https://arxiv.org/api/X1TNQKbxwg3dhTsVPzXMkixAkSs 2026-03-22T12:04:40Z 8090 15 15 http://arxiv.org/abs/2510.04017v2 2026-03-16T19:30:40Z

Foundation models for weather science are pre-trained on vast amounts of structured numerical data and outperform traditional weather forecasting systems. However, these models lack language-based reasoning capabilities, limiting their utility in interactive scientific workflows. Large language models (LLMs) excel at understanding and generating text but cannot reason about high-dimensional meteorological datasets. We bridge this gap by building the first agentic framework for weather science. Our framework includes a Python code-based environment for agents (ZephyrusWorld) to interact with weather data, featuring tools including a WeatherBench 2 dataset indexer, geolocator for geocoding from natural language, weather forecasting, climate simulation capabilities, and a climatology module for querying precomputed climatological statistics (e.g., means, extremes, and quantiles) across multiple timescales. We design Zephyrus, a multi-turn LLM-based weather agent that iteratively analyzes weather datasets, observes results, and refines its approach through conversational feedback loops. We accompany the agent with a new benchmark, ZephyrusBench, with a scalable data generation pipeline that constructs diverse question-answer pairs across weather-related tasks, from basic lookups to advanced forecasting, extreme event detection, and counterfactual reasoning. Experiments on this benchmark demonstrate the strong performance of Zephyrus agents over text-only baselines, outperforming them by up to 44 percentage points in correctness. However, the hard tasks are still difficult even with frontier LLMs, highlighting the challenging nature of our benchmark and suggesting room for future development. Our codebase and benchmark are available at https://github.com/Rose-STL-Lab/Zephyrus.

2025-10-05T03:34:08Z Sumanth Varambally Marshall Fisher Jas Thakker Yiwei Chen Zhirui Xia Yasaman Jafari Ruijia Niu Manas Jain Veeramakali Vignesh Manivannan Zachary Novack Luyu Han Srikar Eranky Salva Rühling Cachay Taylor Berg-Kirkpatrick Duncan Watson-Parris Yi-An Ma Rose Yu http://arxiv.org/abs/2603.15439v1 2026-03-16T15:43:32Z

The two-dimensional propagation of small-amplitude waves through an infinite periodic array of freely-floating rectangular floes is considered under the assumptions of inviscid linearised wave theory. Fluid gaps between adjacent floes allow a complex interaction of the fluid with heave, surge and pitch motions. In particular, the presence of fluid resonance in the vertical channels between floes has a significant influence on wave propagation around certain critical frequencies. Bloch-Floquet theory is used and encodes the wavenumber for propagating waves into periodic boundary conditions. Solutions of the resulting boundary-value problem posed in a fundamental cell are formulated in terms of integral equations in which the three rigid body modes of the problem are treated individually. The dispersion relationship between frequency and wavenumber is expressed in terms of the vanishing of a 3 x 3 determinant which encodes the hydrodynamic coupling between the modes. Accurate numerical solutions are determined using Galerkin's method to approximate solutions to the integral equations. A particular focus of the paper is determining simple explicit approximations for the dispersion relation by assuming the gap between adjacent floes is small compared to the submerged draft of the floe. Approximations are shown to compare well to numerical results for a large range of gap sizes and some surprising results emerge for low-frequency wave propagation. This is particularly relevant to the application area that motivates this study: the modelling of wave propagation through broken ice. Supplementary Material: https://github.com/LloydDafydd/pre-prints/blob/caa889f1ee1eb51980785271bd8d27ce5213bda7/wave-propogation-fff-supp-mat.pdf

2026-03-16T15:43:32Z 22 pages, 14 figures Lloyd Dafydd Richard Porter http://arxiv.org/abs/2603.15358v1 2026-03-16T14:36:47Z

