We compare WWLL lightning events with residuals less than 20 microseconds that occurred on 6, 7, 14, 20, and 21 March 2003 in the range of 40° to 55°W, 15° to 25°S to events in the same range measured by a land-based local Brazil lightning detection network, the Brazilian Integrated Network (BIN) [Pinto and Pinto Jr., 2003; Pinto Jr. et al., 2003b]. Figure 1 shows the region of interest in Brazil. We study these data because the BIN data had already been procured by our group for use in the sprite balloon campaign in 2002–2003 in Brazil [Holzworth et al., submitted, 2003].
Figure 1. Boxed area of Brazil shows the region of comparison between lightning location networks used in this study. As there were no WWLL receivers in South America, this region of Brazil is a low-coverage region of the WWLL network.
Download figure to PowerPoint
 BIN consists of 21 sensors in the region of interest, with an overall stated detection efficiency of 80% of all cloud-to-ground lightning strokes. However, return stroke peak current affects efficiency in certain ranges. Detection efficiency of events with peak current greater than 50 kA is 90% with a location accuracy of less than 1 km. For events with peak current less than 10 kA, detection efficiency could be as low as 30% with approximately 5 km location accuracy. The return stroke peak current measurement also includes uncertainty due to assumed lightning return stroke speeds, as expected for detectors of this type [e.g., MacGorman and Rust, 1998]. BIN cites an uncertainty of 20–30% for strokes with peak current greater than 10 kA and up to 100% uncertainty for strokes with less than 10 kA peak current [Pinto Jr., personal communication, 2003].
 In the five day time period, 671 WWLL events and 63,893 BIN events were reported in the region of interest. Taking into account the 80% accuracy of BIN and the limitation that BIN measures only CG lightning strokes [Pinto Jr. et al., 2003a], a rough estimate of the percentage of all lightning events measured in this region is about 0.3%. The percentage is slightly lower in this “worst case” region than the 1.1% average global detection efficiency calculated above. To measure the accuracy of WWLL, we compare the data sets from the two networks to find “shared” events. A lightning stroke is assumed to be shared if each network measures an event within the same 3 ms and 50 km. According to these criteria, 289 of the 671 WWLL events are common to the BIN stroke data.
 The shared events have an average return stroke peak current of 85.7 kA, as measured by BIN. In contrast, the average peak current of the entire BIN dataset is 33.3 kA, suggesting that the WWLL network only detects large discharges that exceed an approximate “threshold” in return stroke peak current. The histogram in Figure 2 represents this threshold by comparing the BIN peak current distribution of the entire BIN data set to only the BIN events which were also observed by WWLL. Overall, a greater fraction of the strokes have negative polarity, as expected. However, for discharges with an absolute value of peak current less than 50 kA the figure shows that there is a larger proportion of such low-current lightning in the overall BIN dataset (black) than in the subset made up of only shared events (white). For events with peak current greater than 50 kA, the relative pattern is reversed. This pattern illustrates that WWLL detection is biased towards lightning strokes with large peak currents.
Figure 2. Histogram of return stroke peak currents measured by the BIN, shown in 50 kA bins centered on the single peak current value noted beneath them. The two outermost bins contain data for all strokes with peak current greater or less than 175 kA. Distributions are shown for shared WWLL-BIN events (white) and all measured BIN events (black).
Download figure to PowerPoint
 Next, we estimate the spatial and temporal accuracy of WWLL by analyzing the shared events. Time differences between shared strokes are on average 0.06 ± 0.2 ms. Note that the time resolution of the WWLL is 1 microsecond, while the BIN time resolution is 1 nanosecond. To calculate location offsets for WWLL strokes relative to their shared BIN events, we plot each shared BIN event at (0, 0) and determine the east-west and north-south deviation of the WWLL positions (Figure 3). WWLL events have a mean deviation of 3.2 km north (dashed-dotted line), 7.3 km east (dotted line) from BIN events. The plotted ellipse of one standard deviation encompasses (0, 0), indicating no statistically significant difference in the location of the shared events. The elongation of the data spread is possibly a systematic error due to VLF propagation in the Earth-ionosphere waveguide, although this matter must be investigated further. Location errors might be improved by using an enhanced location finding algorithm that incorporates ionospheric propagation. In analyzing the data for random error, we find that the absolute location error is 20.25 ± 13.5 km for WWLL network observations in this part of Brazil.
