Variability in tropical cyclone (TC) activity is a matter of direct concern for affected populations. On interannual and longer time scales, variability in TC passage frequency can be associated with total TC frequency over the concerned ocean basin [basinwide frequency (BF)], the spatial distribution of TC genesis in the basin [genesis distribution (GD)], and the preferable track (PT) that can be considered as a function of genesis locations. To facilitate investigation of mechanisms responsible for the variability, the authors propose an approach of decomposing anomalies in the passage frequency into contributions of variability in BF, GD, and PT, which is named the Integration of Statistics on TC Activity by Genesis Location (ISTAGL) analysis. Application of this approach to TC best track data in the western North Pacific (WNP) basin reveals that overall distribution of the passage frequency trends over the 1961–2010 period is mainly due to the PT trends. On decadal time scales, passage frequency variability in midlatitudes is primarily due to PT variability, while the BF and GD also play roles in the subtropics. The authors further discuss decadal variability over the East China Sea in detail. The authors demonstrate that northward shift of the PT for TCs generated around the Philippines Sea and westward shift for TCs generated in the eastern part of the WNP contribute the variability with almost equal degree. The relationships between these PT shifts and anomalies in environmental circulation fields are also discussed.
Tropical cyclones (TCs) are generated over tropical and subtropical oceanic regions and travel long distances toward landmasses and midlatitudes, causing natural disasters on their way. The severe impacts of the TCs have led to numerous efforts to understand average TC behavior and variability and to perform future projections.
In this study, we examine variabilities in TC passage frequency, which is defined as the number of the TCs observed in a prescribed area or segment within a given period of time. This variable is not related to the duration of TCs staying in the area; they are counted only once during their duration. Frequency of TC landfall on a particular country is a typical example of the passage frequency.
Variability in the passage frequency is caused by several aspects of TC behavior. In the first place, basinwide TC frequency (BF) influences the passage frequency. Second, spatial distribution of TC genesis frequency in the basin [genesis distribution (GD)] also influences the passage frequency, since a preferable track (PT) that TCs tend to take depends strongly on their genesis locations. Third, the PT from a fixed region may also vary because of variations of mean translation velocity of TCs in individual locations.
Such variabilities are strongly influenced by environmental conditions. Several dynamic and thermodynamic parameters are proposed for the diagnosis of TC genesis conditions, such as lower-tropospheric relative vorticity, vertical wind shear, tropospheric humidity, and thermal instability (e.g., Gray 1975; Emanuel and Nolan 2004; Camargo et al. 2007a). TC translation is considered to be primarily due to the advection of TC potential vorticity anomalies by environmental flow averaged over the deep troposphere or the steering flow (George and Gray 1976). Most of the studies dealing with TC translation variability have discussed its mechanisms in terms of those of the steering flow (Ho et al. 2004; Wu et al. 2005; Tu et al. 2009).
To discuss mechanisms responsible for passage frequency variability from the viewpoint of the environmental conditions, it will be fruitful to decompose the variability into contributions from those in the BF, GD, and PT. In this paper, we will propose a method of this decomposition. There have been similar attempts to facilitate discussion of the passage frequency variability. Ho et al. (2004), Kim et al. (2005), and Liu and Chan (2008) examined variability in the passage frequency divided by the BF for individual years, and thus it can be said that these studies separated the contribution of the BF from those of the GD and PT. Camargo et al. (2007b,c) categorized TCs into seven clusters in terms of their track and examined number of cluster members to discuss variability in the TC activity. Differences between the clusters are mainly related to the GD and overall feature of the PT. Wu and Wang (2004) utilized a trajectory model in which TCs are advected by a prescribed steering flow and climatological-mean beta drift to estimate the contribution of the PT in future climate. Wu et al. (2005) also used the trajectory model to discuss observed trends. Compared with these previous efforts, the method proposed here can distinguish the three contributions through solely analyzing TC track data without a help of numerical models.
