Papers for

climate modelers

Papers whose findings have a practical use for this group, as judged from the abstract. Open a paper to read what it means in practice.

HybridCT speeds up 3D cloud imaging for better climate predictions

HybridCT: ML Accelerated Tomography for Climate Relevant 3D Cloud Observations

Abstract: Low-level clouds are one of the largest sources of uncertainty in climate predictions, partly because we lack detailed three-dimensional (3D) measurements of their internal structure. Here, we present HybridCT, a cloud tomographic reconstruction scheme accelerated by machine learning (ML) that recovers 3D cloud extinction directly from passive multi-angle observations. An ML emulator maps radiances from multiple viewing angles to cloud optical thickness (COT), producing path-integrated projections, and a Simultaneous Algebraic Reconstruction Technique (SART) explicitly inverts these projections to retrieve 3D extinction. Training observations are generated from synthetic 3D cloud fields using radiative transfer simulations for nine viewing angles between $\pm$70$^{\circ}$. HybridCT reconstructs extinction with RMSE 13.3 km$^{-1}$ in 3.5 s, compared to 22.7 km$^{-1}$ in approx. 30 min for the state-of-the-art physics-based tomography framework AT3D, a speedup exceeding 500$\times$. HybridCT thus provides a scalable route toward large-scale observations of 3D cloud structure relevant to radiative-budget interpretation and evaluation of high-resolution cloud simulations.

Thu 12 NovMachine Learning
The gist
Low-level clouds are hard to measure in three dimensions, but understanding their shape is important for predicting the climate. The authors created HybridCT, a faster method that uses machine learning to quickly turn pictures of clouds from different angles into detailed 3D cloud images. This method is over 500 times faster than previous tools while still being accurate, which could help scientists better study clouds and improve climate models.
Open → 2611.16255v1

Multifidelity operator inference improves model accuracy with scarce data

Multifidelity Operator Inference for Non-intrusive Model Reduction under Data Scarcity

Abstract: Many-query analyses, in which a computational model is repeatedly evaluated, play an important role in uncertainty quantification and optimization for engineering design. However, many-query analyses can be prohibitively expensive when considering high-fidelity numerical models, especially those governed by partial differential equations (PDEs). Surrogate modeling replaces costly simulations with cheap approximations, enabling practical uncertainty quantification and design workflows. Machine learning (ML) approaches to surrogate modeling have gained significant traction in the development of surrogate models due to their non-intrusive nature; they do not require access to the underlying simulation code. The Operator Inference (OpInf) approach, in particular, fits reduced operators to state and corresponding time derivative data. However, high-fidelity training data is scarce due to limited computational budgets, and OpInf models trained on such scarce data often exhibit high variance, a critical liability in high-risk engineering design. We propose Multifidelity Operator Inference (MF-OpInf), which reduces this variance by supplementing scarce high-fidelity snapshots with abundant low-fidelity data drawn from coarser discretizations of the same problem. MF-OpInf handles data across varying fidelity levels by utilizing a multifidelity linear regression framework to estimate the reduced operators. Numerical experiments on model PDE problems demonstrate that MF-OpInf predicts PDE solution fields with higher accuracy compared to the high-fidelity-only OpInf with the same computational budget.

Tue 10 NovMachine Learning
The gist
Computational models help engineers understand complex systems but can be very slow and expensive to run many times. The authors found a way to combine a few detailed simulations with many cheaper, less detailed ones to create faster and more accurate simplified models. This approach, called multifidelity operator inference, works better than just using the expensive data alone. It helps when there is limited high-quality data but plenty of lower-quality data available.
Open → 2611.13624v1

Single precision ocean model offers long-term climate accuracy and faster runs

FESOM2.8 in single precision: climate fidelity over six decades and performance on CPUs and GPUs

Abstract: Halving the floating-point precision of an ocean model can make it substantially faster, but whether the rounding errors stay negligible over climate time scales has rarely been tested. We implement single precision in the global ocean-sea-ice model FESOM2 and compare 60-year hindcasts with double-precision controls. A first implementation lost salt steadily, accumulated a heat-content deficit of 129 ZJ, and shifted major ocean fronts. Controlled experiments trace the drift to the tracer update: small increments added to a large salinity value are rounded away, and the increments themselves are biased. Accumulating tracer tendencies and layer thicknesses in 64 bits, and carrying the rounding error of each tracer update into the next, removes the drift for at most 8% additional runtime. Single- and double-precision simulations then differ in heat content and mean state about as much as two double-precision runs with slightly perturbed initial conditions, sea ice differs by at most 0.2%, and he salinity difference is four orders of magnitude smaller than before. In the coupled model AWI-ESM3, single precision adds no detectable heat source or sink over 60 years. Single precision speeds up the 60-year hindcasts by a factor of 1.35, and the model by up to 1.66 on meshes of up to 5 km resolution; on GPUs, a C++/Kokkos version runs up to 1.27 times faster and needs 43% less memory. Whether reduced precision is safe for climate integrations is decided by long-term budgets, not by short tests.

Sun 8 NovMathematical SoftwarePerformance
The gist
Using fewer decimal places in ocean climate models can speed up simulations, but it's unclear if this affects long-term accuracy. The authors tested a single precision version of the FESOM2 ocean model over 60 years and found initial problems with salinity and heat content drift. By improving how small changes are tracked, they reduced errors to negligible levels, matching the accuracy of standard double precision runs. This method runs about 35% faster on CPUs and also speeds up GPUs while using less memory.
Open → 2611.11248v1

AI weather model powers fast earth system model with realistic climate

Rescene-ESM: an Earth system model built around a frozen AI weather model

Abstract: Earth system models underpin climate projections, but their cost limits how many scenarios, ensemble members and impact pathways can be explored. Machine-learning weather models forecast as skillfully as physics-based models at a fraction of the cost, yet coupled to an ocean they have produced weak El Nino variability or have learned it from climate-model output, and none carries the climate through rivers to the people exposed to it. Here we present Rescene-ESM, an Earth system model whose coupler, YAHO, yokes a frozen 20.8-million-parameter AI weather model to physically formulated ocean, sea-ice, land, river, carbon and energy components constrained by observations. Adding the couplings one at a time shows what each contributes: over a climatological ocean the AI atmosphere varies from year to year by a third of the observed amount over the oceans, and an interactive mixed layer, an equatorial wave ocean, the residual of the observed sea surface temperature (SST) equation as stochastic forcing and the land's moisture flux raise this step by step and bring El Nino and its teleconnections, leaving the mean climate almost unchanged. The full model simulates about 1,000 years per day on one graphics processing unit, holds the observed climate for a thousand years without drift, reproduces the observed 1979-2022 global warming, carries forced change with a physically based structure around the frozen network (transient climate response 1.60 K), routes runoff to 212 river gauges with a median seasonal-cycle correlation of 0.79 and, driven by emissions, separates the heatwave exposure of an SSP1 and an SSP3 world by a factor of 4.7 in 2081-2100, three-quarters of it through the warmer climate.

Sun 8 NovMachine Learning
The gist
Climate models are usually really slow and costly, so it's hard to explore many future scenarios. The authors built an earth system model that uses a fixed AI weather model combined with physical ocean, land, sea-ice, river, and carbon components. This model can simulate a thousand years per day on one GPU, keeping the climate stable and reproducing observed warming patterns. It also captures important climate features like El Nino and routes river runoff realistically, helping to estimate future heatwave exposure under different emission scenarios.
Open → 2611.11066v1

Stable probabilistic models improve long-term forecasts for complex weather systems

Stable probabilistic emulation of partially observed dynamical systems

Abstract: Predicting the long-term behavior of partially observed dynamical systems such as the coupled atmosphere-ocean system is central to societal decision-making. Motivated by the success of deep learning emulators in weather forecasting, data-driven climate models represent a promising approach. However, climate emulation poses distinct challenges: models must account for unresolved fast variables introduced by unavoidable coarse-graining, reproduce long-term statistics, and remain stable over arbitrarily long rollouts and under external perturbations. These requirements motivate stochastic data-driven emulators, but modern generative models lack stability guarantees. We introduce forced-dissipative stochastic interpolants, a generative framework for learning stochastic reduced dynamics. Its architecture uses structural constraints inspired by turbulent dynamics, which provide stability by design while retaining the flexibility of stochastic interpolants to represent complex probabilistic dynamics. Building on the theory of Markov-chain stability, we establish a Foster--Lyapunov drift condition and prove geometric ergodicity for our emulator, together with stability under steady external forcing. We evaluate the framework on the two-timescale Lorenz-96 system, a prototype of partially observed multiscale turbulence. The resulting emulator remains stable over long integrations and under finite-amplitude forcing, with accurate prediction of forced responses of both the mean and variance.

Fri 6 NovMachine Learning
The gist
Predicting weather or climate over long periods is tough because we can only observe part of the system and many small fast changes are hidden. The authors designed a new kind of model that learns from data how these hidden parts behave, while making sure predictions don’t go off track even after a long time. Their method uses ideas from physics to keep the model stable and reliable. They tested it on a simplified weather system and showed it can predict changes accurately and stay stable under different conditions.
Open → 2611.09317v1

Spin neural operators improve learning of vector and spin fields on sphere

Spin Spherical Fourier Neural Operators: Efficient Equivariant Learning for Vector and Spin Fields on the Sphere

Abstract: Many fields of physical interest on the sphere, such as wind, the polarisation of the cosmic microwave background and the shear of weak gravitational lensing, are spin fields, whose values change phase when the local frame rotates. We extend spherical Fourier neural operators (SFNOs) to fields of any spin, through lifted spherical convolutions that are exactly rotationally equivariant. Lifting spin fields to Mackey functions on the rotation group provides the conceptual framework, while computation is performed exactly and efficiently through their harmonic representation and by appealing to sampling theorems on the sphere and rotation group. We construct spin SFNOs (S$^2$FNOs) from these building blocks. We demonstrate empirically that rotational equivariance holds to machine precision for these layers. We achieve state-of-the-art performance on existing benchmarks, and introduce new benchmarks on which spin-aware models outperform their scalar counterparts.

Tue 3 NovComputer Vision and Pattern RecognitionMachine Learning
The gist
Some important natural phenomena on the globe, like wind directions and cosmic light patterns, are represented by special fields called spin fields. These fields change depending on how you look at them locally, which makes them tricky for computers to understand. The authors developed a new kind of neural network layer that respects these changes exactly and works efficiently by using mathematical properties of the sphere. Their method performs better than previous ones on tests and works precisely even when the data is rotated.
Open → 2611.04389v1

Earth system model parameterizations improved by language model agents

Can LLM Agents Improve Imperfect Earth System Model Parameterizations?

