Quarterly Journal of the Royal Meteorological Society Q. J. R. Meteorol. Soc. 143: 2809–2817, October 2017 A DOI:10.1002/qj.

3129

Assessment of the World Wide Lightning Location Network

(WWLLN) detection efficiency by comparison to the Lightning
Imaging Sensor (LIS)
Rodrigo E. Bürgesser*
Group of Atmospheric Physic, FaMAF, Universidad Nacional de Córdoba, IFEG-CONICET, Córdoba, Argentina
*Correspondence to: R. E. Bürgesser, FAMAF, Ciudad Universitaria, 5000 Córdoba, Argentina.
E-mail: burgesse@famaf.unc.edu.ar

In this study, strokes detected by the World Wide Lightning Location Network (WWLLN)
were compared to the flashes detected by the Lightning Imaging Sensor (LIS) between the
years 2012 and 2014. To evaluate the WWLLN detection efficiency, the strokes detected by
WWLLN in the field of view of LIS were determined and matched with the flashes detected
by LIS. The spatial and time distribution of the strokes detected by WWLLN show a good
correlation with the flashes detected by LIS despite the low detection efficiency reported for
WWLLLN. The analysis shows that WWLLN is capable of detecting more than one stroke
per flash with a global multiplicity of 1.5. However, not all strokes detected by WWLLN in
the field of view of LIS could be assigned to a flash. These unmatched strokes show a spatial
and time distribution, as well as an energy distribution, similar to those of the matched
strokes. The unmatched strokes seem to correspond to cloud-to-ground flashes which are
not well detected by LIS. Based on matched flashes and multiplicity, a correction of the
WWLLN data was derived. With this correction, a global lightning flash rate of 60 fl s-1 was
obtained and a map of the corrected flash density detected by WWLLN was performed. The
spatial distribution of WWLLN multiplicity and detection probability are available for the
community.

Key Words: lightning; detection efficiency; WWLLN; LIS

Received 11 November 2016; Revised 11 July 2017; Accepted 18 July 2017; Published online in Wiley Online Library 5
September 2017

1. Introduction lightning data for discrete times. These data are good enough to
generate lightning annual climatology but not to study the time
Lightning data are widely used in many different fields. For evolution of a given storm or a storm system.
instance, lightning data are used for aviation operations, insurance With the launch of the Geostationary Operational Environ-
companies and meteorological agencies. Many studies have mental Satellite R series (GOES-R) (now GOES-16), a new optical
been using lightning data as a proxy to detect thunderstorms sensor, the Geostationary Lightning Mapper, was deployed to
(Mecikalski et al., 2013; Xu et al., 2013), to generate alerts for continuously detect the lightning activity over America and its
severe weather (Schultz et al., 2011; DeMaria et al., 2012), and adjacent ocean region. The Geostationary Lightning Mapper
as an indicator of climate change (Reeve and Toumi, 1999; (GLM) is an optical sensor that detects total lightning (in-cloud,
Romps et al., 2014). The discharge process and the different cloud-to-cloud and cloud-to-ground) activity over the western
atmospheric processes, where lightning has a fundamental role, hemisphere (http://www.goes-r.gov/). The GLM delivers light-
are research topics in which much effort has been invested in ning measurements similar to those of LIS but it provides a
the past years. Thus, the importance of lightning data has led to continuous lightning detection. Therefore, this instrument will
the development of several detection systems with the purpose of provide high-quality data for forecasting severe storms and con-
obtaining high-quality data. vective weather but only over America and its adjacent ocean
The Lightning Imaging Sensor (LIS), on board the Tropical region (Goodman et al., 2013).
Rainfall Measuring Mission satellite (TRMM, https://trmm.gsfc Different ground-based lightning detection networks, regional
.nasa.gov) has been operated from December 1997 to April 2015, and global, have been developed to provide real and continuous
providing lighting data over the tropical region of Earth (Christian detection over time. Short-range lightning detection networks
et al., 1992; Boccippio et al., 2002). However, due to the satellite such as the Lightning Detection Network (LINET: Betz et al.,
movement, the observation time of a given location is about 80s 2009), the National Lightning Detection Network (Cummins
(Christian et al., 2000) and, therefore, LIS provides high-quality and Mur, 2009), and the Canadian Lightning Detection Network


