Atlantic tropical cyclones downscaled from climate reanalyses show increasing activity over past 150 years

The study addresses whether long-term trends in Atlantic tropical cyclone (TC) activity are real or artifacts of incomplete historical observations. While the North Atlantic constitutes only about 12% of global TCs, its record extends to 1851 and is relatively better observed than other basins, especially pre-satellite. Prior efforts have adjusted historical counts for presumed missing storms using ship and land observation statistics, concluding no detectable trends in hurricanes or major hurricanes over 1851–present. However, these corrections rely on assumptions (e.g., detection thresholds, wind distributions, reporting practices) and tend to privilege a null hypothesis of no trend, limiting detectability of true changes. The paper proposes an alternative: dynamical downscaling of century-scale climate reanalyses that assimilate mainly surface pressure and SST, to infer past TC activity independent of direct storm observations. The purpose is to reassess Atlantic TC trends, evaluate potential biases, contrast with global TC behavior, and examine physical drivers (regional vs global climate influences) of variability and change.

The paper reviews the IBTrACS global TC archive, noting sparse and error-prone records outside the Atlantic and western North Pacific prior to satellites. It summarizes work by Vecchi, Knutson and colleagues that estimated missing Atlantic storms by sampling contemporary tracks with historical ship/land observing statistics. These studies hinge on assumptions about detection thresholds, wind distributions relative to storm centers, land observing practices, and track continuity. They do not account for shifts in storm geography or evolving ship avoidance strategies, and inherently bias toward the null of no trend, thereby reducing any real trend signal. The review distinguishes lack of detectable trend from lack of actual trend, emphasizing that violations of assumptions could conceal real changes. It also notes ongoing improvements and incompleteness in digitized marine observations (ICOADS) used for such corrections compared with broader historical sources used in TC reconstructions.

Three twentieth-century climate reanalyses were downscaled: NOAA 20th Century Reanalysis v2c (1851–2014), NOAA v3 (1836–2015), and ECMWF CERA-20C (1901–2010). These systems assimilate surface pressure, SST, sea ice (and marine surface winds for CERA-20C) into global forecast models, accounting for historical radiative forcings. For each year and each reanalysis, 100 North Atlantic and 100 global synthetic TC tracks were generated using a statistical–deterministic downscaling approach. Inputs include monthly mean SST; atmospheric temperature and humidity (used to compute monthly mean potential intensity following Bister and Emanuel); and synthesized daily 850/250 hPa winds constructed from monthly means, variances, and covariances with geostrophic turbulence spectra. The basin is randomly seeded in space and time with weak proto-vortices (seeding tapered toward the equator). Vortices are advected using a beta-and-advection trajectory model (weighted 250/850 hPa winds). Intensity evolves via a deterministic coupled ocean–atmosphere, axisymmetric, quasi-balanced model with parameterized vertical shear effects and time-interpolated environmental fields. Seeds that fail to reach >7 m s−1 within 2 days are discarded; only storms reaching a lifetime maximum wind of at least 40 kt (~20 m s−1) are retained. The process continues until the preset number of successful storms is obtained. Annual frequency is taken proportional to the ratio of successful to total seeds, with a single proportionality constant set by matching the time mean to reliable historical periods. Atlantic metrics analyzed include: (i) landfalling events (every center crossing from sea to any land mass counted as a separate event; storms must attain ≥40 kt during lifetime); (ii) basin hurricane counts (≥63 kt lifetime max); (iii) basin major hurricane counts (≥95 kt lifetime max). Continental U.S. landfall frequency counts each storm at most once. Power Dissipation Index (PDI) at U.S. landfall is computed as the annual sum of the cube of maximum wind at landfall. Time series are smoothed with 7-year running means to damp high-frequency noise. Downscaled counts are multiplicatively rescaled to match the mean of the historical data over reference periods (e.g., last 50 years for Atlantic; 1990–2010 for global). Statistical inference uses Poisson regression for count data and ordinary linear regression otherwise. To gauge sensitivity to reanalysis internal variability and sampling, a second ensemble member of CERA-20C was also downscaled for global metrics. Potential biases from assimilated TC data, evolving observation density/instrumentation, and SST dataset differences across reanalyses are examined; surface pressure trends over genesis regions and their thermodynamic implications for potential intensity are analyzed. A Genesis Potential Index (GPI) is computed from reanalysis fields to relate environmental changes to genesis rates and to separate regional vs global contributions using a linear decomposition of potential intensity into global and regional components.