Numerical weather prediction has long been constrained by the computational bottlenecks inherent in data assimilation and numerical modeling. While machine learning has accelerated forecasting, existing models largely serve as "emulators of reanalysis products," thereby retaining their systematic biases and operational latencies. Here, we present FuXiWeather2, a unified end-to-end neural framework for assimilation and forecasting. We align training objectives directly with a combination of real-world observations and reanalysis data, enabling the framework to effectively rectify inherent errors within reanalysis products. To address the distribution shift between NWP-derived background inputs during training and self-generated backgrounds during deployment, we introduce a recursive unrolling training method to enhance the precision and stability of analysis generation. Furthermore, our model is trained on a hybrid dataset of raw and simulated observations to mitigate the impact of observational distribution inconsistency. FuXiWeather2 generates high-resolution ($0.25^{\circ}$) global analysis fields and 10-day forecasts within minutes. The analysis fields surpass the NCEP-GFS across most variables and demonstrate superior accuracy over both ERA5 and the ECMWF-HRES system in lower-tropospheric and surface variables. These high-quality analysis fields drive deterministic forecasts that exceed the skill of the HRES system in 91\% of evaluated metrics. Additionally, its outstanding performance in typhoon track prediction underscores its practical value for rapid response to extreme weather events. The FuXiWeather2 analysis dataset is available at https://doi.org/10.5281/zenodo.18872728.

2026-03-16T14:36:47Z Xiaoze Xu Xiuyu Sun Songling Zhu Xiaohui Zhong Yuanqing Huang Zijian Zhu Jun Liu Hao Li http://arxiv.org/abs/2603.15247v1 2026-03-16T13:19:32Z

Meaningful scores for forecast verification are essential for developing reliable forecasts, and there has been much effort to develop scores that align well with human perceptions of forecast quality. Whilst many of these scores have intuitive interpretations, relatively little is known about how these scores rank different forecasts, and how scores reflect forecast error. We theoretically explore the behaviour of two scores that fall within the `neighbourhood' paradigm of spatial verification; the Fractions Skill Score (FSS) and Brier Divergence Skill Score (BDnSS). We investigate how each score ranks forecasts with two types of error; errors in the mean frequency (corresponding to intensity or shape errors) and errors in the standard deviation (corresponding to errors in spatial structure, such as blurring or excess noise). We find that under many situations the FSS assigns higher scores to forecasts that over-predict mean frequency, thus theoretically confirming the need to use the FSS with percentile thresholds. Both scores assign higher scores to smoother forecasts in many situations, a reflection of the `double penalty' problem; however, we observe that size of this effect is larger for the BDnSS than the FSS, showing that the FSS under some situations is less susceptible to the double penalty problem than the BDnSS.

2026-03-16T13:19:32Z Bobby Antonio http://arxiv.org/abs/2603.15200v1 2026-03-16T12:40:30Z

Side-scan sonar (SSS) imagery is susceptible to geometric distortions caused by platform motion instability, which degrade geometric consistency and limit downstream analyses such as mosaicking and perception. Conventional correction methods typically rely on navigation and attitude measurements, which are often unreliable in real ocean conditions. This unreliability necessitates blind geometric correction from a single distorted image, a highly ill-posed problem. To address this issue, we propose a physically motivated knowledge distillation framework for blind geometric correction of SSS imagery. Specifically, a teacher network is trained using paired distorted and geocoded reference images to learn distortion-related geometric differences, and this knowledge is transferred to a student network that performs correction using only a single distorted image during blind inference. To ensure physically plausible deformation estimation, we design a parametric decoder that represents distortions as row-wise affine transformations consistent with the SSS line-scanning imaging mechanism. To compensate for the absence of reference information during blind inference, a hallucination context module is introduced to approximate the teachers geometric reasoning from distorted features under a multi-level distillation scheme. In addition, a differentiable forward warping strategy is adopted to handle the non-bijective deformation characteristics of SSS imagery in an end-to-end manner. Extensive experiments on multiple datasets show that the proposed method outperforms state-of-the-art baselines and generalizes well across different platforms and acquisition conditions.

2026-03-16T12:40:30Z Can Lei Hayat Rajani Valerio Franchi Rafael Garcia Nuno Gracias Huigang Wang Wei Qiang http://arxiv.org/abs/2603.18044v1 2026-03-16T12:36:22Z

In complex systems, events occur at irregular intervals that inherently encode the underlying dynamics of the system. Analyzing the temporal clustering of these events reveals critical insights into the non-random patterns and the temporal evolution. Existing techniques can effectively quantify the overall clustering tendency of events using global statistical measures. However, these macroscopic approaches leave a critical gap, as they do not attempt to investigate the dynamics of individual clusters. Analyzing individual clusters is essential, as it helps comprehend the local interactions that actively drive the system dynamics, which may be obscured by global averaging, while simultaneously revealing the time scales involved. To address these limitations, we propose a complex network-based framework for analyzing clustering of events occurring at irregular intervals. The framework establishes connections using arrival times, transforming the time series into a network. Network properties are then used to quantify the clustering. Further, a community detection algorithm is used to identify individual clusters in time series. We illustrate the method by applying it to standard arrival processes, such as the Poisson process and the Markov-modulated Poisson process. To further demonstrate its scope, we apply the method to two diverse systems: the time series of droplet arrivals in turbulent flows and the R-R intervals in electrocardiogram (ECG) signals.

