCS. Luet al. (2009)discussed the charge moment changes ofthe sprite-producing +CG lightning, typically 1500-3200 C km^-1, which is much greater than that of normal CG lighting. Yang et al. (2013a, b)compared thesprite-generated MCS to other thunderstorm types, and suggested that the convection was stronger insprite-generated thunderstorms, but there was no obvious difference in the microphysical characteristics. 3. 3 Lightning activity in typhoons

Although lightning in tropical cyclones varies, themean lightning density presents an obvious three-circlestructure during most of the typhoon stage(Molinari et al., 1994; Pan et al., 2010; Zhang et al., 2012). Thepeak lightning density appears in the outer rain b and. A lower lightning frequency is found in the eyewall region, and lightning frequency trends to be zero in theinner rain b and. Figure 6 shows that the lightningoccurring in the outer rain b and has an asymmetricdistribution, with the highest density mainly locatedin the deep convection area.

Fig. 6. Lightning distribution in Typhoon Qiangwei. (a)Lightning(dark dots)during 0200-0600 UTC 27 September2008, superimposed over visible cloud imagery from GMS-6 data at 0600 UTC, and (b)composite lightning distributionat the mature stage of Typhoon Qiangwei.

Figure options

The lightning density at the cyclone center ishigher while the storm strengthens than at the weakening stage(Abarca et al., 2011). Generally, the lightning in the eye wall of a typhoon increases sharply asthe wind speed reaches its maximum, so the lightningdensity in the eye wall could indicate the intensity of astrengthening typhoon(Pan et al., 2010). Price et al. (2009)analyzed lightning data detected by WWLLNfor 58 hurricanes during 2005-2007, and showed thatlightning frequency and hurricane strength(i. e., maximum sustained winds)were positively correlated, withan average correlation coefficient of 0. 86. Pan et al. (2014)studied lightning in 69 tropical cyclones overNorthwest Pacific, and found that in more than half ofthe weak(levels 1-3) and strong(levels 4-5)typhoons, the peak value of lightning usually occurred before themaximum wind speed was attained. Yang et al. (2011)analyzed lightning during different intensity stages of46 tropical cyclones using the distribution characteristics of radar reflectivity and ice scattering from theTRMM data, and found that the spatial distributionof lightning differed for storms at different intensities. Diurnal variation of lightning above the sea has beenshown to have two peak values, occurring respectivelyin the afternoon and morning(Pan et al., 2013). 3. 4 Compact intr a-cloud discharge and str ongconvection

In the last two decades, a special type of lightning discharge named CID(Compact Intra-cloud Discharge)or NBE(Narrow Bipolar Event), has causedwidespread concern. CIDs are different from regularlightning discharge processes. This kind of dischargehas a small discharge scale and a short duration ofabout 10-20 μs, with extremely strong high and lowfrequency radiation, one order of magnitude largerthan that of regular intra-cloud discharge(Smith et al., 1999; Zhu et al., 2007, 2010; Wang et al., 2012). Based on 3D CID location results, Wu et al. (2011)found that the discharge height of negative CIDs wasgenerally comparable to that of the tropopause, and the negative CIDs were much less abundant than positive CIDs. These results indicated that negative CIDswere probably produced in extremely vigorous thunderstorm processes, and may therefore reveal strongconvective activity. CIDs tend to occur at top and middle levels of thunderstorms. These discharges canbe produced not only in regions with strong radarecho, but also in regions with lower than 30-dBZ radarecho in the later stage of a thunderstorm. The heightof CIDs might be related to the ionosphere height, butno significant relation has been found. Using LMS, Wang et al. (2012)analyzed 236 CIDs observed inBinzhou, Sh and ong Province during the summers of2007-2008, and found that CIDs occurred at altitudesbetween 7 and 16 km with a peak power ranging between 12 and 781 kW in the 267-273-MHz b and. Lüet al. (2013)reported that NBEs in the Daxing'anlingregion tended to occur during the relatively active period of lightning discharges. NBEs tended to clusteraround the area of a particular convective core withhigh radar echo, and most gathered at the front area ofconvective cores. The position of NBEs moved consistently with the movement of the particular convectivecores. 4. Relationship of lightning to dynamic and microphysical processes and precipitation

