Pith. sign in

REVIEW 3 major objections 5 minor 97 references

This paper argues that the current trajectory of AI weather and climate modelling — built in the Global North on observation-sparse reanalysis data — will automate and amplify, rather than close, the global divide in climate information.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-02 18:40 UTC pith:WWGJW7UI

load-bearing objection A useful synthesis of known AI-climate inequality critiques, with a load-bearing output-level claim that rests on a single preprint. the 3 major comments →

arxiv 2603.05710 v2 pith:WWGJW7UI submitted 2026-03-05 physics.ao-ph cs.AIcs.LG

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
show 3 more authors
Paula Checchia Aude Carreric Francisco Doblas-Reyes
This is my paper
classification physics.ao-ph cs.AIcs.LG
keywords AI weather forecastingclimate inequalityGlobal Southreanalysis biasERA5forecast skill gapfairness metricsknowledge co-production
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper argues that the current wave of AI weather and climate modelling is not democratising access to climate information; it is automating and amplifying the North–South divide. Frontier models are almost all developed in the Global North, trained on reanalysis and climate-model data that are least constrained by observations in the tropics, and evaluated on benchmarks that inherit those gaps. The authors trace this through the full lifecycle — inputs, processes, and outputs — from biased training data and concentrated computing infrastructure to forecast products that are least reliable for the most vulnerable regions. They conclude that without deliberate restructuring — public data infrastructure, fairness-aware metrics, and knowledge co-production — AI will entrench environmental injustice rather than relieve it. A sympathetic reader would take this as a warning grounded in existing evidence, not a new empirical demonstration.

Core claim

On the paper's own terms, the central claim is that inequality in climate information is not incidental to AI development but structural: it enters at the data-acquisition stage (reanalysis products like ERA5 are effectively 'maps without gaps' that hide weak tropical observations), is reinforced during training and evaluation (gradient descent on global mean-squared error concentrates learning in data-dense Northern latitudes, and benchmarks reuse the same biased datasets), and emerges in outputs as regional skill gaps, exemplified by an analysis showing that ECMWF's AIFS improves forecasts mainly in wealthier, more densely populated regions. The paper's proposed response is a Climate Digit

What carries the argument

The paper's organising device is the input-process-output lifecycle of AI systems, used as a lens to show how biases accumulate. Its load-bearing data objects are the reanalysis and climate-model archives — ERA5 and CMIP6 — which provide seamless global grids but embed an illusion of uniform quality: in observation-sparse regions, values are model-generated rather than measured. The process-stage mechanism is gradient descent on R/MSE, which the paper argues prioritises data-dense regions and smooths extremes; the output-stage mechanism is benchmark evaluation built on the same reanalysis, so gaps are validated rather than corrected.

Load-bearing premise

The paper's argument leans on the assumption that the forecast-accuracy gap seen in traditional models — and so far in one AI model (ECMWF's AIFS) — holds for frontier AI weather models generally; if global AI models trained on reanalysis do not reproduce the regional gap, the inequality claim weakens.

