Early start of the North Atlantic hurricane season with warming of the oceans

Historical data on tropical cyclones

All historical data for Atlantic tropical cyclones are taken from the HURDAT2 dataset2 for 1900–2020. Historical information about the CONUS Tropical Storm Warning and Warnings can be found in the National Hurricane Center Tropical Cyclone Reports for each TC from April 1 to May 31 that formed in the 2012-2020 period for which watches or warnings were required CONUS31,32,33,34,35,36,37.

ACE is a built-in metric that takes into account the frequency, intensity and duration of storms. ACE38 the calculations are performed from the six-month intensity data HURDAT2, with the ACE contribution for each sustained wind maximum six hours v in knots calculated:

$$ {ACE} = frac {{v} ^ {2}} {10000} $$


and all items for which a TC or subtropical cyclone (STC) has v> = 34 nodes added together to produce annual totals. Data on extratropical cyclones are not taken into account.

USACE calculations are performed as described in the methodology of ref. 6. Starting from the HURDAT2 data, we consulted the list of hurricane landings of the Atlantic Oceanographic and Meteorological Laboratory39 identify systems that impacted CONUS in the period 1966-1981 when HURDAT2 does not explicitly include landing points. We then used existing track maps to qualitatively estimate a landing location and intensity data point at the nearest hour, which we manually added to the HURDAT2 dataset for these storms. These position and intensity data were linearly interpolated with a temporal resolution of one hour. We use this hourly dataset to calculate the hourly ACE from the maximum sustained winds v in knots from:

$$ {AC} {E} _ {h} = frac {{v} ^ {2}} {60000} $$


To produce an annual USACE time series, we applied a space earth mask to this hourly ACE metric to exclude all TC positions further than 0.5 ° from any part of the CONUS. We have selected the 0.5 ° buffer to correspond to the typical radius of the maximum winds of a mature TC and to account for possible small observation errors, particularly at the beginning of the recording.40. For the purposes of the counts, we consider a “landfall” as the circulation center of a TC or STC with v> = 34 nodes passing within 0.5 ° of the CONUS for any interpolated time interval of one hour. A single TC or STC can make multiple CONUS landings, but can only make a maximum of one landfall in each of the coastal regions shown in Fig. 4a from ref. 6. To find the annual values ​​of USACE, we added this series on the CONUS and temporally on each hour of the year. All time series, distributions, and counts in this document consider the annual cycle of TC activity starting at 0000 UTC on March 1 of the specified year and ending at 2359 UTC on February 28 or 29 of the following year. This choice is made to best align with the North Atlantic SST average annual low, and therefore January and February TC activity such as 2016 Hurricane Alex41 it is considered part of the annual cycle of activity of the North Atlantic TC of the previous year. The January-February TC and STC formations are less than 0.5% of the total annual formations in the period 1950-2020.

Quantile regression and trend calculations

Trends in the start date of the storm named Atlantic were calculated from the annual least squares ordinary regression of the number of six-hour periods between 0000 UTC on 1 March and the first six hours HURDAT2 TC or STC with sustained winds> = 34 knots with respect to the year, subsequently converted into a rate of days per year-1. This calculation was repeated to exclude any TC or STC with fewer than 8 half-yearly HURDAT2 entries with maximum sustained winds> = 34 knots in order to test the influence of short-lived systems on the overall trend. Trends in the CONUS initial impact date were calculated from the annual least squares ordinary regression of the number of one-hour periods between 0000 UTC on 1 March and the first interpolated one-hour HURDAT2 TC or STC entry with sustained winds> = 34 knots passing within 0.5 ° or less of CONUS with respect to the year, converted into a rate of days per year-1. The significance of all trends was assessed with a Mann-Kendall test42.43. The proportions of TC activity occurring in the current official definition of hurricane season are reported as a percentage of all initial TC and STC formations from March 1 to February 28/29 with sustained winds> = 34 knots, CONUS TC landings and STC with sustained winds> = 34 knots, called storm days, ACE and USACE, occurring between 0000 UTC on June 1st and 2359 UTC on November 30th. These data are calculated in average trailing windows of 50 years to take into account the multidecennial variability in the activity of the Atlantic CT.13.

Trends in percentile cut-off dates for ACE (both a climatology in which short-lived CTs are included, and one in which they are excluded based on the above criteria) and USACE are found using quantile regression44. Quantiles divide the cut-off dates at which the specified percentage of the total annual ACE and USACE is reached, ranging from 1% to 99% at 1% intervals, into equal subsets. Quantile regression applies ordinary least squares regression to the conditional quantiles of the response variables ACE and USACE, which are taken at regular intervals from the cumulative distribution of ACE and USACE over the period 1979-2020 and 1900-2020, respectively, for the regressions performed over the year. The sensitivity of the calculated trend to the choice of the time series length is verified by repeating this quantile regression methodology for the initial years of the cumulative distribution functions for ACE of 1950–1990 and USACE of 1900–1990, with a fixed end year of 2020 .

