1. Introduction
Tropical cyclones (TCs) carry heat and energy from the tropics to higher latitudes and also bring precipitation regulating global climate. However, they are treated as one of the biggest natural threats to life and property. There are many factors such as weather conditions, beta effect, and sea surface temperature (SST) that affect TC tracks (Fisher 1958; Chan and Gray 1982; Franklin et al. 1996; Chan 2005; Fisher 1958). Large-scale circulation of atmosphere can effectively influence TC surrounding flow and then TC motion (Chan and Gray 1982). TCs and their environment can also interact to modify the surrounding flow (Wu and Emanuel 1995), and then the vortex is steered.
TC trajectories vary on the seasonal (Gray 1979; Harr and Elsberry 1991), interannual (Chan 1985), and decadal time scales (Ho et al. 2004; Yang et al. 2018). Accumulation of historical TC data makes research on TC track variabilities on long time scales possible. El Niño–Southern Oscillation (ENSO) is the most important climate variability to influence TC tracks on interannual time scale. Besides the southeastward displacement of TC genesis, a remarkable decrease in the northwest quadrant of the western North Pacific (WNP) was found in boreal summer and fall during the El Niño developing years (Chia and Ropelewski 2002; Wang and Chan 2002; Camargo et al. 2007b). There are longer-living and more intense TCs in El Niño years and the contrary in La Niña years (Camargo and Sobel 2005). The tracks tend to show more northward movements and less landing in El Niño years, but in La Niña years there are more possibilities of landing. Recent studies further revealed that different types of ENSO [eastern Pacific (EP) ENSO and central Pacific (CP) ENSO] have different effects on TC track patterns in the WNP (Mei et al. 2015; Zhao and Wang 2016) by corresponding SST, vertical wind shear (VWS) and midlevel relative humidity.
For longer time scale variability of TC tracks, Liu and Chan (2008) showed that the first three empirical orthogonal function (EOF) modes of typhoon (strong TCs in the WNP are also called typhoons) track density from 1960 to 2005 are all associated with decadal variations. For example, the first mode is related to a north–south dipole distribution of 500-hPa geopotential anomalies, which affects the intensity of the subtropical high and the westward midlatitude steering flow, and thus affects TC tracks. This dipole pattern corresponds to the Pacific decadal oscillation (PDO). Landing or not is an important factor of TC tracks. The landfall activity of typhoons along the coast of China during July–September (JAS) exhibits significant interdecadal variation, which is caused by the combined effects of the PDO and SST changes in the tropical Indian Ocean and the western Pacific (IO-WP) on atmospheric circulation. Besides, TC genesis and intensity also show the feature of decadal variation. The correlation between the El Niño Modoki index and TC genesis has increased since 1998 when the PDO switched from a warm to cold phase and more La Niña and CP El Niño events occur. Compared with 1979–97, the reduced low-level relative vorticity and increased VWS caused fewer TCs in 1998–2015. Thus, the genesis demonstrates multidecadal oscillations, especially for intense TCs (Lee et al. 2012). For TC intensity, Chan (2008) has revealed that the category 4 and 5 typhoons on the Saffir–Simpson (Saffir 1977; Simpson and Riehl 1981) scale from 1960 to 2005 vary with a period of 16–32 years, suggesting that variations of the planetary-scale oceanographic and atmospheric conditions on similar time scales govern the formation, intensification, and movement of TCs.
Decadal or longer time scale climate variability can also exert influence on TC activity by modulating the relationship between TC activity and interannual climate variability (Girishkumar et al. 2015; Wang and Liu 2016; Zhao and Wang 2019). The interannual relationship between ENSO and annual rapid intensification (RI) number of TCs in the warm PDO phases is strong and statistically significant. In the cold PDO phases, however, there is no significant correlation between ENSO and RI on the interannual time scale. The enhancement of the interannual ENSO–RI relationship in the warm PDO phases is mainly attributable to the change of the environmental vertical wind shear (Wang and Liu 2016). Zhao and Wang (2019) showed a significant reduction of the number of TCs and a stronger interannual relationship between ENSO and TCs since 1998, in a cold phase of the PDO, which can be ascribed to changes of low-level relative vorticity, vertical wind shear, and midlevel relative humidity. However, such modulation of decadal variability of TC tracks has not been discussed.
This present study focuses on TC track variabilities and addresses how the PDO modulates the relationship between TC tracks and ENSO. The datasets and methods are described in section 2. In section 3 we analyze the results of TC track cluster analysis. Section 4 discusses the EOF analysis of TC track density. A summary and discussion are presented in section 5.
2. Data and methods
a. Data
The best track data in the WNP of 1950–2017 are from the Joint Typhoon Warning Center (JTWC) best track dataset (Chu et al. 2002; https://www.metoc.navy.mil/jtwc/jtwc.html?western-pacific). Following Camargo et al. (2007a), our analyses only consider TCs with tropical storm intensity or higher and tropical depressions (TDs) are not included. According to the intensity of TCs on the Saffir–Simpson scale (Saffir 1977; Simpson and Riehl 1981), we define TCs as follows: tropical storms (TSs), typhoons (TYs; categories 1–2), and intense typhoons (ITYs; categories 3–5). Totally we have 1747 typhoons in our work. We defined landfall where the center of the TC intersects the coast, which could be an island (Camargo et al. 2007a) that is larger than 100 km2 due to TC data spatial resolution.
The data quality prior to 1970s was considered low because there was no satellite observation available. Some of the weaker storms may be missing, especially for the storms over the open ocean. In this study we repeated the analysis using the data after 1979 only and the results are similar. Therefore, our conclusions are robust and do not depend on the time period chosen. Moreover, although the variables of TCs from the best track datasets by different agencies differ slightly from each other (Lowry 2008), our cluster analysis still shows robust conclusions for different datasets.
The daily large-scale atmospheric environment variables are from the U.S. National Centers for Environmental Predication–National Center for Atmospheric Research (NCEP–NCAR) with a spatial resolution of 2.5° × 2.5° (Kalnay et al. 1996) from 1950 to 2017. Anomalies are defined relative to the whole period climatology of 1950–2017. Before analysis, a 3-day running mean is performed to remove higher-frequency information of the data.