Strong convective weather systems are characterized by dynamic, microphysical, and electrochemicalprocesses. Dynamic processes are characterized bystrong updrafts, down bursts, and horizontal windshear. Microphysical processes include all kinds ofparticle growth and phase change. The electricalprocesses denote electrification, charge distributionchanges, and discharge process. All of these processesinfluence each other. Usually, lightning occurs wherean updraft is strong enough to support the coexistenceof a mixed phase region with graupel and liquid water. Carey and Rutledge(1998)found that CG lightning had a very close relationship with graupel. CGlightning was generally located in a strong radar echoregion, but was not completely consistent with thestrong updraft. Based on radar data, lightning detection and electric field sounding, Bruning et al. (2007)observed a multi-cell thunderstorm. They proved thatthere were many graupel particles at the occurrenceof the first lightning, and that the strong electric fieldwas caused by electrification from ice particle collisions in the updraft. Lightning frequency and ice particlecontent were positively correlated.

Zhou et al. (2002)found that lightning and convective precipitation were correlated, and lightningcould therefore be used to estimate precipitation ingeneral convective weather. Meanwhile, lightning frequency and unstable stratification maximum energycould indicate the occurrence and development of convectional weather(Zhou et al., 1999). Feng et al. (2007)combined ground-based radar and lightning location data with LIS, PR, and TMI data from theTRMM satellite, and found that convective precipitation comprised more than 85% of the total precipitation in hailstorms, and the correlation of the lightning and convective precipitation could be used to identifythe convective precipitation area effectively. Based onmulti-sensor data from the TRMM satellite, Yuan and Qie(2010a)studied lightning activity and its relationship with the precipitation structure of a strong squallline in South China, and found that the majority oflightning occurred in the convective region. The vertical profile of the maximum radar reflectivity of theconvective cell may be a good indicator of lightningfrequency and convective intensity. For the cell accompanied by the highest frequency of lightning, theradar reflectivity at each height level was usually thelargest, and the reflectivity lapse rate was the smallestover the frozen layer, and vice versa, for the cell withthe lowest frequency of lightning. Zheng et al. (2004)analyzed a frontal cyclone system that occurred in theHuaihe River basin, and the results showed that lightning occurred in the strong convective precipitationcloud influenced by the cold front; no lightning occurred in the warm front. Precipitation profiles withhigh lightning frequency usually contain a high densityof ice particles. The precipitation profile of a thunderstorm with different lightning frequency is different. The higher lightning frequency corresponded tothe greater precipitation above 5-km height inside thethunderstorm. Hence, thunderstorms with more frequent lightning had more ice particles above the frozenlayer(Ma et al., 2012).

To realize lightning parameterization in a globalor mesoscale model, it is necessary to obtain a quantitative relation between the lightning activity and thedynamic and microphysical parameters of the thunderstorm. Research in this area has most recentlyused the TRMM satellite multi-sensor data. Yuan and Qie(2008)found that lightning frequency and parameters of thunderstorms, such as the top of thunderstorm, frozen layer thickness, and minimum temperature, showed an exponential relationship; for both aprecipitation system or a convective cell, the relationship between lightning frequency and ice-phase precipitation content from 7 to 11 km remained relativelystable, and the correlation coefficient was greater than0. 7. Their work provides a possible lightning parameterization scheme that might be used widely in regional or global models. 5. Charge structure of a thunderstorm: Observation and simulation