c 2017 Royal Meteorological Society
2810 R. E. Bürgesser

(Dockendorff and Spring, 2005) provide regional coverage of

total lightning with high detection efficiency but these networks
do not provide information over oceanic regions or remote
locations. The development of long-range detection networks
such as the World Wide Lightning Location Network (WWLLN,
http://wwlln.net), Vaisala GLD360 (Vaisala, 2009) and the Earth
Networks Total Lightning Network (ENTLN: Heckman, 2014)
allow a world-wide and real-time lighting detection but with a
lower detection efficiency than short-range detection networks
and satellite detection systems.
The World Wide Lightning Location Network (WWLLN,
http://wwlln.net) was developed as an answer to the need for
a world-wide and real-time detection system. This network can
provide quality lightning data over oceanic and remote regions,
where regional networks are not operational. The network also
presents a continuous detection over time which is not provided
by satellite instruments. The lightning data provided by WWLLN
has been used in many studies (e.g. Garreaud et al., 2014;
Shevtsov et al., 2015; Virts et al., 2015; Soula et al., 2016), but
a better characterization of WWLLN detection efficiency would
be beneficial and would allow a more extensive use of the lightning
data provided by this network. Several attempts have been made to
study the detection efficiency and the location and time accuracy
of WWLLN (Abarca et al., 2010; Abreu et al., 2010; Hutchins et al.,
2012a; Rudlosky and Shea, 2013; Thompson et al., 2014; among
others).
Abarca et al. (2010) used the lightning data provided by the
National Lightning Detection Network (NLDN) to assess the
WWLLN capability over the United States between 2006 and
2009. These researchers found an improvement on the WWLLN
capability to detect cloud-to-ground flashes over the years, with
a bias of the network to detect the most energetic flashes. Abreu
et al. (2010) evaluated the WWLLN data during 2008 using the
lightning data of the Canadian Lightning Detection Network
(CLDN). The study was performed on the lightning data detected
over a region centred on southern Ontario (Canada) and found
that WWLLN is most sensitive to high-peak-current lightning
strokes. Also, this study showed that changes in the ionosphere Figure 1. (a) Ratio between the matched strokes and total strokes detected by
affect the WWLLN detection efficiency with a better ability of WWLLN in the field of view of LIS for a constant distance threshold and with
the network during local midnight and worst ability during local different time thresholds and (b) for a constant time threshold and with different
noon. Rudlosky and Shea (2013) evaluated the lightning data distance thresholds. Black circles represent a distance (time) threshold of 0 km
provided by WWLLN relative to the data provided by LIS during (0 ms), open circles 5 km (50 ms), black triangle 10 km (100 ms), open triangle
20 km (200 ms), black square 30 km (300 ms), open square 40 km (400 ms) and
2009–2012 over the western hemisphere. In this study, flashes black diamond 50 km (500 ms) in panel (a) ((b)). The vertical black line runs
detected by LIS were matched with strokes detected by WWLLN through the selected time (a) and distance (b) thresholds. The symbol × marks
to determine the detection efficiency of WWLLN relative to the selected thresholds.
LIS. They found an improvement in the detection efficiency
of WWLLN through the analysed years over the region under
study. Thompson et al. (2014) compared the lightning group’s is due to the addition of new stations and to the upgrade of the
data detected by LIS with the lightning stroke data detected by detection algorithm. Therefore, periodic studies are necessary to
WWLLN. The study was performed over the western hemisphere assess the current WWLLN detection efficiency. Thereupon, the
during 18 months between January 2010 and June 2011. During aim of this study is to evaluate the lightning data detected by
this period, WWLLN showed 11% of coincidence during the WWLLN between the years 2012 and 2014 using the lightning
entire region under study with higher values over oceanic regions data provided by LIS. This analysis would give a new insight on
than over continental regions. WWLLN and LIS detection capabilities, which would improve
Hutchins et al. (2012a) developed a model to calculate the the quality of the lightning data provided by both systems.
relative detection efficiency of the WWLLN using the detected Finally, based on the analysis performed, a correction method
energy per stroke. This model allows the correction of the global for WWLLN data is developed. This correction method would
stroke density detected by WWLLN as if this network had a allow a broad use of the lightning data provided by WWLLN,
uniform spatial and temporal coverage. Hutchins et al. (2012a) expanding the capability of this network.
applied the model to the lightning data detected by WWLLN from
April 2009 through to October 2011. They found an increase in the 2. Methodology
global average relative detection efficiency over the years, which
is highly correlated with the total number of operational stations The data used in this study came from the Lightning Imaging
of the network. Therefore, the lightning activity of different and Sensor (LIS) on board the Tropical Rainfall Measuring Mission
distant regions can be compared since the model compensates for satellite, and from the World Wide Lightning Location Network.
the uneven global coverage of the network and for the variations in The LIS instrument is a sensor designed to detect the total
propagation of the very low frequency signal. However, the model lightning activity, intra-cloud and cloud-to-ground lightning, of
does not provide an absolute detection efficiency and only allows any storm that passes through the field of view of its sensor.
the performance of comparative studies on lightning activity. The LIS instrument detects the optical pulses associated with
These studies of the WWLLN capability have shown a changes in cloud brightness at each pixel. Pixels on LIS sensors
continuous improvement of this network. This improvement exceeding the background threshold are defined as an event.


c 2017 Royal Meteorological Society Q. J. R. Meteorol. Soc. 143: 2809–2817 (2017)
 WWLLN Detection Efficiency Relative to LIS 2811

(a)

(b)

(c)

Figure 2. (a) Spatial distribution of the flashes detected by LIS, (b) the strokes detected by WWLLN in the LIS field of view and (c) WWLLN detection efficiency
relative to the LIS, using a 2◦ × 2◦ spatial resolution. Note the different scales in the upper panels. [Colour figure can be viewed at wileyonlinelibrary.com].