Sensitivity test of the synoptic environment of quantile regression

The dependence of the given ACE and USACE percentile thresholds on synoptic environmental factors is verified by performing quantile regressions of these response variables with respect to GPI, SST, 200 T, RH and VWS. These regressions are performed by the cumulative distribution functions of ACE and USACE over the period 1979-2020, due to the superior quality of global reanalyses since 1979. The SST time series used in quantile regressions is calculated from the monthly mean of April and May ERSSTv522 SST fields, averaging 10–36 ° N, 100–70 ° W excluding portions of this box on the Pacific Ocean. The spatial resolution of ERSSTv5 is 2 °. The time series RH, 200 T and VWS are calculated from the monthly average of April and May ERA521 fields. The primary GPI time series used in this study23 it is calculated from the monthly average ERA5 fields for April and May. The GPI value is calculated at each 0.25 ° grid point in this box for the equation23.25:

$$ {GPI} = {abs} ({10} ^ {5} {{ eta}}) ^ {2} {* left ( frac {{{{{{ rm {RH}}}}} }} {50} right)} ^ {3} {* left ( frac {{{{{{ rm {PI}}}}}}} {70} right)} ^ {3} * { (1 + 0,1 {{{{{ rm {VWS}}}}}})} ^ {- 2} $$


with η the absolute vorticity of 850 hPa, RH the 700 hPa RH, PI the potential intensity, a theoretical maximum intensity TC given SST and a specified atmospheric column25.45and VWS is vertical wind shear between 850 and 250 hPa. The GPI statistic was developed using an adaptation procedure of these variables in the NCEP / NCAR reanalysis46 to the seasonal and spatial climatological recording of global cyclogenesis events. These grid point GPI values ​​are then averaged over 10–36 ° N, 100–70 ° W separately for April and May, then averaged over the two months to produce the GPI time series used in quantile regressions.

An alternative form of GPI24 was used at several points in this study for comparative purposes using the same data sources described above. This GPI construction has the form:

$$ {GPI} = | {{{{{ rm { eta}}}}}} | ^ {3} , * , {{ chi} ^ {- frac {4} {3}} , * , left ( right. {{{{ rm {MAX}}}}}} (({{{{{ rm {PI}}}}}} – 35m {s} ^ { -1}), 0)} ^ {2} * {(25 {{ms}} ^ {- 1} + {{{{{ rm {VWS}}}}}})} ^ {- 4} $ $


with η, PI and VWS as above, e

$$ chi = frac {{s} _ {b} – {s} _ {m}} {{s} _ {0} ^ {*} – {s} _ {b}} $$


with χ a measure of the wet entropy deficit in the middle troposphere egbS.m, ex*0 wet entropies of the boundary layer and of the middle troposphere and wet entropy of saturation of the sea surface24.

SST threshold TC genesis and SST trend

Since the daily SST values ​​are not available from the ERSSTv5 dataset, the spatial coverage of the 26.5 ° C threshold value is calculated from the daily average of ERA521 SST fields, which have a spatial resolution of 0.25 °. The ERA5 SSTs are drawn from several analyzes in the period 1950-2020, including HadISST2.1.0.0 before 1961, HadISST2.1.1.0 between 1961 and mid-200747and OSTIA since mid-200748. The native time resolutions of these analyzes are one month, five days, and one day, respectively. However, because SSTs change slowly over time, using these datasets scaled to a uniform daily resolution is unlikely to introduce significant observational bias into the SST threshold analysis. The proportion of the 10–36 ° N, 100–70 ° W box for which this criterion is met is calculated by applying a land mask and excluding the Pacific Ocean and dividing the number of daily grid points exceeding 26.5 ° C for the total number of non-terrestrial and non-Pacific grid points. ERSSTv5 monthly mean SST trends are calculated from normal least squares regressions over the year. The global mean box or SST trends are calculated after excluding the Earth grid points. Supplementary Figure 7 was the only case where ERA5 SST values ​​were used instead of ERSSTv5 monthly averages as the source of SST values.

Optimized hurricane season limits

The objective start and end dates for the Atlantic hurricane season are calculated from 50-year time series of named storm formations, storm impacts named CONUS, storm days named Atlantic, Atlantic ACE and USACE, with an intra-season smoothing filter of 15 days applied, normalized for total activity within the window. This results in a daily time series of the leveled percentage of total TC activity occurring every day from March 1 to February 28/29 within the 50-year window. A middle-out algorithm starts at the climatological peak of Atlantic CT activity in mid-September and adds one day at a time to the target season until the period that includes the fewest contiguous days is reached for a specified percentage activity threshold . Thresholds tested are 95%, 97%, and 99% of total activity for each of the TC metrics.

Leave a Reply

%d bloggers like this: