Notice: Wiley Online Library will be unavailable on Saturday 27th February from 09:00-14:00 GMT / 04:00-09:00 EST / 17:00-22:00 SGT for essential maintenance. Apologies for the inconvenience.
 A new technique has been developed to operate spatially extended VLF/LF lightning detection networks in a pseudo 3-D mode in order to facilitate discrimination of intra-cloud and cloud-to-ground lightning. Time-of-arrival measurements are carried out with high precision to distinguish VLF/LF-emission regions of lightning discharges in higher altitudes and near ground, respectively. This is accomplished by utilizing deviations of arrival times measured at sensor stations close to lightning events as compared to arrival times expected on the basis of 2-D propagation paths. Successful functioning of the method in networks of common dimensions requires that both signal processing and event time tagging is achieved with an accuracy level of about 1 μs. Results are presented from a new network which has been operating continuously since May 2003 and covers southeast Germany in an area of approximately 300 × 400 km. VLF/LF source emission heights of ∼5–15 km have been identified.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 Lightning detection networks operate in many regions of the world and are intended to report occurrence of ground strikes. While it is undisputed that VLF/LF networks function successfully a certain fraction of lightning is not reliably identified as either cloud-to-ground (CG) or intracloud (IC) discharge; this ambiguity occurs in cases involving both relatively low and very high peak currents [Cummins et al., 1998; Orville et al., 2002; Jacobson et al., 2000; Suscynsky et al., 2000]. Smith et al.  report source height determinations for cases where ionospheric reflections show up in VLF/LF electric-field-change signals. For a long time, numerous 3-D techniques have been developed, especially interferometric measurements in the VHF range [Taylor, 1978; Proctor et al., 1988; Krehbiel et al., 1999], as well as combined observations with VHF and VLF methods [Jacobson et al., 2000], all of which allow excellent tracking of lightning channels and CG - IC discrimination. In large networks, however, these high precision tools have not been used for economic reasons and because of difficulties which arise in the online treatment of large data sets. We argue that progress with respect to reliable CG identification in VLF/LF networks may be attained by further improvements of common pulse shape discrimination based on a combination with a more direct pseudo 3-D observation of VLF/LF emission heights. In this Letter we present a new technique to approach this goal.
2. Munich VLF/LF Network
 Since May 2003 a 6-station network started continuous operation (Figure 1). Basically, the network utilizes technical standard procedures. At each site, two crossed loops are used and arranged such that the magnetic flux Bx(t) and By(t) is measured as a function of time without need for integration; the antenna handles frequencies from ∼1 kHz to more than 1 MHz but the upper limit of the operating range is set to values between 200 and 400 kHz. The higher frequency cut-off serves anti-alias properties and reduces influence of radio signals, such that waveform distortion remains acceptable for the present purposes. Bearing directions are derived from the ratio Bx/By and event time tagging is achieved with common GPS receivers of 300 ns accuracy. Sampling is performed at 1 MHz in a continuous mode and triggered events are analyzed in a parallel circuit so that no dead time occurs. Thresholds are set low enough to detect relatively weak discharges down to ∼1 kA peak current originating from a distance of 100 km (range normalization as commonly defined [see Diendorfer et al., 1998]). As a result, strong signals are detected highly consistent with the network operating commercially in Germany (BLIDS/ALDIS) in view of both event time and ground location, but the number of weak signals below ∼5 kA peak current is dramatically enhanced. Much effort was taken to achieve the standard accuracy (∼1 μs) of arrival times also for low-amplitude signals and complex pulse shapes; this is essential for optimization of 3-D locating and its application to as many signals as possible. A more comprehensive description of the extremely economic yet highly efficient layout will be published in a follow-up paper.
3. Lightning Location
 Common lightning detection networks operate in the VLF/LF range and provide 2-D locating with the principal aim to report ground lightning by measuring return strokes. This task is facilitated by the fact that return strokes emit mainly in the VLF/LF range, whereby - according to general acceptance - the center of radiation lies near the ground level [Rubinstein and Uman, 1990; Nucci et al., 1990]. Typical IC discharges such as stepped leaders, dart leaders, K-events, recoil streamers and related phenomena radiate mainly in the VHF range. Here, radio interferometric methods allow unique tracking and distinction of IC discharges; due to high spatial and temporal resolution 3-D studies are feasible in great detail [Shao et al., 1995]. Likewise it is a fact that the distinct VHF radiation of IC processes is accompanied by VLF/LF emission of intensities varying from insignificant to very pronounced amounts. For example, strong IC-pulses have been observed in the VLF/LF band [Smith et al., 1999; Jacobson et al., 2000], which exhibit features comparable to those of return stroke fields.