Spatially adjacent events during the same observation window are In this study, flashes detected by LIS and strokes detected by
clustered into groups. If a set of groups are sequentially separated WWLLN between the years 2012 and 2014 were analysed. The
in time by no more than 330 ms and in space by no more than flashes detected by LIS are multiple optical pulses that occur in
5.5 km, this set is grouped into a flash. Finally, the flashes that the same storm cell within 330 ms and 5.5 km of each other (Mach
are separated in space by no more than 16.5 km are grouped into et al., 2007). Therefore, reported for each flash are the initial time
areas (Christian et al., 2000; Mach et al., 2007). These different of the flash, the flash duration, the location (latitude/longitude)
products (Events, Groups, Flashes and Areas) are provided by LIS of the flash centroid, its size (area) and the number of groups
and can be related to different lightning-storm characteristics. assigned to the flash. The location, time of occurrence and the
The LIS instrument detects lightning between 38◦ N and 38◦ S radiated VLF energy are reported by WWLLN for each of the
latitude and it has a detection efficiency which depends on the local strokes detected.
time. Boccippio et al. (2002) predict a flash detection efficiency Different studies have compared the strokes detected by
of 93 ± 4% and 73 ± 11% at night and at noon, respectively; WWLLN to flashes and groups detected by LIS (Rudlosky and
while Cecil et al. (2014) present the hourly detection efficiency Shea, 2013; Thompson et al., 2014) to evaluate the WWLLN
for LIS which is within the range of 88% overnight to 69% near detection efficiency. These studies have used different matching
local noon. criteria to assign flashes or groups detected by LIS to the strokes
The WWLLN is a ground-based network with world-wide and detected by WWLLN in order to find the WWLLN detection
real-time detection. The WWLLN detects the very low frequency efficiency relative to LIS. However, these studies did not filter
(VLF) wave packet emitted by lightning. The network uses WWLLN strokes for LIS view times as only WWLLN detection
the Time of Group Arrival (TOGA), along with minimization efficiency relative to LIS was evaluated. In this study only WWLLN
methods, to locate the lightning (Dowden et al., 2002; Rodger strokes during the LIS view times are used so that LIS detection
et al., 2009). The WWLLN preferentially detects cloud-to-ground efficiency relative to WWLLN can also be measured. During
lightning with the greatest peak current and detects, generally, the 3 years under analysis, between 38◦ N and 38◦ S latitude, LIS
a single stroke within a lightning flash (Abarca et al., 2010; detected more than 4 × 106 flashes while WWLLN detected more
Abreu et al., 2010). However, Rudlosky and Shea (2013) found than 600 × 106 strokes. On the areas under LIS observation,
an average multiplicity (number of WWLLN strokes per LIS WWLLN detected over 6 × 105 strokes.
flash) of 1.5 during the period between 2009 and 2012 over the A match between the strokes detected by WWLLN and the
western hemisphere. Recently, an upgrade of the WWLLN allows flashes detected by LIS was performed, using the strokes detected
the network to measure the radiated VLF stroke energy which is by WWLLN in the LIS field of view. A stroke detected by WWLLN
related to the return-stroke peak current (Hutchins et al., 2012b) was assigned to an LIS flash every time the WWLLN stroke met
of the stroke detected by WWLLN. the following criteria:


c 2017 Royal Meteorological Society Q. J. R. Meteorol. Soc. 143: 2809–2817 (2017)
2812 R. E. Bürgesser

• the WWLLN stroke had occurred 200 ms before, during or

200 ms after a LIS flash, and
• the location of the WWLLN stroke is within a distance less
or equal to (Area/π )0.5 + 20 km of the LIS flash centroid.

The time and distance thresholds (200 ms and 20 km) were

selected based on the results of different tests using different
threshold values. Figure 1 shows the ratio between the matched
strokes and the total strokes detected by WWLLN in the LIS
field of view using different time and distance thresholds. As
can be observed on Figure 1, the values of 200 ms and 20 km
for time and distance correspond to values where an increment
of these values results in a no longer significant increment on
the number of the matched strokes. On the other hand, more
restrictive criteria strongly diminished the number of the matched
strokes. Therefore, these broad criteria were selected to ensure
the identification of all possible matches.

3. Results and discussion

Figure 2 shows the spatial distribution, using a spatial resolution

of 2◦ × 2◦ , of the flashes detected by LIS (upper panel), the
distribution of the strokes detected by WWLLN in the areas
under LIS observation (middle panel) and the relative detection
efficiency (lower panel) between 2012 and 2014. The relative
detection efficiency was computed as the ratio between the strokes
detected by WWLLN, in the areas under LIS observation, and the
flashes detected by LIS.
Both detection systems present a similar global pattern of flash
and stroke distributions with lightning detected preferably over
land. A Pearson correlation coefficient of 0.62 (p < 10-4 ) was
computed between both lightning datasets. The high lightning
activity centres over Africa, South America and the Maritime
Figure 3. (a) Local time distribution of the flashes detected by LIS (solid line)
Continent, which have been long known (Whipple, 1929), could and the strokes detected by WWLLN in the field of view of LIS (dashed line), and
be observed on both maps. However, WWLLN shows a better (b) local time distribution of WWLLN detection efficiency relative to the LIS.
ability to detect strokes over Central and North America with
respect to other regions of the globe. This relatively better
performance of WWLLN is in agreement with the findings is evidence that lightning over the oceans has higher energy
of Rudlosky and Shea (2013) and Thompson et al. (2014). than lightning over land (Hutchins et al., 2013; Said et al., 2013)
Meanwhile, WWLLN presents a lower detection capability relative and WWLLN detects preferentially the more energetic lightning
to LIS over South America, Africa and over India (Tibetan (Abarca et al., 2010; Abreu et al., 2010).
plateau). The local time distribution of the flashes detected by LIS
As reported by Rudlosky and Shea (2013) and Thompson et al. and the strokes detected by WWLLN, in the field of view of
(2014), over oceanic regions, there are areas with more strokes LIS, are shown on Figure 3 (upper panel). The datasets are
detected by WWLLN than flashes detected by LIS. This better strongly correlated with a Pearson correlation coefficient of 0.96
ability of WWLLN respect to LIS, over oceanic regions, seems to (p < 10-4 ), which indicates that WWLLN is able to detect the
be the result of a better detection capability of WWLLN over the diurnal cycle on the lightning activity. However, the amplitude of
oceans. Further, the different optical scattering properties of the the diurnal lightning cycle detected by WWLLN was much lower
storms over land and ocean (Peterson and Liu, 2011) together than the amplitude detected by LIS. While LIS reports seven times
with the different properties of lightning flashes that occur over more flashes during local afternoon than during early morning,
continental and oceanic regions also seem to improve the ability WWLLN only reports two times more strokes. This difference
of WWLLN respect to LIS. seems to be due to the low detection efficiency of the network
Boccippio et al. (2000) reported that the flashes detected by the over Africa and over South America, which have been reported
Optical Transient Detector (OTD) and by LIS are brighter and as the main contributors to the peak on the lightning diurnal
larger over the ocean than the flashes detected over continental cycle (Blakeslee et al., 2014). These differences on the detected
regions. However, they were unable to assess if these differences amplitude of the diurnal lightning cycle are reflected in the local
are due to difference in the flashes features or due to differences time distribution of the relative detection efficiency (Figure 3,
in the optical scattering properties of the storms. Different studies lower panel). The relative detection efficiency increases in early
(Peterson and Liu, 2013; Bitzer et al., 2016; Peterson et al., 2017) local morning and reach a maximum values at around 0900 (local
had shown that the flashes detected by LIS over oceanic regions time). After that, the relative detection efficiency decreases to a
are larger, with a longer duration and brighter than the flashes minimum value during local afternoon, which correspond to the
detected over continental regions. Also, Peterson et al. (2017) diurnal cycle maximum.
reported that flashes over the ocean contain more groups and are The good agreement between the spatial and time distributions
more prone to propagate horizontally than flashes over land. of both lightning datasets shows that, despite the lower detection
On the other hand, WWLLN is capable of detecting strokes with efficiency of the WWLLN, this network is able to detect the main
lower energy over the ocean with respect to land given that the features of the global lightning activity.
VLF signal is less attenuated over the oceans than over land during With the matching criteria adopted, 434 553 WWLLN strokes
signal propagation. Therefore, WWLLN has a bias to detect the were assigned to 288 047 LIS flashes. The amount of strokes
highest energy strokes over land and more lower-energy strokes matched is around 66% of the total strokes detected by WWLLN
over oceanic regions (Hutchins et al., 2012b). Furthermore, there in the LIS field of view. The number of strokes matched to