Abstract: Earth system models (ESMs) rely on parameterizations to represent unresolved processes such as convection, cloud microphysics, and turbulence. However, translating advances in physical understanding into stable, validated parameterizations and integrating them into ESMs often takes years. Agentic AI could accelerate this process and enable more systematic exploration of design choices. Using warm-rain collision-coalescence as a test case, we evaluate agents' ability to synthesize scientific literature, reimplement published bulk parameterizations, and propose new functional forms as differentiable JAX code. We compare conventional parameterizations, data-driven emulators, and agent-generated closures against high-fidelity superdroplet simulations, repeating unattended experiments across ten language models. With expert guidance or iterative feedback, agents produced executable, differentiable, mass-conserving code, although unattended performance depended strongly on the scheme and language model. Iterative feedback lowered median rollout error for eight of ten models. Retuned physical schemes matched or outperformed many learned parameterizations at substantially lower computational cost, but two of 100 unattended agent-generated closures achieved slightly lower rain-mass error than the most accurate retuned scheme on held-out simulations.

Mon 2 NovMachine Learning
The gist
Earth system models simulate weather and climate but rely on simplified formulas called parameterizations to represent small processes that computers can't directly model. The authors tested whether language model agents can help create better formulas by reading scientific papers, coding new versions, and comparing them to detailed simulations. They found that with expert help or iterative feedback, these agents produced working and physically consistent code, sometimes even outperforming traditional versions. However, results varied depending on the language model and task, showing promise but not yet a clear replacement for human experts.
Open → 2611.03543v1

Low-rank clebsch-gordan coupling improves spherical weather forecasts

Low-Rank Clebsch--Gordan Coupling for Equivariant Spherical Forecasting

Abstract: Spherical-harmonic neural operators provide an efficient representation for global forecasting, but their discretized pointwise nonlinearities can break rotational equivariance through aliasing. Clebsch-Gordan (CG) tensor products provide exactly equivariant interactions between spherical-harmonic representations, but their computational cost grows rapidly with spectral bandwidth and has limited their use at forecasting scale. We introduce LCG-SFNO, which combines low-rank CG interactions within a controlled low-degree band with coefficient-space residual propagation and an exactly projected quadratic nonlinearity, preserving SO(3) equivariance throughout the learned operator. To make the CG interaction practical, we further develop SSHCG, a sparse GPU implementation with a factorized alternative for wider coupled bands. On spherical shallow-water dynamics and the ERA5-based WeatherBench benchmark, LCG-SFNO reduces rotational error to numerical precision (more than three orders of magnitude below SFNO on shallow water and five orders of magnitude below it on WeatherBench), while matching SFNO on shallow-water forecasting and substantially improving forecast skill relative to SFNO on WeatherBench. Matched ablations show that CG coupling accelerates convergence, while SSHCG keeps the coupling to about 10 percent of a training step at the operating bandwidth, making exact equivariant coupling practical at forecasting scale.

Mon 2 NovMachine Learning
The gist
Forecasting global weather patterns often involves working with data on spherical surfaces, but existing methods can break symmetry when predicting changes, leading to errors. The paper introduces a new technique that keeps the math perfectly aligned with the Earth's rotation, reducing rotational errors dramatically and improving forecast accuracy. The authors also developed efficient software to make this method practical for large-scale weather prediction.
Open → 2611.03411v1

Cascading short forecasts helps predict complex flows from coarse data

Generative Forecasting from Coarse-Grained Observations: When Does Cascading Help?

Abstract: We study probabilistic forecasting of dynamical systems such as fluids and waves from coarse observations. Since the unseen fine scales influence the dynamics, the same coarse initial state can evolve into a whole distribution of futures. Generative forecasters may produce visually plausible samples, but it is hard to tell whether they sample the right conditional distribution, since trajectory data contain a single future for each initial state. We address this problem by generating reference data in a hierarchical way: for each coarse initial state, we generate many different fine-scale components and numerically evolve the full states. Using these conditional ensembles, we show that standard diffusion models and stochastic interpolants make substantial distributional errors. We then show that cascading shorter forecasts greatly reduces these errors for the turbulent flows and waves we study, but we also find and characterize cases in which it hurts. Since the coarse process is not Markov, cascading discards information about the unresolved fine scales even with exact transitions. To understand this, we propose a solvable analytic model which shows that the optimal cascade depth decreases as the coupling between the unresolved and observed scales gets stronger, and increases as the dependence of the forecast on the observation becomes more oscillatory. Consistent with our experiments, cascading is detrimental when the direct forecast is ``easy'' to learn, for example Gaussian as in the passive scalar case. We release our datasets as a first benchmark for probabilistic forecasting with ground-truth conditional ensembles.

Sat 31 OctMachine Learning
The gist
Predicting how things like fluids or waves move can be tricky when you only see rough or coarse data because the hidden small details affect the future. The authors made new test data sets showing many possible detailed versions of the future from a single coarse start. They found that common forecasting methods often get the full range of outcomes wrong. Breaking a forecast into many small steps, called cascading, often improves predictions but sometimes makes them worse depending on the problem. They also developed a simple math model to explain when cascading helps or hurts.
Open → 2611.00686v1

Quantum method boosts chances of rare event simulations

Quantum amplification of improbable futures

Abstract: Some consequential futures consist of entirely plausible steps arranged into an unlikely sequence, such as ordinary weather anomalies combining into an extreme event or local equipment failures propagating into a widespread outage. Because the sequence as a whole carries little probability, it may be rarely observed in simulation. Quantum mechanics offers a way to change those odds: ideal amplitude amplification raises the chance of drawing such futures while preserving their relative likelihoods. For a five-state process six steps ahead, 13,888 of 15,625 futures are rare, yet together they account for about one draw in 40. We built the complete circuit in IBM's Qiskit and checked every part against an exact enumeration; four simulated rounds raised the share of rare futures drawn from 2.5% to 98.0%. On IBM's Kingston quantum computer, a two-step reference circuit that marks rare futures from a precomputed list drew them in 68.4% of shots against 43.5% for its control (z = 50); at three steps (about 1,170 two-qubit gates) the predicted signal reversed sign. Because this model's probabilities are known, every result could be checked against exact values; the mechanism is intended for simulators whose probabilities are hard to calculate, such as climate, epidemic and financial models.

Thu 29 OctEmerging Technologies
The gist
Some important future events happen when a series of ordinary steps line up in a very unlikely way, like unusual weather patterns causing a big storm. These rare sequences are hard to catch using normal simulations because they happen so infrequently. The authors show how quantum computing can increase the chances of seeing these rare futures by amplifying their probabilities without changing their relative odds. They tested their approach on IBM's quantum computers and simulators, raising the chance of drawing rare outcomes from a tiny fraction to nearly all cases in certain tests. This technique can help in complex models where calculating exact probabilities is too hard.
Open → 2610.41024v1

Weather downscaling improved by learning operators and normalizing guidance

Operator-Aware Posterior Sampling for Multivariate Weather Downscaling

Abstract: Statistical downscaling is the task of reconstructing high-resolution meteorological fields from coarse inputs. A principled approach formulates downscaling as a Bayesian inverse problem, using a diffusion prior over high-resolution fields and incorporating the coarse field only at inference through a likelihood. This decouples the prior from the coarse source, allowing one pretrained model to adapt to different sources. However, two challenges limit this flexibility: (1) when coarse and target fields come from different reanalysis systems, no fixed coarsening operator can capture their relationship; and (2) guidance gradients can differ by orders of magnitude across variables, making a single guidance strength ineffective. This paper addresses both challenges within the guidance loop, leaving the prior unchanged. We learn the forward operator from paired data and normalize the gradient per channel. We show that this normalization corresponds to steepest descent under a block-max norm, allowing a single guidance strength across variables. Our code is publicly available at https://github.com/CarloSaccardi/operator-aware-weather-downscaling.

Thu 29 OctComputer Vision and Pattern Recognition
The gist
It is hard to turn low-resolution weather data into detailed, high-resolution weather maps, especially when data sources vary. The authors found problems with current methods when the data comes from different systems and when different weather variables influence the process unevenly. They fixed these by teaching the model how to translate between data sources and by adjusting how the model balances different weather data. Their improvements work without changing the fundamental model used to create the detailed maps.
Open → 2610.39592v1

Data driven learning of operators for forecasting stochastic dynamics

Learning Kolmogorov Backward Operators for Spectral Analysis and Forecasting: A Data-Driven Variational Galerkin Framework

Abstract: For a time-homogeneous Markov process, the fixed-lag solution operator \( (T_τf)(x)=\mathbb E[f(X_{t+τ})\mid X_t=x] \) of the Kolmogorov backward equation evolves observables linearly and deterministically, even when the dynamics are stochastic. We approximate this operator using a two-stage method where we first approximate this operator directly from data, without knowledge of the governing equation (e.g., Fourier Neural Operator), and then compress the Galerkin projection onto a finite-dimensional observable space that is fixed in advance or selected variationally. The error analysis traces the learning error of the estimator, the sampling error of the Galerkin inner products, and the invariance and reconstruction properties of the observable space to the computed spectra and forecasts. The sampling error enters only through a multiplicative factor. Near an isolated spectral cluster, the invariance defect of a subspace controls its distance from the target invariant subspace. The forecast error is controlled by the one-step error of the reduced model and the projection error of the observable; when the measure is invariant and the reduced model preserves constants and is stable on mean-zero observables, the bound is uniform in the horizon. For deterministic dynamics, a learning rate for Fourier neural operators (FNOs) gives one explicit bound for the learning error. Numerical experiments illustrate spectral approximation and forecasting, including long-rollout stability and a comparison with extended dynamic mode decomposition (EDMD) when fewer transition pairs than basis functions are observed.

Thu 29 OctMachine Learning
The gist
Forecasting how systems that change randomly behave over time can be tricky. This paper shows a way to learn the rules for predicting future outcomes directly from data, without knowing the exact system equations. They do this in two steps: first, they build a model that represents the system’s time evolution, and second, they simplify this model to focus on key features. Their approach helps produce reliable long-term predictions and measures the accuracy of the learned model.
Open → 2610.39580v1

Exponential-Nemytskii operators improve modeling of time-dependent fields

Exponential-Nemytskii Neural Operators: Learning Field Evolution Across Step Sizes

Abstract: In variable-step operator learning, agreement at observed step sizes does not by itself determine predictions at other temporal resolutions. We propose an operator whose step-size dependence is prescribed by a first-order Duhamel approximation of semilinear dynamics. It combines exact linear propagation with a local nonlinear function of the field and its derivatives, allowing evaluation below the observation interval without supervision densely sampled across step sizes. Under suitable conditions and at fixed spatial resolution, we prove that agreement between models at two step sizes controls their prediction differences throughout a prescribed range. We also give conditions for local identifiability of both components from two step sizes, supporting interpretation of the learned structure. A latent extension applies the same construction to learned fields. We evaluate the base model on two synthetic problems and the extension on two real datasets against several baselines.

