Classification of Small Negative Lightning Reports at the KSC-ER

At present the lightning flash density is a key input parameter for assessing the risk of occurrence of a lightning strike in a particular region of interest. Since it is known that flashes tend to have more than one ground termination point on average, the use of ground strike point densities as opposed to flash densities is more appropriate. Lightning location systems (LLSs) do not directly provide ground strike point densities. However, ingesting their observations into an algorithm that groups strokes into respective ground strike points results in the sought-after density value. The aim of this study is to assess the ability of three distinct ground strike point algorithms to correctly determine the observed ground-truth strike points. The output of the algorithms is tested against a large set of ground-truth observations taken from different regions around the world, including Austria, Brazil, France, Spain, South Africa and the United States of America. These observations are linked to the observations made by a local LLS in order to retrieve the necessary parameters of each lightning discharge, which serve as input for the algorithms. Median values of the separation distance between the first stroke in the flash and subsequent ground strike points are found to vary between 1.3 and 2.75 km. It follows that all three of the algorithms perform well, with success rates of up to about 90 % to retrieve the correct type of the strokes in the flash, i.e., whether the stroke creates a new termination point or follows a pre-existing channel. The most important factor that influences the algorithms' performance is the accuracy by which the strokes are located by the LLS. Additionally, it is shown that the strokes' peak current plays an important role, whereby strokes with a larger absolute peak current have a higher probability of being correctly classified compared to the weaker strokes.

Journal of Geophysical Research: Atmospheres, 2014

We present a detailed evaluation of performance characteristics of the U.S. National Lightning Detection Network (NLDN) using, as ground truth, Florida rocket-triggered lightning data acquired in 2004-2012. The overall data set includes 78 flashes containing both the initial stage and leader/return-stroke sequences and 2 flashes composed of the initial stage only. In these 80 flashes, there are a total of 326 return strokes (directly measured channel-base currents are available for 290 of them) and 173 kiloampere-scale (≥1 kA) superimposed pulses, including 58 initial continuous current pulses and 115 M components. All these events transported negative charge to the ground. The NLDN detected 245 return strokes and 9 superimposed pulses. The resultant NLDN flash detection efficiency is 94%, return-stroke detection efficiency is 75%, and detection efficiency for superimposed pulses is 5% for peak currents ≥1 kA and 32% for peak currents ≥5 kA. For return strokes, the median location error is 334 m and the median value of absolute peak current estimation error is 14%. The percentage of misclassified events is 4%, all of them being return strokes. The median value of absolute event-time mismatches (the difference in times at which the event is reported to occur by the NLDN and recorded at the lightning triggering facility) for return strokes is 2.8 μs. For two out of the nine superimposed pulses detected by the NLDN, we found optical evidence of a reilluminated branch (recoil leader) coming in contact with the existing grounded channel at an altitude of a few hundred meters above ground.

20th International Lightning Detection …, 2008

Journal of Geophysical Research, 2011

1] We evaluated performance characteristics of the U.S. National Lightning Detection Network (NLDN) using rocket-triggered lightning data acquired in 2004-2009 at Camp Blanding, Florida. A total of 37 negative flashes that contained leader/return stroke sequences (a total of 139) were triggered during these years. For all the return strokes, locations of channel terminations on the ground were known exactly, and for 122 of them currents were measured directly using noninductive shunts. The NLDN recorded 105 Camp Blanding strokes in 34 flashes. The resultant flash and stroke detection efficiencies were 92% and 76%, respectively. The median absolute location error was 308 m. The median NLDN-estimated peak current error was −6.1%, while the median absolute value of current estimation error was 13%. Strokes in "classical" triggered flashes are similar to regular subsequent strokes (following previously formed channels) in natural lightning, and hence the results presented here are applicable only to regular negative subsequent strokes in natural lightning. The flash detection efficiency reported here is expected to be an underestimate of the true value for natural negative lightning flashes, since first strokes typically have larger peak currents than subsequent ones.

1991

The frequency distributions of the peak magnetic field associated with the first detected return stroke of positive and negative cloud-to-ground (CG) flashes was studied using lightning data from northeastern Colorado. These data were obtained during 1985 with a medium-to highgain network of three direction finders (DFs). The median signal strength of positive flashes was almost two times that of the negatives for flashes within 300 km of the DFs, which have an inherent detection-threshold bias that tends to discriminate against weak signals. This bias increases with range, and affects the detection of positive and negative flashes in different ways, because of the differing character of their distributions. Positive flashes appear to have a large percentage of signals clustered around very weak values that are lost to the medium-to-high gain Colorado detection system very quickly with increasing range. The resulting median for positive signals thus could appear to be much larger than the median for negative signals, which are more clustered around intermediate values. When only flashes very close to the DFs are considered, however, the two distributions have almost identical medians. The large percentage of weak positive signals detected close to the DFs has not been explored previously. They have been suggested to come from intracloud discharges and thus are improperly classified as CG flashes. Evidence in hand, however, points to their being; real positive, albeit weak CG flashes. Whether or not they are real positive bound flashes, it from magnetic DF networks.