What would settle it

Compute regional forecast skill (CRPS or RMSE) for three or more frontier AI weather models, stratified by country income and verified against in-situ observations in the tropics rather than reanalysis. If no systematic degradation appears in low-income regions, the paper's central claim is not supported.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • If the paper is right, frontier AI weather models will continue to show the smallest skill gains in the tropics and in low-income countries, where forecast skill already lags; the democratisation narrative is false in intrinsic terms.
  • Fairness-aware, population-weighted, and vulnerability-weighted evaluation metrics would be needed to stop treating open ocean and dense cities as equally important; current benchmarks cannot detect the divide they inherit.
  • A Climate Digital Public Infrastructure — open data, open compute, and open benchmarks — becomes a precondition for equitable AI, not a nice-to-have.
  • AI systems trained on sparse and skewed data risk maladaptation: climate impact models and early-warning dashboards can be misleading when they cannot see the populations they claim to protect.
  • LLM-based climate services will amplify existing geographic and linguistic biases in climate knowledge unless training corpora and evaluation are diversified.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Editorial extension: if fairness-aware loss functions become standard, one testable consequence is a modest skill trade-off in data-dense mid-latitudes; the paper does not quantify this cost.
  • Editorial extension: the paper's logic implies upstream observations matter more than model architecture; expanding tropical surface and upper-air observation networks should improve AI forecast skill in low-income regions more than additional compute.
  • Editorial extension: the AIFS fairness result is a single model; repeating the regional-skill analysis on other frontier models would harden or weaken the central claim.
  • Editorial extension: the paper assumes reanalysis-based benchmarks reflect real-world skill in the tropics; evaluating AI models against in-situ station data could show the gap to be larger or smaller than reported.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. This perspective paper argues that the current trajectory of AI in weather and climate information systems risks automating and amplifying the North–South divide. It frames the issue as a cascade across three stages: inputs (reanalysis and climate model data with observation gaps in the tropics; biased health, biosphere, and text corpora), processes (concentrated HPC infrastructure, benchmark datasets inheriting ERA5/CMIP6 limitations, R/MSE optimization that allegedly prioritizes data-dense regions), and outputs (persistent income-based skill gaps, uneven benefits from AIFS, limited evidence of real-world value of early warning systems). The paper proposes remedies: a Climate Digital Public Infrastructure, a shift to data-centric AI, fairness-aware and human-centric evaluation metrics, and knowledge co-production.

Significance. If the central claim is correct, the paper raises an urgent governance issue for AI4Earth: frontier AI weather models may not be democratizing forecasting but rather entrenching existing asymmetries. The paper's strengths are its breadth and concrete documentation: Appendix Tables A1–A3 list frontier models, their ERA5/CMIP6 training data, and the concentration of HPC centers in the Global North and East Asia. It also connects diverse domains (weather, carbon cycle, health, LLMs) with a consistent framework. However, the paper is a perspective, not an empirical study, and its output-level claim rests on two cited preprints [52, 53] whose methodology is not audited and whose results are not sufficient to establish the headline conclusion. The argument is internally consistent, and the proposed remedies are plausible, but the empirical bridge from input/process asymmetries to output-level inequality needs substantial reinforcement.

major comments (3)
  1. [§2.3, paragraph 2] The empirical bridge for the output-level claim is Linsenmeier & Shrader [52] plus a single fairness analysis of ECMWF's AIFS [53]. [52] measures forecasts as received by users, including national downscaling, local observation networks, and dissemination capacity—not the skill of a globally centralized AI model. The reference's own annotation ('most countries need to develop their own forecasts...') in the bibliography confirms this. [53] reports relative improvements against IFS HRES, not absolute skill gaps between income groups. The sentence 'this trend continues with AI-powered models' is therefore stronger than the cited evidence. The authors should either add an income-stratified absolute-skill comparison on a second model, or explicitly reframe this as a hypothesis with a concrete test.
  2. [§2.2, paragraph 4] The proposed mechanism—'steepest descent in the global loss function is determined by data-dense regions'—is imprecise. ERA5 is a globally gridded reanalysis; every grid cell contributes equally to the loss unless explicitly weighted. Observation density does not add training samples; it affects the reliability of the target. The correct mechanistic argument is target noise heteroscedasticity: noisy targets in observation-sparse regions make the gradient less informative and may bias the solution toward well-constrained regions. This distinction matters because the paper's process-level causal story underpins the entire cascade. The authors should rewrite this passage to state the target-noise argument explicitly and, if possible, support it with a reference or small illustrative calculation.
  3. [§2.3 / §3] The paper does not quantify the relative contribution of training-data concentration versus deployment infrastructure to the claimed output-level gap. The recommendation for a Climate Digital Public Infrastructure covers both, but the diagnosis would be more actionable if the authors separated the two. At present, the paper risks attributing all observed disparities to upstream data and training choices, whereas deployment barriers (connectivity, API access, national capacity) are at least as important. A clear statement of what is known and what is unknown about this decomposition would strengthen the policy recommendations.
minor comments (5)
  1. [Reference [52]] The reference entry contains a first-person annotation ('You might wonder how this can be true...') that appears to be an authorial note accidentally left in the bibliography. It should be removed and replaced with a proper citation to the preprint.
  2. [Reference [53]] The DOI string is duplicated at the end of the reference: '174802937.77365288/v2 174802937.77365288/v2'. Also, both [52] and [53] are preprints or working papers; the text should indicate their preprint status.
  3. [Table A1, first data row] The model name 'Ardavak Weather' is likely a typo for 'Aardvark Weather'. Please check the official spelling.
  4. [§2.2, paragraph 4] The notation 'Root/ Mean Squared Error (R/MSE)' is awkward; consider writing 'Root Mean Squared Error (RMSE)' or explaining that the argument covers both RMSE and MSE.
  5. [Figure 1] The bottom panels 'qualitatively illustrate' the intensity of agency, transparency, infrastructure, and demand. A brief definition of the scales would help readers interpret the figure, since it is central to the narrative.

