Abstract

Scaling relations connecting storm electrical generator power (and hence lightning flash rate) to charge transport velocity and storm geometry were originally posed by Vonnegut in the 1960s. These were later simplified to yield simple parameterizations for lightning based upon cloud-top height, with separate parameterizations derived over land and ocean. It is demonstrated that the most recent ocean parameterization 1) yields predictions of storm updraft velocity, which appear inconsistent with observation, and 2) is formally inconsistent with Vonnegut's original theory. Revised formulations consistent with Vonnegut's original framework are presented. These demonstrate that Vonnegut's theory is, to first order, consistent with recent satellite observations. The implications of assuming that flash rate is set by the electrical generator power, rather than the electrical generator current, are examined. The two approaches yield significantly different predictions about the dependence of charge transfer per flash on storm dimensions, which should be empirically and numerically testable. The two approaches also differ significantly in their physical explanations of regional variability in lightning observations.

1. Introduction

Dramatic (2–3 order of magnitude) spatial variability is found in global observations of lightning activity (Orville and Henderson 1986; Christian et al. 1999a; Boccippio et al. 2000b). Explanation of this variability is contingent upon

  1. understanding why the underlying deep convective spectra (updraft magnitude, ice content) vary regionally, and

  2. understanding how these spectra map to lightning production.

Such understanding is a necessary precursor to physically based attempts to infer thunderstorm properties from lightning observations. This study explores one framework in which to pursue problem 2), specifically scaling representations of a hypothetical quasi-steady-state thunderstorm with simple geometry. The intent of the study is not to suggest that such representations are adequate substitutes for the full time-dependent problem, rather to examine the implications (and predictions) of such approaches when carried to their logical conclusions.

A fundamental scaling relationship between thunderstorm electrical generator power, generator current (charge transport velocity) and storm geometry was originally derived by Vonnegut (1963, hereafter V63). This assumes that a thunderstorm can be conceptualized as a quasi-steady-state electrical dipole, that the dipole charge centers are of comparable size, and that this size scales with storm dimensions. With an additional assumption of monotonic mapping between generator power and lightning flash rate, this scaling forms one possible theoretical basis for linking lightning to thunderstorm dynamics, microphysics, and geometry. Subsequent simplifications to the approach translate the original parameters to more readily observable parameters such as cloud-top height, and appear to be consistent with observation (Williams 1985, hereafter W85; Ushio et al. 2001). The simplified approach was used by Price and Rind (1992, hereafter PR92) and Price et al. (1997), who used continental-based calibrations of the relationship to derive oceanic parameterizations of the same. Subsequently, Michalon et al. (1999) attempted to modify these parameterizations to account for land–ocean cloud condensation nucleus (CCN) spectrum differences, and Anyamba et al. (2000) used the inferred highly nonlinear relationship between lightning and cloud height to test a cloud-based proxy for lightning against integrated global lightning measurements.

PR92 sought to proxy global lightning rates from observable parameters (cloud-top height), and were constrained by the limited validation data available at that time. The uncertainty in this approach was acknowledged in Price et al. (1997). More recently, the deployment of low-earth orbit lightning sensors with high detection efficiency and little bias [the Optical Transient Detector (OTD) and Lightning Imaging Sensor (LIS) sensors, (Christian et al. 1996; Christian et al. 1999b)] and possible future deployment of geostationary lightning mappers (Christian et al. 1989) lead to the possibility that the reverse calculation may be possible, that is, inference of storm properties from lightning observations. Recent numerical modeling advances [e.g., the inclusion of breakdown mechanisms in the South Dakota School of Mines and Technology Storm Electrification Model (SEM) by Helsdon et al. (1992) and in the National Severe Storms Laboratory cloud model by Mansell (2000)] also allow direct testing of lightning-convection relationships. Physically based theoretical relationships may provide one framework in which to interpret both the new empirical data and modeling results. It is thus appropriate to reexamine the most recent simplifications to Vonnegut's theory (W85) and an extrapolation of it (PR92) for consistency with both the original theory and latest observations.

Vonnegut's basic theory, assumptions (both implicit and explicit), and subsequent simplifications are presented in section 2. In section 3, it is demonstrated (using satellite-observed storm flash rate distributions) that inversion of the PR92 ocean parameterization yields predictions about marine storm updraft velocities that may be at odds with the limited in situ measurements available, and it is shown that the parameterization involves a change of variable that is inconsistent with Vonnegut's original theory. Section 4 reexamines some of the basic assumptions, and identifies those now amenable to direct testing. Section 5 explores the significant assumption that flash rate scales as generator power, rather than generator current; the two approaches yield significantly different physical explanations of observed f(z) variability, predict very different geometric weightings for flux/generator current variability, and yield testable predictions of the dependence of per-flash charge transfer on flash rate itself. Regardless of the “correct” flash rate mapping, investigation of these approaches clearly defines parameters of interest for interpretation of both observational data and numerical modeling studies.

2. Scaling history, assumptions, validation, and application

V63 formulated a scaling relation for the electrical power generated by a thunderstorm. The following assumptions were either implicit or explicit in this formulation.

  1. A thunderstorm can be conceptualized as an electrical dipole.

  2. The horizontal and vertical scales of the two dipole charge centers are comparable and vary with storm scale.

  3. A storm generator current can be conceptualized as a net charge transport velocity, which maintains the dipoles.

  4. Time variation of the generator current is small and the generator is (as required under continuity) balanced by other (unspecified) currents.

Assumption 1) is a conventional representation of storm electrical structure (Wilson 1920). While more recent empirical studies have suggested that a tripole representation is more accurate (Simpson and Scrase 1937; Williams 1989), or, in some cases, that a multipole structure appears to hold (Stolzenburg et al. 1998), these empirical studies also confirm that the main negative and upper positive charge centers in storms dominate in terms of charge density. Since the region at and between these charge centers appears to be the dominant location of microphysically induced local charge separation (Takahashi 1978; Saunders 1994), this appears to be a reasonable working assumption.

Assumption 2) has not been rigorously tested (at least in terms of horizontal scales), and there may be reason to believe that upper positive charge centers are larger in spatial extent due to cloud-top divergence. This is likely to be a factor-of-2 type inaccuracy, at best. The assumption of horizontal and vertical scale similarity (and covariance with storm size) places constraints on the applicability of this approach to storms with aspect ratio ∼1, although separation into horizontal and vertical scale components (i.e., a different charge geometry model) is not precluded under the basic theory.

Assumption 3) is a simple description of how net convergence or divergence of differentially charged particles (driven by fall speeds and storm updrafts) yields bulk charge separation. Assumption 4) implies that all quantities discussed below refer to appropriately time-averaged values.

Under these four assumptions, the electrical generator power is given as the product of generator current and the potential drop between the dipoles:

 
P = IΦ.
(1)

Assuming tangential, spherical charge centers with volumetric charge density ± ρ and radius R, maintained by a generator current density J = ρVQ (charge density × charge transport velocity; generator current IJR2),

 
Pɛ–1(ρVQR2)(ρR2).
(2)

Under this charge geometry, the dipole separation is given by zd = 2R, and area A = πR2; hence,

 
Pɛ–1ρ2VQAz2d,
(3)

(recognizing that under this geometry, it is assumed that R grows as zd). This formulation follows W85, but captures the essence of V63. It is important to note (as V63 did) that the generator power is dominated by geometric terms (width and depth of the dipole region), with a linear modulation by the generator current density. This places strong constraints on what terms may dominate when comparing two storms of comparable shape and size.

The next assumption is that the net power dissipation by lightning varies monotonically with generator power, and in absence of an empirically confirmed functional form, linear mapping is implicitly assumed:

 
PlCP,
(4)

(C is a scaling constant and is dimensionless). A secondary assumption (by W85, but not by V63) relates the flash rate linearly with lightning power dissipation, hence storm cell flash rate fc is assumed to vary linearly with P. Hence,

 
fcγCP,
(5)

where γ–1 must have units of energy (work) for dimensional consistency. The implicit assumption is thus that each flash is responsible for a constant electrostatic energy. W85 notes that limited field data may support the assumption that fcP; when flash rate is compared with gravitational power due to falling precipitation (not necessarily generator power), evidence of a constant γ is found. It is important to note that the linear fcP mapping may reasonably be expected to fail in low flash rate storms where other dissipation mechanisms (conduction currents, precipitation currents) might dominate over net lightning current.

Further simplifications are required to make Eq. (5) empirically testable or operationally useful. A common assumption is that zd (or R) varies as cloud-top height zt, although other measures (e.g., the altitude of a fixed radar reflectivity contour) could also be reasonable proxies. The zt approximation could be justified by observations that in many storms the lower negative charge region remains relatively constant in height,1 and that most upper positive charge is carried on small ice crystals with negligible terminal velocity (hence dipole separation can be linearly approximated by cloud-top height). Hereafter, zt will thus be denoted simply as z.