Then, we will apply this method to observed variability over the western North Pacific (WNP) basin (0°–50°N, 100°E–180°), where TC occurrences are most frequent in the world. Over this basin, the passage frequency exhibits large variability with various time scales. There have been a number of studies examining interannual variability. Some studies argued that the El Niño–Southern Oscillation (ENSO) played a major role (e.g., Chan 2000; Chen et al. 2004; Camargo et al. 2007c; Kim et al. 2011), primarily due to a difference in mean genesis location between the ENSO phases (Chan 1985; Dong 1988; Lander 1994; Chia and Ropelewski 2002; Camargo et al. 2007a).
On decadal and interdecadal time scales, Yumoto and Matsuura (2001) and Matsuura et al. (2003) noted a large variability in total TC frequency in the WNP basin, with maximum values during the mid-1960s and early 1990s and minimum values during the mid-1970s. Ho et al. (2004) compared June–September passage frequency between the 1951–79 and 1980–2001 periods. Their Fig. 4 indicated that the frequency in the 1980–2001 period is significantly lower than that in the 1951–79 period over the Philippines Sea and a southern part of the East China Sea (ECS). They also showed that the frequency in the latter period is higher over a northern part of the South China Sea (SCS). In contrast, Wu et al. (2005) showed that, during the 1965–2003 period, passage frequency over the ECS in the June–October season increased, while that over the SCS decreased (their Fig. 2). The discrepancy between these two studies may be due to a difference in the analysis period, target season, and the analysis procedure. In fact, Liu and Chan (2008) showed remarkable decadal variability in the passage frequency, partly due to the Pacific decadal oscillation through modulation of the subtropical high. Recently, Tu et al. (2009) reported that the number of TCs approaching Taiwan, which is located between the SCS and ECS, has increased sharply since the year 2000. As seen above, the passage frequency variability over the ECS has attracted considerable interest of climate researchers and thus will be discussed in detail with the aid of the decomposition method proposed in this study. Our particular interest is on variability with time scales longer than the interannual scale, which is referred to here as “decadal variability” for brevity.
The reminder of the paper is organized as follows: Section 2 explains datasets analyzed in this study and the decomposition method. In section 3, we present attribution of linear trends and magnitude of decadal variability over the WNP, followed by a more detailed examination of the variability over the ECS. In section 4, we discuss probable causes of the PT variability responsible for the passage frequency variability over the ECS. Finally, conclusions of this study are presented in section 5.
2. Data and methodology
The TC best track dataset over the WNP basin analyzed in this study is provided by the Regional Specialized Meteorological Centers–Tokyo Typhoon Center, Japan Meteorological Agency (available at http://www.jma.go.jp/jma/jma-eng/jma-center/rsmc-hp-pub-eg/trackarchives.html). The dataset provides the positions of the TC centers, their intensity in terms of minimum surface pressure and maximum sustained wind, and their size for every 6 h. The dataset also includes timing information of extratropical transitions. While the dataset covers a period from 1951 to the present, we use the data for the 50-yr period of 1961–2010, primarily to ensure the data credibility.
We define the position of TC genesis as the first position where the maximum 10-min-mean wind speed exceeds 34 kt (∼17 m s−1), which is the threshold for the tropical storm (TS). The TCs are tracked until they experience extratropical transition or until their 10-min-mean wind speed falls below 34 kt.
Figure 1 outlines climatological (1961–2010 mean) TC activity. The basinwide annual TC frequency over the WNP basin is 26.5 TCs per year, which is approximately one-third of the global frequency. The TC activity exhibits clear seasonality, with a maximum monthly count in August and a minimum in February (Fig. 1a). In the peak TC season of July–September, TCs are typically generated between 5° and 30°N, with three local maxima over the SCS, northeast of the Philippines, and at around 140°E near the Mariana Islands (Fig. 1b). The TCs tend to translate northwestward over the subtropics south of 30°N and northeastward to its north (Fig. 1c). As a result of the genesis distribution and translation vector, the passage frequency has a maximum over an area east of Taiwan. Bands with high frequency extend westward over the northern SCS and northeastward toward western Japan from the maximum.
