Huracán: Intense Tropical Cyclone Dynamics

• Huracán is a term for intense tropical cyclones defined by sustained hurricane-force winds and characterized by warm-core rotation, vortex dynamics, and complex wind-field modeling.
• Advanced forecasting approaches, including observation-driven ensemble systems with neural network components, improve prediction of hurricane tracks and intensification.
• Exposure analyses quantify global person-day impacts, integrating wind thresholds, sociodemographic indices, and climate-risk modeling to assess hurricane hazards.

Huracán, conventionally rendered in English as hurricane, is the regional name for the most intense class of tropical cyclone. In meteorological usage it denotes a warm-core, rotating storm whose sustained winds reach the hurricane-force threshold on the Saffir–Simpson scale; in contemporary research the term also indexes a wide technical literature on exposure analysis, vortex dynamics, statistical and machine-learning forecast systems, boundary-layer turbulence, and climate-risk modeling. A distinct modern proper-name usage, "Huracan," refers to an observation-driven end-to-end ensemble weather-prediction system (Jiang et al., 24 Oct 2025, Ni et al., 25 Aug 2025).

1. Definition, thresholds, and formation conditions

A huracán is defined meteorologically by the sustained-wind speed in a storm’s core circulation. Using the Saffir–Simpson wind scale for consistency with International Best Track Archive for Climate Stewardship (IBTrACS) definitions, the operative thresholds are as follows.

Wind regime	Threshold
Tropical storm–force	$V \ge 17.5\ \mathrm{m/s}$ $(\ge 63\ \mathrm{km/h}\approx 34\ \text{knots})$
Hurricane–force	$V \ge 32.9\ \mathrm{m/s}$ $(\ge 119\ \mathrm{km/h}\approx 64\ \text{knots})$; $V \ge 74\ \mathrm{mph}$

In the exposure framework used for global assessment, $V$ is the 1-minute sustained surface wind speed at the population-centroid of each second-level administrative unit (ADM2), modeled with the Willoughby parametric wind-field model as storms pass within 250 km. The onset of gale-force or tropical-cyclone-force wind is marked at $17.5\ \mathrm{m/s}$, while hurricane-force conditions begin at $32.9\ \mathrm{m/s}$ (Jiang et al., 24 Oct 2025).

Research on storm formation identifies a standard environmental set of preconditions. Hurricanes require sea-surface temperatures above a threshold often taken near $26.5\,^\circ\mathrm{C}$, a convectively unstable troposphere, ample mid-level moisture, weak vertical wind shear, and some initial vorticity or seed disturbance. In seasonal diagnostics, vertical shear is often represented by

$\Delta V = V_{200} - V_{850},$

where $(\ge 63\ \mathrm{km/h}\approx 34\ \text{knots})$0 and $(\ge 63\ \mathrm{km/h}\approx 34\ \text{knots})$1 are the environmental wind vectors at 200 hPa and 850 hPa respectively. The 2025 Atlantic retrospective further states that anomalies exceeding about $(\ge 63\ \mathrm{km/h}\approx 34\ \text{knots})$2 in sea-surface temperature can boost potential intensity (Powell, 28 Nov 2025).

2. Global exposure, historical rank, and sociodemographic structure

A recent global assessment of the 2024 calendar year evaluated tropical-cyclone exposure against the historical period 1980–2024. It used IBTrACS v4 storm tracks, the Willoughby wind-field model for daily peak sustained winds at ADM2 population centroids, Global Human Settlement Layer population grids, and the Global Gridded Relative Deprivation Index. Exposure was measured in person-days:

$(\ge 63\ \mathrm{km/h}\approx 34\ \text{knots})$3

with temporal trends estimated by

$(\ge 63\ \mathrm{km/h}\approx 34\ \text{knots})$4

In 2024, IBTrACS recorded 94 named storms. Tropical cyclone-force wind affected an estimated 429,902,820 people, corresponding to 5.5% of global population, and hurricane-force wind affected an estimated 59,672,600 people, corresponding to 0.8%. Total exposure reached 507,926,767 tropical-cyclone-force person-days and 60,228,900 hurricane-force person-days.