Wed 28 OctMachine Learning
The gist
Predicting how things change over time is hard, especially when measurements are taken at irregular time steps. This paper introduces a method that combines exact linear changes with flexible nonlinear adjustments to better model how a field evolves across different step sizes. The authors show theoretically that understanding the model’s behavior at two time scales helps predict its performance across a range of scales. They also test their approach on simulated and real data, demonstrating improved predictions compared to existing methods.
Open → 2610.39041v1

Improving inference for costly simulations using smarter variable choices

Bayesian Inference of Expensive Simulators: Survey and Comparison of Design Choices

Abstract: We present a survey of approaches to Bayesian inference with simulator forward models, focusing on settings with prohibitively expensive simulators. In this domain, Bayesian optimization-based methods have emerged as the most promising approach. We unify existing methods into a general framework, enabling consistent reasoning about shared algorithmic structure across the fragmented literature. Within this framework, we identify the primary design choices and perform a systematic empirical comparison of their effects on inference efficiency in terms of required simulation budget. A key result of our study is that inference efficiency is often dominated by the choice of surrogate-model response variable, as it directly governs the complexity of the modeled function. Specifically, we demonstrate that the prevalent scalar log-likelihood is a poor response variable, as it discards crucial information. Based on this insight, we introduce the concept of custom \emph{proxy variables}: transformations of simulator outputs engineered to retain more information and provide smoother response surfaces, substantially improving inference efficiency over standard response variables. We provide actionable guidance and a numerically robust Julia package to facilitate practical adoption of Bayesian optimization for simulator inverse problems.

Tue 27 OctMachine Learning
The gist
Simulators help us understand complex systems but can be very slow to run. The authors looked at ways to make Bayesian inference faster when using these slow simulators. They found that the way simulation results are represented heavily impacts the speed and accuracy of inference. By creating special transformed variables called proxy variables, they keep more useful information and make the process smoother and more efficient. They also provide practical tools and advice for applying these improvements.
Open → 2610.37284v1

Climate model uses lightweight atmosphere emulator for stable long-term simulation

FASST: A Compact Stateless Atmosphere Model for Long-Term Ocean Coupling

Abstract: Coupled climate simulations are computationally expensive, largely because of the atmosphere. We introduce FASST, a coupled climate model that retains the prognostic POP2 ocean while replacing the atmosphere and sea ice with lightweight U-Net emulators. The atmospheric emulator is diagnostic, predicting surface fluxes from the evolving ocean-ice state every $6$ simulated hours, while the sea-ice emulator advances the ice state monthly. For $140$ years of unforced integration, FASST maintains a stable global-mean climate without SST restoring, with post-transient global-mean SST of $291.30 \pm 0.22\mathrm{K}$ compared with $291.25 \pm 0.10\mathrm{K}$ in CESM$2$ piControl. FASST also captures interannual Niño-$3.4$ variability, although with weaker amplitude than CESM ($0.62\mathrm{K}$ versus $1.02\mathrm{K}$). These results demonstrate that a stateless learned atmosphere can sustain stable, free-running coupled climate simulations over century-scale integrations.

Mon 26 OctMachine Learning
The gist
Simulating the Earth's climate is very expensive because the atmosphere part requires many calculations. The authors created FASST, a model that keeps a detailed ocean simulation but replaces the atmosphere and sea ice parts with simpler learned models. Their approach can simulate 140 years of climate data stably without extra corrections and roughly matches key ocean temperature patterns. This suggests simpler atmosphere models can still support long-term climate studies.
Open → 2610.36297v1

Efficient weather forecasting model offers high resolution and lower compute cost

Efficient probabilistic weather forecasting at quarter-degree resolution with two-sided transport attention

Abstract: High-resolution probabilistic weather forecasting remains computationally demanding. Here we present Astra, a 0.25$^\circ$ ensemble model trained on eight H200 GPUs. It combines Kantorovich Attention, a two-sided transport attention in which each region chooses where it reads from and writes to, so every region sends and receives at token resolution, with a self-featurised continuous ranked probability score (CRPS) that also scores forecasts in the model's multiscale features. In WeatherBench 2, Astra matches GenCast's CRPS averaged over five upper-air targets and days 1-10 with 9.8% of its hardware-normalised training compute. Verified against ERA5, its CRPS and ensemble-mean root-mean-square error are lower than the ECMWF ensemble's on all 600 variable, level and lead-time combinations, and its CRPS is lower than GenCast's on 73.5% of them. A 15-day member needs 39 times fewer floating-point operations than GenCast's GPU implementation, and 50 members take under 40 min on one consumer GPU, within 21 GB.

Mon 26 OctMachine Learning
The gist
Predicting the weather accurately usually needs a lot of computer power, especially at fine detail. The authors created Astra, a weather forecasting model that works at a quarter-degree resolution using less computing resources. Astra uses a special attention method that lets different parts of the forecast interact in both directions to make better predictions. Tests show Astra matches or beats existing models while using far fewer calculations and running on accessible hardware like a single consumer GPU.
Open → 2610.35545v1

Generative AI methods improve modeling and validation of turbulent flows

Generative AI in Turbulence: Foundations, Applications, Validations and Benchmarks

Abstract: Generative AI offers new approaches to turbulence problems in which incomplete observations or unresolved d.o.f. admit multiple physically plausible solutions. Potential applications include stochastic closures, super-resolution, data assimilation, uncertainty quantification, and the generation of Eulerian and Lagrangian flow ensembles. However, turbulence is not a conventional image-generation problem. Its multiscale dynamics, intermittency, extreme events, rough velocity fields and intrinsically stochastic transport make visual realism, pointwise error, and low-order spectral agreement insufficient measures of physical fidelity. We examine where generative AI provide novel capabilities with emphasis on diffusion models, score-based methods, and related probability-transport formulations. We discuss their use in unconditional generation, probabilistic reconstruction, reduced-order modeling, temporal evolution, and mesh-free representations. We argue that validation must address turbulence properties: energy transfer, structure functions, anomalous scaling, pdf tails, coherent structures, conservation laws, temporal consistency, ensemble diversity and generalization. Finally, we discuss potential limitations in validation, baseline, data availability and reproducibility, and the need for a benchmark ecosystem combining open data, standardized challenges, hidden tests, physics-aware metrics, and documented failures. This paper serves as a bridge between two communities. For AI practitioners, we provide a fair and accessible introduction to turbulence, essential for appreciating the challenges posed by the field and for assessing the physical relevance of generative models. For turbulent practitioners, we provide an introduction to the key technological aspects of Generative AI, clarifying what these tools actually do, and where their potential and current limitations lie.

Mon 26 OctArtificial IntelligenceMachine Learning
The gist
Turbulent flows, like swirling smoke or water currents, are complicated and hard to predict because they have many possible outcomes and patterns. The authors explore how generative AI—computer methods that create data from learned patterns—can help model and simulate these complex flows more realistically. They point out that common measures for images don’t capture important physical details of turbulence, so better ways to check these AI models are necessary. The paper discusses which AI techniques work best and highlights the need for shared data and tests to compare methods fairly.
Open → 2610.35123v1

Tensor sampling and recovery improve dynamical system reconstructions

Dynamics-Aware Sampling and Recovery for Tensor-Valued Dynamical Systems

Abstract: We study sampling and recovery for tensor-valued dynamical systems evolving under the t-product, using partial spatial observations over time. Although the t-product Fourier transform yields slicewise matrix dynamics, recovery does not generally decouple across slices. Eigenvalues shared across Fourier-domain frontal slices couple the corresponding eigenspaces into aggregated spectral blocks, which determine the tensor-specific recovery geometry and space--time tradeoff. We establish deterministic identifiability conditions for the initial tensor state. In the diagonalizable case, these are frame conditions on the aggregated blocks. We extend them to non-diagonalizable systems through generalized eigenspaces and Jordan structure. A temporal--spectral factorization of the finite-horizon stacked recovery system underlying Tensor Dynamical Sampling (DS) yields a decoupled two-stage formulation once the temporal threshold is reached. Under independent spatial subsampling, we prove high-probability recovery guarantees using empirical Gram matrices on the aggregated blocks. The bounds depend on aggregated multiplicities, incoherence, and population conditioning. Stability estimates separate the spatial frame factor from the temporal Vandermonde factor and show that increasing temporal depth may worsen numerical conditioning even after identifiability is ensured. We implement Tensor DS and evaluate it on synthetic and real data. Phase transitions track the predicted blockwise sampling thresholds, and stability experiments illustrate Vandermonde conditioning. Experiments on sea surface temperature and turbulence velocity data show that a learned t-product system operator generally yields more accurate and structurally faithful reconstructions than vectorized low-rank, Tucker-type, graph-regularized, graph-spectral, and vectorized dynamical sampling baselines.

Sun 25 OctInformation Theory
The gist
This paper studies how to best sample and reconstruct complex evolving systems that can be represented as tensors—multi-dimensional arrays—rather than simple vectors or matrices. The authors explore how information shared across different frequency components of these tensors affects recovery and identify mathematical conditions that guarantee successful reconstruction from limited spatial observations over time. They show that while observing more time steps can help, it may also introduce numerical difficulties due to complex matrix structures. Their methods are tested on both synthetic data and real-world examples like sea surface temperatures and turbulent flows, demonstrating better accuracy and structure preservation compared to existing techniques.
Open → 2610.34621v1

Predicting complex system changes by combining learned models

Extrapolating learned dynamics across bifurcations based on generalized synchronization

Abstract: Predicting dynamics across bifurcations by extrapolating trained models is crucial in diverse fields of physics. Generalized synchronization provides a framework for understanding dynamical reconstruction by neural networks, especially in the context of reservoir computing. Building on this framework, we derive a formula combining pretrained readouts to reconstruct unseen chaotic dynamics. Using classical Lagrange interpolation theory, we derive rigorous bounds on readout errors and conjugacy residuals. Numerical examples demonstrate accurate prediction of chaotic dynamics across bifurcations.

Sun 25 OctMachine Learning
The gist
Predicting how complex systems behave when they suddenly change is hard. The authors found a way to use multiple trained neural network parts together to guess what happens in new situations they haven't seen before. Their math shows this guessing can be very accurate. They tested their approach and proved it works for chaotic systems that change behavior sharply.
Open → 2610.34471v1

Clustering time series from dynamical systems using occupation kernels

The Clustering Occupation Kernel (CLOCK) Method for Time Series Arising from Dynamical Systems

Abstract: We introduce the Clustering Occupation Kernel (CLOCK) method, a novel representation-based framework for clustering time series arising from dynamical systems. CLOCK embeds observed sequences into a reproducing kernel Hilbert space (RKHS) via a time-averaged occupation kernel, enabling trajectory comparison through their RKHS norm. For systems governed by differential equations, this formulation rigorously approximates the kernel mean embedding of the invariant Sinai-Ruelle-Bowen (SRB) measure, meaning our dissimilarity metric approximates the maximum mean discrepancy (MMD) between attractors. To accommodate general time series, we introduce time lags via Takens' Embedding Theorem to reconstruct the phase space. Crucially, CLOCK natively handles trajectories of unequal lengths and irregularly-spaced observations. We validate the proposed metric using spectral clustering, demonstrating superior performance against benchmark methods on both synthetic datasets and a real-world intracranial large-scale electroencephalogram (EEG) dataset. Finally, ablation studies on the Lorenz-96 system are consistent with the theoretical convergence rates of the occupation kernels and highlight the method's strong robustness to zero-mean Gaussian noise.

