ATDnet Detection Efficiency and Cloud Lightning Detection Characteristics from Comparison with the HyLMA during HyMeX SOP1
1. Introduction
High-quality real-time lightning data are increasingly important for severe weather monitoring and forecasting. Most of the European countries are covered by a lightning location system (LLS), either operated by the national meteorological service or by a private company or other organization. There is a trend toward international cooperation of ground-based and satellite-based (geostationary) lightning detection. For example, in 2001 several countries started a cooperation named EUCLID (the European Cooperation for Lightning Detection) that, as of 2015, incorporates 149 sensors and covers most of Europe (Schulz et al. 2015). The GOES-R satellite, which will host the Geostationary Lightning Mapper (GLM) instrument (Goodman et al. 2013), is due to be launched in October 2016. The first EUMETSAT Meteosat Third Generation (MTG) satellite is scheduled for launch in 2019 and will include the lightning imager (LI) instrument (MTG-LI) (Biron et al. 2008) that will provide continuous lighting data across most of Europe, Africa, and parts of the Atlantic from space.
Ground-based LLSs typically rely on the detection of radio frequency electromagnetic fields produced by lightning. CG flashes contain one or more strokes that involve a downward leader and an upward return stroke. Return strokes produce powerful radio emissions in the very low-frequency (VLF)/low-frequency (LF) ranges that can be detected by LLSs (e.g., Nag et al. 2015). Detected strokes can be grouped into “flashes” on the basis of predefined spatial and temporal criteria (e.g., Anderson and Klugmann 2014); multiple strokes observed to follow the same lightning channel or proximate channels may be described as a single flash by a human observer. The largest radio emissions of cloud lightning are related to the initial breakdown process (Rakov and Uman 2003).
All LLSs have their limitations and most importantly not all strokes and flashes are detected. Detection efficiency (DE) is used to quantify the fraction of detected strokes or flashes compared to the real number of occurred strokes/flashes. The detection efficiency may be used to refer to cloud-to-ground (CG) or intercloud/intracloud (IC) flashes, leading to a flash DE, or detection efficiency of CG strokes (stroke DE). A flash DE is likely to be higher than a stroke DE, as a flash is considered to be detected even if only one stroke of the flash was detected (e.g., Rakov 2013). Depending on the nature of the reference data, absolute or relative DE could be measured.
The absolute detection efficiency is very difficult to measure, as it requires a method of observation that detects every flash or stroke, in order to be compared with the LLS being assessed. Available studies use rocket-triggered lightning (Jerauld et al. 2005; Chen et al. 2012), strikes to tall structures (Lafkovici et al. 2008; Diendorfer 2010) and video observations (e.g., Idone et al. 1998) as ground truth data. Such samples are often small and contain only CG flashes and strokes. It is also known that the properties of rocket-triggered lightning are different from natural lightning characteristics (Nag et al. 2015). As a result the number of available absolute DE estimations for existing LLSs is limited.
It is often easier to measure the relative DE of one system compared to another—that is, one lightning dataset is taken as “truth”—and the detection efficiency of the other network is calculated by dividing the number of coincident strokes or flashes by the total number detected by the reference network (e.g., Abarca et al. 2010; Lagouvardos et al. 2009). Some regions are covered by multiple operational LLSs, making it possible to compare each system against the other, as is done by Poelman et al. (2013b).
Usually only a ground flash or a stroke DE is measured, as ground-based networks that use VLF/LF radio emissions mainly detect CG strokes, which also pose the greatest threat to lives and infrastructures. For most of the modern short-range LLSs, including EUCLID (Poelman et al. 2013b) and the lightning detection network (LINET) (Betz et al. 2009) in Europe, and the U.S. National Lightning Detection Network (NLDN) (Biagi et al. 2007; Nag et al. 2011) in America, ground flash DE above 90% has been measured in their core regions. Cloud flash DE is rarely measured, as it can be more difficult to quantify the true number of IC discharges. However, in a recent study Murphy et al. (2014) have estimated that the upgraded NLDN is capable of detecting 30%–58% of cloud flashes.
Much less is known about the DE of long-range LLSs. There are fewer long-range LLSs than short-range systems: examples include the University of Washington Worldwide Lightning Location Network (WWLLN) system (Lay et al. 2004), the Vaisala Global Lightning Dataset 360 (GLD360) system (Said et al. 2010), the National Observatory of Athens ZEUS network (Lagouvardos et al. 2009), Sferics Timing and Ranging Network (STARNET) (Dentel et al. 2014), and the Arrival Time Difference Network (ATDnet) operated by the Met Office. Such systems can cover whole continents as well as seas and oceans that are not observable using short-range networks. This means that they can provide valuable data for intercontinental flight routes, for example. There is also an increased interest in using long-range LLS data for the calibration/validation of geostationary lightning sensors in the future. However, a better understanding about the capabilities and limitations of long-range LLSs is required.
In the present paper, the ATDnet flash DE is measured against the HyMeX Lightning Mapping Array (HyLMA). ATDnet is a VLF long-range LLS operated by the Met Office (Bennett et al. 2011). During the study period, ATDnet consisted of 11 sensors in and around Europe operating in the central frequency of 13.733 kHz. The effective range of the system encompasses Europe, northern Africa, and northern parts of the Atlantic.
ATDnet sensors detect “atmospherics” (sferics). Sferics are electromagnetic waves in the VLF range that propagate in the earth–ionosphere waveguide and can be generated by CG return strokes (Rakov and Uman 2003). The system takes the advantage of the long propagation paths of sferics, which makes it possible to cover large areas with only a limited number of sensors.
