The study evaluates whether Earth‑observation foundation models encode geographic location information by attempting to predict coordinates from their embeddings. Using 284 verified European solar farms, the authors tested Tessera v1, Tessera v1.1, and AlphaEarth, finding that all three models contain recoverable geographic data. AlphaEarth showed the strongest correlation between embedding distance and geographic distance, while both Tessera variants outperformed Sentinel‑2 controls, suggesting that geographic information should be considered when auditing such models.
By Peiwen Zhang, Kristie Hu, Jovana Knezevic, Shunde Yin, Kyle Gao
arXiv:2608.27521v1 Announce Type: new
Abstract: Many dynamical processes unfold on the sphere but the default scientific machine learning architectures are Euclidean. Applying these architectures on...
By Till Muser, Giovanni Abati, Ivan Dokmani\'c
arXiv:2602. 13416v2 Announce Type: replace Abstract: The proliferation of data-driven models in weather and climate sciences has marked a significant paradigm shift, with advanced models demonstrating exceptional skill in medium-range forecasting.
By Haiwen Guan, Dibyajyoti Chakraborty, Moein Darman, Troy Arcomano, Ashesh Chattopadhyay, Romit Maulik
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.
By Pedro Sousa (Department of Computer Science, University of Cambridge), Will Tebbutt (Department of Engineering, University of Cambridge), Sadiq Jaffer (Department of Computer Science, University of Cambridge), Robin Young (Department of Computer Science, University of Cambridge), Anil Madhavapeddy (Department of Computer Science, University of Cambridge), Richard E. Turner (Department of Engineering, University of Cambridge)
arXiv:2607. 29527v1 Announce Type: cross Abstract: A defining problem of the Anthropocene is to model the physical Earth and human societies as one coupled system, yet no learned representation spans their observational breadth.
By Carlos Rodriguez-Pardo, Massimo Tavoni
arXiv:2607. 17037v1 Announce Type: new Abstract: High-resolution atmospheric data are required to resolve mesoscale and localized meteorological structures, however such datasets remain limited in many regions of the world.
By Evangelia Rafaela Frastali, Achyut Paudel, Maryam Golbazi, Frank Liu
The study presents a spatially aware deep learning framework that retrieves all‑sky tropospheric temperature and humidity profiles from the Meteosat Third Generation Flexible Combined Imager (FCI) without relying on numerical weather prediction background fields. Using a Residual U‑Net trained on 14 months of collocated FCI observations and CERRA reanalysis data, the model achieves temperature biases below 0.4 K and relative humidity standard deviations between 12–20 %, with modest performance degradation under cloud cover. Ablation and feature‑sensitivity analyses confirm that incorporating spatial context across all 16 FCI channels, including visible and near‑infrared bands, improves retrieval accuracy, especially beneath cloud tops.
By Alejandro Salgueiro, Johannes Rausch, Julie Th\'er\`ese Villinger, Angela Meyer
The Meteosat Third Generation (MTG) Flexible Combined Imager (FCI) offers new opportunities for tropospheric temperature and humidity profiling, at higher spatio-temporal resolutions and expanded spec...
arXiv:2602. 00392v2 Announce Type: replace Abstract: Geographic data is fundamentally local.
By Arjun Rao, Ruth Crasto, Tessa Ooms, David Rolnick, Konstantin Klemmer, Marc Ru{\ss}wurm
arXiv:2608. 04230v1 Announce Type: new Abstract: Deep learning models for scientific spatio-temporal downscaling often minimize reconstruction error while failing to preserve physically meaningful multi-scale structure.
By Parth Doshi, Priyanka Aravindan, Vaishnav Vaidheeswaran, Md Mahbub Alam, Gabriel Spadon
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.
By Mikko Partio, Leila Hieta, Ossi Laine
arXiv:2606. 31248v1 Announce Type: cross Abstract: Kilometer-scale convection shapes precipitation extremes, tropical organization, and cloud feedbacks, but most global atmospheric models approximate these processes at 25-100 km resolution.
By Zeyuan Hu, Akshay Subramaniam, Noel Keen, Tao Ge, Jaideep Pathak, Mohammad Shoaib Abbas, Suman Ravuri, Karthik Kashinath, Naser Mahfouz, Peter Caldwell, Mike Pritchard, Noah Brenowitz