Crowdsourcing the Frontier: Advancing Hybrid Physics-ML Climate Simulation via a $50,000 Kaggle Competition

2026-03-08T21:12:33Z

Subgrid machine-learning (ML) parameterizations have the potential to introduce a new generation of climate models that incorporate the effects of higher-resolution physics without incurring the prohibitive computational cost associated with more explicit physics-based simulations. However, important issues, ranging from online instability to inconsistent online performance, have limited their operational use for long-term climate projections. To more rapidly drive progress in solving these issues, domain scientists and machine learning researchers opened up the offline aspect of this problem to the broader machine learning and data science community with the release of ClimSim, a NeurIPS Datasets and Benchmarks publication, and an associated Kaggle competition. This paper reports on the downstream results of the Kaggle competition by coupling emulators inspired by the winning teams' architectures to an interactive climate model (including full cloud microphysics, a regime historically prone to online instability) and systematically evaluating their online performance. Our results demonstrate that online stability in the low-resolution, real-geography setting is reproducible across multiple diverse architectures, which we consider a key milestone. All tested architectures exhibit strikingly similar offline and online biases, though their responses to architecture-agnostic design choices (e.g., expanding the list of input variables) can differ significantly. Multiple Kaggle-inspired architectures achieve state-of-the-art (SOTA) results on certain metrics such as zonal mean bias patterns and global RMSE, indicating that crowdsourcing the essence of the offline problem is one path to improving online performance in hybrid physics-AI climate simulation.

Machine Learning of Vertical Fluxes by Unresolved Midlatitude Mesoscale Processes

2026-03-08T16:27:48Z

Machine learning (ML) can represent processes unresolved in coarse-resolution Earth system models (ESMs) by learning from high-resolution climate data. Such ML parameterization approaches have been primarily tested in idealized setups where they have focused on deep convection. It remains largely unexplored whether these approaches could be used in a more targeted fashion to learn vertical fluxes resulting from midlatitude mesoscale processes, such as slantwise convection and frontal dynamics in extratropical cyclones, which are not well represented in ESMs. To address this, we employ a variable-resolution CESM2 simulation with a refined area over the North Atlantic (14-km grid refinement) that resolves such midlatitude mesoscale processes. We train an artificial neural network to predict vertical profiles of mesoscale moisture, heat, and momentum fluxes from the perspective of a coarse-resolution (111-km grid) model. Our results show that a large number of features are required to achieve reasonable model performance when data come from the midlatitudes of real-geography atmospheric simulations, especially when coarse-grained vertical velocities, which we show are not representative of vertical velocities in a coarse-resolution model, are excluded as inputs. Feature importance analysis reveals the importance of vertically non-local information in temperature, moisture, and the meridional wind. We suggest that these non-local relationships capture the influence of cold air outbreaks and fronts on mesoscale fluxes. Our results demonstrate the importance of vertically non-local processes, clarify the regime-dependent predictability of mesoscale fluxes, and identify variables most informative for their parameterization, providing guidance for improving ESMs with ML and advancing our understanding of multi-scale interactions in the midlatitudes.

Understanding the Evolution of Global Atmospheric Rivers with Vapor Kinetic Energy Framework

2026-03-06T23:16:25Z

Atmospheric rivers (ARs) often cause damaging winds, rainfall, and floods. However, the physical mechanisms governing their evolution remain poorly understood. To close this gap, we perform a global Vapor Kinetic Energy (VKE) budget analysis. Using two formulations of VKE, we show that ARs are governed by similar mechanisms regardless of ocean basins. ARs intensify primarily through the conversion of potential energy to kinetic energy (PE-to-KE), with horizontal convergence of vapor kinetic energy providing a secondary contribution in some regions. ARs decay mainly through condensation and turbulent dissipation, while their propagation is governed by the downstream convergence and upstream divergence of vapor kinetic energy. We also find PE-to-KE conversion varies spatially and strengthens in regions of greater baroclinic instability or enhanced topographic lifting, e.g., along North America's west coast. Collectively, these findings demonstrate that the VKE framework provides a powerful diagnostic for how physical processes shape AR evolution and regional variability.

