Decadal Variations of Intense Tropical Cyclones over the Western North Pacific during 19482010

The influence of global warming on tropical cyclone (TC) intensity has been extensively discussed over the past several decades (Emanuel, 1987, 2005; Knutson et al., 1998, 2010;Knutson and Tuleya, 2004; Bender et al., 2010). In the western North Pacific (WNP), studies have suggested an increase in intense TCs (i.e., categories 4 and 5 on the Saffir-Simpson scale; hereafter, Cat45) since the 1970s (Webster et al., 2005; Elsner et al., 2008), while longer TC records suggest that such an increasing trend in TC intensity is a part of interdecadal variations over the WNP basin (Chan. 2006, 2008). Other studies have indicated that the increasing number of Cat45 TCs since the 1970s is detectable only in the Joint Warning Typhoon Center (JTWC) best track dataset (Kamahori et al., 2006; Wu et al., 2006; Song et al., 2010; Ren et al., 2011), suggesting substantial uncertainty in historical TC intensity records. It is clear that understanding the possible impacts of global warming on TC intensity is complicated by various natural variations and uncertainty in historical TC data.

Some studies have been conducted on the decadal and interdecadal variations in TC frequency and tracks over the WNP basin (Yumoto et al., 2001; Matsuura et al.,2003; Yumoto et al., 2003; Ho et al., 2004; Liu and Chan,2008; Kim et al., 2010). Based on the Regional Specialized Meteorological Center of Tokyo (RSMC) TC dataset since 1951, Yumoto and Matsuura, (2001), Yumoto et al., (2003) and Matsuura et al., (2003) found significant variations in TC frequency with a period of about 20 years and suggested that the interdecadal variability was associated with increased (decreased) SST in the central and eastern equatorial Pacific, which strengthens tropical westerlies (easterlies), leads to the eastward (westward) extension (retreat) of the monsoon trough and an anomalous cyclonic (anticyclonic) circulation east of the Philippines, and thus an increasing (decreasing) annual TC frequency. (Ho et al., 2004) contrasted the track change between the two periods of 1951-79 and 1980-2001 with the JTWC best track data and linked the interdecadal change to the westward expansion of the North Pacific subtropical high. (Liu and Chan, 2008) also suggested a significant interdecadal variability of the TC tracks in the WNP basin during the period 1960-2005. Based on a TC trajectory model, Wu et al., (2005) showed that two prevailing TC tracks in the WNP basin have shifted significantly westward over the past four decades due mainly to changes in large-scale steering flows, while TC activity in the South China Sea decreased.

Despite the abovementioned work, relatively few studies have examined TC intensity changes in the WNP basin on the decadal and interdecadal time scales. Using the JTWC best track data during the period 1960-2005, Chan (2006, 2008) showed prominent interdecadal variations in the frequency of Cat45 TCs with a period of about 18-32 years and argued that the variability was due to changes in thermodynamcial and dynamical factors (e.g., SST, low-level vorticity, moist static energy and vertical wind shear), while Wu and Wang, (2008) argued that changes in the proportion of Cat45 TCs over the past three decades were associated with changes in TC formation locations and prevailing tracks. Furthermore, some studies have suggested that TC maximum wind speeds in the JTWC dataset before 1973 were overestimated (Emanuel2005, 2007). Currently, the uncertainty involved in these datasets has become an important issue in understanding the possible influence of global climate change on TC activity in the WNP basin. Given the considerable uncertainty in historical TC records, the question arises as to whether there are significant decadal and interdecadal variations of Cat45 TCs in the WNP basin, which was the main objective of the present reported study. The knowledge gained from such an analysis may help to improve our understanding of the decadal variation of Cat45 TC frequency and provide important background and useful predictors for improving the potential for decadal-scale prediction.

The remainder of the paper is organized as follows. The datasets and processing used in the study are described in section 2. The methodology and dynamically-derived climate changes in the basin-wide Cat45 TC frequency are presented in section 3. The controlling factors for decadal variability of Cat45 TC frequency are identified in section 4, and the associated large-scale pattern is investigated in section 5. Finally, a brief summary is given in section 6.

