• Bell, G. D., and M. Chelliah, 2006: Leading tropical modes associated with interannual and multidecadal fluctuations in North Atlantic hurricane activity. J. Climate, 19 , 590612.

    • Search Google Scholar
    • Export Citation
  • Chan, J. C. L., 1985: Tropical cyclone activity in the Northwest Pacific in relation to the El Niño/Southern Oscillation phenomenon. Mon. Wea. Rev., 113 , 599606.

    • Search Google Scholar
    • Export Citation
  • Chan, J. C. L., 2000: Tropical cyclone activity over the western North Pacific associated with El Niño and La Niña events. J. Climate, 13 , 29602972.

    • Search Google Scholar
    • Export Citation
  • Chu, J-H., C. R. Sampson, A. S. Levine, and E. Fukada, 2002: The Joint Typhoon Warning Center tropical cyclone best-tracks, 1945–2000. Naval Research Laboratory Tech. Rep. NRL/MR/7540-02-16, 112 pp.

  • Elsner, J. B., 2003: Tracking hurricanes. Bull. Amer. Meteor. Soc., 84 , 353356.

  • Elsner, J. B., and B. Kocher, 2000: Global tropical cyclone activity: A link to the North Atlantic Oscillation. Geophys. Res. Lett., 27 , 129132.

    • Search Google Scholar
    • Export Citation
  • Emanuel, K. A., 2005: Increasing destructiveness of tropical cyclones over the past 30 years. Nature, 436 , 686688.

  • Frank, W. M., 1977: The structure and energetics of the tropical cyclone. II. Dynamics and energetics. Mon. Wea. Rev., 105 , 11361150.

    • Search Google Scholar
    • Export Citation
  • Frank, W. M., 1982: Large scale characteristics of tropical cyclones. Mon. Wea. Rev., 110 , 572586.

  • Gray, W. M., C. W. Landsea, P. W. Mielke Jr., and K. J. Berry, 1993: Predicting Atlantic basin seasonal tropical cyclone activity by 1 August. Wea. Forecasting, 8 , 7386.

    • Search Google Scholar
    • Export Citation
  • Henderson-Sellers, A., and Coauthors, 1998: Tropical cyclones and global climate change: A post-IPCC assessment. Bull. Amer. Meteor. Soc., 79 , 1938.

    • Search Google Scholar
    • Export Citation
  • Lander, M. A., 1994: An exploratory analysis of the relationship between tropical storm formation in the western North Pacific and ENSO. Mon. Wea. Rev., 122 , 636651.

    • Search Google Scholar
    • Export Citation
  • Lander, M. A., and C. P. Guard, 1998: A look at global tropical cyclone activity during 1995: Contrasting high Atlantic activity with low activity in other basins. Mon. Wea. Rev., 126 , 11631173.

    • Search Google Scholar
    • Export Citation
  • Pielke Jr., R. A., C. Landsea, M. Mayfield, J. Laver, and R. Pasch, 2005: Hurricanes and global warming. Bull. Amer. Meteor. Soc., 86 , 15711575.

    • Search Google Scholar
    • Export Citation
  • Simpson, R. H., 1974: The hurricane disaster potential scale. Weatherwise, 27 , 169186.

  • Wang, B., and J. C. L. Chan, 2002: How strong ENSO events affect tropical storm activity over the western North Pacific. J. Climate, 15 , 16431658.

    • Search Google Scholar
    • Export Citation
  • Webster, P. J., G. J. Holland, J. A. Curry, and H-R. Chang, 2005: Changes in tropical cyclone number, duration, and intensity in a warming environment. Science, 309 , 18441846.

    • Search Google Scholar
    • Export Citation
  • View in gallery

    Time series of the global annual number of tropical cyclones (counts) for four intensity categories: tropical storms, weaker hurricane-strength tropical cyclones (categories 1–2), stronger tropical cyclones (categories 3–4), and the most intense storms (category 5).

  • View in gallery

    As in Fig. 1, but for the annual global number of storm days in the four intensity groups.

  • View in gallery

    Time series of the MEI ENSO index, the annual NAO index, the annual global number of all tropical cyclones (TS–category 5), and the global number of very intense tropical cyclones (categories 4–5) for the period 1985–2003.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 442 299 23
PDF Downloads 301 241 21

The Interannual Variability of Tropical Cyclones

View More View Less
  • 1 Department of Meteorology, The Pennsylvania State University, University Park, Pennsylvania
© Get Permissions
Full access

Abstract

This paper examines the interannual variability of tropical cyclones in each of the earth’s cyclone basins using data from 1985 to 2003. The data are first analyzed using a Monte Carlo technique to investigate the long-standing myth that the global number of tropical cyclones is less variable than would be expected from examination of the variability in each basin. This belief is found to be false. Variations in the global number of all tropical cyclones are indistinguishable from those that would be expected if each basin was examined independently of the others. Furthermore, the global number of the most intense storms (Saffir–Simpson categories 4–5) is actually more variable than would be expected because of an observed tendency for storm activity to be correlated between basins, and this raises important questions as to how and why these correlations arise. Interbasin correlations and factor analysis of patterns of tropical cyclone activity reveal that there are several significant modes of variability. The largest three factors together explain about 70% of the variance, and each of these factors shows significant correlation with ENSO, the North Atlantic Oscillation (NAO), or both, with ENSO producing the largest effects. The results suggest that patterns of tropical cyclone variability are strongly affected by large-scale modes of interannual variability. The temporal and spatial variations in storm activity are quite different for weaker tropical cyclones (tropical storm through category 2 strength) than for stronger storms (categories 3–5). The stronger storms tend to show stronger interbasin correlations and stronger relationships to ENSO and the NAO than do the weaker storms. This suggests that the factors that control tropical cyclone formation differ in important ways from those that ultimately determine storm intensity.