An essential technique in lightning and thunderstorm research is electric field sounding. This methodhas the major advantage of directly detecting the electric field inside a thunderstorm, and reflecting the vertical distribution of the charge along the soundingpath. Electric sounding experiments have played anessential role in recognizing the charge distribution ofdifferent thunderstorms, and promoting the research ofthunderstorm electrification mechanism, charge structure and its relationship with storm dynamics and lightning initiating mechanisms(Marshall et al., 1995;Stolzenburg et al., 1998). 5. 1 Sounding of the char ge structur e inside athunderstorm

In as early as 1937, Simpson and Scrase(1937)obtained direct scientific evidence of the tripole chargestructure inside thunderstorms by using the electricfield sounding. This simplified tripole charge structuremodel remained widely accepted for more than half acentury. More recent electric field sounding data haveshown that the real charge structure is much morecomplex for most of the thunderstorms. Marshall etal. (1995)observed 11 electric field soundings and corresponding thermodynamic parameters for multicell and supercell thunderstorms in the southern GreatPlains of the US. They found that the charge structure in the weak updraft area was complex, with 7-9charge regions below 10 km, while the charge structure was simple in the strong updraft area with 3-5 charge regions below 10 km. Using 49 electricfield soundings, Stolzenburg et al. (1998)investigated three categories of thunderstorm system, including MCSs, supercells in the southern Great Plains, and small thunderstorms in New Mexico mountain region of the US. Their results showed that the convective regions of the three types of thunderstorms hadthe same basic charge structure: in the updraft areathere were four vertical, alternating polarity chargeregions and the lowest charge zones were positivelycharged. Figure 7 graphically displays their findings. The charge structure outside the updraft region(stillin the convective region)included six vertical alternating polarity charge regions, with the lowest chargeregion still positively charged. The height of thebasic charge regions inside the three thunderstormsincreased in line with the updraft of the thunderstorm.

Fig. 7. Conceptual model of the charge structure in the convective region of a thunderstorm. [From Stolzenburg et al., 1998]

Figure options

Tessendorf et al. (2007)revealed the existenceof inverted charge structure of thunderstorms. Using 3D LMA data from the STEPS experiment, theyanalyzed lightning radiation sources of two thunderstorms, and found that one thunderstorm showeda normal tripole charge structure with the middlelayer negatively charged. Although the upper negative shield charge region was not reflected, the chargestructure was consistent with electric field soundingresults in view of the charge region involved in thedischarge. The other thunderstorm had an invertedcharge structure with positive charge at the middlelevel, opposite to a normal charge structure. Theinverted charge structure usually occurred in supercell storms, accompanied by tornadoes or other disastrous weather(MacGorman et al., 2005). Zhang et al. (2004)utilized the LMA lightning data to analyze thelightning characteristics of a supercell storm and foundthat the spatiotemporal distribution of lightning holes and rings corresponded to updrafts and downdrafts. Lightning holes usually appeared before a tornado, and became most obvious during the tornado stage. During the period of high frequency +CG lightning, the main convective region of the thunderstorm had aninverted tripole charge structure, with +CG lightningproduced by the positive charge at the middle level. The -CG lightning was mainly located in the anvil region of the thunderstorm. The cloud anvil showed adipole charge structure, and a high frequency of -CGlightning was generated in the upper negative region(Zhang et al., 2004).

Over the inl and plateau of China, thunderstormsoften present with a special charge structure. Qie et al. (2005, 2009)comprehensively analyzed the surface electric field and lightning charge source locationof a large number of thunderstorms. Some thunderstorms showed an abnormally large positive chargeat the lower level, and the proportion of this typeof thunderstorms gradually increased with average altitude, adding to the evidence for the predominanceof larger lower positive charge region(Liu et al., 1989). Based on corona probe sounding techniques, the charge structure of a thunderstorm that occurredin Gansu Province, China was observed by Zhao et al. (2009). Their results showed that the thunderstorm presented a negative charge region at the middle level, and positive charge regions at the upper and lower levels with the lower positive region largerthan the normal tripole, further confirming the abovementioned conclusions. Recently, based on 3D localization of wideb and electric field change pulses, Li et al. (2013) analyzed the charge structure of a thunderstorm in Qinghai Province, China. They found an inverted dipole charge structure at the development and mature stage of the thunderstorm, with four chargelayers(positive-negative-positive-negative)at the dissipating stage, at heights of 5. 0, 4. 0, 3. 0, and 1. 8 km, respectively. 5. 2 Simulation of char ge structur e in thunderstorms