ATDnet detections are referred to as “fixes.” Fixes are located using data from a minimum of four ATDnet sensors (also referred to as outstations). The long baselines between ATDnet sensors mean that only relatively powerful radio emissions will be detected. As CG strokes tend to emit the most powerful sferics in the VLF range, whereas cloud lightning processes generally emit weaker sferics (e.g., Cummins and Murphy 2009), it is often assumed that long-range LLSs are only capable of detecting CG lightning.
Studies into other long-range VLF networks have demonstrated recently that this assumption is oversimplified, however, and that VLF networks are in fact capable of detecting some IC discharges (Jacobson et al. 2006). The long-range system ZEUS clearly detects ICs but with lower DE because ICs tend to emit weaker sferics than CGs (Lagouvardos et al. 2009). A recent comparison of ATDnet with the Météorage short-range network in France demonstrated that ATDnet detected a large proportion of flashes that the short-range network categorized as IC discharges (Poelman et al. 2013a), indicating but not proving that ATDnet detects IC discharges.
As such, there is evidence that some fraction of ICs is detected by ATDnet. A quantitative measure of the extent of fixes corresponding to CG and IC flashes in ATDnet data, however, requires observations that can be achieved only through direct observational studies, or by using specialized observational techniques.
The best available system for detailed total lightning data with nearly 100% DE is the lightning mapping array (LMA) developed by the New Mexico Institute of Mining and Technology (Rison et al. 1999; Krehbiel et al. 2000). The LMA detects leader steps (commonly referred to as source points) in developing lightning channels in the 60–66-MHz frequency range, with a source point location rate of up to more than 10 000 points per second. The system has three-dimensional location accuracy within the area covered by the array of only a few tens of meters or better. As such, it can be used to reliably map out the three-dimensional development and structure of a lightning channel.
The system has some limitations that need to be considered when using the LMA data. Very high-frequency sensors can only detect source points that are in line of sight—that is, not beyond the horizon or obscured by terrain—thus, LMA range is limited by the vicinity of sensors. Negative leaders radiate much more strongly than positive leaders, and negative and positive lightning channels often develop at the same time. Thus, some underestimation of positive channels is inevitable (Defer et al. 2015). Also, an additional effort is needed to group individual LMA source points into flashes and to determine the type of flashes.
As part of the HyMeX project (Drobinski et al. 2014), a lightning mapping array (HyLMA) was deployed for over two months in the south of France (Fig. 1). The HyLMA consisted of 12 sensors and was operational during the HyMeX project Special Observation Period 1 (SOP1), which lasted from 5 September to 6 November 2012 (Defer et al. 2015). The obtained data could be used as ground truth data, as a reliable reference against which LLSs could be validated. Thus, it offers a unique opportunity to examine the ATDnet capabilities and limitations near the core region of the network.
(a) Locations of the HyMeX SOP1 LMA sensors as red squares and (b) border of the study area as a black rectangle.
Citation: Journal of Atmospheric and Oceanic Technology 33, 9; 10.1175/JTECH-D-15-0256.1
The main objective of the present paper is to provide the first comprehensive quantitative estimation of ATDnet flash detection efficiency. The main focus is on IC detection, in order to investigate how IC characteristics affect ATDnet capability to detect them. Variations in IC DE between different storms and during the same storm are also examined in order to understand whether they are caused by variations in IC characteristics or related to system limitations.
Section 2 describes the approach used in comparing HyLMA and ATDnet data, and section 3 represents the main results. Section 4 provides a discussion of the results, and section 5 concludes the study.
2. Data and methods
a. Data
The ATDnet data were retrieved from the Met Office archive. The data included a date, time, 2D location (latitude and longitude), and error estimation for each ATDnet fix. The error estimation is computed as the root-mean-square of the major axis and the minor axis of the error ellipse in which a fix is located with a 95% probability. ATDnet reports fixes to 0.1-μs precision. No discrimination between cloud and ground lightning is provided. Only “good” fixes that pass predefined error estimation and waveform quality criteria are used in ATDnet data products. All ATDnet good fixes in the study area during the study period were included in the analysis.
The LMA data were obtained from the HyMeX SOP1 online database (Rison 2012). The ASCII data files consisted of individual LMA source points, including their times, 3D locations (latitude, longitude, and altitude), chi-squared errors, and contributing stations. The time resolution of the data was 1 ns and the source points were not grouped into flashes. At least six contributing stations are needed in order to locate an LMA source point (Defer et al. 2015). In the present study, any LMA source point with fewer than seven contributing stations or the maximum chi-squared error more than five were excluded from the analysis. The seven-station threshold was selected to exclude the source points with only the minimum number of contributing stations and to increase the confidence in the dataset. A higher threshold was avoided in order to preserve enough source points for reliable interpretation of lightning channels.
b. Selection of storms
The initial stage of the analysis was to find storms with suitable data. Daily ATDnet activity was reviewed in order to determine the level of storm activity from 1 September to 6 November 2012, corresponding to a period of a few days before the start of HyMeX SOP1 when the HyLMA was already active to the end of SOP1. The majority of days showed no activity at all. Several further days showed little activity, or storms that were poorly located relative to the LMA network, and as such would not provide sufficiently reliable data. In the end, six candidate days were chosen for further investigation: 5, 11, and 24–26 September, and 26 October 2012.
Of these storms, it was necessary to find storms where there was sufficient activity to provide good statistics, but also few enough flashes to avoid problems due to the large amount of temporally overlapping flashes, making individual flashes difficult to distinguish. The storm on 24 September was extremely intense and hence difficult to analyze. On 26 September the activity was generally outside of the LMA, and on 26 October the data were contaminated by a large number of spurious source points, making it difficult to distinguish between real flashes and LMA errors. The remaining three storms on 5, 11, and 25 September 2012 were found to be suitable for the study. See Table 1 for details of the storms.
