Comparison of PDD and National Lightning Detection Network Data

Overview of the National Lightning Detection Network Data

We acquired stroke-level lightning event data from the National Lightning Detection Network (NLDN) (Global Atmospherics, Inc.) for the period April-September 1998. A description of this specific, custom, data set has been provided by Jacobson et al. [2000]. A general description of the NLDN has been provided by Cummins et al. [1998].

In summary, for a given candidate stroke, data received from no fewer than 3 of 59 Lightning Positioning and Tracking (LPATS) III time-of-arrival (TOA) sensors located within the continental United States were reprocessed using relaxed criteria in order to maximize the detection probability of intracloud (IC) and distant cloud-to-ground (CG) events. The reprocessed data were provided in a stroke-level format with microsecond timing precision.

Specifically, no maximum limit on the range between the event and the sensors was applied, and the reprocessing accommodated ionospherically propagated signals. The resultant event data included "unverified" event locations from very distant CG discharges, occurring thousands of kilometers outside the NLDN network, as well as very energetic IC events within or near the network. No polarity was assigned to those very distant CG events that likely experienced ionospheric pathing.

In the remainder of section 4, we will make reference to "range-corrected" event times. An important caveat that the reader needs to understand is that NLDN event locations outside of the NLDN network (e.g., distant CG events) likely have much larger error ellipsoids than their "within-network" counterparts. Thus any attempt to correct correlated PDD event times will result in range-corrected PDD event times with an inherently larger error compared to the case when the PDD/NLDN correlations occur within the NLDN network. We accept this error as necessary in order to examine NLDN-detected events occurring over the ocean.

In section 4.2 we present a case study illustrating PDD performance against a string of NLDN-reported lightning events overflown by FORTE. We then discuss the PDD's relative stroke-level and flash-level performance against NLDN-reported events. Last, we examine the PDD's detection efficiency versus NLDN-reported positive, negative and unknown polarity lightning events. We emphasize that we will not ascribe an absolute detection efficiency to the PDD versus lightning. We will only ascribe a relative detection efficiency to the PDD versus NLDN-reported lightning events contained in our NLDN data set.

A Case Study Using National Lightning Detection Network Data

To demonstrate the claim that the PDD detects lightning, we compared NLDN data to those event data acquired by the PDD during passes over the continental United States. To illustrate how the two data sets compare during a typical pass, we selected a line of thunderstorms that was overflown by FORTE off the northeastern coast of the United States on April 27, 1998. This comparison is made to confirm absolutely the claim that the PDD detects lightning rather than some other phenomenon.

For the purpose of illustrating a simple comparison, we selected a period of 215 s starting at 0722:20 UTC when the trailing edge of the PDD FOV covered a portion of the North American coastline. The subsatellite point moved from northwest to southeast. During this time period, NLDN reported 503 cloud-to-ground strokes, of which 325 occurred within the PDD FOV. During this same period the PDD detected 178 events having the same characteristic optical waveform as the signal shown in Figure 4a. After range correcting the PDD event times with respect to each and every NLDN event location, we found 61 coincidences whereby PDD events occurred within 500 s of a NLDN event. Plate 1 shows the time-integrated PDD FOV. Also shown are the NLDN events (black crosses) occurring within the PDD FOV. The NLDN event locations for which coincident PDD events exist are further highlighted by red diamonds.

Thus, for this squall line the PDD detection efficiency against NLDN-reported cloud-to-ground lightning strokes was 19%. These 61 PDD-NLDN close coincidences also represent 34% of all PDD events during this interval. We infer that most PDD-detected events probably occur within the cloud where the NLDN detection efficiency is poor, an inference that will be supported in section 4.3. The important point to draw from the comparison in this section is that the PDD does indeed detect optical signals from lightning.

PDD Stroke/Flash Detection Efficiency Versus NLDN-Reported Events

We endeavored to assess the detection efficiency of the PDD against NLDN-reported cloud-to-ground strokes by employing the same 6-month NLDN event data set described in section 4.1. To obtain this detection efficiency, we divided the number of PDD-NLDN coincident events by the number of NLDN events occurring within the PDD FOV. In this exercise, we modified the NLDN event times by correcting for the optical signal transit time from the NLDN location to the FORTE location. A temporal coincidence window of 1 ms, centered on a given PDD event time, was used to define stroke-level coincidences. For flash-level coincidences a temporal window of 1200 ms, centered on a given PDD event time, was employed.