c 2017 Royal Meteorological Society Q. J. R. Meteorol. Soc. 143: 2809–2817 (2017)
 WWLLN Detection Efficiency Relative to LIS 2813

Figure 4. Spatial distribution of the computed multiplicity using a 2◦ × 2◦ spatial resolution. [Colour figure can be viewed at wileyonlinelibrary.com].

the flashes indicates that WWLLN is capable of detecting more

than one stroke per flash. Defining the multiplicity as the ratio
of the number of matched WWLLN strokes to the number of
matched LIS flashes, WWLLN presents a global mean multiplicity
of 1.5. Although, WWLLN is capable of detecting multiple strokes
during an LIS flash, 70% of the matched flashes were coincident
with a single stroke.
Figure 4 shows the spatial distribution of the multiplicity
using a spatial resolution of 2◦ × 2◦ . The higher values of the
multiplicity are found over oceanic regions and over Central
America. These regions also show the higher relative detection
efficiency of WWLLN relative to LIS (Figure 2, lower panel)
as reported by Rudlosky and Shea (2013) and Thompson et al.
(2014). This better detection ability of WWLLN over the oceans
seems to be due to the bias of this network to detect strokes with
lower energy over oceanic regions (Hutchins et al., 2012b) and
the strength of strokes over these regions. Further, as reported
by Peterson et al. (2017), the flashes detected by LIS over ocean
contain more groups than the flashes over land, which implies that
flashes over oceanic regions probably have higher multiplicity.
Figure 5 shows the difference between the time of occurrence of
the matched strokes and the initiation time of the matched flashes
(upper panel) and the distance between the matched strokes and
the flash centroid (lower panel). Given that the flash duration
and area of the flashes are used in the matching criteria, the time
differences and distances of the matched strokes can be larger
than the threshold value used, as can be observed in Figure 5. The
mean (median) time difference and distance are 114 ms (24 ms)
and 10 km (9 km), respectively. As can be observed, almost 90% of
the matched strokes show a positive time difference which implies
that these matched strokes were detected after the initiation of
the matched flash. Although the parameters calculated cannot Figure 5. (a) Difference between the time of occurrence of the matched strokes
be taken as the WWLLN accuracy since different aspects of a and time initiation of flashes. Data are grouped into a 10 ms bin size. (b) Distance
lightning stroke are compared, these values can be used as an between the matched strokes and the flash centroid. Data are grouped into a
0.5 km bin size.
indicator of the WWLLN detection ability. The values obtained
agree with the accuracy of the WWLLN reported by Rudlosky
and Shea (2013).
The remaining WWLLN strokes (221 732 strokes, ∼34%) distribution of the ratio between the unmatched and matched
detected in the LIS field of view were not assigned to any strokes (Figure 6, lower panel) does not show a clear geographical
LIS flash (unmatched strokes). As mentioned before, different dependence on the location of the unmatched strokes.
time and distance thresholds were examined. Using values larger Figure 7 (upper panel) shows the local time distribution of the
than 200 ms and 20 km (up to 500 ms and 50 km) only improved matched (dashed line) and unmatched strokes (solid line). The
the number of matched strokes up to 5% of the total strokes local time distribution of the LIS detection efficiency, adapted by
(Figure 1). With a more restrictive matching criteria (0 ms and the values reported by Cecil et al. (2014, Tab. 2), is also shown on
0 km), the number of matched strokes decreased to around 40% Figure 7 (upper panel, dotted line). The time distribution between
of the total strokes detected in the LIS field of view. Given that the matched and unmatched strokes shows a high correlation,
previous studies do not show evidence that WWLLN detects false with a Pearson correlation coefficient of 0.92 (p < 10-4 ). An
positives (Lay et al., 2004; Rodger et al., 2005, 2006; Jacobson et al., analysis of the energy reported by WWLLN, using the regression
2006; Abarca et al., 2010; Abreu et al., 2010), these unmatched method proposed by Hutchins et al. (2013), does not show a
strokes indicate that WWLLN is capable of detecting strokes that significant difference between the energy distribution of the
are not detected by LIS. matched and unmatched strokes.
Figure 6 shows the spatial distribution, with a spatial resolution On Figure 7 (lower panel) is shown the hourly variation of
of 2◦ × 2◦ , of the matched strokes (upper panel), unmatched the ratio between the matched strokes and the total strokes
strokes (middle panel) and the ratio of unmatched to matched detected by WWLLN in the areas under LIS observation (solid
strokes (lower panel). A similar pattern of the spatial distribution line) and the LIS detection efficiency (Cecil et al., 2014) (dotted
of the matched and unmatched strokes can be observed with line). As can be observed, the ratio between the matched strokes
a Pearson correlation coefficient of 0.95 (p < 10-4 ). The spatial and the total strokes shows a good agreement with the hourly