We use the monthly Extended Reconstruction Sea Surface Temperature (ERSST) v5 (Huang et al. 2017) from the National Climatic Data Center (NCDC) of the National Oceanic and Atmospheric Administration (NOAA). Anomalies are defined relative to the study period climatology of 1950–2017. The Niño-3.4 index is the average of SST anomaly within the Niño-3.4 box (5°S–5°N, 170°–120°W; Barnston et al. 1997). Like Camargo et al. (2007b) and Goddard and Dilley (2005), we define El Niño and La Niña years according to the value of the Niño-3.4 index averaged over the months of July–October (JASO), spanning the peak of the typhoon season. The years (25% of the 68-yr period) with the largest and smallest 17 values of JASO Niño-3.4 index in the period 1950–2017 are defined as El Niño and La Niña years, respectively; the remaining 34 years are classified as neutral years. The PDO index (https://psl.noaa.gov/data/correlation/pdo.data) is used for defining the PDO phases (warm/cold phase), which is constructed by using the monthly SST collected by Physical Sciences Division (PSD) of Earth System Research Laboratory (ESRL) in NOAA. The PDO index is defined as the leading principal component of monthly SST anomalies in the North Pacific Ocean poleward of 20°N (Mantua et al. 1997).
b. Method
Generally, studies classify the trajectories to form several types and each type shares common features. Thus, complex tracks analysis can be simplified among types. There are many ways to do this classification, such as subjectively classifying the typhoons into two principal track types (Sandgathe 1987; Harr and Elsberry 1991; Lander 1994, 1996), namely recurving and straight moving track types. Objective methods are also used to classify TC trajectories. The K-means method (MacQueen 1967) has been used to study the WNP TCs (Elsner and Liu 2003), North Atlantic TCs (Elsner 2003), and North Atlantic extratropical cyclone trajectories that are of at least 3 days duration, and 6-h interval latitude–longitude positions were converted into 24-dimensional vectors to cluster (Blender et al. 1997). Harr and Elsberry (1995) used fuzzy cluster analysis and EOF to describe the three clusters of TC tracks in the WNP: straight-moving, recurving, and north oriented tracks. We use a technique that was first developed by Gaffney and Smyth (1999) and widely used in previous studies (Camargo et al. 2007a,b, 2008; Kossin et al. 2010; Mei and Xie 2016) to group typhoons into different clusters based on the geographic locations of their genesis and the subsequent tracks (Gaffney and Smyth 1999, 2005; Gaffney 2004). The description of this technique parallels that of Camargo et al. (2007a) and Mei and Xie (2016). The finite mixture model (e.g., Everitt and Hand 1981) provides the main basis for the clustering method. It uses a convex linear combination of component density functions to represent the data distribution and it models highly non-Gaussian (or multimodal) densities by a set of basic component densities. Here the regression mixture models replace the marginal component densities with conditional density components to extend the standard mixture modeling framework. The new conditional densities are functions of the TC’s position conditioned on an independent variable (i.e., time). The component densities model a cyclone’s longitudes and latitudes versus time through quadratic polynomial regression functions, which are used to fit the geographical “shape” and initial TC position of the trajectories (Gaffney et al. 2007; Gaffney 2004). The probabilistic approach and mixture model framework allow the component probability density function to be defined on nonvector data and easily accommodate TC tracks of different lengths. Each trajectory (i.e., each cyclone track) is assumed to be generated by one of K different regression models, each having its own shape parameters. The model is fit to the data by maximizing the likelihood of the parameters, given the dataset. Each track can be assigned to the mixture component (and thus the cluster) that was most likely to have generated that track given the model. In other words, the assigned cluster has the highest posterior probability given the track. More detailed discussion of the cluster methodology, its advantages, and examples are given by Gaffney et al. (Gaffney and Smyth 1999, 2005; Gaffney 2004; Gaffney et al. 2007) and by Camargo et al. (2007a,b).
Statistical analysis methods are used in our work, such as Pearson correlation analysis, composite analysis, EOF analysis, and linear regression analysis. Statistical significance of the difference in composites and Pearson correlation is assessed using the Student’s t test. Statistical significance of the linear regression is assessed using the F test.
c. Choice of cluster number
To select the most appropriate number of clusters for the WNP TC tracks in 1950–2017, we calculate log-likelihood and within cluster error as the function of the number of clusters, which is from 1 to 15. Log-likelihood is an index that indicates how good the cluster analysis of TC tracks is in this probabilistic model and then can be used as one criterion to decide the most suitable number of clusters. Figure 1a shows that log-likelihood increases with the number of clusters, indicating that the larger the number of clusters is, the better the TC tracks cluster. But this relationship is not linear; when the number of clusters exceeds a certain value (K ≥ 8), the goodness of cluster effect increases slowly. In addition, the within cluster error curve in Fig. 1b indicates that the total dispersion of each type of TC tracks varies with cluster numbers. The smaller the value of within cluster error is, the better the TC tracks cluster. Again, when the number of clusters exceeds a certain value (K ≥ 8), it shows no significant change in cluster error. Therefore, we initially select K = 5–8 as the candidates for the number of clusters.
(a) Log-likelihood values and (b) within-cluster error for different number of TC track clusters. The log-likelihood values are the maximum of 16 runs and the cluster error values are the minimum of 16 runs, both obtained by a random permutation of the TCs given to the cluster model. (c) Maximum correlation among the clusters for each total cluster value, between monthly NTC or ACE and Niño-3.4 in the periods of 1950–2017. Significant correlations are shown with black asterisks.
Citation: Journal of Climate 35, 20; 10.1175/JCLI-D-21-0381.1
Further, in order to decide the most optimized number of clusters from 5 to 8, we examine the mean regression trajectories of TCs in 1950–2017 in the WNP for 5, 6, 7, and 8 clusters, respectively (not shown). When K is equal to 5 (8), the mean regression trajectories of each cluster are simple (complex), and it is not suitable to characterize the TC tracks. Because our work focuses on the relationship between the TC tracks and ENSO, we hope that the types of TC tracks associated with ENSO can be effectively stripped out by cluster analysis. The maximum correlation coefficient of monthly number of TCs (NTC) or accumulated cyclone energy (ACE) and the Niño-3.4 index calculated from the ERSST dataset for each total cluster number shown in Fig. 1c suggest that when K is equal to 7, the combination of two correlations is the highest. Considering all the analysis above, the most appropriate number of clusters is chosen to be 7 for this study, which is the same as in Camargo et al. (2007a).
3. TC track clusters in different ENSO/PDO phases
a. TC track clusters
Before we investigate the relationship between TC tracks and climate variability, we repeat the cluster analysis of Camargo et al. (2007a) using a longer time series of historical observations. The TC tracks of the WNP in 1950–2017 are clustered into clusters A–G (Fig. 2). The mean regression curves are represented by gray open circles that are drawn at the same time interval (one day). All 1747 TCs were assigned to one of the seven clusters. The seven mean regression curves show that there are three recurved trajectories and four straight-moving tracks. Table 1 summarizes various features of TC tracks and intensity of the seven clusters, in which percent values represent the proportion within all TCs of each cluster. Clusters A, B, and C account for approximately 60% of total number of TCs, which are dominant clusters, but these three clusters have less intense typhoons and less mean duration. Clusters E, F, and G are the opposite. The genesis latitudes of clusters A–C are around 15°–20°N, while clusters D–G are around 6°–10°N and account for about 40% of the total number of TCs. Clusters A, B, and D are generated west of 140°E, while the other four clusters are generated east of 140°E. The tracks of clusters B, D, and F are similar, and they are all aggregated on a narrow southeast–northwest strip, especially for clusters D and F. From the mean regression curves, we can find that the translation speed is higher at the initial stage than that during the recurving stage, and can increase after TCs have recurved such as cluster E. Landfall of TCs is related to the genesis position and track. Clusters A, B, D, and F have a large percentage of landing. For clusters A and B, the cause of landing is that their genesis positions are close to land. Clusters D and F have straight-moving tracks toward land. Besides, many TCs have experienced landing more than once during their lifetime, such as clusters B, D, and F, which passed through many islands. The correlation coefficients of annual NTC/ACE and Niño-3.4 index for each cluster and all TCs are in boldface if they meet the 95% significance level. Cluster G has the largest positive correlation with Niño-3.4 index, and clusters E and C follow by descending correlation order. Cluster A has the negative correlation with Niño-3.4 index.