The monthly SST data used in the study were from the National Oceanic and Atmospheric Administration (NOAA) reconstructed SST dataset (version 3), with a horizontal resolution of 2°×2° (Smith and Reynolds, 2004). The wind data were obtained from the National Centers for Environmental Prediction-National Center for Atmospheric Research (NCEP-NCAR) reanalysis dataset, with a horizontal resolution of 2.5°× 2.5°(Kalnay et al., 1996). The TC information from the JTWC best track dataset included TC center positions and intensity at six-hour intervals in the WNP basin. (Wu and Zhao, 2012) indicated that the JTWC dataset is more reliable compared to other TC best track datasets available for the basin. Chan, (2008) also argued that the intensity records from the JTWC dataset are relatively reliable. For this reason, we used the JTWC dataset as the observation in this study. In the JTWC best track dataset, TC intensity was mainly estimated from aircraft reconnaissance in the pre-satellite era, which started around 1945, but was discontinued in 1987 (Landsea et al., 2006). Since then, the Dvorak technique, which was developed during the 1970s, has been used to estimate TC intensity from satellite imagery and other satellite-based measurements (Dvorak, 1975; Landsea, 2007). The evolution of the technique of TC intensity estimation can lead to inhomogeneity in TC intensity records, and thus caution should be taken to address long-term variations in TC intensity.

Emanuel, (2005) extensively discussed the evolution in measurement and estimation techniques throughout historical records of TC wind speeds. Subsequently, Emanuel, (2007) provided additional evidence supporting the need for a downward adjustment of intensities in the early part of the record using the refined combined wind-pressure relationship described in (Emanuel, 2005), and in agreement with Landsea’s (1993) earlier analysis. In the present study, following (Emanuel, 2005) (see online supporting information), we first adjusted the maximum wind speeds in the JTWC dataset prior to 1973. As shown in Fig. 1, the annual TC number was reduced prior to the 1970s following this adjustment. The annual Cat45 TC frequency was also substantially reduced prior to the 1970s (Fig. 2a). As a result, the heightened Cat45 TC activity around 1960 in the historical (hereafter, unadjusted) TC intensity records, which has been argued as the peak of interdecadal variations (Chan. 2006, 2008), vanishes in the adjusted TC intensity time series. Instead, pronounced decadal variations of the annual Cat45 TC frequency can be found, with the most prominent peaks occurring around 1969, 1991 and 2004 (Fig. 2a).

Figure 1.  The annual unadjusted (solid) and adjusted (dashed) number of tropical cyclones (TCs) reaching tropical storm strength in the JTWC best track dataset over the WNP during the period 1948-2010. TC maximum wind speeds in the JTWC dataset prior to 1973 were adjusted using the pressure-wind relationship, as described in (Emanuel, 2005).

Figure 2.  (a) Unadjusted (green), adjusted (black) and simulated (red) annual number of Cat45 TCs in the JTWC dataset over the WNP basin. A 5-yr running average was applied to the time series. (b) Normalizations of adjusted (dashed) and simulated (solid) annual number of Cat45 TCs over the WNP basin. The time series were detrended and a 5-yr running average was also applied.

Variations in the adjusted and unadjusted Cat45 TC frequencies during the period 1948-2010 were examined with a spectral analysis. As shown in Figs. 3a and b, all the unadjusted, adjusted and simulated Cat45 TC frequencies show interannual cycles, although their significant spectral peaks are not exactly the same. On the decadal scale, a significant spectral peak occurs at about 25 years in the unadjusted Cat45 TC frequency, while a 12-18-yr cycle can be found in the adjusted and simulated Cat45 TC frequency. Despite the edge effects in the spectral analysis, it was found that the results are robust because the significant periods are essentially the same when we used wavelet analysis (not shown). Zhao et al., (2011) suggested that interannual variations in TC intensity are dominated by the interannual variability of the monsoon trough associated with SST changes, but the associated physical mechanisms of the decadal variations of Cat45 TC frequency are not well understood in the literature, due mainly to short TC records and uncertainty in historical intensity records.