Corresponding author address: William M. Frank, Department of Meteorology, The Pennsylvania State University, 503 Walker Building, University Park, PA 16802. Email: frank@ems.psu.edu

Abstract

This paper examines the interannual variability of tropical cyclones in each of the earth’s cyclone basins using data from 1985 to 2003. The data are first analyzed using a Monte Carlo technique to investigate the long-standing myth that the global number of tropical cyclones is less variable than would be expected from examination of the variability in each basin. This belief is found to be false. Variations in the global number of all tropical cyclones are indistinguishable from those that would be expected if each basin was examined independently of the others. Furthermore, the global number of the most intense storms (Saffir–Simpson categories 4–5) is actually more variable than would be expected because of an observed tendency for storm activity to be correlated between basins, and this raises important questions as to how and why these correlations arise. Interbasin correlations and factor analysis of patterns of tropical cyclone activity reveal that there are several significant modes of variability. The largest three factors together explain about 70% of the variance, and each of these factors shows significant correlation with ENSO, the North Atlantic Oscillation (NAO), or both, with ENSO producing the largest effects. The results suggest that patterns of tropical cyclone variability are strongly affected by large-scale modes of interannual variability. The temporal and spatial variations in storm activity are quite different for weaker tropical cyclones (tropical storm through category 2 strength) than for stronger storms (categories 3–5). The stronger storms tend to show stronger interbasin correlations and stronger relationships to ENSO and the NAO than do the weaker storms. This suggests that the factors that control tropical cyclone formation differ in important ways from those that ultimately determine storm intensity.

Corresponding author address: William M. Frank, Department of Meteorology, The Pennsylvania State University, 503 Walker Building, University Park, PA 16802. Email: frank@ems.psu.edu

1. Introduction

For the past several decades there has been a persistent myth that the annual global number of tropical cyclones is observed to be more stable than would be expected given the large variability observed within the individual cyclone basins. Two popular explanations for this perceived observation have been proposed. Most references to the myth have invoked it as evidence that there are important interactions between the storms and the global climate such that storm-induced feedbacks tend to limit the global number of cyclones that can form each year (e.g., Henderson-Sellers et al. 1998). In other words, enhanced storm activity in one basin somehow suppresses the net activity in the other basins.

While this view remains popular, an alternative explanation has arisen based on recent studies showing anticorrelations between storm numbers in some basins that are related to known modes of large-scale variability. For example, Lander and Guard (1998), Lander (1994), Gray et al. (1993), Chan (1985), Bell and Chelliah (2006), and others have analyzed relationships between El Niño–Southern Oscillation (ENSO) and storm numbers in various cyclone basins. Such studies usually show a strong negative correlation between storm activity in the North Atlantic and northeast Pacific associated with ENSO (though some of the studies also show positive correlations between other pairs of basins, e.g., Lander and Guard 1998). This alternative view is that negative correlations between storm numbers in different basins increase the observed variability within the affected basins without having a significant effect on the global variability. This explanation of the apparent stability of the global storm number does not require storm-induced feedbacks between basins.

One reason that the myth has drawn interest is that the first proposed explanation implies the existence of significant relationships between tropical cyclones and climate. For example, it could be evidence that tropical cyclones produce negative feedbacks to the general circulation that ultimately limit their numbers. It might also have important implications for estimating the characteristics of tropical cyclones in past and future climates. In their post–Intergovernmental Panel on Climate Change assessment of tropical cyclones and climate change Henderson-Sellers et al. (1998) noted this connection between the myth and the issue of hurricanes in future climates, which is a topic of much current interest (e.g., Emanuel 2005; Webster et al. 2005; Pielke et al. 2005).

Whatever the cause of the observed storm climatology, the myth is based on the belief that observations show that when there is an above-average number of storms in one basin, then the sum of the storm numbers in the other global basins will usually be somewhat below average. We will refer to this as compensation. The myth is intriguing, and the lead author confesses that he is one of the many who has passed it on to a large number of students and other colleagues over the years. Alas, as will be shown below, it also turns out to be false.

Lander and Guard (1998) first suggested that the annual variability of the global tropical cyclone number might be larger, rather than smaller, than that expected from the observed variability in each individual basin. They examined data from 1969 to 1995 and noted that the standard deviation of the global total of all tropical cyclones was around 10, while the standard deviation of the individual basins averaged around 4. Their study included tropical storms (TS) and hurricanes/mature cyclones (i.e., all storms with estimated maximum winds greater than 17 m s−1). They suggested that perhaps the annual number of tropical cyclones (TCs) is “destabilized by positive correlations among some or all of the TC basins.” However, they did not pursue this aspect of their analysis to the point of determining whether the global TC variability truly shows significant anticompensation.