To discuss causes of variability in the passage frequency from the viewpoint of large-scale environmental fields, we analyze the 40-yr European Centre for Medium-Range Weather Forecasts (ECMWF) Re-Analysis (ERA-40; Uppala et al. 2005) for the 1961–78 period and the ECMWF Interim Re-Analysis (ERA-Interim; Dee et al. 2011) for the 1979–2010 period. Since the horizontal resolution is different between ERA-40 (2.5° × 2.5°) and ERA-Interim (1.5° × 1.5°), the latter dataset is interpolated into the resolution of the former. We also analyze the sea surface temperature (SST) of Centennial in situ Observation-Based Estimates (COBE) SST (Ishii et al. 2005), having 1° × 1° resolution.
b. Integration of statistics on TC activity by genesis location analysis
Anomalies in the passage frequency result from those in the BF, GD, and PT. Here, we propose a method of decomposing the passage frequency anomalies into these three contributions, through transformation of the passage frequency equation.
First, we define passage probability p(A, Ag), which is the probability that TCs generated in an area Ag translate to subsequently pass over an area A. In this paper, both the areas A and Ag are 5° × 5° grid boxes, unless otherwise stated. Figures 2a,b show examples of spatial distribution of climatological passage probability for TCs generated in the (10°–15°N, 125°–130°E) and (10°–15°N, 140°–145°E) areas, respectively, in the July–September period. These spatial distributions of the passage probability represent the PT of the TCs. The passage probability for the TCs generated in the former area (Fig. 2a) exhibits a band of high probability extending in a northwest direction from the genesis region to the northern SCS, indicating that most of the TCs take northwestward track without turning. On the other hand, the passage probability for the TCs generated in the latter area (Fig. 2b) indicates that the TCs tend to initially translate northwestward and subsequently turn northeast toward Japan. Differences between Figs. 2a,b indicate strong dependency of the PT on the genesis location.
The frequency of TCs generated in the area Ag to pass over the area A is equal to a product of the passage probability and the genesis frequency in Ag. Therefore, the passage frequency over A, F(A), can be expressed as
where C is the BF in the unit of number per 3-month season and g(Ag) is the ratio of the genesis frequency in Ag divided by C and referred to as the GD [note that the genesis frequency in Ag equals C × g(Ag)]. To estimate F(A), we count number of TCs that form in and/or travel through individual 5° × 5° grid boxes, unless otherwise stated. All these TC statistics can be estimated individually from the TC track data.
Using this equation, we can attribute the passage frequency anomalies in a given period with respect to the climatological average, which is defined here as the 1961–2010 mean. The passage frequency Eq. (1) for the climatological average is
where overbars indicate the climatological averages. Likewise, the equation for a given period is
where primes indicate the anomalies for the given period with respect to the climatological averages.
Equation (4) represents the decomposition of the passage frequency anomaly F′. The first term on the right-hand side represents contribution from the anomaly in the BF. It indicates how the BF anomaly varies the passage frequency under the condition that the GD and PT are unchanged. The second and third terms represent contributions from the anomalies in the GD and PT, respectively. Note that the GD anomalies contribute to the passage frequency not only in the genesis location but also along the climatological PT. The last term is the sum of the nonlinear terms (NL) with respect to the three anomalies of the BF, GD, and PT. We name this method the Integration of Statistics on TC Activity by Genesis Location (ISTAGL) analysis.
One of the issues involved in the statistical studies of TCs is to determine the target season. On one hand, we aim to set the season as long as possible to analyze tracks of a large number of TCs in order to reduce sampling errors. On the other hand, statistical behaviors of TCs tend to exhibit remarkable seasonality (e.g., Camargo et al. 2007b), which may make it difficult to interpret the statistical results if analyzed all together. Considering this situation, we examine the seasonality in climatological monthly GD and PT to determine the target season to be adopted in this study.
Table 1 shows correlation coefficients of the spatial distribution of the climatological monthly GD between months during June–November over the WNP basin with 5° × 5° resolution. The correlation coefficient reaches values as high as around 0.8 between July, August, and September, indicating that the GD in these three months is similar to each other (Figs. 3a,b). On the other hand, the GD in October, when the monthly TC frequency is as high as that in July (Fig. 1a), differs a lot from those in July, August, and September. In October, the frequent TC genesis area shifts southward compared to the preceding three months (Fig. 3c). The GD in June is also less similar to those in July–September. These features are consistent with Chia and Ropelewski (2002), who demonstrated that monthly-mean locations of TC genesis for June and October are shifted southward from those for July, August, and September.
