Introduction

Tropical cyclones (TCs) are accompanied by heavy precipitation, storm surges, and strong winds, and often cause tremendous damage and casualties in the landfalling and adjacent coastal areas1,2,3,4. Once TC makes landfall, its intensity generally decays rapidly due to the reduction of available moisture and the increase in friction2,5,6, and thus its destruction tends to be confined to a period of several days and to a limited areas inland. As the climate warms, the warmer air contains more moisture, which allows TCs to reach a stronger intensity and decay more slowly7,8,9. Since the damage caused by TCs tends to increase with the maximum sustained wind speed (i.e., TC intensity)10,11, TCs with a slower rate of decay can cause greater potential damages to inland areas from strong winds and heavy precipitation. Moreover, when a TC decays more slowly, it tends to maintain its intensity (and destruction potential) for longer, and may penetrate further inland, resulting in greater risk to regions further inland12,13. Therefore, it is essential to understand the changes in the decay of the intensity of landfalling TCs and the associated precipitation responses, hence enhancing the adaptation to these changes and mitigation of mitigate potential losses in coastal areas.

A significant slowdown In the decay of landfalling TCs in recent decades has been reported in the North Atlantic7,14,15 and western North Pacific (WNP)16. However, Chan et al.17,18 pointed out the large uncertainties in observed TC decay, i.e., the trends in the TC decay rate are highly sensitive to the study area, research period, dataset, and method used. As for the possible impact factors of the changes in TC decay, Li and Chakraborty7 attributed the slowdown of TC decay to the higher moisture stocked in hurricanes due to the warmer sea surface temperatures (SST). In contrast, Chan et al.17 proposed that, rather than the SSTs, it is the effective area of moisture supply determined by the track upon landfall, which is more important in affecting TC decay. Song et al.16 found that the changes in TC decay timescale in the WNP were closely correlated to the soil moisture, low-level vorticity, and high-level divergence.

TC-induced heavy precipitation often triggers destructive flooding in coastal regions2,19,20,21. Previous studies have reported that changes in TC activity, such as the intensification of maximum wind, increasing TC rainfall rate22, and slower movement, can increase TC-induced flood risk in coastal regions3,22,23,24. Comparatively, the relationship between TC-induced precipitation and the rate of TC decay has received less attention. As a slower decay is closely related to the higher moisture carried by TCs, which eventually falls as precipitation, we expect that TC-induced precipitation should increase with a slower TC decay25,26. Lu et al.27 found that the slower a TC decays, the larger the number of rainstorm stations impacted over the Guangxi province of China during 1981–2001. However, observational evidence on the impacts of changes in the decay of landfalling TCs on precipitation is still limited.

In this study, we first examine the long-term trend and temporal evolution (i.e., differences between subperiods) of decay timescale based on 141 landfalling TCs over China during 1967–2018. Here, the TC best track data since the satellite era (i.e., 1967 onwards) is used to avoid data inhomogeneity28,29. The relationship between changing TC decay and possible impact factors such as SST and TC characteristics (e.g., intensity, translation speed, longitude, and latitude of landfall) are evaluated. The impact of changes in the rate of TC decay on TC-induced precipitation is examined based on 3-hourly precipitation data during 1979–2018. We find that the decay rate of landfalling TCs in China has significantly slowed during 1967–2018. The TC decay rate is slower over eastern mainland China than over southern mainland China, which together with the eastward shift of TC landfall locations leads to the slowdown of TC decay over China. Moreover, the slowdown of TC decay increases the probability of precipitation with higher intensities and larger total amounts, and thus leads to greater flood risk in China’s coastal regions.

Results

Temporal changes in decay timescale of landfalling TCs

A total of 148 landfalling TC events during 1967–2018 over China were selected based on the selection criteria (see “Methods” section). The decay timescale τ, a parameter used to characterize the decay rate of TC intensity (i.e., greater τ values indicate a slower decay after landfall), was calculated for each landfalling TC event7,18. To prevent the influence of outliers, 7 landfalling TC events with exceptionally large τ values (greater than twice the standard deviation above the mean) were excluded, and finally 141 landfalling TCs were used to analyze the changes in decay timescale in this study. The tracks and τ values of 141 landfalling TCs are shown in Fig. 1a, b, respectively, for the two half periods (1967–1992 and 1993–2018). Spatially, TCs with greater τ value are more likely to appear in eastern mainland China (EC) where, for example, the number of TCs with τ value larger than 40 h is 16, which is three times the number (i.e., 5) in southern mainland China (SC; Fig. 1b). The probability density functions (PDFs) of τ during 1967–1992 and 1993–2018 show that the PDF of τ in the second period (i.e., 1993–2018) is significantly shifted toward higher values compared to the first period (i.e., 1967–1992), indicating a greater probability for τ reaching higher values in 1993–2018 than in 1967–1992 (Fig. 1c).

Fig. 1: Features of the decay of landfalling TCs over China during 1967–2018.
figure 1

Panel a shows the tracks of 141 landfalling TCs originating from the WNP over 1967–1992 (blue) and 1993–2018 (red), and a dotted box (TC pre-landfall region) bounded by 15°N, 30°N, 110°E, and 130°E. Panel b shows the spatial distribution of the 141 landfall events over China. Each circle represents the TC center 24 h after landfall over 1967–1992 (blue) or 1993–2018 (red), with its size proportional to the decay timescale τ of the event. Panel c shows the histograms and probability density functions (PDFs) of τ during the two periods of 1967–1992 (blue) and 1993–2018 (red), respectively. The difference between the two PDFs in Panel c is tested by Kolmogorov-Smirnov (K-S) test (see the p value). The error bars in histograms correspond to ±1 standard deviation in each bin, which are calculated based on bootstrap sampling method (randomly resampled with replacement in each time series). Panel d shows the time series of τ during 1967–2018. In panel d, the moving black line shows temporal change in τ; the straight black line is the corresponding linear regression line derived from ordinary least squares and the dotted black line is the 95% confidence band about the regression line. The error bars correspond to ±1 s.e.m (standard error of mean) of all events in the 5-year window centered on a given year (because the double 3-year smoothing was applied).

