2.1 CSU-CHILL multiparameter Doppler radar

2.1.1 Description of CSU-CHILL radar and measured variables

Table 2.1 lists the operational characteristics of the CSU-CHILL radar. It is a dual linearly polarized S-band Doppler radar. The radar alternately transmits horizontally polarized and vertically polarized microwave radiation, and simultaneously detects both polarization states via dual receivers.

For the STERAO-A project (including the two case studies to be discussed in Chapters 4 and 5) a plan-position indicator (PPI) sector scanning strategy was used. Full volume coverage of storms with the preservation of good temporal resolution (~ 6 minutes or less) was the goal. Individual elevation sweeps typically were separated by 1° or less. Sector boundaries were set in order to contain most or all of the echo exceeding 30 dBZ. At times, however, the size or proximity of the storm forced certain compromises in the scanning strategy in order to maintain good temporal resolution, including the following: decreased elevational resolution, particularly at high elevation angles; not scanning to the top of the storm echo (i.e., “topping” the storm); and more restrictive azimuthal boundaries so that the storm was not completely contained within the sector. Additionally, during the storm of 12 July 1996, two different and conflicting scanning strategies were alternately employed: the one described above, and a second one associated with another field project occurring at the same time. The latter field project was concerned mainly with sampling low elevation angles with PPI sectors, but also employed range-height indicator (RHI) scans. The STERAO-A scanning strategy had priority, but unfortunately (for the purposes of this study) during 12 July 1996 the second scanning strategy was employed at times. This decreased the temporal resolution of the full-volume sector scans around these times.

The multiparameter variables measured by the radar included horizontal reflectivity (Z_h), radial velocity (V_r), differential reflectivity (Z_dr), linear depolarization ratio (LDR), correlation coefficient at zero lag (RHO_hv), and differential phase (PSI_dp). These variables give information on the size, shape, orientation, thermodynamic phase, and radial velocity of hydrometeors in a bulk sense. Through the use of these variables, it is possible to map regions of hail, rain, and mixed-phase precipitation, as well as to distinguish large (> 2 cm diameter) hail from other hydrometeor types and to determine rain and hail rates with good accuracy (e.g., Doviak and Zrnic, 1993; Balakrishnan and Zrnic, 1990a,b; Zrnic et al., 1993).

A brief discussion of the radar variables used in this study is necessary in order to understand how they can be used to distinguish between precipitation types and to determine precipitation amounts. This discussion is in large part adapted from Carey and Rutledge (1996, 1997).

Horizontal reflectivity measures the amount of power backscattered by individual radar volumes. For scattering by spherical targets with diameters much smaller than the wavelength of the incident radar beam (i.e., the Rayleigh scattering regime; r < 0.07·LAMBDA, where LAMBDA is the wavelength of the radar), reflectivity depends on the sixth power of the hydrometeor diameter. Hence, this reflectivity factor is most sensitive to the largest particles.

The radial velocity variable measures the speed of the targets, but only along the radial direction (i.e., away from or toward the radar). This variable is determined through standard Doppler radar techniques. (See Doviak and Zrnic, 1993, for a review of these.)

Differential reflectivity measures the reflectivity-weighted mean axis ratio of hydrometeors. It is defined, in decibels (dB), as

where Z_v is the reflectivity from the vertically polarized wave. Because Z_dr is reflectivity-weighted, it is also sensitive to the dielectric factor and the size of the scattering hydrometeor. For example, water particles have a higher dielectric constant than ice particles, so a raindrop with the same size and axis ratio as a graupel particle will have a higher differential reflectivity than the graupel particle. Raindrops with diameters greater than 1 mm are deformed into oblate spheroids by aerodynamic drag as they fall. Furthermore, they tend to fall with their maximum dimension oriented horizontally (Pruppacher and Beard, 1970). Thus, Z_h > Z_v and Z_dr is positive in rain. Also, Z_dr increases with drop size. Hail tends to be spherical in shape, or if it is not, it tends to tumble as it falls (Aydin et al., 1984). Thus, regions of hail tend to have a differential reflectivity near zero. Rain and hail mixtures can also have Z_dr near zero provided the hail is large enough to dominate the reflectivity. It is possible to have negative Z_dr in regions where the hail is prolate in shape (i.e., the vertical axis is longer than the horizontal axis -– Zrnic et al., 1993; Hubbert et al., 1997). Also, for S-band radars like the CHILL, non-Rayleigh scattering by large, wet, oblate hail can drive differential reflectivity negative (Aydin and Zhao, 1990). Differential attenuation also is capable of driving Z_dr negative (Bringi et al., 1990).