Main characteristics of the studied storms.
c. Processing the LMA data
The study area is located inside the array of HyLMA sensors between 43.7°–44.5°N and 3.7°–4.7°E (Fig. 1b). This helped to avoid poorly located source points outside of the array. For LMA data processing and analysis, the study area was divided into a 0.01° × 0.01° grid (corresponding to a box size of approximately 1.11 km × 0.80 km at this latitude), referred to as the grid.
Two Python scripts were developed to group LMA source points into LMA flashes. The first returned the number of LMA source points per grid box for every 0.1 s of data. Data for all the boxes of the grid with at least one LMA source point were saved into a file with time and location information.
The second script searched for LMA flashes in the output files of the first script. The gridcell data were processed in time order. If at least one grid cell with four or more LMA source points was found during a 0.1-s time bin, then a new LMA flash was started. If one or more cells with at least four LMA source points were found during the next 0.1-s time bin, then the LMA flash was continued. The flash was continued until a time bin with no grid cells with four or more LMA source points was met. The time of the flash, its duration, and total number of the source points were then saved into an LMA flash file.
To ensure that the time and space thresholds used had no negative impact on the results, all of the computed flashes were later checked manually by plotting their LMA source points (a detailed description below). It was found that there were a few cases when the scripts had failed to discriminate between coincident flashes in the different parts of the study area. All such examples were corrected manually to avoid the results being affected by errors in source point grouping.
Next, plots of the unprocessed LMA source points were created for each LMA flash. Each figure included plots of latitude versus height, longitude versus height, and latitude versus longitude, along with a plot of the time-versus-altitude evolution of an LMA flash (Figs. 2 and 3). The 3D structure of the flash could be determined from the three different 2D subplots. If one or more concurrent ATDnet fixes were found, then their times, locations, and error ellipses were plotted on the source plot of the corresponding LMA flash.
An example source plot for a LMA CG flash detected by ATDnet on 25 Sep 2012. The LMA source points are represented as time colored dots. The concurrent ATDnet fixes are depicted as triangles on the altitude plots and as hexagons with surrounding error ellipses on the plan view. Note that the cloud part and the ground part of the flash were detected by ATDnet.
Citation: Journal of Atmospheric and Oceanic Technology 33, 9; 10.1175/JTECH-D-15-0256.1
An example source plot for an LMA IC flash detected by ATDnet on 25 Sep 2012. The LMA source points are represented as time colored dots. The concurrent ATDnet fix is depicted as a triangle on the altitude plots and as a hexagon with the surrounding error ellipse on the plan view. Note that the time of the ATDnet fix corresponds to the IB.
Citation: Journal of Atmospheric and Oceanic Technology 33, 9; 10.1175/JTECH-D-15-0256.1
All of the plots of LMA source points were then examined visually to determine flash type. If a flash exhibited many LMA source points below 2 km with a clear time evolution of a ground channel (or channels), then it was classified as a ground flash (CG). All the flashes with at least one ground contact were considered to be CGs, even if the ground contact was preceded or followed by significant cloud activity (Fig. 2). Only flashes without obvious LMA activity below 2 km were classified as cloud flashes (IC). Both intracloud (within the same storm cell) and intercloud (between different storm cells) flashes were included in the IC sample (Fig. 3), as it was not possible to distinguish between these types without additional information about the cloud extent.
For some flashes that extended into low altitudes, the number of LMA source points below 2 km was rather low and/or the clear time evolution of a ground channel was missing, making it difficult to determine whether a ground contact took place. All such LMA flashes were classified as unclear (U).
A reduction in the quality of the data was observed for many of the flashes that were located outside the array of LMA sensors. Source points detected beyond the limits of the LMA array would generally have a higher location uncertainty. They would also be less likely to be detected by the LMA sensors due to the increased distance between the source and the sensors. The line-of-sight requirement for the detection of source points outside of the network also meant that low-altitude points would be less likely to be detected by a sufficient number of sensors.
Such flashes frequently exhibited large spatial dispersion of LMA source points and thus appeared incoherent on the source point plots. This would have caused problems in type classification and resulted in large quantities of U flashes. To prevent this, the study area was selected so that it was inside the array of LMA sensors. To be included in our sample, a flash had to be at least partly within the study area.
d. Estimating ATDnet detection efficiency
Visual examination of all the source point plots was carried out to determine which flashes were detected by ATDnet. If a concurrent ATDnet fix was found for an LMA flash and the error ellipse of the ATDnet fix at least partly overlapped with the source points of the LMA flash, then the LMA flash was considered to be detected by ATDnet. Concurrent observations with no LMA source points within the error ellipse of the ATDnet fix were normally counted as ATDnet misses. However, there were a few exceptional cases when a small error ellipse of an ATDnet fix was surrounded by branches of an LMA flash but none of the source points fell within the error ellipse. Those cases were also counted as ATDnet matches.
If more than one ATDnet fix corresponded to one LMA flash (e.g., subsequent strokes of a CG or different branches of an IC were detected), then it was still counted as one ATDnet match. Sometimes not only the return stroke(s) but also the cloud processes of CGs were detected by ATDnet. All those flashes were still counted only once as detected CGs.
As ATDnet’s capability to detect ICs is the main focus of the study, a more detailed investigation was carried out for the cloud flashes. ATDnet’s ability to detect an early and often vertical part of ICs—that is, initial breakdown—was examined. If a concurrent ATDnet observation existed within 0.1 s of the beginning of a detected IC, then it was considered as initial breakdown detection (Fig. 3).