This study first examines the question of whether or not the myth is true. That is, does the global annual number of cyclones vary less than would be expected from the observed variability within the individual storm basins? Correlations between the numbers of storms and storm days in different basins are computed to identify relationships between storm activity in the different basins. The data are then examined using factor analysis to determine the modes of tropical cyclone variability, their effects on the global storm number and days, and the relationships between these modes and two known large-scale modes of variability: ENSO and the North Atlantic Oscillation (NAO). As noted above, several authors have found correlations between ENSO and storm numbers in certain basins. On the other hand, Elsner and Kocher (2000) performed a factor analysis of global tropical cyclone activity and found that the strongest factor was correlated with the NAO rather than ENSO. The current study utilizes a different form of factor analysis that focuses on factors that are associated with ENSO and the NAO in order to determine their respective roles in tropical cyclone climatology.

2. Data

The average annual number of tropical cyclones (maximum winds of 17 m s−1 or more) is usually computed to be between 80 and 90, while the number that reach 33 m s−1 (called hurricanes, typhoons, or severe tropical cyclones, depending upon location) is around 40 to 50. Numbers vary slightly depending upon the period averaged and datasets used.

This study requires a global dataset, and the primary sources of such data are the archives assembled by operational forecast centers around the world. The data available and procedures used to estimate tropical cyclone intensity have varied greatly between regions and in time, and considerable care should be taken with their use. These archives are places where angels fear to tread.

The data used here are taken from archived tropical cyclone best track archives. They were downloaded from Prof. K. Emanuel’s Web page at the Massachusetts Institute of Technology (MIT; http://wind.mit.edu/~emanuel/home.html) and were checked against original source files from warning centers. Data were grouped into five tropical cyclone basins to be consistent with recent studies of tropical cyclone climatology. These basins are North Atlantic (AT), northeast Pacific (EP), northwest Pacific (WP), North Indian Ocean (IO), and Southern Hemisphere (SH). However, combining the southwest Pacific and South Indian Ocean into a single basin may not be desirable for some studies.

After examining the data from each basin and the various published discourses on the data characteristics, we came to the conclusion that no intensity data should be used from years earlier than 1983. One can go back further in the WP and AT basins, but in the others (particularly the IO and SH), there are too many missing or inaccurate intensities for this type of study. Further, because the Joint Typhoon Warning Center did not go back earlier than 1985 during its attempt to reconcile and correct earlier datasets (Chu et al. 2002), we chose that year as a more conservative starting point. Thus, our data are from the 19-yr period from 1985 to 2003. Only data for systems with tropical storm or stronger intensities are used, as there are inconsistencies between the basins in the reporting of tropical depressions and disturbances.

The two variables that we examined were storm counts and storm days. Storm count is the annual number of storms in each basin that fall within an intensity group. The intensity is defined by the maximum strength that the storm reached during its lifetime. Storm days are the number of days during which a storm existed within a certain intensity category. In this case the intensity category is defined by the intensity at each advisory time. (Storm days are determined by summing all of the 6-hourly reports in each intensity category and dividing the totals by 4.) Storm counts reflect the frequency of cyclogenesis in the basin but only provide one data point on intensity: the maximum observed winds. Storm days are a superior measure of the integrated effects of the storm population over the full storm life cycles but convey less information about the number of storms. Storm counts and days are often closely related, but in some basins the relationship between them is more complex. For example, during an El Niño year storms in the NW Pacific basin tend to form farther to the east than normal because of the eastward shifts in the SST maximum and deep convection. This results in longer-than-normal storm lifetimes (e.g., Chan 2000; Wang and Chan 2002).

An intense storm, such as one that is counted as a category 5, spends part of its life cycle in each of the weaker-intensity bins during both its intensification and decay phases. As a result, interannual variations in storm days for different intensity categories tend to be strongly correlated with each other. This is not true of the storm counts that are binned by their peak lifetime intensity category. Figures 1 and 2 show the raw-data global storm counts and storm days, respectively, for several intensity categories. In general, the storm day numbers tend to vary more smoothly than the counts, but both show similar patterns.

During our analysis period the average global number of hurricane-strength storms (categories 1–5) was 49 with a standard deviation of 6.7. There was also an average of 38 tropical storms each year. The storm day averages were 183 for hurricanes and 259 for tropical storms. Table 1 shows the global mean storm counts and days by intensity category.

It is important to note that tropical cyclones do not typically maintain a strong intensity (categories 3–5) for long periods of time. From Table 1 it can be seen that the average category 5 cyclone spends only about 1.2 days at that intensity. For all storms that reach category 4 or 5 intensity (16 per year on average), the mean time spent at category 4 or higher is 1.9 days per storm. For the entire intense storm group (categories 3–5) the average number of storms per year is about 24, and the average time that each storm spends at an intensity of category 3 or higher is 2.3 days. The raw numbers of storm numbers and storm days are shown in the appendix.