Linear depolarization ratio measures the ratio of the cross-polar signal power to the co-polar power. It is defined as

where Z, the reflectivity factor, has units of mm⁶ m^-3, h and v represent the horizontal and vertical polarizations, respectively, and the first subscript stands for the backscattered electric field and the second stands for the incident electric field. If the vertically (horizontally) polarized incident electric field is aligned with either the major or minor axis of an oblate target, then there can be no horizontally (vertically) polarized backscattered field. Thus, for such a situation, LDR approaches negative infinity. However, the CHILL radar’s integrated cross-polarization isolation limits the lowest value of LDR to approximately -35 dB (Mueller et al., 1995). In thunderstorms, most oblate hydrometeors tend to wobble as they fall, so there is a distribution of canting angles which raises LDR. Irregularly shaped hydrometeors (i.e., hail with lobes) increases LDR as well. Also, LDR increases as precipitation particles become more oblate (provided they are canted) or their dielectric factor increases (Frost et al., 1991). For rain, canting is generally negligible, so LDR varies between -27 dB and -30 dB. Low-density graupel has a smaller dielectric factor than rain, so it has a lower LDR than rain, typically less than -30 dB (Frost et al., 1991). Wet aggregates in the radar bright band exhibit very high depolarization. Depolarization in hail also can be very significant because hailstones often exhibit complex motions (e.g., tumbling) as they fall (Knight and Knight, 1970). Tumbling wet hailstones can exhibit LDR values around -20 dB, with maximum values approaching -10 dB (Aydin and Zhao, 1990).

Differential phase is defined as

where PHI_dp is the differential propagation phase, and DELTA is the backscatter differential phase, both in degrees. The differential propagation phase is defined as PHI_dp = PHI_hh - PHI_vv, where PHI_hh is the cumulative phase shift for horizontally transmitted and received polarized radiation during the round trip from radar to resolution volume and back, and PHI_vv is the same for vertically polarized electromagnetic waves. Differential propagation phase results from propagation through an anisotropic medium such as rain. Backscatter differential phase results from backscatter in a radar resolution volume containing particles large enough so that they scatter in the non-Rayleigh regime. At S-band, hail is often large enough to cause such non-Rayleigh scattering (especially when wet), so DELTA must be separated from PHI_dp through the use of filtering techniques. This study used the filtering technique of Hubbert et al. (1993).

The specific differential phase, K_dp, was used in this research. It is calculated from the range derivative of PHI_dp. After filtering PSI_dp to remove the contribution from DELTA, K_dp can be estimated using finite difference techniques. Given measurements of PHI_dp at two different ranges, r₁ and r₂, K_dp can be estimated as

Specific differential phase is not sensitive to isotropic scatterers, such as quasi-spherical or tumbling hail. It is, however, extremely sensitive to anisotropic scatterers such as oblate raindrops. Thus, in a mixed phase environment, K_dp will be sensitive only to the rain portion. From theoretical considerations, Jameson (1985) showed that specific differential phase is linearly related to the liquid water content of precipitation, which allows for accurate rainfall measurements, especially in heavy rain (> 60 mm h^-1; Chandrasekar et al., 1990). Because K_dp is basically insensitive to hail, this variable provides accurate rain rate estimates even in mixed-phase environments. Specific differential phase can be used in conjunction with Z_h to separate the contributions of rain and hail to total reflectivity, and thus allow for estimates of the hail rate (hail mixing ratios) in addition to the rain rate (rain mixing ratios - Balakrishnan and Zrnic, 1990b; Carey and Rutledge, 1997).

