We propose a statistical approach to tornadoes modeling for predicting and simulating occurrences of tornadoes and accumulated cost distributions over a time interval. This is achieved by modeling the tornadoes intensity, measured with the Fujita scale, as a stochastic process. Since the Fujita scale divides tornadoes intensity into six states, it is possible to model the tornadoes intensity by using Markov and semi-Markov models. We demonstrate that the semi-Markov approach is able to reproduce the duration effect that is detected in tornadoes occurrence. The superiority of the semi-Markov model as compared to the Markov chain model is also affirmed by means of a statistical test of hypothesis. As an application we compute the expected value and the variance of the costs generated by the tornadoes over a given time interval in a given area. he paper contributes to the literature by demonstrating that semi-Markov models represent an effective tool for physical analysis of tornadoes as well as for the estimation of the economic damages to human things.
Observations in the 171 AA channel of the Atmospheric Imaging Assembly of the space-borne Solar Dynamics Observatory show tornadoes-like features in the atmosphere of the Sun. These giant tornadoes appear as dark, elongated and apparently rotating structures in front of a brighter background. This phenomenon is thought to be produced by rotating magnetic field structures that extend throughout the atmosphere. We characterize giant tornadoes through a statistical analysis of properties like spatial distribution, lifetimes, and sizes. A total number of 201 giant tornadoes are detected in a period of 25 days, suggesting that on average about 30 events are present across the whole Sun at a time close to solar maximum. Most tornadoes appear in groups and seem to form the legs of prominences, thus serving as plasma sources/sinks. Additional Halpha observations with the Swedish 1-m Solar Telescope imply that giant tornadoes rotate as a structure although clearly exhibiting a thread-like structure. We observe tornado groups that grow prior to the eruption of the connected prominence. The rotation of the tornadoes may progressively twist the magnetic structure of the prominence until it bec
Solar magnetized "tornadoes", a phenomenon discovered in the solar atmosphere, appear as tornado-like structures in the corona but root in the photosphere. Like other solar phenomena, solar tornadoes are a feature of magnetized plasma and therefore differ distinctly from terrestrial tornadoes. Here we report the first analysis of solar "tornadoes" {Two papers which focused on different aspect of solar tornadoes were published in the Astrophysical Journal Letters (Li et al. 2012) and Nature (Wedemeyer-Böhm et al. 2012), respectively, during the revision of this Letter.}. A detailed case study of two events indicates that they are rotating vertical magnetic structures probably driven by underlying vortex flows in the photosphere. They usually exist as a group and relate to filaments/prominences, another important solar phenomenon whose formation and eruption are still mysteries. Solar tornadoes may play a distinct role in the supply of mass and twists to filaments. These findings could lead to a new explanation to filament formation and eruption.
The barbs or legs of some prominences show an apparent motion of rotation, which are often termed solar tornadoes. It is under debate whether the apparent motion is a real rotating motion, or caused by oscillations or counter-streaming flows. We present analysis results from spectroscopic observations of two tornadoes by the Interface Region Imaging Spectrograph. Each tornado was observed for more than 2.5 hours. Doppler velocities are derived through a single Gaussian fit to the Mg~{\sc{ii}}~k~2796Å~and Si~{\sc{iv}}~1393Å~line profiles. We find coherent and stable red and blue shifts adjacent to each other across the tornado axes, which appears to favor the interpretation of these tornadoes as rotating cool plasmas with temperatures of $10^4$ K-$10^5$ K. This interpretation is further supported by simultaneous observations of the Atmospheric Imaging Assembly on board the Solar Dynamics Observatory, which reveal periodic motions of dark structures in the tornadoes. Our results demonstrate that spectroscopic observations can provide key information to disentangle different physical processes in solar prominences.
Solar tornadoes are dark vertical filamentary structures observed in the extreme ultraviolet associated with prominence legs and filament barbs. Their true nature and relationship to prominences requires understanding their magnetic structure and dynamic properties. Recently, a controversy has arisen: is the magnetic field organized forming vertical, helical structures or is it dominantly horizontal? And concerning their dynamics, are tornadoes really rotating or is it just a visual illusion? Here, we analyze four consecutive spectropolarimetric scans of a prominence hosting tornadoes on its legs which help us shed some light on their magnetic and dynamical properties. We show that the magnetic field is very smooth in all the prominence, probably an intrinsic property of the coronal field. The prominence legs have vertical helical fields that show slow temporal variation probably related to the motion of the fibrils. Concerning the dynamics, we argue that 1) if rotation exists, it is intermittent, lasting no more than one hour, and 2) the observed velocity pattern is also consistent with an oscillatory velocity pattern (waves).
