The Washington PostDemocracy Dies in Darkness

Analysis: Atlantic hurricane season is growing longer

Hurricane Otto, Nov. 24. (CIMSS, University of Wisconsin)

Just after noon on Thanksgiving Day, as the Lions and Vikings were kicking off in Detroit and tens of millions of political arguments were kicking off in America’s dining rooms, Hurricane Otto was making landfall in Central America.

Costa Rica and Nicaragua are continuing to recover from this destructive storm, but Otto’s unusual track and intensity are reigniting a debate among meteorologists about whether the Atlantic hurricane season, officially spanning June 1-Nov. 30, is lengthening.

I have a conducted an analysis that shows the hurricane season has in fact grown longer since 1979, shrinking winter vacations for hurricane forecasters and requiring coastal residents to remain vigilant for even longer periods.

The 2016 season not only ended with an unprecedentedly intense November landfall from Hurricane Otto, but also began with the precocious development of Hurricane Alex in the open Atlantic in mid-January and Tropical Storm Bonnie drenching the Carolina coast on Memorial Day weekend.

Other recent out-of-season renegades include 2015’s Tropical Storm Ana, the earliest U.S. landfall in history, and Tropical Storm Beryl, which struck Jacksonville, Fla., with 65 mph winds on Memorial Day 2012.

While these recent case examples offer interesting anecdotal support for lengthier seasons, scientists began investigating whether the typical start and end dates of the Atlantic tropical cyclone season could be changing in the mid-1990s after a spate of very active hurricane seasons.

In 2008, James Kossin published evidence that the dates of the earliest and latest Atlantic tropical cyclone formation dates were pushing outward at a rate of about one day per year, and suggested this trend could be linked to warming ocean temperatures. However, more recently Juliana Karloski and Clark Evans found no trend in Atlantic tropical cyclone formation dates when the time frame was extended to 2014, and a 2015 study of season length in climate models led by John Dwyer yielded mixed results.

These varying conclusions suggest taking a broader approach, especially in light of the elevated early- and late-season activity in 2015 and 2016. So I decided to conduct an analysis to see if any real change in the length of the hurricane season had manifested itself.

The figure above shows the number of days between the first and last Best Track* six-hourly historical positions at tropical storm, hurricane and major hurricane intensity within each calendar year since 1979. Length of season has trended higher by about 1-2 days per year for each of these intensity thresholds, which is significant at a 95 percent confidence level for named storms and a 90 percent confidence level for major hurricanes. (The trend is not significant for hurricanes.)

There are several possible explanations for this apparent tendency, including an overall increase in Atlantic tropical cyclone activity. The Atlantic Multidecadal Oscillation, a cycle of warmer and cooler Atlantic Ocean temperatures that has been shown to modulate seasonal tropical cyclone activity, flipped from an unfavorable negative phase to a favorable positive phase around 1995. This phase change is likely responsible for at least some of the observed increase in season length since 1979.

A problem with using storm formation dates as a measure of season length is that in more active hurricane seasons, we would expect more early and late season storms, even without any change in the preferred distribution of storms across the calendar. To address this issue, I looked at the season-relative progress of total Best Track accumulated cyclone energy (ACE) produced by all storms over each hurricane season.

The plot above shows how quickly each hurricane season since 1979 advanced from 100 percent of the total season energy remaining (dark blue) toward completion (dark red) in normalized terms. This way of visualizing hurricane seasons highlights the busiest portions of each year as times of fastest color change and controls for varied levels of overall activity.

In seasons like 1988, in which a strengthening La Niña pushed more of the season’s storm activity later in the year than normal, blues extend further right on the plot. In early-shifted seasons, like the El Niño-influenced 1997, reds reach further left.

(Because strong storms produce much more ACE than the weak and short-lived systems that often develop early or late in the hurricane season, this method also eliminates the bias associated with counting powerful and marginal tropical cyclones equally.)

With a record of the date when each hurricane season from 1979-2016 was a certain percentage complete, we can now test if these annual thresholds are shifting earlier or later.

I explored these trends using a data analysis tool called quantile regression, which is a way of determining how much the expected date of each ACE percentile has shifted over the study period and the uncertainty of that estimate.

As shown above, quantile regression reveals there is no significant date shift across the vast majority of hurricane season in terms of energy. The date at which the Atlantic hurricane season is expected to be 10 percent, 50 percent, or 90 percent complete have not changed significantly since 1979.

However, there is an intriguing and statistically significant trend for very early season activity at a rate of about half a day to a day earlier per year, which is highlighted in red. This means all else being equal, we might expect the first 5 percent of a hurricane season threshold in the Atlantic to be reached about 20 days earlier in 2016 relative to 1979, or approximately July 20 instead of Aug. 10.

(This relationship remains significant if 2016 is excluded from the calculation, so this year’s record four named storms by mid-June aren’t solely responsible for the trend.)

Still, this effect disappears completely in August, by which time stronger storms become more common in the historical record, and hurricane seasons remain likely to be almost 20 percent complete by Aug. 20.

Overall, examining the distribution of tropical cyclone energy within hurricane seasons reveals that much more has stayed the same than changed in the Atlantic over the last four decades.

Hurricane seasons still peak around Sept. 10, and powerful storms remain most likely to form between early August and mid-October (sorry, Otto). However, there is at least some support for Kossin’s theory of a lengthening hurricane season, with the earliest 5 percent of the hurricane season trending significantly earlier since 1979.

An average Atlantic hurricane season produces around 100 units of ACE, so in real terms getting to 5 ACE units more quickly means perhaps shifting a tropical storm or two into June or July.

In the end, drawing any unequivocal conclusions about changes in the length of hurricane season from only 38 years of the highest-quality storm data is probably overstating the case, an unfortunate truth that also makes predicting seasonal activity a challenge. Nevertheless, while Hurricane Otto’s turkey trot doesn’t necessarily mean that you need to worry about unexpected tropical guests for future Thanksgivings, there’s enough statistical evidence for a slight shift earlier in the opening act of hurricane season that you might consider paying a little more attention to the Atlantic in the Mother’s Day to Bastille Day corridor.

* National Hurricane Center operational advisory intensities are substituted for Best Track data for 2016 Atlantic tropical cyclones.

The author, Ryan Truchelut, is co-founder and chief meteorologist of WeatherTiger, a weather consulting and seasonal forecasting company. An experienced hurricane researcher, he has a BS in geoscience from Princeton University and a PhD in meteorology from Florida State University. Follow: @wx_tiger