Introduction

According to the data from the International Best Track Archive for Climate Stewardship (IBTrACS), approximately half of the total number of tropical cyclones (Category 1 and above, hereinafter TCs or typhoons) that enter the South China Sea (SCS) from western North Pacific pass through a narrow gap between Luzon and Taiwan: the Luzon Strait (Fig. 1a)1. These Luzon Strait TCs (LSTCs) remain above the ocean throughout their lives, minimally affected by terrain1, and nearly all of them make landfall along southern China and Vietnam, causing considerable property losses and human sufferings2, 3. The coasts of southern China and Vietnam have in fact been identified to be amongst the world’s most vulnerable to floods from TCs in future projected sea-level rise under climate change scenarios4, 5. The majority of LSTCs reached their maximum intensities in the warm waters east of Luzon and in the Kuroshio, and then weakened in the SCS1. The fact that the majority of LSTCs weaken after passing through the Luzon Strait suggests the existence of some common environmental factors in the ocean and/or atmosphere which are unfavorable to TC-intensity in SCS, thus making SCS a natural “buffer” which shields southern China and Vietnam from otherwise stronger typhoons from the open western Pacific. This study seeks to describe and understand what factors contribute to the buffer and how the buffer has changed over inter-decadal periods based on observations and reanalysis data. It should be noted that the same environmental factors exist, and our results would apply also to TCs that enter the SCS south (north) of the Luzon Strait; however, the intensities of these TCs may also be strongly affected by the mountainous terrain of Luzon (Taiwan). The LSTCs represent an ideal limiting group of TCs with the maximum potential of reaching their lifetime maximum intensities (LMIs) inside SCS. Besides satisfying our scientific curiosity, knowledge about the behaviors and variability of LSTCs and their dependence on the buffer can help improve storm risk assessments, which are required to formulate plans for long-term adaptation and storm preparedness. We will show that the buffer has weakened since the mid-1980s and early 1990s – coincident with the beginning of rapid rise in sea surface temperature (SST) and sea level6, 7.

Figure 1
figure 1

(a) Tracks of typhoons from western Pacific into SCS through the Luzon Strait, from 1951–2013, updated from Sun and Oey1 using the latest IBTrACS observations. For each track, colors show 6-hourly wind speeds normalized by the maximum wind speed for that track; darkest brown color corresponds to normalized wind speeds >0.95, and indicates where and when the typhoon was near its LMI. Top left shows the total TCs into SCS (278), subdivided into those passing through the Luzon Strait (i.e. the LSTCs, 137), and south of the strait (141, tracks not shown). Red contours show mean absolute dynamic topography MADT (m) from AVISO http://www.aviso.oceanobs.com/. (b) The number of LMIs per 1o × 1o grid (normalized by the maximum number = 6), according to tracks in (a). Tracks within 50 km of the northern (southern) tip of Luzon (Taiwan) were excluded. Maps plotted using MATLAB Version#R2012a (7.14.0.739) 64-bit (glnxa64) (https://www.mathworks.com/support/sysreq/previous_releases.html).

Full size image

Results

Figure 1a (dark brown color along each track) shows the location where the TC is near its LMI, and Fig. 1b the number of LMIs per 1° × 1° grid. The majority of LSTCs (70%) reach their peak intensities in the western Pacific (WP), and weaken as they enter the SCS1. The remaining 30% continue to intensify and reach peak intensities inside the SCS, mostly over the warm waters of the northern SCS shelves prior to landfalls (Fig. 1b). The SCS therefore serves as a natural buffer to dampen TCs. What are the dominant oceanic and/or atmospheric environmental factors that control the SCS buffer, and how do they vary over inter-decadal and longer time scales?

PDO and SCS buffer

The Pacific Decadal Oscillation8 (PDO) and the El Niño Southern Oscillation9 (ENSO) can potentially affect the SCS buffer. TC activity depends on the phases of ENSO and PDO3, 10,11,12,13. We focus on inter-decadal changes and begin with an examination of how the SCS buffer varies with PDO. As a measure of the buffer’s strength, we define the ratio:

$${R}_{ty}{={\rm{N}}}_{{\rm{SCS}}}/({{\rm{N}}}_{{\rm{SCS}}}{+{\rm{N}}}_{{\rm{WP}}}){={\rm{N}}}_{{\rm{SCS}}}/{{\rm{N}}}_{{\rm{TOT}}},$$
(1)

