relationships between tropical cyclone intensity and satellite...

JANUARY 1999 103C E C I L A N D Z I P S E R

q 1999 American Meteorological Society

Relationships between Tropical Cyclone Intensity and Satellite-Based Indicators ofInner Core Convection: 85-GHz Ice-Scattering Signature and Lightning

DANIEL J. CECIL AND EDWARD J. ZIPSER

Department of Meteorology, Texas A&M University, College Station, Texas

(Manuscript received 17 December 1997, in final form 10 March 1998)

ABSTRACT

A key component in the maintenance and intensification of tropical cyclones is the transverse circulation,which transports mass and momentum and provides latent heat release via inner core convective updrafts. Thisstudy examines these updrafts indirectly, using satellite-borne observations of the scattering of upwelling mi-crowave radiation by precipitation-sized ice particles and satellite-borne observations of lightning. The obser-vations are then compared to tropical cyclone intensity (defined here as maximum sustained wind speed) andthe resulting relationships are assessed. Substantial updrafts produce large ice particles aloft, which in turnproduce microwave ice-scattering signatures. The large ice, together with supercooled liquid water also generatedby substantial updrafts, is a necessary ingredient in charge separation, which leads to lightning. Various parametersderived from the inner core ice-scattering signature are computed for regions encircling hurricanes and typhoons,and observations of lightning activity or inactivity are analyzed.

High correlations with future tropical cyclone intensity result from the ice-scattering signature parametersmost closely associated with the areal extent of at least moderate precipitation rates. As expected, the relationshipreveals increasing intensity with increasing ice-scattering signature. Indicators of more intense convection yieldless information concerning tropical cyclone intensity. Correlations tend to be of the same sign for both presentcyclone intensity at the time of the satellite overpass and subsequent intensity change. Correlations are higherfor future cyclone intensity than for either of these. The lightning observations are much more limited than themicrowave observations, because the short amount of time in which lightning can be detected may not adequatelyrepresent a particular storm’s electrical activity. The inner core lightning observations show no clear relationshipto tropical cyclone intensification. However, the lightning observations do suggest an increased likelihood ofinner core lightning in weak tropical storms and strong hurricanes/typhoons. In the examination of case studies,the paradoxical situation of much greater lightning frequency in rainbands than in eyewalls is noted.

1. Introduction

Observations of tropical cyclones are limited by theirremote locations over the oceans. Satellites (and in somecases reconnaissance aircraft) provide the most commonmeans for monitoring these storms before they makelandfall, where they rapidly weaken. The tropical cy-clone responds to latent heat fluxes from the underlyingocean surface. The secondary (transverse) circulation ofa well-organized hurricane features low-level inflow tothe eyewall, where latent heat is released in outwardsloping convective updrafts. In response to the upwardbranch of this circulation, subsidence warming in theeye is hydrostatically responsible for the low centralpressure, which, in turn, sustains the primary circula-tion. The central pressure is sometimes referred to as

Corresponding author address: Daniel J. Cecil, Dept. of Meteo-rology, Eller O&M Building, Texas A&M University, College Station,TX 77843-3150.E-mail: [email protected].

the tropical cyclone’s intensity; in this paper, the max-imum sustained wind speed is used to define intensity.

Research aircraft can provide direct measurements ofthe vertical drafts in tropical cyclones. Studies of hur-ricanes using flight-level measurements (horizontal res-olution ;100 m) (Jorgensen et al. 1985) and verticallypointing Doppler radar (horizontal resolution ;750 m)(Marks and Houze 1987; Black et al. 1996) found typ-ical maximum updraft velocities in eyewalls to be lessthan about 8 m s21. More extreme updrafts occasionallyare found, but in general hurricane updrafts are a factorof two weaker than continental updrafts (Jorgensen etal. 1985). Hurricane rainbands were found to haveslightly weaker updraft magnitudes than eyewalls (Jor-gensen et al. 1985). All of these studies (and others)point to the conclusion that tropical cyclones (and trop-ical oceanic convection in general) contain rather mod-est vertical velocities. Typical values for mean verticalvelocities in eyewall updrafts are ;4 m s21, with max-imum vertical velocities typically 7–8 m s21. Black etal. (1996) provides a more thorough review of literatureon this subject.

104 VOLUME 127M O N T H L Y W E A T H E R R E V I E W

Of course, practical constraints limit the availabilityof such data from research aircraft. Although aircraftdata and ground-based radar data can provide detailedinformation on the storms they do sample, satellite datafacilitate the observation of tropical cyclones in remotelocations. Visible and infrared (IR) imagery from geo-stationary satellites provide the most continuous datafor these storms. Data at microwave frequencies frompolar-orbiting satellites, however, are more directly re-lated to precipitation than are those from visible and IRchannels. The upwelling radiation at these microwavefrequencies can therefore be used to assess the latentheat release in the tropical cyclone’s precipitation re-gions. Previous studies (Glass and Felde 1992; Rao andMacArthur 1994; Rodgers et al. 1994; Rodgers andPierce 1995; Rao and McCoy 1997) have related in-formation from microwave data to tropical cyclone in-tensity, future intensity, and intensity change with vary-ing degrees of success.

The 85-GHz microwave channel is sensitive to pre-cipitation-sized ice particles, which scatter the upwell-ing radiation and reduce the brightness temperature(Spencer et al. 1989). This result is termed an ‘‘ice-scattering signature,’’ as the depressed brightness tem-perature indicates the presence of precipitation-sized icealoft. A low 85-GHz brightness temperature can there-fore imply increased updraft strength. Stronger updraftsproduce more supercooled liquid water, and hence largergraupel through riming. The large graupel scatter theupwelling microwave radiation. In conjunction with thestrong updrafts and large ice, increased latent heat re-lease and precipitation can be inferred. We rarely detectthe extreme ice-scattering signatures associated withstrong updrafts and large graupel in tropical cyclones,and explore whether or not such an occurrence is relatedto intensity or intensity change. Normally less extremeice scattering is observed.

Glass and Felde (1992) examined the 85-GHz hori-zontally polarized brightness temperature for a set ofwestern North Pacific tropical storms and typhoons. Thesample included 19 storms from 1987 and 1988, with15 Defense Meteorological Satellite Program SpecialSensor Microwave/Imager (SSM/I) overpasses of inten-sifying storms and 10 overpasses of decaying storms.Boxes having ‘‘radii’’ (half the length of the side) rang-ing from 0.258 to 3.08 lat were constructed about eachstorm, and the percentage of pixels having brightnesstemperatures below threshold values was computed foreach box. The 0.58 radius box generally produced thehighest correlations with tropical cyclone intensity. Theoptimal brightness temperature thresholds were 220 Kand 230 K, with each producing correlations exceeding0.9 for intensifying storms and 0.7 for decaying storms.

Rao and MacArthur (1994) examined western NorthPacific typhoons, primarily from 1987, and the asso-ciated precipitation fields as derived from SSM/I bright-ness temperatures. All four SSM/I frequencies (19.35

GHz, 22.2 GHz, 37.0 GHz, 85.5 GHz1) were utilizedin the precipitation algorithm. Rainfall rates were morehighly correlated with 24-h future typhoon intensitythan with either current intensity or 24-h change of in-tensity. Among boxes with radii ranging from 0.258 to4.08, rainfall rates in the 2.08 radius box produced thehighest correlation (0.68) with future intensity. This cor-relation improved to 0.81 when the sample was limitedto northwestward-moving typhoons, but this reduced thesample size from 27 overpasses to 16. Rao and McCoy(1997) found similar correlations between 85-GHzbrightness temperatures on the left side of the typhoon(relative to the direction of motion) and 24-h futureintensity for nearly the same set of typhoons.

With a dramatically larger sample size, Rodgers andPierce (1995) examined precipitation characteristics de-rived from the SSM/I 85-GHz channel for western NorthPacific tropical depressions, tropical storms, and ty-phoons occurring during the period from 1987 through1992. In this study, the 12-h increase in inner core (1.08radius) rain rate was compared to the 12-h and 24-hchange in maximum wind speed of the tropical cyclone.Cases in which the rain rate decreased over 12 h wereignored. A total of 123 SSM/I observations of tropicaldepressions, 61 observations of tropical storms, and 73observations of typhoons comprised the sample. Resultsfor both 12- and 24-h change in maximum wind weresimilar, with tropical depressions yielding a correlationof 0.25 and tropical storms yielding a correlation of0.51. Surprisingly, the typhoon sample yielded a neg-ligible correlation (0.04).

Rodgers et al. (1994) performed a similar study usingWestern North Atlantic tropical cyclones, but with amuch smaller sample size. Twenty-two observations oftropical depressions, 17 observations of tropical storms,and 11 observations of hurricanes were used. Correla-tions were much smaller than those from the westernNorth Pacific for both tropical depressions and tropicalstorms (0.06 and 0.27, respectively), but the hurricanesample produced a correlation of 0.78. Rodgers andPierce (1995) suggest that differences in sample size ora greater number of steady state/weakening typhoons inthe western North Pacific sample may be responsiblefor this inconsistency. Adverse effects of vertical windshear and decreasing sea surface temperatures (SST) arealso proposed as possibly influencing more storms inthe typhoon sample than in the hurricane sample.

