Cluster Analysis of Western North Pacific Tropical Cyclone Tracks

1
Cluster Analysis of Western North Pacific
Tropical Cyclone Tracks

• Suzana J. Camargo, Andrew W. Robertson,
• International Research Institute for Climate
  Prediction,
• Columbia Earth Institute, Palisades, NY
• Scott J. Gaffney and Padhraic Smyth
• Department of Information and Computer Science,
• University of California, Irvine, CA

2
Introduction

• Identify different track types, their seasonality
  and relation with large-scale circulation and
  ENSO.
• Importance different track types have landfall
  in different regions.
• New clustering technique used.
• Best track dataset JTWC 1950-2002.
• Only tropical cyclones (TCs) with tropical storm
  or typhoon intensity (no tropical depressions).

3
Clustering Technique

• Developed by S.J. Gaffney and P. Smyth
• - S.J. Gaffney and P. Smyth (1999).
• - S.J. Gaffney (2004), Ph.D. thesis, University
  of California, Irvine.
• Mixture of polynomial regression models (curves)
  to fit the geographical shape of the
  trajectories.
• Extension of the standard multivariate finite
  mixture model to allow quadratic functions.
• Enable highly non-Gaussian density functions to
  be expressed as a mixture of a few PDFs.
• Fitting by maximizing the likelihood of the
  parameters.

4
Clustering Technique (Cont.)

• Rigorous probabilistic context for clustering.
• Accommodate easily tropical cyclone tracks of
  different lengths.
• Previous cluster papers k-means method for the
  Western North Pacific TC tracks
• P.A. Harr and R.L. Elsberry, Monthly Weather
  Review 123, 1225-1246 (1985).
• J.B. Elsner and K.B. Liu, Climate Research 25,
  43-54 (2003).

5
Results

• Appropriate number of clusters appears to be six.
• Quantitative (out of sample likelihood) and
  subjective analysis.
• Two main trajectory-types straight-movers and
  recurvers.
• Additional clusters detailed differences in
  shape among these types.

6
ALL TRACKS in EACH CLUSTER
7
Cluster A

• Tend to exist and form in the Philippines Sea and
  South China Sea.
• Most tracks straight moving northwestward.
• Almost all TCs make landfall (Philippines, South
  China, Taiwan, or Vietnan).
• Few TCs cross to Indian Ocean.
• TCs occur year round, with a peak in July to
  October (maximum in September, minimum in March).
• 24 of TCs.

Tracks JJASON 1971-1990
8
Cluster B

• Genesis north and east of the Philippines.
• Dominant tracks are recurving.
• Many make landfall in China, Taiwan, Korea and
  Japan.
• Occurrence annual cycle similar to cluster A.
• 22 of TCs.

Tracks JJASON 1971-1990
9
Interannual correlations NTC per cluster with
total NTC
NTC data1961-2002
10
Interannual correlations NTC among clusters
11
First Position Density per cluster
12
Track Density by Cluster
13
NTC different Intensities
14
ACE per year
15
Interannual correlations NTC with NINO3.4 JASO
Data 1971-2002
16
Tracks Cluster E - ENSO years
17
Track DensityDifference El Niño and La Niña
Cluster B
Cluster E
18
NTC and ACE - ENSO
19
SST anomalies composites
20
Summary

• New clustering technique applied to WNP TC
  tracks.
• Clusters with different properties genesis and
  track regions, intensity, timing.
• Two clusters are strongly related to ENSO
  reflects well known shift in TC activity in the
  region.
• Composites of large scale fields with different
  characteristics for different clusters.

21
Extra slides
22
(No Transcript)
23
NTC per month in each cluster
24
TC Maximum Wind Speed Distribution per cluster
25
ACE per Year and per TC
26
Correlations ACE
27
TC days and TC life time
28
TC Life time per cluster