Circularity Check

0 steps flagged

No significant circularity: the paper is a literature-based synthesis whose claims are grounded in external citations, not in internal fits or self-citations.

full rationale

The paper contains no equations, fitted parameters, or train/predict loop of its own, so the main circularity patterns (self-definitional predicates, fitted inputs renamed as predictions, ansatz smuggled in via self-citation) are absent. Its central chain is: (i) frontier AI weather models are built in the Global North using ERA5/CMIP6 and HPC concentrated in the Global North (documented in Tables A1/A3 and refs [18],[42]); (ii) ERA5/CMIP6 have known regional biases, especially in the tropics (refs [19]-[26]); (iii) benchmarks inherit those data (Table A2); (iv) historical forecast skill gaps by income (ref [52]) persist for AI models, citing one fairness analysis of AIFS (ref [53]); (v) therefore AI risks amplifying the North-South divide. Each link is supported by independent external literature rather than by the paper's own prior results. I checked the reference list against the author list and found no load-bearing self-citation: no reference is authored by Mozaffari, Duarte, Teckentrup, Materia, Charnley, Palma, Baulenas Serra, Bojovic, Checchia, Carreric, or Doblas-Reyes. The paper also states limitations explicitly, e.g., 'there is almost no rigorous evidence that they improve population outcomes' (Sec. 2.3, refs [56][57]), which is an evidence-quality caveat, not a circular move. The strongest critique available is that extrapolating the AIFS fairness preprint [53] to all AI models, and inferring gradient-descent behaviour from data density, are assumptions whose empirical strength is debatable; that is a correctness/evidence concern, not a definitional reduction. The recommended Climate Digital Public Infrastructure is a normative policy prescription consistent with the diagnosis, but the diagnosis is not defined in terms of the prescription. Under the hard rule that circularity must be exhibited by quote and specific reduction, no such reduction exists here.

Axiom & Free-Parameter Ledger

0 free parameters · 5 axioms · 1 invented entities

The paper adds no fitted parameters or invented physical entities. It rests on qualitative domain assumptions about how AI pipelines and reanalysis biases behave, plus a reliance on preprints for key empirical support.

axioms (5)
  • domain assumption The AI model lifecycle can be decomposed into input, process, and output stages across which biases cascade.
    Section 2 frames the entire analysis; it is an analytic schema, not an empirically tested mechanism.
  • domain assumption Global North/South is the operative axis of climate information inequality.
    Throughout; the paper does not consider within-country inequality or other axes as primary.
  • domain assumption Observation sparsity in reanalysis causes systematic AI forecast degradation in those regions.
    Section 2.1 gives regional examples but no global systematic test.
  • domain assumption R/MSE minimization under gradient descent prioritizes data-dense regions and produces blurry forecasts that harm extremes.
    Section 2.2 states this as an optimization fact; the severity and equity consequences are not derived.
  • domain assumption The cited literature, including preprints [52] and [53], accurately supports the equity claims.
    The paper relies on preprints for the AIFS fairness result and the forecast inequality result.
invented entities (1)
  • Climate Digital Public Infrastructure no independent evidence
    purpose: Proposed open-source, publicly governed digital utility to keep climate information a global public good (Section 3).
    A proposed policy construct with no implemented instance or falsifiable test supplied.