Next, to examine similarities in the PT between months, we compare the climatological passage frequency of month m with a hypothetical passage frequency where the passage probability is replaced by that of another month m′. The hypothesized frequency can be expressed as
where subscripts represent months. To reduce sampling errors, genesis grid boxes with less than five cyclogenesis events for either month of the 50-yr period are excluded in the calculation. Table 2 shows the spatial correlation coefficient between these two passage frequencies. The correlation coefficients are higher than 0.9 between July, August, and September, while those between each of these three months and October are lower. Figure 4 suggests that the PT in October largely differs from those in the preceding three months, preferring a more southward track with less recurving TCs. This distinct feature is consistent with results of Camargo et al. (2007b). They grouped TCs into seven clusters according to their tracks and revealed that straight moving TCs in the southern part of the basin, which were classified into their cluster F, are observed frequently only during October–December.
On the basis of these results, we decided to adopt the months of July–September as the target season in this study. Note that LinHo and Wang (2002) reported that the month-to-month differences in environmental conditions in this season are also relatively small compared with those in other months.
While the ISTAGL analysis requires a large number of TCs to reduce sampling errors, TCs are relatively rare events. Therefore, it is questionable whether this method can be applied to anomalies in individual years. Instead, it seems reasonable to analyze anomalies in multiyear periods. In this study, anomalies in forty-four 7-yr periods of 1961–67, 1962–68, … , and 2004–10 are examined. Therefore, we expect that most of the fluctuation with interannual time scales is leveled out, and we will mainly discuss the decadal variability.
To check whether the 7-yr period is long enough for the ISTAGL analysis against sampling errors, we performed 10 000 series of resampling experiments, in which we randomly excluded 20% of the TCs to repeat the ISTAGL analysis. We confirmed that the results presented in this paper are qualitatively insensitive to the random resampling. We also performed another set of sensitivity tests in which the ISTAGL analysis was applied to 5-, 9-, and 11-yr-mean anomalies. We confirmed that results presented here were not affected seriously by these changes.
3. Results of the ISTAGL analysis
a. Linear trends and decadal variability
In this section, we examine linear trends and magnitude of decadal variability in the passage frequency over the entire WNP basin. We performed the ISTAGL analysis for the forty-four 7-yr periods to obtain time series of the passage frequency anomaly and the four contribution terms. Figure 5 presents their linear trends. Yellow (blue) rectangles indicate grid boxes where positive (negative) trends pass the statistical significance and robustness tests. For the statistical significance, Student's t test was adopted and trends that exceeded the 90% confidence level were considered significant. For the robustness, a trend signal is considered robust when more than 90% of the aforementioned resampling experiments exhibited trends that have the same sign as that of the signal.
In the tropics and subtropics, the passage frequency exhibits negative trends over the SCS and the Philippines (Fig. 5a). In the midlatitudes, its trends have a wavelike pattern. A significantly positive trend along 140°E lies between negative ones along 130° and 150°E. A slight positive trend can also be observed over the ECS, although it is not significant. The negative trend along 130°E is collocated with the zonal maximum of climatological passage frequency (Fig. 1c), indicating that the zonal contrast in the passage frequency is becoming less clear. The negative trend over the SCS and the wavelike pattern are broadly consistent with the findings of Wu et al. (2005).