A less commonly used (and less commonly available)approach for judging the convection in a tropical cy-clone is to analyze lightning activity. Most theories forcharge separation (leading to lightning) involve colli-sions between large and small ice particles in the pres-ence of supercooled liquid water (Illingworth 1985).

1 For simplicity, the 85.5-GHz channels are referred to as ‘‘85GHz’’ in this paper.


However, supercooled liquid water is somewhat rare intropical cyclones (Jorgensen et al. 1985; Willoughby etal. 1985; Black and Hallett 1986). This is consistentwith the previously cited studies that reveal rather mod-est updraft strength. The presence of lightning in a trop-ical cyclone may therefore reveal that the convection isparticularly vigorous. Many of the studies of lightningin tropical cyclones have utilized the National LightningDetection Network or similar land-based networks,which only observe lightning within a few hundred ki-lometers of land. Nonetheless, individual reports (e.g.,Black et al. 1986; Lascody 1992; Black et al. 1994;Lyons and Keen 1994; Molinari et al. 1994, 1999; Hen-ning 1997; Samsury et al. 1997) link lightning in theinner regions of tropical cyclones with intensificationof the cyclones. In some of these reports, lightning issuggested as a precursor to subsequent tropical cycloneintensification (Lyons and Keen 1994) or an indicatorof ongoing intensification (Molinari et al. 1999; Hen-ning 1997), whereas others point out the occurrence oflightning before, during, and after intensification (Sam-sury et al. 1997).

The manner in which tropical cyclone inner core up-drafts are related to the intensity of the parent tropicalcyclone is an open question. The 85-GHz brightnesstemperature is physically linked to these updrafts, inthat a depressed brightness temperature requires largeice aloft, which in turn requires a significant amount ofvertical mass flux and latent heat release in the pro-duction of the ice. Similarly, lightning appears to requirea combination of large and small ice particles togetherwith supercooled liquid water. Significant vertical massflux and latent heat release would also be required tomeet these conditions. It is plausible that both of theseobservables should be related to the intensity of a trop-ical cyclone in some way. Previous research has indeedsuggested relationships. Wide varieties of approacheshave been attempted using 85-GHz brightness temper-atures, and generally have found low brightness tem-peratures to be related to tropical cyclone intensity, asexpected. Studies of lightning in tropical cyclones havebeen severely hampered by the difficulty in detectinglightning at remote locations. These studies tend to focuson storms in which both lightning and cyclone inten-sification are observed, but most do not systematicallyconsider storms in which one occurs without the other.

In the present study, several aspects of observed 85-GHz brightness temperature fields are examined fortheir relationships with tropical cyclone intensity. Ob-servations of lightning from the optical transient detec-tor (OTD) (Goodman et al. 1996) are also comparedwith tropical cyclone intensity. For both observables,indicators of convective activity are considered statis-tically for a large number of storms and are also con-sidered in selected case studies. The key questions tobe addressed are the following: 1) Which indicators ofconvective activity provide the most information abouttropical cyclone intensity? 2) How much information

do these indicators provide? 3) Are these indicatorsmore related to tropical cyclone intensity at the time ofthe observation, future intensity, or intensity change?

2. Data and methods

a. SSM/I

The DMSP F11 and F13 satellites carry SSM/I in sunsynchronous orbits with overpasses near local sunriseand local sunset. The SSM/I swath width is approxi-mately 1400 km and the 85-GHz footprint is approxi-mately 13 km 3 15 km (Hollinger 1991). The orbitsprovide at most two ‘‘independent’’ observations of atropical cyclone per day, about 12 h apart. Additionaloverpasses may occur within 1 h of each other, butwould provide somewhat redundant information. Suchoverpasses are not included in the statistical portion ofthe study, but are instead considered in case studies, asthey provide some insight into how rapidly the featuresbeing studied evolve, and how representative a singlesnapshot at 12-h intervals might be. Because the 1400km swath width is less than the distance between orbitsin the Tropics, not every location is viewed twice on agiven day. Due to the swath width, the movement oftropical cyclones, and occasional losses of data, gaps ofup to ;60 h exist in the database.

As mentioned before, upwelling 85-GHz radiation isscattered by precipitation-sized ice, reducing the bright-ness temperature. To differentiate between low bright-ness temperatures due to ice scattering and those dueto the low surface emissivity of the ocean (especiallyevident in the horizontally polarized channel), Spenceret al. (1989) define a polarization corrected temperature2

(PCT) as

PCT 5 1.818TBV 2 0.818TBH, (1)

where TBV is the brightness temperature for the verti-cally polarized channel and TBH is that of the horizon-tally polarized channel at 85 GHz. Spencer et al. (1989)found that the PCT range of 250–260 K is generally athreshold below which precipitating systems are found,with 250 K roughly corresponding to a moderate rainrate (;3 mm h21). Spencer et al. and the Goddard scat-tering algorithm (GSCAT, described by Adler et al.1994) basically utilize a linear relationship between 85-GHz brightness temperature depression and rain rate.Based on rain rate estimates from GSCAT, Mohr andZipser (1996a,b) used the existence of a 225 K PCT(;10 mm h21) to indicate the presence of cumulonim-bus convection. Independently, McGaughey et al.(1996), using high-resolution data from the AdvancedMicrowave Precipitation Radiometer, derived 225 K as

2 Because the 85-GHz signatures respond to scattering instead ofemission, the PCT is not a thermal signal. It should not be confusedwith thermodynamic temperature.


FIG. 1. Typhoon Oscar 85-GHz PCT field at 2117 UTC 15 Sep-tember 1995. The innermost 48 radius is plotted, with range rings at0.58 intervals. Mean PCT for the four innermost 0.58 radius annuliand best-track wind speed (kt) (1 kt ø 0.5 m s21) interpolated totime of overpass and 6-h intervals thereafter are given in the header.

a threshold for tropical oceanic convection. Indeed, 225K pixels at the satellite scale invariably feature inho-mogeneities with lower brightness temperatures in thehigh resolution data, indicative of deep convection.Mohr et al. (1996) noted a dramatic increase in lightningflash rates in continental convection having PCT below200 K.

In the present study, parameters based on the 85-GHzPCT field are computed for a combination of annularand circular regions centered on the tropical cyclones.These parameters are the areal mean (or average) PCT,the minimum PCT, and the percentage area having PCTat or below threshold values of 250 K, 225 K, and 200K. These are referred to as 250 area, 225 area, and 200area in later discussions. Lower thresholds were origi-nally considered but rarely met, providing little infor-mation. Following Mohr and Zipser (1996a,b) and Mohret al. (1996), the 250 K PCT is considered an indicatorof moderate rain, 225 K PCT is considered an indicatorof deep convection, and lower PCTs are considered in-dicators of more intense convection. For convenience,the term ‘‘convection’’ is used rather loosely in thispaper to refer to regions meeting the specified PCTthresholds. The PCT parameters are computed for 0.58-lat wide annuli covering the inner 28 radius of a tropicalcyclone, 18 and 28 radius circles obtained by combiningthe annuli, and the 18–28 radius annulus. Figure 1 dem-onstrates this geometry, with 0.58-wide annuli coveringthe inner 48 radius of Typhoon Oscar (1995). The rangerings are centered on the interpolated best-track posi-tion, which does not exactly match the eye in the PCTfield. This discrepancy is addressed in section 2c. Also

included in the figure are the areal-mean PCT for thefour innermost annuli and the best-track maximum sus-tained wind speed interpolated to the time of the over-pass and to 6-h intervals thereafter. In this example, asmall rain-free eye is enclosed by an eyewall with rel-atively intense convection. The high PCT in the eye(.275 K, as for the oceanic background) combined withthe low eyewall PCT produce an intermediate (243 K)mean PCT for the innermost 0.58 region. The 0.58–18region, located just outside the eyewall, contains littlerain and correspondingly has a high mean PCT (263K). Only the three innermost annuli are fully sampledby the SSM/I in this example. In such cases where thestorm is located near the edge of the SSM/I swath, onlythose regions that are fully sampled are included in thecomputations of the PCT parameters. If nearly simul-taneous overpasses are available from the F11 and F13satellites, whichever provides greater data coverage forthe location of the storm is used.

b. OTD

The OTD is in low earth orbit at 740 km with a 1300km 3 1300 km field of view (Goodman et al. 1996).The OTD monitors an oxygen emission line at 777.4nm to detect intracloud and cloud-to-ground lightningboth night and day. The orbital and field of view char-acteristics place a particular region in the OTD’s fieldof view at most twice per day, for less than 5 min at atime. As with SSM/I, a tropical cyclone may go unsam-pled (or poorly sampled) by OTD for multiple days. TheOTD detects optical pulses (events) associated with thedissociation and excitation of oxygen due to lightning(Christian et al. 1989) and groups them into flashes.Instrumental spatial resolution is on the order of 15 km,but satellite navigation errors typically reduce the lo-cation accuracy to 30 km (D. Boccippio 1997, personalcommunication). The flash detection efficiency at thenominal instrumental threshold is estimated to be rough-ly 55% during both day and night. Overall, less than10% of the flashes reported in the OTD dataset are falsealarms, which are produced by radiation, electronicnoise, or solar glint (K. Driscoll 1997, personal com-munication).

