Tropical cyclones are one of the most devastating natural disasters in terms of their average destructiveness and annual frequency affecting coastal regions. There are ~100-year records of tropical cyclone activity in some regions (
1,
2), but only since 1982 have tropical cyclones been monitored globally by satellites. Poleward migration of the locations of tropical cyclone maximum intensity in the past 40 years has been reported in the global “best-track” data (
3) and is projected to continue regionally in the 21st century (
4). This poleward migration has been traced back to the poleward shift of tropical cyclone formation, which is speculated to be linked to an anthropogenically forced tropical circulation expansion (
5). These trends could change the coastal tropical cyclone risk in the future (
4). However, the poleward migration of the locations of tropical cyclone formation and peak intensity by themselves may not necessarily indicate a change in coastal tropical cyclone risk directly, because these locations are commonly too far away to generate an impact on the coasts. The changes in coastal tropical cyclone activity and landfall frequency are a central and perhaps ultimate concern.
A tropical cyclone landfall is conventionally defined as the intersection of the surface center of a tropical cyclone with a coastline (
6). To date, there has been no firm evidence of global trends of the frequency of tropical cyclones with maximum wind speed above the hurricane-force wind (64 knots) at landfall (
7). However, a “near miss” or “indirect-hit” tropical cyclone track can also cause damage; for example, Sandy in 2012 and Dorian in 2019 skirted along the U.S. coast for a considerable time before making landfall. Investigating the trend of tropical cyclones entering coastal regions is therefore essential to better understand tropical cyclone risk.
Here we consider tropical cyclones, defined by the lifetime maximum intensity (LMI) reaching at least 34 knots (see supplementary materials), during the period 1982–2018. This is the period for which we have the highest confidence in the quality of data and its completeness in global basins (
8). The poleward migration of tropical cyclone activity was first found at LMI (
3). The annual mean distance of the locations of LMI to the nearest land (
Fig. 1A) also shows an evident and statistically significant decreasing trend of –32 ± 31 km per decade (table S1). The fraction of annual tropical cyclones entering coastal regions, defined as the offshore area with a distance to the nearest land less than 200 km (
Fig. 1B), shows a robust increase of 2.2 ± 1.9% per decade (table S1). We also find positive trends of the annual mean lifetime fraction that tropical cyclones spend in coastal regions, at a rate of 2.1 ± 1.2% per decade globally, and in the Northern and Southern hemispheres of 2.1 ± 1.6% and 2.1 ± 1.9% per decade, respectively (table S1). The consistently positive contribution from both hemispheres (table S1) to the three global trends shown in
Fig. 1 suggests that the increase of tropical cyclone activity in coastal regions is a global phenomenon, although regional differences between basins are evident.
The trend of the frequency of coastal tropical cyclones increases very significantly with proximity to land by one cyclone per decade per 1000 km (
Fig. 2 and table S2). There are ~2 ± 2 additional tropical cyclones per decade in the 200-km coastal region globally (
Fig. 2). Consistent with this, the fraction of annual tropical cyclones entering coastal regions (fig. S1A) and the annual-mean lifetime fraction that tropical cyclones spend in coastal regions (fig. S1B) increase with proximity to land. The statistical significance remains by excluding short-lived storms (
9) and using an alternative set of tropical cyclone records (table S3). These clear increases in trend with reduced distance-to-land threshold suggest a global-scale landward migration of tropical cyclone activity.
Considering the geometry of global land-sea distribution, zonal and/or meridional shifts of tropical cyclone locations may lead to a change of tropical cyclone activity in coastal regions. Our analysis shows that from 1982–1999 to 2000–2018, the epochal mean tropical cyclone location in the global basins migrated not only poleward but also westward (
Table 1). The globally epochal poleward and westward shifts are both statistically significant. If we measure the track shift in terms of degrees, the zonal shift is considerably larger than the meridional change. Tropical cyclone activity is being shifted westward in the West Pacific, East Pacific, and North and South Indian Ocean. These four basins account for ~75% of global tropical cyclones for 1982–2018, which accounts for the global mean westward track shifts. There are eastward track shifts in the South Pacific and no significant zonal shifts in the North Atlantic. These basin-wise changes in tropical cyclone locations can also be seen in the density change of tropical cyclone locations between the two epochs (fig. S2).
