Earlier onset of North Atlantic hurricane season with warming oceans

The Atlantic hurricane season is operationally defined as 1 June–30 November, chosen to capture most tropical cyclone (TC) formations given the strongly peaked seasonal cycle of activity. Favorable overlaps of vertical instability, mid-level relative humidity, and low vertical wind shear during late summer to fall drive this seasonality, with over 85% of accumulated cyclone energy (ACE) between 1 August and 31 October. However, several TCs have formed before June 1 in recent years, some requiring U.S. watches or warnings. Early-season storms disproportionately affect populated areas and often produce significant precipitation impacts. This study investigates whether the onset of North Atlantic TC activity is shifting earlier, how such shifts manifest in occurrence- and impact-based metrics (ACE and U.S. ACE, USACE), and what environmental changes are responsible.

Prior studies reported mixed evidence regarding changes in Atlantic TC season onset. One analysis suggested earlier initial named storm formation in parts of the central/eastern Atlantic by ~1 day per year during 1980–2007, while another found no broader trend for 1979–2014. Projections of future season length changes have also been model dependent with conflicting results. Additional literature highlights potential biases from increased detection of short-lived storms in recent decades, best-track uncertainties, and multi-decadal variability in Atlantic TC activity, motivating careful treatment of detection biases and robust statistical methods.

Data: Historical Atlantic TC data were taken from HURDAT2 for 1900–2020. U.S. watches/warnings for pre-season storms (April–May) during 2012–2020 were compiled from National Hurricane Center Tropical Cyclone Reports. ACE was computed from six-hourly HURDAT2 maximum sustained winds v (kt) summing v²/10,000 for entries with v ≥ 34 kt. USACE was computed by interpolating tracks to hourly resolution, applying a 0.5° coastal buffer land mask for the CONUS, and summing hourly ACE where TC centers were within 0.5° of land; landfalls were defined as hourly entries with v ≥ 34 kt within 0.5° of CONUS. The annual TC year was defined from 1 March to the end of February of the following year to align with the SST annual minimum. Short-lived storms were identified as systems with fewer than eight six-hourly entries with winds > 34 kt and excluded in sensitivity tests. Statistical approach: Trends in initial named storm and initial CONUS impact dates were estimated using ordinary least squares regressions of elapsed time (from 0000 UTC 1 March) against year; significance assessed via Mann–Kendall tests. Quantile regressions were used to estimate trends in ACE and USACE percentile threshold dates (1%–99%) versus year for 1979–2020 (ACE) and 1900–2020 (USACE). Sensitivity to analysis window was tested by varying the start year (ACE: 1950–1990; USACE: 1900–1990) with fixed end year 2020. Environmental analysis: The western Atlantic (WATL) region 10–36°N, 100–70°W was analyzed as the primary domain for pre-/early-season TC genesis. Spring (April–May) environmental fields were taken from ERA5 reanalysis (200-hPa temperature, 600-hPa RH, vertical wind shear 850–200 hPa) and ERSSTv5 (SST). Genesis Potential Indices (GPI) were computed using ERA5-based formulations following Camargo/Emanuel/Sobel and an alternative Emanuel formulation. April–May mean GPIs and component variables were averaged over the WATL box. Quantile regressions related ACE/USACE percentile threshold dates to WATL spring GPI and to individual components (SST as a proxy for potential intensity along with 200-hPa temperature, 600-hPa RH, and vertical wind shear). The contribution of each component to onset shifts was estimated by multiplying quantile regression coefficients by observed 1979–2020 trends of each environmental parameter. SST threshold analyses (26.5 °C coverage) used daily ERA5 SST fields to assess changes in the spatial extent of thermodynamically favorable conditions. Objective hurricane season bounds were derived using 50-year trailing averages of activity metrics (named storm formations, CONUS impacts, named storm days, ACE, USACE), 15-day smoothing, and a middle-out algorithm to identify the most compact period capturing 95%, 97%, or 99% of activity.