The correlation coefficient at zero lag between the horizontally and vertically polarized backscattered radiation depends on the distributions of sizes, shapes, and canting angles of the hydrometeors, as well as on the phase shift upon backscatter. The correlation coefficient is high (> 0.97) in rain. It decreases in mixed-phase hydrometeors because the size, shape, and canting angle distributions, as well as the differential backscatter phase, broaden or increase as the ice particle size increases. Correlation decreases are maximum if the reflectivity factors from the ice and water hydrometeors are similar. If either hydrometeor dominates the reflectivity, then the correlation coefficient is weighted toward that hydrometeor type. Several researchers (Balakrishnan and Zrnic, 1990a, Kennedy and Rutledge, 1995; Carey and Rutledge, 1997) have used the correlation coefficient to infer regions of large (> 2 cm) hail and/or to distinguish between large and small (< 2 cm) hail.

2.1.2 Multiparameter radar data analysis methods

CSU-CHILL radar data were edited using the Research Data Support System (RDSS) software developed at the National Center for Atmospheric Research (NCAR; Oye and Carbone, 1981). Most ground clutter and unwanted clear air echo (including anomalous propagation) were removed by thresholding on Z_h (removing data with Z_h < 0 dBZ) and RHO_hv (removing data with RHO_hv < 0.7). The threshold value for horizontal reflectivity does not impact any of the precipitation measurements, since according to the reflectivity-rain rate relationship (Jones, 1955) used for light rain in this study, it corresponds to a rain rate much less than the 1 mm h^-1 threshold used in the precipitation analyses, to be discussed later in this section. The threshold value for the correlation coefficient is well below any value seen in real precipitation data (e.g., Carey and Rutledge, 1997), so it too will not impact the analysis in a negative fashion. Additional clutter and clear air echo were removed manually. All radial velocity folds were corrected using RDSS. Differential phase was filtered using the technique of Hubbert et al. (1993), as mentioned previously. This filter attenuates gate-to-gate fluctuations while preserving physically meaningful trends. Specific differential phase was then calculated by the finite differencing approximation in Eq. 2.4.

All variables were interpolated to a Cartesian grid using the REORDER software package, which also was developed at NCAR. The gridding scheme used a Cressman filter (Cressman, 1959), with a variable radius of influence. For the filtering process, azimuthal spacing was assumed to be 1° , the half-power width provided by the CHILL antenna. Elevational spacing was set at 1° as well, because most elevation sweeps were spaced 1° or less apart. However, during part of the 12 July 1996 storm’s lifetime, when it was closest to the radar, high-elevation sweeps were spaced greater than 1° apart. For these volumes, the elevational spacing in the gridding routine was set at 1.5°. Grid resolution for both storms was set at 1.0 km in both horizontal directions and 0.5 km in the vertical direction. However, the grid resolution should not impact the analysis to any depth, since the variable radius of influence for the Cressman filter takes into account effects of beam widening with distance, and is the major factor in determining the interpolated data values. The choice of grid resolution merely affects how those interpolated values are averaged and distributed.

Storms were gridded so as to include, as best as possible, all echo of 10 dBZ and greater from the cells of interest. After reviewing all of the radar volumes for each case study, cells which did not seem related to the main storm were removed before gridding the data if they couldn’t be excluded through choice of grid geometry. During the 10 July storm, the radar volume scans sometimes did not extend far enough in the azimuthal direction to include related cells, but were opened up at a later time to include these cells. During the 12 July storm, some cells were far from the main storm at one time, but later merged with the main storm. Unfortunately, including these cells in the grid when they were distant was not possible because of computer resource limitations, but they were included later when they were closer. This artificial exclusion of cells from both storms will have an effect on the radar data analysis, in particular the precipitation analyses. This will be taken into consideration when these data are interpreted in Chapters 4 and 5.

The multiparameter variables Z_dr, LDR, and RHO_hv are often corrupted in areas of strong reflectivity gradients because of the effect of mismatched sidelobe patterns. Herzegh and Carbone (1984) found that when the main beam sampled a region of weak reflectivity in the vicinity of a high reflectivity gradient, measurements of differential reflectivity can be dominated by large sidelobe contributions from a more reflective region. A typical method of dealing with this problem is to filter out possibly bad Z_dr data by thresholding on azimuthal and elevational reflectivity gradients. (Radial gradients do not contribute to this problem.) Often these thresholds are range-dependent, becoming more restrictive at greater ranges. In the case of Carey and Rutledge (1996), these thresholds were set by one-dimensional calculations of Z_dr bias using CSU-CHILL antenna patterns and horizontal reflectivity profiles as input. Unfortunately, such thresholding can often remove good data along with the corrupted data, since a given threshold does not always hold true throughout the data. That is, the actual reflectivity gradient threshold which signals the onset of data corruption by sidelobe contributions frequently varies throughout the data (L. Carey, private communication, 1997). Thus, a single filtering threshold, even a range-dependent one, is unsatisfactory.