We do not find any statistically significant global change in cyclone zonal translation velocity, duration, or zonal shifts of cyclone genesis (table S4). The lack of regionally significant zonal genesis location change is different from a reported poleward shift of genesis in the Pacific Ocean basins (
10). The relative changes in meridional cyclone translation velocity in the second epoch agree with the meridional track shifts in all the basins.
Tropical cyclone tracks are primarily determined by the environmental steering flow and genesis location (
11,
12). We find significant enhancement of westward steering at a global scale in all basins except the North Indian Ocean (
Table 1). Consistent general circulation patterns are confirmed in all three reanalysis products (fig. S3, A to C). The global mean change in zonal steering (+0.3 m s
–1;
Table 1) is about 11% of the global mean zonal steering speed (2.6 m s
–1 in ERA5 reanalysis; fig. S3) and global mean cyclone zonal translation speed (2.7 m s
–1). The West Pacific, East Pacific, and South Indian Ocean show both westward track shift and enhanced westward environmental steering between the epochs. The West Pacific is the only basin that shows a consistent link among seasonal westward steering enhancement, larger westward cyclone velocity, and westward track shift. This robust linkage to seasonal mean environmental conditions may be due to the considerably larger fraction of storm days in the season in the West Pacific relative to the other basins (
13).
An environment with reduced vertical wind shear and/or increased potential intensity is favorable for tropical cyclone development (
14,
15) and may also contribute to geographical shifts of tropical cyclone locations (
3). We find a significant relative decrease in vertical wind shear from the east to west globally (
Fig. 3), which is confirmed in individual basins with reanalysis products (fig. S4, A to F). The South Pacific is the only basin where neither steering flow nor vertical wind shear fully explains the regional eastward track shift. However, we do find a relative increase in potential intensity to the east and a decrease to the west in that basin (fig. S4L). In the North Atlantic, the shear weakens only in the central part of the basin (fig. S4C) with no zonal change in either steering (
Table 1) or potential intensity (fig. S4I), which is consistent with the absence of a significant zonal track shift. Further analysis reveals that the zonal potential intensity change is primarily modulated by the variation of the atmospheric convective available potential energy (fig. S5).
Given the dominant role of steering flow in tropical cyclone tracks (
12) and a larger epochal relative change of steering (
Table 1) than wind shear (
Fig. 3) in our analysis, it is very likely that the westward shift of cyclone tracks is mainly due to the enhanced westward steering. Anomalous westward steering may increase tropical cyclone frequency in the west of the basins, and relatively reduced vertical wind shear provides a further favorable environment for cyclones there. The quantitative contribution of the three environmental factors needs further exploration.
Climate modes such as the El Niño–Southern Oscillation (ENSO) and Pacific Decadal Oscillation (PDO), considered as measures of mainly internal unforced variability, can modulate tropical cyclone activities (
16–
19) and global atmospheric circulation in the tropics (
20,
21). We find that ENSO has a very limited impact on annual trends. However, the PDO may substantially contribute to the trends in the Northern Hemisphere (table S1). The epochal steering flow change is similar to the steering difference between PDO warm and cold phases (fig. S3, D to F). The PDO index had a phase change in 1998 (
5), which also separates the first and second epochs (pre- and post-2000) in our analysis into a warm and cold PDO phase, respectively.
Given no strong trend of global annual frequency of tropical cyclones (
22), an increasing trend of annual coastal cyclone fraction agrees with a rising frequency of coastal tropical cyclones shown here. The lifetime fraction trend in
Fig. 1C seems to agree with a recently reported global slowdown of tropical cyclone translation speed (
23) because the annual mean tropical cyclone duration has not changed significantly during the period 1982–2018 (table S4). However, this causal hypothesis needs to be confirmed, as changes in cyclone moving speed are disputed (
24).