Three different ENSO indices were examined: the Southern Oscillation index (SOI), the equatorial SOI, and a multivariate ENSO index (MEI). Although these indices differ, they are highly correlated with each other, and we found that they are also strongly similar in their correlations with tropical cyclones. For brevity, only the MEI is used in the analyses below. Data for the ENSO indices and the annual NAO index were taken from the Climate Prediction Center of the National Oceanic and Atmospheric Administration (NOAA; http://www.cdc.noaa.gov/ENSO/). Figure 3 shows time series of the MEI and NAO indices along with global storm numbers. Examples of the patterns of variability of several variables associated with the MEI and NAO may be seen on the NOAA Web sites (http://www.cdc.noaa.gov/people/klaus.wolter/MEI/ and http://www.cpc.ncep.noaa.gov/data/teledoc/nao.shtml, respectively).

Most of the analyses below compare annual basin storm statistics to annual values of the MEI and NAO. We also examined relationships between the storm statistics and seasonal measures of the MEI and NAO, and the results were generally similar (Tables 4 and 5). The discussion focuses primarily on correlations with the annual indices.

3. Methodology

The first task is to see whether the variance of the observed global storm number differs from the variance that would be expected if each basin varies independently of the others. If each of n basins had the same mean number of storms and the same variance pattern, then the variance of the sum of the five basins would simply be the square root of n times the variance of an individual basin.

The problem is somewhat more complex when each basin has different properties, so we adopted a Monte Carlo method. We created 1000 annual storm totals by randomly drawing a yearly total from each of the five basins, summing them, and repeating the process 1000 times. These statistics were compared with the observed global storm totals.

If the myth is true, then the standard deviation of the observed global storm number should be smaller than the standard deviation of the number created from the random draws. This is because the implied compensation within the Tropics (a negative correlation between any one basin and the sum of the other four) would reduce the variance of the observed global total relative to random draws. Conversely, there could also be anticompensation between basins. This would occur if the positive correlations in storm numbers between various storm basins were larger than any negative ones.

Note that if there is a long-term trend in the global storm numbers, this would create a positive correlation between storm numbers in the various basins that would have nothing to do with the issue of interbasin compensation within a given year. Therefore, we removed the 19-yr trend from each basin or intensity category and repeated the analysis. Only the detrended analyses are shown here. The observed trends were quite small, and they are not discussed further because the data series is too short to compute significant trends. We also examined the effects of shifting the SH dataset in time so that the total storm numbers for that basin were for a single season, rather than a calendar year. The effects of this shift were small and will not be discussed further in this paper.

The various analyses were carried out for each of six intensity categories (tropical storms and hurricanes within categories 1–5 of the Saffir–Simpson scale) and for various combinations of these categories. These analyses were performed for both storm numbers and storm days. Correlations of storm numbers/days between basins were analyzed. Factor analysis was used to determine patterns of variability, and all of the above were related via correlation analysis with the MEI and an annual NAO index to determine the role of ENSO and the NAO in the global patterns.

The method of factor analysis used here included ENSO and NAO indices along with the tropical cyclone statistics when determining the factors. This was done in order to identify factors that were clearly related to these well-documented modes of large-scale tropical variability. This technique differs from the factor analysis performed by Elsner and Kocher (2000), who computed their factors from the storm data alone, then correlated the resulting factors with ENSO and NAO time series. The factor analysis was performed in Minitab using principal component analysis for factor extraction. No rotation was applied to the resulting factors. Those factors that explained less than 10% of the variance will not be shown or discussed, as their physical significance is doubtful.

In the tables showing results of the factor analyses, the individual cell values show the loading of that factor. The loading is the correlation coefficient between that variable and the factor.

4. Results

a. Interbasin compensation and the myth

There is no evidence of net compensation between basins of the annual storm number or the number of storm days (Tables 2 and 3). For all storms (TS–category 5) and for all hurricanes (categories 1–5), only 55% and 27%, respectively, of the random draws produced annual global storm numbers with standard deviations as large as the observed standard deviation (std dev). For there to be statistically significant net compensation between basins, the std dev of the random draws would have to exceed the observed std dev 95% of the time (50% of the time would be expected if the totals were totally random). The equivalent numbers for storm days are 15% for all storms and 36% for all hurricanes. Hence, when all storms are included, there appears to be weak anticompensation between basins.

When only the most intense storms are analyzed (categories 4 and 5), only 6% of the random draws have storm number std devs that exceed the observed std dev, while only 3% of the storm day draws exceed the observed std dev. (These percentages are 66% and 7%, respectively, for a group containing categories 3–5, not shown.) This indicates that the strongest storms exhibit strong anticompensation (i.e., the basin counts are positively correlated, such that an above-normal intense storm number in one basin is likely to occur with above-normal numbers in the sum of the other basins as well). This result exceeds the 95% significance level for storm days and is very close to that level for storm numbers.

