Cloud Clusters and Tropical Cyclogenesis: Developing and … · 2014. 11. 7. · Cloud Clusters and Tropical Cyclogenesis: Developing and Nondeveloping Systems and Their Large-Scale

Cloud Clusters and Tropical Cyclogenesis: Developing and Nondeveloping Systemsand Their Large-Scale Environment

BRANDON W. KERNS AND SHUYI S. CHEN

Rosenstiel School of Marine and Atmospheric Science, University of Miami, Miami, Florida

(Manuscript received 19 September 2011, in final form 13 July 2012)

ABSTRACT

Tropical cyclone (TC) genesis occurs only when there is persistent, organized convection. The question of

why some cloud clusters develop into a TC and others do not remains unresolved. This question cannot be

addressed adequately without studying nondeveloping systems in a consistent manner together with de-

veloping systems. This study presents a systematic approach in classifying developing and nondeveloping

cloud clusters based on their large-scale environments.

Eight years of hourly satellite IR data and global model analysis over the western North Pacific are used. A

cloud cluster is defined as an area of#208-K cloud-top temperature, generally mesoscale in size. Based on the

overlapping area between successive hourly images, they are then tracked in time as time clusters. The initial

formations of nearly all TCs during July–October 2003–10 were associated with time clusters lasting at least

8 h (8-h clusters). The occurrence of an 8-h cluster is considered to indicate theminimumdegree of convective

organization needed for TC genesis. A nondeveloping system is defined as an 8-h cluster that is considered to

be a viable candidate for TC genesis, but was not associated with the TC genesis.

The large-scale environmental conditions of cyclonic low-level vorticity, low vertical wind shear, low-level

convergence, and elevated tropospheric water vapor are statistically more favorable for developing systems.

Generally, the environment became more (less) favorable with time for the developing (nondeveloping)

systems. Nevertheless, many developing (nondeveloping) systems formed (dissipated) in seemingly un-

favorable (favorable) environments within a lead time of ,24 h.

1. Introduction

The dominant mode of organization of convection in

the tropics is mesoscale convective systems (MCSs) with

convective cores and stratiform precipitation (Houze

1977; Zipser 1977). Tropical cyclogenesis [(TC) genesis]—

the initial development of a self-sustaining, warm-core

vortex—does not occur without persistent convection and

low-level convergence (Gray 1968). Much of our knowl-

edge on TC genesis is based on developing systems that

often occur in environments with enhanced low-level

convergence, vorticity, and atmosphericmoisture that are

favorable for TC genesis. However, these favorable en-

vironmental conditions occur much more often than

TC genesis events. What are the sufficient conditions

for TC genesis, and what is the role of mesoscale

convective systems leading up to TC genesis? To answer

these questions, it is necessary to study both developing

and nondeveloping systems in an objective, consistent

manner.

a. Definition of TC genesis

Part of the ambiguity in studying developing and

nondeveloping systems is the definition of TC genesis.

Frank (1987) defined TC genesis as the transition from

a tropical disturbance to a tropical depression (TD),

with a closed surface wind circulation. This is similar to

the definition used by the National Hurricane Center

(NHC) and the Joint Typhoon Warning Center (JTWC).

The subsequent transformations of the TD to a tropical

storm (TS) and the TS to a hurricane are referred as

‘‘development’’ and ‘‘intensification,’’ respectively. Others

have proposed a two-stage genesis process: 1) large-scale

development to describe the development of the envi-

ronment favorable for TC genesis and 2) core formation

(e.g., McBride 1995). Note that convective cloud systems

can modulate their environment through moistening and

Corresponding author address: Dr. Brandon Kerns, Rosenstiel

School of Marine and Atmospheric Science, University of Miami,

4600 Rickenbacker Causeway, Miami, FL 33149.

E-mail: [email protected]

192 MONTHLY WEATHER REV IEW VOLUME 141

DOI: 10.1175/MWR-D-11-00239.1

� 2013 American Meteorological Society

heating. The interaction between deep convection and its

environment is complex. The two stages of TCgenesismay

be impossible to separate in some cases.

Because of the ambiguous nature of defining TC gen-

esis and the general lack of detailed, high-resolution in

situ data over tropical oceans, it is impossible to de-

termine the exact time of TC genesis for most storms,

even at the 6-h temporal resolution of best-track datasets.

Indeed, different operational centers frequently have

different TD initiation times in their best tracks for the

same storm. In this study, the initial appearance of a TD

or TS in the JTWC best track (‘‘best-track genesis’’) is

used as the reference point for TC genesis.

b. Large-scale environment and TC genesis

The large-scale environment, including area-averaged

vorticity, vertical wind shear, low-level convergence, and

moisture content, strongly influences TC genesis (Gray

1968; DeMaria et al. 2001; Schumacher et al. 2009).

However, the actual genesis process depends critically on

convective systems of various scales and their interaction

with the large-scale disturbance (Simpson et al. 1997,

1998; Houze 2010). Tory and Frank (2010) provide a de-

tailed summary on various hypotheses on TC genesis

from both large-scale and mesoscale perspectives.

Tropical cyclone genesis is most favored in regions

with elevated low-level relative vorticity and low verti-

cal wind shear (Gray 1968; McBride and Zehr 1981;

DeMaria et al. 2001). The Saharan air layer with dry air

and high dust concentration may play a significant role

in modulating TC formation in the Atlantic (Dunion

and Velden 2004), though this is still a topic of active

research (Braun 2010). Camargo et al. (2007) suggested

that moisture is most likely to be a limiting factor in the

Atlantic during El Nino years.

While there are statistical differences between the

environment of the developing and nondeveloping dis-

turbances, they cannot be entirely distinguished based

on the large-scale conditions, using the current genera-

tion of atmospheric analysis products (Perrone and

Lowe 1986; Hennon and Hobgood 2003; Hennon et al.

2005; Kerns and Zipser 2009; Fu et al. 2012; Peng et al.

2012). The within-sample variance is comparable to or

larger than the difference between the developing and

nondeveloping sample means. Note that this appears to

be the case regardless of the tracking method. These

results can be interpreted to imply that the large-scale

environment determines the likelihood of TC genesis,

but other factors at smaller scales determine whether

and when genesis occurs. That is, TC genesis is proba-

bilistic to some extent (Simpson et al. 1998). It is also

possible that a more favorable environment may simply

indicate that the large-scale development component of

TC genesis (McBride 1995) is under way, and therefore

the inner core is more likely to form.

Recently several studies have shown that a deep

(up to ;700–600 hPa) meso-a-scale (200–2000 km)

Lagrangian closed circulation in the wave-relative flow

is favorable for TC genesis, especially in easterly waves

(Dunkerton et al. 2009; Wang et al. 2009, 2012). The so-

called Kelvin cat’s eye configuration (referred to as

a ‘‘wave pouch’’) is favorable because it affords a degree

of protection from the harsh environment and provides

a focus point for sustained deep convection and vorticity

accumulation.

c. Convective processes during TC genesis

Several mechanisms have been identified for organi-

zation of convective systems into the core of a TC. Chen

and Frank (1993) showed that long-lived MCSs can

create inertially stable areas in saturated, stratiform rain

regions. This is favorable for the development of the

initial warm core and hydrostatic surface pressure falls

because a greater fraction of the latent heat released

contributes to columnwarming (Schubert andHack 1982;

Rogers and Fritsch 2001). Houze et al. (2009) found that

large areas of saturated, mesoscale ascent led to the

genesis of Hurricane Ophelia (2005). On the other hand,

Bister and Emanuel (1997) proposed that a long-lived

mesoscale rain area (‘‘showerhead’’) and associated

evaporative cooling would lead to a saturated, cold-cored

midlevel vortex. However, it is unclear how the cold

core evolves into a warm-core system for TC genesis to

occur. Ritchie and Holland (1993, 1997) emphasized

the merger of multiple mesoscale convective vortices.

Vortex merger results in increasingly deeper, stronger

vortices, eventually leading to the development of a

TD or TS.