ATDnet IC DE was also studied as a function of LMA source points per flash. Moreover, ATDnet’s capability to detect ICs with different maximum and minimum altitudes and vertical extent was examined. The maximum and minimum altitudes of a flash were determined as the 90th and 10th percentiles of all the source point altitudes, respectively. The vertical extent of a flash was computed as the difference between the maximum and minimum altitude.
In addition, the effect of the horizontal extent of ICs on ATDnet DE was studied. The grid was used for measuring the horizontal extent of flashes. The flash area was computed as a sum of areas of all the boxes of the grid with four or more LMA source points inside.
To address the varying numbers of flashes within the different flash property ranges, 95% confidence intervals were added to average detection efficiencies in the IC characteristics versus DE plots. There is a 95% probability that the true DE of ATDnet under these conditions would fall within these confidence intervals.
3. Results
a. Flash statistics and ATDnet DE
In total, 1891 LMA flashes were included in the study. Of these, 1324 were determined to be IC flashes and 281 were determined to be CG flashes (Table 2). The remaining 286 flashes, which constituted 15.1% of the total sample, were classified as unclear. For the individual storms, the percentage of unclear flashes varied from 12% to 26%.
Number of LMA and ATDnet flashes, ATDnet DE, and percentage of IC flashes for the three storms studied.
The differing nature of the studied storms (Table 1) is evident from the number of flashes and IC fractions (Table 2). About 73% of all the LMA flashes occurred on 25 September and only 5.6% on 5 September 2012. Similar frequencies of ICs and CGs were observed on 5 September, whereas the other two storms had a much higher proportion of ICs. On 11 September over 82% and on 25 September 2012 nearly 85% of all the classified flashes were ICs. On the basis of the literature, 75%–80% of all flashes are ICs (Prentice and Mackerras 1977). Thus, for the two most intense storms, the IC frequency was somewhat higher than for an average storm. As a result, the IC sample is quite large, which allows for a more reliable estimation of ATDnet IC detection capability.
The overall ATDnet DE was approximately 89% for CGs and 24% for ICs (Table 2). The stability of CG DE between the storms is notable, ranging from 86.8% to 89.3%. For ICs, both of the more intense storms gave a DE of approximately 23%–24%. The smaller sample on 5 September 2012 gave an approximately 10% higher IC DE. It is demonstrated later that it was not just a random deviation due to a small sample size but the characteristics of ICs on 5 September 2012 were different.
The highest overall DE of 56% was observed on 5 September 2012 when the IC percentage was the lowest. For the other two storms, the ATDnet overall DE was about 20% lower, which is in accordance with a much smaller fraction of more easily detectable CGs (Table 2). It should be noted that the overall DE is computed by using all the LMA flashes, including unclear flashes, as the flash type is not important here.
b. DE and IC characteristics
For the majority of ICs registered by ATDnet, the initial breakdown (Nag et al. 2009) was detected (Table 3). The initial breakdown (IB) was often the only detected part of an IC. Only rarely was the IB detection followed by additional ATDnet fixes corresponding to later stages of IC flashes.
Total of ICs and IBs detected by ATDnet and ATDnet IB DE.
It was also found that ATDnet IC DE was much higher for flashes with larger vertical extent (Fig. 4). None of the flashes vertically shorter than 1 km was detected, whereas for the flashes with a vertical extent over 4.5 km, DE was around 45%. Meanwhile the general altitude of flashes did not affect ATDnet’s capability to detect them, and the DE did not vary much between low and high flashes (Fig. 5).
ATDnet IC DE as a function of the vertical extent of the flashes (black) and its 95% confidence intervals (orange).
Citation: Journal of Atmospheric and Oceanic Technology 33, 9; 10.1175/JTECH-D-15-0256.1
ATDnet IC DE as a function of the average LMA source point altitude of the flashes (black) and its 95% confidence intervals (orange).
Citation: Journal of Atmospheric and Oceanic Technology 33, 9; 10.1175/JTECH-D-15-0256.1
Flashes with lower minimum or higher maximum altitude were more easily detected by ATDnet (Fig. 6), but the data shown in Fig. 5 demonstrates that the average altitude of a flash is relatively unimportant. Instead, ICs with higher maximum or lower minimum altitude are more likely to have a greater vertical extent and thus more effectively detected. This assumption was generally confirmed by reviewing the plots of LMA sources (not represented here).
ATDnet IC DE as a function of the (a) minimum and (b) maximum altitude of the flashes (black) and its 95% confidence intervals (orange).
Citation: Journal of Atmospheric and Oceanic Technology 33, 9; 10.1175/JTECH-D-15-0256.1
Flashes with a larger horizontal area (Fig. 7) and more LMA source points (Fig. 8) were detected more frequently. However, these findings do not necessarily mean that ATDnet IC DE is directly dependent on those characteristics, as the vertical extent, the horizontal extent, and the number of LMA source points are all related to each other. Flashes with a longer vertical extent usually have more LMA source points and greater horizontal extent, and all of these characteristics are likely to be linked to the total energy released, which is not quantified here. Thus, even if IC DE depends on only one of the three factors, it may seem to be related also to the other two factors. It is demonstrated below that the vertical extent had the greatest impact on ATDnet IC DE.
ATDnet IC DE as a function of the horizontal area of flashes (black) and its 95% confidence intervals (orange).
Citation: Journal of Atmospheric and Oceanic Technology 33, 9; 10.1175/JTECH-D-15-0256.1
ATDnet IC DE as a function of the number of LMA source points per flash (black) and its 95% confidence intervals (orange).