These simplifications yield the basic relationship,

 
fcγɛ–1ρ2VQAz2.
(6)

W85 further simplified this by invoking a scale similarity between horizontal and vertical dimensions (necessary under V63's original charge model, and thus reducing the A and z2 terms to a single length scale dependence of L4):

 
fcγɛ–1ρ2VQz4.
(7)

A further significant simplification replaces VQ with storm updraft velocity w. Here, w is the most appropriate velocity in the vertical profile to describe charge separation between the two charge centers, if such a velocity uniquely exists. Uniqueness or nonuniqueness of the relationship between VQ and an identifiable w may be a significant flaw in this simplification. Thus,

 
fcγɛ–1ρ2wz4.
(8)

The final implicit assumption is that variability in the charge density of the dominant dipole regions is small (and can thus be treated as a prescribed constant), an idea loosely corroborated by electric field soundings through MCSs (Stolzenburg et al. 1998; in such soundings, charge density in these regions does not appear to vary by orders of magnitude between storms). This has, at best, been only partially verified in continental storms.2

It is again noted that the potential direct contribution (under these simplifications) from variability in storm updraft velocity is at best linear, and geometric terms from the size and separation of the dipole centers dominate. V63 hypothesized that updraft velocity would be positively correlated with cloud-top height, and W85 gave empirical evidence from continental, midlatitude radar-based observations to support a linear relationship between w and z. PR92 refined this empirical relationship over land to a power law with exponent 1.09, and presented a similar power-law relationship over oceans with exponent 0.38. Due to the limited maritime data available to PR92, their ocean w(z) fit is almost completely dominated by several outlier data points corresponding to cloud-top heights of 2 km.3 When only data points corresponding to cloud tops above the freezing level are considered, almost any w(z) relationship could be fitted (i.e., w and z appear uncorrelated in their small sample), although oceanic w are consistently lower than continental w for similar z.

The near-linear continental w(z) relationship presented by W85 and PR92 yields the familiar fcz5 representation, where again, the highly nonlinear dependency is causally driven by geometric terms. This carries a single (measurable) parameter (cloud height), and was applied by PR92 to estimate global lightning from satellite [International Satellite Cloud Climatology Program (ISCCP)] cloud-top measurements. It is important to note that linear w(z) mapping carries with it the necessary corollary that fcw5, although this strong nonlinearity completely arises from the geometric contribution, rather than direct influence of w on charge separation.

A fifth-power fc(z) continental relationship is consistent with data presented by W85, as well as by Shackford (1960), Jacobson and Krider (1976), and Livingston and Krider (1978). Observational matching of a fifth-power univariate dependency does not necessarily validate the sequence of simplifying assumptions, it only demonstrates that they are not inconsistent with observation. While large scatter is present in the storm-by-storm data, the fifth-power dependency appears to roughly hold when observations are averaged in altitude. Ushio et al. (2001), using global storm observations from the Tropical Rainfall Measuring Mission (TRMM) LIS and precipitation radar (PR), also found that a fifth-power relationship was consistent with (though not demanded by) the continental storm observations. Perhaps more importantly, these authors found an upper limit in fc attainable by storms of a given z, strongly suggesting that scaling limitations of some sort (either through direct or indirect coupling to z) are at work in nature. PR92 compared their parameterizations over both land and ocean with regional flash rates (bulk production) observed by the Defense Meteorological Satellite Program (DMSP)–Optical Line Scan (OLS) satellite sensor. This test did demonstrate a net empirical usefulness of their parameterizations. However, this was (necessarily) not a strict test of the theory, which holds only for individual storm flash rates. Williams et al. (2000) and Boccippio et al. (2000b) have recently demonstrated that spatial variability in regional bulk flash production is dominated by variability in the frequency of storm occurrence rather than by variability in storm flash rates. Since a parameterization may accrue “skill” simply by activating whenever a deep cloud is present, validation against bulk lightning production is not necessarily a test of the fidelity of its underlying physics. In a similar integrated approach, a fifth-power weighting was used by Anyamba et al. (2000), who applied it to Television and Infrared Observation Satellite (TIROS) Observational Vertical Sounds (TOVS) cloud data and found some coherence of the weighted time series with Schumann Resonance (SR) global lightning measurements at Madden–Julian Oscillation (MJO) timescales.

It is again emphasized that any empirical tests of the univariate scaling relations are, of course, only tests of the observational consistency of the complete chain of assumptions leading to the particular formulation being tested [i.e., they cannot yet test the basic validity of Eq. (5) for lack of observations of the fundamental parameters; other constructions could conceivably yield similar univariate relationships]. Alternatively, numerical cloud modeling results could be used to investigate Vonnegut's original hypothesis [Eqs. (3), (5)] without the need for additional assumptions.

3. Testing Vonnegut's theory and later simplifications

a. Implications of PR92 ocean parameterization

PR92 derived an ocean parameterization for fc(z) based upon an assumption that the same nonlinear fc(w) [derived from the w(z) dependency over land] holds over both land and ocean. Since large-scale observations of oceanic storm flash rate are now available from the LIS sensor, the necessary implications of this assumption (specifically, predictions about the distribution of oceanic updrafts) can be directly tested.

Two years of LIS data from December 1997 to November 1999 are used in this study. Flashes are identified using the LIS version 4 (v4) “flash” data product, which clusters contiguous optical lightning pulses observed from cloud top by the sensor. Preliminary validation of this product following techniques in Boccippio et al. (2000a) suggests that nominal LIS v4 flashes may represent fragmentations of true contiguous channel structures approximately 10%–20% of the time. Storms are identified using the LIS v4 “area” data product, which clusters flashes using a fixed spatial cutoff parameter. The median and mean diameters of LIS v4 optical areas are 26 and 30 km, although the “cores” of LIS areas are typically less than 10 km wide; much of the total diameter appears to derive from multiple scattering of light within clouds.

To simplify analysis, only LIS areas observed between 80–90 s are considered here. These represent the vast majority of LIS-observed areas; storms observed for less than this duration occur during periods of sensor data buffer overflow or near the edge of the LIS field-of-view (FOV), and account for less than 15% of all areas. The LIS flash detection efficiency is conservatively estimated here to be 75% based on preliminary cross-sensor comparison studies (e.g., Thomas et al. 2000).

The limited duration of LIS observation places a lower bound on observable per-cell flash rate fc. While this cutoff is fundamentally probabilistic [observations of one flash (designated f1) in 80 s may occur from a spectrum of true instantaneous flash rates], for simplicity it is considered here to be discrete. The LIS minimum detectable instantaneous flash rate can be approximated by

 
formula

where δt ∼ 80 s, ∼ 0.75, hence fcmin ∼ 1 fl min–1. In the context of PR92's  fc(w) relationship,

 
w = 14.66 fc0.22,
(10)

(w in m s–1, fc in fl min–1), this corresponds to a minimum inferrable updraft of approximately 15 m s–1.

The observed LIS storm instantaneous flash rate probability distributions Pr(fc) over land and ocean (within the ± 35° orbit of the TRMM satellite) are shown in Fig. 1. Here, the truncated spectra are normalized using observations by Nesbitt et al. (2000) of the population of deep (ice-scattering) precipitation features identified using the TRMM microwave imager (TMI) and PR. These authors found that LIS observed no flashes in 50% of land features, and 98% of ocean features. This population of features exists below the LIS 1 fl min–1 cutoff and consists of an unknown mix of nonflashing and weakly flashing cells; the normalized distributions in Fig. 1 simply place these fractions below 1 fl min–1, so the “complete” probability distribution functions (PDFs) integrate to unity. While there is a definitional mismatch between the precipitation features of Nesbitt et al. (2000) and the LIS areas used here, this suffices as a crude estimate (and will not greatly affect the inferences below).

Fig. 1.

Probability distribution of observed storm flash rates Pr(fc) from the LIS sensor, Dec 1997–Nov 1999, using LIS v4 areas as storm definitions, LIS v4 flashes as flash definitions, only 80–90-s observations in the sample, assuming a 75% flash detection efficiency, and normalizing by 0.5 (land) and 0.02 (ocean) following Nesbitt et al. (2000). 

Fig. 1.

Probability distribution of observed storm flash rates Pr(fc) from the LIS sensor, Dec 1997–Nov 1999, using LIS v4 areas as storm definitions, LIS v4 flashes as flash definitions, only 80–90-s observations in the sample, assuming a 75% flash detection efficiency, and normalizing by 0.5 (land) and 0.02 (ocean) following Nesbitt et al. (2000). 

It is obvious from Fig. 1 that the shapes of the observed distributions do not differ dramatically; that is, when storms with flash rates greater than about 1 fl min–1 occur over land and ocean, their flash rates are similar [their means differ by a factor of 2, as reported by Boccippio et al. (2000b)]. This immediately suggests that PR92-predicted land and ocean updraft distributions will also not differ dramatically (in shape).

Once an fc(w) relationship has been either empirically or theoretically derived, the predicted probability of updraft occurrence Pr(w) can thus be calculated through a direct change-of-variable in the probability distribution

 
formula

for the domain (fc > fcmin, w > wmin).