Recently, an approach was used to assess historical TC intensity records (Wu and Zhao, 2012) in which the intensity of each storm was numerically simulated with a TC intensity model (Emanuel et al., 2006; Emanuel et al., 2008). Using this approach, Wu and Zhao, (2012) found that the evolution of basin-wide TC intensity in the JTWC best track dataset can be reasonably well simulated over the period 1975-2007. In addition, they suggested that the Cat45 TC frequency is sensitive to changes in vertical wind shear (VWS) and SST. For this reason, we focused on the annual Cat45 TC frequency and conducted numerical simulations similar to those in (Wu and Zhao, 2012) by extending the study period back to 1948.

Figure 3.  Spectral analysis of the detrended annual number of (a) unadjusted, (b) adjusted and (c) simulated Cat45 TCs during the period 1948-2010 in the WNP basin. The 95% confidence levels with respect to the red noise spectrum are shown by the black dashed line.

The TC intensity model used was an axisymmetric numerical atmospheric model, coupled with a simple one-dimensional ocean model (EmanueL, 2006; Emanuel, 2008). Using the intensity model, Emanuel et al., (2008) explored the influence of various environmental factors on TC intensity. In the present paper, the observed TC tracks during the period 1948-2010, and the corresponding VWS and SST along the TC tracks are presented. The VWS was calculated as the magnitude of the monthly vector difference between 850-200 hPa, and its effect was parameterized in the TC intensity model. Note that the influence of SST changes associated with TC-ocean interactions was not included in this study. All of the observed TCs in the JTWC dataset were allowed to move along the observed TC tracks and their intensity evolution was simulated in the intensity model. The intensity model was initialized with a warm-core cyclonic vortex. The maximum wind speed of the initial vortex was set to be 21 m s^-1 after a series of numerical experiments because the model vortex weakens at the beginning of the simulation, which was a technique also adopted in Wu and Zhao, (2012) and Zhao et al., (2011). Note that the intensity of the initial vortex had little influence on our simulations; very similar patterns and temporal variations, except for the magnitude of TC intensity, were found when we conducted a few sensitivity experiments. The other parameters of the initial vortex were the same as those in Emanuel et al., (2008). The same model setup was used for all the simulations in this study, and the Student’s t-test was used to test the statistical significance at the 95% confidence level (Wilks, 1995).

In the control experiment (CTRL) (Table 1), all of the TCs moved with their observed tracks, experiencing the corresponding observed monthly VWS and SST along the TC tracks during the period 1948-2010. As shown in Fig. 2a, the simulated Cat45 TC frequency agreed well with the adjusted Cat45 frequency. The correlation coefficient was 0.86 during the period 1948-2010, statistically significant with an effective sample size of 51, as referred to by Dawdy and Matalas, (1964). As shown in Fig. 3c, the significant spectral peaks in the simulated Cat45 TC frequency occurred at about 2.5, 4, 5, 9 and 15 years. In agreement with the adjusted Cat45 frequency, the 18-32-yr variability was not statistically significant in the simulated Cat45 TC frequency, suggesting that such variability was due mainly to the different wind-pressure relationships used in the JTWC records during the period 1948-2010.

Given the observed TC tracks used in the simulation, uncertainty in the simulation may have resulted from missing or incomplete TC track records in the JTWC dataset, especially during the pre-satellite era (Landsea, 2007). To examine this, the JTWC best track data were divided into three periods: 1948-64, 1965-72 and 1973-2010, which were chosen based primarily on developments in observational techniques. From 1965, satellite data started to be used to monitor TCs (Landsea, 2007; EmanueL, 2008) and a new wind-pressure relationship was used for estimating TC intensity from 1973 (Emanuel, 2005). We examined the average TC lifetime and the average time for a TC to achieve Cat45 intensity, and the results were as follows. For the three periods, the average lifetime of Cat45 TCs was 6.68, 8.25 and 8.84 days, while the mean time taken to achieve Cat45 intensity was 2.61, 3.13 and 3.30 days, respectively. Further calculations suggested that the mean lifetime during the pre-satellite period was indeed significantly shorter than the two post-satellite periods, indicating incomplete track records in the JTWC dataset during the pre-satellite period. Moreover, Cat45 TCs had relatively short lifetimes in the pre-satellite era.