Efficacy of Scalable Airline-led Contrail Avoidance

2026-03-06T22:08:04Z

Contrails account for a large portion of aviation's contribution to anthropogenic climate change. Navigational contrail avoidance is a promising solution to mitigate the warming caused by contrails. Prior trials testing navigational contrail avoidance have relied on bespoke integrations of contrail forecasts into airline operations. Here, we use a randomized control trial to test the feasibility of dispatcher-led contrail avoidance integrated into standard flight planning operations using a workflow that scales to an airline's entire network. We validated the efficacy of this intervention using satellite imagery and an automated flight-contrail attribution algorithm. Using this system, we observed an 11.6% reduction in contrail formation rate for the 1232 flights marked as eligible for contrail avoidance (intent-to-treat) relative to the flights in the control group (p = 0.011). In the 112 flights that flew contrail avoidance as planned (per-protocol flights), we observed a 62.0% lower contrail formation rate relative to the flights in the control group (p < 0.001). No statistically significant difference in fuel usage was observed between the two groups.

Physics-Informed Diffusion Model for Generating Synthetic Extreme Rare Weather Events Data

2026-03-06T18:54:55Z

Data scarcity is a primary obstacle in developing robust Machine Learning (ML) models for detecting rapidly intensifying tropical cyclones. Traditional data augmentation techniques (rotation, flipping, brightness adjustment) fail to preserve the physical consistency and high-intensity gradients characteristic of rare Category 4-equivalent events, which constitute only 0.14\% of our dataset (202 of 140,514 samples). We propose a physics-informed diffusion model based on the Context-UNet architecture to generate synthetic, multi-spectral satellite imagery of extreme weather events. Our model is conditioned on critical atmospheric parameters such as average wind speed, type of Ocean and stage of development (early, mature, late etc) -- the known drivers of rapid intensification. Using a controlled pre-generated noise sampling strategy and mixed-precision training, we generated $16\times16$ wind-field samples that are cropped from multi-spectral satellite imagery which preserve realistic spatial autocorrelation and physical consistency. Results demonstrate that our model successfully learns discriminative features across ten distinct context classes, effectively mitigating the data bottleneck. Specifically, we address the extreme class imbalance in our dataset, where Class 4 (Ocean 2, early stage with average wind speed 50kn hurricane) contains only 202 samples compared to 79,768 samples in Class 0. This generative framework provides a scalable solution for augmenting training datasets for operational weather detection algorithms. The average Results yield an average Log-Spectral Distance (LSD) of 4.5dB, demonstrating a scalable framework for enhancing operational weather detection algorithms.

Evaluating the Predictability of Selected Weather Extremes with Aurora, an AI Weather Forecast Model

2026-03-06T17:55:08Z

AI weather foundation models now achieve forecast skill comparable to numerical weather prediction at far lower computational cost, yet their predictability for high-impact extremes across dynamical regimes remains uncertain. We evaluate Aurora using an event-based framework spanning tropical cyclones, freezes, heatwaves, atmospheric rivers, and extreme precipitation at lead times from 1 to 21 days. Aurora demonstrates strong short-range (1-7 day) skill across event types, including competitive tropical cyclone track accuracy and high spatial agreement for temperature and moisture extremes. However, a consistent subseasonal failure mode emerges: while large-scale circulation patterns remain moderately skillful at 14-21 day leads, threshold-based extreme intensity collapses as fields regress toward climatology. This divergence indicates that Aurora retains synoptic-scale dynamical structure but loses surface-impact amplitude beyond 7-10 days. The practical predictability horizon for deterministic AI extreme-event forecasting therefore remains constrained by intrinsic atmospheric dynamics.

Wave Attenuation in Drifting Sea Ice: A Mechanistic Model for Observed Decay Profiles

2026-03-06T16:04:24Z

Wave-sea ice interactions shape the transition zone between open ocean and pack ice in the polar regions. Most theoretical paradigms, implemented in coupled wave-sea ice models, predict exponential decay of the wave energy but some recent observations deviate from this behaviour. Expanding on a framework based on wave energy dissipation due to ice-water drag, we account for drifting sea ice to derive an improved model for wave energy attenuation. Analytical solutions replicate the observed non-exponential wave energy decay and the spatial evolution of the effective attenuation rate in Antarctic sea ice.

Machine Learning Based Mesh Movement for Non-Hydrostatic Tsunami Simulation

2026-03-06T11:03:44Z

This study investigates the use of machine learning based mesh adaptivity, specifically mesh movement methods (UM2N), with depth integrated non-hydrostatic shallow water models. Motivation for this comes from the need for models which balance efficiency and accuracy for use in probabilistic coastal hazard assessment. Implementations are built on the discontinuous Galerkin finite-element (DG-FE) based software, Thetis, which leverages the partial differential equation (PDE) framework Firedrake for automated code generation. Verification on benchmark test cases and validation against laboratory measurements of coastal hazards, focusing on tsunami propagation, run-up, and inundation is performed. In these tests, the UM2N-driven meshes help resolve key non-hydrostatic dynamics and yield numerical solutions in close agreement with reference computations and measured data. Numerical results indicate that the UM2N surrogate based approach significantly accelerates conventional mesh movement techniques and has high robustness over long integration periods and under strongly nonlinear wave conditions.