The detection efficiency and short view time limit theOTD’s ability to monitor low flash-rate storms. Thisshould be particularly troublesome for oceanic convec-tion, where lightning frequency is an order of magnitudeless than over continents (Orville and Henderson 1986;Goodman and Christian 1993; Zipser 1994). These fac-tors, combined with the location uncertainty in the light-ning data, limit the conclusions to be drawn concerninglightning in this study. However, given the extremelylow flash rates found in the inner regions of tropicalcyclones, any observation of lightning during such ashort sampling time may be significant (Molinari et al.1999). OTD provides an opportunity to systematicallyexamine some aspects of the electrical activity over re-


mote tropical oceans. Lightning (or the lack of light-ning) in the innermost 18 radius of tropical cyclones isexamined in this study. Molinari et al. (1998) found alocal minimum in flash rates extending 100 km beyondan inner maximum in hurricanes. This result suggeststhat the OTD location uncertainties should not signifi-cantly limit the ability to distinguish between inner andouter core flashes. Because view times vary across theOTD swath, some criteria must be established to de-termine whether a particular orbit adequately samplesa storm. In this study, an orbit is used if (a) any flashesare detected in the innermost 18 radius circle or (b) atleast 80% of this region is viewed for at least 180 s.

c. Tropical cyclone track and intensity information

Tropical cyclone post-analysis best tracks for 1995and 1996 were provided by the National Hurricane Cen-ter (now known as the Tropical Prediction Center) forAtlantic (ATL) and eastern North Pacific (NEPAC) trop-ical cyclones and by the Joint Typhoon Warning Center(JTWC) for western North Pacific (NWPAC) tropicalcyclones. The postanalysis merges available data to pro-duce a (sometimes smoothed) time history of the cy-clone’s center position, maximum sustained winds, andminimum sea level pressure. The available data some-times include ship reports and land- or island-based ob-servations, or aircraft reconnaissance data in accessibleregions of the ATL. Otherwise, satellite estimates oflocation and intensity, relying heavily on the Dvoraktechnique (Dvorak 1984), are used. Important questionsregarding the reliability of these estimates for researchhave been raised (JTWC 1995). In general, all of thebest-track intensities can be questioned, but some es-timate ultimately must be made.

The current project uses maximum sustained windspeed as the definition of ‘‘intensity.’’ Minimum sealevel pressure is also commonly referred to as intensity.Without direct measurements of either of these, the NE-PAC and NWPAC best tracks essentially assume a directrelationship between wind and pressure. The ATL besttracks contain separate information for wind and pres-sure. Pressure was experimented with as the definitionof intensity in this study, with the resulting correlationsslightly lower than for those where wind is used.

The best-track locations and intensities are interpo-lated to the times of SSM/I and OTD overpasses, andalso to 6-h intervals up to 36 h subsequent to eachoverpass. In some cases the interpolated storm centerclearly does not agree with the PCT field. For example,in Fig. 1 the eye in the PCT field is located southeastof the interpolated center. This may be due to inaccu-racies in the best track, smoothing of a trochoidal wob-ble in the track, or parallax effects in the satellite data.For the 1995 storms, subjective adjustments to the best-track positions were considered. The resulting changeswere insignificant in terms of the statistics for the largesample, but significant for some individual cases. The

example in Fig. 1 is not representative of the typicaloffsets.

Tropical cyclones are removed from the databasewhen they move poleward of 358 or within 100 km ofland. This minimizes the number of storms in the samplewhose intensity is being significantly affected by inter-actions with land or by transitions to extratropical cy-clones; the study focuses on intensity and intensitychange when these processes are not at work. The landrestriction is applied for both islands and continents.This removes some intensifying cyclones from the sam-ple, but is preferable to including storms that are weak-ening due to proximity to land, which would compro-mise the experimental design. Land contamination ofthe SSM/I data is not a concern, as PCT is basicallyinsensitive to the background surface. For the statisticalportion of the SSM/I study, only tropical cyclones athurricane/typhoon strength at the time of the satelliteoverpass are considered. Tropical depressions and trop-ical storms are included in the OTD portion of the study,where a greater sample size is needed.

d. Statistical analysis

For the overpasses that remain in the dataset afterthese restrictions are imposed, linear correlations arecomputed between each of the PCT parameters de-scribed in section 2a (for each annular or circular region)and the tropical cyclone intensity at the time of theoverpass, the future intensity at 6-h intervals after theoverpass, and the intensity change between the time ofthe overpass and the 6-h intervals. Correlations are notcomputed using the lightning data because so manyOTD overpasses reveal no lightning associated withtropical cyclones. Instead, the percentage of OTD over-passes that do indicate lightning is examined as a func-tion of tropical cyclone intensity, future intensity, orintensity change. The number of satellite overpasses inthe computations decreases with increasing size of theregion considered (e.g., the inner 18 radius of a stormmay lie within an SSM/I data swath whereas the inner28 radius is only partially sampled). It also decreaseswith increasing time for the future intensity or intensitychange computation (e.g., a storm may move polewardof 358 15 h after the overpass, in which case only 0-,6-, and 12-h intensities remain in the dataset). The num-ber of overpasses used for each computation is givenin Table 1.

e. Case studies

Case studies displaying the evolution of the PCTfields and the observations (or lack thereof ) of lightningdemonstrate some of the reasons for the relationshipsthat emerge from the statistical analyses. They also dem-onstrate some of the usefulness and some of the draw-backs of the data that do not show up in the simpleparameters used for the statistical analyses. ATL Hur-


TABLE 1. Number of SSM/I observations in sample by annularregion and lag time between satellite overpass and tropical cycloneintensity measure.

0–56 km 56–111 km 111–166 km 166–222 km

Atlantic0 h6 h

12 h18 h24 h30 h36 h

107989186837772

101928681787268

89807671676359

82747166635956

Eastern North Pacific0 h6 h

12 h18 h24 h30 h36 h

31302727262625

28272525252524

27262424242423

24232121212120

Western North Pacific0 h6 h

12 h18 h24 h30 h36 h

139129122111103

9891

134124117106

989387

130120114103

959286

117108102

92848177

TABLE 2. Linear correlation coefficient between 08 and 18 radiusPCT parameters and 24-h future hurricane/typhoon intensity.

All ATL NEPAC NWPAC

Mean PCT250 area225 area200 areaMin. PCT

20.650.610.450.20

20.35

20.730.730.540.07

20.21

20.840.740.790.46

20.69

20.620.540.390.26

20.42

FIG. 2. (a) Correlations between 08 and 18 radius PCT parameters and hurricane/typhoon intensity at the time of SSM/I overpass and 6-hintervals afterward: (b) ATL only, (c) NEPAC only, and (d) NWPAC only.

ricane Luis, NEPAC Hurricane Barbara, and NWPACSuper Typhoon Ryan (all from the 1995 season) arepresented for individual study. These storms were se-lected because they are relatively well sampled by theSSM/I and OTD, and because among them they dem-onstrate aspects of the data most relevant to interpretingthe statistical results. They are among the more long-lived and intense storms in the dataset, with shorter-lived storms being more poorly sampled and less con-ducive to individual study. To better understand the cas-es (particularly the larger-scale environments and thetemporal evolution between SSM/I or OTD overpasses),14-km resolution IR imagery is incorporated (but notshown) in this portion of the study.

3. Results

The main questions in this paper ask which of theindicators of convective intensity show the strongestrelationships with tropical cyclone intensity, how strongthose relationships are, and whether the relationshipsare stronger for current intensity, future intensity, orintensity change. For clarity, results involving ice-scat-tering signatures and those involving lightning are pre-sented separately. To aid in interpretation of the results,ice-scattering signatures and lightning are then consid-ered together in case studies of three individual tropicalcyclones.

a. Relationships between ice-scattering signature andhurricane/typhoon intensity

The ice-scattering signatures tend to have higher cor-relations with future storm intensity than with eithercurrent intensity or intensity change. Figure 2 demon-strates the correlations between the PCT parameters forthe 18 radius circle and hurricane or typhoon intensityor future intensity. Table 2 lists these correlations for24-h future intensity. Figure 3 demonstrates similar cor-relations, but with respect to intensity change. In eachof these figures (and also Figs. 4–5 to follow), the neg-ative of the correlations for average PCT and minimumPCT are plotted for clarity, so they will have the samesense as the correlations for area below the thresholdPCT values. When PCT parameters for separate annularregions are considered (Figs. 4–5, Table 3), the corre-lations with future intensity increase inward, as ex-pected. The correlations with intensity change do not


FIG. 3. As in Fig. 2 except for correlations involving intensity change.

FIG. 4. (a) Correlations between 24-h future hurricane/typhoon intensity and PCT parameters for 0.58 wide annular regions. The outerradius (8 lat) of the annulus is plotted on the abscissa: (b) ATL only, (c) NEPAC only, and (d) NWPAC only.

contain a clear trend, but are generally so low that theywill not be discussed. Figure 2 clearly implies that av-erage PCT and 250 area are highly (negatively) corre-lated with each other, and they tend to produce the high-est correlations with intensity and future intensity. In-dicators of more intense convection (such as the min-imum PCT or 200 area) produce consistently lowercorrelations with tropical cyclone intensity and futureintensity. Comparing Figs. 2b–d, it is apparent that thehighest of these correlations emerge from the NEPACsample, followed fairly closely by the ATL sample andnext by the NWPAC sample.