For the period April-September 1998 the PDD detected 80,067 events over the continental United States. During the same period of time, NLDN reported 105,000 stroke-level (positive, negative, and undetermined polarity) events within the PDD field of view. Of these 80,000 PDD events, 6500 occurred within 500 s of a NLDN-reported event lying within the PDD FOV. Figure 5 shows the occurrence frequency distribution of 4439 PDD-NLDN (negative CG) event pairs, for which the PDD event time has been range-corrected, that are correlated within 10 ms. A strong peak in the distribution occurs at 240 s, demonstrating that stroke-level PDD-NLDN coincidences are tightly grouped compared to interstroke timescales. In examining the longer timescales we find that 24,000 PDD events occurred within 600 ms of the same set of NLDN-reported events. These findings translate into an overall PDD stroke detection efficiency, versus NLDN-reported events, of 6% and a flash detection efficiency of 23%. Moreover, only 8% of the PDD event data set is represented by the PDD-NLDN stroke-level coincidences. This latter point argues that the PDD preferentially detects in-cloud optical emissions.

PDD Detection Efficiency Versus Positive, Negative, and Unknown Polarity CGs

Table 1 gives the PDD detection efficiency versus NLDN-reported positive, negative, and unknown polarity cloud-to-ground (CG) strokes for the April-September 1998 period. We reiterate that most unknown polarity CG events occurred far outside of the NLDN network so that the sferic signal likely followed a multihop path to the sensors, militating against polarity determination. The overall PDD detection efficiency for NLDN-reported strokes occurring within the PDD field of view is low. Positive CG strokes are detected with marginally greater efficiency (5.6%) compared to negative CG strokes (4.6%). However, the NLDN CG events having an undetermined polarity were more than twice as likely (13.6%) to be detected as those having a known polarity. There are three possible explanations for the greater detection efficiency for these unknown polarity CG events.

These CG events of unknown polarity are typically distant and located outside of the NLDN array (more often over ocean than land) so that the received signal is suspected of experiencing ionospheric reflection, resulting in ambiguities in polarity determination. There are three arguments that one can employ to explain the greater PDD detection efficiency for these events. The first is that in order for NLDN detection to occur, these distant events are necessarily strong sources. The second explanation is that maritime clouds are typically thinner than their continental counterparts, increasing detection probability for an event of any brightness. Both of these first two explanations are consistent with Christian and Latham's [1998] finding that lightning observed by NASA's Optical Transient Detector (OTD) exhibited a higher mean radiance per flash over ocean compared to land. Also, in an analysis of data from NASA's OTD and Lightning Imaging Sensor (LIS) imaging optical sensors, Boccippio et al. [2000] found that while the mean per cell flash rate was higher over land compared to ocean, the optical signals from oceanic lightning flashes were generally brighter than those over land. Boccippio et al. [2000] also suggested that either the oceanic flashes were more energetic or that the maritime cloud optical depths were consistently less than their continental counterparts.

The third explanation for the relative PDD detection efficiency superiority over ocean is that these untyped CG events are actually misclassified intracloud lightning events, for which we have inferred the PDD detection efficiency to be inherently greater. Whatever the explanation, these distant NLDN events have a higher probability of detection by the PDD than NLDN-reported events occurring within the continental United States.

On the basis of the low overall stroke-level detection efficiencies and the small overlap between the PDD and NLDN data sets, we infer that the PDD preferentially detects in-cloud optical emissions from lightning. Moreover, Suszcynsky et al. [2000] convincingly argued that the PDD-observed optical emissions associated with CG events likely originate from within the cloud rather than from below the cloud. We conclude that the PDD preferentially detects optical lightning emissions originating from within the cloud and that the PDD, in a statistical sense, is blind to below-cloud optical lightning emissions.

Next: Statistical Characte Up: Optical observations Previous: Data Reduction