TC tracks (black lines) of seven clusters during 1950–2017 in the WNP; each cluster mean regression curve is marked in gray open circles at the same time interval (one day).
Citation: Journal of Climate 35, 20; 10.1175/JCLI-D-21-0381.1
Comparison of various features for each cluster during 1950–2017 in the WNP. Percent values represent the proportions within each cluster. TY indicates a category 1–2 typhoon; ITY indicates a category 3–5 (intense) typhoon. The correlation coefficient at the 95% significance level is marked in boldface.
Large-scale environmental flow affects the direction and translation speed of TCs. We use the wind field data from NCEP reanalysis to calculate steering flow by the equation
Composites of steering flow (vectors; m s−1) and vertical wind shear (VWS) (shading; m s−1) for TCs in each cluster and all TCs during 1950–2017 in the WNP. The composites are calculated on whole active days of TCs. We mark the 95% significant areas in black arrows and dots. Each cluster mean regression curve is shown with a yellow line (trajectory) and asterisk (generation location).
Citation: Journal of Climate 35, 20; 10.1175/JCLI-D-21-0381.1
Vertical wind shear (VWS) is another important factor for genesis, trajectory, intensity, and structure of TCs. Weak local VWS is favorable for TC genesis and intensification, and the stronger VWS can also indirectly modulate the TC movement by the horizontal potential vorticity (PV) and beta drift in a baroclinic theory (Chan 2005). The shading in Fig. 3 shows the composite of VWS between 850 and 200 hPa for each cluster. The TC active region usually corresponds to weak VWS. Strong VWS often occurs in the midlatitudes (around 40°N) of the Northern Hemisphere. For clusters A and C, the low VWS region extends more northward compared with the other clusters and thus the TCs of these two clusters can reach higher latitudes (Camargo et al. 2007b). However, for clusters B, D, and F, their activities are significantly suppressed by strong VWS and the possibility of moving to the north is largely reduced. From the TC tracks of clusters A, C, E, and G, they contain a fair number of cyclones that recurve and show northward movement, so they have a high probability of being affected by VWS due to its large meridional change from low to high value.
SST represents the role of ocean in affecting TC tracks. The SST anomaly composite of all TCs (Fig. 4h) shows that SST is slightly anomalously warm (cold) in the east (west) tropical Pacific, indicating the footprints of ENSO and the PDO. For clusters A, B, and D (E and G), the SST composites (Fig. 4) exhibit a cold (warm) anomaly in the equatorial EP, possibly corresponding to La Niña (El Niño), which is in accordance with Camargo et al. (2007b). From Fig. 4 we see that there are counterclockwise wind fields at 850 hPa around mean regression tracks for all the clusters except cluster B, the mean track of which is too short to observe the corresponding large-scale atmospheric pattern. As suggested by Chan (2000) and Camargo et al. (2007b), the counterclockwise wind fields are generally accompanied with positive vorticity. Therefore, above the mean regression tracks for these clusters, positive vorticity exists and contributes to TC activity.
As in Fig. 3, but for composites of SST anomaly (shading; °C) and wind anomaly at 850 hPa (vectors; m s−1) for TCs in each cluster and all TCs during 1950–2017 in the WNP. Each cluster mean regression curve is shown by a red line (trajectory) and asterisk (generation location). We mark the 95% significant areas in black arrows and dots.
Citation: Journal of Climate 35, 20; 10.1175/JCLI-D-21-0381.1
As shown in Fig. 4, TCs of clusters A, B, and D (E and G) more likely happen in La Niña (El Niño) years. From the perspective of ACE, we check the relationship between each cluster and ENSO. ACE represents the integrated characteristics of the number and intensities of TCs occurring in a basin over their lifetime (Camargo and Sobel 2005). The ACE of a TC is calculated by summing the squares of the estimated maximum sustained wind speed of every active tropical storm (wind speed 35 kt or higher; 1 kt ≈ 0.51 m s−1) at 6-h intervals. Figure 5 shows that the ACE of all TCs is larger in El Niño years and smaller in La Niña years, which suggests the role of ENSO. From Fig. 4h, the El Niño–like SST anomaly pattern of all TCs also reveals the role of ENSO. For clusters A and D, the ACE in La Niña years is larger than in El Niño years. It suggests that La Niña favors these two clusters. The influence of La Niña on cluster B is not significant, possibly because the lifetime of TCs in cluster B is short and most of these TCs are formed near the coastline and fail to develop into intense typhoons. For clusters C, E, and G, the ACE in El Niño years is larger than in La Niña years and the ACE of neutral years is in between. It suggests that the TCs of these three clusters tend to happen in El Niño years. The ACE of cluster F shows no significant difference between El Niño and La Niña years. The SST composite corresponding to cluster F (Fig. 4) shows a cold anomaly in the equatorial EP, and this pattern may be related to El Niño Modoki.
Mean ACE (m2 s−2) per year in La Niña, neutral, and El Niño years, for seven clusters and all TCs. The ACE values of all TCs were divided by 3 to match the scale of the value of ACE of each cluster.
Citation: Journal of Climate 35, 20; 10.1175/JCLI-D-21-0381.1
We combine the analysis based on the correlation coefficient of annual NTC/ACE with ENSO (Table 1), composites of SST anomaly (Fig. 4), and mean ACE per year in different ENSO phases (Fig. 5) for each cluster, and conclude that clusters C, E, and G (mainly occurring in El Niño years) and cluster A (happening more often in La Niña years) are significantly correlated with ENSO. Clusters B, D, and F are aggregated on a narrow southeast–northwest band, and their regression curves of the positive and negative phases of ENSO are very close (Fig. 6). The correlations between annual ACE of clusters B, D, and F and Niño-3.4 index are also not significant (Table 1). The mean ACE per year of clusters B and F has no clear difference in different phases of ENSO. The mean ACE per year of cluster D between ENSO phases shows that cluster D is more active in La Niña years but this ENSO effect is weak based on corresponding SST composite analysis (Fig. 4) and correlation between the ACE and Niño-3.4 index (Table 1). So, clusters B, D, and F are not significantly related to ENSO. Therefore, we focus on clusters C, E, G, and A to detail their relationships with ENSO and how they are modulated by the PDO. We noted that Camargo et al. (2007b) suggested that only two clusters correspond to El Niño events. The difference from this present study should be ascribed to using different lengths of historical data.
Composite differences of steering flow (vectors; m s−1) and VWS (shading; m s−1) of each cluster and all TCs between positive and negative phase of ENSO during 1950–2017 in the WNP. We calculate in all TCs active days, and mark the 95% significant areas in black arrows and dots. Each cluster mean regression curve is shown in red (blue) line (trajectory) and asterisk (generation location), for El Niño (La Niña) years, separately.