The conditional Pr(w), predicted by Fig. 1 and Eqs. (10) and (11), is shown in Fig. 2a (these distributions do not integrate to one because of the truncated observed flash rate spectra). Again, the shapes of the distributions do not differ dramatically, and the mean predicted updrafts for the population of >15 m s–1 updrafts in flashing storms are 23 m s–1 over land, and 20 m s–1 over ocean. Prior empirical observations of continental and oceanic updraft spectra have not been presented in ways amenable to direct intercomparison, although Lucas et al. (1994) and Zipser (1994) observe that the means of the upper 10% of low altitude aircraft-observed updrafts over land and ocean appear to differ by factors of 2–3, significantly larger than the ratio of 1.15 predicted here. Again, direct intercomparison is not yet formally possible, but the suggestion is that strongly nonlinear fc(w) relationships, equal over both land and ocean (the core of the PR92 ocean parameterization assumption) might only hold if the population of low or zero flash rate (below the LIS truncation) storms is considered. Since this population may begin to violate underlying assumptions of the scaling law simplifications, this suggests that a global application of Eq. (10) is inconsistent with observation and may be inconsistent with theory.

Fig. 2.

(a) Truncated updraft spectra Pr(w) predicted using the PR92 parameterizations and the LIS-observed flash rate distribution from Fig. 1. Means of the truncated spectra are overlaid (solid circles). The truncated means do not differ appreciably, perhaps contrary to aircraft observations. (b) As in (a) but for Pr(z). Inversion of the PR92fco(zo) parameterization yields nonphysical cloud-top heights.

Fig. 2.

(a) Truncated updraft spectra Pr(w) predicted using the PR92 parameterizations and the LIS-observed flash rate distribution from Fig. 1. Means of the truncated spectra are overlaid (solid circles). The truncated means do not differ appreciably, perhaps contrary to aircraft observations. (b) As in (a) but for Pr(z). Inversion of the PR92fco(zo) parameterization yields nonphysical cloud-top heights.

The quantity Pr(z) can also be predicted, using the approach of Eq. (11). Again using PR92's parameterization, the predicted (truncated) cloud-top spectra are shown in Fig. 2b. Clearly, the PR92 ocean parameterization yields nonphysical cloud-height predictions upon inversion (70-km cloud tops for 1 fl min–1 storms). Note that these are not extrapolations, simply remappings. Noninvertibility, even of a purely empirical parameterization, suggests that key physical variability is not being correctly resolved.

b. A self-consistent approach

A significant formal inconsistency also exists in the derivation of the ocean parameterization of PR92. To understand this (subtle) departure from theory, it is necessary to retrace this derivation. The derivation is presented in generalized functional form, to mitigate confusion by specific empirical fits and parameters.

PR92 began with three functional representations, wl(zl), wo(zo), and fcl(zl). The subscripts l and o denote land and ocean, respectively. Representation wl(zl) is a near-linear empirical power-law relationship (also found by W85). Relationship wo(zo) is a power-law relationship with exponent 0.38 (with caveats on its robustness as described above). Relationship fcl(zl) is a fifth-power relationship empirically observed and, under V63, predicted by Eq. (8) and linear wl(zl). PR92 ,derive Eq. (10) as wl(fcl) = wl[zl(fcl)], then assume wo(zo) = wl[zl(fcl)] and solve for fc(z). However, it is obvious from this assumption that the resultant expression is for fcl(zo), a definitional inconsistency. Also, PR92 cite V63 and W85, yet derive a 5th-power z dependence of work, not power (as they claim), without inclusion of a generator current term VQ (i.e., w, under W85's simplifications). Their implicit assumption is thus that flash rate scales linearly with instantaneous storm electric potential energy, which is restored after each discharge in an unspecified manner. Neglect of the generator current term indirectly leads to the inconsistencies with V63.

The fcl(zo) inconsistency can be more intuitively understood by recalling that, under V63, the strong nonlinearity in Eq. (10) for fc(w) only derives from the fourth-power weighting of the geometric scaling law terms and the near-linear w(z) dependency over land (PR92's derivation differs). Yet this relationship is assumed to be regionally invariant, and provides the functional core for the remainder of the ocean parameterization. This inconsistency yields the PR92  fco(zo) parameterization

 
formula

in which, under V63 and W85, the fundamental fourth-power geometric weighting given by Eq. (8) has been diminished, implying that over oceans,

 
wozo–2.27.
(13)

This conclusion is both likely nonphysical (inverse relationship between updrafts and cloud heights) and inconsistent with the wo(zo) presented by PR92. As demonstrated in the prior section, Eq. (12) also predicts nonphysical cloud-top heights upon inversion. The correct implementation of variable w(z) relationships between land and ocean (should they actually occur) in the context of Vonnegut's theory [as simplified to Eq. (8)] should be,

 
formula

with appropriate substitutions for land or ocean w(z) relationships (theoretical or empirical) and their inverses. Here, the parameter k1 is invariant, as the scaling should yield the same dependency regardless of the functional form of the parameters chosen. This approach is thus completely capable of being calibrated by land-only observations. If different land/ocean w(z) are stipulated, Vonnegut's theory demands differing univariate relationships (both in w and z) over land and ocean, in contrast to PR92's assumption of equivalent fc(w). Further, all regional (land/ocean) variability in flash rates must, in the context of the theory, be fundamentally driven by variability in either true Pr(w), Pr(z), or w(z).

In the specific case where the land and ocean w(z) are considered to be power-law fits (as by W85 and PR92),

 
formula

we thus have,

 
formula

Column 2 of Table 1 shows the resultant power-law weightings using w(z) fits from PR92 and k1 derived from that study.4 Here, the height dependency is more consistently in the fourth- to fifth-power range, while the weak wo(zo) relationship reported by PR92 yields a much more nonlinear fco(wo) dependency.

Table 1.

Univariate scaling relationships from PR92 and this study. Units are z in km, w m s–1, f c in fl min–1. Note that the relationships derived in this study are formally consistent with V63, but may not necessarily optimally predict true over oceans (see text for discussion). Column 2 utilizes PR92 w(z) empirical fits as in section 3b; column 3 applies w(z) fits derived for zt fits derived for zt greater than 6 km, and is assumed to be the “most consistent” implementation of W85's f cPgen approach.

Univariate scaling relationships from PR92 and this study. Units are z in km, w m s–1, f c in fl min–1. Note that the relationships derived in this study are formally consistent with V63, but may not necessarily optimally predict true over oceans (see text for discussion). Column 2 utilizes PR92 w(z) empirical fits as in section 3b; column 3 applies w(z) fits derived for zt fits derived for zt greater than 6 km, and is assumed to be the “most consistent” implementation of W85's f c ∼ Pgen approach.
Univariate scaling relationships from PR92 and this study. Units are z in km, w m s–1, f c in fl min–1. Note that the relationships derived in this study are formally consistent with V63, but may not necessarily optimally predict true over oceans (see text for discussion). Column 2 utilizes PR92 w(z) empirical fits as in section 3b; column 3 applies w(z) fits derived for zt fits derived for zt greater than 6 km, and is assumed to be the “most consistent” implementation of W85's f c ∼ Pgen approach.

Figure 3a shows Pr(w) predicted using the observed LIS Pr(fc), Eq. (11), and the revised relations in column 2 of Table 1. Here, the means of the truncated updraft spectra differ by a factor of 3 (perhaps closer to observations), and the oceanic spectra are in general lower-valued and more tightly constrained. While this may be seen as an improvement, it is cautioned that the lower-valued, narrower oceanic spectra are primarily a result of the observed w underlying PR92's  wo(zo) empirical fit, which are confined to values from about 5–12 m s–1.

Fig. 3.

(a) Truncated updraft spectra Pr(w) predicted using the revised approach, PR92's ,w(z) relationships and the LIS-observed flash rate distribution from Fig. 1. Means of the truncated spectra are overlaid (solid circles). The revised approach predicts a lower-valued and more tightly constrained updraft spectrum over ocean. Note that ambiguity in PR92's  wo(zo) relationship should caution against strict use of the numerical predictions from this approach. (b) As in (a) but for Pr(z). The revised parameterizations predict a deeper population of oceanic cloud tops for the same flash rates.

Fig. 3.

(a) Truncated updraft spectra Pr(w) predicted using the revised approach, PR92's ,w(z) relationships and the LIS-observed flash rate distribution from Fig. 1. Means of the truncated spectra are overlaid (solid circles). The revised approach predicts a lower-valued and more tightly constrained updraft spectrum over ocean. Note that ambiguity in PR92's  wo(zo) relationship should caution against strict use of the numerical predictions from this approach. (b) As in (a) but for Pr(z). The revised parameterizations predict a deeper population of oceanic cloud tops for the same flash rates.

Figure 3b shows Pr(z) predicted using this approach. The new relations, when inverted, yield physically realistic cloud-top height values, and the truncated means in ocean storms are slightly (about 1.5 km) higher than in land storms (i.e., the same flash rate is predicted to occur in deeper storms over oceans than over land). Recall that this refers only to the population of flashing storms; it is cautioned that observational consistency may only be valid for the “reverse” problem, that is, prediction of z from fc. No consideration has been given to the population of nonflashing storms, so the “forward” fc(z) may easily be overpredicted using this formulation. This is also evident when Fig. 3b is compared against direct observations of oceanic Pr(z) spectra, for example, radar echo tops observed during the Coupled Ocean–Atmosphere Response Experiment (COARE) by Johnson et al. (1999), who find a significantly higher frequency of occurrence [e.g., Pr(z = 10 km) = 0.07 km–1].5

Equation (18) is not intended to supplant PR92's operational ocean parameterization for the forward problem, merely to illustrate the necessary consequences of deriving such a parameterization consistent with the underlying theory. Re-examination of the theory's underlying assumptions (including examination of critical velocities, as well as the appropriateness of the chosen observables, e.g., zt) may be required to yield a theoretically consistent and operationally useful forward fc(z) parameterization.

