Analysis of Cyclone Tracks of North Indian Ocean Using Cluster Analysis - Monitoring and Prediction of Tropical Cyclones in the Indian Ocean and Climate Change - page 88

et al., 2000; Ali et al., 2007; Tyagi et al., 2010). Mohanty (1994) has examined
climatic characteristics of tropical cyclones forming in the Bay of Bengal for
the period 1877-1976. Tyagi et al. (2010) have identified various patterns of
interannual variability in frequency of cyclonic disturbances crossing maritime
states of India and their relation with large scale parameters. Changes in the
frequency of tropical cyclones developing over the Arabian Sea and the Bay of
Bengal have also been studied by Singh et al., 2000). Niyas et al. (2009) have
reviewed the studies examining variability and trends in the tropical cyclones
forming over the NIO. There is, however, no study to date that has analyzed
tropical cyclone tracks in the NIO, in terms of their spatio-temporal
characteristics.
Studies of tropical cyclones tracks using cluster analysis typically may be
grouped into those adopting K-means clustering techniques (Blender et al.,
1997; Elsner and Liu, 2003; Elsner, 2003; Nakamura et al., 2009) and those
using regression mixture models (Gaffney, 2007; Camargo et al., 2007).
Considering the lack of previous studies of the North Indian Ocean basin, in
this paper we have analyzed cyclone tracks using the second approach of
regression mixture model to ensure that the cluster identification is robust,
both with regard to the number of clusters and their properties.
3. Data
The data of tropical cyclonic disturbances over the NIO is available from
Regional Specialized Meteorological Centre - Tropical Cyclones, Cyclone
Warning Division, India Meteorological Department (IMD), India. The best
track data (operationally finalized tracks) from 1877 onwards have been
compiled and published in the form of an atlas by IMD. Mohapatra et al. (2012)
suggest that the reliability of data from 1961 is greater due to improvements in
observational capability, including satellite observations. For this reason, we
consider events during 1961-2010 in this study. Further, we consider only those
events which have reached the cyclone intensity (wind speed > 33 knots). A
total of 237 events satisfy the above mentioned criteria and have been used for
the analysis in this study.
4. Methodology
The regression mixture modelling framework (Gaffney, 2004) consists of
regressing physical location (latitude and longitude) against an index that
corresponds to sequential positions of the storm centre. In the literature these
sequential positions are typically from the best track data where they are evenly
spaced in time (6 hours). Thus, this index is also referred to by Camargo et al.
(2007) as a discrete time index, there being a one to one correspondence between
the index and the real time. However, for the data set used in this study, storm
center locations are irregularly spaced in time, and thus there is no exact
correspondence between the index and the real time. For this reason, we have