The influences of ENSO and the subtropical Indian Ocean Dipole on tropical cyclone trajectories in the southwestern Indian Ocean

Authors

  • Kevin D. Ash,

    Corresponding author
    1. Hazards and Vulnerability Research Institute, Department of Geography, University of South Carolina, 709 Bull Street, Columbia, SC 29208, USA
    • Hazards and Vulnerability Research Institute, Department of Geography, University of South Carolina, 709 Bull Street, Columbia, SC 29208, USA.
    Search for more papers by this author
  • Corene J. Matyas

    1. Department of Geography, University of Florida, 3141 Turlington Hall, Gainesville, FL 32611, USA
    Search for more papers by this author

Abstract

El Niño-Southern Oscillation (ENSO) is known to associate with variability of tropical cyclone (TC) trajectories in the southwestern Indian Ocean. However, consideration of ENSO phase alone does not account for all variability of TC tracks within this region. This study demonstrates that the subtropical Indian Ocean Dipole (SIOD) sea-surface temperature anomaly pattern is also significantly associated with variability in southwestern Indian Ocean TC tracks. Hierarchical cluster analysis is employed to group TC trajectories by their initial and final positions. Median monthly values of the Niño-3.4 index and Subtropical Dipole Index corresponding to the life cycles of TCs in each group are compared using non-parametric analysis of variance. The results suggest that both ENSO and SIOD are significantly associated with different types of southwestern Indian Ocean TC trajectories. Furthermore, significant interactions of ENSO and SIOD phases are found to influence certain types of TC tracks using contingency table tests. During simultaneous warm ENSO and negative SIOD phases, TCs moving across the southwestern Indian Ocean tend to follow more southward or southeastward tracks. During neutral or cool ENSO and positive SIOD phases, TCs moving through the southwestern Indian Ocean tend toward more westward trajectories. These findings suggest that use of an SIOD index in addition to an ENSO index could improve intraseasonal to seasonal statistical prediction of southwestern Indian basin TC activity. Copyright © 2010 Royal Meteorological Society

1. Introduction

Tropical cyclones (TCs) are a recurring phenomenon in the southwestern Indian Ocean, most frequently during TC season from November to April, and the inhabited regions of the western rim of the basin are particularly prone to repeated TC impacts. TCs frequently pass near the small islands of Mauritius and Réunion and can cause severe damage and societal disruption, though loss of human life is typically mitigated by the well-organised and executed TC preparedness and warning systems of these islands (Parker, 1999; Roux et al., 2004). Madagascar has endured many devastating TC impacts (Jury et al., 1993; Naeraa and Jury, 1998; Chang-Seng and Jury, 2010b), and the population's social vulnerability to negative impacts from TC strikes is increased due to widespread livelihood dependence on agriculture and the lesser economic status of the country (Brown, 2009). Mozambique is, likewise, in a precarious economic and climatological position, though TC landfalls are not as frequent as in Madagascar (Vitart et al., 2003; Reason and Keibel, 2004; Reason, 2007; Klinman and Reason, 2008; Silva et al., 2010). TC impacts also extend over the open ocean through disruption of busy shipping lanes across the entire southern Indian Ocean, threatening the lives of vessel crews and their valuable cargoes as they steam between maritime economic hubs in Europe, the Americas, South Asia, and the Far East (Chang-Seng and Jury, 2010b).

Large scale, low-frequency modes of ocean-atmosphere variability related to TC genesis in the southern Indian Ocean have been studied previously and are relatively well understood. The El Niño-Southern Oscillation (ENSO), the Madden-Julian Oscillation, and the Indian Ocean Dipole, or Zonal Mode, are all known to influence the spatial pattern and frequency of TC genesis in the southern Indian Ocean (Kuleshov and de Hoedt, 2003; Bessafi and Wheeler, 2006; Ho et al., 2006; Camargo et al., 2007a; Kuleshov et al., 2008; Leroy and Wheeler, 2008; Chan and Liu, 2009; Kuleshov et al., 2009; Vitart et al., 2010). However, given that TCs forming in the central or southeastern Indian Ocean may traverse thousands of kilometers to the west to impact the aforementioned countries on the western fringe of the basin, it is also important to improve understanding of TC trajectory variability in relation to ocean-atmosphere variability at intraseasonal to seasonal time scales. Less research has specifically focused on the topic of intraseasonal or seasonal southwestern Indian Ocean TC track variability, save for the work of Vitart et al. (2003) which suggested an ENSO link to TC track variability in the region. Outside the body of southern Indian Ocean TC research, ENSO, the subtropical Indian Ocean Dipole (SIOD), and tropical temperate troughs (TTTs) have all been shown to play important roles in air–sea interactions within the southwestern Indian Ocean which could influence TC trajectory variability on intraseasonal to seasonal time scales. In this study, we specifically consider the influence of ENSO and SIOD on southwestern Indian Ocean TC tracks, and use the previously established TTT framework for a broader understanding of the mechanism by which ENSO and SIOD may alter TC track patterns in this region.

It has been demonstrated in each of the world's principal TC regions that ENSO can alter the global or regional atmospheric circulation to either change the spatial patterns of TC genesis or affect their directions of movement (Gray and Sheaffer, 1991; Landsea, 2000; Chu, 2004; Camargo et al., 2010). The importance of ENSO influence on the coupled ocean-atmosphere system in the southern Indian Ocean alone is very well documented (van Loon and Rogers, 1981; Pan and Oort, 1983; Meehl, 1987; Karoly, 1989; Nicholson, 1997; Chambers et al., 1999; Klein et al., 1999; Reason et al., 2000; Venzke et al., 2000; Larkin and Harrison, 2002; Hermes and Reason, 2005; Yoo et al., 2006; Huang and Shukla, 2007a). Just as in the northern and southern Pacific Oceans, there is not a significant basin-wide correlation in the southern Indian Ocean between ENSO and TC frequency (Jury, 1993); instead it is the spatial patterns of genesis and directions of movement that exhibit links to ENSO (Revell and Goulter, 1986; Lander, 1994). ENSO has been demonstrated as a significant predictor of southern Indian TC activity at weekly to monthly time scales and is now utilised in operational statistical prediction of TC activity in the region (Leroy and Wheeler, 2008; Vitart et al., 2010).