Citation: Journal of Climate 35, 20; 10.1175/JCLI-D-21-0381.1
b. TC track clusters in El Niño/La Niña years
It has been well known that typhoon tracks are related to ENSO (Wang and Chan 2002; Camargo et al. 2007b) as shown in section 3a. To further explore the difference of the TC track characters between positive and negative phase of ENSO, we select the TCs to categorize two groups for each cluster: the ENSO(+) group for the TCs in El Niño years and the ENSO(−) group for the TCs in La Niña years, and summarize the statistics in Table 2.
Comparison of various features of TCs in positive and negative phases of ENSO (i.e., El Niño or La Niña) of each cluster and all TCs during 1950–2017 in the WNP. Percent values represent the proportions within each cluster or all TCs.
For individual clusters, typhoons of clusters A, B, and D, which account for nearly 53% of the number of all TCs (Table 1), are mainly active in the western part of the WNP (Fig. 2), and they have larger mean NTC per year in La Niña years and comparably more weak TCs with the intensity of TS and TY (Table 2). Cluster A has the largest mean NTC per year and mean track difference between the two phases of ENSO among three clusters (Table 2 and Fig. 6). Typhoons of clusters C, E, F, and G, which account for nearly 47% of the number of all TCs (Table 1), are mainly active in the eastern part of the WNP (Fig. 2). They have larger mean NTC per year in El Niño years, and tend to be more intense, reaching category 3–5 typhoons (i.e., ITY). Clusters C and E have larger mean NTC per year difference (Table 2) and clusters C, E, and G have larger mean track difference between the two phases of ENSO among the four clusters (Fig. 6). The genesis positions of these four clusters tend to shift to the east and they show more northward movement in El Niño years (Fig. 6), especially for cluster E.
For all the TCs, there are larger mean NTC per year in El Niño years, mainly resulting from the activities of clusters C, E, and G (Table 2); cluster A plays the major opposite role. This indicates that the typhoon activity is pushed eastward in ENSO(+) and fewer TCs make landfall. TCs tend to generate in the east and south of the WNP in El Niño years (all TCs in Table 2), which is clearly manifested in clusters C and E (Fig. 6). Moreover, the average TC lifetime is obviously longer in ENSO(+) for clusters C, E, and G, indicating that those TCs are likely to gain more heat from ocean to grow into strong TCs. Differences of track density (see Fig. 8a) also show that the region where TCs are more active in El Niño years than La Niña years overlaps with the area where clusters C, E, G occur. But the region where cluster A occurs in Fig. 8a has slightly more TCs in La Niña years. There are no significant differences of track density in the region where clusters B, D, and F occur (Fig. 8a). This negative anomaly on the west and positive anomaly on the east pattern can also be found in the spatial pattern of EOF1 in section 4a (Fig. 13a), further proving the effect of ENSO on TC tracks.
Figure 6 shows composite differences of VWS and steering flow between positive and negative phase of ENSO for each cluster. The spatial patterns of composite differences for all the clusters share similarities because they are basically a reflection of environmental variables changing between El Niño and La Niña. When the VWS difference is negative, most of the region corresponds to a westward steering flow, and vice versa. For most of the seven clusters it exhibits positive values of VWS difference in the midlatitude region and the southeastern part of the WNP. The absolute value of VWS in the midlatitude is always high and effectively suppress TC activities in both phases of ENSO. However, in the southeastern part of the WNP, the absolute value of VWS is low and thus TC activities are sensitive to changes in different phases of ENSO.
Similarly, composite differences of SST anomaly (shading; °C) and wind anomaly at 850 hPa (vector; m s−1) between positive and negative phases of ENSO are shown in Fig. 7. Because the composite differences of SST anomaly are calculated from the SST anomaly in El Niño years minus that in La Niña years, all composite difference patterns show an El Niño character. In ENSO(+), the positive SST anomaly in the east of tropical Pacific and the eastward expanding warm pool push TC activities southeastward. Wind at 850 hPa is not only an important variable for observing the characteristics of cyclone vorticity and intensity, but also a part of calculating the integral of steering flow. The difference of wind anomaly at 850 hPa between the two phases of ENSO shows that for most clusters and all TCs, there are weak wind anomalies in the opposite direction to the TC movement direction. It can prevent the steering flow from being strengthened, and contribute to elongate the lifetime of TCs by slowing down the TC translation speed in El Niño years. [The TCs in Table 2 show an average lifetime of 8.67 (6.62) days and translation speed of 20.09 (20.26) km h−1 in the El Niño (La Niña) years.]
As in Fig. 6, but for composite differences of SST anomaly (shading; °C) and wind anomaly at 850 hPa (vectors; m s−1) between positive and negative phase of ENSO during 1950–2017 in the WNP. We mark the 95% significant areas in black arrows and dots.
Citation: Journal of Climate 35, 20; 10.1175/JCLI-D-21-0381.1
Focusing on the seven clusters, we can find that the local environmental variables shown in Figs. 6 and 7 demonstrate similar difference patterns between the two phases of ENSO. Only four clusters, A, C, E, and G, are significantly affected by ENSO due to the changes of these local environmental variables. However, for clusters B, D, and F, along their tracks there are in general no significant changes of these environmental variables between ENSO(+) and ENSO(−) and so ENSO exerts little influence. The above ENSO effects can also be shown in cluster analysis on TCs in ENSO(+) and ENSO(−), respectively (figures not shown).
c. TC track clusters in the El Niño/La Niña years stratified by the PDO
TCs can be influenced by the PDO. The number of TCs is obviously larger in the PDO warm phase (27.38 TCs yr−1) than in the PDO cold phase (24.78 TCs yr−1). After using a 7-yr Gaussian filter to smooth the PDO index of 1950–2017, we obtain two cold phase periods (1950–76 and 1998–2013) and two warm phase periods (1977–97 and 2014–17). Considering the ENSO years in each phase of the PDO, we can get four scenarios: PDO(+)ENSO(+), PDO(+)ENSO(−), PDO(−)ENSO(+), and PDO(−)ENSO(−). The TC track cluster statistics of these four ENSO-related categories are summarized in Table 3 and shown in Figs. 8 and 9. Here we use “PDO(+)” to represent the PDO warm phase, and “PDO(−)” for the PDO cold phase. We examine the differences of track density between PDO(+)ENSO(+) and PDO(+)ENSO(−) and between PDO(−)ENSO(+) and PDO(−)ENSO(−) in Fig. 9.
Composite differences of TC track density between (a) ENSO(+) and ENSO(−) years, (b) PDO(+)ENSO(+) and PDO(+)ENSO(−) years, and (c) PDO(−)ENSO(+) and PDO(−)ENSO(−) years.
Citation: Journal of Climate 35, 20; 10.1175/JCLI-D-21-0381.1
Composite differences of annual track density of clusters A, C, E, and G between El Niño and La Niña years in the (left) warm and (right) cold PDO phase during 1950–2017 in the WNP. Each cluster mean regression curve is shown, red lines represent mean regression curves in El Niño years, blue lines represent mean regression curves in La Niña years, separately.
Citation: Journal of Climate 35, 20; 10.1175/JCLI-D-21-0381.1
Comparison of various features of TCs in positive and negative phases of ENSO (i.e., El Niño/La Niña) in the warm (cold) PDO phases during 1950–2017 in the WNP. Percent values represent the proportions within clusters A, C, E, and G.
