Cluster Analysis of Typhoon Tracks. Part II: Large-Scale Circulation and ENSO
SUZANA J. CAMARGO AND ANDREW W. ROBERTSON
International Research Institute for Climate and Society, The Earth Institute at Columbia University, Palisades, New York
SCOTT J. GAFFNEY AND PADHRAIC SMYTH
Department of Computer Science, University of California, Irvine, Irvine, California
MICHAEL GHIL*
Department of Atmospheric and Oceanic Sciences, and Institute for Geophysics and Planetary Physics, University of California,
Los Angeles, Los Angeles, California
(Manuscript received 6 January 2006, in final form 15 November 2006)
ABSTRACT
A new probabilistic clustering method, based on a regression mixture model, is used to describe tropical
cyclone (TC) propagation in the western North Pacific (WNP). Seven clusters were obtained and described
in Part I of this two-part study. In Part II, the present paper, the large-scale patterns of atmospheric
circulation and sea surface temperature associated with each of the clusters are investigated, as well as
associations with the phase of the El Niño–Southern Oscillation (ENSO). Composite wind field maps over
the WNP provide a physically consistent picture of each TC type, and of its seasonality. Anomalous vorticity
and outgoing longwave radiation indicate changes in the monsoon trough associated with different types of
TC genesis and trajectory. The steering winds at 500 hPa are more zonal in the straight-moving clusters, with
larger meridional components in the recurving ones. Higher values of vertical wind shear in the midlatitudes
also accompany the straight-moving tracks, compared to the recurving ones.
The influence of ENSO on TC activity over the WNP is clearly discerned in specific clusters. Two of the
seven clusters are typical of El Niño events; their genesis locations are shifted southeastward and they are
more intense. The largest cluster is recurving, located northwestward, and occurs more often during La Niña
events. Two types of recurving and one of straight-moving tracks occur preferentially when the Madden–
Julian oscillation is active over the WNP region.
1. Introduction
In the first part of this study (Camargo et al. 2007,
hereafter Part I), a new probabilistic clustering meth-
odology based on a regression mixture model was ap-
plied to the best-track dataset of the Joint Typhoon
Warning Center (2005) for 1950–2002. A set of seven
cyclone-track clusters were identified over the western
North Pacific (WNP), three distinct clusters of straight-
moving tracks and four of recurving tracks, with distinct
landfall probabilities. Intensity and seasonality of the
tropical cyclones (TCs), though unknown to the clus-
tering model, were both found to be highly stratified by
the clusters.
The genesis, trajectories, and other characteristics of
TCs over the WNP are known to be strongly affected
by the large-scale environment and by the El Niño–
Southern Oscillation (ENSO). Indeed, previous studies
have capitalized on the relationship between TCs and
large-scale circulation anomalies to define modes of TC
variability (Harr and Elsberry 1995a). In Part II of this
study, we analyze associations between the clusters
identified in Part I, the large-scale atmospheric circula-
tion anomalies, and ENSO. Our goals are 1) to deter-
mine the physical consistency of the TC clusters; 2) to
* Additional affiliation: Département Terre-Atmosphère-
Océan, and Laboratoire de Météorologie Dynamique (CNRS and
IPSL) Ecole Normale Supérieure, Paris, France.
Corresponding author address: Dr. Suzana J. Camargo, Inter-
national Research Institute for Climate and Society, Monell 225,
61 Route 9W, Palisades, NY 10964-8000.
E-mail: suzana@iri.columbia.edu
3654 JOURNAL OF CLIMATE VOLUME 20
DOI: 10.1175/JCLI4203.1
© 2007 American Meteorological Society
JCLI4203
characterize the patterns of atmospheric variability
over the WNP that accompany different modes of TC
behavior; and 3) to isolate potentially predictable com-
ponents associated with these modes and ENSO.