Figure 4 and Table 3 demonstrate that although theparameters associated with the intensity of convectionproduce their highest correlations in the innermost 0.58radius circle, the parameters associated with moderaterainfall produce substantial correlations in both of thetwo innermost 0.58 radius annuli. Combining these an-nuli into a 18 radius circle provides a better depictionof the tropical cyclone, and most of the correlationsinvolving this circle (as in Fig. 2) exceed those involvingeither of the constituent annuli.

In Fig. 2, the highest correlations between 18 radiusareal mean PCT and future intensity occur with 18-hintensity for ATL storms (20.74), 18-h intensity forNEPAC storms (20.86), and 12-h intensity for NWPACstorms (20.65). The total sample (all three basins com-bined) produces a 20.67 correlation with 12-h intensity.More insight into the relationships between mean PCTand intensity can be found by examining scatterplots ofthe data.

As a starting point, Fig. 6 displays the hurricane ortyphoon intensity at the time of the overpass as a func-tion of 18 radius areal mean PCT for all of the SSM/Ioverpasses in the sample. In general, more intense trop-ical cyclones are associated with lower average PCT,but the overall correlation is only 20.54 and there is alarge amount of scatter in the relationship. All but twoof the hurricanes and typhoons that exceed 130 kt (67m s21) at the time of the overpass represent the NWPAC.These account for many (but not all) of the greatestoutliers in the plot. Otherwise the greatest outliers haveextremely low mean PCT (below 250 K) but have sus-tained winds of less than 80 kt (41 m s21). Only one ofthe NEPAC observations has such a low mean PCT.Whether this is a matter of sampling (the 1995 and 1996NEPAC hurricane seasons were not typical, as will bementioned later, and the NEPAC sample size is muchsmaller than the ATL and NWPAC sample sizes) or isa consistent feature of the NEPAC has not been deter-mined.

Figure 7a is similar to Fig. 6, but it shows the 24-hfuture intensity instead of the intensity at the time ofthe overpass. Not surprisingly, the largest outliers to thebest-fit line belong to the NWPAC sample. Several otherfeatures can be noted from Fig. 7a. All of the mostintense storms again belong to the NWPAC sample, andthe least intense storms belong to the NEPAC. Severaloverpasses (some from each basin) find storms with 24-h intensities between 110 and 120 kt (57–62 m s21),but with a wide range of mean PCT. In fact, once thefuture intensity exceeds 90 kt (46 m s21), hardly any


FIG. 5. As in Fig. 4 except for correlations involving 24-h intensity change.

FIG. 6. Intensity (kt) (1 kt ø 0.5 m s 21) at the time of the SSM/I overpass as a function of 08–18 radius areal mean PCT. Lettersindicate the basin producing each data point: A 5 ATL, E 5 NEPAC,and W 5 NWPAC.

TABLE 3. Linear correlation coefficients between PCT parametersfor 0.58 wide annular regions and 24-h future hurricane/typhoon in-tensity for ATL, NEPAC, and NWPAC combined.

0–0.58 0.5–18 1–1.58 1.5–28

Mean PCT250 area225 area200 areaMin. PCT

20.480.480.410.22

20.41

20.620.540.340.07

20.24

20.400.24

20.0120.0720.17

20.300.170.040.11

20.27

relationship with mean PCT can be found. On the otherhand, 90% of the storms with 24-h intensities exceeding90 kt (46 m s21) are associated with mean PCT of 260K or less.

Considering the basins separately, most of the ATLobservations (Fig. 7b) fall close to the best-fit line,which is similar to the best-fit line for the three basinscombined. The correlation coefficient here is 20.73.The correlation improves to 20.84 in the NEPAC, butthe slope of the best-fit line changes markedly (Fig. 7c).Comparing Fig. 7c with Fig. 7a, the NEPAC storms dofall along either best-fit line, within the range of scatterobserved for other basins. If the observations havingmean PCT below 250 K were removed from the otherbasins, the slopes of the corresponding best-fit lineswould similarly steepen. In contrast to the ATL andNEPAC, a tremendous amount of scatter is found forthe NWPAC (Fig. 7d). The correlation is only 20.62.The best-fit line is similar to that for the total dataset,but large outliers from either line abound. Many of thegreatest outliers are not the result of poorly positioningthe storm center (as might be suspected), but insteadrepresent very intense typhoons whose mean PCT is notcorrespondingly extreme.

If the average PCT is better correlated with futureintensity than current intensity, there should also be acorrelation with intensity change. Figure 8 reveals aweak relationship between this parameter and 24-h in-tensity change. The most rapidly intensifying stormsproduce mean PCT between 250 and 260 K, but otherstorms in the same PCT range weaken dramatically. Justas Fig. 7a indicated that 90% of the storms having 24-h intensities greater than 90 kt (46 m s21) also had meanPCT below 260 K, Fig. 8 indicates that 90% of the

storms in the sample that strengthened by at least 20 kt(10 m s21) in 24 h also had mean PCT below 260 K.Although this threshold certainly does not indicate thatintensification will occur, it does seem to indicate thata hurricane or typhoon has the potential to either reachor maintain (if it has already reached) an extreme in-tensity.

Generally speaking, most of the correlations betweenPCT and intensity change are very low, with a largeamount of scatter, as in Fig. 8. This is the case regardlessof which PCT parameter or annular/circular region isconsidered. An exception is that the intensity changecorrelations are consistently higher in the NEPAC thanin other basins. The highest correlations involving in-tensity change are found between the 1–1.58 radius an-nulus minimum PCT and 24-h intensity change for theNEPAC. This parameter shows no relationship with in-tensity change in the other two basins, but in the NEPACthe correlation reaches 20.69. The NEPAC sample sizeis much smaller than the other two, but a clear rela-


FIG. 7. (a) As in Fig. 6 except for 24-h future intensity (kt) (1 kt ø 0.5 m s21): (b) ATL only, (c) NEPAC only, and (d) NWPAC only.

FIG. 8. As in Fig. 6 except for 24-h change of intensity (kt) (1 ktø 0.5 m s21) after the overpass.

tionship does appear, with weakening storms producinghigh minimum PCT in this region.

b. Relationships between lightning and tropicalcyclone intensity

Due to limitations in the OTD dataset, a large numberof storms is needed to conduct a thorough investigationof possible relationships between lightning and tropicalcyclone intensity. The requirement that a region beviewed for 180 s if no flashes occur can severely restrictthe sample size. However, 1 flash in 180 s would pro-duce a 20 flash per hour flash rate, which many tropicaloceanic mesoscale convective systems fail to produce(Restivo 1995). Reduction of the view-time thresholddoes increase the sample size, but also increases thelikelihood that electrically active storms are assignedzero flash rates simply because they are not viewed forenough time.

Lightning in the inner regions of tropical cyclonesoccurs most often in the weak tropical storms [34–45kt (17–23 m s21) maximum sustained wind] and stronghurricanes or typhoons [.95 kt (49 m s21) maximum

sustained wind] in this sample (Fig. 9a). Fourty-fourpercent and 37% of the OTD observations of these re-spective categories of storms reveal lightning in the in-ner regions. The percentages decrease to 15% for trop-ical depressions [,34 kt (17 m s21) maximum sustainedwind], 32% for strong tropical storms [45–63 kt (23–32 m s21) maximum sustained wind], and 17% for weakhurricanes or typhoons [64–94 kt (33–48 m s21) max-imum sustained wind].

Possible relationships between lightning and tropicalcyclone intensity change appear somewhat ambiguousin this dataset. Because a large majority of tropical de-pression observations contain no lightning and nochange in intensity, only cyclones of at least tropicalstorm magnitude [34 kt (17 m s21)] are considered forintensity change. An examination of the short-term in-tensity trend at the time of an OTD observation revealsthat lightning occurs more often in storms undergoingsmall changes in intensity [#|5| kt (2.5 m s21)] duringthe 6-h period spanning the observation than in thoseundergoing more dramatic intensity changes, whetherpositive or negative. On the longer term, lightning ap-pears to be slightly more likely in storms that intensifymoderately [5–15 kt (2.5–7.5 m s21)] during the 24 hfollowing an observation than in those that weakenmoderately during this time period (Fig. 9b). Fifty-sevenpercent of the observations in the former category and44% of those in the latter include lightning. Thirty per-cent of the observations of storms changing intensityby 5 kt (2.5 m s21) or less include lightning. Only 25%of the observations of the most rapidly intensifying sys-tems and 8% of the observations of the most rapidlyweakening systems include lightning. These percentagesvary depending on the intensity change intervals usedin a computation and on the period of time for whichan intensity change is considered. The most consistentresult is that lightning is found least often in the rapidlyweakening systems and most often in systems that un-dergo small to moderate changes of intensity, whetherstrengthening or weakening. These results focus onlyon whether or not lightning is found by OTD, withoutregard to the number of flashes found. For the vast ma-jority of the observations, five or fewer flashes are noted.


FIG. 9. Fraction of OTD samples revealing lightning in the inner 18 radius of tropical cyclones, as a function of (a) tropical cycloneintensity (kt) (1 kt ø 0.5 m s21) at the time of the overpass or (b) 24-h intensity change (kt). See section 2 for details.