During El Niño, TC genesis is more frequent over the southwestern Indian Ocean (west of about 75°E–85°E) than in the eastern ocean (Evans and Allan, 1992; Kuleshov and de Hoedt, 2003; Ho et al., 2006; Kuleshov et al., 2008). Favourable conditions for deep convection in the western ocean are generated following the occurrence of easterly lower tropospheric wind anomalies south and west of Sumatra during early austral spring (late September–November), which excite a westward-propagating oceanic Rossby wave that, coupled with increased poleward Ekman transport associated with the easterly anomalies in the tropical eastern part of the basin, induce an unusually deep pool of warm sea surface temperature anomalies (SSTA) centered near 15°S, 60°E (Chambers et al., 1999; Jury et al., 1999; Reason et al., 2000; Xie et al., 2002; Jury and Huang, 2004). Cooler SSTA in the southeastern Indian Ocean and anticyclonic vorticity from the lower tropospheric easterly anomalies account for a reduction of TC genesis, most notably in late austral spring and early summer (Ho et al., 2006; Camargo et al., 2007a; Kuleshov et al., 2008; Kuleshov et al., 2009). Conversely, during an ENSO cool event, the western portion of the basin is characterised by reduced TC genesis while the eastern (east of about 75°E–85°E) experiences more, with positive SSTA, increased mid-tropospheric relative humidity, and increased lower tropospheric cyclonic vorticity contributing to an environment conducive to more frequent TC genesis (Wolter, 1987; Ho et al., 2006; Camargo et al., 2007a; Kuleshov et al., 2008; Chan and Liu, 2009; Kuleshov et al., 2009).

As for the tracks of southern Indian TCs in relation to ENSO, Vitart et al. (2003) note that zonal steering flow averaged over 850–200 hPa across the tropical and subtropical portions of the basin is more westerly (easterly) during El Niño (La Niña). Consequently, Mozambique should be at greater risk for landfall during La Niña, whereas westerly steering flows are suggestive of increased incidences of re-curving TCs just east of Madagascar during El Niño (Jury and Pathack, 1991; Vitart et al., 2003; Reason and Keibel, 2004; Ho et al., 2006; Camargo et al., 2007a; Kuleshov et al., 2009). Chang-Seng and Jury (2010a) agree that La Niña is associated with increased frequency of longer-lived and more intense TCs in the southwestern Indian basin. However, though ENSO is the most important source of climate variability in the global tropics and subtropics, it does not account for all variability in the global tropical and subtropical ocean–atmosphere system, nor does it account in totality for the variability of trajectories amongst southwestern Indian TCs (Fauchereau et al., 2003; Vitart et al., 2003; Huang and Shukla, 2007b; Klinman and Reason, 2008).

During El Niño in austral summer, tropical temperate troughs (TTTs) influence atmospheric conditions in the TC regions of the southwestern Indian Ocean (Harangozo and Harrison, 1983; Lyons, 1991; van den Heever et al., 1997; Washington and Todd, 1999). The TTTs frequently shift 30°–35° eastward and collocate with the deep warm pool over the tropical western ocean (Lindesay et al., 1986; Mason and Jury, 1997; Cook, 2000; Tyson and Preston-Whyte, 2000; Nicholson, 2003; Fauchereau et al., 2009; Pohl et al., 2009; Manhique et al., 2011). This eastward shift of the regional tropical convective maximum and accompanying poleward outflow is consistent with the non-linear El Niño teleconnection framework of Hoerling et al. (1997), and places within a broader synoptic context the positive zonal wind anomalies observed in previous studies of the tropical and subtropical southwestern Indian basin (Reason et al., 2000; Yoo et al., 2006). Heretofore, the ENSO-TTT framework has not been applied to the southwestern Indian TC genesis or track research, as only Mavume et al. (2009) briefly mention TTTs as potentially important in relation to TC tracks within the Mozambique Channel.

The SIOD has also been identified as an important source of ocean–atmosphere variability in the southern Indian Ocean. It is represented by the second empirical orthogonal function of southern Indian Ocean tropical and subtropical SSTA (Behera et al., 2000; Behera and Yamagata, 2001; Qian et al., 2002; Suzuki et al., 2004; Huang and Shukla, 2007b), and SIOD phases have been suggested to exhibit variations independent of the simultaneous ENSO phase (Behera and Yamagata, 2001; Fauchereau et al., 2003; Washington and Preston, 2006; Huang and Shukla, 2007b). The pattern is most prominent during the austral warm season and is characterised in the positive (negative) mode by cool (warm) SSTA in the southeastern Indian Ocean, whereas the southwestern Indian Ocean is simultaneously warm (cool). The mechanisms for cooling in the eastern pole during a positive mode are strengthened southeasterly trade winds and resultant enhanced ocean surface evaporation and mixing, while the western warm pole develops concurrently with increased poleward Ekman transport of warm SSTs from tropical latitudes combined with decreased equatorward cold air advection and ocean surface evaporation (Reason, 1999; Venzke et al., 2000; Behera and Yamagata, 2001; Qian et al., 2002; Hermes and Reason, 2005; Chiodi and Harrison, 2007; Huang and Shukla, 2007b). In negative mode, the SSTA poles are generally reversed with cold air advection and equatorward Ekman pumping occurring over the southwestern Indian Ocean in conjunction with more frequent cold frontal passages, while warm air advection is located more frequently over the southeastern part of the basin.

There is evidence in the literature to suggest that eastward shifts of the African TTTs are not only associated with ENSO, but also with the negative phase of the SIOD. Reason (2002) and Fauchereau et al. (2009) noted the strong similarities between the positive SIOD SST and wind anomaly patterns, and the atmospheric anomalies associated with TTTs over southern Africa. This is important because it raises the possibility of local SSTA influence in the variability of TTTs, which have often been attributed to the influence from ENSO. During a negative SIOD phase, a persistent eastward shift of TTTs over the southwestern Indian Ocean should influence TCs to follow more re-curving poleward or even eastward tracks (Parker and Jury, 1999; Chang-Seng and Jury, 2010b). Therefore, if negative SIOD is associated with eastward phase shifting of TTTs, and if SIOD phases can exhibit variations independent of the simultaneous ENSO phase (Fauchereau et al., 2003; Washington and Preston, 2006; Huang and Shukla, 2007b), then an interactive consideration of both ENSO and SIOD is warranted to better account for variability in southwestern Indian Ocean TC trajectories.