where NSCS (NWP) is the number of LSTCs per typhoon season from June through November that intensify (i.e. have LMIs) in SCS (WP), and NTOT = NSCS + NWP is the total number of LSTCs per season. The R ty  ≤ 1, and a smaller (larger) ratio indicates a stronger (weaker) buffer when less (more) LSTCs intensify in SCS. For the analysis period from 1951 to 2013, the R ty is larger (mean = 0.34) during negative PDO (denoted by −PDO) years, and smaller (mean = 0.16) during positive PDO (denoted by + PDO) years (Fig. 2a). The R ty and PDO appear to be anti-correlated, Corr(R ty , PDO) = −0.45 (Fig. 2b); however, the sample is too short for their connection to be statistically meaningful, and another measure is described below. Nonetheless, it appears that a larger percentage of typhoons would gain strength and reach maximum intensity in SCS during the −PDO than the +PDO years after passing through the Luzon Strait.

Figure 2
figure 2

(a) Numbers of WP-(red bars) and SCS-(black bars) intensifying LSTC’s based on the IBTrACS data from 1951–2013; top display shows values of R ty averaged over periods of positive and negative PDOs indicated by vertical dashed lines. (b) R ty anomaly normalized by its standard deviation σ (s.d. = 0.27) (black bars) and the PDO index (red dashed line), 6-year running mean applied to both; a tapered smoothing is applied to end points; top-left display shows the correlation between R ty and PDO and the corresponding p-value. (c) Probability distribution of 6-hourly intensity changes (ΔV) of LSTCs in SCS sampled by positive and negative PDOs. Units are knots [1 knot (kt) = 0.51 m s−1] per 6 hours. Data are in bins of 5-kt resolution. Error bars show ±σ from the mean probability for that bin.

Full size image

A number of studies3, 14 have shown that during −PDO (+PDO) years, typhoons tend to form farther west (east) in the tropical North Pacific (see Supporting Information Fig. S1), resulting in more (less) typhoons entering the South China Sea (Fig. S2). Between 1951 and 2013, the average number of LSTCs per season is 2.6 during the −PDO years, and 1.2 during the +PDO years. However, since R ty is a ratio, a larger supply of LSTCs (i.e. NTOT) during −PDO years does not necessarily lead to a higher percentage of typhoons that intensify in SCS, i.e. to a larger R ty . The probability distribution of intensification rates (ΔV) of LSTCs in SCS in fact shows that the mean intensification rates between +PDO and −PDO years are well separated. For a wide range −10 ≤ ΔV ≤ 15 kt per 6 hours, the intensification rates are positive during −PDO and negative during +PDO (Fig. 2c). The result is consistent with the apparent anti-correlation between R ty and PDO, noted above. Changes in R ty are therefore more likely caused by changes in the oceanic and atmospheric environments, which will be examined next.

Environmental conditions

TC internal dynamics play an important role in intensity changes15. To study inter-decadal variation of LSTCs, we examine instead how large-scale environmental conditions have evolved, and seek some quantifiable metric(s) to describe the change. Environmental factors known to affect storm intensity are: (i) relative humidity (RH)16,17,18,19,20,21,22, (ii) environmental vertical wind shear V s of the horizontal wind16, 20, 23,24,25,26,27,28, (iii) large-scale vorticity (VOR)16, 24, (iv) SST16, 20, 24, 29,30,31,32, (v) maximum potential intensity (MPI)20, 28, 32, and (vi) the depth of upper-ocean warm layer, measured for example by the depth Z 26 of the 26 °C isotherm1, 15, 18, 32, 33. For a shallow Z 26, typhoon winds can more easily mix and entrain cold subsurface water to the surface, lowering the SST and reducing the storm intensity or even killing it. On the other hand, for thick Z 26, the SST may remain above 26 °C despite the strong wind mixing, allowing the storm to maintain its strength or even to intensify. We calculate TC maximum potential intensity, MPI(SST) (m s−1), which depends on SST and the structure of the overlying atmosphere34, 35. Price36 suggested that, due to mixing by the strong TC wind, a temperature (denoted here by Tmix) depth-averaged over some depth hmix would more realistically approximate the SST under the TC; the idea has been applied in various TC-genesis and intensity-change studies37,38,39. Here we calculate Tmix by assuming a linear temperature profile between the surface and Z 26 below the surface, which is then averaged over hmix:

$${{\rm{T}}}_{{\rm{mix}}}=\text{SST}\,-\,({\rm{SST}}\,-\,26){\text{h}}_{\text{mix}}/(2{Z}_{26}).{H}_{v}(\text{SST}-26),$$
(2)

where H v is the Heaviside function. A value of hmix = 80 m is used37. Equation (2) shows that, as the Z 26 deepens, Tmix increases to approach the temperature of the warmed ocean surface where the SST is warmer than 26 °C, thus modeling the sheltering characteristic of a deep and warm upper-ocean layer. On the other hand, when Z 26 decreases, Tmix becomes increasingly cooler than SST, thus modeling the cooling due to the more ready entrainment of cold water from beneath the shallower warm surface layer. As SST ≤ 26 °C, Tmix is set to SST. The Z 26 is calculated using the monthly WOA data plus an anomaly Z 26 ’. The ocean is divided into 2 layers separated by the 26 °C isotherm. Monthly Z 26 and density difference Δρ/ρo between the 2 layers are calculated using WOA40, 41 (Fig. S3); then the sea-level anomaly (SLA = η’)42, 43 from AVISO is used to obtain Z′26 ≈ η′ρo/Δρ. The MPI based on Tmix is denoted as MPI(Tmix).

Intensity Change Index (ICI)

We develop an index to describe the above environmental effects on TC intensity change (see Methods):

$$ICI={(\frac{RH}{50})}^{0.8}\,{(\frac{MPI}{70})}^{3.2}\,{(1+0.1\times WSH)}^{-1.6}.$$
(3)

The ICI is similar in form to the Genesis Potential Index (GPI)44, 45. However, while GPI is useful for studying preferred locations of TC genesis as a function of environmental conditions, we show below that ICI is a preferred index for studying TC intensity change.

Composites for +PDO and −PDO phases

We composite the above variables according to +PDO and −PDO during the satellite altimetry period from October 1992 to 2013 (Fig. 3). We focus in the northern SCS (15–24N & 110–121E where all LSTCs pass) and use the notation Δ−+(α) to denote the difference in the variable α: −PDO composite minus +PDO composite. Since +PDO is from 1992~2005 and −PDO from 2005~2013 (see Fig. 2), Δ−+ also represents the recent decadal change between the two epochs. We find that Δ−+(MPI) based on MPI(SST) is very weak, and therefore show only MPI(Tmix) and the corresponding ICI in Fig. 3e,f. The Δ−+(WSH) decreases while Δ−+(VOR), Δ−+(Z 26 ) and Δ−+(MPI) increase. According to the above cited works, these changes favor TC-intensification and contribute to increased Δ−+(ICI) (Fig. 3f), consistent with Fig. 2c that the intensification rates are positive during −PDO and negative during +PDO.

Figure 3
figure 3

Composites of (a) RH, (b) WSH, (c) VOR, (d) Z 26 , (e) MPI and (f) ICI for +PDO and −PDO from 1992 to 2013. Since +PDO is from 1992~2005, while −PDO is from 2005~2013, each pair of plots approximately describes the decadal change between the two epochs. Stipples indicate regions where the composite mean is significant. Dashed box in northern SCS shows the main area of interest. In each panel abscissa is longitude and ordinate is latitude. Maps plotted using MATLAB Version#R2012a (7.14.0.739) 64-bit (glnxa64) (https://www.mathworks.com/support/sysreq/previous_releases.html).

Full size image

We evaluate the contributions of the individual variables to TC intensity changes. We first compare with the literature the magnitude of the contribution; we then quantitatively assess the contribution. The Δ−+(RH) ≈ +5% (Fig. 3a) is similar to the values previously reported as TC changes intensity16, 19, 21, 22; however, the magnitude is smaller than the values of ±10% that distinguish active and inactive TC periods46, and in SCS the climatological RH is high, ≈70% or more at 600 hPa, which is generally favorable for TCs irrespective of the phases of PDO. A sufficiently strong WSH > 10 m s−1 can rip a TC apart, weakening or even destroying it23,24,25,26,27,28. However, the Δ−+(WSH)≈−0.8 m s−1 (Fig. 3b) is small in magnitude relative to the climatological WSH of ~8 m s−1 over SCS1, 47, 48, and is weak compared to changes of 3~5 m s−1 found previously for significant changes in TC intensity49. The Δ−+(VOR) ≈ −2 × 10–6 s−1 (Fig. 3c) is similar to the values previously found in studies of TC intensity change16, 50, but the composite is insignificant in northern SCS. The increased Δ−+(Z 26)≈ + 16 m (Fig. 3d) is a significant fraction (~25%) of the climatological mean Z 26 = 50~60 m in northern SCS (Fig. S3). A shallow Z 26 allows cold subsurface water to be more easily mixed and entrained to the surface by typhoon winds41 lowering the SST which in turn weakens the storm1, 19, 29,30,31,32,33.