FIG. 10. Hurricane Luis best track, 1200 UTC 27 August–0000UTC 11 September 1995. Positions are plotted every 6 h. Intensitiesare plotted in kt (1 kt ø 0.5 m s21).

No clear tendency for either strengthening or weakeningappears in the cases of higher flash counts.

c. Case studies

The statistics just presented were designed to quantifyaspects of the ice-scattering signature or electrical ac-tivity in order to study a large number of observations.By examining individual cases, we can go beyond theobjective formulas used to compute ice-scattering sig-nature parameters; the details of storm structure thatresult in a given mean PCT, for example, can be ana-lyzed. The cases were selected to display a variety ofstorm life cycles, to demonstrate how the computed pa-rameters relate to cyclone intensity, and to demonstrate

how these parameters sometimes fail to represent cy-clone intensity.

1) ATL HURRICANE LUIS (1995)

Hurricane Luis evolved from a very compact storm(in terms of the spatial extent of its ice-scattering sig-nature) in the eastern Atlantic into a very broad systemas it crossed the northeastern Leeward Islands. Aftermaintaining maximum sustained winds of 110–120 kt(57–62 m s21) for seven days, Luis turned northwardand weakened (Fig. 10). Its long lifetime as a majorhurricane provides ample opportunity for study.

During its formative stages, Luis experienced veryintense convection. As a tropical depression on 28 Au-gust and a tropical storm (Fig. 11a) the following day,several areas with PCT below 175 K were found. Light-ning was also observed during this time by OTD (Fig.12a). The intense convection fluctuated, as evidencedby the lack of lightning near Luis’s center and the re-duced area of extreme ice scattering in subsequent ob-servations. Luis remained only a minimal tropical stormuntil 30 August, when a 60-h intensification from 40 kt(21 m s21) to 115 kt (59 m s21) began. The developmentof a closed eyewall drastically reduced the mean PCTfor the inner regions of Luis. Whereas the 08–18 arealmean PCT had generally been around 270 K during thetropical storm stage, it was near 250 K after the eyewallformed.

Figures 11b,c demonstrate fluctuations in PCT on arather short temporal scale. At 0841 UTC 2 September(Fig. 11b), the F13 satellite found PCT below 175 Kin the rainband about 1.58 east of the center of Luis.Much of this rainband had PCT below 225 K with arelatively large area below 200 K. When the F11 sat-ellite observed Luis less than 90 min later (Fig. 11c),only a small portion of the rainband had PCT below225 K, although the minimum PCT was again below175 K. It should be noted that data from such ‘‘nearcoincident’’ overpasses by the F11 and F13 satellites


FIG. 11. As in Fig. 1 except for Hurricane Luis at (a) 0748 UTC 29 August 1995, (b) 0841 UTC 2 September 1995, (c) 1004 UTC 2September 1995, and (d) 0952 UTC 3 September 1995.

displayed no systematic tendency for one satellite toobserve lower PCT than the other. At 1244 UTC, OTDviewed Luis for 2.5–4 min, observing three lightningflashes in the rainband but none in the eyewall.

The 08–18 areal-mean PCT continued to lower as Luismaintained 115–120 kt (59–62 m s21) intensity. At 0952UTC 3 September (Fig. 11d) Luis produced the lowestsuch mean PCT (232 K) of any storm in this study. Alarge outer eyewall nearly enclosed the eye with PCT

below 225 K, whereas remnants of an inner eyewallcontributed further to the low mean PCT. Luis had justreached peak intensity [120 kt (62 m s21)] and wouldremain at this intensity for two days. Subsequent SSM/I observations featured less intense eyewall ice-scatter-ing signatures, but the large area with low (if not ex-treme) PCT continued to produce quite low areal-meanPCT values (mostly below 250 K). Multiple eyewallswere often evident, which may have been a limiting


FIG. 12. Luis OTD lightning observations at (a) 2348 UTC 28 August 1995 and (b) 1303 UTC 6 September 1995. The amount of time(seconds) in the OTD’s field of view is contoured. Contour interval is 30 s. The innermost 48 radius is plotted, with range rings at 0.58intervals. Best-track wind speed (kt) (1 kt ø 0.5 m s21) interpolated to time of overpass and 6-h intervals thereafter is given in the header.

factor on Luis’s intensity. Previous studies (e.g., Wil-loughby 1990) have established that an outer eyewalloften causes the dissipation of an inner eyewall and windmaximum, thus modifying the storm’s intensity.

Most of the OTD observations of hurricanes in thisstudy suggest that lightning is rather limited in thesestorms, at least for the short sampling duration of theOTD. That may be the case, but Fig. 12b demonstratesthat highly electrically active rainbands do exist and canbe detected by the OTD. Three consecutive OTD over-passes on 6 September reveal such rainbands in Luis’ssoutheast quadrant. Many flashes were detected even inregions that spent less than 60 s in the field of view.

To describe Hurricane Luis in terms of the parametersthat were analyzed statistically, time series of some ofthese parameters are presented in Fig. 13. For the largesample of storms, 08–18 radius mean PCT produces thehighest correlations with intensity. It roughly corre-sponds to the total latent heat release in that region,under the assumption of a linear relationship betweenbrightness temperature and rain rate (e.g., Spencer et al.1989; Adler et al. 1994). This parameter, along withLuis’s intensity, is plotted in Fig. 13a. At the other ex-treme is the minimum PCT, which represents the inten-sity of the most highly convective pixel. Lightning isalso related to the most highly convective areas. Theseparameters are plotted in Fig. 13b. Unlike in the statis-tical analysis, any F11 or F13 SSM/I overpass availableis plotted here. That is, if both satellites sample Luis atsunrise on the same day, both observations are plotted.Lightning data are plotted only for overpasses in whicheither lightning is observed in the 08–18 radius region

or that region is sampled for at least 180 s withoutlightning.

As indicated by minimum PCT, Luis’s most intenseconvection occurred early during the tropical stormstage, then the convection weakened as the larger-scalestorm developed (Fig. 13b). The lone OTD observationof inner core lightning also occurred early in Luis’s life(Fig. 13b). On the other hand, the large, radiometricallycold eyewall resulted in very low areal mean PCT forthe inner regions of Hurricane Luis as the storm ap-proached and then maintained 110–120 kt (62 m s21)intensity for several days. A wide range of areal-meanPCT were encountered while Luis basically maintainedpeak intensity (Fig. 13a), with no apparent response tochanges in the already low mean PCT. The hurricane’sintensity appears to have been modulated by eyewallreplacement cycles during this time.

2) NEPAC HURRICANE BARBARA (1995)

Hurricane Barbara produced at least 100 kt (51 ms21) maximum sustained winds for five days in July1995 in the eastern North Pacific before dissipating inthe Central Pacific (Fig. 14). Barbara quickly developedfrom a weak tropical storm to a hurricane, with a rapidintensification episode [at least 42 hPa of deepening in24 h, as defined by Holliday and Thompson (1979)]following. The hurricane weakened slightly from 115kt (59 m s21) to 100 kt (51 m s21), then reached a peakintensity of 120 kt (62 m s21). Sustained weakeningfollowed as the storm moved over cooler waters. BothSSM/I and OTD data are available at the onset of rapid


FIG. 13. (a) Time series of 08–18 radius areal mean PCT (*) andbest-track intensity (kt) (1 kt ø 0.5 m s21) (solid line) of HurricaneLuis. Here ‘‘L’’ along the intensity curve indicates that the storm waswithin 100 km of land. Horizontal line marks 64 kt sustained winds.(b) Time series of 08–18 radius minimum PCT and lightning in Hur-ricane Luis. Lightning is plotted as log (flashes per minute), withzero lightning along the dotted line.

FIG. 14. Hurricane Barbara best track, 1800 UTC 7 July–0000 UTC18 July 1995. Positions are plotted every 6 h. Intensities are plottedin kt (1 kt ø 0.5 m s21).

intensification, and also during the period in which Hur-ricane Barbara fluctuated between 100 and 120 kt (51–62 m s21).

PCT as low as 137 K were observed near the centerof Barbara early in its tropical storm stage. Four light-ning flashes were observed, also near the center, in 200s. In contrast, PCT remained generally in the 225–250K range as precipitation wrapped around Barbara’s cen-ter at the onset of rapid intensification (Fig. 15a). Ice-scattering signatures were less intense but much moreorganized at this point compared to the tropical stormstage. An OTD overpass near the onset of rapid inten-sification revealed no lightning associated with Hurri-cane Barbara.

After an initial peak intensity of 115 kt (59 m s21)on 10 July, Hurricane Barbara weakened to 100 kt (51m s21) as the eye disappeared from IR imagery. Figure15b demonstrates a distinct eye in the SSM/I data at0216 UTC 12 July, which is obscured in the IR data.The eye reappeared in IR imagery later on 12 July, witha subsequent reintensification of the storm to 120 kt (62m s21). During this reintensification, Barbara’s spatialextent decreased dramatically (Fig. 15c), with the eye-

wall nearly contained by the 0.58 radius circle and theentire region with an ice-scattering signature nearly con-tained by the 18 radius circle at 0204 UTC 13 July.Strong eyewall convection persisted, however, with PCTagain reaching below 200 K at 1440 UTC. As Barbaraeventually weakened, the tiny area of reduced PCT (Fig.15d) bore little resemblance to what one normally thinksof as a hurricane’s precipitation field, despite best-trackwinds near 100 kt (51 m s21).