We average the τ value of landfalling events for each year and apply a double 3-year smoothing7 to this time series to investigate the variations of τ at an interannual scale. The decay timescale τ shows a significant increasing trend of 1.6 h∙decade−1 (p < 0.01), and has increased by 45% from 1967 to 2018 (Fig. 1d). When employing no smoothing (using the original time series) or including the outliers, the increasing trend in the decay timescale τ is still significant with a slope of 1.9 h∙decade−1 (p < 0.1), and 2.3 h∙decade−1 (p < 0.01), respectively (Supplementary fig. 1). The significant increase in τ values indicates the slowdown of landfalling TC decay during 1967–2018 over China. The TC intensity 24 h after landfall was 34% of the intensity upon landfall in 1967, and increased to 47% in 2018. These results indicate that, compared to 50 years ago, TCs at present are more able to maintain their intensity after landfall, and thus have stronger and longer-lasting destructive potential on human society and the environment.

Possible factors affecting the decay timescale of landfalling TCs over China

As mentioned in the Introduction, SST has been reported as an essential factor influencing the decay timescale of landfalling TCs30,31,32. Besides SSTs, TC track characteristics, such as landfall location, intensity, and translation speed, are also to be potential factors influencing TC decay. First, TC tracks are changing systematically under global warming. This change in tracks has been noted in, for example, the poleward shift in the latitude of TC tracks33,34,35,36, and different motion characteristics (such as translation speed) in different regions37,38. Second, τ is calculated based on the TC intensity past the point of landfall, and TCs landfalling with different intensities may show differences in rate of decay39. Third, lower translation speed of TCs over land may allow moisture supply from the ocean for a longer time, which could enhance the storm moisture and thus promote slower decay of landfalling TCs7. We therefore investigate the relationship between SST/TC track characteristics and the decay timescale of landfalling TCs over China.

SST

We compute the seasonal-mean (i.e., June–September, the season with most of TC occurrences) SST during 1967–2018 over three different oceanic regions, i.e., the TC pre-landfall region, the development region, and the genesis region (Fig. 1a and Supplementary fig. 2a, c, e). Then, we smooth the SST time series in the same way as the τ time series, and superpose them and the τ time series. The SSTs in the three regions show consistent inter-annual fluctuation features with τ, and are significantly (p < 0.01) correlated with τ (Fig. 2a and Supplementary fig. 2b, d, f). The correlations are 0.38, 0.33, and 0.37 for TC pre-landfall, TC development, and TC genesis regions, respectively (Supplementary fig. 2). In TC pre-landfall region, the SSTs show a significant (p < 0.01) upward trend at a rate of 0.13 K∙decade−1 during 1967–2018 (Supplementary fig. 3a), and the correlation is the highest with 0.38 among the three regions. This suggests that increasing SSTs over adjacent oceanic areas play an important role in TC decay after landfall over mainland China. These results are consistent with previous studies7, who found significant correlation between τ and SST over the TC pre-landfall, development, and genesis regions of the North Atlantic. They also proved the important role of moisture in modulating TC decay by means of idealized simulation. Our analysis of the connection between SST, moisture, and TC decay shows significant correlation between SST, integrated water vapor, and τ, which further supports the impacts of warmer SST and higher moisture on slower TC decay based on long-term observations (Supplementary fig. 4).

Fig. 2: Association between various factors and the decay timescale τ during 1967–2018.
figure 2

Panels af show the relationship between τ and SST (units: K), TC landfall longitude (units: °E), TC landfall latitude (units: °N), landfall intensity (units: m∙s−1), average translation speed over the first four inland positions (\({{\rm{v}}}_{{\rm{t}}}\); units: m∙s−1), and coastline-perpendicular translation speed (\({{\rm{v}}}_{{\rm{t}}}\sin {\rm{\alpha }}\) shown in Supplementary fig. 5; units: m∙s−1), respectively. The correlation (i.e., r and p values) is estimated using the Spearman method. The thick black line is the corresponding linear regression line derived from ordinary least squares, the dotted black line indicates the 95% confidence band about the regression line, and the error bars correspond to ±1 s.e.m (standard error of mean) of all events in the 5-year window centered on a given year (because the double 3-year smoothing was applied). In panel a, SST is the seasonal-mean (i.e., June–September) values over the pre-landfall region shown in Fig. 1a.

Landfalling location

We examine the changes in landfall centroid location (i.e., longitude and latitude) of TCs and their correlations with τ. We find a significant increasing trend in the longitude centroid during 1967–2018 (0.23 degrees∙decade−1, p < 0.05; Supplementary fig. 3b) and a significant positive correlation between τ and the longitude centroid (r = 0.34, p < 0.05; Fig. 2b), i.e., the larger the longitude centroid, the greater the τ. By contrast, the long-term trend in the latitude centroid, and its correlation with τ are not statistically significant (Fig. 2c and supplementary fig. 3c). Therefore, the slowdown of the TC decay rate in China could related to the eastward shift of in TC landfall location. To further investigate the impact of longitude on the decay timescale of landfalling TCs, we divide the coastal region of China affected by landfalling TCs into EC and SC (Fig. 1b), over which the average centroid longitudes of TCs are 117.8°E and 111.8°E, respectively. As shown in Table 1, the mean value of τ is significantly greater in EC than in SC during 1967–2018, 1967–1992, and 1993–2018 (Fig. 1b and Table 1). Compared to the first half period (i.e., 1967–1992), both the frequency and proportion of TCs making landfall over EC during the second half period (i.e., 1993–2018) have increased. The increases in relative frequency of landfalling TCs in EC (which decay at a relatively slower rate) could result in the slowdown of TC decay rate in China.