Fri 23 OctMachine Learning
The gist
Clustering different time series data from complex systems like weather or brain signals can be difficult, especially when the data varies in length and timing. The authors introduce a new method called CLOCK that represents these sequences in a special mathematical space to compare their patterns more accurately. This method works even when the data is irregular or noisy and can handle different lengths of sequences. They tested CLOCK using simulated models and real brain EEG data, showing it often groups similar time series better than other methods.
Open → 2610.33354v1

Observation assimilation transformer improves weather forecasts from sparse stations

Observation Assimilation Transformer: Probabilistic Weather Analysis from Sparse Station Observations

Abstract: We present the Observation Assimilation Transformer (OAT), a geometry-aware neural process that corrects and refines a prior numerical weather prediction (NWP) field into a high-resolution posterior field by assimilating sparse observations. OAT is an encoder-decoder transformer that considers a set of observed locations and their NWP prior to condition predictions at a set of arbitrary unobserved query locations. Its sparse attention mechanism makes it scalable to thousands of station observations and depends on relative positions between locations, never on the absolute coordinates. The model allows for partially missing observations, predictions at arbitrary continuous locations without grid projection, and outputs are parametric distributions. A station-randomized training algorithm is developed with continuous ranked probability score (CRPS) loss. We test OAT on two datasets - one at regional scale with hundreds of stations, and one at continental scale with thousands of stations. On these datasets OAT outperforms several baselines and shows well calibrated predictions. In an illustrative case study, OAT's downscaling capabilities are demonstrated by downscaling ERA5 (~31km) to COSMO (1km), where OAT achieves lower mean absolute error (MAE) and CRPS for temperature and precipitation on test stations compared to COSMO.

Tue 20 OctMachine Learning
The gist
Weather forecasts use data from many weather stations, but sometimes these observations are few and spread out. The authors created the Observation Assimilation Transformer (OAT), a model that improves weather predictions by smartly combining weather data and prior forecast information. It can make detailed, location-specific predictions even where there are no direct observations. Tests showed OAT gives more accurate and reliable weather details, for example turning coarse weather data into finer, more precise forecasts.
Open → 2610.28327v1

Permafrost temperature forecasting improved with spectral neural networks

Permafrost augmented simulation and forecast using hybrid modeling through spectral neural networks

Abstract: Purpose - Global warming is impacting several crucial Earth systems, including the permafrost. This study investigates the possibility of performing fast and reliable simulation and forecasting of permafrost behavior, addressing a critical gap in the existing literature related to high computation times and energy costs of the simulations. Methodology - A numerical cryohydrogeological study on water and energy transfer in a forested, continuous permafrost site has been conducted using parallel high-performance computing to simulate the permafrost dynamics. The numerical results reasonably replicate the measured soil temperature profiles and the dynamics of infiltrated waters. This work uses a spectral network to achieve a fast, reliable, and stable surrogate model for the simulations to forecast over long periods of time. The surrogate is also enhanced by assimilating experimental data through the development of a hybrid modeling framework. Findings - The proposed approach improves the physical simulation while maintaining the stability of the dynamic model over long periods of time. The results reveal that the proposed hybrid modeling remains stable in integration and forecasting. The final hybrid solution demonstrates a significant improvement in simulation accuracy when tested against a test set of measurements. Predictions of the permafrost behavior in the coming decade are later made based on the CMIP6 climate change scenarios. Originality - This research offers a novel perspective on permafrost modeling that extends the current understanding of its behavior under different climate change scenarios. It shows a higher-than-expected temperature increase of the permafrost. Moreover, the stable forecasting of climate change effects is demonstrated.

Sat 17 OctComputational Engineering, Finance, and Science
The gist
Permafrost, the permanently frozen ground found in cold regions, is warming due to climate change, but simulating these changes accurately takes a lot of time and computing power. The authors created a new method that uses fast spectral neural networks combined with real data to simulate and predict permafrost temperature changes reliably over long periods. Their approach not only speeds up predictions but also enhances accuracy and stability of the forecasts. This method projects a higher-than-expected increase in permafrost temperatures in the next decade under climate change scenarios.
Open → 2610.24665v1

3D turbulent flow fields reconstructed from coarse data with neural networks

High-fidelity Three-Dimensional Flow Field Reconstruction Driven by Turbulent Fluid Dynamics

Abstract: High-fidelity three-dimensional turbulent flow fields are essential for scientific and engineering analyses, yet obtaining them through direct numerical simulation or high-resolution measurements remains expensive. Super-resolution reconstruction from coarse-grained data therefore offers a promising alternative, but remains fundamentally challenging because fine-scale structures are intrinsically three-dimensional and dynamically coupled across scales. Existing data-driven approaches still struggle to recover multiscale volumetric structures and their coherent temporal evolution. To address these limitations, we propose a turbulence-aware framework for time-resolved three-dimensional flow reconstruction. The framework couples a multiscale neural operator, designed to recover volumetric structures, with a state space model to promote temporal consistency in the reconstructed dynamics. In addition, Navier-Stokes residuals are incorporated as soft constraints to regularize the reconstruction toward physically plausible dynamics. A mixture-of-experts mechanism is also introduced to support unsupervised adaptation to unknown coarse-graining. Across evaluations on canonical turbulent systems and atmospheric reanalysis data, the proposed framework reduces reconstruction errors relative to learning-based and classical baselines, while better preserving multiscale spatial structure and coherent temporal evolution. These results validate the effectiveness of the proposed framework in reconstructing high-fidelity three-dimensional flow fields from coarse-grained data.

Tue 13 OctArtificial IntelligenceMachine Learning
The gist
Simulating or measuring complex 3D turbulent flows with high detail is difficult and expensive. The authors created a method that can take rough, low-detail flow data and reconstruct the fine, detailed flow patterns over time using neural networks guided by physics rules. Their method better captures the flow’s three-dimensional shapes and how they change with time compared to previous approaches. This helps scientists and engineers study fluid flows more easily and cheaply.
Open → 2610.18045v1

Spectral deferred correction speeds up turbulent flow simulations

Space-Time Parallel Direct Numerical Simulation of Rayleigh-Bénard Convection

Abstract: Rayleigh-Bénard convection (RBC) is a model for turbulent flow and challenging to simulate by direct numerical simulation (DNS). Adequate resolution of all relevant length scales requires very fine grids in space and stability restrictions can dictate small step sizes. In this paper, we investigate the use of parallel spectral deferred correction (SDC) as a high-order time-stepping method to reduce wall-clock times of direct numerical simulation of RBC. Performance results on JUSUF (CPUs) and JUPITER (GPUs) with two different codes show parallel SDC can offer some concurrency along the temporal axis plus larger maximum stable time steps. Taken together, this allows for a reduction of wall-clock time required for simulations over one free-fall time by having to compute fewer steps and being able to mitigate the higher cost per time step through temporal parallelization.

Tue 13 OctDistributed, Parallel, and Cluster Computing
The gist
Simulating turbulent flows like Rayleigh-Bénard convection takes a lot of computer time because the simulations must capture detailed changes in both space and time. The authors show a way to speed up these simulations by using a special kind of time-stepping method called parallel spectral deferred correction, which allows several steps to be computed at once and supports bigger time steps without losing accuracy. They tested their method on powerful CPU and GPU computers and found it cuts down the total time needed to simulate the flow. This makes studying complex fluid motions more efficient.
Open → 2610.17942v1

Generative model improves data assimilation with big model errors

Accounting for model error in ensemble variational data assimilation through learned generative dynamics

Abstract: Many approaches have recently been proposed to combine data assimilation with machine learning for improved forecasts with geophysical models. Here, we focus on combining advanced sequential data assimilation schemes with a generative forecasting model which accounts for complex model errors. This generative model provides a stochastic approximation of the conditional distribution of the successor state given the current state. The proposed variational formulation is a novel one-step weak-constraint 4D-Var scheme in which a learned conditional stochastic forecast replaces the prescribed additive Gaussian model error representation. We derive and implement this formulation as an ensemble variational algorithm, the generative iterative ensemble Kalman filter (IEnKF-G), which generalises the iterative ensemble Kalman filter with additive Gaussian model error (IEnKF-Q). We extend the method to a smoother with an assimilation window spanning multiple time steps and propose a simplified variant. The IEnKF-G is evaluated on a stochastic variant of the Lorenz 1963 model to assess its robustness. It is then tested on a multi-scale variant of the one-dimensional Lorenz 1996 model, in which the resolved variables are affected by model error originating from the hidden small-scale variables. The algorithm is then compared to an ensemble Kalman filter that adaptively mitigates model and sampling errors, and to the IEnKF-Q. The IEnKF-G is generally more accurate in the tested configurations, with substantial gains under partial observations and significant model error. This demonstrates the potential of generative data-driven models when paired with classical data assimilation methods, particularly when model error is substantial.

Mon 12 OctMachine Learning
The gist
Forecasting complex systems like weather often involves combining observed data with computer models, but these models can make mistakes. The authors developed a new way to use machine learning to better represent these mistakes within a data assimilation method, which helps improve predictions. They replaced simpler error assumptions with a learned model that captures complicated errors more realistically. Tests on simulated weather models showed this approach makes forecasts more accurate, especially when only partial data is available or the model errors are large.
Open → 2610.17333v1

Sensitivity fidelity reveals true physical learning beyond forecast accuracy

Sensitivity Fidelity Reveals Physical Dependencies Beyond Forecast Accuracy in Deep-Learning Earth-System Models

Abstract: Deep-learning models increasingly rival numerical simulation in Earth-system forecasting, yet forecast error alone cannot determine whether their learned input-output dependencies are physically consistent. Here we introduce sensitivity fidelity as an additional criterion: alongside prediction error, model responses to input perturbations should be examined against a trusted reference or established physical knowledge. In an idealized Rayleigh-Bénard convection benchmark, four surrogate architectures achieved low forecast errors yet encoded markedly different sensitivities. Quantitative comparison with a matched adjoint distinguished the models far more sharply than forecast accuracy and showed that predictive skill and sensitivity fidelity need not coincide, while finite-perturbation experiments further linked these differences to the propagation of initial-condition uncertainty. We then applied this strategy to El Niño-Southern Oscillation prediction, where no adjoint reference exists and learned sensitivities showed event-wise robustness, validity under finite perturbations, and consistency with established precursors. These results identify gradient-based evaluation as a discriminating complement to forecast-error metrics, providing a practical test of whether data-driven Earth-system models learn physically credible dependencies.