Citation: Journal of Atmospheric and Oceanic Technology 33, 9; 10.1175/JTECH-D-15-0256.1
c. Time evolution of IC DE
The time evolution of IC DE was studied for the storm on 25 September 2012, which peaked in intensity from 1020 to 1250 UTC. During that time the IC frequency was high enough to provide useful statistics in small time windows with a 5-min time step (Fig. 9). ATDnet IC DE varied from around 10% to 50% between the individual time bins. Variations in DE were accompanied by changes in the vertical and horizontal extent and the average source point number of flashes.
(a) Total number of LMA flashes, (b) ATDnet IC DE vs vertical extent, (c) number of LMA source points, and (d) horizontal extent of LMA flashes for 5-min time bins during 1020–1250 UTC 25 Sep 2012. The black line in (b)–(d) represents ATDnet IC DE.
Citation: Journal of Atmospheric and Oceanic Technology 33, 9; 10.1175/JTECH-D-15-0256.1
There was a positive relationship between the IC DE and the vertical extent of flashes with a Pearson’s correlation coefficient of r = 0.50 (Fig. 9b). Meanwhile, the relationships between IC DE and the number of LMA source points (Fig. 9c) and between IC DE and the horizontal extent of flashes (Fig. 9d) were less well defined with r = 0.20 and r = 0.10, respectively. Changes in IC frequency during the study period (Fig. 9a) did not strongly affect ATDnet DE (r = −0.16).
A notable period in the time evolution of DE was between 1215 and 1230 UTC. During this period the vertical extent of flashes decreased and so did the DE (Fig. 9b). At the same time the horizontal extent (Fig. 9d) and LMA source point number (Fig. 9c) increased. This situation hints that changes in the vertical extent of ICs affect ATDnet DE more than changes in the horizontal extent and LMA source point number.
Another interesting finding is that ATDnet general sensitivity to the ICs was probably higher at the beginning of the 2.5-h period. It can be seen that until around 1045 UTC, the IC DE was relatively high compared to the vertical extent of flashes (Fig. 9b).
It was also checked whether IC DE was related to the overall number of incoming waveforms from all over the network to eliminate the possibility that an abundance of waveforms temporarily overloads the ATDnet central processor and leads to lower DE. The analysis showed that there was no relationship between the number of incoming waveforms and ATDnet IC DE during the studied period (r = 0.03).
As the vertical extent of ICs was found to affect ATDnet DE more clearly than other flash properties, it was averaged for all the studied storms to check whether the differences in IC DE (Table 2) were related to variations in the vertical extent. The mean vertical extent of ICs on 5 September was 3.5 km, which is greater than 3.1 km on 11 September and 3.0 km on 25 September. This difference could explain the somewhat higher IC DE for the storm on 5 September.
4. Discussion
ATDnet CG DE was found to be 89% (Table 2) for the storms analyzed in this study, which is quite high for a long-range LLS. For comparison, the Vaisala GLD360 system is estimated to detect 86%–92% CGs over the continental United States (Demetriades et al. 2010), ZEUS was found to be capable of detecting approximately 25% of CGs over central and western Europe (Lagouvardos et al. 2009), and WWLLN DE was estimated to be 10.3% over the contiguous United States in 2009 (Abarca et al. 2010).
On the basis of the current ATDnet configuration (locations of sensors), it can be assumed that the performance demonstrated in this analysis is representative of large areas in Europe. In fact the HyLMA was located well within the perimeter of the ATDnet but slightly south of its core region with the four nearest ATDnet sensors located 370, 935, 1040, and 1090 km away . Thus, in the ATDnet core region, including south of the United Kingdom, north of France, the Netherlands, and Belgium, DE may be even higher as there are four ATDnet sensors within a radius of 700 km.
The vast majority of the sferic propagation paths between the study area and the nearest ATDnet sensors are over land, which is characterized by lower conductivity and thus higher attenuation of VLF signals. The same is true for most of Europe, so similar DE could be expected for other areas within the perimeter of ATDnet, including central and western Europe. Small spatial variations in DE due to surface features such as mountains are probable over the whole ATDnet range, but it is not possible to estimate their impact on the basis of one study area.
It must be kept in mind that ATDnet is not a global LLS. As its operational sensors are located in and around Europe, its DE is expected to gradually decrease with increasing distance from Europe. Thus, there are large areas in the world where “global” networks like GLD360 and WWLLN are expected to perform better than ATDnet (it should be noted that global networks will still have regional variability in their performance and DE). A comparison with those networks might be of benefit in the future to more precisely quantify the spatial range of ATDnet.
The results revealed that ATDnet is capable of detecting 24% of IC lightning. It is known that other VLF systems, such as WWLLN (Rodger et al. 2006) and ZEUS (Lagouvardos et al. 2009), also detect some cloud lightning but their exact IC DE is not available. Sufficiently sensitive short-range networks are capable of detecting more ICs than ATDnet, as their shorter baselines between sensors can more easily detect IC radio emissions that are generally weaker. For example, NLDN DE has reported to be 30%–58% for ICs (Murphy et al. 2014).
The analysis showed that 67% of the ATDnet IC detections were IB detections (Table 3). It is not surprising, as the strongest radio emissions in the IC are associated with IB (Rakov and Uman 2003). Thus, the related sferics can be detected more easily by a relatively sparse network of sensors. Moreover, ATDnet’s capability to detect IBs at VLF is in good agreement with the finding that IBs are often detected simultaneously by VHF and VLF/LF measurements (e.g., Betz et al. 2008).
It was demonstrated that IC characteristics affected ATDnet DE. Flashes with greater vertical (Fig. 4) and horizontal (Fig. 7) extent and more LMA source points (Fig. 8) were more easily detected. The fact that the three parameters were found to be interrelated made it more difficult to say which of them is paramount. The correlation coefficient analysis suggested that the vertical extent is more important than other IC properties. The idea is further supported by the time evolution of IC DE on 25 September 2012.