4. Understanding Vonnegut's theory

a. Implications of the revised approach

The revised univariate scaling law implementations can be used to assess interdependencies even in absence of rigorous wo(zo) observations. First, in all likelihood, the parameters al and ao are greater than 0 (cloud-top height is positively correlated with updraft speed). This has the necessary implication that flash rate–height relationships will always be at least fourth-power-weighted [if all assumptions leading up to Eq. (8) hold]. Equation (18) also demands that all variability in observed flash rates (e.g., between land and ocean) is determined by either: 1) differences in the underlying Pr(z), or 2) differences in the w(z) relationships, that is, kl, ko, al, and ao.

Noting that al and ao are likely >0, it is also evident that fc(w) relationships will be at least linear in w (as given by V63). Furthermore, in the possible case that ao is less than or equal to 1, this yields at least a fifth-power fc(w) dependency, and possibly much higher [PR92's ,wo(zo) yields an 11.5 exponent, as in Table 1]. Also, the fact that Eq. (19)'s land and ocean variants are not necessarily equal means that a given observed flash rate (e.g., fcmin for LIS) implies different (and likely lower) updrafts over oceans (contrasting PR92's assumption). This has the important implication that, under V63 and W85, regionally invariant univariate relationships between lightning and storm properties are not necessarily predicted to exist. Interestingly, the cause [regionally variant relationships between updrafts and realized cloud depth, w(z)] of this conclusion could be seen as implicating variability in, for example, vertical buoyancy profiles [“shape of the CAPE” (convective available potential energy)] already suspect as a factor in regional convective spectrum differences. V63 and W85 simply provide (geometrically based) energetic weighting on the potential impacts of such variability on flash production.

b. Effects of truncated LIS observations

The LIS effective minimum observable flash rate of 1 fl min–1 places some restrictions on the use of these data to test updraft spectra predictions. In the case of PR92, the inferences were limited to updrafts greater than about 15 m s–1; under the revised approach, the limit is about 7 m s–1. The effects of including lower flash rate storms in the sample have been examined by testing candidates for the probability distribution of fc < 1 fl min–1 storms.

Recalling section 3, Nesbitt et al. (2000) found that 98% of deep (ice-scattering) precipitation features over ocean were not observed by LIS to flash (50% over land) (i.e., they had fc < fcmin, possibly 0). Arbitrary hypothetical low flash rate extrapolations were constructed by assuming various percentages of these storms were truly flashing at a modest rate of 1 fl 3 min–1 (the remainder were assumed nonflashing and ignored, hence truncated updraft spectra were again predicted).

Inclusion of a population of low flash rate storms did little to alter the mean predicted updrafts over land. Inclusion of these storms over ocean did appreciably (and expectedly) alter the means. However, factor-of-2 differences in the predicted land–ocean updraft truncated means only occurred if a perhaps unreasonably large fraction (50%–90%) of all deep ocean cells were assumed to be flashing at rates greater than 0 but below the LIS fcmin. This would be at odds with observations of oceanic lightning frequency reported by, for example, Zipser (1994), or observed by the author with an electric field mill during the Tropical Ocean and Global Atmosphere (TOGA) COARE. This formally reinforces the earlier conclusion of section 3, that a globally invariant fc(w) assumption can only be reconciled with observation if a (perhaps unrealistically) large population of very low (but nonzero) flash rate storms is assumed to exist.

Using the revised parameterizations, factor-of-3 differences in the inferred truncated updraft means are maintained over a wide range of assumed low flash rate storm occurrence. Vonnegut's approach thus does not necessarily require invocation of a large ocean nonzero flash rate population to be consistent with observation.

c. Ambiguity in w(z)

Since strict adherence to V63's theory demands that land–ocean flash rate differences be attributed either to differences in Pr(w), Pr(z), or w(z), it is necessary to examine the sensitivity of these results to ambiguity in w(z). Figure 4 shows the empirical (w, z) data reported in PR92. As noted earlier, consideration of only data for z > 6 km leaves considerable ambiguity in the “true” wo(zo), should such exist. Sensitivity is tested by assuming a range of (al, ao) (from 0.1–5.0), fitting appropriate (kl, ko) to the PR92 observed data, and repredicting updraft and cloud-top spectra.

Fig. 4.

Maximum updraft velocity–cloud-top height data w(z) summarized by PR92 for land (diamonds) and ocean (stars). Overlaid are new best fits for data corresponding only to cold (z greater than 6 km) clouds (kl = 0.0615, al = 2.39, ko = 0.74, and ao = 0.92).

Fig. 4.

Maximum updraft velocity–cloud-top height data w(z) summarized by PR92 for land (diamonds) and ocean (stars). Overlaid are new best fits for data corresponding only to cold (z greater than 6 km) clouds (kl = 0.0615, al = 2.39, ko = 0.74, and ao = 0.92).

Figure 5a shows the χ2 of the empirical fits over land when only z > 6 km data are considered. Clearly a broad range of wl(zl) power-law exponents yield plausible fits, with optimal correlation corresponding to al = 2.39. Figures 5b and 5c show the predicted updraft and cloud-top height spectra, in which the logarithm of the PDFs are raster-plotted (i.e., each vertical “strip” of these figures corresponds to a predicted PDF using a given a). The overall shapes of the predicted spectra are fairly invariant over the broad range of low χ2 values, suggesting little sensitivity to the precise wl(zl) fit. Over ocean (Figs. 6a–c), the range of plausible ao is even broader (consistent with the large scatter in the raw data), with an optimal fit for ao = 0.92. Again, the sensitivity to the specific power law formulation is low. For updrafts, the ocean spectra are consistently narrow and low-valued (compared to land), regardless of the functional form of the fit. For all spectra except land updrafts, the truncated means (white overlay lines) are almost completely insensitive to the specific functional form of w(z); as a further illustration, Figs. 7a,b show the spectrum predictions using PR92's  w(z) fits and the newly computed w(z) fits above; despite significant differences in a, the spectra do not differ dramatically.6 Univariate relations corresponding to the new fits are shown in the third column of Table 1.

Fig. 5.

(a) The χ2 of power-law fits to PR92's ,wl(zl) data in which al is allowed to vary from 0.1–5. (b) Logarithm of predicted land updraft velocity PDF for each fit. Contour levels correspond to 0.1 (black), 0.05, 0.01, 0.005, 0.001, and 0.0005 (lightest grey) probabilities. PDFs have not been normalized by the results of Nesbitt et al (2000). Thick white line overlay shows the mean of each truncated spectrum. (c) As in (a) but for cloud-top height.

Fig. 5.

(a) The χ2 of power-law fits to PR92's ,wl(zl) data in which al is allowed to vary from 0.1–5. (b) Logarithm of predicted land updraft velocity PDF for each fit. Contour levels correspond to 0.1 (black), 0.05, 0.01, 0.005, 0.001, and 0.0005 (lightest grey) probabilities. PDFs have not been normalized by the results of Nesbitt et al (2000). Thick white line overlay shows the mean of each truncated spectrum. (c) As in (a) but for cloud-top height.

Fig. 6.

(a) The χ2 of power-law fits to PR92's  wo(zo) data in which ao is allowed to vary from 0–5. (b) As in Fig. 5b, but for ocean data. (c) As in Fig. 5c, but for ocean data.

Fig. 6.

(a) The χ2 of power-law fits to PR92's  wo(zo) data in which ao is allowed to vary from 0–5. (b) As in Fig. 5b, but for ocean data. (c) As in Fig. 5c, but for ocean data.

Fig. 7.

(a) The Pr(w) predictions from the revised parameterizations using the PR92  w(z) fits (section 3b.) and the optimal w(z) fits of Fig. 4 (section 4c.). (b) As in (a) but for Pr(z).

Fig. 7.

(a) The Pr(w) predictions from the revised parameterizations using the PR92  w(z) fits (section 3b.) and the optimal w(z) fits of Fig. 4 (section 4c.). (b) As in (a) but for Pr(z).

This essentially indicates that the z dependence in w(z) is unable to significantly constrain the theory, and land–ocean differences are (here) driven by the small dynamic range of observed w in PR92's (wo, zo) data. These data effectively placed bounds on ocean Pr(w), which, when mapped through Vonnegut's theory for storm energetics, are consistent with observed flash rate spectra in ocean flashing storms. Succinctly, the (limited) available data on Pr(fc), Pr(w), Pr(z), and w(z) over land and ocean are consistent with (or alternatively, cannot refute) Vonnegut's scaling theory. Equivalently, univariate relationships correctly derived from V63 and constrained by observed w(z) must yield internally self-consistent results.

d. Re-examining the underlying assumptions

It is emphasized that observational consistency has only been demonstrated for the population of flashing storms, or the reverse problem of inferring storm properties from observed flash rates. There is good reason to expect that the forward prediction problem will fail under this approach, for lack of consideration of nonflashing deep storms [apparently dominant over ocean (Nesbitt et al. 2000)]. This suggests that assumptions in the simplifications to Eq. (3) must be reexamined, tested or revised. The most fundamental assumption, that flash rate is determined by generator power, is examined in section 5. The following are also likely candidates for reexamination:

  1. AA(w, z): in situ observations suggest that the area of updraft cores may be related to their intensity (LeMone and Zipser 1980; Zipser and LeMone 1980). This could introduce a slight nonlinearity, invalidating the strict horizontal–vertical scale similarity invoked by W85. This would not violate Eq. (3), which allows separability in the geometric terms, but would require fine-tuning of the univariate relationships. This is not a fundamental obstacle, and continued observation (and reporting) of updraft core areas through in situ or overflying aircraft observations will help facilitate such tuning.

  2. ztzd : this hypothesis is now indirectly testable through the current and planned deployment of very high-frequency (VHF) time-of-arrival 3D lightning channel mapping systems [such as the National Aeronautics and Space Administration (NASA) and New Mexico Institute of Mining and Technology Kennedy Space Center (KSC) Lightning Detection and Ranging System (LDAR) and Lightning Mapping Array (LMA) networks (Poehler and Lennon 1979; Rison et al. 1999)]. The distance between the upper and lower branches of intracloud channels should provide reasonable estimates of zd, which can be directly compared against various measures of zt. This again is thus not a fundamental limitation, as more optimal zd proxies (either IR or radar-based) can presumably be identified.

    The assumption (under V63) that the volume of charge centers scales with storm size (rather than just the area) is perhaps suspect given more recent observations of the relative stratification of the main charge centers into thin layers, but this can easily be rectified by a more realistic charge model.

  3. wρ2VQ, or wVQ and ρ2 is invariant: this is perhaps the most likely point of failure in simplifications to Vonnegut's theory. In addition to considerations such as critical velocities for lightning occurrence (Zipser 1994), which could conceivably be integrated into the theory, it could also be argued that updrafts themselves play only an indirect role in determining the net generator current density. Specifically, the large-scale charge separation can be interpreted as a net convergence (or divergence) of charged particles due to differential terminal velocities: 
    formula
    Here, D is ice particle diameter [= D(T, w)], we integrate over the ice particle spectrum, and ρnet represents the net charge associated with each particle diameter in the spectrum. Parameter Vt is particle terminal velocity. The particle distribution is indirectly determined by w (available water supply, particle collision rates), as is ρnet(D) (for the same reasons). In this representation, it is clear that the vertical profile of particle terminal velocities may be at least as important as w as a direct term in the generator current density, especially for low updraft storms (i.e., w near Vt). The indirect influence of w (determination of the particle spectrum and charging rates) could easily be highly nonlinear. Any overall linear mapping that might exist between w and J (stipulated in V63's theory and W85's simplifications) would thus be fortuitous but not demanded.

    Such mapping [i.e., Eq. (20) above] is at least partially determinable from explicit numerical cloud models, which include local charge separation terms and track charge density on various hydrometeor categories. Thus, while wVQ may be a fundamental flaw in simplifications to V63, an appropriate mapping between the two may not be beyond our reach.

In summary, (1)–(3) above represent insufficiently tested assumptions in the simplifications to V63, which have at best limited and heuristic observational support. It is again emphasized that none are fatal to Eq. (3), and that all are testable using contemporary observational platforms and/or numerical models. Vonnegut's original theory (which even in simplified form appears not to be inconsistent with available observations) may thus yet hold promise as a physically based underpinning for the “inverse” problem of relating lightning flash rate observations to meteorological storm properties.7

5. An alternate approach: Flash rate set by generator current

The idea that flash rate is set by generator power is a key underpinning of W85. Implicit in this assumption are that

  1. lightning power dissipation scales linearly as generator power production, and

  2. flash rate scales linearly as lightning power dissipation,

or alternatively, that unspecified mappings in 1) and 2) combine to yield a linear dependency fcγP. However, it is also plausible to consider that net lightning current matches net generator current (or at least scales linearly with it); that is, fcIl , IlI, and hence fcI. This approach was, for example posited in a time-dependent model by Driscoll et al. (1992) and used to retrieve generator currents from observational data by Driscoll et al. (1994). One possible scaling derivation (again quasi-steady-state) from this alternate approach is presented below. A dipole geometry perhaps more consistent with recent observation (circular plates) is employed, although the specific charge configuration is not critical to the overall comparison with Vonnegut's hypothesis.

a. Charge geometry

Two circular plates are stipulated with radius R, separation distance zd, and equal and opposite areal charge densities σ. The potential difference between the plates (z = 0 is the lower plate, z = zd is the upper plate) is

 
formula

For convenience, call the geometric term above G1(R, zd). This is not dimensionless; it has units of length. Parameter G1 is shown for a range of plausible (R, zd) values in Fig. 8a. For R > 2zd , G1zd ; for R < 2zd , G1 is nonlinear in R. The vertical electric field Ez (neglecting a constant offset) is

 
formula

The field is maximum along the axis and at the plates:

 
formula

For convenience, call the geometric term above G2(R, zd). Note that G2 is dimensionless, and shown in Fig. 8b. For very large R, this correctly reduces to the infinite plate result (no dependence on zd). For very small R, the zd dependence also drops out. For intervening values, G2 varies nonlinearly but only takes on a small range of values (factor of 2 variation).

Fig. 8.

(a) Geometric factor G1 for plausible dipole dimensions—for R > 2zd (right of the dashed line), G1zd; (b) geometric factor G2, which varies by only a factor of 2 for plausible scales, (c) 1/G2, which also varies by only a factor of 2; and (d) G1/G2—for R > zd (right of the dashed line), G1/G2zd/2.

Fig. 8.

(a) Geometric factor G1 for plausible dipole dimensions—for R > 2zd (right of the dashed line), G1zd; (b) geometric factor G2, which varies by only a factor of 2 for plausible scales, (c) 1/G2, which also varies by only a factor of 2; and (d) G1/G2—for R > zd (right of the dashed line), G1/G2zd/2.

b. Estimation of flash rate

To compute flash rate, assume that the interflash time is given by the time δt it takes the generator current to convey enough charge to bring the dipoles fully to breakdown field strength. An efficiency of lightning charge removal η is included:

 
formula

where Jgen is the generator current density (prescribed by storm microphysics and dynamics, i.e., fixed here); σcrit is the charge density transferred to the dipoles required to build a critical field for breakdown, Ecrit (which varies with thermodynamic and microphysical parameters, but is assumed constant); and zd is prescribed (assumed set by storm microphysics and dynamics). It is derived from Eq. (23):

 
formula

(1/G2 is shown for reference in Fig. 8c). It is already implicit from Eq. (26) that this approach will predict at best a weak dependence of fc on storm area, deriving completely from G2 in σcrit. Flash rate is thus given by

 
formula

As stipulated, the flash rate here balances the generator current, and hence ρgenVQ is the dominant term; the geometric correction adds at best a factor of 2 variability. Hence fcIgen essentially implies fcJgen. Additional weightings by area R2 and z2d are not predicted as in V63 and W85. The prediction is thus that if fc varies dramatically with storm geometry (as observations suggest, i.e., as z5t), it is because ρgenVQ covaries with geometry, not because of energetically based geometric weighting. This is a highly important physical distinction between the explanations provided by the two approaches.

c. Lightning power dissipation

Under this framework, power P again is a current run through a potential difference, and the lightning current is assumed to match the generator current. Hence,

 
formula

This is dimensionally consistent in units of Watts. For R > zd , G1/G2zd /2 (Fig. 8d), hence,

 
formula

(still dimensionally consistent). This is the expected result, since the potential is assumed given by a critical field occurring over a given dipole separation distance. The dimensional difference in zd from V63 arises because of the assumption of 2D charge plates, in contrast to V63's spherical charge centers.

d. Charge transfer per flash

The assumptions that flash rate scales with generator power or current carry different implications for the necessary charge transfer per flash Q (since both stipulate a steady state, should satisfy current continuity, and yield different predictions for fc). Since W85 implies a fixed electrostatic energy dissipation per flash γ–1, the charge transfer per flash will necessarily vary inversely with storm potential difference. Alternatively, in an approach which stipulates fcIgen, it might appear that a fixed charge transfer per flash is predicted. However, since fc predicted under this approach essentially (to within a factor of 2) varies with Jgen, charge transfer per flash in this approach will necessarily vary directly with storm area (to maintain the stipulated lightning–generator current balance).

Under W85, since the lightning current is operating in the same Φ as the generator, lightning power dissipation is given by

 
formula

Recalling the implicit assumptions in Eqs. (4) and (5),

 
formula

or, in the context of scale similarity invoked by V63, W85,

 
QL–2.
(34)

The implication is thus that larger dipoles (deeper or wider storms, under scale similarity between the dipoles and storms, as assumed by V63 and W85) yield less charge transfer per flash (while supporting a higher fc because of their geometry). This seems intuitively palatable for deep/high flash rate storms, but seems counterintuitive for large-scale, low flash rate clouds with large dipole moment change flashes (e.g., the stratiform regions of mesoscale convective systems, winter storms, and possibly oceanic storms). Geometry, in those cases and under fcγP, seems to work in the wrong direction.