 Traditionally, lightning detection networks disentangle CG from IC discharges by means of pulse shape and amplitude considerations. In particular, a number of waveform features have been identified which are quite characteristic, such as pulse rise time, peak-zero time, bipolarity, or multiple peak structure. In the past, a wealth of publications has presented instructive and unquestionable examples for different patterns of IC and CG pulses. It turns out, however, that despite the usefulness of these criteria a substantial amount of pulses remain to be correctly identified. For example, the minimum pulse width (peak to threshold time) for classification of CG strokes was changed from 10 μs to 7 μs in the U.S. National Lightning Detection Network (NLDN) [Wacker and Orville, 1999a, 1999b] and from 11 μs to 6 μs in the ALDIS network [Diendorfer et al., 1998], and Heavner et al.  ask for more than 30 μs fall-off time. Furthermore, after many years of research and extensive evaluations of the NLDN, Cummins et al. , Wacker and Orville [1999a, 1999b] and Orville et al.  go as far as to suggest that all signals of positive sign and with peak currents less than 10 kA be regarded as IC events. But even large peak currents frequently raise problems: the NLDN identifies a class of high intensity IC discharges as observed by Jacobson et al.  and Smith et al.  as CG strokes (see below). Under these circumstances it may be helpful to add a new technique.
4. Pseudo 3-D Technique
 Time-of-arrival (TOA) represents a standard procedure for lightning location with high accuracy. With GPS-based timing modern networks report event time precision of the order of 1 μs [Cummins et al., 1998]. This level of time accuracy is particularly realistic in networks with relatively small sensor baselines because short travel distances between stroke location and sensor site reduce the influence of finite ground conductivity and ionospheric reflections. When one relies on the reasonable assumption that CG strokes emit VLF/LF radiation at low heights (mostly near the ground) while IC discharges must emit from substantial elevation within the clouds, it is a straightforward and conclusive idea to exploit differences of travel times from the two centers to the sensor station (Figure 2). To give an example for the effect size, we note that emission from a source height of 10 km produces a TOA-delay of 3.3 μs (1.7 μs) compared with ground level propagation when recorded at a sensor distance of 50 km (100 km). This difference should be detectable in modern networks, provided that at least one sensor is not further away from the lightning than approximately 100 km.
 The 3-D procedure was implemented in the Munich network and its principal functioning is explained in three steps: first, the usual 2-D locating algorithm is applied to determine ground locations. Second, for the sensor nearest to the lightning the difference between the measured event time and the one expected on the basis of travel time between the 2-D located point of lightning and the sensor is determined. Figure 3 displays the distribution of these time differences for a thunderstorm detected within the network area on May 8, 2003. The 2-D model should suffice to describe data for CG-strokes and one expects both symmetry around dT = 0 and a distribution width compatible with average time errors. However, a pronounced asymmetry becomes evident: whenever the emission center is elevated the signal travel time to the sensor is longer as compared with ground wave propagation; consequently, the arrival time is delayed. Indeed, Figure 3 proves that signals are detected which exhibit the predicted delay and thus identify IC discharges. When ground locations are determined without consideration of source heights one must reckon with either increased errors or inconsistent data sets. Interestingly, common networks often discard data from the sensor closest to the lightning because of saturation effects, pulse shape or other unspecified considerations - according to our reasoning a loss of valuable information.
 In a third step, we allow for longer travel paths in the locating algorithm according to the scheme shown in Figure 2. As expected, this leads to a better matching with measured arrival times and recalculation of the time differences yields a much more symmetric distribution, close to a normal error distribution (Figure 3). For routine online evaluation the quoted three steps are replaced by direct implementation of a height parameter into the optimization algorithm, i.e., the locating procedure is carried out directly with four parameters instead of the three traditional ones (event time and 2-D ground location).
Table 1 presents two independent examples for source height determinations utilizing a 2-stroke flash detected approximately 36 km south of sensor #1. In a 2-D analysis acceptable ground locations were obtained, but the corresponding least squares values approached the upper limit: it turned out that sensor #1 produced TOA-deviations as large as ∼8 μs which can not be reconciled with any of the errors discussed below. When we allow non-zero source heights and calculate expected arrival times by utilizing the hypotenuse TP instead of TH (Figure 2) the inconsistency disappears for both strokes. In this particular example, the resulting heights of ∼16 km are unusual but not implausible because IC discharges of this type (‘narrow negative bipolar pulses’) have been frequently observed with various techniques [Smith et al., 1999, 2004]. Interestingly, the waveforms resemble the ones for return strokes so that both BLIDS/ALDIS and NLDN networks misinterpret the discharges as CG strokes.