c 2017 Royal Meteorological Society Q. J. R. Meteorol. Soc. 143: 2809–2817 (2017)
2814 R. E. Bürgesser

(a)

(b)

(c)

Figure 6. (a) Spatial distribution of the matched strokes, (b) the unmatched strokes and (c) the ratio of unmatched to matched strokes, using a 2◦ × 2◦ spatial
resolution. [Colour figure can be viewed at wileyonlinelibrary.com].

distribution of LIS detection efficiency with both distributions 4. Correction to WWLLN lightning data
showing minimum values at local noon. However, the ratio
between the matched strokes and the total strokes shows several The analysis performed and previous studies had shown that
oscillations which seem to be due to the temporal resolution WWLLN had a low detection efficiency, which also shows
used in the calculations. Despite these oscillations, the hourly spatial and temporal variations. This low detection ability, and
variations of LIS detection efficiency seem to explain less than its variations, do not allow for the use of the lightning data
10% of the unmatched strokes. Therefore, the detection efficiency provided by WWLLN on more accurate studies of global lightning
of LIS cannot fully explain the unmatched strokes found in this activity and to perform comparative studies of the lightning
study. Thomas et al. (2000) compared the observation from the climatology between different regions. Several studies had tried
Lightning Mapping Array (LMA) with LIS data. These researchers to find the detection efficiency of the network using the lightning
found an excellent correlation between both datasets. However, data provided by different ground-based network and satellite
they reported that LIS presents different detection efficiencies for systems (Lay et al., 2004; Jacobson et al., 2006; Rodger et al., 2009;
intra-cloud (IC) and cloud-to-ground (CG) discharges. While Abarca et al., 2010; Abreu et al., 2010; Rudlosky and Shea, 2013;
almost all IC discharges detected by LMA were detected by Thompson et al., 2014). However, these studies are spatially and
LIS, only 60% of CG discharges were detected by LIS. The CG temporally limited due to the lack of available lightning data.
discharges not detected by LIS correspond to CG discharges Hutchins et al. (2012a) presented a model based on the stroke
confined below an altitude of 7 km. Given that LIS is an optical energy detected by WWLLN that allows computation of the global
sensor that detects luminous events, the amount of light, from a stroke distribution as if the WWLLN showed a uniform global
discharge, that can reach the sensor will depend on the optical detection efficiency. Nevertheless, this model does not provide an
depth between the discharge and the cloud surface. Therefore, overall absolute detection efficiency for the network. Therefore,
optical sensors present different detection efficiencies for IC and the development of an algorithm which allows computation of
CG, as was reported by Thomas et al. (2000), with a bias to detect the detection efficiency of WWLLN will expand the capability
IC discharges. On the other hand, WWLLN is more sensitive to of the network and facilitate a broad use of the lightning data
CG discharges than to IC discharges (Abarca et al., 2010). This provided by WWLLN. Thus, a correction method of the WWLLN
bias of the network is due to the fact that WWLLN detects the VLF lightning data is proposed based on the data provided by LIS.
radio wave emitted by lightning and that CG discharges radiate Let F be the number of flashes that actually occurred in the
stronger in VLF than IC discharges. Therefore, the unmatched LIS field of view. Then, the probability that LIS detects a flash
strokes seem to be due to the different detection bias reported for can be defined as the ratio between the flashes detected by this
both detection systems. However, further analysis is needed to instrument (F LIS ) and F,
identify different case-studies which could provide an insight on
these unmatched strokes. PLIS = FLIS F −1 . (1)


c 2017 Royal Meteorological Society Q. J. R. Meteorol. Soc. 143: 2809–2817 (2017)
 WWLLN Detection Efficiency Relative to LIS 2815

