Geo Intelligence.

RESEARCH & SOURCES

On the foundations of Earth foundation models

Communications Earth & Environment · CAS Tier 1 · Top (2025)

Paper authors：Xiao Xiang Zhu · Zhitong Xiong · Yi Wang · Adam J. Stewart · Konrad Heidler · Yuanyuan Wang · Zhenghang Yuan · Thomas Dujardin · Qingsong Xu · Yilei Shi

DOI: 10.1038/s43247-025-03127-x

Earth systemsRemote sensing & space

AI reading guide

AI summaryProposes 11 features of an ideal Earth foundation model, including multisensor integration, awareness of location and scale, and physical consistency, and discusses evaluation and research directions.

This reading guide is not the authors’ abstract. Consult the original source.

This record provides source metadata. A detailed editorial analysis has not yet been published.

Read the original source ↗

CAS 2025 major-category tier 1 and Top. This journal-level label does not assess this individual paper.

Related research

Research articleNatureCAS Tier 1 · Top (2025)

A foundation model for the Earth systemOriginal title · English

Authors Cristian Bodnar · Wessel P. Bruinsma · Ana Lucic · Megan Stanley · Anna Allen · Johannes Brandstetter · Patrick Garvan · Maik Riechert · Jonathan A. Weyn · Haiyu Dong · Jayesh K. Gupta · Kit Thambiratnam · Alexander T. Archibald · Chun-Chieh Wu · Elizabeth Heider · Max Welling · Richard E. Turner · Paris Perdikaris

AI summaryPretrains on diverse geophysical data and adapts to forecasting weather, ocean waves, air quality and tropical cyclone tracks, exploring a foundation model for multiple Earth-system tasks.

AuroraFoundation modelsMultitask forecasting
Details & notesRead original ↗
PerspectiveNature Reviews Earth & EnvironmentCAS Tier 1 · Top (2025)

Differentiable modelling to unify machine learning and physical models for geosciencesOriginal title · English

Authors Chaopeng Shen · Alison P. Appling · Pierre Gentine · Toshiyuki Bandai · Hoshin Gupta · Alexandre Tartakovsky · Marco Baity-Jesi · Fabrizio Fenicia · Daniel Kifer · Li Li · Xiaofeng Liu · Wei Ren · Yi Zheng · Ciaran J. Harman · Martyn Clark · Matthew Farthing · Dapeng Feng · Praveen Kumar · Doaa Aboelyazeed · Farshid Rahmani · Yalan Song · Hylke E. Beck · Tadd Bindas · Dipankar Dwivedi · Kuai Fang · Marvin Höge · Chris Rackauckas · Binayak Mohanty · Tirthankar Roy · Chonggang Xu · Kathryn Lawson

AI summaryReviews how differentiable computation can jointly train physical components and neural networks to support process understanding, parameter estimation and geoscientific prediction.

Differentiable modellingPhysical constraintsHydrological modelling
Details & notesRead original ↗
Research articleNatureCAS Tier 1 · Top (2025)

Neural general circulation models for weather and climateOriginal title · English

Authors Dmitrii Kochkov · Janni Yuval · Ian Langmore · Peter Norgaard · Jamie Smith · Griffin Mooers · Milan Klöwer · James Lottes · Stephan Rasp · Peter Düben · Sam Hatfield · Peter Battaglia · Alvaro Sanchez-Gonzalez · Matthew Willson · Michael P. Brenner · Stephan Hoyer

AI summaryCombines a differentiable dynamical core with neural-network parameterisations to study weather forecasting and climate simulation, including physical consistency over long integrations.

NeuralGCMDifferentiable modelsPhysics and AI
Details & notesRead original ↗
PerspectiveNature Machine IntelligenceCAS Tier 1 · Top (2025)

Enhancing reproducibility in hybrid Earth system modelsOriginal title · English

Authors Min Chen · Zhiyi Zhu · Thorsten Wagener · Niklas Boers · R. Dietmar Müller · Josef Strobl · Gustau Camps-Valls · Michael Batty · Anthony J. Jakeman · Olaf Kolditz · Stefano Nativi · Maria Antonia Brovelli · Felix Creutzig · Pankaj Kumar · Paul Whitehead · C. Michael Barton · Dichen Liu · Peilong Ma · Zaiyang Ma · Fengyuan Zhang · Bo Zhang · Peng Hou · Guonian Lü

AI summaryDiscusses numerical stability, workflow transparency and differences in computing resources, and proposes a framework for assessing and supporting reproducibility in hybrid Earth system models.

ReproducibilityHybrid modellingRHEM
Details & notesRead original ↗