Global Abiotic Sulfur Cycling on Earth-like Terrestrial Planets

2026-03-06T10:30:19Z

Sulfur is a redox active element that may have helped mediate an electron flow that kickstarted life and which presently is an essential element for all life on Earth. Despite current uncertainties in global sulfur fluxes, modeling sulfur's abiotic cycling through Earth's deep history is important for understanding the impact of a planet wide biosphere on sulfur geochemical cycling and availability and vice versa. We present here an open-source, dynamical box model for estimating global sulfur fluxes and concentrations among surface and deep Earth reservoirs over Earth history, allowing tracking and estimation of the sulfur distribution in planetary reservoirs over deep time in the absence of life. While the main model presented here does not take into account the abrupt evolution of redox-shunting biosynthetic pathways such as oxygenic photosynthesis, we also modeled the abiotic sulfur cycle before and after a Great Oxidation Event-like transition on Earth-like planets. Our results suggest a considerably distinct chemical makeup of sulfur content in marine sediments in the absence of life on an Earth-like planet, leading to a marine sediment sulfate content two orders of magnitude larger than on present-day Earth and a marine sediment sulfide content 4 orders of magnitude lower than on present day Earth, attributable to the lack of microbial sulfur metabolism. This model could be useful for understanding sulfur cycling on potentially habitable exoplanets.

DAISI: Data Assimilation with Inverse Sampling using Stochastic Interpolants

2026-03-06T06:03:04Z

Data assimilation (DA) is a cornerstone of scientific and engineering applications, combining model forecasts with sparse and noisy observations to estimate latent system states. Classical high-dimensional DA methods, such as the ensemble Kalman filter, rely on Gaussian approximations that are violated for complex dynamics or observation operators. To address this limitation, we introduce DAISI, a scalable filtering algorithm built on flow-based generative models that enables flexible probabilistic inference using data-driven priors. The core idea is to use a stationary, pre-trained generative prior that first incorporates forecast information through a novel inverse-sampling step, before assimilating observations via guidance-based conditional sampling. This allows us to leverage any forecasting model as part of the DA pipeline without having to retrain or fine-tune the generative prior at each assimilation step. Experiments on challenging nonlinear systems show that DAISI achieves accurate filtering results in regimes with sparse, noisy, and nonlinear observations where traditional methods struggle.

The Rise of AI in Weather and Climate Information and its Impact on Global Inequality

2026-03-05T22:07:21Z

The rapid adoption of AI in Earth system science promises unprecedented speed and fidelity in the generation of climate information. However, this technological prowess rests on a fragile and unequal foundation: the current trajectory of AI development risks further automating and amplifying the North-South divide in the global climate information system. We outline the global asymmetry in High-Performance Computing and data infrastructure, demonstrating that the development of foundation models is almost exclusively concentrated in the Global North. Using three different domains, we show how this infrastructure inequality continues through models' inputs, processes and outputs. As an example, in weather and climate modelling, the reliance on historically biased data leads to systematic performance gaps that disproportionately affect the most vulnerable regions. In climate impact modelling, data sparsity and unrepresentative validation risk driving misleading interventions and maladaptation. Finally, in large language models, dependence on dominant textualised forms of climate knowledge risks reinforcing existing biases. We conclude that addressing these disparities demands revisiting the three phases, i.e. models Input, Process and Output. This involves (i) a perspective shift from model-centric to data-centric development, (ii) the establishment of a Climate Digital Public Infrastructure and human-centric evaluation metrics, and (iii) a move from producer-consumer dynamics toward knowledge co-production. This integration of diverse knowledge systems would truly democratise compute sovereignty and ensure that the AI revolution fosters genuine systemic resilience rather than exacerbating inequity.

Global stability of the Atlantic overturning circulation: Edge state, long transients and boundary crisis under CO$_2$ forcing

2026-03-05T20:25:36Z

The Atlantic Meridional Overturning Circulation (AMOC), a crucial ocean current system, could transition to a weak state. Despite severe associated climate impacts, assessing the AMOC's response under global warming and its proximity to possible critical thresholds remains difficult. To understand future Earth system stability, a global dynamical view is needed beyond the local stability analysis underlying classical early-warning methods. Using an intermediate-complexity climate model, we explore the stability landscape of the AMOC for different atmospheric CO$_2$ concentrations. We explicitly compute the edge state (or Melancholia state), a chaotic saddle on the basin boundary separating the strong and weak AMOC attractors found in the model. While being unstable, the edge state can govern the transient climate for centuries, supporting centennial AMOC oscillations driven by atmosphere-ice-ocean interactions in the North Atlantic. At increased CO$_2$ levels projected for the near future, we reveal a boundary crisis where the current AMOC attractor disappears by colliding with the edge state. Under crisis overshoot, long chaotic transients due to "ghost states" lead to diverging ensemble trajectories under time-varying forcing. Rooted in dynamical systems theory, our results offer an explanation of large ensemble variance and apparent "stochastic bifurcations" observed in earth system models under intermediate forcing scenarios.