To correct the lightning data provided by WWLLN, given

that the values of PW were derived using flashes, it is necessary
to cluster the strokes detected by this network into flashes.
Different methods have been proposed to cluster the strokes
into flashes using the times of occurrence and locations criteria.
An alternative approach to determine the number of WWLLN
flashes is using the multiplicity computed in the previous analysis
(Figure 4). Therefore, the number of strokes detected by WWLLN
on each grid box was divided by the value of multiplicity of the
corresponding grid box to group the strokes into flashes. This
approach differs from traditional stroke to flash algorithms but is
expected to provide a good estimation of the number of flashes
detected by WWLLN.
These estimated flashes were corrected using the values of
PW found. Figure 9 shows the corrected flash density based on
the strokes detected by WWLLN between 2012 and 2014. The
correction applied enhances the lightning activity over the areas
where the WWLLN showed lower detection capability (South
America, Africa and over India). The corrected flash density and
the flash density computed with the lightning data provided by
LIS show a high correlation with a Pearson correlation coefficient
of 0.96 (p < 10-4 ).
WWLLN detected around 6 strokes s-1 , which represent a
lightning flash rate of 4 flashes s-1 using a mean global multiplicity
of 1.5. The mean value of PW found in this study is 0.07 and
therefore, the lightning flash rate detected by WWLLN is 60 fl s-1 .
This value is in agreement with the values determined by satellite
measurements (Christian et al., 2003; Cecil et al., 2014).
The corrected lightning activity shows good agreement with the
lightning activity reported by other detection systems, which give
confidence in the proposed method. However, the multiplicity
and the values of PW used in the correction depend on the
Figure 7. (a) Local time distribution of the matched (dashed line) and unmatched matching criteria used to compute the matched flashes between
(solid line) strokes, and the local time distribution of the detection efficiency of both datasets. The use of a more restrictive criterion strongly
LIS (dotted line) as reported by Cecil et al. (2014). (b) Local time distribution of diminished the number of the matched strokes and therefore
the ratio between the matched strokes and the total strokes (solid line) and the
local time distribution of the detection efficiency of LIS (dotted line) as reported decreased the detection probability values. Further analysis
by Cecil et al. (2014). is needed to study the robustness of the correction method
proposed.

In the same way, the probability that WWLLN and LIS detect 5. Summary
the same flash can be defined as the ratio between the matched
flashes (F M ) and F, The flashes detected by LIS were compared to the strokes detected
by WWLLN in the LIS field of view between the years 2012
PW,LIS = FM F −1 . (2) and 2014. During these 3 years, LIS detected more than 40 × 105
flashes on the tropical region while WWLLN detected around
Given that WWLLN and LIS are independent detection 6 × 105 strokes in the field of view of LIS.
systems, the probability PW,LIS can also be defined as, The events detected by both systems show a similar spatial
distribution with a high time correlation showing that WWLLN
PW,LIS = PLIS PW , (3) performance is good enough to detect the main features of
tropical lightning despite the low detection efficiency reported for
where PW is the probability that WWLLN will detect a flash. From this network.
these equations, it is possible to derive an expression for PW as Using a matching criterion, almost 70% of the strokes detected
follows, by WWLLN in the LIS field of view were matched with flashes
detected by LIS with a mean (median) time difference and distance
PW = FM FLIS −1 . (4) of 114 ms (24 ms) and 10 km (9 km), respectively. Although most
of the flashes were matched with single strokes, several matched
The expression obtained for PW is independent of the real LIS flashes had multiple coincident WWLLN strokes, which
number of flashes (F). Therefore, in order to calculate the means that WWLLN is capable of detecting more than one stroke
detection probability of WWLLN, it is not necessary to correct per flash with a mean multiplicity of 1.5.
LIS data by its detection efficiency. However, these equations are Around 30% of strokes detected by WWLLN in the LIS field
valid if both detection systems, LIS and WWLLN, do not present of view were not assigned to any flash detected by LIS. These
false positives. unmatched strokes show a spatial, time and energy distribution
Figure 8 shows the PW distribution computed using Eq. (4) for similar to that of the matched strokes. These unmatched strokes
a 2◦ × 2◦ spatial resolution. PW values range between 0.01 and 0.1 seem to correspond to cloud-to-ground lightning given that LIS
(1% and 10%) for continental regions while over oceanic regions shows a lower detection efficiency for this type of discharge while
PW shows values higher than 0.2 (20%). The map shows a good WWLLN preferentially detects these kind of events. Although
agreement with the analysis performed by Rudlosky and Shea satellite detection systems such as LIS and GLM present high
(2013) and Thompson et al. (2014) for the western hemisphere, detection efficiencies, the results of this study show that optical
with higher probability over Central and North America and over detection systems can miss a significant amount of lightning
oceanic regions. events, particularly cloud-to-ground flashes in the lower part of


c 2017 Royal Meteorological Society Q. J. R. Meteorol. Soc. 143: 2809–2817 (2017)
2816 R. E. Bürgesser

Figure 8. Spatial distribution of the WWLLN detection probability (PW ) using a 2◦ × 2◦ spatial resolution. [Colour figure can be viewed at wileyonlinelibrary.com].