Simpson et al. (1997) present a case in which the in-

teraction of multiple MCSs leads to development of

a ‘‘nascent eye’’ (e.g., the initial warm-core vortex) in

the subsidence regions between the convective systems.

Dolling and Barnes (2012) also related the nascent eye

of a TS tomesoscale subsidence. Harr et al. (1996) found

that the initial round of MCSs modifies the subsynoptic

environment such that the newMCSs organize in bands,

leading to the development of some western North Pa-

cific TCs. Hendricks et al. (2004) and Montgomery et al.

(2006) emphasized the role of ‘‘vortical hot towers’’ as-

sociatedwith strong updrafts in convective cores, which is

important for concentrating vorticity near the surface.

In all the above studies as well as operational practice,

persistent and organized convection is vital for TC gen-

esis.However, themeaning of ‘‘persistent and organized’’

can be difficult to define and is qualitative in many of

these studies.

JANUARY 2013 KERNS AND CHEN 193

d. Tracking developing and nondeveloping systems

Two types of tracking methods have been used in

previous studies: ones based primarily on dynamic fields

and others based on infrared satellite data. Thorncroft

and Hodges (2001) used an automated tracking method

using filtered 850-hPa relative vorticity to develop a 20-

yr climatology of Atlantic African easterly wave (AEW)

activity. Hopsch et al. (2007, 2010) have also used this

vorticity tracking technique. Kerns et al. (2008), Fu et al.

(2012), and Peng et al. (2012) use variations of tracking

filtered relative vorticity. Dunkerton et al. (2009) focus

on the intersection of the wave trough and the critical

layer, which is viewed as the focal point of TC genesis.

The use of clouds as tracers for developing and non-

developing disturbances in the tropics goes back to the

first meteorological satellite images. Gray (1968) con-

sidered tropical disturbances to be ‘‘distinctly orga-

nized cloud and wind patterns in the width range of

100–600 km, which possess a conservatism in time of at

least a day or more.’’ This concept was apparently moti-

vated by observations that midlatitude cloud patterns

generally correspond with circulation features of similar

scale (Hayden 1970). This viewpoint crystallized into the

concept of the synoptic disturbance ‘‘cloud cluster’’

(Gray 1973; Ruprecht and Gray 1976; McBride 1981).

Ruprecht and Gray (1976) described the cloud cluster

quite qualitatively: it is ‘‘a bright, solid white blob’’ in

satellite imagery. McBride (1981) described the synoptic

disturbance cloud cluster as ‘‘a loosely organized collec-

tion of deep convective clouds and covered . . . by a thick

cirrus shield.’’ Note that they used the same dataset as

Ruprecht and Gray (1976), which had considerable lim-

itations in terms of temporal resolution. Modern satellite

data allows for a much more detailed analysis of con-

vective systems through their life cycles.

More recent studies using synoptic cloud cluster con-

cept as a means to identify synoptic-scale disturbances

include Perrone and Lowe (1986), Hennon andHobgood

(2003), Hennon et al. 2005, and Hennon et al. (2011).

Hennon et al. (2011) described an objective method for

tracking these synoptic-scale cloud systems globally.

Cloud clusters may ‘‘disappear’’ for up to 12 h and still be

tracked as the same system, as long as redevelopment

occurs within a reasonable radius. Note that redevelop-

ment of convection does not always imply coherent,

continuous propagation of a synoptic weather system. In

fact, large areas of cloudiness can occur without an easily

identifiable synoptic-scale wind disturbance (Simpson

et al. 1967, 1968).

An alternative approach to objectively define and track

cloud clusters is to focus on mesoscale organization.

Williams and Houze (1987) defined cloud clusters using

an infrared brightness temperature threshold of 208 K,

tracking them in time as time clusters to study winter

monsoon convection in the vicinity of Borneo. This def-

inition of cloud clusters/time clusters emphasized co-

herent mesoscale convective systems, in contrast to the

synoptic-scale cloud clusters discussed above. In this

framework, the multiple convective ‘‘burst’’ events lead-

ing up to TC genesis observed by Zehr (1992) are related

to the development and decay of individual mesoscale

cloud clusters. The cloud cluster/time cluster tracking

framework allows for quantifying multiscale variability

from the mesoscale up to intraseasonal and interannual

time scales (e.g.,Mapes andHouze 1993; Chen et al. 1996;

Chen and Houze 1997a,b). The cloud cluster identifica-

tion and tracking used in these studies are completely

automated and objective, in contrast to previous studies

which relied on manual disturbance identification and

tracking (McBride and Zehr 1981; Perrone and Lowe

1986; Hennon and Hobgood 2003; Kerns et al. 2008;

Dunkerton et al. 2009; Fu et al. 2012; Peng et al. 2012).

Defining developing and nondeveloping systems based

on the cloud cluster tracking method using hourly satel-

lite data is less ambiguous and easy to reproduce in both

operational and research environments.

The organization of this manuscript is as follows. The

data used are described in section 2. The tracking of cloud

clusters/time clusters and vorticity maxima are described

in sections 3 and 4. Classification between developing and

nondeveloping clusters is described in section 5. Section 6

presents an illustrative example of a developing and

nondeveloping case. The results of the large-scale envi-

ronment of developing and nondeveloping systems as

well as their evolution in time are presented in sections 7

and 8. Finally, conclusions are given in section 9.

2. Data

a. TC tracks

The JTWCbest track contains the 6-hourly coordinates

and intensity of named TCs and unnamed depressions

and subtropical cyclones. The best-track data are based

on postseason reanalysis of the available data for each

storm. Because of the need to infer storm initiation using

satellite estimates (e.g., Dvorak 1975), the initial ap-

pearance of a TD or TS in the best-track data is consid-

ered to be a reference point for defining a developing

system in this study, which may not necessarily be the

precise time of TC genesis.

The storms considered in this study occurred in 08–358N, 1008–1608E during July–October, 2003–10 (Fig. 1).

This includes: 84 typhoons, 39 tropical storms, and 17

tropical depressions. Of these, 4 typhoons and 1 trop-

ical storm entered the study region from the east, and 3


typhoons formed in late June then tracked through the

study region in July. Hurricane Ioke (2006) also entered

the study region from the east. For these storms the

precursors to TC genesis are not identified in this study.

b. Satellite data

The geostationaryMultifunctional Transport Satellites

(MTSAT) infrared (IR) channel (10.5–11.5 mm) data

from July–October 2003–10 are used for identification of

cloud clusters. The data are at 0.058 (approximately 5 km

near the equator) resolution. Hourly satellite data are

used to track the evolution of cloud clusters through their

life cycle.

The Tropical Rainfall Measuring Mission (TRMM)

Microwave Imager (TMI) retrieved column-integrated

water vapor [or total precipitable water (TPW)] and

optimum interpolated sea surface temperature (SST)

data are used for examining cloud cluster environmental

moisture and SST. The retrieved data are provided by

Remote Sensing Systems (RSS). For SST, daily fields are

used. The daily fields are optimum interpolation com-

posites of TRMM TMI data and AMSRE data at 0.258spatial resolution (Gentemann et al. 2004). They are

corrected for the diurnal cycle to;0800 local time using

an empirical function of solar insolation, wind speed,

and local time of observation (Gentemann et al. 2003).

SST data are averaged over an area with a radius of 18,and valid observations must have at least 50% coverage.

This criterion is effective as a land mask to eliminate

cases near or over land.

For the water vapor fields, the version 4 average of the

past 3 days of TMI data (bmaps_v04 d3d data) is used.