The approach adopted in this work was not to perform this thresholding during the editing stages of data analysis. Instead, the unthresholded data were gridded, and any analysis software which use Z_dr, LDR, and/or RHO_hv first calculated the local reflectivity gradient in Cartesian space. Then, based on a gradient threshold, a decision was made to include or exclude the particular data from the analysis. In this manner, sensitivity tests with different thresholds could be performed more easily than the previous method. Calculating the reflectivity gradient in Cartesian space is not strictly correct, however, as the natural coordinates of the radar are spherical, and gradients in spherical coordinate space are what cause the sidelobe contamination problem. A more rigorous approach would be to calculate the gradients in spherical coordinates before gridding the data, perhaps creating a separate data field for reflectivity gradient which then can be gridded along with the other data for easy comparison and sensitivity testing. This approach will be explored in future work.

In this thesis there are two objectives focused on the radar data analysis: 1) calculate near-surface rain and hail rates, and thus make estimates of the storm-total mass flux for each precipitation type, and 2) distinguish between regions of different hydrometeor types in a bulk sense.

Rain and hail rates are calculated through the following algorithm, which is summarized in Table 2.2. If the specific differential phase at a grid point is above the noise level, 0.25° km^-1, then it is used to calculate the rain rate at that point following the method of Sachidananda and Zrnic (1987). If K_dp < 0.25° km^-1, then a reflectivity-rain rate relationship (commonly known as a Z-R relationship) from Jones (1955) is used to calculate the rain rate. If this rain rate is between 1 and 20 mm h^-1, then this rain rate is used in the flux calculations described below. If the Z-R relationship gives a rain rate higher than 20 mm h^-1, then it is assumed that the reflectivity is almost entirely due to hail (K_dp < 0.25° km^-1), and no rain rate is calculated at the grid point. Differential reflectivity also could be used to assess the possibility of hail contamination (Golestani et al., 1989), but such analysis would be subject to possible errors due to reflectivity gradients, as well as other problems which will be discussed later.

The hail rate is calculated by first using K_dp to determine the reflectivity factor (in mm⁶ m^-3) due to rain at a grid point by substituting the K_dp-R relationship from Sachidananda and Zrnic (1987) into the Z-R relationship from Jones (1955), then subtracting this rain reflectivity factor from the horizontal reflectivity factor. The difference is the reflectivity factor due to hail, which is then input into the hail reflectivity-hail rate relationship of Carey and Rutledge (1997), which is based on the Cheng and English (1983) hail size distribution, to determine hail rate in mm h^-1 liquid equivalent. However, if Z_hail is not within 7 dB of Z_rain, then the hail rate is assumed negligible, after Balakrishnan and Zrnic (1990b). Note that this method assumes Rayleigh scattering by hail, which may not be valid for S-band radars like the CHILL, especially in large hail.

There are some important sources of error in these radar-derived precipitation rates. Differential phase provides superior estimates of rain rate, especially at high rain rates (> 60 mm h^-1), but is still subject to errors on the order of 10-20% based on raindrop size distribution variation and radar estimation errors alone (Chandrasekar et al., 1990). Radar reflectivity-rain rate relationships provide comparable errors, but only when rain rates are low (< 20 mm h^-1) - otherwise they are much larger (Chandrasekar et al., 1990). Thus, 10-20% serves as a lower bound for errors in rainfall estimates, with an upper bound around 30-40% for those times when assumptions are grossly violated. These errors will affect rain rate calculations as well as the hail rate calculations. In addition, hail rate calculations are subject to further errors. After the initial rain rate calculation, the computation of reflectivity factor due to rain is subject to errors on the order of a few dBZ (Doviak and Zrnic, 1993). The choice of the Cheng and English (1983) hailstone size distribution may not be valid at any specific time during hailfall, and furthermore as mentioned above the assumption of Rayleigh scattering for hail is not always valid either, especially for large hail. Hail rate is inherently difficult to measure properly via in situ methods, so it is difficult to accurately quantify the magnitude of the error associated with these radar-inferred hail rates. However, estimates of error magnitudes on the order of 50% or more probably are not unreasonable, especially in regions where the assumptions about hail size distribution and Rayleigh scattering are not valid. For example, the given algorithm could significantly overestimate hail rate during those times that hail reflectivity is caused mostly by a few large hailstones, rather than the broader distribution predicted by Cheng and English (1983).

