IBTrACS is a global tropical cyclone dataset maintained by the National Oceanic and Atmospheric Administration (NOAA). NOAA compiles post-storm records of best-estimate TC parameters from multiple regional meteorological agencies, providing information such as storm position, central pressure, and maximum sustained wind speed at regular time intervals (typically every 3 h). At every 3 h time step, new values are provided where possible; however, a spatial realisation of the wind field (or other parameters) is not available. IBTrACS serves as a consolidated and quality-controlled reference for historical TC activity. Table  shows which agencies' data are available in IBTrACS, in categories of data temporal resolution; those in bold are used in the following methods in this study. The differences between agencies pose different challenges throughout the analysis and will be detailed throughout.

Table  shows a comparison between the three regions: the time frame of TCs captured for this analysis, the total TCs in each region for the given time frame, the total TCs that have at least one data point with the needed parameters (maximum wind speed, RMW, minimum sea level pressure), and the number of IBTrACS data points used. The time frame of ECUS is limited by the availability of the RMW parameter and is further explained in Sect. . Table  shows a regional comparison of the grid resolution and the total number of grid points.

While three regions are being evaluated in this study – Taiwan, Japan, and ECUS – the primary focus is the coastal areas, as this is where most offshore structures are located (e.g. ). Given that TCs can be very intense systems, the TC extreme winds can still have a significant impact on areas hundreds of kilometres away from the eye, and therefore there needs to be consideration of the region chosen. The exact extent of how far damaging winds can reach is a complex subject . To ensure all potential TC effects are captured, a wide area has been chosen to evaluate, as shown by Fig. a, b, and c. In these figures, a subset of IBTrACS data is shown to illustrate which IBTrACS data points are included in our calculation. A region of approximately 500 km from the most northern, eastern, western, and southern points are included in the map, as shown by the black square. Any data points outside of the black box have been excluded, represented as dark-blue dots in the figure. The purple dots over land are also excluded from this study. The green dots represent included data points.

For the ECUS region, data have been used from the pre-compiled US agencies' variables within IBTrACS. All data for the Taiwan region come from the World Meteorological Organisation Regional Specialised Meteorological Centre in Tokyo. It is operated by the Japanese Meteorological Agency (JMA), responsible for official typhoon forecasts in the western North Pacific. The JMA dataset spans from 1951 to the present. For the Japan region, data from four agencies were combined to maximise the data availability. As previously mentioned, certain parameters are needed, and therefore merging data from different agencies can help to fill data gaps. A hierarchy of preference was established: JMA, HKO, CMA, US. This order of preference is based on the number of available data points for the Japan region, with JMA offering the highest count of data points and the US agencies the least.

Different agencies report the maximum wind speed of TCs using different averaging periods. To ensure consistency across datasets, all wind speeds have been standardised to a 10 min average to align with IEC standard 61400-1. Shorter averaging periods tend to return higher wind speeds, whereas a longer averaging period results in lower wind speeds as fluctuations of turbulence are smoothed over time. derived algorithms to convert wind speeds measured at 10 m between different averaging periods. The conversion factors used here are provided in Table . The recommended procedure to convert a shorter averaging period to a lower averaging period is as follows:

The wind speed conversion factor (CF) for converting x minute wind speed to y minute wind speed, where x<y, is given by CF=Gy,3600Gx,3600, where Gx,3600 is a conversion value that gives the highest x second mean (gust) wind speed (for example, 600 s–10 min) within 3600 s.

The Holland model, developed by , is a conceptual, empirical model to extrapolate one point of data into a wind field. Figure  illustrates a singular data point being extrapolated to an entire wind field. The following use of the Holland model follows the method in . First, the B parameter is calculated: 1B=VmaxKM2⋅ρ⋅e⋅1(Pn-Pc), where Pn is the minimum sea level pressure, Pc is the ambient pressure, ρ is air density, Vmax is the maximum wind speed at 10 m, e is Euler's number, and the constant KM≈0.7.

Using the B parameter, the gradient wind speed can be calculated at distance r from the TC centre: 2Vg=Pn-Pcρ⋅B⋅R0rB⋅exp⁡-R0rB2, where R0 is the RMW.

The RMW parameter is collected only by US agencies. Prior to 2001, the data availability of this parameter was limited to at most 7 data points per year, restricting the use of the US data to be from 2001 onwards.