Because the tropical ocean is relatively moist in the

boundary layer, most of the variance in TPW is due

to variations in the midlevel water vapor content. The

disadvantage of the polar-orbiting satellite TMI and

Advanced Microwave Scanning Radiometer for Earth

Observing System (EOS) (AMSR-E) data is the limited

temporal sampling and gaps between swaths, as well as

rain contamination. This is alleviated by using 3-day-

averaged data, at the expense of some temporal vari-

ability. Because of the 48 averaging radius used and the

relatively small TPW gradients within the west Pacific

(WPAC) warm pool, the loss of information from the

temporal averaging does not significantly affect the

results. For TPW, cases are included only if there are

TPW data for at least half of the 48 radius. Cases withoutsufficient coverage (,50%), usually due to land cover-

age, are not considered in the statistical analysis of de-

veloping and nondeveloping cases.

c. Global analysis data

The National Centers for Environmental Protection

(NCEP) final analysis (FNL) is used to identify and

track TC-related vorticity maxima as well as vorticity

maxima that did not develop into TCs. It is also used to

determine the large-scale environment of the clusters

such as the vertical wind shear and near-surface con-

vergence. The FNL analysis incorporates additional

observations that were not available for inclusion in the

real-time NCEP Global Forecast System (GFS) analy-

sis. However, because of the sparse observational net-

work over most of the WPAC, the NCEP FNL relies

heavily on the GFS model first guess. In particular, it is

sensitive to the model convective parameterization. An

upgrade to the GFS model physics, including convective

parameterization, was implemented on 28 July 2010,

which could potentially affect the results for the last 3

months of the 8-yr study. Nevertheless, the upgrade is not

expected to affect the main conclusions of this study. The

data are 6-hourly and at 18 resolution. Area-averaged

fields are calculated from the analysis by averaging all

grid points within a 68 radius from the center. The most

recent available analysis data are used for each case (e.g.,

a case at 0500 UTC uses data from 0000 UTC).

3. Tracking cloud clusters

a. Cloud clusters

Following Williams and Houze (1987), cloud clusters

are identified as contiguous areas of IR brightness tem-

peratures below a threshold value encompassing a con-

tiguous area of at least 5000 km2 (equivalent diameter of

80 km), but in some cases over 100 000 km2 (350 km),

including meso-b- and small meso-a systems (Orlanski

1975). Theminimumarea threshold is arbitrary; however,

the subset of long-duration convective systems we focus

on attained areas of well above this threshold. The IR

FIG. 1. The number of TC genesis events within a 2.58 radius ofeach point (contours) and genesis locations (plus signs) for July–

October 2003–10. Contours are for every 1 event, starting from 3,

and bold for 5 and above.


threshold is 208 K based on previous studies over the

western Pacific during the Tropical Ocean and Global

Atmosphere Coupled Ocean–Atmosphere Response

Experiment (TOGA COARE; Mapes and Houze 1993;

Chen et al. 1996). There is generally a close correspon-

dence between 208-K cloud area and contiguous meso-

scale areas of rainfall observed by radar. The choice of

208 K filters out many nonprecipitating anvil clouds.

While the correspondence between cold clouds and pre-

cipitation is not one-to-one (Arkin and Ardanuy 1989),

most, if not all, of the 208-K cloud clusters represent

convective systems that at one point in time had a signif-

icant precipitating area.

Note that inmany previous studies synoptic-scale cloud

clusters were tracked, using warmer brightness tem-

perature thresholds and/or subjective manual tracking

methods. This study focuses on objectively tracked me-

soscale convective systems defined by 208-K cold cloud

tops roughly corresponding to rain areas. After the cloud

clusters are identified, the center of the cloud cluster is

defined as the geometric center of the,208-K area. The

size of the cloud cluster is defined as the diameter that

the cloud cluster would have if it were a circle with the

same area.

Figure 2 shows a map of objectively identified cloud

clusters (dark blue shading) at 0000 UTC 8 September

2008. The largest cloud clusterwith an area of 90 400 km2

(size of 340 km) near 168N, 1268E is associated with Ty-

phoon Sinlaku. The rainbands surrounding Sinlaku also

appear as cloud clusters. Several other relatively large

cloud clusters, the largest of which is;75 900 km2 (size

; 310 km), are not associated with TCs, but with a

monsoon disturbance in the South China Sea. Another

area with several large cloud clusters is the intertrop-

ical convergence zone (ITCZ; roughly 58–108N) east of

1458E.

b. Time clusters

Time clusters are cloud clusters that can be tracked in

time with various lifetimes and sizes. Cloud clusters that

overlap by at least 50% or 10 000 km2 between con-

secutive hourly satellite images are considered to be part

of the same time cluster. Time clusters may consist of

cloud clusters that merge and/or split between consec-

utive frames. Note that the term time cluster refers only

to convective systems with continuity in time between at

least two consecutive images.1

FIG. 2. EnhancedMTSAT-1R infrared satellite image at 0000 UTC 8 Sep 2008. Blue regions

are where cloud-top temperature is lower than 208 K, representing the extent of the cloud

clusters. Cyan circles indicate the geometric centers of cloud clusters with the circle area

proportional to the cluster area.

1 Time cluster tracking resets at the beginning of each month.

This results in some long-lasting time clusters occurring at the end

of the month not being continuously tracked into the next month.

Because of the limited number of 8-h clusters lasting from one

month into the next month, this does not affect significantly affect

the results of this study.


At any given time, a time cluster may consist of more

than one cloud clusters that merged at a future time or

split at a previous time, and therefore belong to the same

time cluster. The center and area of the time cluster at

a given time are the geographical center and the com-

bined area of all the constituent cloud clusters, re-

spectively. The maximum area (size) associated with the

lifetime of the time cluster is the maximum area (size)

among the time cluster snapshots. As inTOGACOARE,

time clusters with longer duration generally attained

larger sizes (Fig. 4a and Chen and Houze 1997a,b).

The results presented in rest of this study will use time

clusters exclusively to represent convective cloud sys-

tems. The terms ‘‘cluster,’’ and ‘‘time cluster’’ are used

interchangeably from here on.

Similar to the near-equatorial region during TOGA

COARE (Chen et al. 1996), most time clusters move

westward for a few hours and remain relatively small

(Fig. 3) Only a small minority of time clusters are as-

sociated with TCs. The development and life cycles of

Supertyphoons Sinlaku, Hagupit, and Jangmi and Trop-

ical Storms Mekkhala and Higos in September 2008 are

clearly associated with long-lived time clusters (time

clusters lasting at least 24 consecutive images are high-

lighted). For the case of Sinlaku, the highlighted cluster

lasted 159 h and could be tracked starting 9 h prior to

best-track genesis. Hagupit and Higos also have west-

ward-moving groups of time clusters that are easily

identified as precursors to the storm formation (e.g., de-

veloping clusters). In addition to the developing systems,

there are long-lived time clusters that move westward,

often in distinct envelopes.

c. Initial tropical cyclone time clusters

As a first step to determine the properties of time

clusters that develop into TCs, the initial TC time cluster

for each TC that formed within the study area in July–

October 2003–10 (e.g., Fig. 1) is identified objectively

as follows. It is the longest-lasting time cluster among

either 1) the largest time cluster present at the time of

initial TD appearance in best track or 2) the first time

cluster that formed within 24 h after the initial TD ap-

peared in the JTWC best-track data. For a few weak,

disorganized storms (e.g., unnamed TDs and minimal

tropical storms with intermittent convection), a domi-

nant time cluster could not be tracked until up to 24 h

after best-track genesis.

For all TDs and TCs considered in this study from

July–October 2003–10, the initial TC time clusters lasted

at least 8 h, while the time clusters at later stages of the

TChad a range of sizes and durations similar to the set of

all time clusters (Fig. 4). A time cluster that persists for

at least 8 h (e.g., nine consecutive images) is referred to

as an 8-h cluster. There are 48 609 time clusters tracked

in the study region during July–October 2003–10. A total

of 4379 of them (e.g., 9%) lasted at least 8 h. Physically,

8-h clusters are convective systems with long durations,

which are likely to have large stratiform rain areas at

some time in their lifetime and are likely to persist be-

yond the time of dominant diurnal surface forcing (e.g.,

Chen and Houze 1997a).