For individual clusters, clusters A, C, E, and G account for nearly 23%, 18%, 13%, and 8% of the total number of all TCs (Table 1). Clusters A and C exhibit some modulation of the PDO on the relationship of TCs and ENSO. Cluster A has the same mean NTC per year in El Niño and La Niña years, respectively, in the PDO warm phase. It has larger mean NTC per year in La Niña years in the PDO cold phase (Table 3). Cluster C has larger mean NTC per year in La Niña years in the PDO warm phase, but larger mean NTC per year in El Niño years in the PDO cold phase (Table 3). However, the enhancement of track density in the PDO cold phases for clusters A and C is not distinct (Figs. 9b and 9d) in the regions along with their mean tracks. Clusters E and G show significant PDO modulation on the relationship between the TCs and ENSO. Clusters E and G have larger mean NTC per year in El Niño years in both the PDO warm and cold phases (Table 3), and the differences of track density in Figs. 9e–h are enhanced in the PDO warm phases. It is interesting that there are larger differences for genesis positions in the PDO cold phase (Table 3 and Fig. 9).
For all the TCs, in terms of mean NTC per year, more TCs occur in ENSO(+) than in ENSO(−) (26.53 vs 23.06 for all TCs in Table 2). This difference is slightly larger in the PDO warm phase (28.71 vs 25.67) than in the PDO cold phase (25.00 vs 22.50), especially for cluster G (Table 3). As shown in Fig. 8b in the PDO warm phase, the track density difference between two ENSO phases is enhanced compared with Fig. 8a, especially in the regions of clusters E and G (Fig. 9). The regions where other clusters exist show smaller TC density difference in Fig. 8b. Such contrast between different ENSO phases in the PDO cold phase is much weaker (Fig. 8c). Therefore, we conclude that there are more significant differences for the TC tracks between positive and negative phases of ENSO in the PDO warm phase for clusters E and G, which have significant relationships with ENSO. However, we note that for clusters A and C no such conclusion can be drawn, even though they are also significantly correlated with ENSO. The causes are investigated again by exploring the environmental factors.
For each cluster (A, C, E, and G), we show the composite differences of steering flow and VWS during TCs’ active days in El Niño years minus that in La Niña years in the PDO warm and cold phases, respectively (Fig. 10). The patterns of composite differences in the PDO cold phases are similar to the patterns of Fig. 6, but in the PDO warm phases the composite differences in the regions over clusters E and G trajectories are more significant, suggesting that local steering flow and VWS are the important factors controlling the PDO modulation on the relationship between TC tracks of clusters E and G and ENSO.
Composite differences of steering flow (vectors; m s−1) and VWS (shading; m s−1) of clusters A, C, E, and G between El Niño and La Niña in the (left) warm and (right) cold PDO phase during 1950–2017 in the WNP. We calculate composites in all TCs active days, and mark the 95% significant areas in black arrows and dots. Each cluster mean regression curve is shown in a white or green line (trajectory) and asterisk (generation location) for El Niño or La Niña, respectively.
Citation: Journal of Climate 35, 20; 10.1175/JCLI-D-21-0381.1
The composite differences for SST anomaly and 850-hPa wind anomaly in the PDO warm and cold phases are shown in Fig. 11. The stronger El Niño–like SST pattern and eastward tropical wind field can lead to more eastward or southward shift of the TC genesis positions in the PDO warm phase. The differences of SST anomaly in PDO(+) are more pronounced, and we can find more regions with significant difference of SST in the PDO warm phase over the TC activity region, especially for clusters E and G. Figures 11g and 11h show the difference of wind anomaly at 850 hPa between the two ENSO phases in the positive and negative phases of the PDO for cluster G, respectively. We can see that there is a stronger reversed wind anomaly in the direction of TC movement in the PDO warm phase than in the PDO cold phase. This reversed wind anomaly can weaken the steering flow, so it leads to longer lifetime and lower translation speed for cluster G TCs in El Niño years than in La Niña years in the positive phase of the PDO (Table 3). However, in the negative phase of the PDO, the reversed wind anomaly is not obvious, so there is no significant difference of TC lifetime and translation speed between the two ENSO phases.
As in Fig. 10, but for composite differences of SST anomaly (shading; °C) and wind anomaly at 850 hPa (vectors; m s−1) between El Niño and La Niña in the (left) warm and (right) cold PDO phase during 1950–2017 in the WNP. We mark the 95% significant areas in black arrows and dots. Each cluster mean regression curve is shown in red or magenta line (trajectory) and asterisk (generation location) for El Niño or La Niña, respectively.
Citation: Journal of Climate 35, 20; 10.1175/JCLI-D-21-0381.1
The environmental factors analysis reveals that VWS, steering flow, SST, and wind vorticity contribute to the PDO modulation on the relationships between TC tracks and ENSO. However, these influences are only on clusters E and G. For clusters A and C, which are also correlated with ENSO, their corresponding composite (VWS, steering flow, SST, and 850-hPa wind) differences between PDO(+)ENSO(+) − PDO(+)ENSO(−) and PDO(−)ENSO(+) − PDO(−)ENSO(−) are indistinct, especially in the regions of cluster A and C trajectories. We further choose TCs in PDO(+)ENSO(+), PDO(+)ENSO(−), PDO(−)ENSO(+), and PDO(−)ENSO(−) and perform cluster analysis respectively (figure not shown), supporting the above conclusions.
4. EOF analysis on TC track density
In this section, we use EOF analysis to examine the relationships between TC tracks and ENSO (PDO). Yearly track density from 1950 to 2017 are constructed based on JTWC dataset. Tropical cyclones with intensity as TS or higher are selected and in total 1747 TCs are included (the same as in section 3). The numbers of TCs that pass over a grid box of 5° latitude × 5° longitude are counted. If a TC stays on the same grid box for more than one time step successively, it is only counted once so that the scenario of many storms passing through a grid box can be distinguished from that of a storm staying in the same grid box for a long time (Liu and Chan 2008).
To perform EOF analysis on track density anomaly, we need to remove the trend. The spatial pattern of TC track density trend is shown in Fig. 12a. During the period of 1950–2017, the activity of TCs increases in the southeastern part of the WNP, SCS, and East China Sea (ECS), and decreases above 40°N. The annual number of TCs in the WNP has no significant trend (Fig. 12b), but there is a declining trend for TCs of recursive track clusters and an increasing trend for TCs of straight track clusters that move from the east toward the continent. Therefore Fig. 12a agrees well with Fig. 12b, both suggesting that under the background of global warming there are more TCs moving toward the coasts of China, leading to higher chances of landfall.
(a) Pattern of trend of TC track density in spatial in 1950–2017. (b) Annual NTC of all TCs in western North Pacific (black line), the sum NTC of straight-moving clusters B, D, F, and G (red line), and the sum NTC of recurving clusters A, C, and G (blue line). Dashed lines show their linear trends during 1950–2017.
