One spring day a few years ago, my one-year-old son and I went for our usual daily walk around our neighborhood to look at clouds, colorful flowers, and lizards. On this particular afternoon, soon after cresting what passes for a hill in Northern Florida, we were chased back home by a formidable outer band of Tropical Storm Alberto. While I should have done a more thorough job of checking the radar prior to leaving home, I do have some excuse for not expecting brisk winds and a torrential downpour from a tropical cyclone: it was not yet, in fact, hurricane season.
Alberto, which caused 18 deaths and $125 million in flood damage in May 2018, is not alone in impacting the continental United States outside the June 1-November 30 bounds of hurricane season. In fact, Alberto was one of seven tropical or subtropical storms since 2012 to require U.S. tropical storm watches or warnings to be hoisted by the National Hurricane Center before June 1.
In our recent paper in Nature Communications, I, along with co-authors Phil Klotzbach, Erica Staehling, Kim Wood, Daniel Halperin, Carl Schreck, and Eric Blake, investigate whether initial tropical cyclone (TC) activity and U.S. landfall risks are shifting earlier in the year, and the potential causes of these trends. As my experience shows, understanding trends in TC activity onset has real-world implications. Not only do pre-season or early-season storms disproportionately impact populated areas, especially through excessive precipitation, but they strike at a time when the general population (or even a sleep-deprived hurricane forecaster) is not primed to prepare for such threats or react to warnings. With around 30% of TC casualties arising from flooding, any divergence between perception and reality of risk is of serious concern.
We began our study by using quality-controlled TC historical data from the National Hurricane Center reaching back to 1900 for U.S. landfalls and 1979 for Atlantic storms. The 1979 start represents the earliest time for which observational tools like satellite imagery were sufficiently available to be confident that early- and pre-season storms in the open ocean were not systematically going undetected. For landfalls, 1900 is regarded as the year for which the full U.S. coastline was populated with enough weather stations and observers to detect TC strikes.
Using this data, we show that there is a significant trend towards earlier onset of North Atlantic TC activity. The date at which the first few percent of overall TC activity within a hurricane season occurs is shifting earlier at a rate exceeding five days per decade since 1979. The initial continental U.S. named storm landfall is trending earlier at a rate of about two days per decade since 1900. To address whether these trends might be caused by the increased detection of short-lived storms in recent years due to improved observational tools, we also repeated all of our analyses with and without named storms lasting less than 48 hours. The shift towards earlier TC development and U.S. landfall risks was accentuated when excluding these "shorties."
Satisfied that these trends were real, we then used an analytical technique called quantile regression to investigate why more tropical storms are developing and making landfall in May and early June in recent Atlantic hurricane seasons. We found that additional pre-season and early-season activity is linked to spring environmental conditions becoming more conducive for TC formation, particularly in western portions of the subtropical Atlantic Ocean and the western Caribbean Sea. Some parameters that matter for whether storms form, like moisture and wind shear, were shown not to have changed much in these areas since 1979, or to only have a weak influence on early season activity.
Rather, we found that increased TC activity in the pre-season and earliest weeks of hurricane season is primarily driven by warming ocean temperatures in the western Atlantic. For each 1°C of ocean warming in this area, the initial few percentiles of Atlantic TC activity are expected to shift about one month earlier. With spring warming trends in the Caribbean and western Atlantic accelerating to 0.3°C per decade since 1995, ocean temperature shifts roughly explain the observed change in season onset, even normalizing for quieter and busier years.
This suggests that the threshold date at which the first percentiles of TC activity and U.S. landfalls occur might continue to shift earlier, irrespective of whether Atlantic hurricane seasons overall become busier, quieter, or stay around the same in the future. While by definition, the initial percentiles of the season only comprise a small portion of total Atlantic TC activity, pre-season and early season TCs can have outsized societal impacts, particularly with respect to dangerous flash flooding. Alberto's impacts in 2018 underscore these rising risks; another example of an extreme, destructive TC precipitation event in the opening days of hurricane season is 2001's Tropical Storm Allison, which deposited up to 35" of rain and caused nearly $10 billion in damage in eastern Texas.
So, what does all this mean for hurricane season itself? As it turns out, the concept of "hurricane season" currently lacks a precise, scientific definition, and actually originated as the annual operating dates of a telegraph line connecting U.S. Weather Bureau offices in the 1930s. While official efforts are underway to develop a more rigorous delineation, that hurricane season is a social construct underscores the additional considerations like public communication, net risk mitigation, and emergency management activities in determining when to draw the line. We found long-term trends earlier in optimized hurricane season start dates based on initial Atlantic named storm formations, U.S. named storm landfalls, and Accumulated Cyclone Energy occurring over the U.S., all of which suggest the season could be defined empirically as beginning prior to June 1 based on the last 50 years of data.
Overall, the additional ocean temperature increases likely in the coming decades will probably continue to widen the mismatch between the actual onset of TC risks to the U.S. coastline, and the current start date of hurricane season on June 1. Our study shows initial Atlantic TC formation is shifting earlier by at least half a day per year on average, and reasonably may be expected to continue to do so given the likely cause of the changes. Should the definition of hurricane season adapt to include portions of May? Our study suggests that perhaps it should, at least from a quantitative perspective. After all, it's no fun to get caught in the rain without an umbrella.