2026-03-16T12:36:22Z 33 pages, 17 figures Ambedkar Sanket Sukdeo K. Shri Vignesh Sachin S. Gunthe T Narayan Rao Amit Kumar Patra R. I. Sujith http://arxiv.org/abs/2603.15127v1 2026-03-16T11:24:03Z

Accurate prediction of tropical cyclones remains a major challenge for both numerical weather prediction and emerging artificial intelligence weather prediction systems. While recent global AI models have demonstrated strong skill in large-scale circulation prediction, they often struggle to represent the mesoscale structures critical for tropical cyclone intensity and precipitation. Here we develop the Hybrid Intelligent Typhoon System (HITS), a regional AI forecasting framework for medium-range typhoon prediction over the Asia-Pacific region, trained on a newly constructed 9-km high-resolution typhoon reanalysis dataset. The model combines regional autoregressive prediction with large scale dynamical constraints from the state-of-the-art ECMWF Artificial Intelligence Forecasting System (AIFS), allowing it to remain dynamically consistent with the evolving large-scale circulation while resolving mesoscale structures. HITS is further extended with a structure-aware perceptual training strategy (HITS-LPIPS) that improves the representation of convective and typhoon rainband structures. Experiments show that the hybrid framework substantially improves precipitation structure and typhoon intensity forecasts compared with both purely autoregressive regional AI models and standalone AI downscaling approaches. In particular, HITS-LPIPS reduces intensity errors by up to 47.8% relative to AIFS at a 72 hour lead time and produces a near-unbiased wind-pressure relationship for simulated typhoons. These results demonstrate that dynamically constrained regional AI systems provide a promising pathway for improving medium-range typhoon prediction.

2026-03-16T11:24:03Z 17 pages;6 figures Zeyi Niu Wei Huang Sirong Huang Zhuo Wang Mu Mu Mengqi Yang Xinhai Han Haofei Sun Zhaoyang Huo Bo Qin http://arxiv.org/abs/2510.09530v4 2026-03-16T08:31:38Z

The environments of severe thunderstorms that produced hail were examined using 171 proximity soundings (2005-2024) archived in the 3rd Data Release of Project SWAP. These soundings were categorized based on their geographical occurrence into three hail-prone environments across Luzon, Philippines. Key parameters describing instability, vertical wind shear, and moisture were calculated to assess the environmental conditions for hail production. The probability of hail occurrence, expressed as a function of W$_{\text{MAX}}$ ($\sqrt{2 \times \text{CAPE}}$) and 0-6 km bulk shear (DLS), revealed patterns distinct from those reported in other regions. Hail events in Luzon were most likely under high CAPE conditions, where boundary-layer moisture was sufficient, mid- and low-level lapse rates were steep, and lifting condensation levels were high. Surprisingly, weak DLS was common across Luzon hail environments, diverging from existing severe weather climatologies, yet large DCAPE indicated environments conducive to damaging wind events. When DLS was replaced with the shear magnitude between the cloud base and equilibrium level, the probability of hail occurrence increased, better aligning with global severe weather climatologies. This finding is supported by hodograph analyses, which show largely unidirectional wind profiles: strong speed shear aloft but weak directional shear in the low-levels. Parameters such as W$_{\text{MAX}}$SHEAR, W$_{\text{MAX}}$SHEAR$_{\text{LCL-EL}}$, and BWD$_{\text{LCL-EL}}$ emerge as potential discriminators between non-severe and severe thunderstorms capable of producing hail, and as useful metrics for assessing convective storm severity in Luzon and possibly countrywide. Finally, two recurring severe setups conducive to hail were identified: (1) an easterly regime associated with trade winds, and (2) a westerly regime linked to the Asian summer monsoon.