The 08–18 radius mean PCT was identified earlier asbeing highly correlated with future intensity. This pa-rameter remained somewhat low (250–260 K) for mostof Hurricane Barbara’s life, until rising above 270 K asthe storm weakened (Fig. 16a). Although no correlationis seen with intensity change, a low mean PCT suggestedthat the minimal hurricane on 9 July had the potentialto become a major hurricane, as it did the followingday. The continued low mean PCT (resulting from anactive, well-developed eyewall) after Barbara’s initialweakening on 11 July confirmed that the storm couldnot yet be discounted. The OTD sampled Barbara duringmany stages of the storm’s life cycle, including the rapidintensification period, but detected lightning only at thebeginning of the tropical storm stage (Fig. 16b).

3) NWPAC SUPER TYPHOON RYAN (1995)

While meandering in the South China Sea, Ryan grad-ually strengthened to typhoon intensity (Fig. 17). In-tensification continued as Ryan accelerated northeast-ward, passing near Taiwan with 130 kt (67 m s21) in-tensity. Weakening followed as the storm approachedand made landfall on Japan. Ryan was well sampled by


FIG. 15. As in Fig. 1 except for Hurricane Barbara at (a) 1352 UTC 9 July 1995, (b) 0216 UTC 12 July 1995, (c) 0204 UTC 13 July1995, and (d) 0236 UTC 15 July 1995.

SSM/I throughout the intensification from tropical stormto super typhoon. Nearly coincident overpasses of theSSM/I and OTD on 21 September provide a direct com-parison of 85-GHz PCT and short view time lightningobservations for a super typhoon reaching its peak in-tensity.

The intensity of convection in the inner regions ofRyan fluctuated but the degree of spatial organizationin the PCT field increased as Ryan developed throughits tropical storm and typhoon stages. Extreme ice scat-

tering (with PCT near 140 K) was encountered in theeyewall at times, such as 2213 UTC 19 September (Fig.18a) and 0915 UTC 21 September (Fig. 18c). Althoughthe total area experiencing rather low PCT (below 225K) remained impressive, the minimum eyewall PCT at2201 UTC 20 September (Fig. 18b) and 2148 UTC 21September (Fig. 18d) only reached near 200 K.

The fluctuations in intensity of eyewall convectionon 21 September are also suggested by OTD. Five light-ning flashes were detected at 0908 UTC (Fig. 19a) when


FIG. 16. As in Fig. 13 except for Hurricane Barbara.

FIG. 17. Typhoon Ryan best track, 1200 UTC 14 September–1800UTC 24 September 1995. Positions are plotted every 6 h. Intensitiesare plotted in kt (1 kt ø 0.5 m s21).

the eyewall was in view for approximately 40–100 s.Geolocation appears to be questionable for these flashes(the platform carrying OTD may have been temporarilyunstable), but they are clearly associated with the intenseeyewall convection depicted by the SSM/I. The eyewallwas in view for a slightly longer time (60–120 s) at2148 UTC (Fig. 19b), but this time no flashes weredetected in the eyewall region. Minimum PCT at thistime was near 200 K. That no flashes were detected byOTD in less than 2 min certainly does not mean thatthe eyewall had become electrically inactive, but at leastsuggests a reduced flash rate compared with earlier inthe day. This is consistent with the much higher mini-mum PCT found at 2148 UTC compared with 0915UTC. Even though view time decreased to the east ofthe eyewall, 19 flashes associated with the rainband 28east of the center were detected by OTD in a 70-s span.PCTs as low as 128 K were detected with this band.The correspondence seen here between PCT and elec-trical activity is not surprising, given the importance oflarge ice to the scattering of 85-GHz radiation and tothe charge separation process. In other cases, however,the correspondence has not always been as straightfor-ward.

Another OTD overpass at 0822 UTC 22 Septemberindicated lightning in both the eyewall and the rainbandlocated east of Ryan’s center. View times were againless than 120 s and decreased to the east of the center,

but the rainband produced more lightning (11 flashes)than the eyewall (2 flashes). Within hours of this OTDobservation, Ryan began to weaken. SSM/I observa-tions later that day continued to identify intense con-vective cells in the rainband, but also revealed a greatlydiminished eyewall with the weakening storm.

Of the cases presented individually, Super TyphoonRyan provides the clearest example of a relationshipbetween 08 and 18 areal mean PCT and future tropicalcyclone intensity (Fig. 20a). The best-track intensityincreases as the mean PCT decreases, then peaks afterthe mean PCT shows a slight increase. Further increasesin mean PCT accompany the decrease in Ryan’s inten-sity as the storm approaches Japan. The 243 K meanPCT at 2213 UTC 19 September is an exception to thetrend of best-track intensity smoothly following themean PCT. The intense convection at this time was alsodepicted in IR imagery, and leads to a question con-cerning the accuracy of the best-track intensities, whichwill be addressed in section 4. Inner core lightning wasobserved twice while Ryan was at or near its peak in-tensity, including one OTD overpass just before weak-ening began (Fig. 20b).

4. Discussion

High correlations (;20.7) are found between innercore areal-mean PCT (and 250 area) and future hurri-cane/typhoon intensity. OTD lightning observations re-veal an increased likelihood of lightning in weak trop-ical storms and strong hurricanes, but no clear trendsare found connecting lightning to intensity change. Giv-en the inherent limitations of OTD in such a study, the


FIG. 18. As in Fig. 1 except for Typhoon Ryan at (a) 2213 UTC 19 September 1995, (b) 2201 UTC 20 September 1995, (c) 0915 UTC21 September 1995, and (d) 2148 UTC 21 September 1995.

lightning results are inconclusive. Individual case stud-ies reveal both advantages and drawbacks in using SSM/I imagery to analyze tropical cyclones, while also pro-viding new information with the lightning observations.

The first aspect of these results to address is the highercorrelations between inner core ice-scattering signaturesand future intensity than between ice-scattering signa-tures and either present intensity or intensity change.Why are the correlations highest for future intensity? Arespectable correlation with present intensity was ex-

pected (and found); after all, satellite-derived intensityestimates of tropical cyclones basically attempt to mea-sure the organization and vigor of the cyclone’s con-vection (Dvorak 1984). If the latent heating implied bythe ice-scattering signature provides more energy thanwould be necessary to sustain the system, an intensi-fication of the cyclone’s circulations should result. Thesestatements mean that the ice-scattering signature is re-lated to both current intensity and intensity change, withthe combined effect being an increased correlation with


FIG. 19. As in Fig. 12 except for Typhoon Ryan at (a) 0908 UTC 21 September 1995 and (b) 2148 UTC 21 September 1995.

FIG. 20. As in Fig. 13 except for Typhoon Ryan.

future intensity. That is, there may be a temporal lagbetween the production of the ice-scattering signatureand the response of the primary circulation.

The next results to address are the high correlationsbetween indicators of moderate rainfall (250 area and

average PCT, which are highly correlated with each oth-er) and tropical cyclone intensity, compared to thosebetween indicators of intense convection (minimumPCT and 200 area) and tropical cyclone intensity. Glassand Felde (1992) reported that the area having hori-zontally polarized brightness temperatures below 220–230 K produced high correlations with intensity. Thisrange of brightness temperatures typically correspondsto 230–250 K PCT. Considering the various reports ofconvective bursts associated with deepening tropical cy-clones (e.g., Lyons and Keen 1994; Molinari et al. 1994;Black et al. 1994), why do the indicators of intenseconvection produce such low correlations? The param-eters that produce the highest correlations with intensityrespond primarily to the mesoscale nature of the pre-cipitation, as opposed to the convective scale. The min-imum PCT parameter depends on a single 13 km 3 15km pixel. Although intense convection is sometimesassociated with significant intensification of the larger-scale vortex, such a small region of intense convectionmay have little impact on the cyclone’s intensity whenother organized precipitation (e.g., a well-formed eye-wall) is lacking.

In contrast to each of the results discussed thus faris the relatively high correlation mentioned for NEPAC18–1.58 radius minimum PCT and tropical cyclone in-tensity change. The ATL and NWPAC produce low cor-relations with intensity change for this parameter, andthe 18–1.58 region generally produces lower correlationswith intensity than do regions inward of 18 radius. Whatis peculiar about the NEPAC sample, and why does the18–1.58 region stand out? The NEPAC sample includesa higher percentage of decaying storms than the ATLand NWPAC samples, due to the equatorward extent of


FIG. 21. Time series of Digital Dvorak T numbers (solid circles)and best-track intensity (open circles) for Typhoon Ryan. Here Tnumber 5 corresponds to 90 kt (46 m s21) maximum sustained winds,7 corresponds to 140 kt (67 m s21) (from JTWC 1995).

cool waters in this basin. The NEPAC storms that weak-ened rapidly had very high minimum PCT beyond 18radius, with the spatial extent of the ice-scattering sig-nature decreasing dramatically with time. This helpedproduce relatively high correlations for all of the PCTparameters with intensity change (Fig. 5c) and similarlywith future intensity (Fig. 4c).