In Section 2, we describe the clustering procedure employed to reduce a sample of 191 TCs to 6 types based on their genesis locations within 3 subregions of the southern Indian basin main development region (54°E–110°E) and then further subdivided by their directions of movement from each genesis subregion. We also outline the analysis of variance (ANOVA) tests used to infer differences of association with SIOD and ENSO between the final six TC trajectory groups. Section 3 conveys the clustering procedure results, presents the ANOVA test results, and compares different TC trajectory groups that share common genesis regions. Expanded discussion of the results and potential applications are given in Section 4, and Section 5 summarises the important findings from this study and offers possible directions for future research.

2. Data and methods

2.1. Tropical cyclone data

The TC trajectories in this study are from the Joint Typhoon Warning Center (JTWC) best-track dataset for the Southern Hemisphere (Chu et al., 2002). Because this research focuses on the southwestern Indian Ocean, we consider only those storms which passed west of 90°E longitude during their respective life cycles. The 90th meridian is chosen as the west–east boundary in accordance with the official forecasting areas of responsibility assigned by the World Meteorological Organization (WMO) to Regional Specialized Meteorological Centers (RSMC) La Réunion (west of 90°E) and Perth (east of 90°E) (Caroff, 2009). The JTWC best-track data are preferred over the RSMC La Réunion best-track archive for this study because the La Réunion dataset does not extend as far to the east of 90°E as the JTWC data due to the RSMC administrative boundary. By using the JTWC archive, we include more complete trajectories of TCs that formed east of 90°E, crossed that meridian, and continued westward into the southwestern Indian basin.

Additionally, we considered only TCs that reached a maximum lifetime intensity of at least 30 m s−1 maximum 1-minute sustained wind. This proviso addresses TC data quality concerns through inclusion of stronger and better organised TCs, which are less prone to position-fix errors than weaker, poorly organised tropical systems (Yip et al., 2006). These cautious steps are deemed necessary due to the lack of complete and permanent geostationary satellite coverage over the southern Indian region prior to May 1998 (Kossin et al., 2007; Chang-Seng and Jury, 2010a; Kuleshov et al., 2010). Despite the limited satellite coverage, Knaff and Sampson (2009) note that JTWC TC data are suitable for analysis from about 1980. In keeping with their suggestion, we assembled a sample of 191 TCs from the recent thirty-year period spanning 1979–2008. An equally important consideration of studying TCs from this period is to research principally the ocean–atmosphere interactions operating after the known Indo-Pacific region climate shift of 1976–1977, as spatial patterns of ocean–atmosphere variability associated with ENSO and SIOD interactions are known to differ on multi-decadal time scales (Trenberth, 1990; Zinke et al., 2004; Terray and Dominiak, 2005).

2.2. Clustering procedure

To test for influences from ENSO and SIOD, we first break the 191 trajectories into groups using cluster analysis. Cluster analysis is widely used in geophysical research for classification or data exploration, and a well-structured clustering solution, whether or not it reveals the natural modality of the data, can aid in the discernment of mechanisms that shape differences and similarities of complex events or phenomena (Gong and Richman, 1995; Wilks, 2006). Cluster analysis has been employed recently as a tool to research variability in TC trajectories in the northwestern Pacific (Elsner and Liu, 2003; Camargo et al., 2007b; Choi et al., 2009), the northeastern Pacific (Camargo et al., 2008), and the North Atlantic (Elsner, 2003; Kossin et al., 2010). The present study represents the first known application of cluster analysis to analyse TC track variability in the southwestern Indian Ocean.

We implement a two-stage agglomerative hierarchical clustering procedure using a Euclidean distance measure and the group average linkage method. Agglomerative hierarchical clustering begins with each observation as its own group and then iteratively joins the two closest groups until there is one group that includes every observation (Lattin et al., 2003). The cophenetic correlation may then be used as a diagnostic of the strength of a clustering structure in tandem with a dendrogram (or a tree diagram) that illustrates the structure of linkages carried out by the clustering algorithm. Cophenetic correlation, therefore, indicates how well the visual structure illustrated by the dendrogram reflects real clustering structures in the data. If the cophenetic correlation approaches or exceeds 0.8, visual inspection of the dendrogram may allow for discernment of an appropriate number of clusters by cutting the dendrogram where the average dissimilarity between the groups jumps considerably between grouping iterations (Romesburg, 1984). At both stages of the clustering procedure for this study, the cophenetic correlations exceed 0.7. By this method, the final number of clusters extracted may not always reflect the true modality of the data, but does facilitate analysis and interpretation of the data (Wilks, 2006).

The two-stage clustering solution employed in our study allows for incorporation of known physical mechanisms related to ENSO that influence variability in spatial patterns of southern Indian TC genesis, while also further subdividing the TCs by the directions of their movement. It has been shown previously in observational studies that El Niño (La Niña) is associated with more TCs forming west (east) of about 75°E–85°E (Ho et al., 2006; Camargo et al., 2007a; Kuleshov et al., 2008). This is also theoretically justified in that a favourable pool of warm SSTA and an unusually deep thermocline are present in the southwestern Indian Ocean during strong ENSO warm events (Xie et al., 2002; Jury and Huang, 2004; Camargo et al., 2007a; Kuleshov et al., 2009), while unfavourable negative SSTAs, lower tropospheric easterly anomalies, and elevated mean sea level pressure (MSLP) are present in the southeastern Indian basin (Reason et al., 2000; Larkin and Harrison, 2002). It follows that a stratification of the longitudes of TC genesis points should echo the established regional physical characteristics of ENSO. Thus, the first stage is a univariate clustering considering only the initial longitude of each TC, from which five clusters are extracted.

Qualitative comparison of the TC genesis clusters (Figure 1) with previous research suggests that stage 1 of the analysis approximates the stratification of TC genesis associated with ENSO influences as described above (refer to Figures 3 and 4 from Kuleshov et al., 2008, and Figure 2(c) from Ho et al., 2006). The three large groups in the main development region of the southern Indian Ocean contain 89% of the 191 TCs in the initial cluster analysis. The western (eastern) group is in the favourable genesis area during a warm (cool) phase of ENSO. Whereas Ho et al. (2006) placed the boundary separating the west and east genesis regions at 75°E, Kuleshov et al. (2008) suggested 85°E. Therefore, in stage 1 of the analysis, the central group represents a region of transition where the differing influences of ENSO phases on TC genesis are not well defined. As the three large groups across the main body of the southern Indian basin comprise the majority of TCs in the sample, only these are carried forth for the second stage of clustering.