2025-10-10T16:43:51Z 42 pages, 18 figures, 2 tables. Accepted and Published to Annals of Geophysics (AG; Atmospheric Sciences) Annals of Geophysics, 69, 1, A105, 2026 Generich H. Capuli 10.4401/ag-9484 http://arxiv.org/abs/2510.08107v3 2026-03-15T13:46:50Z

Persistent systematic errors in Earth system models (ESMs) arise from difficulties in representing the full diversity of subgrid, multiscale atmospheric convection and turbulence. Machine learning (ML) parameterizations trained on short high-resolution simulations show strong potential to reduce these errors. However, stable long-term atmospheric simulations with hybrid (physics + ML) ESMs remain difficult, as neural networks (NNs) trained offline often destabilize online runs. Training convection parameterizations directly on coarse-grained data is challenging, notably because scales cannot be cleanly separated. This issue is mitigated using data from superparameterized simulations, which provide clearer scale separation. Yet, transferring a parameterization from one ESM to another remains difficult due to distribution shifts that induce large inference errors. Here, we present a proof-of-concept where a ClimSim-trained, physics-informed NN convection parameterization is successfully transferred to ICON-A. The scheme is (a) trained on adjusted ClimSim data with subtracted radiative tendencies, and (b) integrated into ICON-A. The NN parameterization predicts its own error, enabling mixing with a conventional convection scheme when confidence is low, thus making the hybrid AI-physics model tunable with respect to observations and reanalysis through mixing parameters. This improves process understanding by constraining convective tendencies across column water vapor, lower-tropospheric stability, and geographical conditions, yielding interpretable regime behavior. In AMIP-style setups, several hybrid configurations outperform the default convection scheme (e.g., improved precipitation statistics). With additive input noise during training, both hybrid and pure-ML schemes lead to stable simulations and remain physically consistent for at least 20 years.

2025-10-09T11:44:47Z Helge Heuer Tom Beucler Mierk Schwabe Julien Savre Manuel Schlund Veronika Eyring http://arxiv.org/abs/2603.14253v1 2026-03-15T07:20:22Z

This study develops a novel framework within the Weather Research and Forecast Model for modeling aerosol-cloud-lightning interactions. The framework explicitly represents aerosol-cloud interactions by prescribing aerosols with two configurations: an idealized setup, where both cloud condensation nuclei (CCN) and ice nucleating particles (IN) are assumed to have a single chemical composition and spatially uniform distributions; and a quasi-realistic configuration, with multi-species aerosols assigned spatially varying distributions, where hygroscopic components act as CCN, dust particles act as IN, and all aerosol species influence radiative transfer. Cloud microphysics is coupled with detailed charge separation and discharge processes to enable the lightning simulation. The framework is evaluated using two thunderstorms in Guangdong, China. For an isolated storm, the model successfully reproduces the observed tripolar charge structure (positive-negative-positive), demonstrating its capability in simulating cloud electrification. For a frontal storm, it captures well the observed precipitation and lightning, and shows that increasing CCN suppresses the rainfall while enhancing the lightning. Higher CCN concentrations produce more numerous but smaller cloud droplets, which suppresses the coalescence into rain droplets, allows a greater number of droplets to loft into the upper troposphere, and forms more but smaller cloud ice particles. This boosts graupel-ice collisions, intensifies non-inductive charging, strengthens the upper positive charge and the vertical electric-field gradient, ultimately increasing the lightning frequency. In contrast, no significant aerosol-induced invigoration of updrafts is observed. These results highlight the dominant role of aerosol microphysical effects over dynamical invigoration in modulating thunderstorm electrification and lightning activity.

2026-03-15T07:20:22Z 27 pages, 10 figures Weishan Wang Guoxing Chen Yijun Zhang Jen-Ping Chen Dong Zheng Liangtao Xu http://arxiv.org/abs/2603.13902v1 2026-03-14T11:30:16Z

An analysis was conducted of broadband sound pulses received by a vertical array in Lake Kinneret (Israel). For most frequencies within the pulse frequency bands, the array is sparse. The application of the autoproduct transform made it possible to approximately reconstruct the signals that would be received after the emission of pulses at low frequencies for which the array is dense. Using the coherent state method developed in quantum theory, a transition has been made from representing the reconstructed field as a function of depth and time to its distribution in the 'depth-angle-time' phase space. Due to the absence of multipath, the intensity distribution in this space should be weakly sensitive to variations in environmental parameters. In accordance with this expectation, the distribution found is close to the result of its calculation using an idealized (range-independent) waveguide model. It has been shown that this intensity distribution can be used as input data for a neural network when solving the problem of sound source localization in an underwater waveguide. In the examples considered, the neural network is trained on synthetic data, i.e., data obtained from theoretical calculations.

2026-03-14T11:30:16Z A. L. Virovlyansky http://arxiv.org/abs/2603.13899v1 2026-03-14T11:20:42Z

A method for calibrating a monopole sound source in a water tank with reflective side walls and bottom is considered. The idea of the method is based on the phenomenon of anomalous transparency of the water-air boundary for a sound source located at a shallow depth. This boundary plays the role of a filter that prevents waves reflected from the side walls and bottom from entering the air. For a shallow source, the field in the air will be approximately the same as for a source located at the same depth in a homogeneous water half-space. This field is described by a well-known analytical formula that makes it possible to estimate the source strength in water based on the sound intensity level measured in air.

2026-03-14T11:20:42Z A. L. Virovlyansky M. S. Deryabin A. A. Prokhorov A. Yu. Kazarova V. K. Bakhtin http://arxiv.org/abs/2602.19494v2 2026-03-13T17:35:03Z

We develop a trajectory-based Koopman method for sea surface temperature (SST) that lifts annual SST segments using a signature kernel -- a reproducing kernel Hilbert space (RKHS) kernel that compares paths via iterated-integral features -- and learns the one-year shift operator. By operating on annual trajectory segments rather than instantaneous fields, the method encodes finite-time history, which helps capture memory effects in SST-only evolution. The resulting operator improves out-of-sample multi-year forecast skill relative to a climatology baseline and reveals coherent spectral modes. We implement the approach via kernel extended dynamic mode decomposition (EDMD) on signature-kernel Gram matrices, yielding a single pipeline for forecasting and spectral diagnostics of high-dimensional SST dynamics.

2026-02-23T04:28:39Z 21 pages, 6 figures Nozomi Sugiura Satoshi Osafune Shinya Kouketsu http://arxiv.org/abs/2603.07227v2 2026-03-13T15:44:40Z

We quantify changes DeltaQ in 100-year return values for regional annual maxima and minima of near-surface atmospheric temperature from output of five CMIP6 models, for five of the Earth's desert regions, over the interval (2025,2125). We use generalised extreme value (GEV) regression to characterise changes in extremes, considering a range of different parametric forms for the variation of GEV parameters with time, and coupling models for different scenarios so that they provide a common GEV tail in the first year of observation. Parameters are estimated using Bayesian inference. We perform a simulation study using ground truth models generating data qualitatively similar to the CMIP6 output, to assess the relative performance of different information criteria in selecting models from a set of candidates, to minimise error in predictions of DeltaQ. The Bayesian information criterion (BIC) provides best performance, out-performing the divergence and widely-applicable information criteria in particular. Using BIC-selected GEV regression models, we estimate joint posterior distributions of DeltaQ over three forcing scenarios, for different combinations of region, GCM and climate ensemble. Estimates show a consistent trend across regions, GCMs and climate ensembles, of DeltaQ increasing with climate scenario for both regional annual maxima and minima. Aggregating posterior distributions over climate ensembles and GCMs, we find evidence for significant increases in DeltaQ for regional annual maxima under more severe forcing scenarios for all desert regions. Similar but weaker and less significant trends are observed for regional annual minima.

2026-03-07T14:17:48Z Callum Leach Kevin Ewans Philip Jonathan http://arxiv.org/abs/2603.12515v1 2026-03-12T23:24:04Z

Earth's climate stability, characterized by the global radiative feedback parameter ($λ$), varies decadally due to changing surface temperature patterns. Recent variations in $λ$ are poorly understood as coordinated model simulations typically end in 2014. We apply a convolutional neural network trained on climate model simulations to observation-based surface temperature reconstructions to estimate variations in $λ$ up to 2025. We find that $λ$ reached a minimum (maximum stability) around the mid 1990s ($λ\simeq\SI{-3}{Wm^{-2}/K}$), but has since weakened significantly ($λ\simeq\SI{-2}{Wm^{-2}/K}$). We confirm these results with climate model simulations extended to 2022. The recent $λ$ weakening is not significantly affected by El Niño Southern Oscillation or Pacific Decadal Oscillation. Attribution reveals that warming in the subtropical Northeast Pacific is an important driver of the recently weakened feedback, confirmed by targeted experiments in E3SMv2. Our approach enables near real-time monitoring of Earth's climate stability.

2026-03-12T23:24:04Z 7 pages, 3 figures; supplemental information (9 figures, 3 tables) Senne Van Loon Maria Rugenstein Mark D. Zelinka Timothy Andrews