The JMA dataset does not provide the RMW, but the 50 kt radius is provided. Specifically, the longest and shortest radii for 50 kt winds are reported, from which the RMW can be estimated. These data are available from 1977 onward. The JMA dataset only covers the western North Pacific, and the 50 kt radii are not available for the ECUS region.

The Holland model outputs wind speed at gradient level , the height at which surface friction no longer significantly affects the wind flow. Therefore, the wind speed needs to be scaled to a height that is of use. For this paper, the wind speed has been scaled to 100 m. suggest that, for TCs, the layer below the broad maximum wind speed of 500 m, approximately follows the logarithm of the altitude. To scale the wind, firstly, the geostrophic drag law is solved for the friction velocity : 3G=u*κln⁡u*fcol⋅z0-A2+C2, where u* is the friction velocity, G is the geostrophic wind, κ is the con Kármán constant, fcol is the Coriolis parameter, z0 is the surface correction parameter, and A and C are constants defined by the neutral conditions. The constants are set as A=1.8 and C=4.5. Using the friction velocity, the wind speed is scaled to a different height using the logarithmic wind law 4u(z)=u*κln⁡zz0, where z is the target height. Combining the geostrophic wind law with the logarithmic profile of the wind is a well-established method and already implemented into systems such as the Wind Atlas Analysis and Application Program (WAsP) .

The wind speeds were originally scaled to 10 m for comparison against the IBTrACS dataset. The approach was performed for varying surface correction parameters for each region. The optimal surface correction parameter was selected by evaluating, on average, how well the scaled wind speeds match the 10 m IBTrACS data. The adjusted values for the surface correction parameter identified are 5e-6, 9e-6, and 1e-5 m for the Taiwan, Japan, and ECUS regions, respectively. Note that although the roughness length would usually be used, here a surface correction parameter is used in its place. The rationale for this approach is discussed in detail in Appendix .

The commonly accepted deep ocean surface roughness value is 0.2 mm . However, shows that the eyewall profile follows the logarithm of the altitude using a surface roughness of 0.07 mm.

To scale to 100 m, the same method was performed using the adjusted roughness length.

For each individual grid point and for each year, the maximum wind speed at 100 m is obtained following Sect.  and . Using the annual maxima, the Gumbel distribution estimates U50. This method follows the approach described in and , in which Xm is the ascending sorted annual maxima (for each grid point), with n samples.

The sample mean is computed as 5b0=1n∑i=1nXm,i and the weighted mean 6b1=1(n-1)n∑i=1n(i-1)⋅Xm,i.

The Gumbel distribution parameters α and β can be calculated such that 7α=ln⁡22b1-b0 and β=b0-γα, where γ is Euler's constant = 0.57721. The wind speed for the return period T is 8UT=-1αln⁡ln⁡TT-1+β,

and the standard deviation associated with the wind speed for the return period T is 9σT=πα6n1+1.14kT+1.1kT2, where kT=6π-ln⁡ln⁡TreturnTreturn-1-0.577.

Using the yearly maximum for the Gumbel fit can result in the discarding of other large storms that occur in the same active year. This, in turn, could lead to the tail retaining less information than other approaches. Since U50 is derived in the Gumbel distribution by extrapolating the fitted tail to an exceedance probability, the estimate could carry additional uncertainty. A peak-over-threshold approach using the generalised Pareto distribution was tested, but the grid resolution produced too few exceedances per point. This led to unstable and often negative shape parameters. The Gumbel distribution remains a standard choice, but for this procedure, the resulting U50 represents the wind level associated with exceedance of the yearly maximum, rather than the full set of extreme storms that may occur within a year.

In Fig. a, U50 at 100 m is presented for the Taiwan region. The region expands from latitude 12 to 33.5° N and longitude 110 to 131.5° E. Taiwan is located in the centre of the figure, with the Philippines to the south and mainland China to the west.