- Downscaled North Atlantic TC metrics (landfall events, basin hurricanes, basin major hurricanes) show statistically significant upward trends over the full periods of the reanalyses. All regression curves exhibit small p-values; the largest is 0.004 (NOAA v2c landfall counts), indicating robust increases. The smallest trends are generally from NOAA v2c. - Agreement between downscaled and observed (uncorrected IBTrACS) Atlantic metrics is strong since ~1900 for the newer reanalyses; downscaled values exceed observations before 1900, consistent with undercounting in early historical records. - Poisson regression of downscaled landfall counts against historical data over 1901–2010 yields p-values of 0.09 (NOAA v2c), 3.8×10−6 (NOAA v3), and 2.5×10−5 (CERA-20C), supporting consistency in variability where records overlap. - U.S. landfall frequency (storms ≥40 kt, each storm counted once): both observed and all downscaled series show upward linear trends over the full record significant at p=0.05. Over the 20th–21st centuries, downscaled trends remain small but significant, whereas the observed trend is neither positive nor statistically significant, consistent with prior studies. - U.S. landfall Power Dissipation Index (PDI) shows statistically significant increases in all three downscalings over the full record and also over the 20th–21st centuries, implying increasing destructive wind potential at landfall. - Globally, downscaled annual TC numbers show little net change over the last century: NOAA v2c indicates a significant trend (direction differs by reanalysis), NOAA v3 an opposing significant trend, and CERA-20C no significant trend. Downscaled global major hurricanes show significant upward trends in the two NOAA reanalyses largely before 1920; CERA-20C shows no significant trend. A second CERA-20C ensemble member yields order-unity differences in variability and trends, indicating that global trends are not robust relative to internal variability and sampling noise. Net global picture: little systematic change in TC frequency; any underlying climate trends would require far larger synthetic samples and multiple reanalysis ensemble members to detect robustly. - A pronounced Atlantic hurricane drought in the 1970s–1980s appears in observations and all three downscalings; evidence points to anthropogenic sulfate aerosols (and associated radiative effects of increased mineral dust via Sahel drought) as key contributors. Post-1990 uptick aligns with declining sulfate emissions following clean air regulations. - A marked increase in Atlantic activity during the 1920s (peaking in the 1930s) is reproduced and is consistent with elevated tropical North Atlantic SSTs during the Dust Bowl era. - Environmental controls: Atlantic GPI correlates strongly with downscaled genesis (Atlantic r: 0.84, 0.94, 0.93 for NOAA v2c, v3, CERA-20C; global r: 0.24, 0.87, 0.76 respectively). In the Atlantic Main Development Region, potential intensity anomalies explain much of the variability, with a decomposition showing regional climate variability dominates over the global component. The direct global contribution to potential intensity increases by ~0.9 m s−1 (highly significant) over the record, but regional variability is the primary driver of TC changes. - Bias assessment: Reanalysis assimilation of historical TC data would be expected to depress potential intensity (via tropospheric warming and drying) and genesis, potentially inducing negative, not positive, downscaled trends; no global step change circa 1980 is seen despite satellite-era assimilation increases. Upward trends in reanalysis surface pressure over genesis regions (~1–2 hPa since the 19th century) would, if entirely artificial, imply only ~1.2 m s−1 spurious potential intensity increase in the Atlantic—far smaller than the ~6 m s−1 actual increase—while similar artificial increases in other basins are not associated with increased downscaled activity, arguing against observational artifact as the main cause.

The downscaling results indicate that Atlantic TC activity has increased over the past 150–180 years, with a distinct mid-century lull and recent resurgence, and that early historical records likely undercount TCs. This contrasts with statistically corrected historical analyses that find no detectable trend but is consistent with the notion that those methods privilege the null hypothesis and can underestimate true trends. The findings suggest that regional climate variability—particularly changes in potential intensity driven by regional SST patterns and atmospheric conditions—dominates Atlantic TC variability, while the direct global-warming contribution is secondary but non-negligible. The 1970s–1980s hurricane drought aligns with elevated sulfate aerosol forcing and associated mineral dust radiative effects via Sahel drought; the post-1990 rebound aligns with reduced aerosols following air-quality regulation. The 1920s–1930s peak corresponds to warm tropical North Atlantic conditions during the Dust Bowl era. In contrast, global TC metrics show weak or insignificant trends and lack robustness across reanalyses and ensemble members, underscoring the special nature of Atlantic TC climatology and the strong role of regional drivers. Potential assimilation biases and observational artifacts were examined and found unlikely to explain the positive Atlantic trends. Overall, the results support a view in which Atlantic TC changes are predominantly driven by regional climate variability and aerosol forcing superimposed on a smaller global-warming signal.

This study provides an independent, dynamics-based reconstruction of historical TC activity using downscaling of three century-scale reanalyses. It finds: (i) significant increases in Atlantic TC activity since the 19th century, including stronger landfall destructive potential; (ii) a mid-century hurricane drought linked to anthropogenic aerosols; (iii) modest direct global-warming contributions with regional climate variability dominating Atlantic changes; and (iv) minimal, non-robust global trends in TC frequency and intensity metrics over the last century. These results reconcile undercounting in early observational records while revealing real long-term Atlantic increases obscured by detection limitations. Future research should: expand ensemble sizes and synthetic track counts to better isolate weak global signals; refine reanalysis inputs (e.g., historical SST and surface pressure datasets) to reduce potential artifacts; improve understanding and quantification of aerosol and dust impacts on regional TC environments; and integrate downscaling with coupled climate model projections to assess future changes in TC frequency, intensity, rainfall, and landfall risk—especially given robust thermodynamic expectations of increased TC rainfall and modeled increases in future TC activity, particularly in the Northern Hemisphere.

- Potential biases in reanalyses (e.g., evolving observation density, instrumentation changes, assimilation of historical TC data) could impart artificial trends in environmental fields; while analyses suggest these would more likely suppress than enhance downscaled trends, residual biases cannot be ruled out. - The downscaling uses monthly mean thermodynamic inputs and synthesized winds on coarse grids; it does not explicitly resolve mesoscale processes or storm–environment feedbacks beyond the axisymmetric intensity framework. - Annual frequencies are calibrated via a proportionality constant and depend on the seeding and success ratio assumptions; multiplicative rescaling to recent means may mask absolute biases. - Global trend detection is limited by random variability from finite (100) tracks per year and single/limited reanalysis ensemble members; order-unity differences between CERA-20C ensemble members indicate limited robustness for weak global signals. - Historical observational datasets (IBTrACS, ICOADS) remain incomplete and inhomogeneous, complicating validation and comparison, especially pre-1900 outside the Atlantic. - Differences among reanalysis SST datasets (SODAsi.2, SODAsi.3, HadISST2) can affect downscaled TC variability and trends; attribution across datasets remains uncertain.