However, the adjusted Cat45 TC frequency prior to the satellite era may be reasonable. First, owing to aircraft reconnaissance during 1948-1972, and the relatively longer duration of this period, Cat45 TCs should have had less chance to be missed in the JTWC dataset than weak TCs. Second, the locations where TCs reached Cat45 intensity occurred mostly in the western part of the WNP basin, and were thus well covered by the aircraft reconnaissance missions operated from Guam (Fig. 4).

Figure 4.  The climatological mean SST (dashed contours) and mean vertical wind shear (solid contours) during the period 1948-2010 over the WNP basin. Dots represent locations of the first occurrence of Cat45 TC intensity reported in the adjusted JTWC dataset. The green box highlights the location of the territory of Guam, where JTWC provided tropical cyclone watches and warnings prior to 1999.

In addition, the simulated Cat45 TC frequency was highly consistent with that in the adjusted dataset. Despite the relatively short duration of the pre-satellite era in the dataset, both the simulated and adjusted Cat45 TCs showed a similar rapid intensification process. It was noted that simulated TC intensification is smaller than observed (Wang and Zhou, 2008). In the present study, the observed (simulated) Cat45 TCs on average took 48 hours to increase about 32 m s^-1 (23 m s^-1 )in the maximum wind before reaching Cat45 intensity during the period 1973-2010. The average time for a TC to achieve Cat45 intensity in the model was about 3.3 days, which also compared well with observations. Associated with the lifetime of 6.68 days in the pre-satellite period, there was enough time to allow TCs to reach the Cat45 intensity in the model.

As shown in Fig. 2a, the adjusted records and simulation clearly indicate that the frequency of Cat45 TCs persistently increased with an almost linear trend over the past 60 years, although the simulated trend was smaller than the adjusted. While further study is needed to verify the increasing trend, we focus here on the decadal variability of the Cat45 TC frequency. For this purpose, all variables mentioned in the following discussion were detrended to reduce the influence of the long-term trend and a 5-yr running average was also used to reduce the interannual influences. The influences of SST and VWS on the climate change of TC intensity have been extensively discussed in previous studies (Goldenberg et al.,2001; Emanuel, 2005, 2008; Webster et al., 2005; Wu et al.,2008; Zhao et al., 2011; Wu and Zhao, 2012). Wu and Wang, (2008) suggested that changes in the basin-wide TC intensity can be associated with shifts of the prevailing TC tracks. We examined the individual contributions of changes in TC tracks, SST and VWS to the decadal variability of Cat45 TC frequency by conducting three sensitivity experiments.

As shown in Table 1, Experiment T-Clim (V-Clim) was the same as CTRL, but with the SST (VWS) along the TC tracks using the climatological mean during the period 1948-2010, while experiment VT-Clim was run with both the SST and VWS fixed in the climatological mean environment. The simulated annual Cat45 TC frequency in the three sensitivity experiments were well correlated with that in the CTRL experiment (correlation coefficients all exceeded 0.8) (Fig. 5a), suggesting that neither SST nor VWS was the primary controlling factor for the decadal variations (Fig. 5a). The influence of SST (VWS) changes on the decadal variability of the basin-wide Cat45 TC frequency can be obtained by contrasting V-Clim (T-Clim) and VT-Clim, respectively. As shown in Fig. 5b, the simulated difference of Cat45 TC frequency between the experiments was weakly correlated with CTRL (both correlation coefficients were about 0.09), also suggesting that SST (VWS) change had little direct influence on the decadal variability of Cat45 TC frequency. However, the combined effect of SST and VWS changes, which was examined by contrasting CTRL and the difference between CTRL and VT-Clim, was much larger than the sum of their individual contributions (Fig. 6a). Its correlation with the simulated Cat45 frequency was 0.35, significant at the 95% confidence level. This indicates that a combination of SST and VWS changes contributes to the decadal variability in the basin-wide Cat45 TC frequency.