Figure 9. Flash density detected by WWLLN between 2012 and 2014 corrected by multiplicity and PW using a 2◦ × 2◦ spatial resolution. [Colour figure can be viewed
at wileyonlinelibrary.com].

the storms. Therefore, in order to obtain high-quality lightning Christian HJ, Blakeslee RJ, Goodman SJ. 1992. ‘Lightning imaging sensor (LIS)
data, the different detection systems, satellite and ground-based, for the Earth observing system’, NASA-TM-4350. NASA Marshall Space
Flight Center: Huntsville, AL.
should be used together.
Christian HJ, Blakeslee RJ, Goodman SJ, Mach DM. 2000. ‘The Algorithm
A correction to the lightning activity detected by WWLLN is Theoretical Basis Document (ATBD) for the Lightning Imaging Sensor
proposed which allows for a more extensive use of the lightning (LIS)’. http://eospso.gsfc.nasa.gov/eos_homepage/for_scientists/atbd/docs/
data provided by the network. Using the correction, WWLLN LIS/atbd-lis-01.pdf (accessed 20 February 2017).
presents a lightning flash rate of 60 fl s-1 , which is in agreement Christian HJ, Blakeslee RJ, Boccippio DJ, Boeck WL, Buechler DE, Driscoll
KT, Goodman SJ, Hall JM, Koshak WJ, Mach DM, Stewart MF. 2003.
with reported values. Global frequency and distribution of lightning as observed from space by
the Optical Transient Detector. J. Geophys. Res. 108: 4005. https://doi.org/
Acknowledgements 10.1029/2002JD002347.
Cummins KL, Mur MJ. 2009. An overview of lightning locating systems:
The author wishes to thank the World Wide Lightning History, techniques, and data uses, with an in-depth look at the US NLDN.
IEEE Trans. Electromagn. Compat. 51: 499–518.
Location Network (http://wwlln.net), collaboration among over
DeMaria M, DeMaria RT, Knaff JA, Molenar D. 2012. Tropical cyclone
50 universities and institutions, for providing the lightning lightning and rapid intensity change. Mon. Weather Rev. 140: 1828–1842.
location data used in this article. He is grateful to the NASA Dockendorff D, Spring K. 2005. ‘The Canadian Lightning Detection
Global Hydrology Resource Center DAAC for LIS data. This work Network – novel approaches for performance measurement and network
was supported by SECYT-UNC, CONICET and FONCYT. The management’. In Proceedings, WMO Technical Conference on Instruments
and Methods of Observation (TECO), Bucharest, Romania, 34–37.
author wants to thank the anonymous reviewers for their useful
Dowden RL, Brundell JB, Rodger CJ. 2002. VLF lightning location by time
comments and suggestions that greatly improved the quality of of group arrival (TOGA) at multiple sites. J. Atmos. Solar–Terr. Phys. 64:
this article. 817–830.
Garreaud RD, Nicora MG, Bürgesser RE, Ávila EE. 2014. Lightning in western
References Patagonia. J. Geophys. Res. Atmos. 119: 4471–4485. https://doi.org/10.1002/
2013JD021160.
Abarca SF, Corbosiero KL, Galarneau TJ Jr. 2010. An evaluation of the Goodman SJ, Blakeslee RJ, Koshak WJ, Mach D, Bailey J, Buechler D, Carey
WorldWide Lightning Location Network (WWLLN) using the National L, Schultz C, Bateman M, McCaul E Jr, Stano G. 2013. The GOES-R
Lightning Detection Network (NLDN) as ground truth. J. Geophys. Res. geostationary lightning mapper (GLM). Atmos. Res. 125: 34–49.
115: D18206. https://doi.org/10.1029/2009JD013411. Heckman S. 2014. ‘ENTLN status update’. In XV International Conference on
Abreu D, Chandan D, Holzworth RH, Strong K. 2010. A performance Atmospheric Electricity, 15–20 June 2014. Norman, OK.
assessment of the World Wide Lightning Location Network (WWLLN) Hutchins ML, Holzworth RH, Brundell JB, Rodger CJ. 2012a. Relative detection
via comparison with the Canadian Lightning Detection Network (CLDN). efficiency of the World Wide Lightning Location Network. Radio Sci. 47:
Atmos. Meas. Tech. 3: 1143–1153. RS6005. https://doi.org/10.1029/2012RS005049.
Betz HD, Schmidt K, Laroche P, Blanchet P, Oettinger WP, Defer E, Dziewit Hutchins ML, Holzworth RH, Rodger CJ, Brundell JB. 2012b. Far-field power
Z, Konarski J. 2009. LINET – an international lightning detection network of lightning strokes as measured by the World Wide Lightning Location
in Europe. Atmos. Res. 91: 564–573. Network. J. Atmos. Oceanic Technol. 29: 1102–1110.
Bitzer PM, Burchfield JC, Christian HJ. 2016. A Bayesian approach to assess Hutchins ML, Holzworth RH, Virts KS, Wallace JM, Heckman S. 2013.
the performance of lightning detection systems. J. Atmos. Oceanic Technol. Radiated VLF energy differences of land and oceanic lightning. Geophys.
33: 563–578. Res. Lett. 40: 2390–2394.
Blakeslee RJ, Mach DM, Bateman MG, Bailey JC. 2014. Seasonal variations in Jacobson AR, Holzworth R, Harlin J, Dowden R, Lay E. 2006. Performance
the lightning diurnal cycle and implications for the global electric circuit. assessment of the World Wide Lightning Location Network (WWLLN),
Atmos. Res. 135: 228–243. using the Los Alamos Sferic Array (LASA) as ground truth. J. Atmos.
Boccippio DJ, Goodman SJ, Heckman S. 2000. Regional differences in tropical Oceanic Technol. 23: 1082–1092.
lightning distributions. J. Appl. Meteorol. 39: 2231–2248. Lay EH, Holzworth RH, Rodger CJ, Thomas JN, Pinto O Jr, Dowden
Boccippio DJ, Koshak WJ, Blakeslee RJ. 2002. Performance assessment of the R. 2004. WWLL global lightning detection system: Regional validation
Optical Transient Detector (OTD) and Lightning Imaging Sensor. Part I: study in Brazil. Geophys. Res. Lett. 31: L03102. https://doi.org/10.1029/
Predicted diurnal variability. J. Atmos. Oceanic Technol. 19: 1318–1332. 2003GL018882.
Cecil DJ, Buechler DE, Blakeslee RJ. 2014. Gridded lightning climatology from Mach DM, Christian HJ, Blakeslee RJ, Boccippio DJ, Goodman SJ, Boeck
TRMM-LIS and OTD: Dataset description. Atmos. Res. 135: 404–414. WL. 2007. Performance assessment of the Optical Transient Detector and


c 2017 Royal Meteorological Society Q. J. R. Meteorol. Soc. 143: 2809–2817 (2017)
 WWLLN Detection Efficiency Relative to LIS 2817