Table 1 Comparison of decay timescale τ for landfalling TCs over SC and EC

Landfall intensity

We note that there is a significant increasing trend in the intensity of landfalling TCs during 1967–2018 over China (0.40 m∙s−1∙decade−1, p < 0.05; Supplementary fig. 3d), and a weakly significant positive correlation between τ and landfall intensity (r = 0.27, p = 0.06; Fig. 2d). This result indicates that τ increases with the enhancement of TC landfall intensity, which is consistent with previous studies40,41 who indicated that TC decay rate is proportional to its landfalling intensity. However, another some studies42 found a weak relationship between TC decay rate and the initial intensity of landfall. Therefore, in the future, whether TC decay rate is significantly affected by its landfall intensity should be future investigated (such as using numerical modeling) with giving deep physical mechanisms.

Translation speed

We examine the relationship between the average translation speed over the first four inland positions (\({v}_{t}\) hereafter)/the vertical component of its coastline (i.e., coastline-perpendicular translation speed, \({v}_{t}\sin \alpha\) hereafter; Supplementary fig. 4) and τ of landfalling TCs over China, and find no significant relation between them (Fig. 2e, f). The trend analysis shows that neither the changes in \({v}_{t}\) or \({v}_{t}\sin \alpha\) are statistically significant during 1967–2018 (Supplementary fig. 3e, f). These results indicate that the TC translation speed (\({v}_{t}\) and \({v}_{t}\sin \alpha\)) likely plays a minor role in the observed increase in τ over China (\({v}_{t}\) and \({v}_{t}\sin \alpha\)).

Overall, the mean decay timescale τ over China’s coast has increased from 29.11 h during 1967–1992 to 30.76 h during 1993–2018. The relative contributions of warming SSTs, eastward landfall location of TCs, and other factors (including the increasing landfall intensity) to this increasing τ are estimated as 60.6%, and 37.0%, and 2.4% (see “Methods” section for estimating these contributions), respectively.

Reasons for the difference in the decay timescale between EC and SC

As shown in Fig. 2b and Table 1, we find a faster slowdown in TC decay after landfall in EC than in SC. We thus explore the causes of this difference in the decay timescale between EC and SC. As discussed above, the landing position of TCs, TC landfall intensity, and translation speed could affect the variation in τ. Therefore, we first investigate the difference of TC track characteristics between EC and SC.

Variation in the genesis position of a TC can affect the TC track at sea43 and its maximum intensity positions44, which can potentially lead to changes in the location of landfalling TCs. We examine the variation in the genesis position (i.e., the first position with TC intensity >10.8 m∙s−1, including its longitude and latitude) and its correlation with landfall position (i.e., landfall longitude and latitude) for the selected 141 TCs. There are significant (p < 0.05) increasing trends in genesis longitude (0.60 degrees∙decade−1; Supplementary fig. 6a), and an insignificant decreasing trend in genesis latitude (−0.14 degrees∙decade−1; Supplementary fig. 6b), indicating an eastward shift of TC genesis position. TCs move northwestward due to the beta effect (i.e., a meridional vorticity gradient due to Coriolis effect)34, thus the more easterly a TC genesis position is, the more northerly it moves and makes landfall in EC. The correlation analysis also shows that the landfall longitude and latitude of TCs are significantly (p < 0.01) correlated with the TC genesis longitude, with correlations of 0.40 and 0.43 (Supplementary fig. 6a, b), respectively, while the correlations between landfall position and genesis latitude (Supplementary fig. 6c, d) are not significant. This further demonstrates that the more easterly the TC genesis longitude is, the more easterly and northerly its landfall location will be.

The displacement of the TC genesis location can lead to changes in the TC tracks and affect the duration of TCs at sea45. Comparing the characteristics of TCs making landfall in EC and SC, we observe that the mean genesis location of TCs making landfall in EC (mean genesis location: 138.5°E, 15.9°N) is more northeasterly (p < 0.01) than that of TCs making landfall in SC (mean genesis location: 131.5°E, 14.1°N; Fig. 3a, b). Meanwhile, the mean duration over sea of TCs landing in EC (i.e., 181.6 h) is longer than in SC (i.e., 170.9 h; Fig. 3c), although this difference is insignificant. This indicates that the eastward genesis location of TCs could make the TC path more tortuous and lead to an increase in the duration of the TCs at sea. TCs that lasts longer at sea may carry more moisture. Our analysis of the 500-hPa specific humidity in EC and SC within the 6 h before and after the TC landfalls shows that the specific humidity during TC landfalls in EC is significantly higher than that in SC (Fig. 3f). This indicates that the increased duration of TCs at sea potentially increases their moisture content. The more moisture carried by the TC upon landfall, the slower the subsequent rate of decay7. Therefore, the higher moisture content carried by TCs landing in EC also contributes to their slower decay. However, we find that coastline-perpendicular translation speed (\({v}_{t}\sin \alpha\) hereafter) and landfall intensity do not explain the greater τ value in EC than in SC. The \({v}_{t}\sin \alpha\) of TCs after landfall is significantly (p < 0.01) higher in EC than in SC (Fig. 3d), which is not favorable for TCs bringing more moisture when landing in EC. There is no significant difference in TC landfall intensity between EC (mean values: 27 m∙s−1) and SC (mean values: 26.9 m∙s−1), suggesting that the difference in the decay of landfalling TCs between EC and SC is not related to the TC landfall intensity (Fig. 3e).