The results clearly show that the myth is false. That is, there is not a tendency for enhanced storm activity in one basin to be compensated by reduced net storm activity in the others. In fact, for the more intense storms there appears to be significant anticompensation, or positive correlations, between basins in agreement with the findings of Lander and Guard (1998). This raises the question as to what causes these correlations, which may be related to known large-scale modes of variability. The analyses that follow explore patterns of tropical cyclone variability and their relationships to ENSO and the NAO.

b. Interbasin correlations

Correlations between annual storm numbers and storm days in the individual basins are shown in Tables 4 and 5, respectively, for three intensity groups: all storms (TS–category 5), weak storms (TS–category 2), and strong storms (categories 3–5). Because the time series consist of data from 19 yr, a correlation must have an absolute value of 0.46 to be significant at the 95% level. However, each table shows a set of 10 interbasin correlations. This means that in each table there is slightly less than a 50% probability that one of these 10 correlations would have an absolute value of at least 0.46 from random chance. Correlations below this level may well be meaningful in combination, but the relationships they imply cannot be substantiated by the individual correlations. Therefore, factor analysis is applied in section 4c below to isolate coherent modes of interannual variation from background noise.

The total number of tropical cyclones in the Atlantic is negatively correlated with storm numbers in each of the other four basins, and the coefficients are of magnitude 0.31 or larger for all basins except the Indian Ocean. The northeast Pacific is positively correlated with all basins except the Atlantic, though none of the positive correlations exceeds 0.26. The northwest Pacific is negatively correlated with all basins except the northeast Pacific. None of the 10 interbasin correlations exceeds 0.34 in absolute magnitude.

The pattern of the interbasin correlations of storm numbers for the weaker storms (TS–category 2) is somewhat similar to those for all storms, with 7 of the 10 correlations having the same sign. However, the magnitudes of the correlations are generally weaker than those for all storms.

The correlations for strong-storm numbers (categories 3–5) are generally stronger than for weak storms or all storms, with 2 of the 10 correlations exceeding 0.46 and four exceeding 0.40. The patterns of the correlations for strong storms are much different from those of weak storms. Seven of the 10 interbasin correlation pairs have different signs between Tables 4b and 4c. For example, the number of strong storms in the AT is negatively correlated with strong-storm numbers in the EP (−0.46) and positively correlated with such storms in the IO (0.41). Weak-storm numbers in the AT correlate positively with the EP (0.19) and negatively with the IO (−0.31).

The different patterns of variation between strong and weak tropical cyclones indicate that the two groups respond differently to changes in environmental conditions. This suggests that some of the modes of large-scale interannual variability in the Tropics affect the percentage of storms that reach the strong-storm intensity threshold. Further research is needed in this area.

The interbasin correlations for storm days (Tables 5a–c) show patterns that closely resemble those for storm numbers, as expected. The magnitudes of the storm day correlations tend to be slightly larger than those for storm numbers for the all-storms and strong-storm categories and slightly smaller for the weak-storm category.

The bottom four rows in each panel of Tables 4 and 5 show correlations (r) between the MEI and NAO indices and the storm counts and days. The first two rows are correlations using the annual indices, while the third and fourth rows show correlations with the seasonal indices, as described above. The correlation patterns are generally similar for the annual and seasonal indices, although the seasonal correlations tend to be slightly stronger. The discussion will focus on the annual indices for brevity. The two indices are only weakly correlated with each other at about −0.10 (annual) to −0.12 (seasonal). The positive phase of the MEI (El Niño) is associated with an increase in total global storm counts (0.20 correlation), and this is due entirely to an increase in the number of category 3–5 storms (r = 0.36). The MEI is even more strongly correlated with total storm days (r = 0.48 for all storms, 0.52 for strong storms). During an El Niño (positive MEI) there is a major decrease in storm numbers and days in the AT (e.g., Gray et al. 1993) with corresponding increases in activity in all other basins, particularly the EP and WP. This pattern is seen for the all-storms group, but it is much stronger for the strong-storm group than for the weak-storm group. The MEI shows virtually no correlation with storm days in the AT and EP basins. Noteworthy are the much larger correlations between the MEI and WP for storm days than for storm numbers. This could result from the well-known eastward shift in the WP region of storm formation during El Niño, which causes longer tracks and lifetimes (e.g., Wang and Chan 2002). Because the weak-storm group in the WP is negatively correlated with the MEI for storm numbers and uncorrelated for storm days, El Niño is also associated with an increase in the fraction of storms that intensify to category 3 or higher in the WP.

c. Factor analysis

The patterns of interbasin correlation and anticorrelation suggest the existence of one or more modes of interannual variability in tropical cyclone numbers and storm days. Factor analysis is performed to deduce these modes from the detrended time series. The first analysis uses the global storm numbers and storm days in each separate intensity category (TS–category 5) along with the MEI and the annual NAO index described above. Because there are six intensity categories, the data consisted of a time series of eight-dimensional vectors. Results are similar using the other two ENSO indices (SOI and equatorial SOI, not shown). The inclusion of the MEI and NAO indices increases the likelihood that the analysis will identify modes that are associated with ENSO and/or the NAO.

The second analysis uses the individual basin numbers of total storms and storm days with all intensity categories combined as well as the MEI and NAO indices. Because there are five basins, this data series consists of seven-dimensional vectors.