During summer and early fall, monsoonal southwest-
erly winds extend eastward at low levels from Asia and
carry moisture into the equatorial WNP. These wester-
lies meet moist low-latitude easterlies over the western
North Pacific and form a confluence zone that is con-
ducive to convection (e.g., Holland 1995) and TC for-
mation (Briegel and Frank 1997; Lander 1996; Ritchie
and Holland 1999). Easterly waves do not penetrate
very far into the monsoonal westerlies: they tend to
slow down and turn poleward, leading to accumulation
of wave energy in the confluence region (Holland
1995). This interaction can potentially lead to the trans-
formation of these waves into cyclones. Another
mechanism of TC formation is the preconditioning of
the environment in the wake of a previous cyclone
(Frank 1982; Holland 1995; Sobel and Bretherton 1999;
Kuo et al. 2001; Li et al. 2003).
The waves affecting the WNP span a wide range of
spatial and temporal scales (Holland 1995; Sobel and
Bretherton 1999; Kuo et al. 2001; Li et al. 2003). They
include the large-scale, eastward-moving, 30–60-day
Madden–Julian oscillation (MJO; Madden and Julian
1972); the convectively coupled N ⫽ 1 equatorial
Rossby waves (Wheeler and Kiladis 1999); westward-
moving easterly waves that propagate in the trade wind
belt of the central Pacific (Reed and Recker 1971; Lau
and Lau 1990) and have a characteristic period of 6–9
days; and mixed Rossby–gravity equatorial waves, with
a spectral peak at 4–5 days (Itoh and Ghil 1988; Lieb-
mann and Hendon 1990; Dickinson and Molinari 2002).
The monsoonal flow often exhibits a life cycle of sev-
eral weeks (Holland 1995), connected to the MJO,
which may thus cause TC tracks to vary intraseasonally
as well.
Another important feature of the low-level circula-
tion of the WNP is the subtropical anticyclone, which
separates the near-equatorial confluence zone from the
midlatitude westerlies. It is well known that TCs usually
do not move into the center of the subtropical high, so
the latter’s location and intensity will influence TC tra-
jectories (Harr and Elsberry 1995a). In the upper tro-
posphere, the anticyclone is more diffuse (Holland
1995), with extensive equatorial easterlies. Tropical up-
per-tropospheric troughs (TUTTs) can cause changes
in cyclone development and tracks (e.g., Sadler 1978;
Montgomery and Farrell 1993; Ferreira and Schubert
1999). Sadler (1978) identified the presence of a TUTT
as conducive to cyclogenesis in the WNP, while Fer-
reira and Schubert (1999) discussed the formation of
TUTT cells east of the TCs.
The relationship of ENSO and the total number of
cyclones in the WNP has been investigated by several
authors (e.g., Atkinson 1977; Chan 1985; Lander 1994a;
Chan 2000) with varying results, due to differences in
data and techniques used in these studies. In a recent
study, Wang and Chan (2002) observed an increase in
the number of TCs in that basin during strong El Niño
events.
Other studies have identified an important ENSO
impact on TC genesis location over the WNP (Pan
1982; Chan 1985; Chen et al. 1998; Wang and Chan
2002; Chia and Ropelewski 2002), with a displacement
of the mean cyclone genesis region to the southeast
(northwest) during El Niño (EN) [La Niña (LN)] years,
respectively. This displacement is known to lead to dif-
ferent track characteristics as well. There is an en-
hanced tendency of the cyclones in EN years to recurve
northeastward and reach more northward latitudes
(Wang and Chan 2002). In EN years the cyclones also
have a tendency to have longer lifetimes than during
LN years (Wang and Chan 2002), to be more intense
(Camargo and Sobel 2005), and to form in greater num-
bers over the central Pacific region (Chu and Wang
1997; Clark and Chu 2002); some of the latter then
move into the WNP. Wu and Wang (2004) discussed
the anomalous large-scale winds associated with differ-
ent ENSO phases, consistent with the different track
types. Sobel and Camargo (2005) argued that WNP cy-
clones could play an active role in the development of
EN events.
In this paper, we stratify the influence of ENSO
phases on cyclones by examining the interannual vari-
ability of each TC cluster’s frequency of occurrence.