Fig. 3: Differences in track characteristics of the 141 TCs making landfall in EC and SC during 1967–2018.
figure 3

Panels af show the boxplots of TC genesis longitude (units: °E), TC genesis latitude (units: °N), over-sea TC duration (units: hours), coastline-perpendicular translation speed (\({{\rm{v}}}_{{\rm{t}}}\) sinα; units: m∙s−1), TC landfall intensity (units: m∙s−1), and mean 500-hPa specific humidity (units: kg∙kg−1) during the 6 h before and after landfall in EC and SC, respectively. The black dot represents the mean values of the corresponding categories. Student’s t-test is used to test the difference (see p value) between EC and SC.

To investigate the dynamic mechanisms associated with the difference in decay of landfalling TCs over EC and SC, we examine the environmental variables during June-September of the 10 years with the largest longitude (i.e., 1977, 1988, 1989, 1990, 2005, 2006, 2007, 2013, 2014, and 2015) and the 10 years with the smallest longitude (i.e., 1967, 1968, 1969, 1970, 1971, 1993, 1994, 2002, 2003, and 2018) of the TC landing center during 1967–2018. We find that favorable environmental conditions (including dynamic and thermodynamic conditions) play a key role in slowing down the decay of landfalling TCs. In EC, there is positive anomalous low-level relative vorticity (Fig. 4a) and anomalous counterclockwise rotating airflow (Fig. 4d), which are conducive to maintaining a strong counterclockwise rotating cyclone after TC landfall. Meanwhile, the negative anomalous mid-level vertical velocity (increase in upward motion; Fig. 4b) facilitates the maintenance of the central updraft after TC landfall. The abnormally weak vertical wind shear (Fig. 4d) also favors the maintenance of the TC warm structure.

Fig. 4: Composite differences of environmental variables during June-September between the 10 years with the largest longitude and the 10 years with the smallest longitude of the TC landing center during 1967–2018.
figure 4

Panels ah are for 850-hPa relative vorticity (units: 10−6∙s−1), 500-hPa vertical wind (units: Pa∙s−1), 600-hPa relative humidity (units: %), vertical wind shear (units: m∙s−1) and steering flow (units: m∙s−1), sea surface temperature (units: K), water vapor flux (units: kg∙m−1∙s−1), genesis potential index, and soil moisture (units: m3∙m−3), respectively. In panel d, colored shadings indicate vertical wind shear, and black vectors indicate steering flow. In panel f, colored shadings indicate integrated water vapor, and black vectors indicate vapor flux. In panels ah, the areas with black/white dots indicate significant anomalies at the 0.1 significance level, as tested by the Student’s t test, and the red border indicates the EC and SC areas in China.

Abnormally warmer SSTs in the WNP (Fig. 4e) provide greater energy and moisture for TC formation and development. The relatively high mid-level relative humidity (Fig. 4c) and the significant increase of integrated water vapor (Fig. 4f) promote the release of latent heat from water vapor condensation, thus maintaining the warm core structure of the TCs. Additionally, the anomalously higher soil moisture (Fig. 4h) in EC is conducive to higher thermal conductivity and thus longer-lasting latent heat fluxes, which are conducive to the maintenance of TCs after landfall. However, the favorable environmental conditions mentioned above are not available in SC, which may explain the faster decay in SC than in EC.

During the 10 years with the largest landfall longitude of TCs, the dynamic (positive anomalous low-level relative vorticity, negative anomalous mid-level vertical wind, and weaker vertical wind shear; Fig. 4a, b, d), and thermodynamic (including high 600-hPa relative humidity, increasing SST and integrated vapor transport; Fig. 4c, e, f) conditions favorable to TC formation and development extend beyond China to the eastern WNP. We surmise these favorable conditions in the eastern WNP enabled more TC genesis and growth in this area. To assess this hypothesis, we calculate the Genesis potential Index (GPI, see “Methods” section) over WNP, and find positive anomalies in the GPI in most of the WNP. Large values of GPI are identified in areas with a longitude range of 120°E–150°E and latitude range of 8°N–17°N. The mean genesis longitude and latitude of the 141 landfalling TCs analyzed in this study is 135.0°E, and 15.0°N, respectively. Comparatively, the TC genesis potential during the 10 years with the largest landfall longitude is more easterly and northerly. The eastward shift of TC genesis location enables the TCs to last longer for β-drift (i.e., the process of TCs moving poleward and westward) over the WNP34, resulting in the northwestward movement of TCs. In addition, TCs preferentially land in EC due to the northward steering flow over the range of 118°E-130°E and 18°N-30°N (Fig. 4d). These analyses also explain the dynamical mechanisms behind the more easterly genesis position of the TCs making landfall in China.

Impact of slower decay of landfalling TCs on precipitation

The slowing down of the decay of landfalling TCs implies that the affected areas face longer-lasting and stronger storms, which may bring about heavier rainfall. To examine the impact of the slower decay of landfalling TCs on precipitation, we first calculate the change rate of precipitation over the 24- and 48-hour periods after TC landfall. It should be noted that, limited by the length of 3-hour precipitation data, we only analyzed the 104 landfalling TCs and associated precipitation during 1979–2018 in this section. TC-induced precipitation is identified at every 3-hourly timestep (see Methods), and for each timestep, the mean and maximum precipitation amount of all grid points affected by TCs are defined as Pmean and Pmax, respectively. On this basis, Pmean24, Pmax24, Pmean48, and Pmax48 are the Pmean and Pmax values within 24- and 48-h after TC landfall, respectively (Supplementary fig. 8). Then the change rate of Pmean24, Pmax24, Pmean48, and Pmax48 are calculated as the ratio of the slope of linear regression to the mean value (i.e., slope of linear regression/mean value x 100%; see Supplementary fig. 8r for an example). The change rate can be positive and negative: a positive change rate indicates an increase in precipitation intensity over time, and vice versa. Then, we analyze how the change rate of Pmean24, Pmax24, Pmean48, and Pmax48 vary with different τ ranges (i.e., 0-10th, 10-20th, 20-30th, …, 90-100th).