pith-pipeline@v1.3.0-alltime-deepseek · 15543 in / 9531 out tokens · 88040 ms · 2026-08-02T18:40:47.263886+00:00 · methodology

0 comments
read the original 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.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

97 extracted references · 6 canonical work pages

  1. [1]

    Nature525(7567), 47–55 (2015)

    Bauer, P., Thorpe, A., Brunet, G.: The quiet revolution of numerical weather prediction. Nature525(7567), 47–55 (2015)

    2015

  2. [2]

    Nature566(7743), 195–204 (2019)

    Reichstein, M., Camps-Valls, G., Stevens, B., Jung, M., Denzler, J., Carvalhais, N., Prabhat, F.: Deep learning and process understanding for data-driven earth system science. Nature566(7743), 195–204 (2019)

    2019

  3. [3]

    Nature Geoscience17(10), 963–971 (2024)

    Eyring, V., Gentine, P., Camps-Valls, G., Lawrence, D.M., Reichstein, M.: Ai-empowered next-generation multiscale climate modelling for mitigation and adaptation. Nature Geoscience17(10), 963–971 (2024)

    2024

  4. [4]

    a calibration tool for parameterization improvement

    Couvreux, F., Hourdin, F., Williamson, D., Roehrig, R., Volodina, V., Ville- franque, N., Rio, C., Audouin, O., Salter, J., Bazile, E.,et al.: Process-based climate model development harnessing machine learning: I. a calibration tool for parameterization improvement. Journal of Advances in Modeling Earth Systems 13(3), 2020–002217 (2021)

    2020

  5. [5]

    Nature Communica- tions15(1), 9191 (2024)

    Plésiat, É., Dunn, R.J., Donat, M.G., Kadow, C.: Artificial intelligence reveals past climate extremes by reconstructing historical records. Nature Communica- tions15(1), 9191 (2024)

    2024

  6. [6]

    Earth System Science Data Discussions2025, 1–28 (2025) 10

    Guo, Z., Li, W., Ciais, P., Sitch, S., Werf, G.R., Bowring, S.P., Bastos, A., Mouil- lot, F., He, J., Sun, M.,et al.: Reconstructed global monthly burned area maps from 1901 to 2020. Earth System Science Data Discussions2025, 1–28 (2025) 10

    1901

  7. [7]

    Nature Reviews Physics7(1), 6–20 (2025)

    Bracco, A., Brajard, J., Dijkstra, H.A., Hassanzadeh, P., Lessig, C., Monteleoni, C.: Machine learning for the physics of climate. Nature Reviews Physics7(1), 6–20 (2025)

    2025

  8. [8]

    Nature climate change11(11), 966–972 (2021)

    Callaghan, M., Schleussner, C.-F., Nath, S., Lejeune, Q., Knutson, T.R., Re- ichstein, M., Hansen, G., Theokritoff, E., Andrijevic, M., Brecha, R.J.,et al.: Machine-learning-based evidence and attribution mapping of 100,000 climate impact studies. Nature climate change11(11), 966–972 (2021)

    2021

  9. [9]

    Nature Machine Intelligence6(12), 1556–1565 (2024)

    Huang, Y., Bathiany, S., Ashwin, P., Boers, N.: Deep learning for predicting rate-induced tipping. Nature Machine Intelligence6(12), 1556–1565 (2024)

    2024

  10. [10]

    Bulletin of the American Meteorological Society105(3), 567–583 (2024) https://doi.org/10.1175/BAMS-D-23-0196.1

    McGovern, A., Bostrom, A., McGraw, M., Chase, R.J., Gagne, D.J., Ebert- Uphoff, I., Musgrave, K.D., Schumacher, A.: Identifying and categorizing bias in ai/ml for earth sciences. Bulletin of the American Meteorological Society105(3), 567–583 (2024) https://doi.org/10.1175/BAMS-D-23-0196.1

    work page doi:10.1175/bams-d-23-0196.1 2024

  11. [11]

    npj Climate Action, 2 (1),

    Debnath, R., Creutzig, F., Sovacool, B., Shuckburgh, E.: Harnessing human and machine intelligence for planetary-level climate action. npj Climate Action, 2 (1),