Precipitation fluxes are computed by summing all the precipitation rates over the whole horizontal grid at a specific vertical level (usually the lowest level - 0.5 km above ground level, or AGL -– in this case). Fluxes are calculated for the entire storm complex, not individual cells. While a cell-by-cell precipitation analysis may be preferable, in practice it would be difficult to do properly due to the possibility of the choice of analysis region geometry improperly influencing results. That is, it is often difficult to clearly distinguish between two or more cells in multicellular convection, especially on the outskirts of cores and when cores are merging or splitting. In these situations it is obvious that analysis geometry could play a major role in precipitation flux computations, making the analysis results more an artifact of the geometry than real precipitation rates. Though the possibility of quantitative cell-by-cell analysis is lost, computing fluxes for the entire storm complex removes this potentially large source of error. However, qualitative cell-by-cell analysis of radar reflectivity structure and lightning will be performed.

These precipitation analyses are similar to the methods employed by Carey and Rutledge (1997). However, there are some differences. Carey and Rutledge (1997) split the rain rate calculations into three parts, one for light rain (1 < R < 20 mm h^-1), one for moderate rain (20 < R < 60 mm h^-1), and one for heavy rain (R > 60 mm h^-1). For light rain they used the Z-R relationship of Jones (1955), and for heavy rain they used the R-K_dp relationship of Sachidananda and Zrnic (1987), but for moderate rain they used a reflectivity-differential reflectivity relationship, which is also from Sachidananda and Zrnic (1987). However, this relationship can be prone to hail contamination, so they used a Z_dr correction method based on the results of Golestani et al. (1989) when the probability of hail contamination was deemed high. This method was tested on the data for the 10 July storm. The results will be discussed in Chapter 4, but it appears that the flux calculation method of Carey and Rutledge (1997) still may be prone to hail contamination despite the correction methods, at least for the 10 July storm.

The precipitation calculation method employed in this thesis avoids this hail contamination problem by not using differential reflectivity in any way. It is especially desirable to use only K_dp and Z_h for the precipitation calculations for 10 July, as this storm was at a very long range from the radar. At long ranges the other multiparameter variables are more easily susceptible to sidelobe contamination, and thus are less useful.

Table 2.3 lists the multiparameter variable matrix employed in order to distinguish between different bulk hydrometeor species. This study employed Z_h, Z_dr, LDR, K_dp, and RHO_hv to determine regions of large hail, small hail, large hail mixed with rain, small hail mixed with rain, and rain only. This matrix is identical to the one used by Carey and Rutledge (1997). The matrix itself, which is applicable only when the environmental temperature is at or above freezing (i.e., near the surface), is a modified form of the one proposed by Doviak and Zrnic (1993). Carey and Rutledge (1997) adjusted these values in order to incorporate more recent modeling and observational results, and to maximize agreement with surface reports of precipitation type and size. Note that in this study the main concern is to distinguish between large and small hail, whether they are mixed with rain or not. Note that multiparameter radar measurements are not unique, in that more than one hydrometeor mix could give the same set of multiparameter radar measurements. Thus, this matrix is only valid as long as its assumptions are valid, and when they are not, large errors (which are difficult to quantify) could result. Thus, this matrix is best used to infer the main hydrometeor types present, and to gain a sense of the basic trends occurring.

2.2. ONERA VHF lightning interferometer

This discussion of the ITF is adapted from Mazur et al. (1997), Laroche et al. (1994), and private communications between the author and ONERA researchers, notably E. Defer and Dr. P. Laroche. The ITF uses interferometric techniques to map VHF emissions from lightning in three dimensions.