Increasing computing speeds have made it possible to study thunderstorm electrification and discharge processes by using high resolution numericalsimulations. The simulation of dynamic-electrificationcoupling has already become an important researchtopic. There are three main advantages of simulations. First, complex interaction of the microphysical processes with the electrification, discharge, and dynamicprocesses can be unraveled. Second, simulations canprovide high spatiotemporal resolution data for different physical processes, going some way to remedythe deficiency of thunderstorm observ ations. Third, all kinds of electrification, discharge mechanisms, and theoretical hypotheses can be effectively and economically verified.

The noninductive electrification mechanism isconsidered to be the major electrification mechanismin strong convective weather(Takahashi, 1978; Saunders et al., 1991). In this mechanism, charge separation is caused by bouncing collisions between large and small ice particles in the mixed phase region, and doesnot require an external electric field. The noninductive electrification mechanism results in a fast charging rate. Moreover, the reversal temperature, whichaffects the polarity of charge transferred, is a key factor in determining the charge structure. The inductiveelectrification mechanism of hydrometeor particles dependent on the environmental electric field also playsan important role at the initial stage of the electrifi-cation of thunderstorms.

With the gradual improvement of models and increased availability of observ ational data, the simulated charge structure of thunderstorms has becomeincreasingly accurate. Mansell et al. (2005)comparedfive different noninductive electrification schemes, using a 3D thunderstorm model that considered both theinductive and the noninductive electrification mechanisms. They found that three had a normal tripolecharge structure(negative region in the middle), whilethe other two were mainly dependent on the riming ofgraupel particles. Although lightning propagation and breakdown processes are relatively complicated, it ispossible to establish lightning discharge parameterization schemes on the basis of the electrification modelwith a high temporal and spatial resolution. A largenumber of studies on charge structure and dischargeprocesses were based on the thunderstorm model developed from hailstorm models in China(Yan et al., 1996; Zhang et al., 2000; Sun et al., 2002; Tan et al., 2007). The simulation of real thunderstorms is alsogathering attention of more researchers(Guo et al., 2007; Zhou and Guo, 2009; Liu et al., 2014). Tan et al. (2006)adopted the lightning parameterization from Mansell et al. (2002)into a 2D high resolution thunderstorm model, and the lightning discharge channelstructure and propagation characteristics were reproduced well. The model resolution reached 12. 5 m, and the simulation results well represented the dischargecharacteristics of the channel and the development ofbi-directional leaders.

In order to simulate large-scale thunderstormsmore realistically, simulations of charge structurebased on the mesoscale NWP(numerical weather prediction)models have been developed in recent years. Huang et al. (2008)simulated lightning activity bycoupling electrification and discharge processes in themesoscale model GRAPES(Global/Regional Assimilation and Prediction System)of China, which provided a background field for a nested cloud-scalemodel. Li et al. (2012) and Liu et al. (2014)introduced two noninductive electrification schemes. T ak ahashi(1978; abbreviated as Takahashi78) and Saunders and Peck(1998; abbreviated as Saunders98)implanted lightning discharge parameterization schemesinto the RAMS(Regional Atmospheric Model System)model v6. 0, and found that the simulated thunderstorm showed a tripole charge structure under theTakahashi78 electrification scheme and changed froma dipolar to tripolar charge structure under the Saunders98 scheme. The simulated lightning frequency wasconsistent with observ ation. Xu et al. (2012)simulated the charge structure of a supercell in the WRF(Weather Research and Forecasting)model, which wascoupled with electrification and discharge processes. The results showed a tripolar charge structure with apositive charge region between -40 and -60℃, a mainnegative charge region between -10 and -30℃, a lowerpositive region near the 0℃ layer, and a maximum total charge density approximated to 2 n Cm^-3. Thesimulated results also showed that the squall line presented a dipolar charge structure, and the maximumcharge density was less than that in the supercell. During the mature stage of the squall line, the simulatedlightning activity was similar to observation. 6. Assimilation of lightning data and forecasting of severe convective weather