The highest values of U50 are observed east of longitude 120° E, corresponding to areas east of Taiwan and the Philippines. In contrast, in the west of Taiwan and the Philippines, the wind speed is slower by approximately 10–20 m s⁻¹. The majority of TCs forming in the western Pacific form to the south-east of Taiwan , where they can intensify before weakening as they approach or move over land . Therefore, it is expected to see the highest winds south-east of Taiwan, before reaching land. The spatial distribution of the wind field looks somewhat consistent with previous studies based on IBTrACS data, such as and . To compare with the results of and , U50 at 10 m is also provided here (Fig. d). The maximum value of U50 at 10 m in this area is 72.7 m s⁻¹, which is 72 m s⁻¹ in and 70.4 m s⁻¹ in . Figure d here also includes more detailed spatial variability of U50 at 10 m than and due to their larger spatial grid spacing of 1° × 1°. It should be noted that the spatial distribution of Fig. a and d is very similar, and the 10 m results simply return smaller values. The detailed spatial distribution of U50 at 100 m in this study is similar to the results in (their Fig. 7a), which was based on the CFSR reanalysis data with grid spacing of about 40 km, where the values of U50 were corrected to an equivalent temporal resolution of 10 min. While present results at 50 and 150 m using CFSR, MERRA 2, and ERA5, the resultant dataset from the project has also been made available. Examining the 100 m results, the spatial distribution appears most similar to the CFSR output, with the peak located to the east of the Philippines. Meanwhile, for the ERA5 and MERRA 2 results, the peak is located further north. The maximum U50 at 100 m in this analysis is 84.3 m s⁻¹, which is closest to the ERA5 estimate of U50 of 86.7 m s⁻¹.

The 95 % confidence interval calculated using Eq. () in connection with the use of Gumbel distribution is shown in Fig. b. It widens in areas where U50 increases, peaking at approximately 15 m s⁻¹, and it narrows to approximately 2 m s⁻¹ at the smallest. In this context, the 95 % confidence interval reflects the variability associated with the fitted Gumbel distribution. It does not represent the uncertainty of all input parameters. Uncertainties relating to the entire process are discussed in Sect. . Given that the Holland model is applied to every IBTrACS data point that fits the criteria, all grid points have the same number of data points, which are listed in Table . To illustrate which areas in each region contain the highest wind speeds, Fig. c has been developed. It shows the number of data points above 19 m s⁻¹ from the Holland model at each grid point. NOAA defines a tropical storm when the 1 min average winds reach 34 kt at 10 m. By converting 34 kt to m s⁻¹, a 10 min average, and scaling the wind speed height to 100 m, the threshold becomes 19 m s⁻¹. These methods are all described in Sect. . The same reasoning holds true in Sect.  and .

In Fig. a, U50 at 100 m is presented for the Japan region. The region expands from latitude 21 to 55.5° N and longitude 120.6 to 156° E. Japan is located in the centre of the figure, with South Korea, North Korea, Russia, and China located to the west. In this region, the genesis of TCs typically occurs at lower latitudes, and they move in a north-west direction. The Coriolis force turns the TCs farther north, and once it has reached mid-latitudes, the TC turns eastward due to westerlies in addition to the Coriolis force . As they move north, the sea water temperature typically becomes cooler and the TC weakens , hence the higher U50 values occurring at latitudes lower than 36° N while the TC still holds intensity. In Fig. c, showing the number of values each grid point has above 19 m s⁻¹, there are far fewer data points above 36° N latitude in comparison to below 36° N latitude, and above 46° N latitude, there are few to no data points giving an explanation as to why wind speeds approach 0 m s⁻¹ in this region. However, this does not mean that the extreme wind speed in these areas is negligible, as other weather processes can take place, but this study solely focuses on tropical cyclone extreme wind speed. This clearly demonstrates the uncertainty associated with too few data samples.

The southern coastline of Japan is the most exposed, although both the south-western and south-eastern coastlines still experience elevated wind speeds due to TCs. As in the Taiwan region, the 95 % confidence interval increases with wind speed, reaching its highest values in areas where there is substantial and strong TC activity.

The spatial distribution as well as the magnitude of U50 at 100 m in Fig. a is quite similar to that in (their Fig. 7a) for the overlapping area, including the patchy patterns. Such levels of detail are absent in and due to the coarser spatial resolution. The results presented here differ spatially from . In , peak wind speeds occur primarily below 35° N and up to 140° E, with a north-eastward trailing feature. In contrast, this analysis shows a less pronounced decrease beyond 140° E and does not capture this north-eastward feature. This could be due to the difference in sample size at high latitudes, as the dataset used in does not exhibit the same reduction in data availability with increasing latitude. The maximum U50 in this study is lower than all values used in ; however, the ERA5 and MERRA 2 maxima are shifted farther north into the Japan region, where as the CFSR results are more consistent with those presented here, aside from a single pronounced peak in the analysis.

In Fig. a, U50 at 100 m is presented for the ECUS region. The region expands from latitude 22 to 57.5° N and longitude 88.5 to 57° W. The east coast of the US is shown, along with the top of the Caribbean in the south.