Note that this general relationship (Q varying inversely with storm scale) is insensitive to the particular dipolar charge configuration assumed. If V63 and W85 had, instead, stipulated thin charge plates (as above), Q would vary as

 
formula

that is, QL–1 (where, for R > 2zd, L is essentially zd and any R dependence drops out). This is illustrated in Fig. 9a, which shows 1/G1 for plausible scales. This general (inverse scaling) result derives because Eqs. (33) and (34) describe the implicit assumptions in W85 (fixed electrostatic energy dissipation per flash), not the formal geometry-dependent details.

Fig. 9.

(a) Geometric factor 1/G1, the geometric dependency of charge transfer per flash Q predicted by Vonnegut's fcPgen theory. (b) Parameter Q explicitly predicted by the fcIgen approach, assuming η = 1 and Ecrit = 300 kV m–1; Q appears too high—lightning efficiency η = 0.1 might yield more plausible results.

Fig. 9.

(a) Geometric factor 1/G1, the geometric dependency of charge transfer per flash Q predicted by Vonnegut's fcPgen theory. (b) Parameter Q explicitly predicted by the fcIgen approach, assuming η = 1 and Ecrit = 300 kV m–1; Q appears too high—lightning efficiency η = 0.1 might yield more plausible results.

Under the fcIgen approach derived here,

 
formula

which essentially predicts that charge transfer per flash varies directly with storm area (for Rzd, the zd dependence essentially drops out). This is again necessary as fcIgen is effectively equivalent to fcJgen, to within a factor of 2; hence, Q must vary with area to balance net current. That is,

 
QL2.
(38)

This is quite different from the W85 prediction. Here, larger storms only yield slightly higher flash rates because of their geometry (and if the observations claim they yield significantly higher flash rates, it is because ρgenVQ somehow covaries with L) and also yield more charge transfer per flash. Flash rates are more driven by generator current than storm geometry (by stipulation). Predictions of Q from Eq. (37), assuming η = 1, Ecrit = 300 kV m–1, are shown in Fig. 9b. Clearly, predicted Q are unrealistically high, suggesting η ≪ 1 (a value of 0.1 could yield plausible results).8

Differences between an L–2 or L–1 dependence and an L2 dependence should be empirically testable based on measurements of Q and storm geometry across a spectrum of storm scales.

e. Prediction of updrafts and cloud-top heights

The fcIgen approach can also be used to examine predictions of univariate scaling relationships. Constraining this application are observations suggesting that fc is highly nonlinearly dependent on zt, while the predicted fc has no (significant) direct z dependency. This has the necessary implication (under this approach), that I = I(z), or alternatively that I depends on a storm parameter, which itself is highly nonlinearly coupled with z. Ignoring the weak geometric dependency G2 (which, under scale similarity, is a constant anyway), we have, for land and ocean:

 
formula

substituting (κl, αl, kl, al) or (κo, αo, ko, ao) as appropriate for land or ocean. The (κ, α) must be determined by empirical fits [e.g., PR92's empirical fcl (zl)]. Leaving η as a free parameter, it is clear that

 
formula

As with fcPgen, the existence of regionally variant w(z) requires regionally variant fc(w). However, under fcPgen, Jw is stipulated [regional microphysical variability in J(w) is disallowed]. Under fcIgen, no such restriction is imposed, and under empirical constraints, J(w) must vary regionally, presumably reflecting variability in local microphysics.

The univariate scaling relations using (κ, α) from empirical fc(z) fits by PR92 and Ushio et al. (2001) are shown in Table 2. Ushio et al. (2001) analyzed data for August 1998 from the TRMM LIS and PR, for spatial domains including Northern Hemisphere only, and the Tropics. Fits were made to raw (fc, z) data and data averaged in altitude (as by W85). The tropical results (along with predictions from the fcPgen approach) are shown in Figs. 10 and 11. Since the z spectra shown in Fig. 11 are simply remappings of the climatological Pr(fc) through purely empirical fits, they illustrate the disagreement between fcPgen predictions and observations. Specifically, the fcPgen cloud-top spectra appear to roll off too quickly, consistent with the earlier observation that they disagree with spectra reported by Johnson et al. (1999) [over oceans, the empirical z spectra are now consistent with Johnson et al. (1999)'s spectra]. This of course does not validate the fcIgen approach, since the empirical fits are stipulated.

Table 2.

Univariate scaling relationships using the generator current approach (fcIgen). The z relationships under this approach must be observed empirically. The fc(z) fits for both raw data and averaged into height bins (approach used by W85).

Univariate scaling relationships using the generator current approach (fc ∼ Igen). The z relationships under this approach must be observed empirically. The fc(z) fits for both raw data and averaged into height bins (approach used by W85).
Univariate scaling relationships using the generator current approach (fc ∼ Igen). The z relationships under this approach must be observed empirically. The fc(z) fits for both raw data and averaged into height bins (approach used by W85).
Fig. 10.

Updraft spectra Pr(w) predicted by the fcPgen and fcIgen approaches. The fcPgen results are as in Fig. 7. (a) Here, fcIgen based on empirical fc(z) from PR92. (b) As in (a), but from Ushio et al. (2001), Tropics in August. Dual curves denote fc(z) fits computed by averaging in z (as by W85), and by taking raw (fc, z) values.

Fig. 10.

Updraft spectra Pr(w) predicted by the fcPgen and fcIgen approaches. The fcPgen results are as in Fig. 7. (a) Here, fcIgen based on empirical fc(z) from PR92. (b) As in (a), but from Ushio et al. (2001), Tropics in August. Dual curves denote fc(z) fits computed by averaging in z (as by W85), and by taking raw (fc, z) values.

Fig. 11.

As in Fig. 10, but for cloud-top height spectra Pr(z). Note that in the fcIgen approach, these are simply empirical fits to the observed data.

Fig. 11.

As in Fig. 10, but for cloud-top height spectra Pr(z). Note that in the fcIgen approach, these are simply empirical fits to the observed data.

Note that since deeper oceanic cloud tops correspond to the same flash rate as over land, if any degree of scale similarity between zd and zt holds, this approach predicts Qo > Ql (for the same flash rate), and if any degree of scale similarity between R and zt holds, this approach may predict QoQl. Since the shapes of land and ocean fc spectra are similar (Fig. 1) and differ primarily in frequency of occurrence, this prediction may be testable by examination of the relative frequency of occurrence of high dipole moment change flashes observed over tropical land and ocean by long-range extremely low-frequency (ELF) or SR techniques (at comparable ranges).

For updrafts (Fig. 10), the results are again dominated by the dynamic range in the constraining w(z) data. The fcIgen approach predicts higher mean updrafts over both land and ocean, and the land means are significantly higher than under fcPgen (30–40 m s–1 vs 23 m s–1). The large scatter in the underlying (fc, z) data (nonrobustness of the empirical fits against averaging or spatial domain) and the narrow dynamic range of the ocean spectra suggests that discrimination between the two theories will not be possible by comparison with observed ocean updraft spectra. Such discrimination might be possible over land. Overall, the land–ocean differences are again dominated by w(z), although for different reasons {microphysics, i.e., J[z(w)], rather than cloud geometry, i.e., Φ[z(w)]}.

f. Summary

In summary, W85's (fcP) approach has flash rate set, essentially, by storm geometry, with a small direct contribution from generator current density (which is stipulated to vary linearly with updrafts, i.e., Jw). Under W85, charge transfer per flash varies inversely with storm scale, is insensitive to generator current, and is “what it needs to be” to have lightning power dissipation match generator power production under a prescribed potential difference (equivalently, each lightning flash transfers a constant electrostatic energy). Under this approach, J is predicted to be significantly smaller over ocean (for a given observed flash rate). Predicted cloud-top height spectra appear to be lower than observed.

In the alternative (fcI) approach, flash rate is set directly by generator current density with slight modulation by storm geometry. Since flash rate observationally varies dramatically with storm scale, the implication is that generator current density must covary with scale (for unspecified reasons). Charge transfer per flash varies directly with storm scale (primarily area), and is what it needs to be to maintain near-critical breakdown fields under the inferred flash rate (equivalently, each lightning flash is responsible for a constant current). When constrained by empirical w(z) relationships, the implication is that J is nonlinearly dependent on w, and that this relationship is regionally variant, with exponent ∼2 over land, and ∼5–15 over ocean.9 Under this approach, J is stipulated to be identical over land and ocean (for a given observed flash rate). When driven by empirical fc(z), the approach predicts higher valued updraft spectra than W85's approach.

Empirical testing of predictions of Q and its dependence on storm scale are needed to distinguish between the two approaches from observational data analysis. This ambiguity is nontrivial, as J is presumably the quantity of greatest interest in the pursuit of storm property inferences from flash rate observations (through its coupling to updrafts w, ice q, etc.).

The two approaches are shown schematically in Figs. 12a,b. “Open boxes” and dashed lines denote empirical fits or parameterizations. The sequencing of generator-driven electrical parameters is simply the order in which they might be computed under a prescribed Jgen and geometry in a fully time-and-space dependent calculation. The fcPgen approach [under empirical w(z) constraints] predicts regionally variant fc(w) relationships; since microphysical contributions are disallowed under its Jw assumption, it is clear that such regional variability in this approach must be driven by forcing and environmental effects on the relationship between updrafts and geometry. The fcIgen does not explicitly describe the generator current–meteorology connection and allows for microphysical contributions. Under empirical w(z) and fc(z) constraints, it predicts regionally variant fc(w) and J(w). In this approach, regional variability (variability in forcing and environment) may manifest itself in J(w) by influencing local microphysics (although the exact influence is not explicitly described).