Using the ISTAGL analysis, these trends are decomposed into the four contributions, BF, GD, PT, and NL (Figs. 5b–e). Among them, the PT contribution plays the most vital role (note that the contour intervals for the passage frequency trend and the PT contribution are twice as large as those for the other three terms). In the midlatitudes north of 25°N, the PT contribution is the primary cause of the wavelike pattern in the passage frequency trends. Both of the two positive signals over the ECS and along 140°E are significant and robust. The trends of the PT contribution north of 25°N and west of 150°E are positive on average, indicating an increased likelihood of TCs approaching East Asian countries in recent years. The trends of the BF contribution, whose spatial pattern reflects that of the climatological passage frequency by definition, are significantly negative over the entire domain, which results from a decreasing trend in the total TC frequency in the basin over the last 50 yr (Yumoto and Matsuura 2001; Matsuura et al. 2003). The trends of the GD contribution are large and significant in the eastern part of the basin. Positive trends are generally observed to the south of 30°N, accompanied by negative ones to their north. The trend of the NL contribution is larger around Japan than elsewhere, although its magnitude is generally smaller than the PT contribution.
The passage frequency exhibits a decreasing trend over northern SCS, which was also pointed out by Tu et al. (2009). They related this trend to a reduction in the local TC genesis frequency, which is consistent with the negative trend of the BF contribution. Our ISTAGL analysis further reveals that the PT contribution also plays a role in the decreasing trend of the passage frequency. Magnitudes of the trends of the BF and PT contribution are comparable to each other for this area. We will further mention variability in the PT contribution over SCS in the next subsection.
Note that the interdecadal variability in the TC activity is large over WNP and may mask long-term trends (Chan and Liu 2004; Chan 2006; Trenberth et al. 2007). Therefore, some portion of the 50-yr trends shown here may possibly be a part of this variability.
Figure 6 shows the root-mean-square (RMS) of the time series of the four contribution terms. The RMS of the PT contribution is much larger than that of the other three terms in the midlatitudes, with two maxima around eastern Japan and over the ECS. In the subtropics, in addition to the PT contribution, the BF and GD contributions are also influential. While the RMS of the PT contribution is the largest among the four terms around 120°–130°E, the RMS of the GD contribution is the largest over northern SCS, around 130°–140°E, and around 160°E. The BF contribution plays a comparable role to the PT contribution over the northern SCS and an area to the east of Taiwan. As for the NL contribution, the RMS is considerably smaller than those of the other three terms in most parts of the subtropics and midlatitudes. This suggests that the attribution of the passage frequency anomaly on decadal scales can generally be discussed in the quasi-linear framework of contributions from BF, GD, and PT.
Relative magnitude among the three linear contributions, however, depends on the size of the grids used for calculation of the passage frequency [A in Eq. (1)] and the genesis frequency [Ag in Eq. (1)]. When we performed the ISTAGL analysis using the 10° × 10° grid boxes for both A and Ag (Fig. 7), relative contribution of the PT becomes smaller in the subtropics while those of the BF and GD become larger. On the other hand, dominance of the PT contribution over the midlatitudes is still observed. A probable reason for the reduction of the PT contribution in the subtropics is that the 10° × 10° resolution is so coarse that it is sometimes difficult to resolve the track shifts there, as the degree of the shifts is expected to be small near the genesis location. In general, the adoption of coarser grids for A reduces the RMS of the PT contribution. Nevertheless, there are several aspects that are consistent with previous results with the use of the 5° × 5° grid boxes, including dominance of the PT contribution in the midlatitudes, importance of the BF and GD contributions in the subtropics, and considerably small RMS of the NL contribution.
b. Passage frequency over the ECS
In this section, we discuss attribution of decadal variability in the passage frequency over the ECS in detail. In particular, we examine the number of TCs that entered the ECS region from its southern boundary, which is defined by the segment (25°N, 120°–130°E) in this study (Fig. 9b). This segment is located just north of the maximum climatological passage frequency and is almost perpendicular to the mean translation vectors (Fig. 1c). The average number of these TCs per 7-yr period in the July–September season is approximately 20 (141 cases in 50 yr), which is approximately 40% of all TCs that pass over the 25°N latitudinal line.
We can also apply the ISTAGL analysis for the passage frequency to the southern boundary of the ECS instead of the area A. Figure 8 shows the time series of the 7-yr-mean passage frequency anomaly and the BF, GD, and PT contributions. The passage frequency exhibits a marked decadal variability superimposed on an overall increasing trend (∼1.5 in 50 yr). It exhibits three maxima at the beginning of the time series (mid-1960s), early 1990s, and mid-2000s. It is apparent that all of the three contribution terms influence the passage frequency variability. Among them, the PT contribution is the primary factor. In fact, the correlation coefficient between the passage frequency and the PT contribution is +0.75. As for the linear trend, the increasing trend of the passage frequency is solely due to the PT contribution, consistent with Fig. 5.