Citation: Journal of Climate 35, 20; 10.1175/JCLI-D-21-0381.1
Following Liu and Chan (2008), we first averaged the density data in a 3 × 3 grid sliding domain to reduce the noise level, and then the number was divided by the total annual number of TCs over the whole basin to obtain the percentage value to reduce the effect of the interannual variation of total annual TC number, and finally we detrended the annual track density. Corresponding to annual track density, the Niño-3.4 index, the PDO index, and environment variables were averaged over July–October to represent the status of these variables during the TC peak season.
a. Interannual variability
A 2–7-yr bandpass filter is applied to the annual track density for EOF analysis to capture the interannual variability. The leading two modes account for about 50% of total variance. The spatial pattern of EOF1 exhibits a dipole structure with negative loading on the west and positive loading in the WNP on the east (Fig. 13a), and its associated principal component PC1 and Niño-3.4 index are significantly correlated with correlation coefficient at 0.64 (Fig. 13c). It is obvious that the EOF1 is linked to ENSO. In ENSO(+), TC activities increase in the regions of clusters C, E, and G (see section 3 and Fig. 8a). Figure 13e shows that the relationship between PC1 and ENSO with significant correlation coefficient does not change much between PDO(+) and PDO(−). It reflects that the PDO modulation is weak on the relationship between EOF1 and ENSO.
First two modes of EOF analysis for TC track density that are 2–7-yr pass filtered. (a),(b) Spatial modes; mean regression curves of clusters A, C, E, and G in section 3a are marked in black lines. (c),(d) Normalized PC time series (black lines) and Niño-3.4 index (red lines). (e),(f) 10-yr sliding correlation between normalized PC time series and Niño-3.4 index during 1950–2017 (black lines; red asterisks indicate 95% confidence level) and the PDO index (blue lines), calculated by an 11-yr running mean.
Citation: Journal of Climate 35, 20; 10.1175/JCLI-D-21-0381.1
EOF2 is also correlated with ENSO (Fig. 13d) with a northeast–southwest dipole pattern (Fig. 13b). The running correlation between PC2 and Niño-3.4 index becomes statistically significant in the 1980s–1990s when the PDO is in its warm phase (Fig. 13f). Moreover, this correlation is also strong after 2012 but not significant, perhaps due to the short and incomplete warm phase of the PDO (Fig. 13d). The findings above reveal that both first two EOF modes are related with ENSO but only for the second mode; its correlation with ENSO is modulated by the PDO and the correlation is particularly enhanced during the PDO(+). As we know from the previous section, the relationship between clusters E and G and ENSO is enhanced in the PDO warm phase. The regions where cluster E is active are well overlapped with the maximum loading of EOF2 spatial function. But for cluster G, it is located near the boundary of two dipoles, not completely overlapped with EOF2 maximum loading. This is possibly because there are not sufficient TC numbers for cluster G, and therefore the shape of cluster G’s mean track is not robust. All in all, the modulation effect of the PDO on TC activity is well exhibited from these two different perspectives (cluster analysis for TC trajectories and EOF analysis for TC track density). Both conclusions are well in accordance, indicating the linkage between clusters E and G and the spatial pattern of the EOF2. EOF3 and higher-order modes of EOF explain little variance and are not discussed.
The relations between TC activity and ENSO/PDO need to be explained by environmental factors that affect TC tracks. Regressions of SST anomaly, 850-hPa wind, steering flow, and VWS anomalies onto PC1 and PC2 in the whole period are shown in Figs. 14a–d, as are regressions in the PDO warm phases in Figs. 14e–h and in the PDO cold phases in Figs. 14i–l. It is likely a strong (developing or weak) El Niño–type SST anomaly pattern corresponding to EOF1 (EOF2) in Figs. 14a and 14b. For the regressions of VWS (Figs. 14c,d), there is a zonal band located in the midlatitudes and a dipole pattern in the tropics of the Pacific. This pattern for EOF1 is more distinct than EOF2. The regression of steering flow onto PC2 demonstrates an anticyclone pattern within the TCs active region 10°–40°N, 120°–160°E, which does not exist in the regression pattern onto PC1 (Figs. 14c,d). These regressions explain well how ENSO affects TC tracks by two modes. The environmental factors (local SST, VWS, and steering flow) associated with PC2 (Figs. 14f,h,j,l) are affected more obviously during the PDO(+) compared with PC1 (Figs. 14e,g,i,k).
(left) Regressions of (a) SST anomaly (shading; °C) and 850-hPa wind anomaly (vector; m s−1) and (c) VWS (shading; m s−1) and steering flow (vector; m s−1) onto normalized PC1 in the whole period. (e),(g) and (i),(k) As in (a) and (c), but for regressed patterns onto PC1 in the PDO warm phase and cold phase, respectively. (right) Regression patterns onto PC2. Stippling indicates linear correlation coefficient at a 0.05 significance level for SST and VWS anomaly. Black arrows for steering flow and 850-hPa wind mean the linear correlation coefficient at a 0.05 significance level; mean regression curves of clusters A, C, E, and G in section 3a are marked in red lines.
Citation: Journal of Climate 35, 20; 10.1175/JCLI-D-21-0381.1
b. Decadal to multidecadal variability
The first two modes of EOF analysis on TC track density after a 7-yr low-pass filter is applied account for almost 66.7% of total variance (figures not shown). Both PC1and PC2 are insignificantly correlated with the PDO index because the effective number of degrees of freedom has dramatically decreased after 11 years running mean of the PDO index, but the PDO plays an important role in affecting the decadal to multidecadal variability of TC activity (Liu and Chan 2008). However, a comprehensive analysis of the direct relationship between TC activity and the PDO is not our focus and need to be addressed in future study.
5. Conclusions and discussion
Our analysis has revealed that the PDO modulates the relationship between ENSO and tracks in the WNP. TC trajectories in the WNP during 1950–2017 are clustered into seven clusters (clusters A–G), including three recurved trajectories and four straight-moving tracks. These clusters are distinguished well by number of TCs, intensity, lifetime, genesis position/month, landing, and track. From the spatial pattern of SST anomaly of each cluster and accumulated cyclone energy (ACE) in different phases of ENSO for each cluster, we find that cluster A is dominated by La Niña, and clusters C, E, and G are dominated by El Niño. ENSO exerts its influence on these clusters mainly through steering flow and VWS. However, such ENSO effects on quantity, genesis location, and tracks of corresponding clusters are revealed to be enhanced in the positive phase of the PDO only for clusters E and G (which generate in southeastern part of the WNP and undergo a long lifetime and track) because the PDO explains little variance of environmental factors that control TC activity in the regions where clusters A and C (generated and occurring in the northerly part of the WNP) are located. This finding is also supported by TC track density analysis. The two leading modes of EOF analysis of TC track density are both significantly correlated with ENSO. The enhancement of ENSO effects during the PDO positive phases is displayed significantly by the second mode through impacting local SST, VWS, and steering flow. Therefore, from both Lagrangian (cluster analysis) and Eulerian (EOF analysis) viewpoints, we draw consistent conclusions about the PDO modulation on the relationship between ENSO and typhoon tracks.
As we extended the time series of TC data for study from 1950 to 2017 compared to a range of 1950–2002 in Camargo et al. (2007a,b), our cluster analysis demonstrates seven clusters quite similar to theirs, suggesting the robustness of cluster analysis results. However, differences still exist. We have three clusters corresponding to El Niño rather than two in their study. Although the consistency of TC track cluster analysis need further verification when longer reliable historical record is available, the time period of this study permits investigation on decadal time scales. There are direct and indirect two ways for the PDO to influence on TC activity. Motivated by previous studies about the PDO modulation on the relationship between ENSO and rapid intensification, we particularly focus on such modulation on ENSO and TC tracks in this study. A comprehensive analysis about decadal variability of TC activity and the PDO influences need to be completed next.