Fig. 12.

Schematic representation of the (a) fcPgen and (b) fcIgen approaches. Symbols are as follows: generator power, P; electrostatic energy, U; electric field, E; flash rate, f; potential, Φ; charge distribution, Q; generator current, I; generator current density, J; storm geometry, G; storm updraft speed, w; and hydrometeor spectra, q. Open boxes denote empirical relationships, assumptions, or parameterizations. “TBD” indicates a generator current density/storm parameter relationship which is to be determined under the fcIgen approach; either empirically [as here via fc(z)] or through numerical modeling.

Fig. 12.

Schematic representation of the (a) fcPgen and (b) fcIgen approaches. Symbols are as follows: generator power, P; electrostatic energy, U; electric field, E; flash rate, f; potential, Φ; charge distribution, Q; generator current, I; generator current density, J; storm geometry, G; storm updraft speed, w; and hydrometeor spectra, q. Open boxes denote empirical relationships, assumptions, or parameterizations. “TBD” indicates a generator current density/storm parameter relationship which is to be determined under the fcIgen approach; either empirically [as here via fc(z)] or through numerical modeling.

Since both approaches, when constrained by observed w(z), predict regionally variant fc(w), the observation that Pr(w) varies regionally is by itself insufficient to explain regional lightning variability. A complete explanation must also consider why, for example, similar updrafts over land and oceans yield different cloud geometry (under fcPgen) or why similar updrafts over land and oceans yield different generator current density (under fcIgen).

Under either approach, geometry G and generator current density J are fundamental parameters of interest (especially in the context of understanding and/or parameterizing a lightning–microphysics or lightning–dynamics relationship). This places added emphasis on the importance of calculating or observing G and J during aircraft and ground-based electrification studies, and of reporting them during explicit modeling studies.

6. Conclusions

The history of scaling relationships between thunderstorm electrical energetics and thunderstorm dynamic and geometric properties has been reviewed, with particular care given to describing the (often implicit) assumptions made when simplifying the underlying theory. The particular implementation of these relationships derived by PR92 for application to oceanic domains has been shown to yield unrealistic predictions of oceanic updraft spectra and nonphysical predictions of cloud-top heights, and to contain a formal inconsistency with Vonnegut's original theory.

It is emphasized that the operational utility of PR92's ocean parameterization for prediction of bulk (long timescale) lightning production is not addressed in this study, merely its inconsistency with the underlying theory from which it purports to derive, and its (inferred) inconsistency with the latest oceanic instantaneous storm flash rate observations and limited knowledge of oceanic updraft spectra. The empirical and theoretical inconsistencies should strongly caution against further attempts to improve the PR92 ocean parameterization from a theoretical standpoint [e.g., inclusion of the effects of CCN spectra (Michalon et al. 1999)], although of course purely empirical fine-tuning is not precluded. The inconsistencies should further caution against use of these parameterizations for the reverse problem (inference of storm properties from lightning observations themselves), although the parameterizations were never intended for this purpose.

The revised univariate scaling relations [Eqs. (18), (19)] are consistent with Vonnegut's original scaling framework and provide significant constraints on what the true univariate scaling dependencies under this theory should be, even in absence of adequate observations over oceanic regions. Using these revised parameterizations, predictions of land and ocean updraft and cloud-top height spectra from the observed flash rate distributions yield physically plausible results. This is primarily a demonstration that univariate relationships correctly derived from Vonnegut's theory must yield internally self-consistent results. The utility of these univariate relationships for the application that motivated PR92 (forward prediction of flash rates from observed cloud-top heights) has not been demonstrated here. An important implication of the revised approach is that regionally invariant univariate relationships between lightning and storm properties are not necessarily expected to exist.

Three significant assumptions in the simplified scaling laws are identified for further testing. Two, better estimation of the dipole separation distance using proxy parameters, and identification of storm cell area/height interdependency (or lack thereof), are tractable using existing data or data currently being collected, and represent “fine-tuning” of the approach. The third, the mapping between updraft velocity and generator current density, is of greater importance, is relevant for any approach, and is in principle derivable from numerical cloud models with explicit microphysics and electrification. Scaling theory, even if not operationally applied, thus helps provide an interpretation and analysis framework (and motivation) for observational or modeled data.

An alternative scaling approach, assuming flash rate varies directly with generator current, was examined. This approach necessarily yields little direct (geometrically based) dependence of flash rates on storm scales. Observations of strong flash rate/scale dependencies thus imply that generator current density varies with storm scale, for unspecified (and not necessarily directly causal) reasons. Vonnegut's approach (under Williams' simplifications) implies that charge transfer per flash will vary inversely with storm scale (especially depth), while the generator current approach implies that charge transfer per flash varies directly with storm scale (primarily area). Vonnegut's approach stipulates that, over land, generator current density varies linearly with updrafts w, and in absence of claims to the contrary, this must be assumed to apply over oceans as well. The generator current approach, paired with empirical data, requires generator current density to vary as w2 over land, and as w5w15 over ocean. For a given observed flash rate, Vonnegut's approach predicts significantly smaller generator current density over oceans, while the generator current approach stipulates that it be the same over land and ocean. Neither approach predicts a regionally invariant flash rate/updraft relationship, which implies that observations of regional variability in updraft spectra are insufficient, by themselves, to fully explain regional variability in lightning. Vonnegut's approach suggests that flash rate observations should be normalized by storm area and depth to capture variability in generator current density, while the generator current approach does not. For physically based inference of storm properties from lightning flash rate observations, the distinction is important, as generator current density is the parameter most closely coupled to meteorological parameters of interest. Determination of whether flash rate is governed by power or current (perhaps through examination of the predictions of dependence of charge transfer per flash on storm geometry) is thus required to resolve the normalization issue.

Acknowledgments

Thanks are given to Earle Williams, Kevin Driscoll, William Koshak, Walt Petersen, Steve Nesbitt, Stan Heckman, and Tomoo Ushio for extensive discussions and suggestions, and early access to data analyses in progress. This research was partially supported by NRA-99-OES-03, under the direction of Dr. Ramesh Kakar.