Several researchers (e.g., Warwick et al., 1979; Richard and Auffray, 1985; Rhodes et al., 1994) have used interferometric lightning mapping in the past with good success. In radio interferometry, the difference in arrival times of electromagnetic waves at two different antennas is translated into a differential phase, which can then be used to calculate the angle to the radiation source. Through the use of several antenna pairs, the source can be located in three-dimensional space. The baselines connecting the antenna pairs are measured in wavelengths and half-wavelengths in order to resolve ambiguities in the arrival angle.

The ONERA ITF system itself consists of two main receiving stations. One station consists of two separate antenna sets – the elevation sensor, which measures the elevation angle of the sources; and the azimuthal sensor to determine the azimuthal direction to the source. The second receiving station senses only azimuthal information. Each antenna set consists of circular, vertically polarized electric dipole arrays. The operational characteristics of the ONERA ITF are listed in Table 2.4. Figure 2.1 shows a schematic of the combined azimuth and elevation receiving station. The azimuth-only receiving station is similar in appearance, though it is shorter and lacks the elevation sensor. Figure 2.2 demonstrates how the two stations combine to give reconstruction of a VHF source in three-dimensions.

During the STERAO-A project, the two receiving stations were separated by a baseline approximately 40 km in length, oriented roughly northwest to southeast. (The southernmost receiving station was the azimuth-only sensor.) A spatial resolution of 1-2 km is obtained in the two lobes (east and west) whose boundaries are defined by a distance of approximately 50 km off the baseline. (See Figure 1.1 for a schematic representation of the ITF lobes during STERAO-A.) The ITF cannot unambiguously resolve the positions of VHF sources along or near the baseline, so the best ITF cases occurred when storms were situated within one of the high-resolution lobes.

The three basic types of discharges identified by the ITF are the recoil streamer, the negative leader, and the spider discharge. See Uman (1987) for a detailed review of these types of discharges. Recoil streamers and spider discharges generally are associated with intra-cloud (IC) lightning, whereas negative leaders are associated with cloud-to-ground (CG) lightning. It is important to note that these discharges do not exhaust the list of possible lightning discharges, only those that emit significant VHF radiation within the ITF’s band (110-118 MHz). In particular, events involving positive breakdown (e.g., positive leaders associated with positive CGs) do not emit much VHF radiation, nor do return strokes associated with CG lightning. (They are lower frequency emitters.)

The ITF’s elevational resolution is set to optimize the resolution of IC lightning data. At low elevation angles (i.e., less than 8°), the resolution degrades from the value listed in Table 2.4. Thus, the ITF does not resolve negative leader processes associated with CG lightning as well as say, recoil streamer processes associated with IC lightning. Additionally, after analysis of the ITF data from STERAO-A, it was found that the elevational resolution of the sensor was less than the value expected based on the results of past field projects involving the ITF (P. Laroche, private communication, 1997). Thus, the true elevational resolution of the ITF was coarser than the 0.5° value listed in Table 2.4. This has an impact on the mapping of both IC and CG lightning, but it will affect the resolution of CG data the most since the low-elevation resolution is by nature more coarse.

During post-processing of the data using the Analyse software developed by ONERA, flashes were identified and classified based on the criteria listed in Table 2.5. These criteria are based on current knowledge of the various VHF-emitting lightning discharge processes (P. Laroche, private communication, 1997). The criteria govern how the software groups individual localizations (i.e., single VHF sources) into VHF radiation bursts, and how it groups these bursts into separate flashes. The software distinguishes between IC and CG lightning on the basis of an altitude threshold on negative leader processes. Related negative leaders occurring below a given altitude, set by observations from previous field experiments with the interferometer, are classified as CG flashes. Above this altitude they would comprise an IC flash. It is important to note that spider discharges associated with ICs often resemble negative leaders associated with CGs. Thus, a group of spider discharges occurring below the threshold altitude could be incorrectly classified as a CG flash. Thus, the known elevational resolution problems with the ITF during STERAO-A could have profound impacts on the proper classification of IC and CG lightning. This possibility will be explored in Chapter 3.