4. Tracking vorticity maxima

To help identify the 8-h clusters leading up to a TC

genesis (i.e., the precursors to TC genesis), an objective

vorticity maximum tracking algorithm is used to track

the system in time prior to best-track genesis. AGaussian

smoothing function was applied to the NCEP FNL rela-

tive vorticity at 850 hPa. The weighting function is a

Gaussian bell curve with standard deviation of three grid

points (38, ;330 km) extending out nine grid points

(;990 km) in each direction. This smoothing filters out

small-scale features that are not fully resolved by the

analysis. Kerns et al. (2008) showed that coherent vor-

ticity maxima could be tracked back several days prior to

initial formation of some TCs.

Vorticity maxima are first identified and then tracked

in time by matching them with the closest one within 38radius in the next time frame using the 6-hourly data.

Tracks end when a vorticity maximum can no longer be

found within 38 at the next analysis time. The minimum

threshold value of the vorticity tracking is 23 1026 s21.

It is a low threshold chosen to be as inclusive as possible.

Vorticity maxima can be seen moving generally

westward, often together with coherent envelopes of

cloud clusters (Fig. 5). Some are evidently precursors

leading to TC formations and others are not associated

with TC genesis. For example, Typhoons Hagupit and

Higos were associated with well-defined vorticity fea-

tures and pre-TC vorticity centers that could be tracked

from east of 1608E (outside of the cluster tracking do-

main). The vorticity maxima associated with each TC

are found by searching within 38 of the best-track data.

Most vorticity maxima that are associated with best-

track TDs are within 18 distance of the best track (not

shown). The vorticity maxima tracks extend up to sev-

eral days prior to TC formation and can cross a signifi-

cant part of the basin during that time (Fig. 6a).

5. Classification of developing and nondevelopingsystems

Because the initial developments of all TCs during

July–October 2003–10 are associated with an 8-h cluster,

the occurrence of an 8-h cluster is considered to indicate


aminimumduration and level ofmesoscale organization

that is necessary, but not sufficient, for TC genesis to

occur. Therefore, only 8-h clusters over the ocean, which

are not already TCs, are included as candidates for

further identification of developing and nondeveloping

systems. The 8-h minimum duration threshold greatly

reduces the number of time clusters to only those that

have a potential to develop into a TC and, therefore, are

considered a viable candidate to TC genesis.

a. Developing systems

A developing system is defined as a time cluster or a

group of time clusters that last at least 8 h and are as-

sociated with a tractable vorticity maximum that pro-

vides a way to identify the precursors leading up to TC

formation. Developing systems include pre-TC clusters

that are backtracked in time using the vorticity maxima

tracks. For many storms the initial TC time cluster that

FIG. 3. Time–longitude diagram of cloud clusters and time clusters within the domain shown

in Fig. 2, for September 2008. Black circles represent cloud cluster locations. The size of the

circles is proportional to the size of the cloud clusters. Time clusters are connected with red

lines. Time clusters lasting at least 24 h are highlighted with yellow circles and dark blue lines.

TC best tracks (green lines) are labeled by storm names (unnamed tropical systems are labeled

by storm number). The initial appearance of each storm as a TDor TS in the JTWCbest track is

indicated with a thick, black circle.


existed at the time of initial appearance of a TD in the

best track (e.g., initial TC clusters) could be tracked

several hours back before genesis as a time cluster and in

some cases more than a day (Fig. 6b). Additionally, any

8-h cluster passing within 58 of a pre-TC vorticity track is

also considered to be a developing system (Fig. 7). The

developing systems have a similar range of lead times as

the pre-TC vorticity maxima (Figs. 6a,b).

b. Nondeveloping systems

A nondeveloping system is defined as an 8-h clus-

ter that is not a TC cluster and is not previously iden-

tified as a developing cluster (Fig. 7). As illustrated by

Fig. 5, many nondeveloping systems are associated with

westward-moving vorticity features and trackable vor-

ticity centers.

Among the 31 289 time clusters tracked over the

ocean, 3144 were 8-h clusters (Fig. 7). A total of 398 of

the 3144 8-h clusters were associated with active TCs and

are excluded as candidates for TC genesis. Among the

other 2746 8-h clusters, 435 (15.8%) were developing

systems and 2311 (84.2%) were nondeveloping systems.

6. Examples of developing and nondevelopingsystems

a. Precursors to Typhoon Hagupit (2008)

As shown in Figs. 3 and 5, Typhoon Hagupit was a

case with a well-defined sequence of precursor devel-

oping clusters going back several days prior to TC for-

mation. Further details on the evolution of vorticity and

time clusters leading up to the development of Hagupit

are shown in Fig. 8. An envelope of latitude-averaged

vorticity with an embedded vorticity center can be traced

back to east of 1608E at least as far back as 12 September,

the 8-h clusters associated with the vorticity maximum

began on 14 September, and a 4-day series of 8-h clusters

occurred during 15–18 September, leading up to the ini-

tial appearance of the TD in best track at 1800 UTC 18

September.

The 8-h clusters occurred preferentially to the south of

the center of vorticity, and remained skewed to the

south even after the initial TD had formed. This is

probably due to the effect of large-scale vertical shear on

the system (Black et al. 2002; Corbosiero and Molinari

2002; Lonfat et al. 2004). Also note that the vorticity

maximum weakened somewhat on 16 September, but

then it intensified after ;1 day of sustained 8-h cluster

activity. As shown in the relative vorticity swath map

(e.g., the maximum vorticity along the track of the

vorticity center), the relative vorticity gradually increased

as the system moved westward on 17–18 September. It

is likely that the persistent, organized convection on

16–18 September played a strong role in modifying the

mesoscale environment. That is, the time clusters can

be considered distinct precursors to the TC genesis. The

convection also likely contributed to the preconditioning

of the large-scale vorticity and moisture leading up to the

formation of Hagupit.

b. Nondeveloping system

A sequence of nondeveloping systems during 24–30

September 2010 was associated with a westward-moving

envelope of vorticity maxima, which is shown as an

FIG. 4. Time cluster duration and maximum size (e.g., equivalent diameter) for (a) all time clusters and (b) time

clusters passing within 38 of JTWC best track (e.g., TC time clusters). In (b), dominant clusters associated with the

initial formation of the TC are indicated as solid black dots, and other TC clusters at a later stage in the storm

evolution are drawn as open circles. The vertical dashed line indicates the 8-h duration threshold, and the arrow

indicates time clusters lasting at least 8 h (8-h clusters).


example in Fig. 9. In this case, a vorticity center could

be tracked over 3 days moving from ;1408E to the

Philippines, along ;128N. The envelope of enhanced

vorticity can be traced back farther to the east, but not the

vorticity center. Several 8-h clusters occurred along and

to the south of the vorticity center track. Based on the

vorticity swath map, the vorticity center attained an am-

plitude similar to pre-Hagupit on 15 September 2008, but

FIG. 5. Time–longitude diagram of 8-h clusters and NCEP FNL relative vorticity for Sep-

tember 2008. The background color shading is 850-hPa relative vorticity averaged from 58–258N along each longitude grid. The 8-h cluster tracks are denoted as magenta circles with the

circle area proportional to the time cluster area. Vorticity maxima lasting at least 24 h are

drawn as triangles. JTWC TC best tracks are drawn as thick, black lines.


despite of the collocated 8-h clusters and the vorticity

maxima over a period of 3 days before moving into the

vicinity of the Philippine islands, the system did not in-

tensify further into a TC. This nondeveloping system

eventually moved over the Philippines. Note that TC

genesis does occur in close proximity to the Philippines

(Fig. 1), which means the islands were not necessarily a

factor in this case, especially since it failed to develop

over a 3-day period prior to approaching the islands.

Clearly, the collocation of relative vorticity and 8-h

clusters is not sufficient to determine whether TC genesis

will occur, and it is necessary to consider other factors.