Region	TC-force exposed	H-force exposed
Global	429,902,820 (5.5%)	59,672,600 (0.8%)
Western Pacific	298,701,690 (6.9%)	49,645,170 (1.1%)
South-East Asia	118,107,940 (8.2%)	—
Americas	75,697,540 (4.0%)	9,891,650 (0.5%)
Europe	8,444,540 (1.7%)	—
Africa	6,973,980 (0.6%)	692,080 (0.06%)

Historically, 2024 tropical cyclone-force exposure ranked 12th highest since 1980, while hurricane-force exposure ranked 10th. Linear trends over 1980–2024 were positive in absolute terms, with $(\ge 63\ \mathrm{km/h}\approx 34\ \text{knots})$5 tropical-cyclone-force person-days per year and $(\ge 63\ \mathrm{km/h}\approx 34\ \text{knots})$6 hurricane-force person-days per year, both with $(\ge 63\ \mathrm{km/h}\approx 34\ \text{knots})$7. After normalization by global population, however, per-capita exposure showed no significant trend ($(\ge 63\ \mathrm{km/h}\approx 34\ \text{knots})$8). This suggests that population growth and urban expansion are important mediators of changing exposure burdens.

The sociodemographic distribution of exposure was heterogeneous. Globally, 56.5% of tropical-cyclone-exposed areas fell in the top two Global Gridded Relative Deprivation Index quartiles, whereas hurricane-exposed areas were disproportionately less deprived, with 54.1% in the bottom two quartiles. Urban or peri-urban ADM2s, defined by population density greater than 1,500 cap/km², accounted for 62% of hurricane-force exposure but 48% of tropical-cyclone-force exposure. Areas with higher child dependency ratio and infant mortality rate experienced 20% greater per-capita exposure, controlling for storm frequency ($(\ge 63\ \mathrm{km/h}\approx 34\ \text{knots})$9). Regionally, South-East Asia’s 118 M person-days of tropical-cyclone-force exposure occurred in ADM2s with 88.9% in moderately high or high deprivation, while the Western Pacific’s 49.6 M hurricane-force person-days were largely concentrated in low-deprivation urban centers (73.6%) (Jiang et al., 24 Oct 2025).

3. Dynamical structure, helicity, and rotating stratified turbulence

In fluid-dynamical analyses, hurricane structure is often discussed in terms of helicity, vorticity, rotation, stratification, and wave–eddy coupling. Helicity is defined by

$V \ge 32.9\ \mathrm{m/s}$0

where $V \ge 32.9\ \mathrm{m/s}$1 is the velocity field and $V \ge 32.9\ \mathrm{m/s}$2 the vorticity. Physically, it measures corkscrew motion or mirror-symmetry breaking. In atmospheric practice, the lower-tropospheric analogue is frequently storm-relative helicity; values of order $V \ge 32.9\ \mathrm{m/s}$3 or more are associated with rotating updrafts in supercells and with hurricane spiral-band structure.

Direct numerical simulations of decaying rotating stratified turbulence in the Boussinesq framework show that helicity can be generated spontaneously through the joint action of buoyancy and rotation. Under a quasi-linear balance, the horizontal helicity density satisfies

$V \ge 32.9\ \mathrm{m/s}$4

or equivalently

$V \ge 32.9\ \mathrm{m/s}$5

For $V \ge 32.9\ \mathrm{m/s}$6, the amount of helicity produced is correctly predicted by this balance equation. Outside that regime, helicity production remains persistent up to $V \ge 32.9\ \mathrm{m/s}$7, Reynolds number $V \ge 32.9\ \mathrm{m/s}$8, $V \ge 32.9\ \mathrm{m/s}$9, and $(\ge 119\ \mathrm{km/h}\approx 64\ \text{knots})$0.

The hurricane relevance of this result is explicit in the same work. Typical tropical-latitude values are stated as $(\ge 119\ \mathrm{km/h}\approx 64\ \text{knots})$1 and $(\ge 119\ \mathrm{km/h}\approx 64\ \text{knots})$2, giving $(\ge 119\ \mathrm{km/h}\approx 64\ \text{knots})$3–$(\ge 119\ \mathrm{km/h}\approx 64\ \text{knots})$4. Although that lies beyond the simulated range, the persistence of helicity generation up to $(\ge 119\ \mathrm{km/h}\approx 64\ \text{knots})$5 suggests that related wave–eddy couplings, modified by moisture, latent heating, and shear, can sustain large-scale helicity in real hurricanes. The source term $(\ge 119\ \mathrm{km/h}\approx 64\ \text{knots})$6 links helicity production to vertical velocity–buoyancy-shear correlations, including inertial–gravity wave motions in the eye and rainbands. Observed hurricane storm-relative helicity in the lowest 3 km often exceeds 150–300 $(\ge 119\ \mathrm{km/h}\approx 64\ \text{knots})$7 and is described there as correlated with intensification rates (Marino et al., 2012).