Due to a wide detection range, small terrain effects, and continuous observ ation, lightning data havea potential application in the monitoring, early warning, and forecasting of strong convective weather systems. With the accumulation of high quality lightninglocation data, lightning data assimilation has becomean important research topic.

The exploration of lightning data assimilationmethods has received substantial attention. Alexander et al. (1999)were early investigators of lightningdata assimilation techniques. They established a relationship between lightning and precipitation ratio witha classic image processing method using microwavesounding data and CG lightning data, and applied itin the MM5 model. This lightning data assimilationtechnique was shown to improve the 12-24-h precipitation forecast of a superstorm event. Mansell et al. (2007)added NLDN and LMA lightning into the coupled ocean-atmosphere mesoscale prediction system(COAMPS), and utilized lightning data to control theconvective parameters of the model, producing simulated precipitation that accurately matched with theobserv ations. Li et al. (2008)obtained a relationshipbetween lightning and convective precipitation by using the TRMM satellite data, and assimilated the convective precipitation retrieved by LIS data into the initial field of the ARPS(Adv anced Regional PredictionSystem)model. To a certain extent, the predicted center and intensity of heavy rain in the Jianghuai regionwas improved. Based on MM5 model and the lightning and precipitation data of the TRMM satellite, the relation between lightning and precipitation ratewas established by Pessi and Businger(2009). Theyadded the relation into the MM5 model, and obtainedreasonable results for a low-pressure system in NorthPacific. Combining the methods from Papadopouloset al. (2009) and Mansell et al. (2007), Ran and Zhou(2011) conducted a nudging assimilation of water vapor and cloud hydrometeors using TRMM lightning data. Improvements in short-term rainfall forecasts were achieved for three short-term precipitationevents. The distributions of the hydrometeor particles in a thunderstorm play an important role in lightning occurrence. Fierro et al. (2012)established thenudging function of the water vapor mixing ratio and graupel mixing ratio with lightning frequency, and thehigh-precision lightning data were assimilated into theWRF model. A thunderstorm with a tornado waswell simulated, and the convection forecast was signifi-cantly improved. Recently, Qie et al. (2014)established empirical relations between total lightning flashrate and the ice particle(graupel, ice, and snow)mixing ratio. The constructed nudging functions wereused in the WRF model, and they found that the representation of convection was significantly improvedone hour after the total lightning data assimilation, even during the assimilation period. The precipitation center, amount, and coverage were all much closerto the observation in the sensitivity run with lightningdata assimilation than in the control run without lightning data assimilation.