Figure 1.

Map of five cluster solution of southwestern Indian Ocean tropical cyclones clustered by initial longitude. Only the three main development subregions (western, central, eastern) are considered in subsequent analysis. Note that this study only considers TCs which formed or passed west of 90°E during their life cycles

Figure 2.

Six TC trajectory clusters, arranged according to group size, within the main development regions for the southern Indian Ocean between 54°E and 110°E. a) C1, eastern genesis/southwest–south movement; b) C2, central/west-southwest; c) C3, western/west-southwest; d) C4, western/south-southeast; e) C5, eastern/west; f) C6, central/south-southeast

Figure 3.

Sea surface temperature anomaly composites for the southern Indian Ocean corresponding to the six TC trajectory clusters presented in Figure 2. SDI West (27°S–37°S, 55°E–65°E) and SDI East (18°S–28°S, 90°E–100°E) regions are shown as boxes in each image. Positive (negative) values are symbolised by dark solid (light broken) lines with units in °C

Figure 4.

Sea surface temperature anomaly composites for the equatorial Pacific Ocean corresponding to the six TC trajectory clusters presented in Figure 2. Niño-3.4 region (5°N–5°S, 120°W–170°W) is shown with a box in each image. Positive (negative) values are symbolised by dark solid (light broken) lines with units in °C

The second stage subsequently applies a bivariate cluster analysis within each of the three genesis clusters in the main TC development regions bounded by 54°E and 110°E. Each genesis group is further subdivided into eastward and westward trajectories using the final latitude and longitude of each respective TC. One of the principal goals of this study is to investigate the relationship of ENSO and SIOD to the directions of TC movements. The second stage of clustering therefore is intended to further subdivide TCs within each region that move mostly westward and threaten land from those that move southward or eastward and remain at sea. The two-stage solution, incorporating both the initial and final geographic locations of the TCs, results in six groups of trajectories within the main body of the tropical and subtropical southern Indian Ocean. These are presented in greater detail in Section 3.1.

2.3. Analysis of variance

The six trajectory groups are compared in terms of their median monthly values of SSTA indices representing both ENSO and SIOD through ANOVA. Each TC is assigned the monthly values of the index corresponding to the month(s) spanning each TC's life cycle. If a storm's life cycle bridged two months, the two index values are averaged. For ENSO, five-month running means of standardised Niño-3.4 region anomalies (N3.4, 5°N, 5°S, 170°W, 120°W) are used (data obtained online at http://www.cpc.noaa.gov/data/indices/), and for SIOD the Subtropical Dipole Index (SDI) is used (obtained at http://www.jamstec.go.jp/res/ress/behera/iosdindex.html). SDI is calculated by subtracting the east pole SSTA (18°S–28°S, 90°E–100°E) from the west pole SSTA (27°S–37°S, 55°E–65°E) (Behera and Yamagata, 2001).

A non-parametric rank test, the Kruskal-Wallis (KW) test with ties adjustment (Kruskal and Wallis, 1952; Higgins, 2004), is used to compare the medians of N3.4 and SDI for the six principal trajectory groups. The null hypotheses for the tests are that the median values of N3.4 and SDI are not significantly different across all six TC clusters. The alternative hypotheses are that there exist significant differences in at least one pair of TC groups in their median values of N3.4 or SDI. To ensure the results of the KW tests are appropriately interpreted, Modified-Levene Equal-Variance Tests are also applied (Brown and Forsythe, 1974). Dunn's rank sums procedure is used to test statistical significance in multiple comparisons (Dunn, 1964).

To complement the ANOVA tests, SSTA composites are also constructed for each cluster using the National Oceanic and Atmospheric Administration (NOAA) Optimum Interpolation (OI) SST Version 2 (Reynolds et al., 2002). An SST average is extracted based on the initial and final dates for each respective TC's life cycle. Monthly SST averages are also created based on the period 1981–2008. To obtain SST anomaly composites for each TC, the 26-year monthly composite is subtracted from each respective TC event composite. For example, TC Jaya's life cycle spanned March–April of 2007. Therefore, the SSTA composite map for Jaya is the SST composite for that TC's life cycle minus the combined March–April 26-year SST composite.

Finally, to test the interaction of ENSO and SIOD phases with respect to the TC trajectory types, three contingency tables are constructed stratifying TCs that formed in the same TC genesis subregions by six categories of ENSO and SIOD interactions: ENSO warm event with positive SIOD (E + S+), ENSO warm event with negative SIOD (E + S−), ENSO neutral with positive SIOD (E∼S+), ENSO neutral with negative SIOD (E∼S−), ENSO cool event with positive SIOD (E − S+), and ENSO cool event with negative SIOD (E − S−). Months in which each TC occurred are assigned as ENSO warm, neutral, or cool phases based on a criteria of at least six consecutive months with + /− 0.4 °C anomalies for five-month running means of the standardised N3.4 index. This is a very similar methodology for determining ENSO phase as outlined by Trenberth (1997) and implemented recently in Lau et al. (2008) and Antico (2009). The principal difference is that our base period for standardisation of the N3.4 index is limited to the study period of 1979–2008. Fisher's Exact Test provides exact p-values for contingency table tests, thus it is applied such that robust results may be obtained despite the small sample sizes (Higgins, 2004).

3. Results

3.1. Cluster analysis

The two-stage cluster analysis assigns the 170 main development region TCs into 6 groups of trajectories, according to their initial longitudes and final latitudes and longitudes. They are ranked by the number of TCs within each group, designated simply as C1 through C6 (Table I). Almost 70% of the 170 TCs are contained within the 3 largest groups (C1-C3), and these display a net southwesterly component of movement. This is to be expected given that typical background conditions would allow for such an average motion around the northwest fringes of the southern Indian subtropical anticyclone.