We extend the above composite analysis for a longer period before the satellite altimetry era i.e. before Oct/1992. To estimate η’ (hence Z′26), we note that SLAs in SCS and warm pool are correlated (Fig. 4a)51. In turn, both are closely related to the fluctuations of the North Equatorial Current bifurcation (NECBF) point near 12~13N east of the Philippines, such that sea level in SCS drops (rises) as the NECBF shifts poleward (equatorward)51,52,53, as shown in Fig. 4b. We therefore use a proxy of NECBF54 to project it to AVISO SLA and reconstruct η’ from 1982 to 2013. We limit our analysis to the period after 1980, since prior to that date the use of the NCEP data for TC parameters may be unreliable55, 56. We repeat the PDO-composite analysis. The results (Fig. S4) are similar to Fig. 3, showing that during −PDO (+PDO) years ICI is positive (negative) and the conditions are more (less) favorable for TC to intensify in SCS.

Figure 4
figure 4

(a) Homogeneous correlation map of mode-1 northern SCS EOF(η’) (57%; domain shown in blue) with AVISO SLA in the western North Pacific from 1993–2013, showing that the SCS η’ is closely related to the SLA fluctuations over the warm pool, in particular to the fluctuations in the vicinity of the NECBF point near 12~13oN. (b) Correlation between NECBF with AVISO SLA in the red domain of (a); dots show correlations significant at the 95% confidence level. In SCS and warm pool, SLA drops (rises) as NECBF shifts north (south) [Chang and Oey 2012]. Maps plotted using MATLAB Version#R2012a (7.14.0.739) 64-bit (glnxa64) (https://www.mathworks.com/support/sysreq/previous_releases.html).

Full size image

Intensity changes

To quantitatively assess the influence of various environmental factors on TC intensity changes in SCS, we use the 6-hourly IBTrACS data to calculate the annual (typhoon season) mean intensification rates. We regress ΔV against the variables (Fig. 5a). We find that Tmix and MPI(Tmix) are significant predictors of typhoon intensity changes in SCS. The SST, RH, WSH and VOR are poor predictors with large p-values. Except for VOR, their combined contribution to ICI is non-negligible however, and results in significant high correlation (r = 0.87, Fig. 5a); thus ICI accounts for 75% of the total variance of TC intensity change in SCS. These findings support the results of the composite analyses. Indeed, the probability distribution of intensification rates sampled by ICI (Fig. 5b) shows that the mean intensification rates between positive and negative ICIs are separated over a wide range up to |ΔV| ≤ 15 kt per 6 hours, and the mean intensification rates are positive for positive ICI and negative for negative ICI.

Figure 5
figure 5

(a) Regression lines of normalized (by s.d. σ) 6-hourly intensity changes (ΔV) of LSTCs vs. the indicated variables, from 1993 to 2013. Legends show the r, p and slopes for ICI, MPI, Tmix which are all significant at the 98% confidence level: shown as solid dark lines; grey dashed lines are for insignificant regressions (p ≥ 0.53). (b) Probability distribution of ΔV sampled by positive and negative ICIs, from 1993 to 2013. Units are knots [1 knot (kt) = 0.51 m s−1] per 6 hours. Data are in bins of 5-kt resolution. Error bars show ±σ from the mean probability for that bin.

Full size image

Summary and Discussion

We have demonstrated that a greater percentage of typhoons entering the SCS (i.e. greater R ty ) intensify during years of −PDO than +PDO. The dominant cause is a substantial (25%) deepening of the Z 26 that tends to shelter the warm surface ocean from cool subsurface waters, which increases T mix . We noted before that T mix increases with both Z 26 and SST and Eq. (2) expresses the ‘sheltering’ property as described previously. Mechanistically, the deepened Z 26 makes it more difficult for TC winds to mix and entrain cooler subsurface waters to the warm surface, increasing the MPI and allowing a greater percentage of TCs to intensify in SCS. Thus we develop ICI based on MPI(T mix ) to describe the TC intensity changes. To assess ICI for regions other than SCS, we calculate the correlation between ICI and ΔV (Fig. 6) over the western North Pacific. It shows a highly significant band of correlation with r ≈ 0.6~0.8 from the Philippines Sea to northern SCS. Such a good predictability is clearly desirable for storm preparedness and risk analysis. As altimetry observations are now routinely available, the ICI can be used to formulate a more accurate TC-intensity prediction scheme.