The gray bands in Fig. 8 indicate one standard deviation of the variables obtained in the resampling experiment described in section 2b, which can be considered as a measure of sampling error. The standard deviation of the PT contribution is large and comparable to that of the passage frequency, while those of the BF and GD contributions are relatively small. This suggests that the major source of the sampling error in the passage frequency comes from the PT contribution.
Next, we discuss possible mechanisms responsible for the PT contribution. Considering the mean translation vector, we can think of two possibilities of track shifts that lead to the PT contribution for the ECS. The first one is that an increase in the probability of TCs moving into the ECS is associated with a decrease in the probability of TCs moving into the SCS. This is probably due to a northward shift of the track east of the Philippines. The other possibility is that the increase is associated with a decrease in the probability of TCs moving into the oceanic areas to the east of the ECS (EECS), due to a westward shift of the track southeast of the ECS. We now discuss the relative importance of these two track shifts utilizing the ISTAGL analysis.
Since only 4% of the TCs passing over the southern boundary of the ECS are generated in the SCS, hereafter we focus on the TCs generated in the tropical WNP (0°–25°N, 120°E–180°) region. Most of these TCs leave the region via the segment (5°–25°N, 120°E) (see the bold line in Fig. 9a) to move into the SCS, via the segment (25°N, 120°–130°E) to move into the ECS, or via the segment (25°N, 130°E–180°) (Fig. 9c) to move into the EECS. Approximately 6% of the TCs weaken to lose their TS intensity before reaching the boundary, which will be excluded from the following analysis. To examine relationship between genesis location and the destinations, Figs. 9a–c shows climatological passage probability to the SCS, ECS, and EECS, through respective segments, as a function of genesis location. High passage probability (≥60%) toward the SCS in the southwestern part of the region (Fig. 9a) indicates that majority of the TCs generated in this region move toward the SCS. In contrast, the TCs generated in the northeastern part tend to move toward the EECS (Fig. 9c). Apart from the area just south of the ECS, the passage probability toward the ECS (Fig. 9b) is generally smaller than one of the other two passage probabilities.
On the basis of Figs. 9a–c, we subjectively define two subregions, W and E, as shown in Fig. 9d. The boundary between these regions is nearly parallel to the mean translation vector (Fig. 1c). Figures 9e,f present climatological passage probability for TCs generated in the regions W and E, respectively. In these figures, for example, the passage probability of TCs generated in region W moving into the SCS is represented as p(SCS, W), as in section 2b. About half of the TCs generated in region W move toward the SCS, while about 60% of the TCs generated in region E move toward the EECS. We will investigate the northward and westward shifts of the preferable track by examining the passage probability variability for the TCs generated in regions W and E, respectively. The climatological TC genesis frequency for regions W and E are 3.9 and 3.8 (3 months)−1, respectively.
Figure 10 shows time series of the passage probability for the two genesis regions. Both p(ECS, W) and p(ECS, E) exhibit large decadal variability. Their amplitudes are comparable to each other and have increased since the 1980s. The sum of these two time series is well correlated with the PT contribution shown in Fig. 8d, with a correlation coefficient of +0.76. Note that p(ECS, W) leads several years to p(ECS, E), while detailed discussion on this feature remains for future study. As for the linear trend (shown by thin lines in Fig. 10), both passage probabilities exhibit increasing trends over the 50 yr, with p(ECS, W) showing a higher rate (+24% in 50 yr) than p(ECS, E) (+18% in 50 yr).
For region W, the decadal variability in p(ECS, W) is clearly related to that in p(SCS, W), with a large negative correlation coefficient of −0.80. The p(SCS, W) also exhibits a decreasing trend (−24% in 50 yr). During the last decade, p(ECS, W) has become slightly higher than p(SCS, W). The correlation coefficient is high and negative (−0.60) even after removing their linear trends. The association between the increase in p(ECS, W) and the decrease in p(SCS, W) implies the northward shift of the track. For region E, the variability in p(ECS, E) has a clear negative correlation with that in p(EECS, E) with a correlation coefficient of −0.87 (−0.86 for the detrended time series). This implies the westward shift of the track.