Besides the application of lightning data to convective weather forecasting, lightning forecasting itself is also an important issue. Zheng et al. (2005)used lightning data for the Beijing area combinedwith sounding data, and found that lightning wasassociated with the potential convective stability index, uplift index, convective available potential energy(CAPE), and potential temperature at 700 hPa. Based on the analysis of the multi-parameter prediction of lightning probability, the diagnostic indicator oflightning forecasting was determined. Strong updraft and sufficient water vapor resulted in more ice particles, which had a direct effect on electrification and discharge(Zheng et al., 2007). Based on the above studies, a lightning monitoring and early warning systemwas developed by the Chinese Academy of Meteorological Sciences, as described by Zhang et al. (2006b). Based on multiple parameters, multi-algorithms integration technology, and weather forecast products, thepotential lightning area and the probability of lightning occurrence in 0-2 h were obtained by the comprehensive prediction method. According to the correlation between lightning density and radar echo, aforecast scheme of CG lightning was established and coupled with the GRAPES model(Wang et al., 2010). It is found that the scheme could forecast the lightning center in 6 h, and the predicted lightning densitywas consistent with the observation for two thunderstorm cases in South China. McCaul et al. (2009)established a regression equation for prediction of lightning density based on the statistical correlation between ice particles and lightning density. The forecasted lightning zone and trend in 6 h for a supercellstorm with a tornado and a hailstorm was consistentwith the observ ations. Barthe et al. (2010)simulated a strong thunderstorm and an air mass thunderstorm in a plateau region using the WRF model. They investigated the lightning forecasting ability byusing different physical quantities(falling ice mass, icewater path, and ice quantity flux, updraft, maximumupdraft, and cloud top) and obtained decent results. Yair et al. (2010)introduced a new lightning potential index into the WRF model, and forecasted theoccurrence of lightning by using the observed lightning data. They found that the forecasting ability forstrong convective weather was much improved in predicting the lightning distribution region and the precipitation of three thunderstorms that occurred in theMediterranean. 7. Conclusion

In this paper, the recent research adv ances inlightning meteorology are reviewed, with a focus onthe research conducted in China. Advanced lightning detection and location technology are summarized, and their important role in the study of lightning meteorology is discussed. Lightning phenomenain different strong convective weather systems such ashailstorm, squall lines, and typhoons are reviewed, and the relationship of lightning and dynamic, microphysical processes and precipitation of thunderstorms arediscussed. The charge structure of a thunderstormis also discussed in terms of both observ ational and simulation results. The lightning data assimilationmethod and lightning data application in strong convective weather forecasting are finally elaborated.

The comprehensive underst and ing of lightning activity is very difficult, because of the complexity inthe strong convection weather systems and the differences between individual storms, in addition to thedifficulty of accurate detection of lightning. Many scientific questions remain unanswered. The current underst and ing of lightning activity in different weathersystems is still very limited in China. High accuracy and high resolution lightning location technology iscrucial for the study of lightning meteorology, and is also the basis of application of lightning data in severeconvective weather forecasting. The technology of 3DVHF radiation source location could map lightningdischarge with high time and spatial accuracy. Theelectric field and comprehensive meteorological soundings are the most direct measurement approaches applied to the underst and ing of electrification in thunderstorms. Both the 3D VHF radiation source location technology and electric field sounding have beenwell developed in the US, but not in China. Althoughthe LMS, which is similar to the LMA, has been developed by the Cold and Arid Regions Environmental and Engineering Research Institute, Chinese Academyof Sciences(Zhang et al., 2010), there is still no capability of real-time lightning location or monitoring. Research into the electric field sounding of thunderstorms is progressing(Zhao et al., 2009), but onlystrong electric fields can currently be detected, and the related data are also very limited. Research alongthis direction should continue to be of high priority inChina.

Electrification, lightning, dynamic and microphysical processes in thunderstorms and their relationships are important scientific issues of lightningmeteorology. It is very important to study the characteristics of lightning activities in different strongconvective weather systems, and also the relationshipsbetween the dynamics and the microphysical structure of thunderstorms based on lightning detection and location networks, by using Doppler-polarizationradar observ ations and the electric field sounding inside thunderstorms. The above mentioned issues arethe key problems in lightning meteorology today, and will provide the theoretical basis for the application oflightning monitoring in forecasting of strong convective weather.

Lightning data assimilation methods and theirapplication to numerical simulations and predictionswill provide supplementary methods to improve shortterm severe convective weather forecasting. With thecomplementary radar data to lightning data, the accuracy of convective activity information in the modelinitial field could be improved, and the ability of theshort-term forecasting of strong convective weathercould be further enhanced. This is an important and promising research direction. In addition, carryingout lightning prediction and improving the ability oflightning disaster forecasting will also be an importantresearch direction and application target of lightningmeteorology in the future.