Lightning Location and Classification With the Great Plains Los Alamos Sferic Array

In early April 2005 Los Alamos National Laboratory deployed a long-baseline electric field change network to study lightning in the Great Plains region of the United States, called the Great Plains Network (GPN) of the Los Alamos Sferic Array (LASA). The network consists of six waveform digitizing stations, with one installation per state in: Colorado, Nebraska, Kansas, New Mexico, Oklahoma and Texas. The GPN LASA is similar to the LASA in north-central Florida with the exception that the diameter is approximately seven times greater than that of the Florida network (1000 versus 150 km). Harlin et al. [2004 Fall AGU Meeting, AE33A-0181] presented a geolocation technique for the Florida network which found individual station arrival times from the peak powers of the detected sferics. This technique had the advantage of locating multiple events per recorded waveform as well as the ability to determine full 3-dimensional locations for the lightning events. Unfortunately, due to the large station baseline distances, these techniques do not work reliably for the GPN and the location and event-type classification algorithms have had to be reworked. As in prior array implementations, the arrival times of a common feature in a lightning sferic are found by cross-correlation techniques. This method choses the absolute time of a feature of interest in a reference station and determines the relative time of the same feature in the remaining stations. Relative times are converted to absolute times and the location of the common lightning features are determined via typical χ2 minimization techniques. Since the geometry of a network with such large baselines does not facilitate accurate source-height determinations, only latitude, longitude and event times are determined using the newly reimplemented cross-correlation technique. Several parameters affect the quality of the derived event times (and hence the value of the reduced χ2 for the solutions) and we address questions such as: What is the optimum amount of down-sampling that can be done (so as to speed the processing but not overly degrade the solution quality)? What range of frequencies should be kept and what range should be filtered out to yield the best arrival times? How reliable can man-made carriers be removed from the data (and to what extent does it help)? What's the best metric for choosing the cross-correlation reference waveform? These issues, as well as a review of the event-type classification algorithms will be presented.