REFERENCES

REFERENCES
Anyamba
,
E.
,
E.
Williams
,
J.
Susskind
,
A.
Fraser-Smith
, and
M.
Fullekrug
,
2000
:
The manifestation of the Madden–Julian oscillation in global deep convection and in the Schumann resonance intensity.
J. Atmos. Sci.
,
57
,
1029
1044
.
Boccippio
,
D. J.
, and
Coauthors
.
2000a
:
The Optical Transient Detector (OTD): Instrument characteristics and cross-sensor validation.
J. Atmos. Oceanic Technol.
,
17
,
441
458
.
Boccippio
,
D. J.
,
S. J.
Goodman
, and
S.
Heckman
,
2000b
:
Regional differences in tropical lightning distributions.
J. Appl. Meteor.
,
39
,
2231
2248
.
Christian
,
H. J.
,
R. J.
Blakeslee
, and
S. J.
Goodman
,
1989
:
The detection of lightning from geostationary orbit.
J. Geophys. Res.
,
94
,
13329
13337
.
Christian
,
H. J.
,
K. T.
Driscoll
,
S. J.
Goodman
,
R. J.
Blakeslee
,
D. A.
Mach
, and
D. E.
Buechler
,
1996
:
The Optical Transient Detector (OTD).
Proc. 10th Int. Conf. on Atmospheric Electricity, Osaka, Japan, ICAE, 368–371
.
Christian
,
H. J.
, and
Coauthors
.
1999a
:
Global frequency and distribution of lightning as observed by the Optical Transient Detector (OTD).
Proc. 11th Int. Conf. on Atmospheric Electricity, Guntersville, AL, ICAE, 726–729
.
Christian
,
H. J.
, and
Coauthors
.
1999b
:
The Lightning Imaging Sensor.
Proc. 11th Int. Conf. on Atmospheric Electricity, Guntersville, AL, ICAE, 746–749
.
Driscoll
,
K. T.
,
R. J.
Blakeslee
, and
M. E.
Baginski
,
1992
:
A modeling study of the time-averaged electric currents in the vicinity of isolated thunderstorms.
J. Geophys. Res.
,
97
,
11535
11551
.
Driscoll
,
K. T.
,
R. J.
Blakeslee
, and
W. J.
Koshak
,
1994
:
Time-averaged current analysis of a thunderstorm using ground-based measurements.
J. Geophys. Res.
,
99
,
10653
10661
.
Helsdon
,
J.
,
G.
Wu
, and
R.
Farley
,
1992
:
An intracloud lightning parameterization scheme for a storm electrification model.
J. Geophys. Res.
,
97
,
5865
5884
.
Jacobson
,
E. A.
, and
E. P.
Krider
,
1976
:
Electrostatic field changes produced by Florida lightning.
J. Atmos. Sci.
,
33
,
103
114
.
Johnson
,
R. H.
,
T. M.
Rickenbach
,
S. A.
Rutledge
,
P. E.
Ciesielski
, and
W. H.
Schubert
,
1999
:
Trimodal characteristics of tropical convection.
J. Climate
,
12
,
2397
2418
.
Krehbiel
,
P. R.
,
R.
Tennis
,
M.
Brook
,
E. W.
Holmes
, and
R.
Comes
,
1984
:
A comparative study of the initial sequence of lightning in a small Florida thunderstorm.
Preprints, Seventh Int. Conf. on Atmospheric Electricity, Albany, NY, Amer. Meteor. Soc., 279–285
.
LeMone
,
M. A.
, and
E. J.
Zipser
,
1980
:
Cumulonimbus vertical velocity events in GATE. Part I: Diameter, intensity, and mass flux.
J. Atmos. Sci.
,
37
,
2444
2457
.
Livingston
,
J. M.
, and
E. P.
Krider
,
1978
:
Electric fields produced by Florida thunderstorms.
J. Geophys. Res.
,
83
,
385
401
.
Lucas
,
C.
,
E. J.
Zipser
, and
M. A.
LeMone
,
1994
:
Vertical velocity in oceanic convection off tropical Australia.
J. Atmos. Sci.
,
51
,
3183
3194
.
Mansell
,
E.
,
2000
:
Electrification and lightning in simulated supercell and non-supercell thunderstorms.
Ph. D. dissertation, University of Oklahoma, 184 pp
.
Michalon
,
N.
,
A.
Nassif
,
T.
Saouri
,
J.
Royer
, and
C.
Pontikis
,
1999
:
Contribution to the climatological study of lightning.
Geophys. Res. Lett.
,
26
,
3097
3100
.
Nesbitt
,
S. W.
,
E. J.
Zipser
, and
D. J.
Cecil
,
2000
:
A census of precipitation features in the Tropics using TRMM: Radar, ice scattering, and lightning observations.
J. Climate
,
13
,
4087
4106
.
Orville
,
R. E.
, and
W.
Henderson
,
1986
:
Global distribution of midnight lightning: September 1977 to August 1978.
Mon. Wea. Rev.
,
114
,
2640
2653
.
Poehler
,
H. A.
, and
C. L.
Lennon
,
1979
:
Lightning detection and ranging system (LDAR): System description and performance objectives.
NASA Tech. Rep. TM-74105
.
Price
,
C.
, and
D.
Rind
,
1992
:
A simple lightning parameterization for calculating global lightning distributions.
J. Geophys. Res.
,
97
,
9919
9933
.
Price
,
C.
,
J.
Penner
, and
M.
Prather
,
1997
:
NOx from lightning. 1. Global distribution based on lightning physics.
J. Geophys. Res.
,
102
,
5929
5941
.
Rison
,
W.
,
R. J.
Thomas
,
P. R.
Krehbiel
,
T.
Hamlin
, and
J.
Harlin
,
1999
:
A GPS-based three-dimensional lighting mapping system: Initial observations in central New Mexico.
Geophys. Res. Lett.
,
26
,
3573
3576
.
Saunders
,
C. P. R.
,
1994
:
Thunderstorm electrification laboratory experiments and charging mechanisms.
J. Geophys. Res.
,
99
,
10773
10779
.
Shackford
,
C. R.
,
1960
:
Radar indications of a precipitation–lightning relationship in New England thunderstorms.
J. Meteor
,
17
,
15
19
.
Simpson
,
J. C.
, and
F.
Scrase
,
1937
:
The distribution of electricity in thunderclouds.
Proc. Roy. Soc. London
,
161
,
309
352
.
Solomon
,
R.
, and
M. B.
Baker
,
1998
:
Lightning flash rate and type in convective storms.
J. Geophys. Res.
,
103
,
14041
14057
.
Stolzenburg
,
M.
,
W. D.
Rust
, and
T. C.
Marshall
,
1998
:
Electrical structure in thunderstorm convective regions. 3. Synthesis.
J. Geophys. Res.
,
103
,
14097
14108
.
Takahashi
,
T.
,
1978
:
Riming electrification as a charge generation mechanism in thunderstorms.
J. Atmos. Sci.
,
35
,
1536
1548
.
Thomas
,
R. J.
,
P. R.
Krehbiel
,
W.
Rison
,
T.
Hamlin
,
D. J.
Boccippio
,
S. J.
Goodman
, and
H. J.
Christian
,
2000
:
Comparison of ground-based 3-dimensional lightning mapping observations with satellite-based LIS observations in Oklahoma.
Geophys. Res. Lett.
,
27
,
1703
1706
.
Ushio
,
T.
,
S.
Heckman
,
D. J.
Boccippio
, and
H. J.
Christian
,
2001
:
A survey of thunderstorm flash rates compared to cloud top height using TRMM satellite data.
J. Geophys. Res.
,
106,
,
24089
24095
.
Vonnegut
,
B.
,
1963
:
Some facts and speculation concerning the origin and role of thunderstorm electricity.
Severe Local Storms, Meteor. Monogr., No. 27, Amer. Meteor. Soc., 224–241
.
Williams
,
E. R.
,
1985
:
Large scale charge separation in thunderclouds.
J. Geophys. Res.
,
90
,
6013
6025
.
Williams
,
E. R.
,
1989
:
The tripole structure of thunderstorms.
J. Geophys. Res.
,
94
,
13151
13167
.
Williams
,
E. R.
,
C. M.
Cooke
, and
K. A.
Wright
,
1985
:
Electrical discharge propagation in and around space charge clouds.
J. Geophys. Res.
,
90
,
6059
6070
.
Williams
,
E. R.
,
K.
Rothkin
,
D.
Stevenson
, and
D. J.
Boccippio
,
2000
:
Global lightning variations caused by changes in thunderstorm flash rate and by changes in the number of thunderstorms.
J. Appl. Meteor.
,
39
,
2223
2230
.
Wilson
,
C. T. R.
,
1920
:
Investigations on lightning discharges and on the electric field of thunderstorms.
Philos. Trans. Roy. Soc.
,
221A
,
73
115
.
Winn
,
W. P.
, and
L. G.
Byerley
,
1975
:
Electric field growth in thunderclouds.
Quart. J. Roy. Meteor. Soc.
,
101
,
979
993
.
Zipser
,
E. J.
,
1994
:
Deep cumulonimbus cloud systems in the Tropics with and without lightning.
Mon. Wea. Rev.
,
122
,
1837
1851
.
Zipser
,
E. J.
, and
M. A.
LeMone
,
1980
:
Cumulonimbus vertical velocity events in GATE. Part II: Synthesis and model core structure.
J. Atmos. Sci.
,
37
,
2458
2469
.

Footnotes

Corresponding author address: Dennis J. Boccippio, National Space Science and Technology Center, NASA/Marshall Space Flight Center SD-60, Marshall Space Flight Center, AL 35812. Email: Dennis.Boccippio@msfc.nasa.gov.

1

This observation seems corroborated in isolated storms by a number of studies reviewed by W85, and by results of Krehbiel et al. (1984) and Heckman and Ushio (1999, personal communication) utilizing time-of-arrival channel and electric field-based charge center locations, but called into question in MCSs and supercells from electric field soundings by Stolzenburg et al. (1998).

2

Alternatively, the assumption wρ2VQ would yield the same result but not demand invariant charge densities.

3

Within the context of the theory, w(z) is a statistical relationship between instantaneously observed separation velocities in the charging zone and dipole separation distance (not a vertical profile of w). Proxying this relationship by maximum updraft velocity and cloud-top height is, again, an observational convenience that may or may not be intrinsically meaningful.

4

Parameter k1 may be estimated by equating Eq. (18) [using the PR92 ,wl(zl) relationship] to PR92's empirical land fc(z) fit and solving. This yields an expression that scales as z–0.19. This term varies negligibly over tropospheric cloud heights and is replaced with a constant value of 0.62, yielding k1 = 1.4314 × 10–5. Alternatively, a fit of Eq. (18) may be made directly to land (fc, z) data, fixing (al, kl) and hence removing any z dependency from k1.

5

Note that the spectra reported by Johnson et al. (1999) also included warm precipitation features, which comprise at least 20%–30% of the overall spectrum. To enable direct comparison with predicted Pr(z) as normalized by the results of Nesbitt et al. (2000), this warm population should be excluded from consideration, yielding Pr(z = 10) ∼ 0.1.

6

Note that under the new w(z) fits, the calibration constant kl must be recomputed; following the same approach as earlier [matching to PR92's empirical land f(z) curve], kl is now estimated at 1.7204 × 10–5.

7

At the very least, belief in Eq. (6) suggests that we should normalize flash rate observations by the geometric terms Az2d, if such are available, if we wish to relate them to generator current density VQ Belief in Eq. (3) further suggests that results from 1D or 1.5D cloud–electrification models that impose spatial scales upon the simulated updrafts [e.g., Solomon and Baker (1998)] should be interpreted with care.

8

Driscoll et al. (1994) argued that the ratio of lightning to generator current (η here) might be 0.33–0.66, while instantaneous discharge charge removal efficiency (perhaps analagous to η here) has been reported as 0.3–0.5 in thundercloud discharges (Winn and Byerley 1975) and 0.2–0.4 in laboratory (polymethylmethacrylate) discharges (Williams et al. 1985).

9

Interestingly, an empirically driven power-law J(w) prediction with very small coefficient and very large exponent might be interpreted as evidence of a critical velocity and an updraft spectrum that rarely strays very high above it.