The change rates of Pmean24, Pmax24, Pmean48, and Pmax48 all increase with increasing τ-quartile intervals (Fig. 5a–d). Only the change rates of Pmean48 and Pmax48 show a significant (p < 0.01) positive correlation with τ-quartile intervals, with correlation coefficients of 0.30 and 0.36, respectively (Fig. 5c, d). This indicates that the slower-decaying TCs are more likely to slow down the decay/growth of precipitation intensity within the 48 h period after landfall.

Fig. 5: Effect of variation in τ on the rate of change of TC-induced precipitation during 1979–2018.
figure 5

Panels ad show the relationship between the change rate in Pmean24, Pmax24, Pmean48, and Pmax48 and associated percentiles of τ. Panels eh show PDFs of the change rate in Pmean24, Pmax24, Pmean48, and Pmax48 associated with slower- (>70th percentile of τ; red) and faster-decaying (<30th percentile of τ; blue) landfalling TCs, respectively. In panels ad, the black dot indicates the mean values of the corresponding boxplot, the red dashed line is the linear regression line obtained by ordinary least squares, and the correlations are estimated using the Spearman method. The K-S test is used to test the difference between the two PDFs in panels eh (see the inset p values).

To further validate the impact of TC decay on the precipitation change rate, we compare the change rate of precipitation over 24- and 48-hours after the landfall of slower- (i.e., >70th percentile of τ) and faster-decaying (i.e., <30th percentile of τ) TCs. Figure 5e–h shows the PDFs of change rate of Pmean24, Pmax24, Pmean48, and Pmax48 of slower- and faster-decay TCs, respectively. The probability of the TC-induced precipitation change rate of the slower-decaying TCs being positive (i.e., increase) is significantly (p < 0.01) greater than that of the faster-decaying TCs, especially for precipitation occurring within 48 h of TC landfall (Fig. 5g, h). This further evidence the finding that the slower-decaying TCs are more likely to maintain or increase their precipitation intensity compared to the faster-decaying TCs, with a more pronounced enhancement of precipitation over 48 h compared to 24 h.

In addition to the change rates in TC-induced precipitation, the impact of the TC decay timescale on the total precipitation amount over the 24- and 48-h are analyzed. Similarly, we compare the PDFs of grid-averaged total amount of precipitation (PTmean hereafter) and the maximum total amount of precipitation (PTmax hereafter) in all grids within 24- and 48-hours, respectively. Compared to faster-decaying landfalling TCs, there is a significant rightward shift in the PDFs of PTmean24, PTmax24, PTmean48, and PTmax48 associated with slower-decaying TCs (Supplementary fig. 9), demonstrating the higher probability of slower-decaying TCs to produce higher total precipitation amounts.

Furthermore, we examine the time-varying characteristics in PTmean24, PTmax24, PTmean48, and PTmax48 for the 104 landfalling TCs during 1979–2018. We find that PTmean24, PTmax24, PTmean48, and PTmax48 all show significant (p < 0.01 or p < 0.05) increasing trends (Fig. 6a, c, e, g), with trends of 2.0 mm∙decade−1, 23.0 mm∙decade−1, 1.9 mm∙decade−1, and 24.2 mm∙decade−1, respectively. We presume that these increasing trends in the total TC-induced precipitation amount may be related to the slower decay of landfalling TCs. Thus, the correlations between these precipitation indicators (PTmean24, PTmax24, PTmean48, and PTmax48) and τ for the 104 landfalling TCs are tested, and all of them show a significant (p < 0.01) positive correlation, with correlations of 0.64, 0.76, 0.67, and 0.70, respectively (Fig. 6b, d, f, h). The slower the decay of landfalling TCs, the higher the mean and maximum total precipitation amount within 24 and 48 h.

Fig. 6: Effect of variation in τ on TC-induced precipitation during 1979–2018.
figure 6

Panels a, b (c, d) show the time series of PTmean24 (PTmax24) and the relation between PTmean24 (PTmax24) and τ, respectively. Panels eh are the same as Panels ad, but for PTmean48 and PTmax48, respectively. In panels a, c, e, and g, moving black line shows temporal change in TC-induced precipitation. In panels b, d, f, and h, the correlations (see r and p values) are estimated using the Spearman method. In panels ah, the diagonal black line is the corresponding linear regression line derived from ordinary least squares, the dashed black line is the 95% confidence band about the regression line, and the error bars correspond to ±1 s.e.m (standard error of mean) of all events in the 5-year window centered on a given year (because the double 3-year smoothing was applied).

The above analysis is for the total TC-induced precipitation amount in unrestricted areas, which consists of all grid points with precipitation occurring in at least one timestep within the 24 (48) h after TC landfall. The same analysis is conducted for the total precipitation amount within restricted areas, which only includes grid points with precipitation occurring in all timesteps within the 24 h after TC landfall. Based on the precipitation in the restricted areas, the grid-averaged total precipitation amount, and the maximum total precipitation amount within the 24 h after TC landfall are defined as re-PTmean24, and re-PTmax24, respectively (see “Methods” section and Supplementary fig. 8). Here we also find significant increases in re-PTmean24 and re-PTmax24 of 13.1 mm∙decade−1 and 16.7 mm∙decade−1 during 1979–2018, respectively (Supplementary fig. 10a, c). Meanwhile, both re-PTmean24 and re-PTmax24 show significant (p < 0.01) and strong positive correlations with τ (r of 0.69 and 0.82, respectively; Supplementary fig. 10b, d). Overall, the result obtained from the restricted area is consistent with the result from the unrestricted area.

Discussion

In this study, we find a significant slowdown in the decay rate of landfalling TCs over the coast of China during 1967–2018. The existence of long-term changes in the decay rate of landfalling TCs in recent decades and its causes are still an on-going debate. The trends in the decay of landfalling TCs reported in previous studies vary by regions, study periods, datasets, TC selection criteria, and other factors7,15,16,17,18. TCs display large regional variation in characteristics due to the different environments in which they are embedded. For instance, the translation speed of TCs is relatively faster in the midlatitudes than in the lower latitudes due to strong westerly winds from the midlatitude jet37,38. TCs in different basins and making landfall by different modes (e.g., hard-strike and lingering modes) also show discrepancies in decay timescale17,18. Additionally, the decay of TC intensity is controlled by a complex interaction between internal and external factors46,47, and thus shows significant individual differences. Since the sample size of landfalling TCs is small, the statistical results are very sensitive to the data processing procedures (e.g., selection criteria). In our study, the threshold of minimum landfall intensity was set at 24.5 m∙s−1. By doing this, the number of selected landfalling TCs is 141, with 2.7 landfalling TCs per year on average. The larger sample size reduces the sensitivity of our results to individual landfalling TC events. Furthermore, the slowdown in the rate of decay of landfalling TCs over China is significant no matter whether we employ the double 3-year smoothing method or not. The significant differences between the PDFs of TC decay during 1967–1992 and 1993–2018 further confirm the higher probability of slower-decaying TCs in the second period (1993–2018). This evidence supports the robustness of our finding that the intensity of TCs after landfalling in China has decayed more slowly in recent decades.

We notice that there is uncertainty in the TC intensity of the best track data due to the changes in monitoring techniques28,48,49. For example, the maximum intensity of TCs was likely to be underestimated because of the inability to adequately monitor the intensities of TCs over the open ocean28,50, and considerable number of weak and short-lived TCs might be missed due to insufficient observational networks29,51. In our study, a series of criteria are applied to select the TCs that used to analyze the decay rate. First, only landfalling TCs are selected, and the decay rate of TC intensity is calculated for the 24 h after the landfalls. Although that the TCs over the open ocean might be missed or underestimated before or at the early stage of satellite era, the intensity estimates of landfalling TCs that impacted coast areas where people lived are more reliable. Second, only TCs landing with intensity no less than 24.5 m∙s−1 and survived more than 24 h after landfalls are selected to analyze the decay rate. In that case, those weak or short-lived TCs were excluded. By applying these two measures, the impact of the uncertainty in TCs over the open ocean, weak and short-lived TCs on our analysis could be substantially reduced. Even though, we suggest more studies to further validate our findings by using both reanalysis datasets and updated tropical cyclone simulation models.

The slower decay of landfalling TCs over China is resulted by multiple factors including the increase in SSTs, the eastward shift of TC landfall locations, and other factors such as enhanced landfall intensity. We find significant increasing trends in SSTs at the point of TC genesis, development, and in pre-landfall regions, with their fluctuations significantly correlated with the TC decay rate. Besides, the TC decay rate is mainly determined by the moisture stocked by TCs and the dynamical and thermodynamic conditions. Here, we find that TCs landing in EC decay slower than in SC, because 1) the easterly TC genesis and landfall locations prolong the duration of the TCs over sea, which results in more moisture carried by the TCs and thus a slower decay of the landfalling TCs, and 2) the dynamical and thermodynamic conditions over EC are conducive to maintaining the TCs after landfall. In that case, the eastward shift of TC genesis and landfall locations causes more TCs making landfall in EC with slower decay rate, which contributes to the slower decay of TCs over China.

SSTs are one of the most important factors affecting the formation, development, and decay of TCs8,52,53. For example, changes in SST can modulate TC frequency and distribution of TC tracks12,54,55. In addition, the distribution of SST can alter the thermodynamic environment and thus affect TC intensity56,57. In this study, we reveal the impact and contribution of changes in SSTs, shift of TC landfall location, and other factors such as TC intensity to the slower decay of landfalling TCs. But the interactions between SST, TC landfall location, and intensity, and their impact on the decay of TCs require further investigation.

The slowdown of the TC decay is not monotonous, but shows large inter-annual variations (e.g., the τ reached a peak in the 1990s). This implies that the changes in TC decay are affected by natural climate variability. The factors such as El Niño Southern Oscillation (ENSO), Pacific Decadal Oscillation (PDO), and the monsoon system can modulate the large-scale circulations, leading to changes in ocean and atmosphere conditions, and thus influencing the inter-annual variations of TC activity over the western North Pacific58,59,60. For example, in El Niño years and strong monsoon trough years with monsoon trough extending eastward, there are more TCs generated in east and TCs tend to be more intense and longer-lived due to the favorable dynamic and thermodynamic environment, which is consistent with our findings that TCs generated in east would stay longer over ocean, and decay slower after landfall.

As for the long-term trends, even though previous work has attributed the changes in TC characteristics such as the intensification of maximum intensity61,62, poleward shift of TC tracks34,63, and the slower translation speed to anthropogenic climate change3,37,38, our understanding of the impact of anthropogenic climate change on TC characteristics is still limited, with low confidence in the detected anthropogenic signal. This is because TCs are complex weather systems with high variability. The observations of TCs suffer from issues of data homogeneity28, while the climate models still need to be improved to extract TC activity accurately and efficiently64. Therefore, the question of whether the slowdown in the decay of landfalling TCs over China detected in this study is caused by anthropogenic or natural climate change is not discussed in this study, and requires separate investigation.