7. Large-scale environment of developing andnondeveloping systems

The developing and nondeveloping systems are as-

sociated with a wide range of environment conditions,

quantified by low-level vorticity, low-level convergence,

vertical wind shear, moisture content, and SST. In the

subsequent analysis, the large-scale environmental con-

ditions are computed at each hourly snapshot throughout

the 8-h clusters’ lifetimes. In addition to developing and

nondeveloping systems, the environment predictors are

also calculated for the first 24 h of the TCs for com-

parison purpose. There is a significant overlap in these

parameters between developing and nondeveloping

systems (Fig. 10). Nevertheless, there are significant sta-

tistical differences (above 95% confidence in the Stu-

dent’s t test) in all the parameters, except SST. This

separation in the environments of the developing versus

nondeveloping systems can be used to obtain a skillful

discrimination between the two subsets of 8-h clusters.

Additionally, the information provided by the time se-

quence of cloud clusters allows the time history of the

developing and nondeveloping systems to be quantified.

a. Environmental conditions

The strongest statistical predictor of TC genesis is

low-level relative vorticity (Fig. 10a). Consistent with

McBride and Zehr (1981), the median relative vorticity

for the developing cases (18 3 1026 s21) is nearly twice

FIG. 6. Scatter diagram of lead time vs distance from the initial tropical depression (TD) formation based on JTWC

best track: (a) pre-TC vorticity maxima and (b) developing 8-h clusters. In (b), the triangle markers are for the initial

TC time clusters, which are trackable directly to the TC.

FIG. 7. Schematic of the classification of 8-h clusters into de-

veloping and nondeveloping systems. The number of time clusters

represented is in parentheses for each category.


that of the nondevelopers2 (9 3 1026 s21). This is much

greater discrimination than Fu et al. (2012) who found

only a 20% difference using filtered vorticity fields. For

reference, the relative vorticity distribution of the first

24 h of the TCs is also shown. As expected, the TCs have

somewhat higher relative vorticity than the developing

systems. This suggests that most of the large-scale rela-

tive vorticity was generated prior to TC genesis. Note

that the system vorticity and the large-scale environ-

ment vorticity cannot be entirely distinguished.

In addition to low-level vorticity, themedian low-level

convergence is ;30% greater for the developing sys-

tems versus nondeveloping systems (Fig. 10b). As with

relative vorticity, the difference between developing

systems and the first 24 h of TCs is less significant.

Vertical wind shear is significantly lower on average

for the developing systems (Fig. 10c). The median ver-

tical wind shear of the developing systems is ;9 m s21

(17 kt). This median value of shear is similar to the

threshold below which rapid intensification of TCs gen-

erally occurs (Kaplan and DeMaria 2003). Half of the

developing systems occurred in an environment with

shear .9 m s21. Also, about 30% of the nondeveloping

cases were associated with shear values ,9 m s21.

Satellite-derived TPW was somewhat higher in de-

veloping cases compared with nondeveloping cases,

but it is not a strong discriminating factor (Fig. 10d).

Hennon and Hobgood (2003) also found TPW to be

a poor predictor in the WPAC. This is probably due to

a generally moist environment compared with the At-

lantic. A similar calculation was done using only the

subset of cases with individual TMI swaths in close

proximity in space and time, and similar results were

obtained (not shown).

Satellite-derived SST did not suggest a significant dif-

ference between developing and nondeveloping systems

FIG. 8. Evolution of vorticity maxima and cloud clusters associated with the genesis of Typhoon Hagupit (2008).

The best track is shown as the thick black line. The vorticity maxima associated with the particular storm is drawn as

triangles. The 8-h cloud clusters associated with the particular storm (e.g., TC clusters and developing clusters) are

drawn as magenta circles, proportional to the cloud cluster size. (left) Time–longitude Hovmoller plot. The back-

ground color shading is relative vorticity at 850 hPa averaged from 58–258N at each longitude. (right) Map of the

tracks of the features shown in (left). The color shading is the maximum relative vorticity within 108 of the vorticity

maximum center.

2 A sensitivity test was performed including only the non-

developing systems tracking within 58 of a vorticity center lasting

over 24 h. The median value of relative vorticity obtained was

;2 3 1026 s21 higher.


(Fig. 10e). This is probably because of the generally warm

SSTs in the WPAC during July–October.

b. Genesis productivity

The preferred areas of TC genesis (Fig. 1) generally

correspond with the preferred areas of 8-h clusters

(Fig. 11a). Nevertheless, the maximum of 8-h cluster

occurrence is in the South China Sea just west of the

Philippine islands, whereas theTCgenesis ismost frequent

east of the Philippines near 1308E. The South China Sea

has a relatively larger number of nondeveloping clusters

than the warm pool region east of the Philippines. The

relative frequency of developing and nondeveloping

clusters is summarized using a genesis productivity simi-

lar to Kerns et al. (2008) and Hennon et al. (2013). The

genesis productivity of 8-h clusters was calculated as the

number of developing time clusters divided by the total

number of candidate time clusters (developing and non-

developing clusters, excluding TCs).

The genesis productivity is greatest over the west

Pacific warm pool near 158–208N, 1308E, which is col-

located with the relatively high occurrence of 8-h clus-

ters (Fig. 12). In contrast, the South China Sea has

relatively low genesis productivity. The peak in genesis

productivity east of the Philippines is farther north than

the maximum track density of both vorticity maxima

and 8-h clusters (Figs. 11a,b). Similar to the Atlantic,

the maximum genesis productivity is located to the

north of the maximum track density of disturbance

centers (Kerns et al. 2008). Note that the latitude of

maximum genesis productivity corresponds to the

maximum latitude—and highest planetary vorticity—

within the geographical region where 8-h cluster occur-

rences are relatively frequent. Motivated by the maxi-

mum of genesis productivity at ;178N, the following is

considered to be an additional predictor of TC genesis in

the WPAC:

ALAT175 jlatitude2 17j ,

where the vertical bars indicate the absolute value.

c. Linear discriminant analysis

The developing systems statistically form in more fa-

vorable environmental conditions (e.g., lower shear,

higher humidity, stronger low-level vorticity, and con-

vergence) than the nondevelopers. Linear discriminant

analysis (LDA) was performed using the five environ-

mental parameters shown in Figs. 10a–e as well as

ALAT17 (section 7b). LDA can determine the overall

FIG. 9. As in Fig. 8, but for the nondeveloping cases associated with a moderately strong vorticity maximum from

27–30 Sep 2010.


discrimination between developing and nondeveloping

systems using an optimal linear combination of multi-

ple predictors (Perrone and Lowe 1986; Hennon and

Hobgood 2003; Kerns and Zipser 2009). The probability

of development from LDA (or posterior probability) is

considered to be an indication of how favorable the large-

scale environment is for TC genesis. A climatological

probability of development is defined as the fraction of

the developing systems in all 8-h cluster instances, which

is ;20%. When LDA probability is significantly greater

FIG. 10. Cumulative distributions of the large-scale environment conditions associated with 8-h cluster snapshots:

(a) 850-hPa relative vorticity, (b) 925-hPa divergence, (c) 850–200 hPa vertical wind shear, (d) TMI 3-day mean

TPW, and (e) TMI/AMSRE SST. Nondeveloping clusters are indicated by the dashed curves. Developing clusters

are indicated by the solid curves. The first 24 h of TCs are indicated by the light, dotted curves. The dynamic fields

from the FNL analysis are based on 08–68 averages in (a)–(c), TPW is based on 08–48 averages in (d), and SST is based

on 18 averages in (e).


(less) than the climatological background probability of

development, the environment can be considered to be

favorable (unfavorable) for TC genesis. An environment

with LDA probability to develop close the climatology is

considered to be marginally favorable for TC genesis.

To determine how effective the LDA is for distin-

guishing between developing and nondeveloping sys-

tems, the leave-one-out (‘‘jackknife’’) cross validation

is used. A single sample is left out, and the LDA is re-

calculated using the remaining samples. This process is

repeated for each hourly instance of the 8-h cluster, to

obtain the probability of that case developing. The prob-

ability of detection is the fraction of developing systems

that were correctly predicted to be developing systems.

The false-alarm rate is the fraction of nondeveloping sys-

tems that were erroneously predicted to be developing

systems. The probability of detection and false-alarm rate

depend on the threshold posterior probability used to

decide when to predict development, which is also re-

ferred to as the decision threshold.

