39 papers
This study demonstrates that a machine learning-based mesh movement approach (UM2N) integrated into the Thetis software significantly accelerates and robustly enhances the accuracy of non-hydrostatic tsunami simulations for probabilistic coastal hazard assessment.
This paper introduces a novel hybrid ensemble forecasting framework that uses spectral nudging to integrate machine-learned large-scale guidance with physics-based mesoscale dynamics, resulting in significant forecast skill improvements of up to two days in the tropics and enhanced tropical cyclone track predictions without compromising storm intensity or ensemble spread.
The paper introduces DAISI, a scalable data assimilation algorithm that leverages flow-based generative models with a novel inverse-sampling step to integrate forecast information and guide conditional sampling, enabling accurate filtering in complex, nonlinear systems where traditional Gaussian-based methods fail.
This paper argues that while AI promises to revolutionize climate information, its current reliance on Global North-dominated infrastructure and biased data risks exacerbating global inequality, necessitating a shift toward data-centric development, shared digital public infrastructure, and co-produced knowledge to ensure equitable outcomes.
This paper presents an open-source dynamical box model to simulate global abiotic sulfur cycling on Earth-like planets, revealing that the absence of life would result in marine sediment sulfate concentrations two orders of magnitude higher and sulfide concentrations four orders of magnitude lower than on present-day Earth.
This paper introduces EnsAI, an AI-based emulator that generates atmospheric chemical ensembles 3,300 times faster than the traditional GEM-MACH model while accurately reproducing meteorology-dependent features and delivering comparable emissions inversion results.
This study presents the first definitive eight-sigma detection of the C3 molecule in Titan's atmosphere using ultra-high-resolution VLT-ESPRESSO observations, confirming a column density consistent with photochemical models and demonstrating the efficacy of exoplanet research techniques for Solar System targets.
This paper proposes that hydrodynamic outflows driven by hydrogen recombination in the extended proto-lunar disk atmosphere, distinct from Earth's compact carbon-dominated atmosphere, effectively removed volatile elements from the Moon's precursor material, thereby explaining the observed volatile depletion in lunar rocks relative to Earth.
This paper proposes a new ensemble filtering approach that inherently localizes the analysis probability density function via variational Bayesian optimization and marginal partitioning, thereby eliminating the need for ad-hoc localization techniques while achieving accuracy and computational efficiency comparable to tuned standard EnKF and ETKF methods.
This paper introduces Climate Downscaling with Stochastic Interpolants (CDSI), a cost-effective, data-driven method that leverages stochastic interpolants to efficiently transform coarse global climate model outputs into accurate high-resolution regional projections, thereby enabling broader ensemble simulations and improved uncertainty quantification compared to traditional Regional Climate Models.
This study develops a convolutional neural network framework to automatically detect and statistically characterize ripple-scale gravity wave structures in mesospheric airglow imagery, thereby advancing the understanding of instability-driven atmospheric dynamics and demonstrating the potential of deep learning in scientific research.
This study demonstrates that enhanced friction and turbulent mixing over Jamaica's mountainous terrain were the primary drivers of Hurricane Melissa's (2025) rapid decay, accounting for a significant portion of the observed wind and energy loss, while additional asymmetric and thermodynamic processes further amplified the weakening.
This study utilizes a hierarchy of global climate models to demonstrate that while extratropical near-surface extreme winds robustly intensify with surface warming, regional projections remain highly uncertain due to inter-model differences in representing the physical dynamics and seasonality of extreme-producing weather systems.
Using a novel lag-linear model derived from energy budget mechanisms, this study demonstrates that ENSO strength will initially increase due to enhanced ocean stratification before declining as the Walker circulation slows, revealing that faster emissions lead to stronger peak variability even with identical total emissions.
This report evaluates the accuracy and consistency of high-frequency ocean wave energy measurements across four buoy types, revealing that Datawell Waverider buoys systematically report higher energy levels than other models, a finding that holds true regardless of the specific metric or evaluation method used.
This paper extends a theoretical model of wave attenuation through broken floating ice of random thickness to finite water depths, utilizing multiple scales analysis to derive an explicit attenuation expression that predicts an eighth-power frequency dependence at low frequencies and shows strong agreement with numerical simulations and field measurements.
This paper presents experimental evidence and a supporting theoretical model demonstrating that the interaction between surface waves and ambient sub-surface turbulence generates a near-surface Eulerian-mean flow that opposes and partially cancels the Stokes drift, thereby significantly altering the vertical redistribution of momentum and the transport of water-borne materials in the ocean.
This paper introduces Latent Data Assimilation (LDA), a machine learning framework that performs Bayesian data assimilation within an autoencoder-learned latent space to effectively capture nonlinear physical constraints and produce physically consistent, high-quality atmospheric analyses that outperform traditional model-space methods.
The paper introduces HLOBA, a hybrid-ensemble data assimilation method operating in a learned latent space that simultaneously achieves high accuracy, computational efficiency, and robust uncertainty quantification by fusing model forecasts and observations through a Bayesian update.