We performed cluster analysis and then assessed the impacts of ENSO/PDO on each cluster. The effects of ENSO/PDO can also be demonstrated by cluster analysis in different phases of climate variability. The contrasts of the shapes of mean clustering curves between different climate variability phases visualize the effects of ENSO/PDO, but they lack the basis for quantitative comparison between each pair of clusters from different climate variability phases, respectively, due to the limitation of cluster analysis.
From the TC track density trend, we also find that TC activity increases in the southeastern part of the WNP, SCS, and ECS, and decreases above 40°N. The annual number of TCs in the WNP has no significant trend, but there is a declining trend for TCs of recursive track clusters and an increasing trend for TCs of straight track clusters (Fig. 12), indicating that under the background of global warming there are more TCs moving toward the coasts of China, leading to higher chances of landfall. The findings on long-term variability and trends of TC activity presented in this study are expected to contribute to further improvement of TC track prediction.
Acknowledgments.
This work is supported by the National Key Research and Development Program of China 2016YFA0601804, Open Financial Grant from Qingdao National Laboratory for Marine Science and Technology QNLM2016ORP0101, Shanghai Typhoon Research Foundation, National Science Foundation of China 41776019, HKUST-SJTU Joint Research Collaboration Fund, and a grant of Shanghai Frontiers Science Center of Polar Science (SCOPS). Chaoming Huang is supported by the National Key Research and Development Program of China 2016YFC1401905. We thank Scott J. Gaffney for providing the Matlab toolbox with the clustering algorithms described in his Ph.D. paper at http://www.datalab.uci.edu/resources/CCT.
Data availability statement.
NOAA ERSST V5 data are provided by the NOAA/OAR/ESRL PSD, Boulder, Colorado, USA at https://www.esrl.noaa.gov/psd/.
REFERENCES
Barnston, A. G., M. Chelliah, and S. B. Goldenberg, 1997: Documentation of a highly ENSO-related SST region in the equatorial Pacific. Atmos.–Ocean, 35, 367–383, https://doi.org/10.1080/07055900.1997.9649597.
Blender, R., K. Fraedrich, and F. Lunkeit, 1997: Identification of cyclone-track regimes in the North Atlantic. Quart. J. Roy. Meteor. Soc., 123, 727–741, https://doi.org/10.1002/qj.49712353910.
Camargo, S. J., and A. H. Sobel, 2005: Western North Pacific tropical cyclone intensity and ENSO. J. Climate, 18, 2996–3006, https://doi.org/10.1175/JCLI3457.1.
Camargo, S. J., A. W. Robertson, S. J. Gaffney, P. Smyth, and M. Ghil, 2007a: Cluster analysis of typhoon tracks. Part I: General properties. J. Climate, 20, 3635–3653, https://doi.org/10.1175/JCLI4188.1.
Camargo, S. J., A. W. Robertson, S. J. Gaffney, P. Smyth, and M. Ghil, 2007b: Cluster analysis of typhoon tracks. Part II: Large-scale circulation and ENSO. J. Climate, 20, 3654–3676, https://doi.org/10.1175/JCLI4203.1.
Camargo, S. J., A. W. Robertson, A. G. Barnston, and M. Ghil, 2008: Clustering of eastern North Pacific tropical cyclone tracks: ENSO and MJO effects. Geochem. Geophys. Geosyst., 9, Q06V05, https://doi.org/10.1029/2007GC001861.
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, 599–606, https://doi.org/10.1175/1520-0493(1985)113<0599:TCAITN>2.0.CO;2.
Chan, J. C. L., 2000: Tropical cyclone activity over the western North Pacific associated with El Niño and La Nina events. J. Climate, 13, 2960–2972, https://doi.org/10.1175/1520-0442(2000)013<2960:TCAOTW>2.0.CO;2.
Chan, J. C. L., 2005: The physics of tropical cyclone motion. Annu. Rev. Fluid Mech., 37, 99–128, https://doi.org/10.1146/annurev.fluid.37.061903.175702.
Chan, J. C. L., 2008: Decadal variations of intense typhoon occurrence in the western North Pacific. Proc. Roy. Soc. London, 464A, 249–272, https://doi.org/10.1098/rspa.2007.0183.
Chan, J. C. L., and W. M. Gray, 1982: Tropical cyclone movement and surrounding flow relationships. Mon. Wea. Rev., 110, 1354–1374, https://doi.org/10.1175/1520-0493(1982)110<1354:TCMASF>2.0.CO;2.
Chia, H. H., and C. F. Ropelewski, 2002: The interannual variability in the genesis location of tropical cyclones in the northwest Pacific. J. Climate, 15, 2934–2944, https://doi.org/10.1175/1520-0442(2002)015<2934:TIVITG>2.0.CO;2.
Chu, J.-H., C. R. Sampson, A. S. Levine, and E. Fukada, 2002: The Joint Typhoon Warning Center tropical cyclone best tracks, 1945–2000. Tech. Rep. NRL/MR/7540-02-16, Naval Research Laboratory, https://www.metoc.navy.mil/jtwc/products/best-tracks/tc-bt-report.html.
Elsner, J. B., 2003: Tracking hurricanes. Bull. Amer. Meteor. Soc., 84, 353–356, https://doi.org/10.1175/BAMS-84-3-353.
Elsner, J. B., and K. B. Liu, 2003: Examining the ENSO–typhoon hypothesis. Climate Res., 25, 43–54, https://doi.org/10.3354/cr025043.
Everitt, B. S., and D. J. Hand, 1981: Finite mixture distributions. Monographs on Statistics and Applied Probability, Springer, 143 pp.
Fisher, E. L., 1958: Hurricane and the sea surface temperature field. J. Atmos. Sci., 15, 328–333, https://doi.org/10.1175/1520-0469(1958)015<0328:HATSST>2.0.CO;2.
Franklin, J. L., S. E. Feuer, J. Kaplan, and S. D. Aberson, 1996: Tropical cyclone motion and surrounding flow relationships: Searching for beta gyres in omega dropwindsonde datasets. Mon. Wea. Rev., 124, 64–84, https://doi.org/10.1175/1520-0493(1996)124<0064:TCMASF>2.0.CO;2.
Gaffney, S. J., 2004: Probabilistic curve-aligned clustering and prediction with regression mixture models. Ph.D. thesis, University of California, Irvine, 281 pp., http://www.ics.uci.edu/pub/sgaffney/outgoing/sgaffney_thesis.pdf.
Gaffney, S. J., and P. Smyth, 1999: Trajectory clustering with mixture of regression models. Proc. Fifth ACM SIGKDD Int. Conf. on Knowledge Discovery and Data Mining, San Diego, CA, Association for Computing Machinery, 63–72.