The Heidke skill score (HSS) is used to determine

the predictive skill relative to making random guesses

(Marzban 1998). HSS ranges from below 0 to 1. HSS.0.3 is considered to indicate a benchmark for a useful

level of skill. Skill scores above the benchmark are seen

for posterior probabilities between 0.15 and 0.50 (Fig. 13).

Using the climatological likelihood (20% LDA proba-

bility to develop), a 70% detection rate with a 28% false-

alarm rate is obtained (Fig. 13). At the threshold with

FIG. 11. The number of tracks passing through each 58 3 58 during July–October 2003–10: (a) time clusters lasting at

least 8 h (8-h clusters) and (b) vorticity maxima lasting at least 24 h.

FIG. 12. The TC genesis productivity of 8-h clusters. Genesis

productivity is defined as the percentage of candidate (e.g., de-

veloping plus nondeveloping) 8-h clusters passing through each 58box that are identified as developing systems.

FIG. 13. The probability of detection (POD, dashed), false-alarm

rate (FAR, dash–dot), and Heidke skill score (HSS, solid) as

a function of the threshold LDA probability of development. The

climatological likelihood of development, which is the fraction of

candidate time cluster snapshots that are developing systems, is

indicated by the vertical gray line (;0.2). The POD, FAR, andHSS

are determined using leave-one-out (‘‘jackknife’’) cross validation.

See text for details.


peak skill as indicated by the HSS (;30% LDA proba-

bility to develop), there is roughly a 50% probability of

detection and a 12% false-alarm rate. Cloud cluster area,

growth rate, and aspect ratio were considered as potential

predictors, but they did not improve the skill, probably

because individual convective systems each go through

their individual life cycles with significant changes to their

morphology over periods of a few hours.

The HSS obtained in this study compares favorably

with previous studies using various tracking methods

and predictors. Hennon and Hobgood (2003) obtained

an HSS of ;0.4 for the 12- and 24-h lead-time periods.

Note that the lead time in this study is significantly

longer (median of 39 h). Hennon et al. (2005) expanded

the range of posterior probability above the benchmark

of HSS 5 0.3 using a neural network classifier, but did

not significantly increase the peak skill score. The range

of posterior probability with HSS . 0.3 in this study is

somewhat greater than their LDA classification, but less

than their neural network classifier. The skill in this

study is comparable to Kerns and Zipser (2009) for the

Atlantic, but significantly lower than their results in the

eastern North Pacific. Peng et al. (2012) and Fu et al.

(2012) did not present skill scores, but their filtered

relative vorticity appears to provide somewhat less

discrimination than the spatial averaged vorticity used

in this study (their Fig. 9). Because relative vorticity is

the strongest predictor, it is likely that the skill of their

studies would be somewhat lower than the current study.

Note that these skill score comparisons are limited be-

cause of the unique mesoscale classification method

used here compared with the synoptic-scale classifica-

tion used in previous studies. Nevertheless, the skill

scores obtained are not fundamentally different than

previous studies using synoptic classification.

Despite the statistically significant differences be-

tween the developing and nondeveloping systems, there

are many nondeveloping systems occur in environments

that are apparently favorable for TC formation (e.g.,

LDA probability to develop .20%); 20% of the non-

developing systems have at least the same chance to

develop as the median of the developing systems. Recall

that the sample size of nondeveloping systems is 2311,

while there are only 436 developing (Fig. 7). For cases

with the LDA probability to develop greater than the

climatological likelihood of development, the fraction

of developing systems is larger than the fraction of

nondeveloping systems (Fig. 14a); however, the num-

ber of nondeveloping cases with a probability to de-

velop larger than the climatological background is

considerably larger than the number of developing sys-

tems with the same probability, up to posterior proba-

bility of 0.4 (Fig. 14b).

8. Evolution of developing and nondevelopingsystems

To better understand and further distinguish the char-

acteristics of the developing and nondeveloping systems,

evolutions of both groups are examined in context of

their large-scale environmental conditions in terms of the

LDA probability to develop. The developing systems are

backtracked in time so that their evolution leading up to

TC genesis can be examined in an evolving large-scale

environment (Fig. 15a). A small group of the developing

systems can be tracked back as long as 8 days prior to TC

FIG. 14. (a) The probability density function (PDF) of posterior probabilities from linear discriminant analysis. (b)

The number of instances (e.g., hourly snapshots) of nondeveloping and developing systems in each bin (0.05) of LDA

posterior probability. The gray vertical line represents the climatological likelihood of an 8-h cluster being a de-

veloping cluster (;0.2).


genesis. Not surprisingly, the environment is not partic-

ularly conducive to TC genesis at that early stage. As the

genesis time approaches, cloud clusters are statistically

more likely to occur in increasingly favorable environ-

ments.On the other hand,many other developing systems

occurred inmoderate to less favorable environments even

with short lead times ,48 h. In fact, the peak of the dis-

tribution at,48-h pregenesis (e.g., contours in Fig. 15a) is

in a marginally favorable environment near the climato-

logical likelihood of TC genesis (;20%).

The evolution of nondeveloping systems is examined

from the formation of a 8-h cluster to its dissipation

(Fig. 15b). In contrast to the developing systems, the

nondeveloping systems statistically occurred in less

favorable environments as time progressed. Themajority

of ‘‘false alarms’’ occurred with clusters lasting ,36 h,

which represents the vast majority of nondeveloping

systems. A minority of very long-lasting (.24 h) clusters

remained in a favorable environment for .24 h prior to

their dissipation. However, 806 (274) of the 2311 non-

developing systems had LDA probability to develop

above 0.2 (0.3) within the first 24 h after cluster initiation.

The distinct evolution of the relatively lasting de-

veloping and nondeveloping systems may indicate that

their large-scale environmental conditions are better re-

solved by the global FNL analysis fields, whereas the

majority of developing and nondeveloping systems

formed in the environments that are less distinguish-

able and/or or unresolved by the global analysis. The

latter presents a major challenge for TC genesis fore-

casting using the large-scale environmental conditions,

which can lead to many false alarms.

The interaction of long-lasting cloud clusters with

their environment may modulate the conditions leading

up to TC genesis, which is unresolved in the current

observational data and global analysis fields. This could

FIG. 15. The time history of LDA probability to develop for (a) developing systems and (b)

nondeveloping systems. Each of the light gray lines represents the entire time history of an

individual 8-h cluster. In (a), time is with respect to the time of TD/TS formation in the JTWC

best track (gray lines). The initiation points of the time clusters are indicated with black dots. In

(b), time is with respect to the initial formation of the 8-h cluster, and the dissipation points are

drawn as triangles. Contours are drawn for the number of snapshots (e.g., hourly points along

the gray lines) in each bin: (a) 24-h and (b) 6-h bins. Both panels use 0.1 LDA probability bins.

Contours are for 10, 20, 50, 100, 200, 300, 400, 500, 1000, and 2000 snapshots for both (a) and (b).

The climatological probability to develop (0.2) is indicated by the black horizontal line in both

(a) and (b).


be due to a single dominant cloud cluster lasting ex-

tremely long (.24 h) or the repeated occurrence of

shorter duration clusters. In developing systems, the

persistent convection can feed back into the large-scale

environment, making it increasingly more favorable.

This interaction is more likely to occur in an environ-

ment with higher relative vorticity, which is more likely

to be inertially stable with a smaller Rossby radius of

deformation in those cases (Schubert and Hack1982). In

nondeveloping systems this positive interaction does not

take place, and instead the environment of the clusters

instead becomes more hostile with time. Determining

the mechanism that allows feedback to the environment

is an ongoing research topic. It is hypothesized that the

mesoscale environment in which convection occurs is

a critical factor in determining to what extent the con-

vection can feed back to the environment (e.g., Chen

and Frank 1993).