2.3 National Lightning Detection Network

The NLDN is comprised of a network of magnetic direction finder (DF) sensors and time-of-arrival (TOA) sensors that spans the contiguous United States. The purpose of the network is to accurately locate, in time and space, the ground strike locations of cloud-to-ground (CG) lightning flashes. The network also gives information about the multiplicity (i.e., number of return strokes) and approximate strength of each flash. This strength estimate can be converted to an estimate of peak current through the following relation (Orville, 1991):

where I_LLP is the signal strength in Lightning Location and Protection (LLP) units, and I_peak is the peak flash current (kA).

Recently, the NLDN was upgraded to its current combination of DF and TOA technology (Cummins et al., 1996). This upgrade was reported by Cummins et al. (1996) to have improved detection efficiency in northeastern Colorado to approximately 90% or more, with a median location accuracy of 0.5 km.

An unexpected result of this upgrade is the inclusion of a previously undetected population of weak positive flashes. These discharges may not be true positive CGs; instead, many could be intra-cloud (IC) discharges which are being falsely identified as positive CGs. It has been suggested that some detected positive discharges with peak currents under 7 kA may be IC discharges (P. Krehbiel, private communication, 1997). However, no positive CGs with peak currents under this threshold were observed in any of the NLDN data for the storms of 10 and 12 July. Therefore, for the purposes of this study, all detected NLDN CGs were considered to be true CG events.

2.4 Field change meters

During the STERAO-A field project, a network of three flat plate antennas, or FCMs, was maintained. Two were located at fixed sites, and one was mobile. The FCMs were essentially identical to the unit used by Carey and Rutledge (1996, 1997). Therefore, the following discussion is adapted from those papers.

The purpose of the FCM is to measure the electrostatic field change associated with both IC and CG lightning (Uman, 1987). Thus, the total flash rate of storms can be estimated with these sensors. In this study, a FCM consisted of a disk-shaped conductor with its long axis parallel to the ground. The disk was mounted on a stand with its accompanying electronics, and then inverted toward the ground to minimize precipitation contamination. Any change in the ambient electrostatic field was accompanied by a change in the voltage difference across the output. The time constant of the resistor-capacitor network was 30 ms for the mobile FCM, and 25 ms for the fixed FCMs. These time constants are long enough to reproduce the field changes by most IC and CG lightning, yet short enough to bring the output voltage back to zero between flashes. Voltage and time data were sampled at approximately 1 kHz. These FCMs were not calibrated, so voltage data could not be used to estimate electrostatic field changes, though signal strengths could be inferred in a relative sense.

The FCMs have a detection range of approximately 35-40 km (B. Rison, private communication, 1997) at the most sensitive gain setting (the only gain setting used during almost all project days, including 12 July). However, the detection efficiency of the instruments is not well known, but probably is a function of signal strength and range to flash. Also, the FCMs do not resolve direction or range to the flash, so it was important to ensure that only one storm was in range at a time, to avoid ambiguity in flash rate estimates for a single storm.

The FCM data were post-processed using an equally weighted running mean filter designed to eliminate a known source of 60 Hz noise. Lightning flashes were then counted in the data by comparing signal amplitudes to the noise amplitude. The signal-to-noise ratio threshold for counting a particular signal as a flash was set by using the flash-counting algorithm on data from fair-weather days (i.e., days when no storms were in range), and determining an SNR ratio that was as low as possible, but would not be susceptible to noise-induced false signals. Return strokes were treated by requiring 500 ms to pass before the next signal could be counted as a flash.

Because of the unknown detection efficiency and the nature of the flash-counting algorithm, the FCMs probably cannot be used to estimate the total flash rate very accurately. However, they can be used to examine trends in flash rates with reasonable confidence, as in Carey and Rutledge (1996, 1997). However, any trends in the data need to be analyzed with caution, as due to the limited range of the instrument and the fact that the detection efficiency itself is probably range-dependent, such trends may reflect storms moving into and out of range, and not necessarily any true changes in the total flash rate.

FCM data wre obtained for part of the 12 July 1996 storm’s lifetime. Unfortunately, no FCM data were obtained for the 10 July storm. The 12 July FCM data will be utilized in Chapter 3 in an intercomparison with the ITF. They will not be used in the case studies. The reason for this is because the data for this day generally reflect the influence of the movement of the storm into and out of range, and therefore not real trends in storm flash rate. Also, the FCM appears to detect far fewer flashes than the ITF, as will be shown in Chapter 3.