4. Statistical, stochastic, and neural representations

Modern hurricane research uses a wide range of data-driven models to represent track evolution, intensity change, and synthetic-event generation. One line of work employs recurrent neural networks trained on HURDAT2. A deep-learning study using 1,665 North Atlantic storms represented each six-hour record by 86 input features and compared Many-To-One with Many-To-Many LSTM architectures. Its best six-hour model, M2M$(\ge 119\ \mathrm{km/h}\approx 64\ \text{knots})$8, achieved mean 6 h MAE $(\ge 119\ \mathrm{km/h}\approx 64\ \text{knots})$9 km on held-out test storms; best 12 h MAE was $V \ge 74\ \mathrm{mph}$0–72 km, and best 24 h MAE $V \ge 74\ \mathrm{mph}$1 km. Excluding the 81-dimensional displacement-probability vector increased 6 h error by 11–16 km and 12 h error by 17–23 km, while Many-To-Many models reduced the iterative error accumulation that degraded Many-To-One multi-step forecasts (Bose et al., 2021).

A related grid-based recurrent neural-network approach represented Atlantic hurricane motion on a $V \ge 74\ \mathrm{mph}$2 grid of 7,256 blocks and forecast six-hourly motion from latitude, longitude, wind speed, pressure-derived variables, direction, distance, and grid index. The system reported MSE $V \ge 74\ \mathrm{mph}$3 and RMSE $V \ge 74\ \mathrm{mph}$4 on train and test data, generated forecasts up to approximately 120 hours, and was described as competitive with methods employed by the National Hurricane Center (Alemany et al., 2018).

Intensity evolution has also been formalized probabilistically. The Markov environment-dependent hurricane intensity model (MeHiM) treats six-hour intensity change as a hidden Markov process with three unobserved states of intensification and transition probabilities conditioned on environmental variables including potential intensity, vertical wind shear, relative humidity, and ocean feedback. In North Atlantic development data spanning 555 storms from 1979–2014, MeHiM reproduced the observed rapid-intensification frequency ($V \ge 74\ \mathrm{mph}$5 kt per 24 h) more faithfully than comparison models: observed 24.9%, OLS 31.0%, finite-mixture regression 20.9%, and MeHiM 25.6% (Jing et al., 2018).

Synthetic catalogs address a different problem: rare-event sampling. HurriCast combines ARIMA, K-Means, and an autoencoder, using ARIMA$V \ge 74\ \mathrm{mph}$6 for temporal count prediction, $V \ge 74\ \mathrm{mph}$7 spatial clusters, and an autoencoder with latent perturbation to generate plausible tracks; on a $V \ge 74\ \mathrm{mph}$8 density grid, the cell-wise Pearson correlation between real and simulated counts was reported as $V \ge 74\ \mathrm{mph}$9 (Gao et al., 2023). WHITS, by contrast, is a non-parametric semi-Markov simulator fit basin by basin to IBTrACS, with transitions conditioned on position, age, forward vector, and local wind speed. Its 10,000-year global synthetic catalog reproduces observed track density and annual hurricane/typhoon-force wind-hit probability across all basins, and precomputing transition probabilities enables generation of 10,000 synthetic years in under 24 h per basin on a multi-core workstation (Nakamura et al., 19 May 2026).

5. Hazard assessment, topography, and infrastructure loading

Hurricane hazard is not limited to the storm core. A recent Hurricane Impact Index for Central America decomposes impact into direct and indirect components on a $V$0 grid at six-hour resolution. The direct effect is restricted to cells within 500 km of the hurricane center and above the event-specific 90th percentile of vorticity, whereas the indirect effect targets cells outside the direct region whose wind-vector direction lies within $V$1 of the principal axis of the closest mountain range above 500 m elevation. Applied to Otto (2016), Nate (2017), Julia (2022), Bonnie (2022), Iota (2020), and Eta (2020), the stacked regional index showed highest values along the Pacific coast of Nicaragua, Costa Rica, and Panama. Hurricane Nate exhibited an indirect impact about five times its direct impact, and Costa Rican spatial patterns aligned well with DesInventar DataCards losses (Camacho et al., 10 Jun 2025).

The engineering literature emphasizes the mismatch between standard offshore design assumptions and major-hurricane eyewall flow. Large-eddy simulations of an idealized Category 5 hurricane showed that the International Electrotechnical Commission Class I thresholds of a 10-minute mean wind of 50 $V$2 and a 3-second gust of 70 $V$3 are exceeded in the eyewall over a depth of 0–200 m. Mean winds exceeded 50 $V$4 from roughly 7 km to 22 km radius, gusts exceeded 70 $V$5 over a similar band and sometimes topped 115 $V$6, gust factors reached 1.7 near the eye–eyewall interface, and vertical veer across the rotor layer reached 15–50 degrees. The same analysis argued that turbines in eyewall flow would require yaw-rate capabilities exceeding about $V$7 (Worsnop et al., 2016).

