AI for Science

Accurate prediction of dust source emissions is critical for mitigating the significant environmental and health hazards posed by dust storms. Traditional forecasting methods often struggle to capture the complex spatiotemporal dynamics of these phenomena. In this paper, we demonstrate that proximity graphs enable Graph Neural Networks (GNNs) to effectively model the intricate spatial and temporal relationships between data points. Specifically, we use proximity graphs--such as Delaunay triangulation, Gabriel graph, k-Nearest Neighbor graph, and Yao graph--as the input for GNNs (including GraphSAGE, Graph Convolutional Networks, and Graph Attention Networks) to perform message passing. Our approach highlights the effectiveness of integrating proximity graphs with GNNs for robust and accurate dust source forecasting. To emphasize the importance of proximity graph representations, we compare our method against GNNs using random graphs for message passing. The results show that GNNs with proximity graphs significantly outperform those with random graphs and are also far superior to Long Short-Term Memory (LSTM) model in dust source emission forecasting.

Probabilistic weather forecasting is undergoing rapid transformation with artificial intelligence (AI). In traditional numerical weather prediction, computing power can limit how well ensemble forecasts approximate the unknown statistical distribution of future states. AI models facilitate larger ensembles and are trained with probabilistic considerations, ideally leading to better uncertainty quantification. Forecasts from these state-of-the-art models are often considered well-calibrated. However, here we show that the statistical coverage of such models, the ultimate measure of calibration, can struggle, especially on extreme events. To address this shortcoming, we employ conformal prediction, a class of statistical methods that mathematically guarantees coverage under no distributional assumptions, unlike previous post-processing techniques. We apply online conformal prediction to temperature and precipitation forecasts (including extremes) of three leading global weather models, GenCast, NeuralGCM, and AIFS-ENS, ensuring calibrated uncertainty at no expense to other probabilistic metrics. This post-processing method can be applied to any forecasting model.

Comprehensive information on cloud microphysical process rates from numerical simulations allows for better understanding of precipitation formation pathways and aerosol-cloud interactions. However, resource limitations often make it impractical to include all microphysical process rates in the model output, limiting in-depth analyses. To address this shortcoming, we introduce PRecover, a data-driven post-processing approach to recover microphysical process rates that are not stored during runtime from standard output of a numerical weather prediction model. In particular, we train random forests, gradient boosting models, and feed-forward neural networks to recover microphysical process rates from a two-moment bulk microphysics scheme in the ICOsahedral Nonhydrostatic (ICON) model. We use cloud variables as input, obtained from high-resolution simulations in a limited-area setup over Europe. Warm-rain and ice microphysical process rates are recovered with a two-step classification-regression approach for both instantaneous and accumulated process rates. As a physics-based baseline, we assess whether process rates can be directly recalculated from stored ICON output variables. Accurate recalculation is possible for process rates such as accretion and self-collection but not for the autoconversion, rain melting or heterogeneous ice nucleation rate. Using PRecover, we successfully recover most of the process rates that are accumulated over output time steps of 10 minutes or less, but the values are increasingly difficult to recover for rates accumulated over longer accumulation intervals. To quantify predictive uncertainty, we provide calibrated prediction intervals through conformalized quantile regression. We demonstrate spatial transferability of the models with two case studies over different regional domains and simulation settings unseen during training.

We investigate the seasonal phase locking of the El Niño-Southern Oscillation (ENSO) in a linear stochastic recharge oscillator (SRO), a damped oscillator with additive noise and a time-dependent growth rate. Phase locking is reflected in the seasonality of the variance of the sea surface temperature anomaly (SSTA). In general, energy drives such a change, whereas entropy governs whether it occurs; phase locking is thus subject to both an energy- and an entropy-based constraint. We quantify this entropy-based constraint using a thermodynamic uncertainty relation (TUR), a fundamental inequality in stochastic thermodynamics. The TUR constrains the tendency of the SSTA variance by the partial entropy production rate, which is dominated by the ratio of forward and backward transition probabilities and quantifies the irreversibility of SSTA transitions. The growth rate governs this irreversibility: its extrema occur in boreal autumn and late winter, and the entropy production rate peaks at both times. These peaks relax the TUR constraint on the tendency of the SSTA variance, so that the variance itself can peak in boreal winter, consistent with observed ENSO phase locking. Conversely, when irreversibility is insufficient, ENSO cannot grow or decay. If this irreversibility were interpreted as dissipated energy, the constraint on ENSO growth and decay would require this dissipation to be exported from the equatorial Pacific. A more realistic model is needed to test this hypothesis and to further explore the physical connection between entropy and dissipated energy.

While geostrophy remains the simplest and most practical balance to extract velocity information from sea-surface height anomaly (SSHa), confusions remain within the oceanographic community to what extent this balance can be applied to altimetric observations with the launch of the Surface Water and Ocean Topography (SWOT) satellite. Given the limited temporal resolution of SWOT, many studies have resorted to claiming that the spatially filtered SSHa fields correspond to the geostrophic component. This introduces the ambiguity of which spatial scale to choose. Here, we build upon the recent developments in internal tide (IT) corrections (Yadidya et al., 2025) and apply a dynamic mode decomposition (DMD)-based method introduced by Lapo et al. (2025) to robustly extract the geostrophic component associated with sub-inertial frequencies from the SWOT one-day-repeat orbit; we distribute the global dataset as a public good. We provide the joint probability density function (PDF) of vorticity and strain, and spectra of SSHa at a few cross-over regions.

The deployment of Time-Series Foundation Models (TSFMs) in physical sciences is hindered by a critical trade-off: while these models encode rich, universal temporal dynamics, they suffer from severe distributional misalignment when applied zero-shot to specific scientific domains, and their computational cost prohibits deployment in edge-computing sensor networks. We address a fundamental challenge: How can we extract latent structural knowledge from misaligned foundation models (FM) to train lightweight, specialized forecasters? We propose Gated Uncertainty-Aware Routing for Distillation (Guard), a novel framework that reframes multiteacher distillation as an instance-wise decision process with two adaptive mechanisms: (1) a Contextual Router that dynamically selects the most relevant teacher based on local input statistics, exploiting complementarity across diverse foundation models; and (2) an Uncertainty-Gated Temperature mechanism that acts as a "circuit-breaker," automatically attenuating distillation strength when teacher confidence diverges from domain reality. We evaluate our proposed lightweight framework on four climate-critical domains: meteorology, ecosystem carbon flux, soil moisture, and energy grids. Our method significantly reduces RMSE relative to a fixed-weight multi-teacher distillation baseline, successfully distilling knowledge from pretrained FMs (teachers) even when they exhibit suboptimal zero-shot accuracy due to distribution shift between the original and target data domains. We demonstrate that these domain-misaligned teachers can still serve as critical correctives, outperforming the globally superior FMs on 28.5% of the hardest instances. Ultimately, this enables high-precision scientific forecasting suitable for resource-constrained edge deployment. Code is available at https://github.com/RupasreeDey/GUARD-KDD2026.

Increased upper-ocean stratification is an unavoidable consequence of global warming and will strongly impact the structure of ocean currents. Using a high-resolution ocean model, we show that intensification of stratification leads to the loss of coherence of the Gulf Stream Extension, replacing its steady eastward path with vigorous, chaotic meanders. This regime shift persists independently of changes in the Atlantic Meridional Overturning Circulation and surface wind forcing. Enhanced meandering under intensified stratification also proves to be a robust feature across both idealized and realistic ocean models that resolve mesoscale eddies, but is not captured by coarse-resolution models that parameterize eddies. The presented findings therefore highlight the need for improved representations of oceanic turbulence in climate projections.

Spatial patterns in observed climate trends remain poorly understood. Here we argue that a warming of land relative to ocean has shaped observed surface temperature and atmospheric circulation trends, including the negative Inter-Decadal Pacific Oscillation (IPO)-like tendency across the Pacific basin. Observed and modeled trends display an overall decline in sea level pressure over the faster-warming land relative to ocean, with a spatial pattern that resembles the seasonal cycle and the response to land heating in idealized climate model experiments. Coupled climate model simulations with historical forcing underestimate the land-sea warming ratio. It is only in the early response of abrupt CO2 quadrupling climate model simulations that climate models are able to recreate the observed land-sea warming ratio, in which case a strengthening of oceanic surface highs and a negative IPO-like surface warming pattern over the Pacific comparable to observed trends are seen. We propose that discrepancies between modeled and observed trends in many climate variables may be explained by the underestimation of the land-sea warming ratio by climate models. Determining the cause of this discrepancy has the potential to constrain projections of future climate change as the underlying mechanism causing climate models to underestimate the land-sea warming ratio discrepancy will set the persistence of this problem.