There was an interesting episode on 25 September 2012 when relatively flat flashes with large horizontal extent and numerous LMA source points prevailed between 1220 and 1230 UTC (Fig. 9). Meanwhile, the flashes were vertically short and more detailed visual examination showed that the vertical IB was sometimes completely missing. As the episode was characterized by low ATDnet IC DE, it indicates that the absence of long vertical lightning channels lowered ATDnet DE, whereas the presence of long horizontal channels did not compensate the effect.
It is also worth mentioning that the vertical whip antennas used by ATDnet sensors are expected to favor detection of vertically polarized radio emissions. Long vertical channels of ICs probably emit vertically polarized sferics that are able to trigger the sensors far enough to meet the four contributing sensor criteria necessary for ATDnet fix location.
Vertically extensive ICs are probably more powerful, as the positive and negative charge regions in a thundercloud are farther apart and stronger charges are needed to generate electric field strong enough to exceed the resistance of the air between the charge regions. The resulting IC is more powerful and a larger amount of charge is transferred via the vertical channel. A stronger vertical channel in turn emits more powerful sferics.
It is likely that vertically extensive ICs are characterized by higher peak currents. Unfortunately, it was not possible to examine the relationship between the peak current of cloud flashes and ATDnet DE, as neither ATDnet nor HyLMA provides estimates of the currents of the detected ICs.
The fact that IC DE is significantly affected by flash characteristics means that DE is not a fixed parameter even if the network setup and propagation conditions are considered to be stable. The results demonstrated that ATDnet IC DE varied not only between the individual storms but also during the same storm and that those variations were in accordance with changing IC properties (Fig. 9). Much larger IC DE variations were observed in the course of the same storm than between the three storms. A wider study with more storms included is needed to estimate how much IC properties and DE can actually vary within and between storms.
The potential influence of changing propagation conditions on ATDnet DE can also be seen on the background of abrupt DE changes related to flash properties. During the study period on 25 September 2012, even vertically shorter flashes were detected rather efficiently between 1020 and 1045 UTC (Fig. 9b). Later at around 1120, 1140, and 1210 UTC, vertically longer flashes were needed to achieve similarly good IC DE. It is possible that such variability can be at least partly accounted for the changes in propagation conditions.
It should be taken into account that the representativeness of the results is somewhat constrained by the small size of the data sample. All three studied storms occurred at daytime and in the same season, that is, late summer/early autumn. Thus, even the good agreement between ATDnet IC and especially CG DE of the studied storms does not necessarily mean that such a level of performance is always the case.
Further study with a larger sample of storms is needed to estimate the influence of changes in propagation conditions in the wider diurnal scale. ATDnet DE is expected to be lower at night due to modal interference between sky wave propagation modes 1 and 2 (Bennett et al. 2011). Future studies should include nighttime storms to quantify the reduction in DE.
In addition to diurnal changes, seasonal changes in DE may exist, especially if ATDnet sensors are saturated during high-activity events in summer. The results revealed no saturation problems during the three examined storms. However, all the storms occurred in September, when lightning activity in Europe and in the Northern Hemisphere is already lower than in summer. Thus, additional DE measurements during periods with exceptionally high activity in the ATDnet range are necessary in order to check whether temporary saturation is a problem in summer.
Larger data samples would also allow for making more reliable conclusions about the relationships between IC characteristics and DE by reducing the width of the confidence intervals. For example, on the basis of analyzed data, it becomes less clear if average DE continues to increase for the flashes vertically longer than 4 km, as the confidence intervals are larger than the differences between the average values (Fig. 4).
It would also be interesting to carry out a similar study with other European lightning location networks included. This would give invaluable information about the capabilities and limitations of different short-range and long-range detection methods. Although the HyMeX LMA was only temporarily deployed, there are more permanent LMAs in Corsica (Coquillat et al. 2015) and in Spain (van der Velde et al. 2012) that could potentially provide useful data for such studies.
5. Conclusions
The Met Office ATDnet long-range VLF lightning location system was validated against the HyLMA deployed in the south of France as part of the HyMeX project Special Observation Period 1. The results revealed that ATDnet detected 89% of the 281 ground flashes detected by the HyLMA over three storm events. In addition, ATDnet detected 24% of the 1324 HyLMA cloud flashes during the three storms. This finding confirms that ATDnet is capable of detecting ICs. This is an important capability, as the majority of lightning discharges are ICs and some storms do not produce CGs at all. On the basis of the results, it can be concluded that ATDnet is able to locate storms with low or no CG activity.
It was demonstrated that ICs with greater vertical and horizontal extent and more LMA sources were more effectively detected by ATDnet. Most of the IC detections were found to be initial breakdown detections, and there was a positive relationship between DE and the vertical extent of ICs. These findings show that the vertical extent of ICs has the greatest impact on DE of the measured characteristic. A more comprehensive study with a larger data sample would be needed to better understand the relationship between flash characteristics and ATDnet detection efficiency.
Acknowledgments
The LMA data were obtained from the HyMeX program, sponsored by Grants MISTRALS/HyMeX and ANR-11-BS56-0005 IODA-MED. We thank Eric Defer for his useful comments and suggestions in the early stage of the study. We also thank the two reviewers of the manuscript for their valuable suggestions.
REFERENCES
Abarca, S. F., Corbosiero K. L. , and Galarneau T. J. , 2010: An evaluation of the Worldwide Lightning Location Network (WWLLN) using the National Lightning Detection Network (NLDN) as ground truth. J. Geophys. Res., 115, doi:10.1029/2009JD013411.