These results suggest that the PT contribution for the passage frequency over the ECS is due to both the northward and westward shifts of the track. These two types play equally important roles in the decadal variability and the 50-yr trend. Moreover, the amplitudes of the decadal variability in the two track shifts have increased since the 1980s. Although the definition of the boundary between regions W and E is subjective, we confirmed that slight shifts of the boundary in zonal direction did not qualitatively affect the results shown here.
As argued in the last section, two types of the track shifts play key roles in the decadal variability of PT contribution for the passage frequency to the ECS. In this section, we discuss relationships between these shifts and environmental circulation fields by using regression analyses. Since a decrease in p(SCS, W) indicates an increase in the probability of TCs taking a more northward track, we adopt −p(SCS, W) as a key variable representing the northward shift. Likewise, we adopt −p(EECS, E) for the westward shift.
In general, the track shifts are considered to be influenced primarily by changes in the steering flow (Ho et al. 2004; Wu et al. 2005; Kim et al. 2005; Tu et al. 2009), as introduced in section 1. In this paper, the steering flow is defined as a mass-weighted vertical average of horizontal wind from 850- to 200-hPa levels. Figure 11a shows the 7-yr-mean July–September steering flow anomaly regressed on −p(SCS, W). The black arrows indicate that the zonal and/or meridional components of the regressed flow are statistically significant. Northerly and westerly wind anomalies are found over southern China and the northern SCS (north of 20°N), respectively, forming a cyclonic anomaly. The westerly anomaly extends eastward to 140°E, although the signals are not significant. In vertical, a significant westerly anomaly can be found throughout the troposphere over the northern SCS, with its amplitude larger in the upper troposphere (Fig. 11b). The westerly anomaly probably contributes to the northward shift of the track in an indirect way. The anomaly tends to prevent TCs from moving farther westward under easterly climatological steering flow. Meanwhile, TCs tend to take the northward track primarily because of the beta drift and climatological southerly steering flow, thus moving toward the ECS. The cyclonic anomaly can also be observed in the difference between the 1965–83 and 1984–2003 periods, examined by Wu et al. (2005). They discussed that the cyclonic anomaly was probably associated with the distinctive tropospheric cooling trend over East Asia (Yu et al. 2004).
The 7-yr-mean steering flow anomaly regressed on −p(EECS, E) (Fig. 12a) shows an anticyclonic circulation anomaly, with its center at around 30°N, 132.5°E. This is accompanied by easterly and southeasterly anomalies, which are consistent with the westward shift of the track. Furthermore, this anticyclonic anomaly is accompanied by a local maximum in the regressed geopotential height anomaly at the 500-hPa level (figure not shown), indicating a westward extension of the North Pacific high. This westward extension is also found in the linear trend (Wu et al. 2005) and changes associated with the climate shift in the late 1970s (Ho et al. 2004; Zhou et al. 2009). The vertical structure of the regressed zonal wind averaged over 130°–135°E (Fig. 12b) shows easterly anomalies in the subtropics (15°–25°N) throughout the troposphere, with a slight southward tilt with height. In the tropics (0°–10°N), the vertical structure is characterized by the first baroclinic structure. In the lower troposphere, the westerly and easterly anomalies in the tropics and subtropics, respectively, constitutes a positive relative vorticity anomaly in between, indicating that the monsoon trough is deeper than normal.
The regressed SST field (Fig. 12c) exhibits a positive maximum located near the dateline on the equator, with positive anomalies expanding northeastward and southeastward. The spatial structure of the regressed SST is reminiscent of the decadal ENSO (Wang and Picaut 2004). In fact, the correlation coefficient between −p(EECS, E) and the 7-yr running mean SST anomalies over the Niño-4 region (5°S–5°N, 160°E–150°W) is +0.66, which is above the 95% confidence level. Furthermore, the circulation field anomalies regressed on the 7-yr-mean Niño-4 SST (figure not shown) is qualitatively similar to that shown in Figs. 12a,b.