Table I. Six main groups of TC trajectories, ranked by number of TCs in each group. Group median initial and final longitudes are in decimal degrees east, and median final latitudes are in decimal degrees south
Cluster designationNumber of TCsMedian initial longitude (°E)Genesis regionMedian final latitude (°S)Median final longitude (°E)Direction of movement
C14094.3Eastern23.482.4SW/S
C23978.0Central25.256.1W/SW
C33964.3Western26.949.8W/SW
C42266.1Western26.568.0S/SE
C51699.1Eastern18.758.0W
C61481.2Central20.583.6S/SE

The largest group is C1 with 40 TCs. These formed between 87°E and 110°E and followed southwestward and southward trajectories (Figure 2(a)). C2 is comprised of 39 TCs which formed between 73°E and 87°E and moved westward or southwestward (Figure 2(b)). Group C2 threatened land with relative frequency (16 of 39 within 200 km of Madagascar, Mauritius, La Réunion, or Mozambique) while C1 storms remained over the open ocean. The third largest group, C3 (39 TCs), originated between 54°E and 73°E with a large proportion (33 of 39) that moved westward or southwestward within 200 km of inhabited areas (Figure 2(c)).

The remaining 30% of TCs comprise the 3 smaller groups. These represent marked departures from the typical net southwestward movement, and are therefore likely candidates for association with notable departures from the more typical southwestward steering flow regime, such as might occur with strong phases of ENSO or SIOD. C4 (Figure 2(d)), in contrast to C3, developed between 54°E and 73°E with a dominant displacement southward and southeastward. Though the 22 TCs in group C4 formed in close proximity to the inhabited islands in the southwestern Indian Ocean, they seldom passed within 200 km of Madagascar or the Mascarene islands. Having formed between 87°E and 110°E, group C5 (Figure 2(e)) is distinguished from C1 by longer distance tracks with more westward trajectories, 10 of which passed west of 60°E and threatened populated regions. The smallest of the 6 trajectory clusters, C6, formed between 73°E and 87°E (Figure 2(f)). Direction of movement is south and southeastward, keeping these systems well out to sea.

3.2. KW ANOVA

3.2.1. ANOVA results

The clustering procedure using initial longitude and final latitude and longitude allows TC trajectory clusters with common geographic origins and diverging paths to be compared. This is accomplished through ANOVA, where we compare the median values of the N3.4 index and the SDI (Table II) first across all groups and then in multiple pair-wise comparisons. The results of the KW tests are to reject H0 of no difference between the median values of both N3.4 and SDI (α = 0.001, p-values 0.00019 and 0.00012, respectively). Modified Levene Equal-Variance tests are also employed, and H0 of homoscedasticity is not rejected for either N3.4 or SDI (α = 0.05, p-values 0.88 and 0.22, respectively). Thus, there is statistical evidence to support the assertion that SSTs in the N3.4 and SDI regions are significantly and contemporaneously associated with different types of southwestern Indian TC trajectories.

Table II. Median standardised anomalies for Niño-3.4 (N3.4) sea surface temperature and Subtropical Dipole Index (SDI)
Cluster IDNiño-3.4SDI
C1− 0.05− 0.19
C2− 0.04− 0.06
C3− 0.220.34
C40.73− 0.70
C5− 0.370.77
C60.160.14

Several other variables were tested using the same KW ANOVA method. Previous studies suggested a linkage between the tropical Indian Ocean Dipole or Zonal Mode (Saji et al., 1999; Webster et al., 1999) and southern Indian TC activity (Leroy and Wheeler, 2008; Chan and Liu, 2009; Vitart et al., 2010). Klinman and Reason (2008) and Chang-Seng and Jury (2010a) noted that variability of the Southern Hemisphere planetary waves could also affect southwestern Indian TC trajectories. Vitart et al. (2010) include the Trans-Niño Index (Trenberth and Stepaniak, 2001) as a significant predictor of southern Indian TC activity in association with ENSO variability, consistent with recent findings of association between ENSO variability and TC activity in the North Atlantic and North Pacific basins (Kim et al., 2009; Chen and Tam, 2010; Kim et al., 2010). Accordingly, we carried out the same ANOVA procedure using TNI, Dipole Mode Index (DMI) (Saji et al., 1999), and an Antarctic Oscillation Index (AAO) (obtained from NOAA at http://www.cpc.noaa.gov/products/precip/CWlink/daily_ao_index/aao/aao_index.html). We did not find significant differences between the six TC trajectory types based upon these three additional indices (results not shown).

3.2.2. Western region TC types: C3 versus C4

To ascertain exactly which TC types are more significantly associated with phases of ENSO and/or SIOD, Dunn's rank-sum procedure is employed for multiple comparisons of the median values of N3.4 and SDI across the 6 trajectory clusters. Results of the multiple comparisons for N3.4 (Table III) and SDI (Table IV) indicate that TC type C4 is significantly more associated with El Niño and the negative SIOD mode than the other 5 trajectory types. The SSTA composites for C4 corroborate these results. The composite SSTA pattern in the Indian Ocean (Figure 3(d)) is suggestive of the negative SIOD, which is characterised by warm anomalies from Madagascar arcing eastward and southeastward toward southwestern Australia (Reason, 1999; Behera et al., 2000; Behera and Yamagata, 2001; Qian et al., 2002; Suzuki et al., 2004). Positive SSTA exceeding 1 °C are also evident between 120°W and 170°W in the N3.4 region (Figure 4(d)).

Table III. Multiple comparison Z-value tests for Niño-3.4 region stratified by the six TC trajectory clusters
Niño-3.4C1C2C3C4C5C6
  • a

    Indicates significant differences of median Niño-3.4 values at α = 0.05.

  • b

    Indicates significance at α = 0.01.

C1 0.79291.02093.2288b1.80090.4907
C2  0.22663.8834b1.19331.0618
C3   4.0759b1.02051.2265
C4    4.2297b2.0611a
C5     1.8720
C6      
Table IV. Multiple comparison Z-value tests for SDI region stratified by the six TC trajectory clusters
SDIC1C2C3C4C5C6
  • a

    Indicates significant differences of median SDI values at α = 0.05.

  • b

    Indicates significance at α = 0.01.