We show that the ENSO dependency is specific to par-
ticular clusters, thus confirming and expanding Wang
and Chan’s (2002) results, and that the EN track types
that dominate in EN versus LN years are well separated
from the others. The fact that different track types pre-
vail during warm versus cold events is important be-
cause it leads to different regions of landfall. Differ-
ences in landfall patterns in different regions of Asia
during ENSO cycles have been discussed in several
studies (e.g., Saunders et al. 2000; Elsner and Liu 2003;
Wu et al. 2004). Here we show the anomalous large-
scale atmospheric circulation patterns associated with
the landfall distributions constructed for each cluster,
including relationships with ENSO.
While ENSO strongly influences the interannual
variability of TC tracks in the WNP, other factors may
also be important, such as the MJO (Liebmann et al.
1994; Sobel and Maloney 2000), stratospheric quasi-
15 JULY 2007 CAMARGO ET AL. 3655
biennial oscillation (Chan 1995), TUTTs (Sadler 1978),
and associations with the Indian summer monsoon (Ku-
mar and Krishnan 2005). Of these, we discuss here
MJO effects. Variability on decadal or longer time
scales in typhoon activity, tracks, and landfall have
been discussed in various studies (Chan and Shi 2000;
Chu 2002; Ho et al. 2004) but are not considered in the
present analysis.
The datasets and methods are described briefly in
section 2. In section 3 we construct composites of at-
mospheric circulation patterns and sea surface tem-
perature (SST) fields for each cluster. The relationship
with ENSO and MJO is investigated in section 4. A
summary and a discussion are presented in section 5.
2. Data and methods
The cyclone data are based on the Joint Typhoon
Warning Center (2007) best-track dataset for the time
interval 1950–2002, as used in Part I. Tropical depres-
sions are present in this dataset, but only TCs with
tropical storm intensity or higher are included in our
analysis. The clustering methodology was applied to a
total of 1393 TC tracks, as described in detail in Part I.
Further methodological details can be found in Gaffney
(2004) and Gaffney et al. (2007), where the method was
applied to North Atlantic extratropical cyclones.
Each track is assigned to one of seven clusters. To
analyze the large-scale environment associated with
each cluster, we use several standard datasets: weekly
SST fields, available from November 1981 (Reynolds et
al. 2002), daily National Oceanographic and Atmo-
spheric Administration (NOAA) outgoing longwave
radiation (OLR) available continuously since 1979
(Liebmann and Smith 1996), and several meteorologi-
cal variables from the National Centers for Environ-
mental Prediction–National Center for Atmospheric
Research (NCEP–NCAR) reanalysis (Kalnay et al.
1996), since 1950. In each case, we use the entire data
record available to us. Anomalies are defined relative
to the 1971–2000 climatology for the reanalysis and
Reynolds datasets, and relative to 1979–2002 for
NOAA OLR. Daily climatologies, obtained using 3-day
running means, are used for atmospheric variables, and
weekly values for SST.
The statistical significance of the resulting compos-
ites was determined using a Monte Carlo test, described
in the appendix. The results were verified in several
selected cases using a hypergeometric statistical test, as
described in Mason and Goddard (2001), which was
found to give very similar results at the 95% signifi-
cance level.
The monthly SST index for the Niño-3.4 area (5°S–
5°N; 170°–120°W; Barnston et al. 1997) was used to
define the phase of ENSO, obtained from the Climate
Prediction Center (CPC) Web site (CPC 2007). We de-
fine EN and LN years according to the value of the
Niño-3.4 index averaged over the months of July–
October (JASO), spanning the peak of the typhoon
season. The 13 yr (approximately 25% of the 53-yr pe-
riod) with the largest and smallest values of Niño-3.4 in
the period 1950–2002 are defined as EN and LN years,
respectively; the remaining 27 yr are classified as neu-
tral years. The EN and LN years defined using our
percentile method correspond to the Northern Hemi-
sphere summers before the peaks of the ENSO events
and correspond well to the ENSO events obtained us-
ing more traditional definitions; see for instance, Tren-
berth (1997), as discussed by Goddard and Dilley
(2005).