Figure 3. Location offsets of shared WWLL-BIN events relative to the BIN-determined discharge position. Each shared BIN event is taken to be at (0, 0) and the corresponding WWLL event is plotted relative to (0, 0). Mean location offset is 3.2 km north (dashed-dotted), 7.3 km east (dotted). One standard deviation (ellipse) encompasses (0, 0).
Download figure to PowerPoint
 In addition to shared events, we consider the unshared WWLL events to determine if they are valid measurements of lightning discharges. By plotting unshared WWLL locations with all BIN locations on each individual day, we find that 300 of the 382 unshared WWLL events lie within 30 km of BIN locations. Because 30 km is of the order of magnitude of a storm system, it seems reasonable that WWLL positions within this range represent valid lightning discharges.
 WWLL events farther than 30 km from any BIN event were classified as outliers, well separated from known storm centers. Data from the 5 days we consider contains 82 outliers. In order to verify whether the outlier events are likely to be valid lightning discharges, we use independent VLF measurements from the balloon campaign [Holzworth et al., Submitted, 2003] as well as raw data from the BIN network.
 The data collection period of the 7 March 2003 balloon flight overlaps with only a few measured WWLL outliers. Even so, balloon data indicate the arrival of a lightning spheric by a spike in AC electric field strength within a millisecond of one of the WWLL outliers (Figure 4) [Holzworth et al., Submitted, 2003; Thomas et al., Submitted, 2003]. As such, we can be confident that this WWLL position was due to a lightning discharge located near the balloon.
Figure 4. Vertical AC-field measurements beginning at 00:45:23 UT from a balloon flight on 7 March 2003. The timing of a WWLL reported outlier event is marked at 00:45:24.173 UT (dashed-dotted line), coinciding to within a millisecond of the spike in the balloon-measured AC-field.
Download figure to PowerPoint
 Having found independent data that verified the existence of lightning associated with one outlier event, we look into better determining the validity of the remaining 81 outliers. By using raw data from the BIN network, we can compare WWLL events to data that may have been discarded in the measurement algorithm, but still contains valuable information. For example, if the minimum number of BIN stations did not detect an event, one or two stations may still have recorded a time and approximate location. The processed data would exclude such an event in the final lightning positions due to large uncertainties in the location. BIN algorithms also exclude IC lightning based on waveform shape, since the network is only interested in measuring CG lightning accurately [Pinto Jr., personal communication, 2003].
 We found that the event at 00:45:24.173 UT on 7 March 2003, measured by WWLL and verified by VLF balloon data, was also present in the raw BIN data, but not in the final processed BIN data. The raw BIN event occurred within 1 millisecond of the WWLL event time, but separated by a distance of 63.3 km. Since raw BIN data has low location accuracy we cannot use these locations for comparison with the WWLL positions. However, we can confirm that the WWLL events are associated with real lightning discharges occurring in the region.
 Of the 82 outlier events, 43 occurred within 1 millisecond of a BIN raw data event. Of these 43, 75% were reported by WWLL to be located within 30 kilometers of the roughly located raw BIN event.
 For the remaining 39 outliers, we analyze the raw BIN data that has been classified as IC lightning and hence discarded. Of these remaining outliers, 25 of the WWLL events occurred within 1 millisecond of BIN-measured IC flashes while 7 were within 10 milliseconds. This information leads us to believe that WWLL can measure IC lightning as well as CG, and is not currently configured to distinguish between the two types.
 Only 7 of the outlying events were not matched to BIN data in some way. Since BIN claims an 90% efficiency for the high peak current events, it is probable that BIN simply missed a few events that WWLL measured.
 This analysis of BIN observations provides good evidence of coincident lightning for ∼99% of the WWLL events. Thus, while the detection efficiency may be low, the false-positive rate is also very low.