The Rise of AI in Weather and Climate Information and its Impact on Global Inequality

Amirpasha Mozaffari; Amanda Duarte; Lina Teckentrup; Stefano Materia; Gina E. C. Charnley; Lluis Palma; Eulalia Baulenas Serra; Dragana Bojovic

physics.ao-phMar 5, 2026

The Rise of AI in Weather and Climate Information and its Impact on Global Inequality

Authors: Amirpasha Mozaffari, Amanda Duarte, Lina Teckentrup, Stefano Materia, Gina E. C. Charnley, Lluis Palma, Eulalia Baulenas Serra, Dragana Bojovic,

Organizations: Earth Sciences Department, Barcelona Supercomputing Center (BSC), Plaça d’Eusebi Güell, 1-3, Barcelona, 08034, Spain. · School of Public Health, Imperial College London, White City Campus, London, W12 7TA, United Kingdom. · Independent Researcher. · Catalan Institution for Research and Advanced Studies (ICREA), Passeig Lluís Companys 23, Barcelona, 08010, Spain.

Abstract

AI development's current trajectory risks automating and amplifying the North-South divide in the global climate information system. Frontier models are built almost exclusively in the Global North, and this inequality continues through inputs, processes, and outputs, from biased training data to unrepresentative validation, disproportionately affecting vulnerable regions. Addressing these disparities requires a Climate Digital Public Infrastructure, evaluation metrics centring well-being, and knowledge co-production to foster resilience rather than inequity.

Figures & tables

Figure 1: The lifecycle of AI model development in weather and climate sciences, tracing the workflow from initial data acquisition to downstream application. The bottom panels qualitatively illustrate the varying intensity of four key dimensions across these stages: the level of human agency and oversight, the degree of transparency and trust, the magnitude of infrastructure requirements, and the existence of market demand.

Appendix figures & tables3 assets

Supplementary material from the paper’s appendix.

Appendix

Model	Year	Developer	Data used for training	Metrics	Training Compute
Aurora ( Bodnar et al., 2025 )	2025	Microsoft Research	Diverse mix of 6 datasets (ERA5, CMCC, IFS-HR, HRES, GFS, GFS Forecasts)	RMSE, MAE	2.5 Weeks on 32 A100
Ardavak Weather ( Allen et al., 2025 )	2025	Univ. of Cambridge, Turing Inst., Microsoft, ECMWF	Raw observations (sat radiances, station data) & ERA5 data.	RMSE, MAE, ACC	100 hours on A100
WeatherNext GenCast ( Price et al., 2025 )	2025	Google DeepMind + Google Research	ERA5 reanalysis data.	CRPS, Spread	5 days on 32 TPUv5
FengWu ( Chen et al., 2025 )	2025	Shanghai AI Lab	ERA5 reanalysis data.	RMSE, ACC	not reported
WeatherNext GraphCast ( Lam and others, 2023 )	2024	Google DeepMind + Google Research	39 years (1979–2017) of ERA5 reanalysis data.	RMSE & ACC	4 weeks on 32 TPU
NeuralGCM ( Kochkov et al., 2024 )	2024	Google, MIT, Harvard & ECMWF	ERA5 reanalysis data.	RMSE, RMSB, CRPS, spread-skill ratio	Not reported

Appendix

Table 1: List of the frontier weather and climate models, developers, training data, metrics, and compute cost (Table A1)

Benchmark name	Year	Datasets that are included
ClimateSet ( Kaltenborn et al., 2023 )	2023	Input4MIPs, CMIP6 (36 climate models)
Weatherbench 1 ( Rasp et al., 2020 )	2020	ERA5
Weatherbench 2 ( Rasp et al., 2024 )	2023	ERA5, IFS Operational Forecasts, GFS
HR-Extreme ( Ran et al., 2025 )	2024	High-Resolution Rapid Refresh (HRRR), NOAA Storm Events
ExEBench ( Zhao et al., 2025 )	2025	CMIP6, ERA5
ClimDetect ( Yu et al., 2025 )	2023	CMIP6, ERA5, JRA-3Q, MERRA-2