We analyze the effect of the slower decay of landfalling TCs on TC-induced precipitation, and find that slowly-decaying TCs are more likely to decelerate the decay of precipitation intensity and induce higher total precipitation amounts within 24 and 48 h of landfall. This result can also be inferred from the longer TC translation duration and the enhanced rainfall rate of TCs9,22. The slowdown of TC translation speed could enhance the impact of slower decay rate on precipitation. As a TC with a slower decay rate can maintain its intensity for a longer duration, this longer-lasting and stronger TC will be stalled over a region for longer duration if its translation speed is slower, which thus causes the higher TC-induced total precipitation amounts at this region3. Additionally, the increasing rainfall rate of TCs, which is more pronounce in the WNP, and for strong TCs (i.e., categories 4 and 5 according to the Saffir-Simpson scale), has been evidenced in previous studies based on model simulations and observations9,22. Guzman and Jiang22 have linked the enhancing TC rainfall rate to the increases in SSTs and total precipitable water in the TC environment, which is consistent with our analysis: the greater the moisture carried by TCs (thus the slower the decay), the more precipitation they produce. Our results suggest that the slowdown of TC decay should be taken into account when assessing socio-economic impacts of landfalling TCs, since it can increase associated hazards such as flooding.

Methods

TC best-track data

The TC best track dataset is provided by the International Best Track Archive for Climate Stewardship (IBTrACS) version 4.065. Multiple TC datasets from different institutions, such as the China Meteorological Administration (CMA), Joint Typhoon Warning Center (JTWC) and Japan Meteorological Agency (JMA), are merged by IBTrACS into one global TC database via collation and correction. Currently, the IBTrACS dataset is widely used in TC-related research24,37,38,66,67. In this study, we adopt 2-minute maximum sustained wind (TC intensity; CMA data merged in IBTrACS), and the latitude and longitude (IBTrACS) of the TC center at 3-h intervals.

When analyzing the decay of landfalling TCs, we use the TC dataset with record interval of 6 h during 1967–2018, which belongs to the post-satellite era and tend to be homogeneous and reliable28,29,68,69,70. In this period, we study all the landfall events that meet four criteria7: 1) TC intensity at the last position before landfall need to be no less than 24.5 m∙s−1 (the minimum intensity for ‘severe tropical storm’ according to CMA); 2) TCs survived more than 24 h after landfall, meaning that there are at least 4 consecutive 6-hour inland position records after landfall; 3) TC intensity cannot increase past landfall; 4) TCs have not undergone an extratropical transition (i.e., evolving into extratropical storm systems) during the first day after landfall. When analyzing the TC-induced precipitation, the TC dataset with a record interval of 3 h during 1979–2018 corresponding to the 3-hourly precipitation data is used.

Gridded precipitation data

The gridded precipitation data with a time resolution of 3 h and a spatial resolution of 0.1°×0.1° during 1979–2018 is derived from the China Meteorological Forcing dataset (CMFD), National Tibetan Plateau Data Center. This dataset, made through fusion of remote sensing products, reanalysis datasets and in-situ station data71,72, is one of the most widely used climate datasets in China owing to its successive period and stable quality73,74,75,76,77,78.

Reanalysis data

Monthly and hourly large-scale environmental variables (such as sea surface temperature, wind field, and specific humidity) with a spatial resolution of \(0.25^\circ\) during 1967–2018 are collected from the European Centre for Medium-Range Weather Forecasts (ECMWF) fifth generation atmospheric reanalysis dataset (ERA5). In this study, vertical wind shear (VWS) is derived by calculating the difference between the wind field at 200 hPa and 850 hPa79, and the 500-hPa wind field is selected as the steering flow80,81,82.

Decay timescale of landfalling TCs

Most of the damage tends to occur within 24 h past TC landfall, therefore, for each TC, the intensity (\(I({\rm{t}})\)) within 24 h past landfall (\({\rm{t}}={\rm{t}}1,{\rm{t}}2,{\rm{t}}3,{\rm{t}}4\)) is computed. \(I({\rm{t}})\) shows an exponential decay as \(I({\rm{t}})=I(0){{\rm{e}}}^{-{\rm{t}}/{\rm{\tau }}}\)83,84, where t is the time after landfall and τ is the decay timescale. The larger the value of τ is, the slower the TC decays. Within the 24 h after landfall, \(I({\rm{t}})\) can be represented well by the exponential model with only one parameter τ7. However, this one parameter exponential model is not appropriate for \(I({\rm{t}})\) of period more than 24 h.

The time series of τ is computed by employing a double 3-year smoothing method to reduce random noise. The corresponding ±1 s.e.m (i.e., standard error of mean) of all events in the 5-year window centered on a given year is shown as an error bar, since we applied the double 3-year smoothing method to the time series. So, the s.e.m is computed as \(({\rm{standard\; deviation}}/\sqrt{N})\), where \(N\) is the number of all events in the 5-year window. The trend in τ is estimated using linear regression based on the ordinary least squares method. The original time series without any smoothing show consistent results (Supplementary fig. 1).

Translation speed

We compute the annual-mean TC translation speed (\({v}_{t}\)) and coastline-perpendicular translation speed (\({v}_{t}\sin \alpha\)), and investigate the relation between them and the decay of landfalling TCs. α is the angle between the local 200-km coastline and the incident direction vector (computed using the coordinates of the first two inland locations; Supplementary fig. 5). The translation speed is calculated using the distance between two neighboring TC centers divided by the record interval (i.e., 6 h). For each TC, we only consider the average translation speed over the first four inland positions. The component of translation speed perpendicular to the coastline is defined as coastline-perpendicular translation speed (\({v}_{t}\sin \alpha\)). An example of calculating the \({v}_{t}\sin \alpha\) of Typhoon Mangkhut (2018) is given in Supplementary fig. 5.