Sun 11 OctMachine Learning
The gist
Deep-learning models can predict events like weather patterns well, but having a low error in predictions doesn't mean the model understands the true physical relationships involved. The authors introduced a way called sensitivity fidelity that checks how models respond to small changes in input compared to trusted references or known physics. They found models with similar prediction scores could behave very differently in these sensitivity checks. This method helps to verify if AI models learn meaningful physical dependencies and can improve trust in their forecasts.
Open → 2610.15540v1

Refined grid sampling speeds up neural modeling of scientific data

Refined Grid Sampling for Efficient Implicit Neural Representations of Large-Scale Scientific Data

Abstract: Implicit neural representations (INRs) offer a powerful framework for modeling high-dimensional physical fields, but their training remains inefficient because evaluating dense coordinate grids is computationally expensive. Existing coordinate-pruning strategies rely on local error functions that can oversample correlated points and incur substantial overhead from full-grid error evaluation. We introduce Refined Grid Sampling (RGS), which combines stratified grid sampling, error-based refinement, and dynamic grid-resolution adjustment. This design reduces spatial redundancy without evaluating errors over the full grid. On high-resolution Navier--Stokes fields, RGS achieves higher reconstruction accuracy within a fixed wall-clock budget and reaches the accuracy of full-grid training up to $6\times$ faster than the evaluated baselines.

Fri 9 OctMachine Learning
The gist
Training neural networks to represent complex physical data usually takes a lot of time because computers must check many points in space. The authors introduce a method called Refined Grid Sampling that smartly samples points without checking the entire grid each time. This new approach reduces wasted effort and helps the neural network learn more accurately and faster. They show their method works well on detailed fluid flow data, achieving similar accuracy much quicker than previous techniques.
Open → 2610.14205v1

Benchmark compares feature selection methods on real data

SelectArena: A Modern Benchmark for Feature Selection on Real-world Tabular Data

Abstract: Selecting a subset of features (columns) from a tabular dataset is a ubiquitous challenge in real-world applications. Industry experts seek to minimize model training costs, climate scientists require an interpretable subset of factors to understand extreme weather events, and clinicians need to choose among diagnostic tests due to financial and time constraints. When looking at the scientific literature to determine which feature selection method to use, practitioners across domains face an unrepresentative landscape of benchmarking. Despite extensive research, existing feature selection benchmarks provide limited guidance because they rely on outdated models, ad hoc evaluation protocols, or artificially generated data that do not reflect real-world problems. To close the gap between scientific benchmarking and real-world applications, we introduce SelectArena, a modern benchmark for feature selection. We compare 15 widely used feature selection methods on 71 manually curated real-world tabular datasets, covering classification and regression problems, using standardized evaluation protocols and state-of-the-art tabular (foundation) models. Our results reveal clear trade-offs across predictive performance, runtime, stability, and validity. Tree-based methods, CART and RFImportance, along with F-Test are Pareto-optimal when taking into account predictive performance and runtime. F-Test, information-theory-, and distance-based methods lead on stability and validity, with F-Test emerging as a balanced choice across our metrics. Search-based methods are slow, unstable and achieve low performance. SelectArena turns these findings into practical guidance and, as an extensible platform, supports future feature selection research.

Fri 9 OctMachine LearningArtificial Intelligence
The gist
Choosing the right features from data tables is important in many fields, but current tests of the best methods often use outdated or fake data. The authors created SelectArena, a new benchmark using real-world datasets to fairly compare 15 common feature selection methods. They found certain methods are better at balancing accuracy, speed, and consistency, helping users pick the right approach for their needs. This benchmark can also be used to test new methods in the future.
Open → 2610.13644v1

Kernel method improves learning for nonlinear system forecasts

Bilevel optimization for data-driven learning of Koopman embeddings using kernel-based autoencoders

Abstract: Koopman operator theory provides a linear framework for analyzing nonlinear dynamical systems and has become a major tool for data-driven modeling. A central challenge, however, is that finite-dimensional approximations computed by methods such as extended dynamic mode decomposition (EDMD) require the dictionary to be specified a priori. Recent machine-learning approaches address this limitation by learning the dictionary from data, predominantly using artificial neural network (ANN) autoencoder architectures. Although kernel methods offer an alternative with greater interpretability and tractability for theoretical analysis, they have received little attention in this setting. We introduce extended dynamic mode decomposition with kernel-based dictionary learning (EDMD-kDL), a kernel-based method for learning finite-dimensional Koopman embeddings directly from data. The method combines ideas from collocation methods and bilevel optimization to simultaneously learn a kernel dictionary and the corresponding Koopman approximation. We evaluate EDMD-kDL against state-of-the-art ANN-based approaches on a range of numerical experiments, including global sea-surface-temperature forecasting and learning directly from video data. Across all tested settings, EDMD-kDL achieves performance comparable to or better than the ANN-based methods. Moreover, in contrast to standard kernel methods, the proposed approach is scalable to large datasets by design since the size of the required kernel matrices depends on the number of collocation points rather than the size of the training dataset.

Thu 8 OctMachine Learning
The gist
It can be hard to make predictions about complex systems that change in nonlinear ways, like the weather or ocean patterns. The authors created a new way to learn simpler linear models from data using kernel methods, which can be easier to interpret than neural networks. Their approach learns the important patterns directly and works well compared to existing methods, even on large datasets like video or sea temperature data. This makes it easier to build accurate and scalable models to understand and forecast nonlinear systems.
Open → 2610.12370v1

Improved Gaussian process flows for easier and faster probabilistic modeling

Feature Space Adaptation for Effortless Gaussian Process Flows

Abstract: Outside the linear-Gaussian regime, conditional sampling from Gaussian processes (GPs) is challenging. Recent methods such as FlowGP (Moss et al., (2026)) can condition on arbitrary non-linear and non-Gaussian statements, but at considerable cost: an expensive iterative and high-dimensional diffusion that requires hand-specified kernel hyperparameters. In this paper, we alleviate two significant drawbacks of FlowGP by (1) introducing kernel approximations that enable scaling to high-resolution domains and (2) proposing a way to obtain the marginal likelihood by measuring the work needed to steer the diffusion towards conditioning statements. We enable, for the first time, hyperparameter optimisation within FlowGP and demonstrate our approach on probabilistic downscaling from areal summary statistics, PDE solution inference on irregular domains, and recovery of sea level anomaly fields from non-Gaussian satellite observations.

Thu 8 OctMachine Learning
The gist
Gaussian processes are a way to make predictions about complex data, but they become hard to use when the data is not simple or linear. The authors improve an existing method called FlowGP by making it faster and easier to tune, so it works better with complicated, real-world data. They do this by introducing new math tricks that speed up the process and let the system learn important settings automatically. Their approach helps with tasks like refining coarse geographical data and filling in missing information from satellite images.
Open → 2610.11459v1

AI agents build competitive El Nino climate models from data

SciExam for ENSO: Can AI Agents Build Climate Models?

Abstract: Language-model agents are increasingly asked to carry out open-ended scientific research, yet their results are usually graded against a known answer, a rubric, or a language-model reviewer, none of which can tell whether a new scientific model is valid. The AI Science Exam for El Nino-Southern Oscillation (SciExam for ENSO) is a benchmark in which agents build low-order stochastic models of ENSO, the dominant mode of interannual climate variability, from real observations. Within a six-hour budget, agents process the observations, write their own diagnostics, which are then frozen, and develop a model using only these diagnostics as feedback. Hidden graders then test whether the model reproduces ENSO's statistics, recovers unobserved variables, and forecasts held-out years, and score a published model in the same way. Across twelve agent systems, six produce models that score higher than the published model, mainly through better reconstruction and forecasting. The simplified forms of the stronger models are each compatible with one of the two competing explanations of ENSO's warm-cold asymmetry, an open debate that the task never mentions. Controlled runs of the top system under varied information suggest that its scores do not come from recalling the dated observational record and that the information it receives shapes how it builds its model. SciExam for ENSO can thus evaluate agent research where no answer is known, and the results suggest that agents can already build competitive models whose structures bear on questions that scientists still debate.

Wed 7 OctArtificial IntelligenceMachine Learning
The gist
Predicting El Nino, a major climate pattern, requires building complex models from observations. The authors created a test where AI agents try to build these models themselves using only real climate data and diagnostics they generate. Some AI agents made models that were better at forecasting and matching El Nino’s behavior than a known published model. The results show AI agents can create useful climate models and reveal different explanations for El Nino’s patterns that scientists are still debating.
Open → 2610.10513v1

Neural pde solvers mimic numerical scheme errors not exact physics

The Symbol of the Surrogate: Measuring Numerical Provenance in Neural PDE Solvers

Abstract: Neural PDE surrogates are trained on numerical solver outputs that contain both physical evolution and solver-specific discretization errors. Because surrogates are also evaluated against held-out trajectories from the same solver, standard benchmarks cannot distinguish fidelity to the exact evolution from imitation of the numerical scheme. We introduce an empirical Fourier-symbol diagnostic that probes a trained surrogate's linearized one-step operator with individual Fourier modes and compares it with both exact-evolution and training-scheme references. To address architectural spectral bias, we train identical networks on schemes with orthogonal dissipative and dispersive signatures and compare their learned operators. In linear advection, the learned surrogates reproduce the training schemes' amplitude and phase errors, with the twin-scheme difference reaching more than 99.8\% of the analytically predicted full-imitation ceiling. The same behavior occurs for a non-local Fourier neural operator and at the operator level for nonlinear Burgers dynamics. These results show that agreement with solver-generated test data does not by itself establish fidelity to the exact evolution. Fourier-symbol measurements provide a direct diagnostic of numerical provenance.

Wed 7 OctMachine Learning
The gist
Neural networks meant to predict physical changes often learn the quirks of the computer methods used to generate their training data, not the true underlying physics. The authors show that evaluating these networks on similar data doesn't reveal this problem. They introduce a technique using Fourier analysis to test whether the networks've learned the actual physics or just the numerical shortcuts. Their tests reveal the networks mostly copy the numerical solver's behavior, not the exact physical evolution.
Open → 2610.09255v1

AI emulators tested for correct response in chaotic system simulations

A Response Theory Probe for Learned Stochastic AI Simulators, Tested on Lorenz-63

Abstract: Machine-learning emulators of chaotic and stochastic systems are usually validated on forecast skill and long-run statistics. Neither certifies that an emulator responds correctly to forcing, the property that projection and attribution studies rely on. Linear response theory makes this testable: the forced response follows from unperturbed correlations through a generalized fluctuation-dissipation relation, and decomposes over the stochastic Ruelle-Pollicott resonances of the Koopman generator. Building on the Koopmanism Response framework, we turn this into a calibrated, mode-resolved test for learned surrogates: each surrogate rollout passes or fails each check, and failure rates are compared with those of independent realizations of the true system. On stochastic Lorenz-63, a three-variable toy model, we evaluate SINDy, an MLP, a reservoir computer, a neural ODE and a neural SDE with learned diffusion, over up to 80 rollouts each. A sparse-regression model with the correct library passes every check at rates consistent with the true system. Invariant-statistics fidelity and response fidelity dissociate in both directions: a quarter of reservoir-computer rollouts pass every invariant-statistics check and match the static susceptibility $χ(0)$, yet misrepresent the slow relaxation modes, while the neural ODE and SDE rarely meet the invariant-statistics floor but recover those modes in three quarters of rollouts. As expected of a time-integrated quantity dominated here by fast relaxation, $χ(0)$ does not separate these cases. For a fixed network, the training formulation (one-step drift, flow map, or multi-step through the integrator) decides which of these properties it gets right.