The results for storm intensity categories are shown in Tables 6 and 7. Only factors that explain at least 10% of the variance are shown. Loadings with absolute values of 0.50 or greater are in boldface font. The largest factor for the storm counts (Table 6) explains 26% of the total variance and shows strong, roughly equal loadings for both the MEI and NAO. There are strong positive loadings for category 4 and 5 storms and also for category 2. This means that the global numbers of storms in these categories are anomalously high during the positive phases of MEI and NAO and low during the negative phases. The positive phase of this mode is associated with both the El Niño phase of ENSO and the positive phase of the NAO (anomalously high pressure in the Azores region). Further, category 3 storms are greatly reduced during this mode. Tropical storm numbers are slightly suppressed. Our interpretation of these results is that during the positive phase of this mode a greater percentage of mature tropical cyclones reach category 4–5 intensity rather than reaching a plateau at category 3. There also seems to be an above-normal tendency for tropical storms to develop into moderate (category 2) tropical cyclones.

Factor 2 (19% of variance) has little ENSO signal but has moderate loading of the positive NAO phase. It is associated with a global increase in category 4 storms and some tendency for more TS than usual, but the numbers are down in all other storm categories, particularly categories 1 and 2. The third factor (17%) is a positive-NAO and negative-MEI mode that exhibits slightly increased storm activity in categories 2–4 but strong decreases in the number of TS and category 5 storms.

When storm days are examined (Table 7), the first factor explains 52% of the total variance, twice the amount explained by the largest factor for storm counts. The largest storm-day factor again represents a mode with positive loadings of the MEI and NAO, but in this case the ENSO signal dominates. All categories show large, positive loadings indicating that the positive (El Niño) phase of ENSO tends to increase the global number of storm days at all intensities.

Factor 2 (18%) in Table 7 strongly resembles factor 3 for storm counts (Table 6), though with a stronger NAO loading. Factor 3 for storm days is not clearly related to any of the storm count factors. Factors 2 and 3 combined explain 31% of the variance, much less than the 52% explained by factor 1.

The factor analyses for individual basins, the MEI, and the NAO are shown in Tables 8 and 9 for storm counts and storm days, respectively. Factor 1 for storm counts explains 32% of the variance. Once again, both the MEI and NAO loadings are positive, and MEI dominates this mode. Thus, when El Niño is active, and to a lesser degree the NAO is positive, there is a tendency for increased storm numbers in all basins except the Atlantic. The strongest loadings are the negative Atlantic and positive northeast Pacific values, which reflect the known dipole between these basins that is associated with ENSO. Factor 1 for storm days (Table 9) is almost identical to factor 1 for storm counts.

Factor 2 for basin storm counts is likewise similar in most respects to the second factor for storm days. These second factors explain 25% and 21% of their respective variances, and they each represent a mode dominated by the NAO with a smaller, negative MEI loading. The second factors are associated with decreased storm counts and days in both the north Indian Ocean and Southern Hemisphere during the positive NAO. However, this NAO mode has very different effects on storm numbers and on storm days in two basins: the northwest Pacific and the North Atlantic. In the Atlantic a positive NAO is associated with normal storm numbers but below-normal storm days. This indicates a shorter mean storm lifetime during the positive phase of the NAO, which is consistent with the observations of Elsner (2003) that the positive NAO is associated with an eastward shift in the longitude of (and hence earlier time of) hurricane recurvature in the Atlantic. In the northwest Pacific this positive-NAO mode is associated with an elevated number of tropical cyclones but only an average number of storm days in the basin. This indicates a decreased average storm lifetime, as in the Atlantic. The reason for the shortened lifetimes in the northwest Pacific is not clear at this time.

Summarizing the results of the four factor analyses, the strongest factor in each case occurs when the MEI and NAO are of the same sign with ENSO tending to dominate. They appear to all be describing a similar physical mode of large-scale variability. When both indices are positive (El Niño phase of ENSO, high pressure in the Azores region), there is a strong tendency toward increased numbers of storms and storm days globally as well as an increase in the average intensity of the storms. Spatially, the first factors show that the positive MEI and NAO phase is associated with increased storm activity in all basins except the Atlantic, which instead shows a sharp decrease in storm activity. The combined effects of these factors show the dominance of positive correlations in storm activity between basins and hence explain why the 80-cyclone myth proves to be false.

The second factors in each of Tables 7 –9, as well as the third factor of Table 6, all describe modes of variability in which the NAO loadings are positive, the MEI is negative, and the NAO correlations have larger magnitudes than the MEI. They too seem to be analyzing a single physical mode. When the NAO phase is positive, these factors are associated with decreased storm activity at the extremes (fewer TS and category 4–5 storms) with increased activity in categories 1–3. They also are associated with decreases in storm activity in the IO and SH as well as increased numbers of storms (but not storm days) in the WP.

5. Conclusions

a. The 80-cyclone myth

This study analyzes tropical cyclone interannual variability from 1985 to 2003. The variability of the total number of global storms with maximum winds greater than 17 m s−1 is found to be indistinguishable from the variability that would arise if the storms in each basin formed randomly in time with the variability observed in that basin. There is no observed tendency for above-normal cyclone activity in one basin to be compensated by net below-normal activity averaged over the remaining basins when all storms are considered. However, there is a tendency for the interannual variability of more intense storms (categories 3–5) to be greater than would be predicted from independent variations in the five basins. This is particularly true for the strongest storms (categories 4–5). Hence, the myth of the 80 cyclones is false.