Anderson, G., and Klugmann D. , 2014: A European lightning density analysis using 5 years of ATDnet data. Nat. Hazards Earth Syst. Sci., 14, 815–829, doi:10.5194/nhess-14-815-2014.
Bennett, A. J., Gaffard C. , Nash J. , Callaghan G. , and Atkinson N. C. , 2011: The effect of modal interference on VLF long-range lightning location networks using the waveform correlation technique. J. Atmos. Oceanic Technol., 28, 993–1006, doi:10.1175/2011JTECHA1527.1.
Betz, H.-D., and Coauthors, 2008: Detection of in-cloud lightning with VLF/LF and VHF networks for studies of the initial discharge phase. Geophys. Res. Lett., 35, L23802, doi:10.1029/2008GL035820.
Betz, H.-D., Schmidt K. , Laroche P. , Blanchet P. , Oettinger W. P. , Defer E. , Dziewit Z. , and Konarski J. , 2009: LINET—An international lightning detection network in Europe. Atmos. Res., 91, 564–573, doi:10.1016/j.atmosres.2008.06.012.
Biagi, C. J., Cummins K. L. , Kehoe K. E. , and Krider E. P. , 2007: National Lightning Detection Network (NLDN) performance in southern Arizona, Texas, and Oklahoma in 2003–2004. J. Geophys. Res., 112, D05208, doi:10.1029/2006JD007341.
Biron, D., De Leonibus L. , Laquale P. , Labate D. , Zauli F. , and Melfi D. , 2008: Simulation of Meteosat Third Generation-Lightning Imager through Tropical Rainfall Measuring Mission: Lightning Imaging Sensor data. Remote Sensing System Engineering, P. E. Ardanuy and J. J. Puschell, Eds., International Society for Optical Engineering (SPIE Proceedings, Vol. 7087), 708706, doi:10.1117/12.794764.
Chen, L., Zhang Y. , Lu W. , Zheng D. , Zhang Y. , Chen S. , and Huang Z. , 2012: Performance evaluation for a lightning location system based on observations of artificially triggered lightning and natural lightning flashes. J. Atmos. Oceanic Technol., 29, 1835–1844, doi:10.1175/JTECH-D-12-00028.1.
Coquillat, S., Defer E. , Lambert D. , Martin J.-M. , Pinty J.-P. , Pont V. , and Prieur S. , 2015: SAETTA: Fine-scale observation of the total lightning activity in the framework of the CORSiCA atmospheric observatory. Geophysical Research Abstracts, Vol. 17, Abstract EGU2015-10123. [Available online at http://meetingorganizer.copernicus.org/EGU2015/EGU2015-10123.pdf.]
Cummins, K., and Murphy M. , 2009: An overview of lightning locating systems: History, techniques, and data uses, with an in-depth look at the U.S. NLDN. IEEE Trans. Electromagn. Compat., 51, 499–518, doi:10.1109/TEMC.2009.2023450.
Defer, E., and Coauthors, 2015: An overview of the lightning and atmospheric electricity observations collected in southern France during the HYdrological cycle in Mediterranean EXperiment (HyMeX), Special Observation Period 1. Atmos. Meas. Tech., 8, 649–669, doi:10.5194/amt-8-649-2015.
Demetriades, N. W. S., Murphy M. J. , and Cramer J. A. , 2010: Validation of Vaisala’s Global Lightning Dataset (GLD360) over the continental United States. Extended abstracts, 21st Int. Lightning Detection Conf./Third Int. Lightning Meteorology Conf., Orlando, FL, Vaisala. [Available online at http://www.vaisala.com/Vaisala%20Documents/Scientific%20papers/6.Demetriades,%20Murphy,%20Cramer.pdf.]
Dentel, L. M., da Rocha B. R. P. , and de Souza J. R. S. , 2014: Evaluation of STARNET lightning detection performance in the Amazon region. Int. J. Remote Sens., 35, 115–126, doi:10.1080/01431161.2013.862604.
Diendorfer, G., 2010: LLS performance validation using lightning to towers. 21st Int. Lightning Detection Conf./Third Int. Lightning Meteorology Conf., Orlando, FL, Vaisala, [Available online at http://www.vaisala.com/VaisalaDocuments/Scientificpapers/1.Keynote-Diendorfer.pdf.]
Drobinski, J., and Coauthors, 2014: HyMeX: A 10-year multidisciplinary program on the Mediterranean water cycle. Bull. Amer. Meteor. Soc., 95, 1063–1082, doi:10.1175/BAMS-D-12-00242.1.
Goodman, S. J., and Coauthors, 2013: The GOES-R Geostationary Lightning Mapper (GLM). Atmos. Res., 125126, 34–49, doi:10.1016/j.atmosres.2013.01.006.
Idone, V. P., Davis D. A. , Moore P. K. , Wang Y. , Henderson R. W. , Ries M. , and Jamason P. F. , 1998: Performance evaluation of the U.S. National Lightning Detection Network in eastern New York: 1. Detection efficiency. J. Geophys. Res., 103, 9045–9055, doi:10.1029/98JD00154.
Jacobson, A. R., Holzworth R. , Harlin J. , Dowden R. , and Lay E. , 2006: Performance assessment of the World Wide Lightning Location Network (WWLLN), using the Los Alamos Sferic Array (LASA) as ground truth. J. Atmos. Oceanic Technol., 23, 1082–1092, doi:10.1175/JTECH1902.1.
Jerauld, J., Rakov V. A. , Uman M. A. , Rambo K. J. , Jordan D. M. , Cummins K. L. , and Cramer J. A. , 2005: An evaluation of the performance characteristics of the U.S. National Lightning Detection Network in Florida using rocket-triggered lightning. J. Geophys. Res. Atmospheres, 110, doi:10.1029/2005JD005924.