Figure 6
figure 6

The correlation between ICI and ΔV. Filled (open) circles denote regions where the correlations are significant at the 95% (90%) confidence level. Maps plotted using MATLAB Version#R2012a (7.14.0.739) 64-bit (glnxa64) (https://www.mathworks.com/support/sysreq/previous_releases.html).

Full size image

Why does Z 26 in SCS deepen for −PDO? Local changes may contribute. The Z 26 can deepen due to a negative wind stress curl over SCS which depresses isotherms by Ekman downwelling. However, the surface wind stress curl (not shown) is positive during −PDO years in SCS and cannot contribute to the deepened Z 26 . The Z 26 can also deepen due to heat transport into SCS from the western Pacific warm pool through the Luzon Strait. However, the Luzon Strait transport is positively correlated with PDO52, so that the heat transport deficit (surplus) during -POD (+PDO) years cannot cause Z 26 to deepen (shoal). Instead, we argue that the deepening of Z 26 in SCS is remotely driven (Fig. 4), closely tied to the deepening of the warm pool waters as PDO turns negative.

Sea level in the western tropical Pacific including the SCS has undergone a rapid rise in recent decades since the early 1990s57; the trend remains significant in SCS during the typhoon season (Fig.S5). The rise is dynamically caused by the strengthening of the easterly trade wind as part of the response to the recent global warming trend6, 58. Independent analysis59 also supports the idea that the strengthening of the Pacific Walker circulation since the mid-1970s can be explained by rising SST. Therefore, since as sea level rises Z 26 deepens, both it and the rising SST lead to increased T mix , hence also a corresponding increase in ICI. Evidence that the Z 26 has indeed increased more rapidly in recent decades compared to earlier decades can be seen by comparing the two Z 26 -composites (Figs 3d and S4d), suggesting a modulation in recent decades by the rapid sea-level rise previously identified57. Should negative PDO continue, and sea level and SST continue to rise, our study suggests that the SCS buffer may further weaken. In that case, one may expect potentially greater number of stronger landfalling typhoons along the SCS coastlines in the coming decades.

Methods

Six-hourly typhoon locations and corresponding maximum wind speeds from 1951 to 2013 were obtained from the IBTrACS dataset60. Monthly atmospheric variables (for RH, WSH and VOR) were from NCEP/NCAR reanalysis data. Weekly satellite sea-level anomaly data from 1992/October to 2013 were downloaded from AVISO, and monthly averaged. Monthly World Ocean Atlas (WOA) data is used to estimate climatological Z 26 . The SST from GHRSST (1982–2013, 1/4o × 1/4o) is used in the calculations of various metrics related to TC-intensity. Monthly PDO time series8 is used to compute positive and negative PDO composites (i.e. arithmetic averages) of various quantities. We focus on long-term variability and, except for the IBTrACS data, monthly climatology is removed from each time series. All quoted trends and correlations are given at the 95% confidence level i.e. the probability that the trend or correlation is due to randomness of the time series is p < 0.05.

We develop an index to describe the environmental effects on TC intensity change:

$$ICI=F({G}_{\varpi };{G}_{RH};{G}_{MPI};{G}_{WSH}),$$

where F and the G’s are some functions of the subscripted variables, ϖ = absolute vorticity at 850 hPa (s−1), RH = relative humidity at 600 hPa, MPI = MPI(Tmix), and WSH = |V s|. The functional F is a product of G’s, and the G’s some fractional power functions, i.e. ratio of polynomials of the variables raised to some power. Using the 6-hourly IBTrACS data, the exponents are calculated using an iterative Newton algorithm to maximize the percentage variance of ΔV described by ICI, yielding Eq. (3). The effect of \({G}_{\varpi }\) is very weak and is omitted in (3). We also considered including the effect of the TC translation speed on TC intensity change28; it too has only minor effects. We compare ICI with GPI. In northern SCS, ICI accounts for 75% of the total variance of ΔV, compared to 30% for GPI.