C1 0.60041.90282.1531a3.1125b1.0592
C2  1.30242.6538b2.6414b0.6178
C3   3.7465b1.61180.3485
C4    4.5597b2.6445b
C5     1.6232
C6      

El Niño is well known to associate with westerly wind anomalies over the southwestern Indian basin during an austral warm season (Vitart et al., 2003; Todd et al., 2004; Yoo et al., 2006). TTTs are, likewise, associated with westerly anomalies over the same geographic region, with vertical extent from the boundary layer up to 500 hPa and a west–east signature that can extend across the entire southern Indian Ocean (Todd and Washington, 1999). Reason (2002) found the TTTs also to vary in association with SSTA dipoles similar in pattern to the SIOD of Behera and Yamagata (2001), and the deepest convection coincident with TTTs has likewise been found to shift from southern Africa to the southwestern Indian Ocean in association with warm ENSO events (Fauchereau et al., 2009; Pohl et al., 2009; Manhique et al., 2011). Given the high significance of El Niño and negative SIOD for C4 relative to all other TC trajectory types, coupled with the propensity of TTTs to extend eastward over Madagascar and the western ocean during both of these phases, this suggests that C4 type TCs are highly likely to occur when ENSO is in warm phase and SIOD is simultaneously in negative mode. This 30°–35° eastward shift in the regional tropical convective maximum and the related eastward shift of poleward convective outflow, and the subtropical jet stream, provides a logical explanation by which C4-type TCs are often swept south and southeastward away from Madagascar. This is consistent with the synoptic-scale teleconnections of Hoerling et al. (1997) and the S-moving TC kinematic and thermodynamic profile of Chang-Seng and Jury (2010b).

Having established the ocean–atmospheric connections for C4, it is now appropriate to compare with C3, which forms in the same western region, yet approaches inhabited regions with greater frequency. This group is significantly different from C4 in median values of N3.4 and SDI (Tables III and IV), suggesting that C3 occurs more frequently during La Niña or a positive SIOD phase. The SSTA composite maps for C3 (Figures 3(c) and 4(c)) depict warm anomalies south and southeast of Madagascar, and cool anomalies in the equatorial Pacific. SSTA composite difference maps for C4 and C3 (Figures 5(a) and 6(a)) further support the importance of the SDI and N3.4 regions in distinguishing between these trajectory types. Enhanced easterly and southeasterly trade winds across the subtropical and tropical southern Indian basin during La Niña tend to coincide with more westward moving TCs that would threaten inhabited land masses with higher frequency such as those in group C3 (Reason et al., 2000; Vitart et al., 2003; Chang-Seng and Jury, 2010a and 2010b). Furthermore, in accordance with a positive SIOD phase, warm SSTA in the southwestern Indian Ocean are associated with increased convective precipitation over southern Africa as the TTTs are more frequently anchored to the Angola thermal low (Todd and Washington, 1999; Reason, 2001; Reason, 2002). For C3-type TCs that turn poleward of 25°S late in their life cycle, interaction with TTTs over the African subcontinent would help explain late re-curvature.

Figure 5.

Sea surface temperature anomaly composite differences in the southern Indian Ocean for (a) C4 minus C3; (b) C1 minus C5; and (c) C6 minus C2. SDI West (27°S–37°S, 55°E–65°E) and SDI East (18°S–28°S, 90°E–100°E) regions are shown as boxes in each image. Positive (negative) values are symbolised by dark solid (light broken) lines with units in °C

Figure 6.

Sea surface temperature anomaly composite differences in the equatorial Pacific Ocean for (a) C4 minus C3; (b) C1 minus C5; and (c) C6 minus C2. Niño-3.4 region (5°N–5°S, 120°W–170°W) is shown with a box in each image. Positive (negative) values are symbolised by dark solid (light broken) lines with units in °C

Further supporting the difference between TC tracks belonging to groups C3 and C4, the result of the Fisher's Exact Test indicates a highly significant association between the TC types (C3 and C4) and the type of interaction between ENSO and SIOD (p-value = 0.0007, Table V). When ENSO was in warm phase and SIOD was simultaneously negative, type C4 TCs occurred frequently, and type C3 TCs seldom occurred. These findings support the assertion that type C4 storms are highly associated with antiphasing of ENSO (warm) and SIOD (negative) wherein conditions are favourable for TC–trough interaction over the southwestern Indian Ocean. In contrast, type C3 storms occur significantly more often than C4 storms when ENSO is neutral and SIOD is positive. In this situation, the Mascarene anticyclone and associated trade winds are present over the central and southwestern Indian basin, and TTTs typically remain anchored over the African subcontinent. These results underscore the insufficiency of an opposing symmetrical El Niño–La Niña paradigm in discerning influences on southwestern Indian TC trajectories without consideration of ENSO-independent variability (Vitart et al., 2003; Klinman and Reason, 2008).

Table V. Southern Indian Ocean western subregion tropical cyclones by type and ENSO and SIOD interactions. E+ represents a warm ENSO event, E∼ is a neutral event, and E− a cool event. S+ signifies a positive SIOD event and S− a negative SIOD event
 C3C4Total
E + S+527
E + S−21214
E∼S+13215
E∼S−549
E − S+303
E − S−628
Total342256

3.2.3. Eastern region TC types: C1 versus C5

The TC trajectory type that develops in the eastern region with a typically westward motion, C5, is most strongly associated with the positive SIOD mode relative to the other clusters. Readily identifiable positive SIOD signals are apparent both in the high significance of C5 in the multiple comparisons (Table IV), and in the SSTA composite maps (Figure 3(e)) which show spatially coherent warm anomalies off the southeast coast of Africa into the southwestern Indian Ocean, as well as cool anomalies extending eastward from Madagascar along 15°S, 25°S. A La Niña SSTA pattern is also visible across the equatorial central Pacific (Figure 4(e)). The cool SSTAs in the southern Indian region are largely induced by decreased air-to-sea latent heat flux and Ekman transport associated with strong trade winds and concomitant equatorward advection of relatively dry mid-latitude air masses across a tight north–south SST gradient (Behera and Yamagata, 2001; Hermes and Reason, 2005; Chiodi and Harrison, 2007; Huang and Shukla, 2007b). The presence of these SSTA infers anomalously strong trade winds across the southern Indian Ocean north of 25°S which could aid in steering TCs on longer duration and lower latitude westward tracks, as in Vitart et al. (2003). Therefore, when a positive SIOD phase occurs in tandem with an ENSO neutral or cool phase, the TCs of type C5 tend to remain at lower latitudes because the seasonally strong subtropical southern Indian anticyclone precludes repeated northward intrusions by planetary waves (L'Heureux and Thompson, 2006) which then fail to foster eastward extension of convective clusters and their poleward outflows associated with the Angola thermal trough and South Indian Convergence Zone (SICZ).