The results suggest that tropical cyclones do not have the broad-reaching effects on the global circulation required for storms in one basin to inhibit or enhance storm activity in another. Rather, they seem to form when conditions permit, and the occurrence of favorable conditions is modulated by larger-scale changes in the tropical circulation. A tropical cyclone may produce some feedback to future storm activity within its own basin during the current storm season, but not to other basins.

Perhaps this finding should not be too surprising. Earlier studies have indicated that within the atmosphere the larger-scale effects of a tropical cyclone are not dramatic. For example, Frank (1977, 1982) found that typhoons export kinetic energy in the upper troposphere, perhaps enough to account for a few percent of the total in limited zonal bands when they are active. Tropical cyclones may increase total rainfall within the Tropics by increasing the evaporation rate. (The anomalous rainfall does not increase the mass flux through the storm, but rather raises the outflow level. This probably affects the tropical tropopause.) They may have more significant and long-lasting effects within the ocean, but we suggest that these would probably be too slow to have strong effects in other basins during the current storm season.

The strongest link between the rate of tropical cyclone formation and the global circulation may be related to the time required to build up enough vorticity in a monsoon trough region to promote instability and hence cyclogenesis. Globally, the majority of tropical cyclones form in or near monsoon troughs. The authors speculate that understanding the temporal cycling of this portion of the Hadley circulation would be a promising avenue to better understanding why the number of cyclones is around 80 per year.

b. Modes of interannual variability

Correlation and factor analyses show that there are strong spatial and intensity relationships between storm activity in different basins, and that significant amounts of this variability are related to ENSO and the NAO, particularly ENSO. The strongest mode of interannual variability in storm numbers and days is one in which the MEI and NAO are of the same phase (e.g., both are positive when El Niño occurs with a positive pressure anomaly in the Azores) with the ENSO signal tending to dominate. When the indices are positive, there is a strong tendency for increased storm numbers and days in all basins except the Atlantic, where activity is suppressed. As a result, there are also increases in the global numbers of storms and storm days. The positive phase of this mode results in increased frequency of tropical storms intensifying into hurricanes as well as a greater fraction of hurricanes reaching the highest intensity categories.

There appears to be at least one weaker but still significant mode in which the NAO loading is stronger than ENSO. This mode does not show a strong effect on overall storm numbers or intensities, but it does reflect a tendency toward increased storm numbers in the northwest Pacific and reduced numbers in the North Indian Ocean and Southern Hemisphere. The reasons for these correlations are unclear. The NAO-dominated mode is also associated with decreased tropical cyclone lifetimes in both the North Atlantic and northwest Pacific.

c. Differences between strong and weak cyclone behavior

There appear to be important differences between the modes of interannual variability of weaker tropical cyclones (TS–category 2) and of stronger cyclones (categories 3–5). The strong-storm group exhibits stronger correlation patterns between basins than does the weaker group. Further, the two intensity groups correlate quite differently with ENSO and with the NAO.

These results suggest that the dominant modes of interannual variability of tropical cyclones tend to modulate the distribution of intensities within a basin as well as the number of storms. They also suggest that the physical processes that affect the number of storms formed are, to some significant degree, distinct from those that affect the intensities that storms reach. Storm formation numbers largely reflect the frequency with which a basin can develop low-level vorticity centers over the vast regions that are thermodynamically adequate for genesis. In contrast, variations in storm intensity patterns are more likely to result from changes in sea surface temperatures and winds over the large areas traversed by mature cyclones. While both formation and maximum intensity may be related to the same variables (such as sea surface temperature), they may be related in ways that are physically and spatially different.

Acknowledgments

This work was supported by the National Science Foundation, Grant ATM-0233881, and the National Aeronautics and Space Administration, Grant NNG05GQ64G.

REFERENCES

  • Bell, G. D., and M. Chelliah, 2006: Leading tropical modes associated with interannual and multidecadal fluctuations in North Atlantic hurricane activity. J. Climate, 19 , 590612.

    • Search Google Scholar
    • Export Citation
  • Chan, J. C. L., 1985: Tropical cyclone activity in the Northwest Pacific in relation to the El Niño/Southern Oscillation phenomenon. Mon. Wea. Rev., 113 , 599606.

    • Search Google Scholar
    • Export Citation
  • Chan, J. C. L., 2000: Tropical cyclone activity over the western North Pacific associated with El Niño and La Niña events. J. Climate, 13 , 29602972.

    • Search Google Scholar
    • Export Citation
  • Chu, J-H., C. R. Sampson, A. S. Levine, and E. Fukada, 2002: The Joint Typhoon Warning Center tropical cyclone best-tracks, 1945–2000. Naval Research Laboratory Tech. Rep. NRL/MR/7540-02-16, 112 pp.

  • Elsner, J. B., 2003: Tracking hurricanes. Bull. Amer. Meteor. Soc., 84 , 353356.

  • Elsner, J. B., and B. Kocher, 2000: Global tropical cyclone activity: A link to the North Atlantic Oscillation. Geophys. Res. Lett., 27 , 129132.

    • Search Google Scholar
    • Export Citation
  • Emanuel, K. A., 2005: Increasing destructiveness of tropical cyclones over the past 30 years. Nature, 436 , 686688.