Detection of C3 in Titan with VLT-ESPRESSO

2026-03-05T16:44:26Z

Titan is regarded as a natural laboratory in the Solar System for studying atmospheric photochemistry and the abiotic production of organic molecules on cold small exoplanets. Since the end of the Cassini-Huygens mission, telescope observations have enabled new detections of increasingly complex carbon-based molecules at infrared and sub-millimetre wavelengths, while the optical regime has been largely overlooked. Following a recent tentative detection of the 405 nm absorption band of C3 in Titan in archived optical VLT UVES spectra at resolving power R = 60000, this work reports an eight sigma detection of the C3 405 nm absorption band in Titan using dedicated ultra high resolution VLT ESPRESSO observations at R = 190000, the highest spectral resolution optical observations of Titan to date. The VLT ESPRESSO spectrum is compared to model spectra of Titan with varying C3 abundances. A chi squared analysis is used to assess the agreement between non solar spectral features and C3 absorption as the C3 abundance is varied, and a Bayesian Markov Chain Monte Carlo fit between model and observed spectra is performed. The chi squared analysis yields an eight sigma detection of C3, consistent with a C3 column density of approximately 1.5E13 cm-2, while the MCMC fit retrieves a C3 column density of 1.47E13 cm-2 at five sigma. These values are consistent with the order of magnitude predicted by photochemical models, which reach parts per million levels in the Titan mesosphere. This work demonstrates the usefulness of instruments and techniques originally developed for exoplanet research when applied to Solar System targets.

EnsAI: An Emulator for Atmospheric Chemical Ensembles

2026-03-05T16:23:10Z

Ensemble-based methods for data assimilation and emission inversions are a popular way to encode flow-dependency within the model error covariance. While most ensemble methods do not require the use of an adjoint model, the need to repeatedly run a geophysical model to generate the ensemble can be a significant computational burden. In this paper, we introduce EnsAI, a new AI-based ensemble generation system for atmospheric chemical constituents. When trained on an existing ensemble for ammonia generated by the GEM-MACH air quality model, it was shown that the ensembles produced by EnsAI can accurately reproduce the meteorology-dependent features of the original ensemble, while generating the ensemble 3,300 times faster than the original GEM-MACH ensemble. While EnsAI requires an upfront cost for generating an ensemble used for training, as well as the training itself, the long term computational savings can greatly exceed these initial computational costs. When used in an emissions inversion system, EnsAI produced similar inversion results to those in which the original GEM-MACH ensemble was used while using significantly less computational resources.

Hydrodynamic outflows of proto-lunar disk volatiles

2026-03-05T16:02:24Z

Volatile elements - those that vaporize at low temperatures - are depleted in lunar rocks relative to terrestrial rocks. This systematic chemical depletion is evidence for vaporization and preferential removal of vapor from proto-lunar materials during the high-temperature processes accompanying lunar origin. Despite the robustness of these observations, the physical processes by which proto-lunar vapors were removed after the giant impact are not yet well-understood. Here, we show that toward the end of post-giant impact cooling history, Earth's atmosphere was dominated by carbon species (e.g., CO) and was spatially compact, behaving as a closed system retaining Earth's volatile inventory, whereas the proto-lunar disk atmosphere was dominated by H and H2 and was spatially extended, developing into a hydrodynamic outflow analogous to the solar wind. We find that equilibrium H2 recombination (2H->H2) in a partially-dissociated disk atmosphere produces a nearly isothermal structure, a feature known to activate outflows. The expected outflow was strong enough to propel proto-lunar volatiles from a Roche-interior (r < 3RE) disk out of Earth's gravity field and to establish a cometary tail composed of volatile elements transporting proto-lunar disk volatiles into interplanetary space. The proposed model suggests that the dichotomy in volatile element abundances between the silicate Earth and Moon is a natural outcome of the hydrodynamical behavior of magma ocean atmospheres and that lunar chemical and isotopic volatile abundances are diagnostic of the radial structure of the proto-lunar disk towards the end of its condensation.