Another possible interpretation of the increasingly

favorable large-scale environment of developing sys-

tems with time is that the large-scale environment is

partly a symptom of the outer-core component of TC

genesis. This process may occur concurrently with inner-

core formation and may be inseparable from it.

9. Conclusions

Tropical convective cloud systems (cloud clusters) are

a key ingredient for TC genesis. The characteristics of

the cloud clusters and their environmental conditions

are examined in terms of developing and nondeveloping

systems. An objective cloud cluster tracking method is

used to track and classify developing and nondeveloping

systems. The key findings are summarized as follows:

d A total of 31 289 time clusters were tracked from July–

October 2003–10 over thewesternNorthPacificOcean.

Only 2311 8-h clusters were considered as viable

candidates for TC genesis, among which 435 (15.8%)

were developing systems and 2311 (84.2%) were non-

developing systems (Fig. 7).d The developing systems generally have greater low-

level vorticity, greater low-level convergence, lower

vertical wind shear, and higher water vapor content

than the nondeveloping systems (Fig. 10). There is no

significant difference in SST.d Linear discriminant analysis (LDA) can skillfully

distinguish between developing and nondeveloping

systems. The Heidke skill score obtained is ;0.4

(Fig. 13), among the highest skill scores presented in the

literature. A probability of detection of 70% and false-

alarm rate of 27% is obtained.d Generally, the environment became more favorable

with time for developing systems with long (1–7 days)

lead time to TC genesis (Fig. 15a), whereas nondevel-

oping systems tend to evolve into unfavorable environ-

ments with time (Fig. 15b).d Many developing (nondeveloping) systems formed

(dissipated) in seemingly unfavorable (favorable)

environments within a lead time (duration) of ,24 h,

which presents a major challenge for TC genesis

forecasting.

Incorporating the 8-h cluster duration can improve

statistical prediction of TC genesis by eliminating many

disturbances with weaker, short-lived convective sys-

tems and providing a more stringent criterion to de-

termine systems that have the potential to develop. In an

operational environment, the cloud cluster tracking

method and LDA classification could be used to provide

an objective, statistical guidance on the potential for

operational invest systems to develop into TCs.

The lack of a more robust separation of the environ-

mental conditions for developing and nondeveloping

systemsmay be in part due to the relative low-resolution

of FNL analysis fields. It is possible that the skill could

be improved by using higher-quality analyses. Al-

though observations from targeted field programs can

provide insights about TC genesis on mesoscale and

subsynoptic scales, the lack of continuous temporal

and spatial coverage makes it difficult to address the

convective–environmental interactions. A combination of

in situ observations and high-resolution, cloud-resolving

modeling over a large region allowing both convective

systems and their environment to interact freely will be

a key to better understand and predict the interaction of

the convection (e.g., long-lasting cloud clusters) with the

environment during TC genesis.

Acknowledgments. The authors wish to thank Mr. Ed

Ryan for his assistance on processing the satellite data

and cloud cluster tracking analysis. The comments and

suggestions from the two anonymous reviewers helped

to improve the manuscript significantly. This work was

supported by the Office of Naval Research (ONR) under

the Tropical Cyclone Structure 2008 (TCS-08) Research

Grant N00014-08-1-0258 and the Impact of Typhoon on

Ocean in the Pacific (ITOP) Research Grant N00014-08-

1-0576. The IR satellite data were processed by Kochi

University, Japan. Microwave OI SST, and TRMM TMI

water vapor retrievals are produced by Remote Sensing

Systems (http://www.remss.com).

REFERENCES

Arkin, P. A., and P. E. Ardanuy, 1989: Estimating climatic-scale

precipitation from space: A review. J. Climate, 2, 1229–1238.


Bister, M., and K. A. Emanuel, 1997: The genesis of Hurricane

Guillermo: TEXMEX analyses and a modeling study. Mon.

Wea. Rev., 125, 2662–2682.

Black, M. L., J. F. Gamache, F. D. Marks Jr., C. E. Samsury, and

H. E. Willoughby, 2002: Eastern Pacific Hurricane Jimena of

1991 andOlivia of 1994: The effect of vertical shear on structure

and intensity. Mon. Wea. Rev., 130, 2291–2312.

Braun, S. A., 2010: Reevaluating the role of the Saharan air layer in

Atlantic tropical cyclogenesis and evolution.Mon. Wea. Rev.,

138, 2007–2037.

Camargo, S., K. A. Emanuel, and A. H. Sobel, 2007: Use of

a genesis potential index to diagnose ENSO effects on tropical

cyclone genesis. J. Climate, 20, 4819–4834.Chen, S. S., and W. M. Frank, 1993: A numerical study of the

genesis of extratropical convective mesovortices. Part I: Evo-

lution and dynamics. J. Atmos. Sci., 50, 2401–2426.

——, and R. A. Houze Jr., 1997a: Diurnal variation and lifecycle of

deep convective systems over the tropical Pacific warm pool.

Quart. J. Roy. Meteor. Soc., 123, 357–388.

——, and ——, 1997b: Interannual variability of deep convection

over the tropical warm pool. J. Geophys. Res., 102 (D22),

25 783–25 795.

——, ——, and B. E. Mapes, 1996: Multiscale variability of deep

convection in relation to large-scale circulation in TOGA

COARE. J. Atmos. Sci., 53, 1380–1409.

Corbosiero, K. L., and J. Molinari, 2002: The effects of vertical

wind shear on the distribution of convection in tropical cy-

clones. Mon. Wea. Rev., 130, 2110–2123.

DeMaria, M., J. A. Knaff, and B. H. Connell, 2001: A tropical

cyclone genesis parameter for the tropical Atlantic. Wea.

Forecasting, 16, 219–233.

Dolling, K., and G. M. Barnes, 2012: Warm core formation in

Tropical StormHumberto (2001).Mon. Wea. Rev., 140, 1177–

1190.

Dunion, J. P., and C. S. Velden, 2004: The impact of the Saharan air

layer onAtlantic tropical cyclone activity.Bull. Amer.Meteor.

Soc., 85, 353–365.

Dunkerton, T. J., M. T. Montgomery, and Z. Wang, 2009: Tropical

cyclogenesis in a tropical wave critical layer: Easterly waves.

Atmos. Chem. Phys., 9, 5587–5646.

Dvorak, V. F., 1975: Tropical cyclone intensity analysis and

forecasting from satellite imagery.Mon. Wea. Rev., 103, 420–

430.

Frank, W. M., 1987: Tropical cyclone formation. A Global View

of Tropical Cyclones, R. L. Elsberry, Ed., Office of Naval

Research, 53–90.

Fu, B., M. S. Peng, T. Li, and D. E. Stevens, 2012: Developing

versus nondeveloping disturbances for tropical cyclone for-

mation. Part II: Western North Pacific. Mon. Wea. Rev., 140,

1067–1080.

Gentemann, C. L., C. J. Donlon, A. Stuart-Menteth, and F. J.

Wentz, 2003: Diurnal signals in satellite sea surface tempera-

ture measurements.Geophys. Res. Lett., 30, 1140, doi:10.1029/

2002GL016291.

——, F. J. Wentz, C. M. Mears, and D. K. Smith, 2004: In situ

validation of Tropical Rainfall Measuring Mission microwave

sea surface temperatures. J. Geophys. Res., 109, C04021,

doi:10.1029/2003JC002092.

Gray, W. M., 1968: Global view of the origin of tropical distur-

bances and storms. Mon. Wea. Rev., 96, 669–700.——, 1973: Cumulus convection and larger scale circulations.

I. Broadscale and mesoscale considerations. Mon. Wea. Rev.,

101, 839–855.

Harr, P. A., R. L. Elsberry, and J. C. L. Chan, 1996: Transformation

of a large monsoon depression to a tropical storm during

TCM-93. Mon. Wea. Rev., 124, 2625–2643.

Hayden, C. M., 1970: An objective analysis of cloud cluster di-

mensions and spacing in the tropical North Pacific.Mon. Wea.