Monthly Weather Review, 1998

A clear association between large peak current cloud-to-ground lightning flashes of positive polarity and sprites and elves in the stratosphere and mesosphere has been previously demonstrated. This paper reports on the first climatology of large peak current cloud-to-ground (LPCCG) lightning flashes compiled from the U.S. National Lightning Detection Network. Analysis of almost 60 million CG flashes from 14 summer months (1991-95) reveals distinct geographic differences in the distribution of positive and negative polarity LPCCGs, arbitrarily defined as flashes with peak currents Ն75 kA. Large peak current positive CGs (LPCϩCGs) are concentrated in the High Plains and upper Midwest, the region in which a large majority of optical sprite and elves observations have been obtained. By contrast, large peak current negative CGs (LPCϪCGs) preferentially occur over the coastal waters of the Gulf of Mexico and the southeastern United States. A total of 1.46 million LPCCGs were found, of which only 13.7% were ϩCGs. Almost 70% of the LPCϩCGs, however, occurred in the central United States (30Њ-50ЊN, 88Њ-110ЊW). The percentage of all LPCCGs that were positive approached 30% in the central United States compared to 4.5% for the remainder of the country. A ϩCG is 3.1 times more likely to exceed 75 kA than is a ϪCG flash on a national basis. Yet in terms of absolute numbers for all ranges of peak current Ն75 kA, negative CGs are clearly dominant. For peak currents Ն75 and 200 kA, negative CGs outnumbered positive CGs by ratios of 6.4 and 4.1, respectively. In the central United States, however, during evening hours the number of LPCϩCGs almost reaches parity with LPCϪCGs. Average stroke multiplicity also exhibited regional differences. Over a half million negative CGs and over 1000 positive CGs were found with multiplicity Ն10.

Journal of Geophysical Research, 1999

A model of return stroke detection by the U.S. National Lightning Detection Network (NLDN) magnetic direction finder (MDF) sensors is used to approximate the pulse width criterion modification made to the sensors during the 1994 upgrade. Decreasing the pulse width detection criterion used by the MDF sensors increases their effective detection range, which increases their sensitivity to weak flashes (because of NLDN network geometry, increasing sensitivity has little effect on detection of strong flashes). Consequently, we observe an increase in the weak flash counts. The increased detection of weak flashes accounts, in part, for the decrease in mean peak currents observed in subsequent years to 1994. In addition to decreasing the mean peak current of detected positive and negative flashes, the NLDN upgrade has apparently had the unwanted effect of increasing the contamination of the positive CG flash data with cloud flashes.

Journal of Geophysical Research, 1998

The detection efficiency (DE) of the U.S. National Lightning Detection Network (NLDN) has been evaluated using a large data set of video observations of cloud-to-ground lightning activity in the vicinity of Albany, New York. These data were acquired during the summers of 1993, 1994, and 1995, the latter being the year of completion of a major upgrade of the network to the improved accuracy from combined technology (IMPACT) configuration. For 1993, we find a flash DE value of 67% based upon 517 cloud-to-ground flashes documented on video. The latter two years yielded both flash and stroke DEs: in 1994, 86% of 893 flashes and 67% of 2162 strokes were detected; in 1995, 72% of 433 flashes and 47% of 1242 strokes were detected. The higher DEs of 1994 relative to 1995 are likely due to additional sensors deployed locally during the initial stage of the IMPACT upgrade. Detection efficiencies were found to vary significantly from storm to storm in each season, likely due to the inherent variability of return stroke characteristics between storms. For a special subset of 92 strokes of known location and measured electric-field change, peak current estimates were generated using the transmission-line model and a return stroke speed of 1.2x108 m/s. This speed was selected, as it is the effective speed used in present NLDN peak current estimates. For this 92-stroke data subset, the stroke DE depended upon peak current: strokes with peak currents greater than 14 kA were almost always detected (39 of 40); below 14 kA, the DE dropped until by 6-10 kA, the stroke DE was only 18% (three of 17). None of 14 strokes with estimated peak currents below 6 kA was detected. If the IMPACT design constraint of an effective 5-kA minimum peak current is applied to our 92-stroke subset, the respective flash and stroke DEs are 84% and 69%; this is consistent with NLDN model predicted performance in this area. As a faster return stroke speed, possibly 1.8x 108 m/s, would seem appropriate, the above cited current values would need to be scaled downward by a factor of 2/3, implying greater actual sensitivity of the NLDN to weaker strokes. However, a commensurate adjustment downward would be required of present NLDN-derived peak current estimates as well.

Journal of Geophysical Research, 1994

Wideband magnetic direction finders have been used to obtain a cloud-to-gro_und lightning flash count for the contiguous United States, an area of 7.7 x 10 6 km 2, for the period 1989 through 1991. Over 46 million flashes to ground were recorded and are divided among the three years, 13.4 million in 1989, 15.9 million in 1990, and 16.9 million in 1991. Maximum flash densities occur in Florida and increase each year, from 9 flashes km -2 (1989) to 13 flashes km -2 (1991). The database contains 1.7 million positive flashes divided among the three years, 0.4 million in 1989, 0.6 million in 1990, and 0.7 million in 1991. In 1990 and 1991 the positive flash density maximum occurred in Florida but elsewhere in 1989. Secondary maximum positive flash densities occur throughout the Midwest. The annual mean percentage of positive flashes in the total lightning count is 3.7% for the period 1989-1991. The interannual variation is small, ranging from 3.1% (1989) to 4.0% (1991). However, in any given year the geographical variation of the percentage of positive flashes is large. The percentage of positive flashes is 2% (1989) at the latitude of Florida but near 25% and higher at the latitudes of the upper Midwest, Maine, and Oregon.