Anomalies in the TC passage frequency are attributable to three aspects of the statistical behavior of TCs: the basinwide TC frequency (BF), the spatial distribution of the genesis frequency in the basin [genesis distribution (GD)], and the preferable track (PT) that TCs tend to take after their genesis. They are influenced by different aspects of environmental conditions. Therefore, the decomposition of the passage frequency anomalies into contributions of anomalies of the BF, GD, and PT will facilitate discussion on mechanisms responsible for the overall passage frequency anomalies. This study proposes a new approach, called ISTAGL analysis, that can perform the decomposition by analyzing solely the TC track data without any information such as environmental circulation fields.
We apply the ISTAGL analysis to the observed decadal variability in the WNP basin over the 1961–2010 period. The target season selected is July–September, because the climatological GD and PT do not vary significantly during this season. The ISTAGL analysis reveals that most of the spatial distribution of the linear trends in the passage frequency is attributed to the PT contribution. Furthermore, the PT contribution plays a dominant role in the decadal variability in the midlatitudes. In the subtropics, not only the PT but also the BF and GD contributions play roles. The nonlinear contribution is only secondary in most parts of the WNP, which implies that the attribution of the decadal variability and trends in the passage frequency can generally be considered on a linear framework of BF, GD, and PT contributions.
The variability in the passage frequency over the southern boundary of the ECS is discussed in detail. The passage frequency exhibits the positive trend with decadal variability, which is mainly attributed to the PT contribution. There are two types of PT shifts that lead to an increase in the passage frequency over the ECS. The first type is the northward track shift for TCs generated in the western part of the tropical WNP region, and an increase in the probability of TCs moving toward the ECS is associated with a decrease in the probability of TCs moving toward the SCS. The other is the westward track shift for TCs generated in the eastern part of the region, which is associated with a decrease in the probability of TCs moving toward the EECS. We quantitatively demonstrate that these two types play comparable roles in the positive trend and the decadal variability of the passage frequency. These two track shifts indicate that more TCs generated within the wider area of the WNP have moved toward the ECS in recent years than in the past.
The mechanisms of the two types of the track shifts are briefly discussed from the viewpoint of the steering flow variability. The regression analysis with an index representing the northward track shift reveals the westerly anomaly of the regressed flow just south and southwest of the ECS. This indirectly contributes to the northward track shift through prevention of the TCs from moving farther westward to reach the SCS. As for the westward track shift, the regressed steering flow anomaly is characterized by the anticyclonic circulation with its center to the east of the ECS, causing easterly and southeasterly anomalies south and southeast of the ECS, respectively, which directly contributes to the westward track shift. The associated SST anomaly and anticyclonic steering flow anomaly are reminiscent of the decadal ENSO.
Compared to the previous studies that addressed the attribution of the passage frequency anomalies, as reviewed in the introduction, a merit of our ISTAGL analysis is that it can directly distinguish between BF, GD, and PT contributions without using numerical models. Furthermore, we can evaluate the nonlinear contribution to judge whether we can discuss the attribution of the passage frequency anomalies in the linear framework of the BF, GD, and PT contributions.
Through the presentation of the application of the ISTAGL analysis in this paper, we demonstrate that this analysis is an effective approach for discussing the attribution of the decadal variability in the passage frequency over the WNP basin. In principle, the ISTAGL analysis can decompose differences between two passage frequency fields, as long as a sufficiently large number of TC samples are considered. Therefore, we foresee further applications of the ISTAGL analysis, such as composite analysis of El Niño and La Niña years and evaluation of the credibility of projected future trends. These analyses will be conducted in our future studies.
This study is supported by the Global Environmental Research Fund (S-5-2 and A1201) of the Ministry of the Environment, Japan, and a Grand-in-Aid for Scientific Research (B-21310120) from the Ministry of Education, Culture, Sports, Science and Technology, Japan.
Current affiliation: Japan Agency for Marine–Earth Science and Technology, Yokosuka, Japan.