We find statistically significant landward migration of tropical cyclone activity. The lack of a statistically significant trend of actual landfall (+1 ± 2 cyclones per decade globally) may be due to the rapid intensity reduction before and during landfall. Major tropical cyclones have decayed more rapidly after LMI (
25). This could stabilize the annual frequency of landfall defined by a fixed intensity threshold. Thus, the lack of a global trend of annual mean landfall frequency, as has been previously reported (
7) and confirmed in our analysis, may not be contradictory to the increase in tropical cyclone activity in coastal regions reported here.
Long-term zonal migration of tropical cyclones, as found in our analysis, has been previously shown directly or indirectly in some individual basins for different metrics. In the West Pacific, westward migration of the tropical cyclone maximum intensity locations was reported (
26), which was related to the strengthened tropical Pacific Walker circulation driven by the zonally enhanced sea surface temperature gradient. No detectable trend of U.S. landfall hurricane frequency has emerged (
22). There has been a decreasing trend of landfall frequency by severe tropical cyclones in eastern Australia (
2). In the North Indian Ocean, the reduction of local wind shear over the Arabian Sea has made tropical cyclone development more favorable, and this may continue in the future (
27).
The relatively limited length of global tropical cyclone observations, limitations of climate modeling, and possible association with the PDO limit our ability to attribute tropical cyclone migration toward coasts to anthropogenic forcing. However, we note that global epochal westward shifts are still present when considering only PDO-neutral years (table S5). The West Pacific LMI distance to land, fraction of coastal cyclones, and coastal time fraction trends since 1950 are still significant after the PDO is removed (table S1). This suggests that factors other than the PDO are also important. It is argued that greenhouse gas emissions have contributed to the observed changes in regional distribution of tropical cyclones since 1980, and yet these trends may not persist in the 21st century (
28).
The enhanced westward tropical cyclone steering may also be consistent with Hadley circulation expansion. The PDO phase change and Hadley circulation expansion are related (
20,
21). The poleward migration of cyclone locations has also been related to the anomalous sinking and rising motions due to the meridional expansion of the Hadley circulation (
5). Long-term zonal shifts of tropical precipitation emerge in atmospheric reanalysis data (
29). The Walker circulation dominates the large-scale zonal motion in the tropics. An attempt has been made to establish a link between the covariability of tropical cyclone genesis locations and both the Hadley and Walker circulations in the West Pacific (
30). The combined Walker and Hadley circulation variation is a plausible avenue for future study to understand the global tropical cyclone changes presented here. The considerable impacts of coastal tropical cyclones warrant further study of the changes in coastal tropical cyclone activity, as identified here, and its future projections and the consequent change in risk.
Acknowledgments
We thank C. Landsea, two anonymous reviewers, and K. Emanuel for helpful comments.
Funding: Supported by the UK-China Research and Innovation Partnership Fund through the Met Office Climate Science for Service Partnership (CSSP) China as part of the Newton Fund.
Author contributions: S.W. and R.T. conceived the study. S.W. performed the analysis. Both authors discussed the results and jointly contributed to writing the manuscript.
Competing interests: The authors declare no competing interests.
Data and materials availability: The tropical cyclone best-track data can be downloaded from the National Centers for Environmental Information website (
www.ncdc.noaa.gov/ibtracs/index.php). The ERA5, MERRA, and NCEP/NCAR reanalysis data are available at the European Centre for Medium-Range Weather Forecasts (
www.ecmwf.int/en/forecasts/datasets/reanalysis-datasets/era5), the NASA Modeling and Assimilation Data and Information Services Center (
https://gmao.gsfc.nasa.gov/reanalysis/MERRA-2/), and the NOAA Physical Sciences Laboratory (
https://psl.noaa.gov/data/reanalysis/reanalysis.shtml), respectively.