arXiv:2606. 26361v1 Announce Type: new Abstract: ML foundation models are able to emulate atmospheric dynamics accurately and efficiently but operate as opaque ``black boxes''.
arXiv:2605.24038v3 Announce Type: replace-cross Abstract: Aurora visibility at a given location requires two physically distinct conditions to hold at once: aurora occurring overhead, governed by sol...
arXiv:2607. 05658v1 Announce Type: cross Abstract: Global Navigation Satellite Systems (GNSS), best known for positioning, also serve weather science, as atmospheric water vapour delays their signals.
Aurora 1. 5 adds 22 more variables, hourly temporal resolution, and probabilistic ensemble forecasting to the Aurora foundation model, making it more useful for real-world weather, climate, and energy applications.
A deep‑learning model that converts dynamical circulation forecasts into precipitation estimates predicts a dry anomaly over central China in summer 2026, with consistent signals from March to May. Retrospective tests show the model performs best in analogue years marked by sustained central equatorial Pacific warming, which promotes a cyclonic circulation that drives northerly winds and moisture divergence, suppressing rainfall. Layer‑wise relevance propagation identifies these northerly winds as the key driver, and perturbation tests confirm that removing them eliminates the dry anomaly, demonstrating a physically interpretable link between AI predictions and climate dynamics.
A deep learning model that converts dynamical circulation forecasts into precipitation estimates predicts a dry anomaly over central China in the summer of 2026, with consistent signals from March to May. Retrospective tests show the model performs best in analogue years marked by sustained central equatorial Pacific warming, which promotes a cyclonic circulation that drives northerly winds and moisture divergence, suppressing rainfall. Layer‑wise relevance propagation identifies these northerly winds as the key driver, and perturbation tests confirm that removing them eliminates the predicted dry anomaly, providing a physically interpretable explanation for the AI forecast.
arXiv:2511. 23043v2 Announce Type: replace-cross Abstract: We present a probabilistic data-driven weather model providing ensembles of high spatial resolution realizations of 87 variables at arbitrary ensemble size and forecast length.
arXiv:2606. 08563v1 Announce Type: new Abstract: While global data-driven models excel at predicting continuous atmospheric variables, three-dimensional hydrometeor forecasting remains challenging due to the zero-inflated, long-tailed distributions of these variables.
arXiv:2608. 12271v1 Announce Type: new Abstract: Global weather reanalyses and forecasts resolve the evolving atmospheric state on coarse grids, but site-specific applications require predictions at arbitrary locations where near-surface conditions also depend on unresolved terrain and land-surface properties.
The study demonstrates that near‑surface weather variability over tens to hundreds of meters can be inferred without resolving atmospheric dynamics by combining sparse weather stations, high‑resolution Earth observation, and coarse atmospheric dynamics. Using this approach, the authors estimate temperature, dewpoint, and wind at 30‑meter resolution across the contiguous United States, achieving 11–28 % error reduction compared to the strongest baseline and recovering nearly half of the temperature variability in median grid cells. The method captures time‑varying differences between locations and produces coherent patterns linked to topography and land cover.
The paper introduces CloudCast v2, a machine‑learning model that forecasts 12‑hour cloud‑cover from satellite‑derived initial conditions. Trained first on the Copernicus European Regional Reanalysis to learn cloud‑evolution dynamics, it is then adapted to real satellite data using conditional flow matching, a generative technique that conditions noise on observed cloud fields and NWP inputs. CloudCast v2 achieves a 10 % reduction in mean absolute error compared to its predecessor and surpasses it in spatial skill after 3–6 hours, extending useful forecasting beyond the typical 1–3‑hour nowcasting window while preserving satellite‑level spatial detail.
The paper introduces an explainable contrastive learning model that maps five monthly cloud and radiation fields into a shared representation space to analyze perturbed parameter ensembles (PPEs) in climate simulations. Trained on two 100-member Community Atmosphere Model version 6 PPEs differing only in the warm rain microphysics scheme (KK2000 vs. TAU-ML), the model achieves over 94% linear classification accuracy while preserving seasonal variability and ensemble spread. The learned representations place satellite observations on the same low‑dimensional manifold as the PPEs, with TAU-ML PPE closer to observations, and Integrated Gradients attribution identifies key regional contributions linked to cloud microphysics, boundary layer turbulence, and deep convection.
arXiv:2607. 20778v1 Announce Type: new Abstract: Weather forecasting foundation models (FMs) are increasingly fine-tuned to predict air quality, offering fast global pollution forecasts at lower computational cost than conventional chemical transport models.