Lightning Imaging Sensor. J. Geophys. Res. 112: D09210. https://doi.org/10 Rudlosky SD, Shea DT. 2013. Evaluating WWLLN performance relative to
.1029/2006JD007787. TRMM/LIS. Geophys. Res. Lett. 40: 2344–2348.
Mecikalski JR, Li X, Carey LD, McCaul EW, Coleman TA. 2013. Regional Said RK, Cohen MB, Inan US. 2013. Highly intense lightning over the oceans:
comparison of GOES cloud-top properties and radar characteristics Estimated peak currents from global GLD360 observations. J. Geophys. Res.
in advance of first-flash lightning initiation. Mon. Weather Rev. 141: Atmos. 118: 6905–6915.
55–74. Schultz CJ, Petersen WA, Carey LD. 2011. Lightning and severe weather: A
Peterson M, Liu C. 2011. Global statistics of lightning in anvil and comparison between total and cloud-to-ground lightning trends. Weather
stratiform regions over the tropics and subtropics observed by the and Forecasting 26: 744–755.
Tropical Rainfall Measuring Mission. J. Geophys. Res. 116: D23201. Shevtsov BM, Firstov PP, Cherneva NV, Holzworth RH, Akbashev RR. 2015.
https://doi.org/10.1029/2011JD015908. Lightning and electrical activity during the Shiveluch volcano eruption on
Peterson M, Liu C. 2013. Characteristics of lightning flashes with exceptional 16 November 2014. Nat. Hazards Earth Syst. Sci. Discuss. 3: 6745–6755.
illuminated areas, durations, and optical powers and surrounding storm Soula S, Kigotsi Kasereka J, Georgis JF, Barthe C. 2016. Lightning climatology
properties in the tropics and inner subtropics. J. Geophys. Res. Atmos. 118: in the Congo Basin. Atmos. Res. 178–179: 304–331.
11727–11740. Thomas RJ, Krehbiel PR, Rison W, Hamlin T, Boccippio DJ, Goodman SJ,
Peterson M, Deierling W, Liu C, Mach D, Kalb C. 2017. The properties of Christian HJ. 2000. Comparison of ground-based 3-dimensional lightning
optical lightning flashes and the clouds they illuminate. J. Geophys. Res. mapping observations with satellite-based LIS observations in Oklahoma.
Atmos. 122: 423–442. https://doi.org/10.1002/2016JD025312. Geophys. Res. Lett. 27: 1703–1706.
Reeve N, Toumi R. 1999. Lightning activity as an indicator of climate change. Thompson KB, Bateman MG, Carey LD. 2014. A comparison of two ground-
Q. J. R. Meteorol. Soc. 125: 893–903. based lightning detection networks against the satellite-based Lightning
Rodger CJ, Brundell JB, Dowden RL. 2005. Location accuracy of VLF World- Imaging Sensor (LIS). J. Atmos. Oceanic Technol. 31: 2191–2205.
Wide Lightning Location (WWLL) network: Post-algorithm upgrade. Ann. Vaisala. 2009. ‘Vaisala Global Lightning Dataset – technology, opera-
Geophys. 23: 277–290. https://doi.org/10.5194/angeo-23-277-2005. tions and application overview’. 9. http://www.vaisala.com/en/products/
Rodger CJ, Werner S, Brundell JB, Lay EH, Thomson NR, Holzworth thunderstormandlightningdetectionsystems/Pages/GLD360.aspx (accessed
RH, Dowden RL. 2006. Detection efficiency of the VLF World-Wide 20 February 2017).
Lightning Location Network (WWLLN): Initial case study. Ann. Geophys. Virts KS, Wallace JM, Hutchins ML, Holzworth RH. 2015. Diurnal and seasonal
24: 3197–3214. https://doi.org/10.5194/angeo-24-3197-2006. lightning variability over the Gulf Stream and the Gulf of Mexico. J. Atmos.
Rodger CJ, Brundell JB, Holzworth RH, Lay EH, Crosby NB, Huang TY, Rycroft Sci. 72: 2657–2665. https://doi.org/10.1175/JAS-D-14-0233.1.
MJ. 2009. ‘Growing detection efficiency of the World Wide Lightning Whipple FJW. 1929. On the association of the diurnal variation of electric
Location Network’. In AIP Conference Proceedings 1118: 15–20. Corte, potential gradient in fine weather with the distribution of thunderstorms
France. over the globe. Q. J. R. Meteorol. Soc. 55: 1–18.
Romps DM, Seeley JT, Vollaro D, Molinari J. 2014. Projected increase in Xu W, Adler RF, Wang NY. 2013. Improving geostationary satellite rainfall esti-
lightning strikes in the United States due to global warming. Science 346: mates using lightning observations: Underlying lightning–rainfall–cloud
851–854. relationships. J. Appl. Meteorol. Climatol. 52: 213–229.


c 2017 Royal Meteorological Society Q. J. R. Meteorol. Soc. 143: 2809–2817 (2017)