Krehbiel, P. R., Thomas R. J. , Rison W. , Hamlin T. , Harlin J. , and Davis M. , 2000: GPS-based mapping system reveals lightning inside storms. Eos, Trans. Amer. Geophys. Union, 81, 21–25, doi:10.1029/00EO00014.
Lafkovici, A., Hussein A. , Janischewskyj W. , and Cummins K. , 2008: Evaluation of the performance characteristics of the North American Lightning Detection Network based on tall-structure lightning. IEEE Trans. Electromagn. Compat., 50, 630–641, doi:10.1109/TEMC.2008.927922.
Lagouvardos, K., Kotroni V. , Betz H.-D. , and Schmidt K. , 2009: A comparison of lightning data provided by ZEUS and LINET networks over Western Europe. Nat. Hazard Earth Sys. Sci., 9, 1713–1717, doi:10.5194/nhess-9-1713-2009.
Lay, E. H., Holzworth R. H. , Rodger C. J. , Thomas J. N. , Pinto O. Jr., and Dowden R. L. , 2004: WWLL global lightning detection system: Regional validation study in Brazil. Geophys. Res. Lett., 31, L03102, doi:10.1029/2003GL018882.
Murphy, M. J., Nag A. , Cramer J. A. , and Pifer A. E. , 2014: Enhanced cloud lightning performance of the U.S. National Lightning Detection Network following the 2013 upgrade. Extended abstracts, 23rd Int. Lightning Detection Conf./Fifth Int. Lightning Meteorology Conf., Tucson, AZ, Vaisala. [Available online at http://www.vaisala.com/Vaisala%20Documents/Scientific%20papers/2014%20ILDC%20ILMC/ILDC-Wednesday/Murphy%20et%20al-Improved%20NLDN%20Performance%20after%202013%20Upgrade-2014-ILDC-ILMC.pdf.]
Nag, A., DeCarlo B. A. , and Rakov V. A. , 2009: Analysis of microsecond- and submicrosecond-scale electric field pulses produced by cloud and ground lightning discharges. Atmos. Res., 91, 316–325, doi:10.1016/j.atmosres.2008.01.014.
Nag, A., and Coauthors, 2011: Evaluation of U.S. National Lightning Detection Network performance characteristics using rocket-triggered lightning data acquired in 2004–2009. J. Geophys. Res., 116, D02123, doi:10.1029/2010JD014929.
Nag, A., Murphy M. J. , Schulz W. , and Cummins K. L. , 2015: Lightning locating systems: Insights on characteristics and validation techniques. Earth Space Sci., 2, doi:10.1002/2014EA000051.
Poelman, D. R., Honoré F. , Anderson G. , and Pedeboy S. , 2013a: Comparing a regional, subcontinental, and long-range lightning location system over the Benelux and France. J. Atmos. Oceanic Technol., 30, 2394–2405, doi:10.1175/JTECH-D-12-00263.1.
Poelman, D. R., Schulz W. , and Vergeiner C. , 2013b: Performance characteristics of distinct lightning detection networks covering Belgium. J. Atmos. Oceanic Technol., 30, 942–951, doi:10.1175/JTECH-D-12-00162.1.
Prentice, S. A., and Mackerras D. , 1977: The ratio of cloud to cloud-ground lightning flashes in thunderstorms. J. Appl. Meteor., 16, 545–550, doi:10.1175/1520-0450(1977)016<0545:TROCTC>2.0.CO;2.
Rakov, V. A., 2013: Electromagnetic methods of lightning detection. Surv. Geophys., 34, 731–753, doi:10.1007/s10712-013-9251-1.
Rakov, V. A., and Uman M. A. , 2003: Lightning: Physics and Effects. Cambridge University Press, 687 pp.
Rison, W., 2012: HyMeX Lightning Mapping Array. New Mexico Institute of Mining and Technology, accessed 24 August 2016, doi:10.6096/MISTRALS-HYMEX.LIGHTNING.LMA.
Rison, W., Thomas R. J. , Krehbiel P. R. , Hamlin T. , and Harlin J. , 1999: A GPS-based three-dimensional lightning mapping system: Initial observations in central New Mexico. Geophys. Res. Lett., 26, 3573–3576, doi:10.1029/1999GL010856.
Rodger, C. J., Werner S. , Brundell J. B. , Lay E. H. , Thomson N. R. , Holzworth R. H. , and Dowden R. L. , 2006: Detection efficiency of the VLF World-Wide Lightning Location Network (WWLLN): Initial case study. Ann. Geophys., 24, 3197–3214, doi:10.5194/angeo-24-3197-2006.
Said, R. K., Inan U. S. , and Cummins K. L. , 2010: Long-range lightning geolocation using a VLF radio atmospheric waveform bank. J. Geophys. Res., 115, D23108, doi:10.1029/2010JD013863.
Schulz, W., Diendorfer G. , Pedeboy S. , and Poelman D. R. , 2015: The European lightning location system EUCLID—Part 1: Performance analysis and validation. Nat. Hazards Earth Syst. Sci., 16, 595–605, doi:10.5194/nhess-16-595-2016.
van der Velde, O. A., Montanyà J. , Romero D. , Pineda N. , and Soula S. , 2012: Ebro Lightning Mapping Array: Sprite-producing lightning and ground-to-cloud-to-ground flashes. Geophysical Research Abstracts, Vol. 14, Abstract EGU2012-6541. [Available online at http://meetingorganizer.copernicus.org/EGU2012/EGU2012-6541.pdf.]