3. TC clusters and the large-scale circulation
We constructed composites of large-scale circulation
patterns associated with each cluster by compositing
the days that correspond to either the first position or
the maximum intensity of each cyclone. The results are
very similar in either case, so we show here only the
composites for the first position. The rationale for
choosing the first position is the potential usage of these
patterns in tracks and landfall forecasts, once the gen-
esis position of a cyclone is known. Besides, the large-
scale circulation patterns that control genesis and track
type are not independent from each other, since they
are associated with different positions and intensities of
the monsoon trough and subtropical ridge (Harr and
Elsberry 1991). In the case of the steering winds, the
composites were constructed using all the days during
which a TC was active. The number of TCs included in
each composite varies by cluster, and according to the
length of the available dataset for each variable (OLR,
SST, wind fields, etc).
a. Steering winds
The movement of cyclones is largely determined by
the ambient flow, the so-called large-scale steering
winds. The midtropospheric levels (700 and 500 hPa)
have been found to have the best correlation with TC
direction and speed (Chan and Gray 1982). The steer-
ing is due to the advection of concentrated relative vor-
ticity associated with the cyclone by the large-scale en-
vironmental winds (Wu and Wang 2004). A highly ide-
alized form of this steering can be found in the motion
of point or Rankine vortices in a two-dimensional ve-
locity field (von Helmholtz 1858; Aref 1983; Ide and Ghil
3656 JOURNAL OF CLIMATE VOLUME 20
1997a,b). Other fundamental contributions to cyclone
motion are the “beta drift,” due to the advection of
planetary vorticity by the TC circulation, and nonlinear
interactions with the large-scale winds and planetary-
vorticity gradient (Fioriono and Elsberry 1989; Wang
and Holland 1996). These contributions tend to cause a
net drift to the northwest, relative to the steering winds,
especially in the case of weak ambient flows (Wang and
Li 1992; Wang and Holland 1996; Carr and Elsberry
1990; Franklin et al. 1996). Various theoretical and ob-
servational studies (e.g., Emanuel 1991; Wu and Eman-
uel 1995; Chan 2005) have also discussed the effects of
a nonuniform background, interaction between cy-
clones and the monsoonal flow, and other issues.
We show in Figs. 1 and 2 wind and anomalous winds
composites at 500 hPa, over all days of each trajectory;
the patterns for 700 hPa (not shown) are very similar. In
each cluster the mean regression trajectory is clearly
consistent with the composite wind fields, indicating the
dominating influence on the former. In all the clusters,
the northwestward drift is evident, either throughout
the trajectory (for the straight movers in clusters B, D,
FIG. 1. Daily wind (500 hPa) composites, for the cyclones (TCs) in each cluster and for all TCs during the period
1950–2002. Shaded regions are statistically significant. The composites are based on all days during which a TC was
active. Also shown are the mean regression trajectory (black line) and genesis location (black asterisk) for each
cluster.
15 J
ULY 2007 CAMARGO ET AL. 3657
F, and G) or at least during its early part (for the re-
curvers in clusters A, C, and E).
The recurving clusters (A, C, and E) are generated in
a region of southeasterly flow, with the recurving oc-
curring at the latitudes between 20° and 30°N, where
the trade easterlies change to subtropical westerlies
that are not reached by the straight movers. Although
there is a good deal of within-cluster spread about the
mean regression trajectory (Fig. 5 of Part I), the latter
have similar shapes, and the main difference between
these three clusters is largely in the different mean gen-
esis locations.
The reasons why the straight movers hit land (clus-
ters B and D) or decay in the latitudes of low surface
winds, near 20°N (clusters F and G), are more complex
and involve genesis position and less northward drift in
their early stages. The genesis location of cluster B, a
straight-moving cluster, is close to the Asian coast, with
FIG. 2. Composites of daily anomalous wind fields at 500 hPa for the cyclones in each cluster and for all TCs
during the period 1950–2002. Also shown are the mean regression trajectory (gray line) and genesis location (gray
asterisk) for each cluster.
3658 JOURNAL OF CLIMATE VOLUME 20