While TCs in group C5 develop in the southeastern Indian basin, they occasionally threaten land as they progress far westward. TCs in the counterpart group C1 also develop in the east but re-curve into the middle latitudes in the central or eastern ocean. The SSTA composite maps for C1 (Figures 3(a) and 4(a)) do not exhibit a strong ENSO warm or cool pattern in the Pacific. However, the data depict a northwest to southeast region of warm SSTA in the central and southeast subtropical southern Indian Ocean which resembles the spatial SSTA pattern of the SIOD negative mode. Maps of the differences between the C1 and C5 composites (Figures 5(b) and 6(b)) show marked differences in SSTA patterns in both the SIOD and ENSO regions. Specifically, the presence of warm SSTA in the central and eastern tropical and subtropical southern Indian basin and cool SSTA in the southwest subtropical Indian Ocean indicates weakened trade winds and increased frequency of frontal intrusions. This would result in stronger cold air advection promoting decreased air-to-sea latent heat flux in the subtropical southern ocean (Behera and Yamagata, 2001; Chiodi and Harrison, 2007). Also, reduced Ekman transport from the relatively weak trade winds to the north, in conjunction with increased air-to-sea heat flux from warm advection ahead of cold fronts, would allow for warm SSTA patterns associated with the negative SIOD phase or El Niño. Similar to TC type C4, troughs penetrating farther north into the central and southeastern Indian basin would influence C1 TCs to re-curve more abruptly into the mid-latitudes than C5 TCs.

Following the same methodology as for types C3 and C4 above, a contingency table of ENSO and SIOD interactions is constructed for types C1 and C5 and Fisher's Exact Test is again applied (Table VI). Results indicate a significant association between the two eastern region trajectory types and ENSO and SIOD phase interactions, albeit with a more liberal alpha level (α = 0.1, p-value = 0.097). Type C5 TC trajectories, the westward moving group, did not occur at all coincident with any ENSO warm phase during our study period, and only twice when ENSO was neutral but SIOD was in negative mode. Type C5 storms occurred most often during positive SIOD phases with only two during La Niña when SIOD was negative. The data suggest for the eastern region that it is not sufficient to ascribe TCs of long duration that move consistently westward only to La Niña. Contemporaneous consideration of SIOD phase is especially relevant when ENSO is neutral, and it is physically consistent that a positive SIOD phase would influence more westward moving TCs, as seen in both trajectory types C3 and C5. Finally, it is also evident in analysis of group C1 that a warm ENSO event combined with a negative SIOD phase results in a higher frequency of re-curving TC trajectories than during an ENSO warm event combined with a positive SIOD mode.

Table VI. As in Table V, but for eastern subregion tropical cyclones
 C1C5Total
E + S+303
E + S−606
E∼S+8816
E∼S−12214
E − S+549
E − S−527
Total391655

3.2.4. Central region TC types: C2 versus C6

The TC group with genesis in the central region and westward and southwestward movement is C2. C2 is only significantly different from C4 in the N3.4 comparisons (Table III), and is only significantly different than C5 in the SDI comparisons (Table IV). However, the latter result should not be interpreted to mean that C2 exhibits a strong association with SIOD that is opposite in sign to the strong association between C5 and the positive SIOD phase. The statistical significance is in a strong positive association (C5) compared to a very weak association (C2), not a strong positive association compared to a strong negative association. It is clear from both the SDI values and the SSTA composite map (Figure 3(b)) that in this study C2 TCs are not associated obviously with either phase of the SIOD. The equatorial Pacific SSTA composite map for C2 (Figure 4(b)) displays a recognisable La Niña pattern east of 180°, though the N3.4 values are not statistically significant in multiple comparisons.

As in the previous sections, it is useful to compare C2 to the counterpart central region TC type C6, which is characterised by more eastward trajectories than C2. These two groups are not statistically different in comparisons of their median values of N3.4 and SDI. Nor are there contiguous warm SSTA in the equatorial Pacific (Figure 4(f)) to suggest an El Niño association. The composite difference maps of C6 minus C2 (Figures 5(c) and 6(c)) identify the equatorial Pacific as a region of note in distinguishing the two TC types, but there is not a significant statistical nor an obvious geospatial difference that would allow for more concise interpretation. Finally, a contingency table is constructed stratifying the TCs by type and by configurations of ENSO and SIOD (Table VII). The results of Fisher's test also suggest no significant association to distinguish TC trajectories in types C2 and C6 by their frequencies during different interactions of ENSO and SIOD phases (p-value = 0.559). When separated from the western and eastern regions, the central region TC trajectories do not display obvious modes relating re-curving and eastward tracks to El Niño and negative SIOD, nor westward tracks to neutral ENSO and positive SIOD or La Niña. Given the small size of group C6 and that half of the TCs in the group were early or late season storms (September–October, or April–May), perhaps it is not surprising that no clear interactive signal between ENSO and SIOD is found in the central region. Both ENSO and SIOD are known to exhibit stronger signals at the height of austral summer as they are phase locked to the seasonal cycle.

Table VII. As in Table V, but for central subregion tropical cyclones
 C2C6Total
E + S+426
E + S−415
E∼S+9413
E∼S−549
E − S+606
E − S−10313
Total381452

4. Discussion

In this study, we build on the ENSO-TC track framework of Vitart et al. (2003) and Klinman and Reason (2008) by considering interactions of the ENSO and SIOD phases. We find that southward and southeastward moving TCs (types C1 and C4) are more likely to occur, particularly in the southwestern Indian Ocean, when ENSO is in warm phase and SIOD is simultaneously in negative mode. Furthermore, westward and southwestward moving TCs (C3 and C5) are likely during a cool ENSO event (relative to warm ENSO), but are most likely to occur when ENSO is neutral and SIOD is in positive mode. A key implication of these results is that periods of either strong westerly or easterly steering flow in the southwestern Indian Ocean should not be attributed solely to any particular ENSO phase, but the simultaneous SIOD phase should be considered as well.