Wu and Wang, (2008) suggested that the shifts in the TC prevailing tracks may have allowed more TCs to follow a longer journey, favoring the development of intense TCs. In the present study, the effect of the TC track change could be derived by contrasting the Cat45 TC frequencies between CTRL and VT-Clim. As shown in Fig. 6b, both of them were well correlated, with a correlation coefficient of 0.82, suggesting that the track change can almost account for the decadal evolution of the basin-wide Cat45 TC frequency, in part because of the combined effect of SST and VWS change.

Figure 5.  (a) Time series of the normalized annual simulated Cat45 TC number in CTRL (solid line), V_Clim (dashed line with closed dots), VT-Clim (dashed line with plus symbols) and T-Clim (dashed line with open dots) during the period 1948-2010, respectively. (b) The SST effect (dashed line with closed dots) and vertical wind shear effect (dashed line with open dots) on the decadal variability of Cat45 TC frequency, which can be examined by contrasting experiments VT-Climand V-Clim (T-Clim), respectively. The solid line as shown in (b) also represents the simulated Cat45 TC frequency from the CTRL experiment.

Figure 6.  (a) Time series of the normalized annual simulated Cat45 TC frequency for CTRL (solid line) and the difference between CTRL and VT-Clim (dashed line), in which the environmental vertical wind shear and SST changed while the TC tracks remain unchanged, indicating the combined effect of SST and vertical wind shear on the decadal variability of Cat45 TC frequency. Similarly, (b) shows the time series of the normalized annual simulated Cat45 TC frequency for CTRL (solid line) and VT-Clim (dashed line), which indicates the effect of track changes on the decadal variability of Cat45 TC frequency. See the text for further detail.

As indicated above, the numerical results suggested the change in TC tracks exerts the dominant influence on the decadal variability of Cat45 TC frequency over the WNP basin. In order to understand how track changes can affect the decadal variations of Cat45 TC frequency, we contrasted (see Fig. 2b) the large-scale patterns between positive phases (1952-57, 1965-73, 1988-96 and 2001-06) and negative phases (1950-51, 1958-64, 1974-87, 1997-2000 and 2007-08). In this section, we focus mainly on the associated large-scale patterns between positive and negative phases since 1965, mainly because this was when satellite data started to be used in locating TC centers; thus, the TC track information in the best track data during the period is relatively reliable.

Figure 7 shows the differences in the TC formation frequency between the positive and negative phases. Compared to the negative phases, the TC formation is enhanced in the southeast part of the WNP basin (10°-20°N, 135°-170°E) in a positive phase, with a moderate decrease over the northwest part (10°-25°N, 120°-135°E). Note that although some TCs over the WNP basin might be missing in the JTWC best track data before the satellite era, we found that the composite difference of TC genesis frequency between the positive and negative phases was similar when we used the data since 1948. This suggests that TC formation locations shift eastward during positive phases, which was also indicated by the mean TC duration. For each individual year, the mean duration was defined as the average duration of all TCs that occurred in the peak TC season (July-September). The mean duration was 5.9 days for positive phases, which was significantly longer than the 4.5 days for negative phases.

Figure 7.  Composite difference of TC genesis frequency (×100) over the peak TC season (July-September) between positive and negative phases. The difference with shading is statistically significant at the 50% level.

Figure 8.  Correlation between the adjusted Cat45 TC frequency over the WNP basin and SST over the peak TC season (July-September) for (a) the period 1965-2010 and (b) 1948-2010. The shaded areas indicate significance at the 95% confidence level.