Mon 5 OctMachine Learning
The gist
Predicting chaotic systems like weather can be done using AI models, but it’s tricky to know if they react correctly when conditions change. The authors developed a new way to test if AI simulators respond properly to changes using theory that links their natural behavior to forced behavior. They tested several AI models on a simple chaotic example (the Lorenz-63 system) and found that some models that look good in normal tests still fail when forced, while others capture forced behavior better but miss traditional checks. Their method helps identify which AI models truly behave like the real system under different conditions.
Open → 2610.06798v1

Transformer attention improves predictions on spherical data efficiently

SO(3)-RoPE for Spherical Transformers

Abstract: Spherical data arise in many scientific applications. Often spherical transformers disregard the geometry of the underlying spherical domain, causing distortions and coordinate singularities near the poles. We introduce SO(3)-RoPE, a relative positional embedding that incorporates spherical geometry into transformer attention through unitary SO(3) representations. Our formulation is SO(3)-equivariant and compatible with FlashAttention, retaining efficiency of vanilla transformers. On shallow water dynamics prediction over a rotating sphere, our SO3ViT outperforms an S2Transformer baseline with lower errors and reduced runtime.

Mon 5 OctMachine LearningArtificial Intelligence
The gist
Predicting things on spherical surfaces, like the Earth, is tricky because normal computer methods don’t handle the sphere’s shape well, especially near the poles. The authors developed a new way to make transformers, a type of AI model, understand the sphere’s geometry better without slowing down calculations. Their method, called SO(3)-RoPE, respects the sphere's rotations and improves both the speed and accuracy of predictions in simulations of shallow water over a rotating planet. This could help models that work with spherical data avoid common errors caused by ignoring the sphere’s shape.
Open → 2610.06229v1

Neural operators uphold exact physics laws for better transport modeling

Poisson-GENERIC Neural Operators: Exact Metriplectic Structure in Function Space via Casimir Entropies

Abstract: Existing thermodynamically consistent neural operators impose the GENERIC degeneracy conditions by projecting the reversible operator onto the complement of the entropy gradient. This makes the operator state-dependent and forfeits the Jacobi identity, so the result is metriplectic-degenerate rather than metriplectic. We instead obtain degeneracy the way GENERIC does. For nonlinear transport, the reversible operator $L$ is the compatible Lie-Poisson pencil $αD+λ(uD+Du)$; otherwise it is a constant, trivially Poisson Fourier multiplier. On an augmented state $(u,s)$ with a latent entropy density, $S=\int s$ is a Casimir of $L$, so $L\,δS/δz=0$ holds identically without projection. The energy combines a fixed mechanical quadratic, a learned gauge-free potential, and a convex internal energy. The friction operator $M=AA^\top$ satisfies $M\,δE/δz=0$ pointwise, and its Onsager parity structure permits diffusion and damping while provably excluding transport. For any parameters, skewness, positivity, both degeneracies, and the Jacobi identity (on the resolved band for the Lie-Poisson term) hold to machine precision. Heat conduction and damped waves admit exact closed-form friction operators, the second law bounds physical energy under a checkable curvature condition, and a discrete-gradient integrator yields exact discrete first and second laws. On four PDEs in 1D and 2D with three backbones (FNO, Transolver, CNO), the model wins 61 of 72 seed-level comparisons against same-backbone unconstrained baselines, learns the exact transport and wave symbols, matches the true dissipation rate within 13% on heat and Burgers, and dissipates nothing on advection. A constant-$L$ ablation isolates the cost of exact Jacobi as the loss of Burgers, while a learned-entropy ablation injects energy on every reversible-irreversible problem.

Sun 4 OctMachine Learning
The gist
Modeling processes like heat flow and wave motion often involves complex physics rules that need to be exactly respected for accuracy. The authors developed a new kind of neural network operator that strictly follows these physics laws in function space, preserving energy and entropy relationships without shortcuts. This approach improves the quality of predictions on several partial differential equations, ensuring correct transport and dissipation behaviors. It also provides ways to integrate these systems over time while keeping physical laws intact.
Open → 2610.05570v1

Stochastic adaptive spectral methods improve predictions for complex equations

Stochastic Adaptive Fourier Decomposition for Operator Learning

Abstract: Fourier Neural Operators (FNOs) offer an efficient paradigm for solving partial differential equations (PDEs). However, FNOs rely on a fixed Fourier basis and hard frequency truncation, which inherently limit their ability to model non-periodic, localized, and fine-scale solution structures. We propose the Stochastic Adaptive Fourier Decomposition Neural Operator (SAFDNO), a spectral neural operator that replaces predefined Fourier modes with an adaptive Takenaka-Malmquist (TM) orthonormal system derived from the theory of Stochastic Adaptive Fourier Decomposition (SAFD). Instead of performing an expensive greedy pole search in classical SAFD, SAFDNO amortizes stochastic pole selection through a neural pole predictor and constructs the adaptive TM system directly from latent features. The resulting operator performs learned filtering on coefficients of analytic branches in the Hardy space with respect to the input-adaptive TM system, preserving the global receptive field of spectral operators while providing a more flexible representation than using fixed spectral bases. Across nine PDE benchmark problems, SAFDNO achieves the best performance on all six regular-grid problems among strong neural operator baselines, with especially notable gains on Darcy, Burgers, and Navier-Stokes, where fixed Fourier modes are often less effective at modeling localized oscillations, sharp transitions, and multiscale structures. SAFDNO also exhibits stronger zero-shot super-resolution performance and shows less performance degradation when deployed on finer discretizations. These results suggest that the input-adaptive TM system provides a promising alternative to fixed spectral representations for neural operator learning.

Sat 3 OctMachine Learning
The gist
Predicting how physical systems behave, like weather or fluid flow, often involves solving complex equations called partial differential equations (PDEs). Previous methods used fixed patterns (Fourier modes) to simplify these predictions but struggled with details like sharp changes or local effects. The authors propose a new approach, SAFDNO, that adapts these patterns based on the input, allowing more accurate modeling of tricky features. This method works better on a variety of test problems and can handle higher-resolution data without losing accuracy.
Open → 2610.04241v1

Physics based operators speed up long term chaotic system forecasts

Physics is the Best Teacher: Consistency Learning for Time-Invariant Operators of Chaotic Dynamics

Abstract: Accelerating the prediction of long-term behavior in chaotic systems is crucial in scientific computing. However, existing methods rely on numerical solvers or autoregressive models that advance one small step at a time, which makes long horizons expensive. We instead view this problem as learning the system's time-invariant evolution operator, which jumps the state across a large time span in a single evaluation. To this end, we derive the consistency equations a time-invariant operator must satisfy, with differential and compositional objectives in physical time. These equations also connect the learned operator to the physics-prescribed instant dynamics, enabling physics embedding in consistency learning. Across five chaotic systems, we find that physics-distilled consistency makes both short-term trajectories and long-term statistics more accurate. The learned operator survives temporal extrapolation and requires one-tenth as many evaluations as autoregressive rollout, offering an efficient route to long-term simulation of chaotic dynamics.

Fri 2 OctMachine Learning
The gist
Predicting how chaotic systems behave over a long time usually takes many small steps, which is slow. The authors propose teaching a computer model to jump forward big time steps in one go by learning rules that always apply to the system. They connect these rules to known physics laws to make the predictions more accurate. This new method predicts complex systems faster, using fewer steps while keeping good accuracy.
Open → 2610.04108v1

Climate downscaling methods vary in preserving spatial accuracy and detail

Beyond Pointwise Error: A Multi-Metric Evaluation of Spatial Climate Downscaling

Abstract: Climate downscaling aims to reconstruct fine scale spatial fields from coarse resolution inputs. Evaluating the quality of these reconstructions is challenging: low pointwise error can come at the cost of fine scale variability, while realistic spatial variability can be achieved with inaccurate local structures. The evaluation metric can therefore change which method appears to perform best. This work presents a multi metric benchmark comparing five spatial downscaling methods on ERA5 temperature, wind, and precipitation fields. Five criteria assess complementary properties: pointwise error, structural similarity, distribution error, spectral error, and gradient error. The results reveal a systematic trade off between spatial fidelity and fine scale variability. Some methods perform best on pointwise and spatially aligned metrics, but lose high frequency content, while others preserve substantially more spectral variability at the cost of less accurately positioned local structures. Consequently, method rankings change across metrics and variables. These results show that there is no single best downscaling method. Multi metric evaluation is therefore essential for assessing which properties of a climate field are preserved.

Thu 1 OctMachine LearningArtificial Intelligence
The gist
Climate downscaling tries to make detailed local weather information from rough, large-area data. The authors show that different ways of doing this have trade-offs: some get local details right but miss subtle patterns, while others keep more natural variation but aren’t perfectly accurate at specific spots. They tested five methods on temperature, wind, and rain data and found no single best solution. They suggest using multiple evaluation methods to judge which downscaling approach works best for different needs.
Open → 2610.01579v1

Cascade model architecture reduces errors in long term neural predictions

CasEm: A Cascade Architecture for Long-Horizon Neural Emulation

Abstract: Autoregressive neural emulators can drift or diverge over long rollouts despite accurate short-term predictions. We introduce Cascaded Emulation (CasEm), a one-way rollout architecture that augments an existing full-state backbone with an independently evolving model of physically specified aggregates. Its forecasts guide corrections to full-state predictions, without feedback from the backbone to the aggregate model. Effective guidance requires aggregates that cover substantial backbone error, remain accurately predictable, and support useful full-state corrections. We derive a finite-horizon error bound that clarifies these three factors and use empirical diagnostics to guide subsystem selection. Across four ODE/PDE benchmarks, CasEm reduces long-horizon rollout errors across diverse backbones and suppresses the trend toward error divergence in both diffusion tasks using Fourier neural operator backbones. In global climate emulation, CasEm with a regional total-water subsystem reduces 10-year full-state time-mean error by 66.6% and 46.3% for frozen ACE and Spherical DYffusion backbones, respectively, while adding less than 3% to inference time.

Mon 28 SeptMachine Learning
The gist
Neural networks that predict future states over long periods can make bigger and bigger mistakes as they go. The authors introduce a new method called CasEm, which adds a second model focusing on key summary features to help correct these errors during long predictions. This second model runs separately and guides corrections without interfering with the main model’s short-term accuracy. Their approach cuts down mistakes in various physics simulations and climate models, making long-term forecasts more reliable.
Open → 2609.34246v1

Generative weather models need new evaluation approaches to reveal quality

StatD2GAN: When Calibration Masks Generator Quality in Held-Out Evaluation of Synthetic Weather Sequences

Abstract: Generative models for multivariate weather series are routinely evaluated with pooled distributional metrics computed after marginal calibration. We show this practice can invalidate architectural conclusions, and rebuild the evaluation of StatD2GAN, a three-discriminator GAN with evolutionary weight adaptation, around a held-out protocol: the final two calendar years of each dataset are held out behind a 168 hour embargo, calibration is fitted on the training block only, and all metrics are computed on the held-out block. Evidence comes from 25 matched (location, seed) pairs across five Koppen-Geiger climates, tested with Wilcoxon signed-rank tests under Holm correction. Four results follow. First, isotonic calibration drives the Kolmogorov-Smirnov distance to within 2% of a per-location noise-and-shift floor for every architecture tested, including a deliberately weak RCGAN baseline, so calibrated marginal metrics cannot discriminate between architectures. Second, the sorted-representation discriminator is the only component whose removal significantly degrades cross-variable dependence (Kendall tau MAE +0.080, Holm p = 0.009), with a regime-dependent effect: near zero in Ankara, above 115% in Dubai and Yakutsk. A rank-transformed variant isolates the mechanism as quantile supervision of the marginals rather than copula matching. Third, physical constraint violations are injected by calibration, not the generator; projection removes them at negligible cost (deltaKS <= 0.003). Fourth, pooled metrics conceal a collapse of between-sequence weekly-mean variability, a proxy for seasonal and regime diversity, in TimeGAN that only sequence-level statistics expose. We recommend floor-referenced marginal evaluation, matched-pair testing, and sequence-level variance decomposition as minimum requirements for calibrated generative pipelines.

Sun 27 SeptMachine Learning
The gist
Synthetic weather data models are often checked by comparing individual weather measurements, but this can hide how well the models capture complex weather patterns. The authors found that calibrating these models inaccurately makes many quality measures look similar, even when the models are very different. They suggest testing on separate time periods without mixing data and looking at whole weather sequences to better judge model performance. Their approach helps spot when models fail to capture seasonal changes or relationships between weather factors.
Open → 2609.33761v1

Aurora-X improves forecasting accuracy across many time series types

Aurora-X: Built for Extreme Time Series Forecasting

Abstract: Time series foundation models (TSFMs) enable cross-domain forecasting, but their development as general-purpose forecasters remains constrained by underexplored training potential and limited architectural versatility. To address these challenges, we introduce Aurora-X, a billion-scale TSFM with a progressive curriculum and a unified architecture. We first use channel-independent pretraining to learn temporal patterns, then introduce cross-variable dependencies, varied context and horizon lengths, and future covariates if available during midtraining. Variable-resolution post-training further enables an adjustable temporal span per token at inference. With fixed model weights, this supports longer histories under a fixed token budget or fewer tokens for the same history, enabling test-time scaling. With a versatile architecture, Aurora-X supports cross-variable modeling, covariate conditioning, and parallel decoding of future patches for probabilistic forecasting. These are supported by a novel pattern-guided mixture-of-experts that expands model capacity through sparse activation and uses shallow patch similarities to constrain deep-layer routing, guiding expert specialization across heterogeneous time series. Furthermore, we propose an implicit quantile network head that predicts arbitrary quantiles to characterize predictive distributions, enhancing probabilistic forecasting flexibility. Comprehensive experiments on GIFT-Eval, TIME, FEV-Bench, TFB, and DAG-Bench demonstrate state-of-the-art forecasting performance against pretrained TSFMs and task-specific supervised models.

Fri 25 SeptMachine Learning
The gist
Time series forecasting tries to predict what will happen next based on past data, like weather or stock prices. The authors introduced Aurora-X, a very large model designed to handle many different types of time series data better than before. It learns patterns step-by-step, including relationships between variables and different time lengths, and it can adapt to using longer or shorter past data when making predictions. Aurora-X also predicts probabilities of future values, providing a fuller picture of uncertainty. Tests show it usually forecasts more accurately than earlier models.
Open → 2609.31038v1

Machine learning models predict climate effects with causal insights

Learning Hierarchical Causal Representations of the Effects of Forcings on Temperature in Climate Models

Abstract: Machine learning (ML) emulators provide a fast and cost-effective method to simulate climate change scenarios after being trained on Earth System Models projections. However, the black-box nature of those data-driven approaches limit the usability and trustworthiness of their outputs and in particular their use as causal attribution tools. Here, we develop a hierarchical causal representation learning framework applied to sea surface temperature fields from a state-of-the-art global climate model. As a key advance over previous work, our framework explicitly models both atmospheric dynamical interactions arising from internal climate variability and forced responses due to changes in atmospheric greenhouse gas and aerosol concentrations. When trained on future climate change scenarios, our method accurately predicts the long-term global mean and regional temperature evolution and shows physically realistic responses to perturbations in greenhouse gas and aerosol concentrations when evaluated on unseen scenarios. Our results underline the potential of causal representation learning frameworks for advancing climate model emulation.

Fri 25 SeptMachine Learning
The gist
Climate models are complex and slow to run, so people use machine learning to simulate future temperature changes faster. The authors developed a new machine learning method that not only predicts temperature changes but also explains how greenhouse gases and aerosols cause these changes. Their method works well on data it hasn’t seen before and matches expected physical behaviors. This approach could help improve trust and understanding of climate model predictions.
Open → 2609.30995v1

Lightweight models improve finer scale climate data prediction accuracy

Lightweight Probabilistic Downscaling from a Deterministic Base Model

Abstract: Climate data downscaling is the task of increasing the spatial resolution of climate data, typically by generating fine-resolution regional climate data from coarse global model output. Recent machine learning (ML) work in the related task of weather forecasting has seen significant improvements due to newly devised training methods and architectural components, but these have not yet benefited downscaling. We adapt two of these methods to create a family of lightweight probabilistic ML downscaling models built on a modified U-Net backbone and evaluate them on the CORDEX-ML-Bench suite for daily maximum temperature and precipitation across three geographic regions: the Alps, New Zealand and South Africa. We find that a two-stage training curriculum, combining deterministic pretraining with probabilistic tuning, transfers well to downscaling, beating the state-of-the-art for RMSE. Our work provides an advancement towards lightweight, probabilistic downscaling models, reducing the current trade-off between computational intensity and distributional fit.

Thu 24 SeptMachine Learning
The gist
Climate models often predict weather and climate at a large scale, but it's useful to have detailed local predictions too. The authors developed smaller, efficient machine learning models to increase the detail of climate predictions, working from coarse global data. They blended two training steps—starting with simple prediction and then tuning for uncertainty—to get better, more accurate results across regions like the Alps and South Africa. Their approach balances accuracy and computational cost better than previous methods.
Open → 2609.29383v1

Functional dynamic mode decomposition learns infinite dimensional systems from data

Functional dynamic mode decomposition: Learning infinite-dimensional systems from data

Abstract: Dynamic mode decomposition (DMD) is a data-driven method that computes the best linear approximation of the underlying dynamical system and decomposes the dynamics into a superposition of characteristic spatiotemporal patterns. Originally introduced by the fluid dynamics community, DMD and its extensions have found widespread use in many other research areas such as molecular dynamics, climate science, engineering, finance, and neuroscience. Applications include dimensionality reduction, forecasting, system identification, control, and spectral clustering. In order to apply DMD to partial differential equations, the spatial domain is typically first discretized using finite difference or finite element techniques, thus implicitly rendering the problem finite-dimensional. We extend projected and exact DMD to infinite-dimensional systems. Rather than estimating matrices from vector-valued observations, our DMD variants learn finite-rank operators from functional data such as observables, densities, or wavefunctions. We show that conventional DMD algorithms can be regarded as special cases of their functional DMD counterparts. All results will be illustrated with the aid of guiding examples. We focus in particular on Koopman, Perron-Frobenius, and Koopman-von Neumann operators associated with graphons, ordinary differential equations, and stochastic differential equations.

Thu 24 SeptMachine Learning
The gist
Dynamic mode decomposition (DMD) is a technique to simplify and understand complex changing systems by breaking them into patterns. The authors extend DMD to work directly with infinite-dimensional data, like continuous functions, rather than limiting it to simpler finite data. This lets them analyze systems modeled by equations involving continuous spaces without approximating them as finite sets. Their approach connects to important mathematical operators used to describe dynamics in areas like networks and random processes, helping to generalize DMD for broader scientific use.
Open → 2609.29159v1

Physical field method improves AI for predicting complex physics systems

PhyMo: A Physical-Field Modality for Multimodal AI4Physics

Abstract: Multimodal learning is emerging as a powerful paradigm for AI for Physics (AI4Physics), where predicting physical systems requires the joint interpretation of heterogeneous observations, measurements, and domain knowledge. However, existing approaches typically represent physical quantities and governing equations as generic numerical or textual tokens, overlooking the physical constraints that determine their spatiotemporal interactions. To address this limitation, we introduce the \textbf{physical-field modality} and propose \textbf{PhyMo}, a physics-grounded multimodal framework that organizes heterogeneous measurements through PDE-associated operators. PhyMo follows a three-stage learning procedure: the physical-field encoder is first pretrained through field reconstruction under PDE residual supervision, its representations are subsequently aligned with visual embeddings in a shared latent space, and the fused multimodal representations are finally processed by corresponding downstream prediction heads. Experiments on five datasets spanning diverse physical environments show that PhyMo achieves state-of-the-art performance, compared to the strongest baseline on each dataset, demonstrating the superiority of PhyMo on multimodal representation learning in AI4Physics.

Wed 23 SeptMachine LearningArtificial Intelligence
The gist
Predicting how physical systems behave often requires combining different types of data and knowledge. Existing AI methods treat physical measurements and equations as simple numbers or text, missing important physical rules that guide how these parts interact in space and time. This work introduces a new way to represent physical data called a physical-field modality, which respects these physical rules. The researchers built PhyMo, a system that first learns to understand physical fields, then aligns this knowledge with visual data, and finally uses the combined information to make better predictions. Tests on multiple physics problems show that PhyMo outperforms previous methods in understanding and predicting physical phenomena.
Open → 2609.27554v1

Time series forecasts improve by learning from past outcomes

When Tomorrow Becomes Today: Self-Evolving Policies for Agentic Time-Series Forecasting

Abstract: Agentic time series forecasting concerns systems whose underlying mechanisms evolve, making the relative effectiveness of numerical models, reasoning strategies, and intervention rules inherently time-varying. Consequently, a time series agent must adapt the forecasts it produces and the orchestration policy that determines which components to trust and how to coordinate them. The deployment process naturally provides supervision for this adaptation as forecast horizons elapse and realized targets reveal the effectiveness of earlier decisions. Committing all numerical expert forecasts and candidate agent paths before target observation allows each realized outcome to evaluate the entire alternative set, providing delayed feedback without additional annotation. However, existing time series agents primarily incorporate prior experience through forecast refinement, reflection, or retrieval, without systematically converting realized outcomes into persistent updates to the joint orchestration policy governing later origins. To exploit this delayed feedback systematically, we introduce TimEvolve, a frozen-backbone time series agent that converts each realized outcome into persistent joint updates of expert trust, agent path selection, and intervention strength. A temporally ordered predict, reveal, and update protocol applies this feedback to subsequent forecasts. Experiments across eight Time-MMD domains show that TimEvolve achieves the best average MSE and MAE ranks among fifteen methods and the lowest errors on both metrics in seven domains. These results demonstrate the value of learning forecasting policies from the futures encountered during deployment.