Rev., 98, 534–540.Hendricks, E. A., M. T. Montgomery, and C. A. Davis, 2004: The

role of ‘‘vertical’’ hot towers in the formation of Tropical

Cyclone Diana (1984). J. Atmos. Sci., 61, 1209–1232.

Hennon, C. C., and J. S. Hobgood, 2003: Forecasting tropical cy-

clone cyclogenesis over the Atlantic basin using large-scale

data. Mon. Wea. Rev., 131, 2927–2940.

——, C. Marzban, and J. S. Hobgood, 2005: Improving tropical

cyclogenesis statistical model forecasts through the applica-

tion of a neural network classifier.Wea. Forecasting, 20, 1073–

1083.

——, C. N. Helms, K. R. Knapp, and A. R. Bowen, 2011: An ob-

jective algorithm for detecting and tracking tropical cloud

clusters: Implications for tropical cyclogenesis prediction.

J. Atmos. Oceanic Technol., 28, 1007–1018.——, and Coauthors, 2013: Tropical cloud cluster climatology,

variability, and genesis productivity. J. Climate, in press.

Hopsch, S. B., C. D. Thorncroft, K. Hodges, and A. Aiyyer, 2007:

West African storm tracks and their relationship to Atlantic

tropical cyclones. J. Climate, 20, 2468–2483.——,——, and K. R. Tyle, 2010: Analysis of African easterly wave

structures and their role in influencing tropical cyclogenesis.

Mon. Wea. Rev., 138, 1399–1419.

Houze, R. A., Jr., 1977: Structure and dynamics of a tropical squall-

line system. Mon. Wea. Rev., 105, 1540–1567.——, 2010: Clouds in tropical cyclones. Mon. Wea. Rev., 138,

293–344.

——, W.-C. Lee, and M. M. Bell, 2009: Convective contribution to

the genesis of HurricaneOphelia (2005).Mon.Wea. Rev., 137,

2778–2800.

Kaplan, J., and M. DeMaria, 2003: Large-scale characteristics of

rapidly intensifying tropical cyclones in the North Atlantic

basin. Wea. Forecasting, 18, 1093–1108.

Kerns, B.W., and E. J. Zipser, 2009: Four years of tropical ERA-40

vorticity maxima tracks. Part II: Differences between de-

veloping and nondeveloping disturbances. Mon. Wea. Rev.,

137, 2576–2591.

——, K. M. Greene, and E. J. Zipser, 2008: Four years of tropical

ERA-40 vorticity maxima tracks. Part I: Climatology and ver-

tical vorticity structure. Mon. Wea. Rev., 136, 4301–4319.Lonfat, M., F. D. Marks Jr., and S. S. Chen, 2004: Precipitation

distribution in tropical cyclones using the Tropical Rainfall

Measuring Mission (TRMM) Microwave Imager: A global

perspective. Mon. Wea. Rev., 132, 1645–1660.

Mapes, B. E., and R. A. Houze, 1993: Cloud clusters and super-

clusters over the oceanic warm pool. Mon. Wea. Rev., 121,

1398–1416.

Marzban, C., 1998: Scalar measures of performance in rare-event

situations. Wea. Forecasting, 13, 753–763.

McBride, J. L., 1981: Observational analysis of tropical cyclone

formation. Part I: Basic description of data sets. J. Atmos. Sci.,

38, 1117–1131.

——, 1995: Tropical cyclone formation. Global Perspectives on

Tropical Cyclones, R. L. Elsberry, Ed., World Meteorological

Organization, 66–105.

——, and R. Zehr, 1981: Observational analysis of tropical cyclone

formation. Part II: Comparison of nondeveloping versus de-

veloping systems. J. Atmos. Sci., 38, 1132–1151.


Montgomery, M. T., M. E. Nicholls, T. A. Cram, and A. B.

Saunders, 2006: A vortical hot tower route to tropical cyclo-

genesis. J. Atmos. Sci., 63, 355–386.

Orlanski, I., 1975: A rational subdivision of scales for atmospheric

processes. Bull. Amer. Meteor. Soc., 56, 527–530.

Peng, M. S., B. Fu, T. Li, and D. E. Stevens, 2012: Developing

versus nondeveloping disturbances for tropical cyclone for-

mation. Part I: North Atlantic. Mon. Wea. Rev., 140, 1047–1066.

Perrone, T. J., and P. R. Lowe, 1986: A statistically derived pre-

diction procedure for tropical storm formation. Mon. Wea.

Rev., 114, 165–177.Ritchie, E. A., and G. J. Holland, 1993: On the interaction of

tropical-cyclone-scale vortices. II: Discrete vortex patches.

Quart. J. Roy. Meteor. Soc., 119, 1363–1379.——, and ——, 1997: Scale interactions during the formation of

Typhoon Irving. Mon. Wea. Rev., 125, 1377–1396.

Rogers, R. F., and J. M. Fritsch, 2001: Surface cyclogenesis from

convectively driven amplification of midlevel mesoscale con-

vective vortices. Mon. Wea. Rev., 129, 605–637.

Ruprecht, E., andW.M. Gray, 1976: Analysis of satellite-observed

tropical cloud clusters. I. Wind and dynamic fields. Tellus, 28,

391–413.

Schubert, W. H., and J. J. Hack, 1982: Inertial stability and

tropical cyclone development. J. Atmos. Sci., 39, 1687–

1697.

Schumacher, A. B., M. DeMaria, and J. A. Knaff, 2009: Objective

estimation of the 24-h probability of tropical cyclone forma-

tion. Wea. Forecasting, 24, 456–471.

Simpson, J., M. Garstang, E. J. Zipser, and G. A. Dean, 1967: A

study of a non-deepening tropical disturbance. J. Appl. Me-

teor., 6, 237–254.

——, E. Ritchie, G. J. Holland, and J. Halverson, 1997: Mesoscale

interactions in tropical cyclone genesis. Mon. Wea. Rev., 125,

2643–2661.

——, J. B. Halverson, B. S. Ferrier, W. A. Petersen, R. H. Simpson,

R. Blakeslee, and S. L. Durden, 1998: On the role of ‘‘hot

towers’’ in tropical cyclone formation. Meteor. Atmos. Phys.,

67, 15–35.

Simpson, R. H., N. Frank, D. Shideler, and H. M. Johnson, 1968:

Atlantic tropical disturbances, 1967.Mon.Wea.Rev., 96, 251–259.

Thorncroft, C., and K. Hodges, 2001: African easterly wave vari-

ability and its relationship to Atlantic tropical cyclone activity.

J. Climate, 14, 1166–1179.Tory, K., and W. M. Frank, 2010: Tropical cyclone formation.

Global Perspectives on Tropical Cyclones: From Science to

Mitigation, J. C. L. Chan and J. D. Kepert, Eds., World Sci-

entific Series on Asia-Pacific Weather and Climate, Vol. 4,

World Scientific, 55–90.

Wang, Z., M. T. Montgomery, and T. J. Dunkerton, 2009: A dy-

namically-based method for forecasting tropical cyclogenesis

location in the Atlantic sector using global model products.

Geophys. Res. Lett., 36, L03801, doi:10.1029/2008GL035586.

——, ——, and C. Fritz, 2012: A first look at the structure of the

wave pouch during the 2009 PREDICT–GRIP dry runs over

the Atlantic. Mon. Wea. Rev., 140, 1144–1163.

Williams, M., and R. A. Houze, 1987: Satellite-observed charac-

teristics of winter monsoon cloud clusters. Mon. Wea. Rev.,

115, 505–519.

Zehr, R. M., 1992: Tropical cyclogenesis in the western North

Pacific. NOAA Tech. Rep. NESDIS 61, 181 pp.

Zipser, E. J., 1977: Mesoscale and convective-scale downdrafts as

distinct components of squall-line circulation.Mon.Wea. Rev.,

105, 1568–1589.


Documents

Cloud Clusters and Tropical Cyclogenesis: Developing and … · 2014. 11. 7. · Cloud Clusters and Tropical Cyclogenesis: Developing and Nondeveloping Systems and Their Large-Scale