We further calculated the correlations between SST and the adjusted Cat45 TC frequency. A similar correlation pattern over the tropical central and eastern Pacific was found for the periods 1965-2010 and 1948-2010, respectively (Fig. 8). There is a clear suggestion that the increased Cat45 TC activity in positive phases is closely associated with SST changes over the tropical central and eastern Pacific. Figure 9 shows the differences of the 850-hPa winds and SST between positive and negative phases during the peak TC activity season since 1965. We also found a similar large-scale pattern between the positive and negative phases when we used the data from 1948-2010 (not shown). To the west of the positive SST differences, westerly wind differences extended from the Philippines Sea to 170°E. (Yumoto and Matsuura, 2001), (Yumoto et al., 2003) and (Matsuura et al., 2003) suggested that increased (decreased) SST in the central and eastern equatorial Pacific strengthens the tropical westerlies (easterlies), and leads to the eastward (westward) extension (retreat) of the monsoon trough. The eastward extension of the monsoon trough can be considered the response to the heating associated with the warming SST anomalies (Gill,1980; Holland1995).

(Ritchie and Holland, 1999) found that more than 75% of TCs in the WNP basin are associated with the monsoon trough, and previous studies have found that TCs tend to form near the eastern end of the monsoon trough (Holland1995; Briegel1997; Ritchie1999). Therefore, we can conclude that the atmospheric response to the warming SST over the tropical central and eastern Pacific leads to the eastward extension of the monsoon trough, providing a favorable large-scale environment for TC formation and resulting in the eastward shift in TC formation locations. As suggested by (Wu and Wang, 2008), the eastward shift in the TC formation locations allows more TCs to follow a longer journey—favoring the development of Cat45 TCs—compared to negative phases.

Figure 9.  Difference of the wind fields (vectors; units: m s^-1 )at 850 hPa and SST (contours; interval for solid contours: 0.1°C) over the peak TC season (July-September). Only differences of wind fields and SST above the 95% confidence level are depicted by vectors and color shading, respectively.

Using JTWC best track data during the period 1948-2010, the decadal and interdecadal variations of Cat45 TCs in the WNP basin and the associated mechanisms were examined. The basin-wide annual Cat45 TC frequency was numerically simulated in a TC intensity model by allowing all of the observed TCs to move along the observed TC tracks. The simulated annual Cat45 TC frequency was in good agreement with observations when the TC intensity prior to 1973 was adjusted with the combined wind-pressure relationship, as in Emanuel, (2005). The simulated and adjusted time series showed decadal variations with three peaks around 1969, 1991 and 2004. Although interdecadal variability with an 18-32-yr period was found in the unadjusted Cat45 TC frequency during the period 1948-2010, it was insignificant in the simulated and adjusted Cat45 TC frequency. We argue that that the interdecadal variability in Chan (2006, 2008) was due mainly to the different wind-pressure relationships used in the JTWC records prior to 1973.

Numerical results showed that changes in TC tracks constitute the most important factor for decadal variations in the Cat45 TC frequency, although the combined effect of changes in SST and VWS also contributes to the decadal variability. Further analysis suggested that the decadal fluctuations in TC tracks are closely associated with the eastward shift in TC formation locations. The warming SST over the tropical central and eastern Pacific leads to the eastward extension of the monsoon trough and then the eastward shift in TC formation locations. As suggested by (Wu and Wang, 2008), the eastward shift in the TC formation locations favors more TCs having a longer journey, and thus also having a greater chance to develop into Cat45 TCs.

It should be noted that the adjusted records and simulation clearly showed that the frequency of Cat45 TCs persistently increased with an almost linear trend over the past 63 years, although the simulated trend was smaller than the adjusted. As suggested in Wu and Wang, (2004) and Wu et al., (2005), long-term changes in prevailing TC tracks in the WNP basin have been identified, which may have contributed to the increase of the Cat45 TC frequency. Previous studies have also argued that local SST change has contributed to the increasing trend of TC intensity in the WNP (Emanuel, 1987; Knutson et al., 1998; Knutson and Tuleya, 2004; Bender et al., 2010). Furthermore, a recent statistical analysis indicates that the observed TC track changes in the WNP are linked to global SST warming (Wang et al., 2011). However, the long-term increasing trend may be subject to uncertainty in TC historical data and further studies are therefore needed.