The SIOD-TC trajectory relationship proposed in this paper may be applied to TC Favio of 2007 to explain its unusual direction of motion south of Madagascar and back to the northwest over Mozambique. Klinman and Reason (2008) noted that such a westward track during a weak El Niño year did not follow the ENSO-TC steering model of Vitart et al. (2003), and suggested that monthly data might be more appropriate than seasonal data in discerning southern Indian TC track variability. Using monthly data, the SDI was strongly positive (1.35) in February 2007, which is consistent with both the warm SSTA south of Madagascar and persistent trade winds which allowed Favio to remain at a lower latitude (rather than re-curving into higher latitudes) and provided a favourable thermodynamic environment for it to re-intensify southwest of Madagascar. While the example of Favio is well suited to demonstrate how SIOD mode can assist in understanding TC track behaviour when ENSO is neutral or weakly in warm or cool phase, it must also be acknowledged that SIOD mode, like ENSO phase, is not a perfect predictor of southern Indian TC track direction.

This study suggests that inclusion of an index for SIOD mode could add predictive power in statistical modelling of southwestern Indian TC occurrence at intraseasonal or even seasonal temporal scales. Indirect evidence in support of this assertion exists in the peer-reviewed literature. Leroy and Wheeler (2008) include the second rotated principal component of Indo-Pacific SSTA as a significant predictor (variable SST2) in their logistic regression model of Southern Hemisphere weekly TC genesis and activity. Their SST2 variable is very similar in construction and spatial pattern (shown in their Figure 5(b), which was adapted from Drosdowsky and Chambers, 2001) to the subtropical SSTA dipole patterns identified in numerous previous studies (Reason, 1999; Behera and Yamagata, 2001; Qian et al., 2002; Suzuki et al., 2004; Huang and Shukla, 2007b). In the updated operational statistical model (Vitart et al., 2010), the SST2 variable is replaced by the tropical Dipole Mode Index of Saji et al. (1999). Our results suggest that an index capturing the subtropical dipole phenomenon (such as SDI) could be at least as important as the tropical mode in modelling of southwestern Indian TC occurrence or track direction. As a proxy measure of the variability in strength and position of the TTTs and southern Indian subtropical anticyclone at subseasonal intervals, it is physically consistent that SDI should be significantly associated with variability of southwestern Indian TC trajectories and, therefore, occurrence at subseasonal intervals. As Hermes and Reason (2005) observe, the SIOD anomalies may not always be adequately captured by the SDI as constructed by Behera and Yamagata (2001). More research is needed to index SIOD-like patterns with varying spatial signatures before these can most effectively be applied in prediction of TC activity or track direction.

The strong tendencies for certain TC trajectories to occur with certain configurations of ENSO and SIOD (Tables V and VI) also raise the possibility that theseconfigurations are not only specific to types of TC trajectories, but actually reflect preferred ocean–atmosphere patterns during southern Indian TC season. Using the same methodology as described in Section 2.1 for assignment of El Niño, neutral, and La Niña months, SDI was negative 64% of the time coincident with El Niño (Figure 7) during November–March over the period 1979–2007, whereas SDI was positive 61% of the time coincident with neutral ENSO conditions. La Niña and both positive and negative SIOD modes occurred with equal frequency during the study period. These monthly percentages are similar to daily percentages for coincidences of Antarctic Oscillation (AAO) and ENSO phases presented in Carvalho et al. (2005) (seen in their Figure 6). There is a strong possibility that the AAO or Southern Annular Mode (SAM) is associated with both ENSO and SIOD and subsequently with TTTs and TC trajectory variability through the modification of planetary waves (Hermes and Reason, 2005; L'Heureux and Thompson, 2006; Klinman and Reason, 2008; Manhique et al., 2011; Chang-Seng and Jury, 2010a).

Figure 7.

Percentage of months during November–March over the period 1979–2007 when ENSO was in cool phase (ENSO−), neutral phase (ENSO∼), or warm phase (ENSO+) and SDI was simultaneously in negative or positive mode

5. Conclusions

The goal of this work was to investigate the influences of both the SIOD and ENSO on southwestern Indian Ocean TC trajectories. We grouped TCs by their genesis regions and subsequent trajectories and then employed non-parametric ANOVA to test for differences of monthly values of N3.4 and SDI corresponding to each TC's life cycle. The results indicate that both N3.4 and SDI are associated with significant differences in the directions of TC trajectories in the study region. Specifically, our work suggests that southern Indian basin TCs originating in the regions 54°E–73°E or 87°E–110°E, and following westward and southwestward trajectories (southward and southeastward trajectories) are significantly more associated with a cool or neutral ENSO phase and a positive SIOD mode (warm ENSO and negative SIOD mode). We do not find the same statistically significant relationships in the central ocean approximately bounded by 73°E–87°E.

The results presented in this study go beyond an ENSO-only model of southwestern Indian TC trajectory variability. When considered together with ENSO, the SIOD (using the SDI) allows a clearer understanding that when ENSO is in warm phase and SIOD is negative the regional ocean–atmosphere patterns strongly favour a shift of tropical temperate troughs (TTTs) over the southwestern Indian Ocean coincident with the warm SSTA pool. In this situation, the TTTs are positioned to frequently steer TCs away from inhabited areas on the western rim of the basin. When ENSO is neutral or in cool phase and SIOD is positive, TTTs do not shift persistently eastward over the southwestern Indian Ocean and well developed tropical systems are frequently steered along more westward and southwestward tracks. It is important to discern the differences in these configurations and could be of great use in the future to provide forewarning perhaps a month or more in advance of enhanced potential for TC strikes to the heavily populated and socially vulnerable nations of Madagascar and Mozambique.

In broad terms, much additional work is needed to understand the asymmetrical and non-linear tropical and extratropical interactions that produce the SIOD SSTA patterns, and how these bear influence on southwestern Indian TC trajectories at intraseasonal and seasonal time scales. In particular, understanding of the relationship between TTTs and TC activity could benefit from deeper inquiry as this topic has not been comprehensively examined in the literature. Future research should also explore alternative indexing procedures to most effectively capture the geographic location and extent as well as magnitude of SIOD events, and test the application of these alternative SIOD indexes for potential improvement of operational intraseasonal to seasonal southwestern Indian basin TC prediction schemes.

Acknowledgements

The authors extend thanks to the two anonymous reviewers for their constructive comments. We also thank Peter Waylen and Tim Fik for their valuable suggestions during the earlier stages of this work. This research was undertaken at the University of Florida as part of the lead author's M.S. thesis. He wishes to thank the UF Geography Department for funding and support.

Ancillary