As the Earth’s atmosphere warms, the atmospheric circulation changes. These changes vary by region and time of year, but there is evidence that anthropogenic warming causes a general weakening of summertime tropical circulation1,2,3,4,5,6,7,8. Because tropical cyclones are carried along within their ambient environmental wind, there is a plausible a priori expectation that the translation speed of tropical cyclones has slowed with warming. In addition to circulation changes, anthropogenic warming causes increases in atmospheric water-vapour capacity, which are generally expected to increase precipitation rates9. Rain rates near the centres of tropical cyclones are also expected to increase with increasing global temperatures10,11,12. The amount of tropical-cyclone-related rainfall that any given local area will experience is proportional to the rain rates and inversely proportional to the translation speeds of tropical cyclones. Here I show that tropical-cyclone translation speed has decreased globally by 10 per cent over the period 1949–2016, which is very likely to have compounded, and possibly dominated, any increases in local rainfall totals that may have occurred as a result of increased tropical-cyclone rain rates. The magnitude of the slowdown varies substantially by region and by latitude, but is generally consistent with expected changes in atmospheric circulation forced by anthropogenic emissions. Of particular importance is the slowdown of 30 per cent and 20 per cent over land areas affected by western North Pacific and North Atlantic tropical cyclones, respectively, and the slowdown of 19 per cent over land areas in the Australian region. The unprecedented rainfall totals associated with the ‘stall’ of Hurricane Harvey13,14,15 over Texas in 2017 provide a notable example of the relationship between regional rainfall amounts and tropical-cyclone translation speed. Any systematic past or future change in the translation speed of tropical cyclones, particularly over land, is therefore highly relevant when considering potential changes in local rainfall totals.
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
This work was supported by NOAA’s National Centers for Environmental Information.Reviewer information
Nature thanks M. Mann and C. Patricola for their contribution to the peer review of this work.
Extended data figures and tables
Extended Data Fig. 1 Time series of annual-mean tropical-cyclone translation speed and their linear trends in varying ocean basins.
Time series are shown for the individual ocean basins: North Atlantic, western North Pacific, northern Indian, eastern North Pacific, and the regions west and east of 100° E in the Southern Hemisphere.
Extended Data Fig. 2 Time series of annual-mean tropical-cyclone translation speed and their linear trends in varying latitude belts.
a, b, Time series are shown for global latitude belts southward (poleward) of 30° S and northward (poleward) of 35° N (a), and in the belts 30°–0° S, 0°–15° N, 15°–25° N and 25–35° N (b). The broader latitude belts defined in the less-active Southern Hemisphere were needed to have a large enough sample of tropical cyclones in each to perform the analyses. The analyses shown here are fairly robust to the choice of latitude-belt bounds.
Extended Data Fig. 3 Best-track tropical-cyclone centre locations for tropical-cyclone translation speeds over land.
Locations (black dots) are shown for the regions considered in Fig. 3 and Extended Data Table 1. a–g, Over-land positions are shown for land areas affected by tropical cyclones globally (a), in the North Atlantic (b), western North Pacific (c), eastern North Pacific (d) and northern Indian (e) basins, and in the regions west (f) and east (g) of 100° E in the Southern Hemisphere.
The histogram shows the distribution of residuals from the trend for the time series of annual-mean global tropical-cyclone translation speed shown in Fig. 1a. The normal distribution is shown in red.
Extended Data Fig. 5 Quantile–quantile plot for the residuals of the global ordinary least-squares regression.
The left tail of the distribution is thinner than the normal distribution, which suggests that the likelihood of an extremely slow translation speed is less than would be found in a normally distributed sample.