  • Frank, W. M., 1977: The structure and energetics of the tropical cyclone. II. Dynamics and energetics. Mon. Wea. Rev., 105 , 11361150.

    • Search Google Scholar
    • Export Citation
  • Frank, W. M., 1982: Large scale characteristics of tropical cyclones. Mon. Wea. Rev., 110 , 572586.

  • Gray, W. M., C. W. Landsea, P. W. Mielke Jr., and K. J. Berry, 1993: Predicting Atlantic basin seasonal tropical cyclone activity by 1 August. Wea. Forecasting, 8 , 7386.

    • Search Google Scholar
    • Export Citation
  • Henderson-Sellers, A., and Coauthors, 1998: Tropical cyclones and global climate change: A post-IPCC assessment. Bull. Amer. Meteor. Soc., 79 , 1938.

    • Search Google Scholar
    • Export Citation
  • Lander, M. A., 1994: An exploratory analysis of the relationship between tropical storm formation in the western North Pacific and ENSO. Mon. Wea. Rev., 122 , 636651.

    • Search Google Scholar
    • Export Citation
  • Lander, M. A., and C. P. Guard, 1998: A look at global tropical cyclone activity during 1995: Contrasting high Atlantic activity with low activity in other basins. Mon. Wea. Rev., 126 , 11631173.

    • Search Google Scholar
    • Export Citation
  • Pielke Jr., R. A., C. Landsea, M. Mayfield, J. Laver, and R. Pasch, 2005: Hurricanes and global warming. Bull. Amer. Meteor. Soc., 86 , 15711575.

    • Search Google Scholar
    • Export Citation
  • Simpson, R. H., 1974: The hurricane disaster potential scale. Weatherwise, 27 , 169186.

  • Wang, B., and J. C. L. Chan, 2002: How strong ENSO events affect tropical storm activity over the western North Pacific. J. Climate, 15 , 16431658.

    • Search Google Scholar
    • Export Citation
  • Webster, P. J., G. J. Holland, J. A. Curry, and H-R. Chang, 2005: Changes in tropical cyclone number, duration, and intensity in a warming environment. Science, 309 , 18441846.

    • Search Google Scholar
    • Export Citation

APPENDIX

Raw Data for Each Year and Basin

Raw data showing the storm counts and storm days for each year and each basin (Tables A1 –A4).

Fig. 1.
Fig. 1.

Time series of the global annual number of tropical cyclones (counts) for four intensity categories: tropical storms, weaker hurricane-strength tropical cyclones (categories 1–2), stronger tropical cyclones (categories 3–4), and the most intense storms (category 5).

Citation: Monthly Weather Review 135, 10; 10.1175/MWR3435.1

Fig. 2.
Fig. 2.

As in Fig. 1, but for the annual global number of storm days in the four intensity groups.

Citation: Monthly Weather Review 135, 10; 10.1175/MWR3435.1

Fig. 3.
Fig. 3.

Time series of the MEI ENSO index, the annual NAO index, the annual global number of all tropical cyclones (TS–category 5), and the global number of very intense tropical cyclones (categories 4–5) for the period 1985–2003.

Citation: Monthly Weather Review 135, 10; 10.1175/MWR3435.1

Table A1. Storm days for each intensity category, summed over all five basins.

i1520-0493-135-10-3587-ta01

Table A2. As in Table A1, but for storm counts (number of storms).

i1520-0493-135-10-3587-ta02

Table A3. Storm days in each basin, summed for all six intensity categories.

i1520-0493-135-10-3587-ta03

Table A4. As in Table A3, but for storm counts.

i1520-0493-135-10-3587-ta04
Table 1.

Average global storm counts and storm days for each intensity category (1985–2003).

Table 1.
Table 2.

Annual global number of storms for three intensity groups: the std dev of these means over the 19-yr sample, the std dev of the 1000 random draws for 19 yr, and the percentage of the random draws whose std dev exceeds the std dev of the observed totals.

Table 2.
Table 3.

As in Table 2, but for storm days.

Table 3.
Table 4.

(a) Correlations between annual storm numbers in the five basins for all storms (TS–category 5). Negative correlations are in italics, and any correlations with an absolute value of 0.46 or above are in boldface font. (b) As in (a), but for TS–category 2 storms. (c) As in (a), but for category 3–5 storms.

Table 4.
Table 5.

(a) As in Table 4a, but for storm days. (b) As in (a), but for TS–category 2 storms only. (c) As in (a), but for category 3–5 storms only.

Table 5.
Table 6.

Factor loadings (correlation coefficients) for the global storm counts by storm intensity category and for the MEI and NAO indices. Negative loadings are in italics, and those with loadings of 0.46 or greater are in boldface.

Table 6.
Table 7.

As in Table 6, but for storm days.

Table 7.
Table 8.

Factor loadings for the basin storm counts (TS–category 5) and the MEI and NAO indices. Only factors that explain at least 10% of the variance are shown. Negative loadings are in italics, and those with absolute values of 0.50 or greater are in boldface.

Table 8.
Table 9.

As in Table 8, but for storm days.

Table 9.
Save