Gaffney, S. J., and P. Smyth, 2005: Joint probabilistic curve-clustering and alignment. Advances in Neural Information Processing Systems 17: Proceedings of the 2004 Conference, L. K. Saul, Y. Weiss, and L. Bottou, Eds., MIT Press, 473–580.
Gaffney, S. J., A. W. Robertson, P. Smyth, S. J. Camargo, and M. Ghil, 2007: Probabilistic clustering of extratropical cyclones using regression mixture models. Climate Dyn., 29, 423–440, https://doi.org/10.1007/s00382-007-0235-z.
Girishkumar, M. S., V. P. Thanga Prakash, and M. Ravichandran, 2015: Influence of Pacific decadal oscillation on the relationship between ENSO and tropical cyclone activity in the Bay of Bengal during October–December. Climate Dyn., 44, 3469–3479, https://doi.org/10.1007/s00382-014-2282-6.
Goddard, L., and M. Dilley, 2005: El Niño: Catastrophe or opportunity? J. Climate, 18, 651–665, https://doi.org/10.1175/JCLI-3277.1.
Gray, W. M., 1979: Hurricanes: Their formation, structure and likely role in the tropical circulation. Meteorology over the Tropical Oceans, D. B. Shaw, Ed., Royal Meteorological Society, 155–218.
Harr, P. A., and R. L. Elsberry, 1991: Tropical cyclone track characteristics as a function of large-scale circulation anomalies. Mon. Wea. Rev., 119, 1448–1468, https://doi.org/10.1175/1520-0493(1991)119<1448:TCTCAA>2.0.CO;2.
Harr, P. A., and R. L. Elsberry, 1995: Large-scale circulation variability over the tropical western North Pacific. Part I: Spatial patterns and tropical cyclone characteristics. Mon. Wea. Rev., 123, 1225–1246, https://doi.org/10.1175/1520-0493(1995)123<1225:LSCVOT>2.0.CO;2.
Ho, C.-H., J.-J. Baik, J.-H. Kim, and D.-Y. Gong, 2004: Interdecadal changes in summertime typhoon tracks. J. Climate, 17, 1767–1776, https://doi.org/10.1175/1520-0442(2004)017<1767:ICISTT>2.0.CO;2.
Huang, B., and Coauthors, 2017: Extended reconstructed sea surface temperature version 5 (ERSSTv5): Upgrades, validations, and intercomparisons. J. Climate, 30, 8179–8205, https://doi.org/10.1175/JCLI-D-16-0836.1.
Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77, 437–472, https://doi.org/10.1175/1520-0477(1996)077<0437:TNYRP>2.0.CO;2.
Kossin, J. P., S. J. Camargo, and M. Sitkowski, 2010: Climate modulation of North Atlantic hurricane tracks. J. Climate, 23, 3057–3076, https://doi.org/10.1175/2010JCLI3497.1.
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, 636–651, https://doi.org/10.1175/1520-0493(1994)122<0636:AEAOTR>2.0.CO;2.
Lander, M. A., 1996: Specific tropical cyclone tracks and unusual tropical cyclone motions associated with a reverse-oriented monsoon trough in the western North Pacific. Wea. Forecasting, 11, 170–186, https://doi.org/10.1175/1520-0434(1996)011<0170:STCTTA>2.0.CO;2.
Lee, H. S., T. Yamashita, and T. Mishima, 2012: Multi-decadal variations of ENSO, the Pacific decadal oscillation and tropical cyclones in the western North Pacific. Prog. Oceanogr., 105, 67–80, https://doi.org/10.1016/j.pocean.2012.04.009.
Liu, K. S., and J. C. L. Chan, 2008: Interdecadal variability of western North Pacific tropical cyclone tracks. J. Climate, 21, 4464–4476, https://doi.org/10.1175/2008JCLI2207.1.
Lowry, M. R., 2008: Developing a unified superset in quantifying ambiguities among tropical cyclone best track data for the western North Pacific. Diploma thesis, Paper 1026, Florida State University, 149 pp.
MacQueen, J., 1967: Some methods for classification and analysis of multivariate observations. Proc. Fifth Berkeley Symp. on Mathematical Statistics and Probability, Berkeley, CA, University of California, 281–297.
Mantua, N. J., S. R. Hare, Y. Zhang, J. M. Wallace, and R. C. Francis, 1997: A Pacific interdecadal climate oscillation with impacts on salmon production. Bull. Amer. Meteor. Soc., 78, 1069–1079, https://doi.org/10.1175/1520-0477(1997)078<1069:APICOW>2.0.CO;2.
Mei, W., and S.-P. Xie, 2016: Intensification of landfalling typhoons over the northwest Pacific since the late 1970s. Nat. Geosci., 9, 753–757, https://doi.org/10.1038/ngeo2792.
Mei, W., S.-P. Xie, M. Zhao, and Y. Wang, 2015: Forced and internal variability of tropical cyclone track density in the western North Pacific. J. Climate, 28, 143–167, https://doi.org/10.1175/JCLI-D-14-00164.1.
Reynes, A., 2003: Environmental steering flow analysis for central North Pacific tropical cyclones based on NCEP/NCAR reanalysis data. Ph.D. thesis, University of Hawaii, 104 pp.
Saffir, H. S., 1977: Design and construction requirements for hurricane resistant construction. Preprints, ASCE Tech. Rep. 2830, American Society of Civil Engineers, 20 pp.
Sandgathe, S. A., 1987: Opportunities for tropical cyclone motion research in the Northwest Pacific region. Tech. Rep. NPS-63-87-006, Naval Postgraduate School, 36 pp.
Simpson, R. H., and H. Riehl, 1981: The Hurricane and Its Impact. Louisiana State University Press, 398 pp.
Wang, B., and J. C. L. Chan, 2002: How strong ENSO events affect tropical storm activity over the western North Pacific. J. Climate, 15, 1643–1658, https://doi.org/10.1175/1520-0442(2002)015<1643:HSEEAT>2.0.CO;2.
Wang, X., and H. Liu, 2016: PDO modulation of ENSO effect on tropical cyclone rapid intensification in the western North Pacific. Climate Dyn., 46, 15–28, https://doi.org/10.1007/s00382-015-2563-8.
Wu, C.-C., and K. A. Emanuel, 1995: Potential vorticity diagnostics of hurricane movement. Part I: A case study of Hurricane Bob (1991). Mon. Wea. Rev., 123, 69–92, https://doi.org/10.1175/1520-0493(1995)123<0069:PVDOHM>2.0.CO;2.
Yang, L., S. Chen, C. Wang, D. Wang, and X. Wang, 2018: Potential impact of the Pacific decadal oscillation and sea surface temperature in the tropical Indian Ocean–western Pacific on the variability of typhoon landfall on the China coast. Climate Dyn., 51, 2695–2705, https://doi.org/10.1007/s00382-017-4037-7.
Zhao, H., and C. Wang, 2016: Interdecadal modulation on the relationship between ENSO and typhoon activity during the late season in the western North Pacific. Climate Dyn., 47, 315–328, https://doi.org/10.1007/s00382-015-2837-1.
Zhao, H., and C. Wang, 2019: On the relationship between ENSO and tropical cyclones in the western North Pacific during the boreal summer. Climate Dyn., 52, 275–288, https://doi.org/10.1007/s00382-018-4136-0.