Genesis potential index

Genesis potential index (GPI) developed by Emanuel and Nolan83 is widely used to study the impact of environmental field changes on TC activity under global warming and it is expressed as follows:

$${\rm{GPI}}={\left|{10}^{5}{\rm{\eta }}\right|}^{\frac{3}{2}}{\times \left({\rm{RH}}/50\right)}^{3}{\times \left({{\rm{V}}}_{{\rm{pot}}}/70\right)}^{3}\times {\left(1+0.1{{\rm{V}}}_{{\rm{shear}}}\right)}^{-2}$$
(1)

where \({\rm{\eta }}\) is the 850-hPa absolute vorticity (units: \({10}^{-5}\,\cdot\, {s}^{-1}\)), RH is the 600-hPa relative humidity (units: %), \({{\rm{V}}}_{{\rm{pot}}}\) is the TC maximum potential intensity (units: \({\rm{m}}\,\cdot\, {s}^{-1}\)), and \({{\rm{V}}}_{{\rm{shear}}}\) is the magnitude of the wind shear between 850 hPa and 200 hPa (units: \({\rm{m}}\,\cdot\, {s}^{-1}\)). The formulation of \({{\rm{V}}}_{{\rm{pot}}}\) defined by Bister and Emanuel84 is calculated as follows:

$${{\rm{V}}}_{{\rm{pot}}}=\sqrt{\frac{{{\rm{C}}}_{{\rm{k}}}}{{{\rm{C}}}_{{\rm{d}}}}\frac{{{\rm{T}}}_{{\rm{s}}}}{{{\rm{T}}}_{{\rm{o}}}}\left({{\rm{CAPE}}}^{* }-{{\rm{CAPE}}}^{{\rm{b}}}\right)}$$
(2)

where \({{\rm{C}}}_{{\rm{k}}}\) is the exchange coefficient of enthalpy, \({{\rm{C}}}_{{\rm{d}}}\) is the drag coefficient, \({{\rm{T}}}_{{\rm{s}}}\) is the SST, \({{\rm{T}}}_{{\rm{o}}}\) is the outflow layer temperature, \({{\rm{CAPE}}}^{* }\) is the convective available potential energy of saturated air over the ocean surface, and \({{\rm{CAPE}}}^{{\rm{b}}}\) is that of boundary layer air.

Calculation of the relative contribution

The relative contributions of increasing SST, eastward shift of TC landfall locations and other factors (including enhancing TC landfall intensity) to the increase in decay timescale τ between two periods are calculated following Li and Chakraborty7. The landfall events are categorized into two periods of equal length (1967–1992 and 1993–2018) and two regions (EC and SC). During 1967–1992 (1993–2018), the mean τ for EC and SC are 33.29 (34.58) h and 25.53 (26.27) h, respectively, and the proportion of landfalling TCs for EC and SC are 46.15% (53.95%) and 53.85% (46.05%), respectively. Meanwhile, the mean τ in China during 1967–1992 (1993–2018) is 29.11 (30.76) hours, with an increase of 1.65 h in the latter period.

If the proportion of landfalling TCs remained unchanged during 1993–2018, the mean τ in China is 30.11 h (46.15% × 34.58 + 53.85% × 26.27 = 30.11), with the increase in SST leading to an increase in τ of 1 h (30.11 - 29.11 = 1 h), so the relative contribution of SST to τ is 60.6% (1/1.65 = 60.6%); If the SST remained unchanged during 1993–2018, the mean τ in China is 29.72 h (53.95% × 33.29 + 46.05% × 25.53 = 29.72 h), and the increment of τ due to the change in positions of landfalling TCs is 0.61 h (29.72-29.11 = 0.61 h), so the relative contribution of eastward landing to τ is 37.0% (0.61/1.65 = 37.0%). Thus, the relative contribution of other factors such as the enhanced TC landfall intensity to τ is 2.4% (100% - 60.6% - 37.0% = 2.4%).

Identification of TC-induced precipitation

A number of previous studies assumed TC-induced precipitation can be estimated as the precipitation within a fixed radius (e.g. 500 km) from the TC center at a given time windows (e.g. ±1 day)85,86,87,88. However, this approach is not entirely appropriate, as TC-induced precipitation is not uniformly distributed with the TC center but varies irregularly with time89. To solve these problems, Ren et al.89,90 proposed the objective synoptic analysis technique (OSAT) to isolate TC-induced precipitation from daily precipitation by imitating the process which weather forecasters use to manually analyze a synoptic map. The OSAT method is now widely used to identify TC-induced precipitation91,92,93,94,95. There are two key steps about this method, including dividing the daily precipitation into several individual rainbelts and distinguishing TC rainbelts based on the distance function between the TC center and the rainbands distribution.

In this study, the OSAT method is adopted to identify the TC-induced precipitation. For each landfalling TC during 1979–2018, we identify TC-induced precipitation in every 3-hourly timestep within 48 h after landfall, and define eight precipitation indices. For each 3-hourly timestep, the mean, and maximum precipitation amount of all grid points affected by TCs are calculated and defined as Pmean and Pmax, respectively. On this basis, Pmean24, Pmax24, Pmean48, and Pmax48 are Pmean and Pmax values within 24- and 48-h after TC landfalls, respectively. Then the change rate of Pmean24, Pmax24, Pmean48, and Pmax48 are calculated as the ratio of the slope of linear regression to the mean value (i.e., slope of linear regression/mean value x 100%). The change rate can be positive and negative. A positive change rate indicates the increasing precipitation over time, and vice versa. To examine the impact of TC decay timescale on the total precipitation amount, we calculate the cumulated precipitation amount of each grid within 24 and 48 h, then define the mean and maximum values as PTmean24, PTmax24, PTmean 48, and PTmax48, respectively. Taking the precipitation events induced by TC Ken (1989) as an example, we show the spatial distribution of TC Ken-induced precipitation at each 3-hourly moment within 48 h after landfall (Supplementary fig. 8a–q) and the change in precipitation intensity over time (Supplementary fig. 8r), respectively. The black-dotted areas in Supplementary fig. 8a–i indicate the restricted precipitation areas that consist of only grid points with precipitation occurring in all timesteps within 24 h after TC landfall. Based on the precipitation occurring within 24 h of TC landfall in the restricted area, the corresponding mean and maximum total precipitation amount (i.e., re-PTmean24 and re-PTmax24) are then calculated.