The results from the ECUS region deviate from expectations. At a granular scale, the eyewall of the TC is evident as it traverses over the region, and the U50 field exhibits a fragmented structure, which is discussed in Sect. . On a macro scale, the spatial distribution of U50 maxima and minima is unexpected. The highest wind speeds were anticipated to occur consistently in the lower half of the region. This distribution was expected due to the positive correlation between latitude and RMW , and TCs in the Gulf of Mexico, on average, have a smaller RMW than those in the Atlantic . The intensity of a TC is inversely related to the RMW , meaning that, on average, storms with smaller RMW tend to have stronger winds than TCs with large RMW.

The results here show that while the highest return level is 74.9 m s⁻¹ at 26.75° N latitude and 64.75° W longitude, the latitude farther north also exhibits return levels which are nearly as high. The wind speeds between latitudes 33 and 40° N, and longitudes 64.5 and 59.5° W are noted as rather high.

The 95 % confidence intervals follow the same pattern as in the previous cases: regions with higher wind speeds correspond to a widening confidence interval. Furthermore, the confidence interval structure also follows the same structure as U50.

The study of , reflecting the reanalysis CFSR wind field, does suggest a stronger U50,100m band south of 30° N, between about 30° N, 75° W and 22° N, 65° W, with the highest value of about 73 m s⁻¹. Due to the Coriolis force and westerlies, this strong U50,100m band turns north-east (their Fig. 7c). Compared to , Fig. b here captures the second, relatively weaker extreme wind band north of 30° N and did not capture the full picture of the first, stronger extreme wind band. We argue that it is, on one side, caused by the few IBTrACS tracks in this region (see Fig. c), and on the other side, related to the application of the Holland model. A detailed discussion is provided in Sect. . The hotspot in the Caribbean is both present in the current study and in . identify a Caribbean hotspot across all three datasets. In addition, the CFSR results show a separate, lower-latitude hotspot in the Atlantic Ocean that is not evident in the ERA5 or MERRA 2 outputs, which this analysis is more inline with. The CFSR and MERRA 2 datasets otherwise place their primary peaks above 35° N. The maximum wind speed in this analysis (74.9 m s⁻¹) is closest in magnitude to the CFSR value of 72.0 m s⁻¹, although the locations of these maxima differ substantially.

The mean U50 and the standard deviation were calculated for each parameter for each set of 100 simulations. Each parameter was varied exclusively, while all others were held constant. Each parameter was varied within its approximate 95 % confidence interval. The regional results are presented in Figs. , and .

As shown in Tables , , and , across all regions, specific individual parameters dominate the overall uncertainty, rather than the interactions, where all parameter interactions contribute below 1 % in all cases.

The results shown in Tables  for Taiwan and for Japan are, as expected, similar. They are in close proximity and have similar data restrictions. In both regions, the B parameter dominates the impact on the overall uncertainty. This could be attributed to the B parameter incorporating the uncertainty of two parameters, leading to a larger combined variance. The RMW has a smaller impact on the Taiwan and Japan regions; however, it contributes significantly to uncertainty in the ECUS region, particularly to the south, which is further discussed in Sect. . In the ECUS region, RMW, wind speed, and the B parameter have similar uncertainty; however, the method takes into account the entire defined region. From Fig. , it is clear that the RMW uncertainty is primarily heightened in the southern half of the region. By breaking the region down further, the contribution to overall uncertainty would likely shift.

Wind speed uncertainty for all regions has a substantial impact, which is expected given its direct role in the U50 calculation. Surprisingly, pressure uncertainty shows little influence in all three regions, suggesting that either the Holland model has low sensitivity to pressure variation or that the uncertainty associated with pressure is minimal. Scaling the wind speed and changing the latitude/longitude of the maximum wind speed appear to have a negligible impact on the overall uncertainty.

Examining the interaction contributions to the total uncertainty in Tables , , and , given that RMW, wind speed, and the B parameter are the biggest individual contributors to uncertainty, it is unsurprising that their interactions are the biggest contributions to uncertainty. Two notable features of the data are observed: (1) the largest interaction is between the RMW and B parameter in Taiwan, accounting for 7.2 %; (2) the wind speed–B parameter interaction in Japan is negative, indicating that the uncertainty of these parameters partially counteract each other. However, given the confidence intervals associated with all interaction contributions, these effects could be close to negligible. Therefore, the primary focus should remain on the individual effects.