The lowest correlations between ice-scattering sig-nature and intensity involve NWPAC typhoons. Thesecorrelations are much lower when the 1995 sample isconsidered separately (Cecil 1997), but the combined1995–96 sample also produces lower correlations forthe NWPAC than for the other basins. Much of thescatter involves very intense or rapidly intensifyingtropical cyclones, which are more often observed in theNWPAC than in the ATL or NEPAC. In the more intensesystems, processes such as eyewall cycles become morecommon (Willoughby 1990); these cycles can signifi-cantly affect a cyclone’s intensity. Even in the absenceof an eyewall cycle, the eyewall location and degree ofsymmetry cannot be determined from PCT parameterscomputed for a 18 radius circle, but such details areclosely linked to the cyclone’s intensity. This result issimilar to the finding by Marks (1985) that the intensityof Hurricane Allen (1980) was related to its eyewallevolution, but the rain rate (thus latent heat release)inside a 18 radius circle showed little response to chang-es in the eyewall. As demonstrated in numerical studies(Shapiro and Willoughby 1982; Schubert and Hack1982), not only the magnitude of the heating but alsothe location of the heating must be taken into consid-eration. A wide variety of eyewall structures may yieldsimilar values for the PCT parameters used here. Thesesimple parameters ultimately cannot make up for patternrecognition.

Another potential cause for lower NWPAC correla-tions is possible inaccuracy in the best-track intensitydata. In the absence of aircraft reconnaissance, tropicalcyclone intensities in the Pacific Ocean are primarilyderived from satellite data, using the Dvorak (1984)technique. These intensity estimates sometimes includeadjustments based on the expected rates of intensifica-tion for tropical cyclones. Super Typhoon Ryan providesa disturbing example of the constraints these expectedintensification rates can impose on the best-track inten-sity. Without these adjustments, objective application ofDvorak’s technique yields intensity estimates of 140 kt(72 m s21) late on 19 September and early on 20 Sep-tember (Lander and Angove 1998) (Fig. 21). The finalbest-track intensity during this time is 80–90 kt (41–46m s21). Ryan is a rather exceptional case, because theobjective application of Dvorak’s technique more oftenfalls roughly in line with the final best-track data. How-ever, as pointed out by Lander in JTWC (1995), theserapid changes in intensity, if they are real, would se-riously compromise the usefulness of the best-track datain studies involving intensity. With more rapid inten-

sifiers in the NWPAC than the other basins, the potentialproblem is greater in the NWPAC.

The OTD lightning data used in this study reveal thatweak tropical storms and strong hurricanes or typhoonsare most likely to feature inner core lightning. The great-est likelihood of lightning occurs with weak tropicalstorms, which is consistent with the frequent observa-tions of extremely low minimum PCT during this stagein the case studies presented. As in the published lit-erature, examples of inner core lightning either preced-ing or coinciding with cyclone intensification are ob-served, but inner core lightning is also observed pre-ceding or coinciding with cyclone weakening and ismost often observed during periods in which the cy-clone’s intensity remains fairly steady. Although thisdoes not necessarily contradict reports that link innercore lightning to intensification, it does indicate that anysuch relationship is far from being clear cut. Usuallyonly a few flashes are observed in a particular area, butexamples of high flash rates occur almost exclusivelywith rainbands, as opposed to eyewalls. This is some-what paradoxical, although not a new result (Samsuryand Orville 1994; Molinari et al. 1994, 1999), giventhat neither Jorgensen et al. (1985) nor Black et al.(1996) find evidence for rainbands having stronger up-drafts than eyewalls. The OTD was not designed forindividual case study, and given its limited view times,any attempts at quantification of the lightning obser-vations should be deemed inconclusive.

That the observations of lightning show little (if any)relationship with tropical cyclone intensity change isconsistent with the result that SSM/I indicators of in-tense convection are only mildly correlated with hur-ricane/typhoon intensity and intensity change. A singleintense convective cell may produce a very strong ice-scattering signature and a great deal of lightning. Thespatial coverage and organization, more so than the in-


tensity, of the precipitation shows the greater relation-ship with overall hurricane/typhoon intensity.

Both of the key components of this study (SSM/I andOTD) involve observations from satellites that encoun-ter the targeted storms at most twice per day. Theseobservations concern convective and mesoscale entitiesoperating on much shorter temporal scales. As the num-ber of observations increases (i.e., observations frommore storms are combined), some trends become clearer.When individual observations are examined, the pos-sibility of fluctuations in convective intensity must beconsidered. Such fluctations were demonstrated in theSSM/I observations of Hurricane Luis; short timescalefluctuations in lightning flash rates have also been re-ported by Molinari et al. (1999). This makes direct com-parison of individual observations potentially extremelymisleading, as one observation may correspond to ashort timescale maximum in convective vigor, whereasthe other corresponds to a short timescale minimum inconvective vigor.

A note should be made concerning the representa-tiveness of the sample in this study. All the storms con-sidered occurred during 1995 and 1996, which wereunusual years for tropical cyclones. In fact, 1995 and1996 were among the most active ATL seasons on re-cord. On the other hand, NEPAC experienced a far be-low normal hurricane season each year. The peculiaritiesof the 1995 and 1996 seasons may have affected theresults presented here in unseen ways.

Many factors seem to influence tropical cyclone in-tensity [e.g., SST (Merrill 1988), vertical wind shear(DeMaria et al. 1993), relative eddy angular momentumflux convergence (DeMaria et al. 1993), potential vor-ticity advection (Molinari et al. 1995), and internal dy-namics (Shapiro and Willoughby 1982)]. Through itsrelationship with rainfall, the ice-scattering signaturecan be expected to respond to each of the factors men-tioned above. Considering the complexities of tropicalcyclones, the correlations between ice-scattering sig-nature and hurricane/typhoon intensity are remarkablyhigh. Particularly noteworthy is that such simple mea-sures as areal-mean PCT contain so much information.

5. Conclusions

This paper uses 85-GHz PCT as a proxy for updraftstrength, convective vigor, and precipitation intensity.With this proxy, relationships are sought between PCTparameters and tropical cyclone intensity, future inten-sity, and intensity change. Of the relationships sought,the strongest involve the spatial coverage of at leastmoderate inner core rainfall. Correlations are of thesame sense for both present intensity (at the time of thesatellite overpass) and subsequent intensity change, butthese effects combine to produce even higher correla-tions for future intensity—that is, a temporal lag existsbetween the production of an ice-scattering signatureand the response of the hurricane/typhoon intensity.

Correlations between mean PCT and 24-h intensityare highest for NEPAC (20.84) and ATL (20.73)storms, with lower correlations (20.62) for NWPACstorms. The greatest scatter in plots of future intensityversus mean PCT tend to be associated with very intenseor rapidly intensifying systems. Such systems occurmore frequently in the NWPAC than in the other basins.It is proposed that such large scatter results from theintense/rapidly intensifying typhoons because details ofthe eyewall(s) are more important in these storms, andare poorly resolved by the simple PCT parameters usedhere. Without in situ observations in these typhoons,the input intensity estimates are also questioned. Thequestions particularly address the legitimacy of intensityestimates for rapidly intensifying systems, again leavingthe NWPAC more susceptible to error.

OTD observations of inner core lightning indicate agreater likelihood of lightning in weak tropical stormsand in strong hurricanes or typhoons than in other trop-ical cyclones. These lightning observations do not ap-pear to be a useful indicator of intensity change. Light-ning is observed in a spectrum of tropical cyclone sit-uations; for example, tropical storms showing no signsof intensification, typhoons undergoing rapid intensifi-cation, and super typhoons that soon decrease in inten-sity. The most common scenario for inner core lightningin this sample involves storms whose intensity changesslowly, if at all. Storms are sampled by OTD for lessthan 5 min at a time, making large flash counts rare.These are encountered, however, most often in rainbandsinstead of eyewalls. The lack of any apparent relation-ship between lightning occurrence and subsequent cy-clone intensity change is consistent with the relativelylow correlations between tropical cyclone intensitychange and PCT-based indicators of intense convection.The lack of a relationship may also be related to thelimitations of OTD in such a study.

These results raise several issues for further study.Would other oceanic basins mimic the high correlationsfound in the NEPAC, the lower correlations found inthe NWPAC, or would they fall somewhere in between,as does the ATL? The highest correlations basically in-volve 08–18 radius latent heat release. Would other mi-crowave channels that are more intimately tied to latentheat release, but have lower resolution than the 85 GHz13 km 3 15 km footprint, likewise produce such highcorrelations? The lack of a relationship between indi-cators of intense convection (such as lightning) and trop-ical cyclone deepening seems to be in conflict with pre-vious reports. Would a more detailed dataset (such aslightning observations from a geostationary satellite)resolve this conflict, and if so, with what result? Finally,could a tropical cyclone intensity forecasting tool bedeveloped based on improvements to the parametersstudied here? At what point would increasing the com-plexity of the parameters cease to improve the resultingforecasts?


Acknowledgments. SSM/I, OTD, and IR data wereprovided by the Global Hydrology Resource Center atthe Global Hydrology and Climate Center, Huntsville,AL. Thanks to the staff there and the Lightning ImagingSensor Scientific Computing Facility at GHCC, whomade the OTD data usable. Thanks also to Chip Guard,Frank Marks, John Molinari, G. V. Rao, Ed Rappaport,Chris Samsury, and two anonymous reviewers who haveprovided useful comments on this manuscript. This re-search was supported by the National Aeronautics andSpace Administration (NAG8-912, NAG5-4699, andNAG5-4796).