Tornadoes are the most violent of all atmospheric storms. In a typical year, the United States experiences hundreds of tornadoes with associated damages on the order of one billion dollars. Community preparation and resilience would benefit from accurate predictions of these economic losses, particularly as populations in tornado-prone areas increase in density and extent. Here, we use a zero-inflated modeling approach and artificial neural networks to predict tornado-induced property damage using publicly available data. We developed a neural network that predicts whether a tornado will cause property damage (out-of-sample accuracy = 0.821 and area under the receiver operating characteristic curve, AUROC, = 0.872). Conditional on a tornado causing damage, another neural network predicts the amount of damage (out-of-sample mean squared error = 0.0918 and R2 = 0.432). When used together, these two models function as a zero-inflated log-normal regression with hidden layers. From the best-performing models, we provide static and interactive gridded maps of monthly predicted probabilities of damage and property damages for the year 2019. Two primary weaknesses include (1) model fitting
(Abridge) On 13 September 2025 around 22 UTC, a localized tornado outbreak affected eastern Philippines, causing significant damage in Camarines Norte. The event developed within an atypical easterly severe weather regime characterized by warm, moist southeasterly flow and strong low-level wind shear associated with an easterly wave trough. A vorticity convergence zone along the inverted trough enhanced low-level rotation, while highly curved streamwise hodographs indicated a favorable environment for supercells and tornadogenesis. At least five vortices were identified, including three tornadic supercells. The Magang tornado was rated IF2.5 (EF3-equivalent) with $\sim$2 km damage path, while the Cahabaan and Napilihan tornadoes were rated IF1 (EF1-equivalent), with Cahabaan producing $\sim$3 km damage path. The remaining vortices were rated IF0 (EF0-equivalent). These tornadoes occurred simultaneously, indicating multiple discrete supercells within the same mesoscale environment and possible inflow-outflow interactions. Dual-polarization radar observations revealed Z$_\text{DR}$ and K$_\text{DP}$ columns, a debris signature in the Magang tornado, and a bounded weak echo region (BW
Rapid and accurate building damage assessment in the immediate aftermath of tornadoes is critical for coordinating life-saving search and rescue operations, optimizing emergency resource allocation, and accelerating community recovery. However, current automated methods struggle with the unique visual complexity of tornado-induced wreckage, primarily due to severe domain shift from standard pre-training datasets and extreme class imbalance in real-world disaster data. To address these challenges, we introduce a systematic experimental framework evaluating 79 open-source deep learning models, encompassing both Convolutional Neural Networks (CNNs) and Vision Transformers, across over 2,300 controlled experiments on our newly curated Quad-State Tornado Damage (QSTD) benchmark dataset. Our findings reveal that achieving operational-grade performance hinges on a complex interaction between architecture and optimization, rather than architectural selection alone. Most strikingly, we demonstrate that optimizer choice can be more consequential than architecture: switching from Adam to SGD provided dramatic F1 gains of +25 to +38 points for Vision Transformer and Swin Transformer families,
Magnetic tornadoes, characterized as impulsive Alfven waves initiated by photospheric vortices in intergranular lanes, are considered efficient energy channels to the corona. Despite their acknowledged importance for solar coronal heating, their observational counterparts from the corona have not been well understood. To address this issue, we use a radiative MHD simulation of a coronal loop with footpoints rooted in the upper convection zone, and synthesize the chromospheric and coronal emissions corresponding to a magnetic tornado. Considering SDO/AIA 171 A and Solar Orbiter/EUI 174 A channels, our synthesis reveals that the coronal response to magnetic tornadoes can be observed as an EUV brightening of which width is ~2 Mm. This brightening is located above the synthesized chromospheric swirl observed in Ca II 8542 A, Ca II K, and Mg II k lines, which can be detected by instruments such as SST/CRISP, GST/FISS, and IRIS. Considering the height correspondence of the synthesized brightening, magnetic tornadoes can be an alternative mechanism for the small-scale EUV brightenings such as the solar "campfires''. Our findings indicate that coordinated observations encompassing the chro
Weather radar is the primary tool used by forecasters to detect and warn for tornadoes in near-real time. In order to assist forecasters in warning the public, several algorithms have been developed to automatically detect tornadic signatures in weather radar observations. Recently, Machine Learning (ML) algorithms, which learn directly from large amounts of labeled data, have been shown to be highly effective for this purpose. Since tornadoes are extremely rare events within the corpus of all available radar observations, the selection and design of training datasets for ML applications is critical for the performance, robustness, and ultimate acceptance of ML algorithms. This study introduces a new benchmark dataset, TorNet to support development of ML algorithms in tornado detection and prediction. TorNet contains full-resolution, polarimetric, Level-II WSR-88D data sampled from 10 years of reported storm events. A number of ML baselines for tornado detection are developed and compared, including a novel deep learning (DL) architecture capable of processing raw radar imagery without the need for manual feature extraction required for existing ML algorithms. Despite not benefitin
There is a growing need to evaluate Large Language Models (LLMs) on complex, high-impact, real-world tasks to assess their true readiness as reasoning agents. To address this gap, we introduce AgentCaster, a contamination-free framework employing multimodal LLMs end-to-end for the challenging, long-horizon task of tornado forecasting. Within AgentCaster, models interpret heterogeneous spatiotemporal data from a high-resolution convection-allowing forecast archive. We assess model performance over a 40-day period featuring diverse historical data, spanning several major tornado outbreaks and including over 500 tornado reports. Each day, models query interactively from a pool of 3,625 forecast maps and 40,125 forecast soundings for a forecast horizon of 12-36 hours. Probabilistic tornado-risk polygon predictions are verified against ground truths derived from geometric comparisons across disjoint risk bands in projected coordinate space. To quantify accuracy, we propose domain-specific TornadoBench and TornadoHallucination metrics, with TornadoBench highly challenging for both LLMs and domain expert human forecasters. Notably, human experts significantly outperform state-of-the-art m