A separate non-stationary hurricane boundary-layer LES, driven by asymmetric gradient-wind forcing and thermodynamic profiles derived from proxy soundings, was validated against field measurements during Hurricanes Harvey and Irma. For Harvey, the simulated 3-s gust at 10 m reached 62.4 $V$8, close to the observed 61.3 $V$9; for Irma, the simulated 3-s gust at 15 m reached 53.5 $17.5\ \mathrm{m/s}$0, close to the observed 54.2 $17.5\ \mathrm{m/s}$1. The same model reproduced turbulence intensities of around 20% in Harvey and 26% in Irma, and its simulated wind spectra in the longitudinal and lateral directions agreed well with observations (Ma et al., 2023).

6. Climate scaling, intervention proposals, and extraterrestrial extension

Attempts to generalize hurricane occurrence statistically include a universal hurricane frequency function in which genesis frequency per month per $17.5\ \mathrm{m/s}$2 cell is expressed as

$17.5\ \mathrm{m/s}$3

with

$17.5\ \mathrm{m/s}$4

$17.5\ \mathrm{m/s}$5, $17.5\ \mathrm{m/s}$6, and $17.5\ \mathrm{m/s}$7. In that framework, a fixed increase in sea-surface temperature produces larger percentage increases in hurricane frequency at higher latitudes and lower baseline temperatures, because $17.5\ \mathrm{m/s}$8 for small $17.5\ \mathrm{m/s}$9 (Ehrlich, 2010).

Not all research in this area is descriptive. One intervention study proposed that hurricane intensity might be reduced before landfall by cooling the hurricane track through ocean mixing with cold deep water. For an upper warm layer of thickness $32.9\ \mathrm{m/s}$0 and a cold deep layer $32.9\ \mathrm{m/s}$1, the minimal work per unit area was given as approximately $32.9\ \mathrm{m/s}$2; for a swath $32.9\ \mathrm{m/s}$3, the required work was estimated as $32.9\ \mathrm{m/s}$4 to cool by about $32.9\ \mathrm{m/s}$5. The same paper stated a coefficient of performance $32.9\ \mathrm{m/s}$6, argued that about ten Shark-class submarines operating for ten hours could supply the needed energy, and derived the damage scaling $32.9\ \mathrm{m/s}$7, so that a 20 percent reduction in maximal wind speed yields $32.9\ \mathrm{m/s}$8. These claims remain a proposal rather than an operational practice (Sirovich, 2020).

The concept of hurricane has also been extended beyond Earth. High-resolution simulations of tidally locked terrestrial planets found that hurricanes can form on some such planets but not on all of them. Faster rotation and hotter conditions near the inner edge of the habitable zone favored more numerous and stronger storms; for rotation periods $32.9\ \mathrm{m/s}$9 days and substellar temperatures of about 310–315 K, genesis was much more active than in cooler or more slowly rotating cases. No hurricanes formed in $26.5\,^\circ\mathrm{C}$0- or He-dominated atmospheres because moist convection was inhibited by mean-molecular-weight effects, whereas Earth-based hurricane theory was described as applicable only when atmospheric compositions are similar to Earth’s (Yan et al., 2020).

7. Huracan as an observation-driven ensemble forecasting system

As a proper name, Huracan designates an end-to-end data-driven system for ensemble data assimilation and medium-range weather prediction. It mirrors a classical numerical weather-prediction workflow by combining an Ensemble Data Assimilation model with an Ensemble Forecast model, but it is observation-driven rather than analysis-driven. Both components share a modified Spherical Fourier Neural Operator backbone in which one of the original multilayer perceptrons is replaced by a Swin-Transformer layer, and the learned spectral filters are compressed by a 12× parameter reduction through representation as linear combinations of $26.5\,^\circ\mathrm{C}$1 basis filters. In inference, the assimilation model produces 48 ensemble initial conditions from raw observations, and the forecast model advances each member for up to 10 days in 6 h increments.

The system was evaluated on a 2024 test set of 100 issue dates and 10 lead days against ECMWF ENS and a hybrid baseline. Its central reported result is that Huracan matches or exceeds the continuous ranked probability score of ECMWF ENS on 526 of 697 variable-by-lead combinations, corresponding to 75.4% of cases. It outperformed particularly on temperature and humidity at virtually all levels, and on wind at the surface and in the upper troposphere, while lagging slightly on geopotential and mean-sea-level pressure in the mid-troposphere at 500 hPa. The observation streams ingested comprised microwave sounders, hyperspectral infrared sounders compressed to 32 latent channels, geostationary imagers, radiosondes, and surface stations. Reported limitations included some under-performance in mid-troposphere geopotential and pressure, some overfitting, and ensemble under-dispersion at intermediate lead times (Ni et al., 25 Aug 2025).