Table 1. Example for 2-D Vs 3-D Analysis of Two Strokesa
 It is an important result that the extracted height distributions, depending on the observed thunderstorm cell, usually peak between some 5 and 10 km, in accordance with typical vertical cell dimensions in Germany. Figure 4 shows an example using the data from Figure 3: the network reported 6700 strokes, in each case recorded simultaneously at 4 or more sensors. The described 3-D analysis identified 2800 (∼40%) substantial elevation emissions and the height distribution shows a peak near 8 km. We attribute these strokes to IC discharges. In turn, 3900 strokes (∼60%) were not compatible with substantial source heights; consequently, CG strokes must be found in this group. We suspect that this latter group may contain a relatively small number of unidentified IC discharges due to error sources discussed below.
 It should be pointed out that the derivation of source heights requires no measurements in addition to the ones already performed in common 2-D locating systems. Furthermore, the technique proceeds without any theoretical assumption and involves no adjustable parameter. Application is particularly easy when a stroke has been recorded by at least 4 sensors. In case of only 2 or 3 reporting sensors, TOA-values must be supplemented by bearing angles; when 4 independent measurements are available, the 3 traditional parameters and the emission elevation can be determined. The source height is weakly correlated with the other 3 parameters so that a highly inconsistent parameter set which does not allow proper 2-D locating cannot be remedied by introducing a fictitious source height.
 In assessing the extracted heights it should be kept in mind that the proposed procedure is not aimed at precise elevation mapping; rather, it might be utilized to facilitate distinction between IC and CG lightning. There are mainly 5 effects which could hinder a successful application of the presented 3-D technique:
 1. Statistical time errors. The measured signal times have statistical errors with a mean of the order of 1 μs. In cases of accidental delays larger than 1 μs, long distance to the sensor from the discharge center and unfavorable locating circumstances, an erroneous emission source height can result. Critical evaluation of the experimental data and computer simulations of the network responses indicate that an error of this kind may occur in about 10% of the cases; however, the false heights accumulate near low values and do not exhibit the characteristic peak structure observed for real data.
 2. Site and path errors. It is well known that time delays occur due to location dependent site errors and from propagation effects such as finite conductivity of the ground and topographic structures. However, an influence of this type would affect all signals traveling a similar path. Furthermore, a network with relatively small sensor baselines – as the Munich network – is less affected.
 3. Systematic time errors due to electronic asymmetries. Electronic time delay and signal processing prior to analysis of the digitized waveform may not be identical at different sensor stations. In our network effects of this kind are small because they do not show up in inter-comparisons of the relative time distributions at the various sensors.
 4. Erroneous onset times. It is essential that arrival times are fixed in precisely the same way at each reporting sensor. While simply shaped pulses can be reliably tagged errors may arise in cases with multiple-peak structures; furthermore, pulse shapes may change with increasing propagation path. Our analysis procedure minimizes this error source by taking into account the entire pattern of pulses.
 5. Near field and saturation effects. It is advantageous to exploit signal information from the sensor closest to the reported lightning location. Measurements of the electric field very near a lightning source have shown various near-field effects which may distort pulse shapes as compared to larger distances; we utilize B(t) and did not find any evidence for pulse shifts. Finally, signal saturation can occur. Our time tagging algorithm continues to function and remains particularly reliable when unsaturated structures show up around the saturated main peak. In many cases verification was obtained from the fact that a non-saturated subsequent stroke within an IC flash yielded an identical height value (Table 1).
 The presented pseudo 3-D technique provides a new tool for online recognition of substantial emission source heights and allows discrimination of CG and IC discharges in a relatively large fraction of reported lightning. It can be implemented in any lightning detection network which relies on TOA techniques. Networks with larger baselines, e.g., NLDN, could also benefit from the suggested source height analysis, especially for events which occur within approximately 100 km from a sensor. Of course, further studies are necessary to better verify the correctness of the presented 3-D procedure and to assess the quoted error sources; possibilities are extensive comparisons with results from i) IC/CG assignments in existing VLF/LF networks, and ii) real 3-D radio interferometric systems. The present method is not intended to compete with such 3-D systems which – at much more expense – resolve short steps in lightning channels. Instead, our method is designed to be simple and to allow for a broad application with respect to better identification of IC discharges, given reasonably small sensor baselines and adequate signal time management. Finally, it might be particularly promising to combine the pseudo 3-D technique with waveform discrimination in order to achieve optimized distinction of CG and IC discharges.
 This work was partially supported by the Heidenhain Stiftung and Nowcast Mobile GmbH. The authors thank the German Weather Service for providing sensor locations within a scientific cooperation and are indebted to G. Diendorfer and F. Heidler for valuable discussions.