• Initial onset is occurring earlier: The first Atlantic named storm date trended earlier by 1.2 days per year (p < 0.005) during 1979–2020, strengthening to 1.4 days per year (p < 0.001) when excluding short-lived TCs. The first CONUS landfall trended earlier by 0.22 days per year (1900–2020; p < 0.05), i.e., about 2 days per decade.
• Early-percentile ACE thresholds are shifting earlier: The 1st–3rd ACE percentile threshold dates shifted earlier by 0.5–1.0 day per year (p < 0.05 for 2nd–3rd; p < 0.01 for 1st) during 1979–2020; excluding short-lived TCs yields ~1 day per year (p < 0.01). For USACE, the 1st–2nd percentile thresholds advanced by ~0.2 days per year over 1900–2020 (p < 0.05).
• Robustness to start year: For ACE 1st-percentile, trends are earlier by 0.15–0.5 days per year for start years 1950–1965 and 0.5–1.0 days per year (p < 0.05) for 1965–1987; excluding short-lived TCs yields 0.5–1.5 days per year (p < 0.05) for start years after 1970. USACE 1st-percentile trends are significant across all windows, 0.2–0.75 days per year earlier for 1900–1969 start years and 0.75–2.0 days per year earlier (p < 0.01) for 1970–1990.
• Environmental linkage: Higher WATL April–May GPI values are associated with earlier ACE threshold dates for the first 10 percentiles (20–30 days earlier per unit GPI; p < 0.05) and earlier USACE thresholds for the first three percentiles (≈40 days per unit GPI; p < 0.05). WATL spring GPI increased by ~0.1 units per decade since 1979 (p < 0.05), implying a 10–15 day earlier ACE onset and ~15–20 day earlier USACE onset since 1979.
• Drivers: Thermodynamic factors dominate. Higher spring WATL SSTs and mid-level RH are significantly linked to earlier onset (ACE first three percentiles p < 0.01; USACE first four percentiles for SST and first 18 percentiles for RH p < 0.05). Only SST shows a significant positive trend since 1979 (~0.15 °C per decade; accelerating to ~0.3 °C per decade since 1995). Relationships with 200-hPa temperature and vertical wind shear are not significant for initial percentiles.
• Magnitude of observed shifts: SST-driven changes alone imply ~25 days earlier onset for 1st-percentile ACE and USACE over 1979–2020. Observations show median 1st-percentile ACE shifted from 6 July (1979–1999) to 16 June (2000–2020; 20 days earlier), and USACE from 22 July to 20 June (31 days earlier).
• Early-season impacts: From 1979–2020, 41% of pre–1 August TCs made CONUS landfall vs 23% for other TCs. While only 6% of basin ACE occurs before 1 August, 17% of USACE occurs before 1 August.
• Operational implications: The portion of TC activity and impacts within the official 1 June–30 November season has declined (e.g., 96% of CONUS landfalls in-season during 1971–2020 vs 99% during 1921–1970), suggesting season bounds that begin prior to 1 June may better capture risk.

The findings demonstrate a statistically significant shift toward earlier onset of North Atlantic TC activity, concentrated in the earliest percentiles of ACE and USACE and robust to detection biases from short-lived storms and variations in analysis window. Quantile regressions link these shifts to more favorable spring thermodynamic conditions in the western Atlantic, particularly rising SSTs, which increase potential intensity and environmental instability. Although upper-tropospheric temperatures have also warmed (~0.15 °C per decade), the increase is not sufficient to offset SST-driven increases in instability, consistent with expected lapse-rate behavior; thus, net instability has increased, advancing the date when large areas meet convective thresholds for genesis. Because pre-/early-season TCs disproportionately affect land and contribute a larger share of USACE than their share of basin-wide ACE, earlier onset has outsized societal implications. The results suggest revising operational season bounds (e.g., including portions of May) to align with the effective onset of TC risk and indicate that continued ocean warming could further advance onset by roughly 0.5–1.0 days per year, independent of changes in total seasonal counts or ACE.

This study provides multiple lines of evidence that the onset of North Atlantic TC activity has shifted earlier in recent decades. Early-season ACE and USACE thresholds are advancing, first named storm formation and initial U.S. landfall dates are trending earlier, and these changes are physically consistent with observed springtime warming of western Atlantic SSTs and associated increases in genesis potential. Thermodynamic changes, primarily SST increases, are the dominant drivers; dynamic factors show weaker or insignificant relationships for onset. The empirical and physically based results support reconsidering the operational start of the Atlantic hurricane season to include portions of May. Future work should refine attribution by quantifying anthropogenic versus natural contributions to spring SST trends, assess model projections of onset shifts under different warming scenarios, explore regional heterogeneity within the WATL, and evaluate implications for early-season hazard preparedness and forecast practices.

• Detection biases: Increased observation capabilities in the satellite era have improved detection of short-lived and weak TCs, potentially biasing onset metrics; sensitivity analyses excluding short-lived storms mitigate but may not fully eliminate this bias.
• Reanalysis and dataset uncertainties: Environmental trends and GPI calculations rely on ERA5 and ERSSTv5; structural changes and inhomogeneities in these datasets over time can affect trend estimates. Best-track positional and intensity uncertainties also propagate into ACE/USACE and landfall timing.
• Statistical assumptions: Quantile regression results can be sensitive to the choice of time window and endpoints; although robustness tests were performed, residual sensitivity remains.
• Proxy choices: Use of SST and 200-hPa temperature as proxies for potential intensity and instability, and the specific GPI formulations, introduce methodological assumptions that may not capture all aspects of genesis physics, particularly baroclinic influences on early-season storms.
• Spatial generalization: Analyses focus on a broad WATL box (10–36°N, 100–70°W); regional variations within this domain may exhibit different behaviors not fully resolved in the basin-wide aggregates.
• Manual adjustments: For some historical storms (1966–1981), landfall points were added manually to HURDAT2 for USACE computations, introducing potential subjective uncertainty.