Appendix

Table 2: List of the most common weather and climate benchmark datasets (Table A2)

Continent	Institution	Country / Location	System Name(s)
Europe	BSC-CNS	Spain (Barcelona)	“MareNostrum 5”
Europe	CINECA	Italy (Bologna)	“Leonardo”
Europe	DKRZ	Germany (Hamburg)	“Levante”
Europe	DWD	Germany (Offenbach)	NEC SX-Aurora TSUBASA
Europe	ECMWF	Italy (Bologna)	Atos BullSequana XH2000
Europe	CSC	Finland (Kajaani)	“LUMI”

Appendix

Table 3: List of selected HPC Centres and Systems for Weather and Climate Modeling (Table A3) adopted and expanded from ( Zaidan et al., 2025 )

Algorithmic Bias

Algorithmic bias studies how automated systems produce or amplify unfair outcomes across demographic groups, and how to measure and reduce them.

See all 100 papers

Artificial intelligence pathways from weather to climate
Oct 7, 2026Tom Beucler, J. David Neelin, Hui Su +10Climate Emulation Climate Modeling
University of Lausanne, Lausanne, Switzerland. · University of California, Los Angeles, Los Angeles, CA, USA. · Hong Kong University of Science and Technology, Hong Kong SAR, China. +6
Geographic Bias and Diversity in AI Evaluation
Apr 28, 2026Zilong Liu, Krzysztof Janowicz, Gengchen Mai +2Algorithmic Bias Language Model Bias Evaluation
University of Vienna, Austria · University of Texas at Austin, USA · University of Wisconsin-Madison, USA +1
Artificial intelligence is creating a new global linguistic hierarchy
Feb 12, 2026Giulia Occhini, Kumiko Tanaka-Ishii, Anna Barford +10Low-Resource Language Processing
University of Cambridge, UK. · Waseda University, Japan. · University College London, UK. +1

Oct 7, 2026J/K move · Enter open · S save

Artificial intelligence pathways from weather to climate

Tom Beucler, J. David Neelin, Hui Su, Shivanshi Asthana+9

University of Lausanne, Lausanne, Switzerland. · University of California, Los Angeles, Los Angeles, CA, USA. · Hong Kong University of Science and Technology, Hong Kong SAR, China.+6

Climate Emulation Climate Modeling Weather Forecasting

Open paper AskPDF arXiv

Deep learning has made rapid advances in weather forecasting: autoregressive models trained on atmospheric reanalyses now rival dynamical models across nowcasting, medium-range, and subseasonal-to-seasonal lead times, producing well-calibrated ensemble forecasts at reduced cost. We review these advances and consider their extension to climate horizons, where the challenge shifts from initial-condition skill to producing reliable statistical responses under altered forcings. AI-powered climate prediction systems must produce credible forced responses to drivers (e.g., greenhouse gases, land-use change) typically outside the observed record. We propose two minimum requirements for AI in climate modeling: (i) external forcing agents must enter explicitly enough to support interventions in which they vary independently; and (ii) robustness must be stress-tested in out-of-distribution regimes, including extremes and counterfactual trajectories. Using leading AI autoregressive emulators and hybrid physics-AI models, we identify development and coupling challenges. Comparing the reported throughput of these models with that of GPU-ported dynamical models highlights how AI can reduce time-to-solution by advancing only the target variables at the required resolution and using longer time steps, rather than integrating a full high-frequency, multivariate state. Diverse AI downscaling strategies can partially substitute for explicit fine-scale resolution, paving the way toward inexpensive local hazard assessment across prediction horizons.

Abstract

Algorithmic Bias

Explore similar work

Artificial intelligence pathways from weather to climate

Geographic Bias and Diversity in AI Evaluation

Artificial intelligence is creating a new global linguistic hierarchy