Mon 21 SeptMachine LearningArtificial Intelligence
The gist
Predicting future trends can be tricky when the system itself changes over time, making some methods better at certain moments. The authors introduce TimEvolve, a new forecasting system that learns from actual past results to improve future predictions and decides which forecasting methods to trust based on experience. Unlike earlier approaches, TimEvolve adapts its strategy continuously by using feedback from real outcomes to update how it combines models and interventions. Tests show it outperforms many existing methods across several scenarios where the underlying systems evolve.
Open → 2609.24862v1

Next generation reservoir computing infers missing parts of complex systems

Inference of Unknown Dynamical Components Using Next Generation Reservoir Computing: From Chaotic Systems to Climate Data

Abstract: We investigate next generation reservoir computing (NGRC) as a data-driven approach for inferring unseen components of dynamical systems. We compare NGRC with traditional reservoir computing (RC) using the Lorenz and Rössler system, where two unknown components are inferred from one given component. For both systems, NGRC achieves accurate results while requiring fewer training data and less computational time than RC. We identified an inverse proportional behavior between the number of time-delayed steps needed for NGRC and the temporal resolution, indicating that the physical time span covered by the delay interval is an important factor in determining the required number of delayed steps. Finally, we apply NGRC to the observational climate data of ENSO (El Niño--Southern Oscillation) and infer one observable from the remaining variables. Despite the noise and complexity of the real-world data, the NGRC shows promising results. Our findings demonstrate the potential of NGRC for efficient inference of unseen components in both controlled dynamical systems and real-world data.

Mon 21 SeptMachine Learning
The gist
Many systems, like the weather or climate, have parts we cannot directly observe, making prediction challenging. The authors used a method called next generation reservoir computing to guess missing parts of these systems from available data. They tested this on well-known chaotic systems and real climate data, showing it works well using less data and time than older methods. This approach might help better understand and predict complex natural phenomena by filling in unknown details.
Open → 2609.24754v1

Chronosphere adapts climate data detail across space and time

Chronosphere: Space-Time Tessellation of Local Climate Experts

Abstract: We introduce Chronosphere, a spatio-temporal neural field that learns representations of climate. A central challenge in geographic representation learning is modeling environmental processes whose spatial and temporal complexity varies widely. Yet existing location encoders typically fix a single level of detail everywhere. Global bases such as spherical harmonics spread capacity uniformly across space and time. Localized bases resolve only predefined regions. Learned tessellations adapt, but are inefficient at representing higher frequencies. Chronosphere unifies these approaches, pairing an adaptive tessellation of learnable sites on the spacetime torus $S^2\times S^1$ with a shared bank of local basis functions. Both where capacity is placed and how much detail each region carries adapt to the data, across space and time. Trained to reconstruct climatology, Chronosphere matches or leads state-of-the-art location encoders across spatial and temporal tasks, with the largest gains under spatial and temporal transfer.

Fri 18 SeptComputer Vision and Pattern RecognitionMachine Learning
The gist
Representing climate data is challenging because weather and climate change in complex ways across different places and times. The authors created Chronosphere, a machine learning method that divides space and time into adaptable regions, each capturing the right amount of climate detail. This approach combines local and global methods to better focus on important areas and times, improving how well climate information is understood and predicted. Their method outperforms existing ones, especially when applied to new locations or times.
Open → 2609.21872v1

OceanMoE improves long-term ocean variable forecasting with adaptive models

OceanMoE: Structured Conditional Sparse Computation for Long-Horizon Multivariate Ocean Forecasting

Abstract: Multivariate ocean forecasting must exploit shared evolution in a coupled ocean system while adapting to the heterogeneous statistical and dynamical characteristics of different prediction variables and locations. Fully shared models may lack the flexibility to handle this heterogeneity, whereas fully independent models discard the common ocean context shared across variables. The key question is how to retain shared context in a unified model while allowing computation to specialize according to the prediction target and local state. We propose OceanMoE, a structured conditional sparse Mixture-of-Experts framework that combines sharing and specialization for multivariate ocean forecasting. OceanMoE fuses cross-variable information to construct target-specific local representations and uses them to perform content-conditioned sparse routing at each spatial location, with the number of active experts adapted to router confidence. In the decoder, routing is augmented with a learned geographic bias parameterized by spherical-harmonic spatial bases, while shared residual and seasonal pathways provide common cross-variable and month-dependent context. Experiments on long-horizon autoregressive ORAS5 forecasting show that OceanMoE lowers aggregate forecasting error in both evaluated settings and maintains lower geometric-mean normalized RMSE than the corresponding baselines over most later rollout months. Routing analyses further show that expert allocation varies with prediction targets and spatial locations. These results support structured conditional computation as a modeling strategy for balancing shared ocean context with adaptive specialization.

Thu 17 SeptMachine Learning
The gist
Forecasting multiple ocean features like temperature and currents together is tricky because they change differently across locations and variables. The paper presents OceanMoE, a model that balances shared ocean information with specialized predictions by selecting the right small groups of experts for each place and variable. This approach helps make better long-term ocean predictions by adapting to local and variable differences while still learning from common ocean patterns. Tests showed OceanMoE reduced forecast errors compared to traditional methods and changed how experts were chosen based on what was being predicted and where.
Open → 2609.19768v1

Variational latent models improve long horizon pde predictions

Stable by Construction: Variational Latent Markov Operators for Long-Horizon PDE Prediction

Abstract: Neural PDE solvers provide efficient surrogates for time-dependent physical systems, but autoregressive prediction over long horizons remains challenging because local errors can induce distribution shift and accumulate under recursive deployment. We develop a variational approach to this problem by introducing latent Markov dynamics in which physical states are represented by latent distributions and evolved through probabilistic transitions. The framework is formulated directly on function spaces and specialized to functional Gaussian models, where structured latent perturbations induce a spectral geometry and variational transition alignment regularizes the learned dynamics. We further analyze how these mechanisms affect autoregressive error propagation, providing a theoretical connection between variational training and long-horizon prediction. We instantiate the framework as the Variational Autoencoding Markov Operator (VAMO), which combines spatially resolved latent fields, structured Gaussian perturbations, and a neural-operator transition. Empirically, we demonstrate the effectiveness of VAMO on several fluid-dynamics benchmarks with prediction horizons extending substantially beyond those represented during training, where it consistently reduces error accumulation and improves rollout stability over several deterministic and noise-injection baselines. Overall, these results highlight variational modeling as a complementary approach to robust long-horizon neural PDE dynamics.

Tue 15 SeptMachine Learning
The gist
Predicting how physical systems change over time using neural networks is hard because small errors add up. The authors developed a method that represents the system in a way that focuses on probabilities and smooth transitions, helping avoid growing mistakes. Their approach models the system’s hidden states with special math tools called variational latent distributions and Gaussian processes. This helps their predictions for things like fluid flow stay accurate even when looking far into the future.
Open → 2609.16621v1

Physics inputs speed up glacier flow simulations on single GPUs

Physics-enriched neural solvers for transient ice-flow simulation

Abstract: Transient glacier simulations with higher-order ice flow require the repeated solution of a nonlinear problem as the geometry evolves. In the online mode of the Instructed Glacier Model, the velocity field is represented by a neural network whose weights are warm-started from the previous time step and updated with a few optimizer iterations. We show that supplying the network with inexpensive input fields derived from low-order ice-flow balances improves this online solver. Unlike residual-based physics-informed neural networks, which incorporate physics through governing-equation penalties in the loss, our approach leaves the governing energy objective unchanged, adding physical structure through the network inputs. Across three real-world glacier configurations, the enriched solver is markedly more robust to solver settings. On the two alpine cases, it also improves the tuned accuracy--runtime trade-off, reducing surface-velocity errors by factors of two to four at fixed runtime and reaching few-percent relative errors with only $10^4$--$10^5$ trainable parameters, far fewer than comparable raw-input baselines. A 300-year Aletsch simulation then completes in under one minute, and the larger Valais domain in about two minutes, on a single GPU---a budget once reserved for much simpler shallow-ice models. Gains are smaller for the fast marine-terminating glacier, where nonlocal stress coupling favors larger or spectral networks. More broadly, the results suggest that enriching a neural solver's inputs with reduced-order physics can make repeated higher-order solves much cheaper, with no training data and no offline training.

Fri 11 SeptMachine Learning
The gist
Glacier flow simulations can be very slow because they require solving complex physics equations repeatedly as the glacier shape changes. The authors improved a neural network-based solver by giving it simple physics-based hints as inputs, which helps the network solve these equations faster and more accurately without needing extra training data. Their method runs fast enough to simulate hundreds of years of glacier movement in just minutes using a single GPU. This work shows a way to make advanced glacier models much more efficient for real-world use.
Open → 2609.12900v1

Bayesian neural network achieves faster precise weather forecasts with uncertainty

4D Parallelism Unlocks Exascale Bayesian Neural Networks for High-Fidelity Atmospheric Modeling

Abstract: We present BEAST, the first-ever Bayesian Swin Transformer for atmospheric forecasting on 0.25$^\circ$ global resolution able to accurately quantify both aleatoric and epistemic uncertainty. To overcome the associated computational bottlenecks, we devise an orthogonal 4D-parallelization scheme that introduces a unique domain-tensor-parallelism strategy and a novel uncertainty parallel method, enabling us to fully leverage GPU capacity and efficiently scale model training. For a 2.4-billion-parameter model, we achieve a peak performance of 3.96 EFLOP/s on 20,480 NVIDIA GH200 GPUs on the JUPITER supercomputer. We train BEAST as a 700-million-parameter model with 96 random weight samples on 384 nodes on 40 years of data for nearly one million gradient updates. This model achieves predictive skill scores competitive with state-of-the-art probabilistic atmospheric AI models and numerical models, and can predict extreme events with exceptional skill, while generating large ensembles 3 to 4 times faster than the current-best AI model. Our contribution unlocks the potential of high-fidelity uncertainty quantification in atmospheric AI models, heralding a new era for AI-based models in climate and Earth system sciences.

Fri 11 SeptArtificial IntelligenceDistributed, Parallel, and Cluster ComputingMachine Learning
The gist
Weather forecasting models need to predict not just what will happen, but also how sure they are about their predictions. The authors created BEAST, a new kind of AI model that uses a special neural network to make highly detailed weather forecasts across the globe and shows how confident it is in those forecasts. They also developed a way to run this huge model efficiently on many GPUs at once, making training faster and more powerful. Their results show the model can predict extreme weather well and generate many forecast possibilities more quickly than other AI models.
Open → 2609.12815v1