REFERENCES

Adler, R. F., G. J. Huffman, and P. R. Keehn, 1994: Global tropicalrain estimates from microwave-adjusted geosynchronous IRdata. Remote Sens. Rev., 11, 125–152.

Black, P. G., R. A. Black, J. Hallett, and W. A. Lyons, 1986: Electricalactivity of the hurricane. Preprints, 23d Conf. on Radar Mete-orology, Snowmass, CO, Amer. Meteor. Soc., J277–J280.

Black, M. L., R. W. Burpee, and F. D. Marks Jr., 1996: Vertical motioncharacteristics of tropical cyclones determined with airborneDoppler radial velocities. J. Atmos. Sci., 53, 1887–1909.

Black, R. A., and J. Hallett, 1986: Observations of the distributionof ice in hurricanes. J. Atmos. Sci., 43, 802–822., H. B. Bluestein, and M. L. Black, 1994: Unusually strongvertical motions in a Caribbean hurricane. Mon. Wea. Rev., 122,2722–2739.

Cecil, D. J., 1997: Relationships between tropical cyclone intensityand satellite based inicators of inner core convection: 85 GHzice scattering signature and lightning. M.S. thesis, Dept. of Me-teorology, Texas A&M University, 125 pp. [Available fromEvans Library, Texas A&M University, College Station, TX77843.]

Christian, H. J., R. J. Blakeslee, and S. J. Goodman, 1989: The de-tection of lightning from geostationary orbit. J. Geophys. Res.,94, 13 329–13 337.

DeMaria, M., J. Baik, and J. Kaplan, 1993: Upper level eddy angularmomentum flux and tropical cyclone intensity change. J. Atmos.Sci., 50, 1133–1147.

Dvorak, V. F., 1984: Tropical cyclone intensity analysis using satellitedata. NOAA Tech. Rep. NESDIS 11, 47 pp.

Glass, M., and G. W. Felde, 1992: Intensity estimation of tropicalcyclones using SSM/I brightness temperatures. Preprints, SixthConf. on Satellite Meteorology and Oceanography, Atlanta, GA,Amer. Meteor. Soc., J8–J10.

Goodman, S. J., and H. J. Christian, 1993: Global observations oflightning. Atlas of Satellite Observations Related to GlobalChange, R. J. Gurney, J. L. Foster, and C. L. Parkinson, Eds.,Cambridge University Press, 191–219.

, and Coauthors, 1996: The optical transient detector: Firstresults. Preprints, Eighth Conf. on Satellite Meteorology andOceanography, Atlanta, GA, Amer. Meteor. Soc., 583–587.

Henning, R. G., 1997: Electrically active mesoscale convective com-plexes and their link to explosive tropical cyclogenesis. Pre-prints, 22d Conf. on Hurricanes and Tropical Meteorology, Ft.Collins, CO, Amer. Meteor. Soc., 129–130.

Holliday, C. R., and A. H. Thompson, 1979: Climatological char-acteristics of rapidly intensifying typhoons. Mon. Wea. Rev., 107,1022–1034.

Hollinger, J. T., 1991: DMSP Special Sensor Microwave/imager cal-ibration/validation final report, vol. II. Naval Research Labo-ratory, 12–25. [Available from Naval Research Laboratory,Washington, DC 20375.]

Illingworth, A. J., 1985: Charge separation in thunderstorms: Smallscale processes. J. Geophys. Res., 90, 6026–6032.

Jorgensen, D. P., E. J. Zipser, and M. A. LeMone, 1985: Verticalmotions in intense hurricanes. J. Atmos. Sci., 42, 839–856.

JTWC, 1995: Annual tropical cyclone report. 289 pp. [Available fromU.S. Naval Pacific Meteorology and Oceanography CommandCenter West, Guam, PSC 489 Box 12, FPO AP 96536-0051.]

Lander, M. A., and M. D. Angove, 1998: Eastern Hemisphere tropicalcyclones of 1995. Mon. Wea. Rev., 126, 257–280.

Lascody, R. A., 1992: A different look at Hurricane Andrew—Light-ning around the eyewall. Natl. Wea. Dig., 17, 39–40.

Lyons, W. A., and C. S. Keen, 1994: Observations of lightning inconvective supercells within tropical storms and hurricanes.Mon. Wea. Rev., 122, 1897–1916.

Marks, F. D., Jr., 1985: Evolution of the structure of precipitation inHurricane Allen (1980). Mon. Wea. Rev., 113, 909–930., and R. A. Houze Jr., 1987: Inner core structure of HurricaneAlicia from airborne Doppler radar observations. J. Atmos. Sci.,44, 1296–1317.

McGaughey, G., E. J. Zipser, R. W. Spencer, and R. E. Hood, 1996:High-resolution passive microwave observations of convectivesystems over the Tropical Pacific Ocean. J. Appl. Meteor., 35,1921–1947.

Merrill, R. T., 1988: Environmental influences on hurricane intensi-fication. J. Atmos. Sci., 45, 1678–1687.

Mohr, K. I., and E. J. Zipser, 1996a: Defining mesoscale convectivesystems by their 85-GHz ice scattering signature. Bull. Amer.Meteor. Soc., 77, 1179–1189., and , 1996b: Mesoscale convective systems defined bytheir 85-GHz ice scattering signature: Size and intensity com-parison over tropical oceans and continents. Mon. Wea. Rev.,124, 2417–2437., E. R. Toracinta, E. J. Zipser, and R. E. Orville, 1996: A com-parison of WSR-88D reflectivities, SSM/I brightness tempera-tures, and lightning for mesoscale convective systems in Texas.Part II: SSM/I brightness temperatures and lightning. J. Appl.Meteor., 35, 919–931.

Molinari, J., P. K. Moore, V. P. Idone, R. W. Henderson, and A. B.Saljoughy, 1994: Cloud-to-ground lightning in Hurricane An-drew. J. Geophys. Res., 99, 16 665–16 676., S. Skubis, and D. Vollaro, 1995: External influences on hur-ricane intensity. Part III: Potential vorticity structure. J. Atmos.Sci., 52, 3593–3606., P. Moore, and V. Idone, 1999: Convective structure of hurri-canes as revealed by lightning locations. Mon. Wea. Rev., inpress.

Orville, R. E., and R. W. Henderson, 1986: Global distribution ofmidnight lightning: September 1977 to August 1978. Mon. Wea.Rev., 114, 2640–2653.

Rao, G. V., and P. D. MacArthur, 1994: The SSM/I estimated rainfallamounts of tropical cyclones and their potential in predictingthe cyclone intensity changes. Mon. Wea. Rev., 122, 1568–1574., and J. H. McCoy, 1997: SSM/I measured microwave brightnesstemperature (TB’s), anomalies of TB’s, and their relationship totyphoon intensification. Nat. Hazards, 15, 1–19.

Restivo, M. E., 1995: The convective structures associated withcloud-to-ground lightning in TOGA COARE mesoscale con-vective systems. M.S. thesis, Dept. of Meteorology, Texas A&MUniversity, 98 pp. [Available from Evans Library, Texas A&MUniversity, College Station, TX 77843.]

Rodgers, E. B., and H. F. Pierce, 1995: A satellite observational studyof precipitation characteristics in western North Pacific tropicalcyclones. J. Appl. Meteor., 34, 2587–2599., S. W. Chang, and H. F. Pierce, 1994: A satellite observationaland numerical study of precipitation characteristics in westernNorth Atlantic tropical cyclones. J. Appl. Meteor., 33, 129–139.

Samsury, C. E., and R. E. Orville, 1994: Cloud-to-ground lightningin tropical cyclones: A study of Hurricanes Hugo (1989) andJerry (1989). Mon. Wea. Rev., 122, 1887–1896., M. L. Black, P. P. Dodge, and R. E. Orville, 1997: Utilizationof airborne and NEXRAD data in the analysis of cloud-to-groundlightning in 1995 and 1996 tropical cyclones. Preprints, 22d


Conf. on Hurricanes and Tropical Meteorology, Ft. Collins, CO,Amer. Meteor. Soc., 125–126.

Schubert, W. H., and J. J. Hack, 1982: Inertial stability and tropicalcyclone development. J. Atmos. Sci., 39, 1687–1697.

Shapiro, L. J., and H. E. Willoughby, 1982: The response of balancedhurricanes to local sources of heat and momentum. J. Atmos.Sci., 39, 378–394.

Spencer, R. W., H. M. Goodman, and R. E. Hood, 1989: Precipitationretrieval over land and ocean with the SSM/I: Identification and

characteristics of the scattering signal. J. Atmos. Oceanic Tech-nol., 6, 254–273.

Willoughby, H. E., 1990: Temporal changes of the primary circulationin tropical cyclones. J. Atmos. Sci., 47, 242–264., D. P. Jorgensen, R. A. Black, and S. L. Rosenthal, 1985: ProjectStormfury: A scientific chronicle 1962–1983. Bull. Amer. Me-teor. Soc., 66, 505–514.

Zipser, E. J., 1994: Deep cumulonimbus cloud systems in the Tropicswith and without lightning. Mon. Wea. Rev., 122, 1837–1851.

relationships between tropical cyclone intensity and satellite...

Documents