  12. [12]

    The Guardian (2025)

    Hall, R., Sample, I.: AI-driven weather prediction breakthrough reported. The Guardian (2025). Accessed: 2025-12-10

    2025

  13. [13]

    Azhar, A.: Fully AI-Driven System Signals a New Era in Weather Forecasting. HPCwire. Accessed: 2025-12-10 (2025). https://www.hpcwire.com/aiwire/2025/ 03/25/fully-ai-driven-system-signals-a-new-era-in-weather-forecasting/

    2025

  14. [14]

    https://arxiv.org/abs/2504.17028

    Khadir, I., Stevenson, S., Li, H., Krick, K., Burrows, A., Hall, D., Posey, S., Shen, S.S.P.: Democracy of AI Numerical Weather Models: An Example of Global Forecasting with FourCastNetv2 Made by a University Research Lab Using GPU (2025). https://arxiv.org/abs/2504.17028

    Pith/arXiv arXiv 2025

  15. [15]

    npj Climate Action4(1), 97 (2025)

    Kuznetsov, I., Jost, A.A., Pantiukhin, D., Shapkin, B., Jung, T., Koldunov, N.: Transforming climate services with llms and multi-source data integration. npj Climate Action4(1), 97 (2025)

    2025

  16. [16]

    arXiv preprint arXiv:2510.24373 (2025)

    Galaz, V., Schewenius, M., Donges, J.F., Fetzer, I., Zhivkoplias, E., Barfuss, W., Delannoy, L., Wang-Erlandsson, L., Gelbrecht, M., Heitzig, J., et al.: Ai for a planet under pressure. arXiv preprint arXiv:2510.24373 (2025)

    arXiv 2025

  17. [17]

    Overland, I., Fossum Sagbakken, H., Isataeva, A., Kolodzinskaia, G., Simpson, N.P., Trisos, C., Vakulchuk, R.: Funding flows for climate change research on africa: where do they come from and where do they go? Climate and Development 14(8), 705–724 (2022)

    2022

  18. [18]

    Quarterly journal of the royal meteorological society146(730), 1999–2049 (2020)

    Hersbach, H., Bell, B., Berrisford, P., Hirahara, S., Horányi, A., Muñoz-Sabater, 11 J., Nicolas, J., Peubey, C., Radu, R., Schepers, D.,et al.: The era5 global reanal- ysis. Quarterly journal of the royal meteorological society146(730), 1999–2049 (2020)

    1999

  19. [19]

    Jiao, D., Xu, N., Yang, F., Xu, K.: Evaluation of spatial-temporal variation performance of ERA5 precipitation data in China. Sci. Rep., 11, 17956 (2021)

    2021

  20. [20]

    Quarterly Journal of the Royal Meteorological Society148(748), 3152–3165 (2022)

    Lavers, D.A., Simmons, A., Vamborg, F., Rodwell, M.J.: An evaluation of era5 precipitation for climate monitoring. Quarterly Journal of the Royal Meteorological Society148(748), 3152–3165 (2022)

    2022

  21. [21]

    Dynamics of Atmospheres and Oceans106, 101442 (2024)

    Silva, Q.P., Moreira, D.S., Freitas, H.C., Domingues, L.M.: Discrepancies between observationandera5reanalysisintheamazondeforestationcontext:Acasestudy. Dynamics of Atmospheres and Oceans106, 101442 (2024)

    2024

  22. [22]

    Scientific reports15(1), 7268 (2025)

    Polasky, A., Sapkota, V., Forest, C.E., Fuentes, J.D.: Discrepancies in precipita- tion trends between observational and reanalysis datasets in the amazon basin. Scientific reports15(1), 7268 (2025)

    2025

  23. [23]

    Environmental Research Letters20(10), 104043 (2025)

    Cattelan, L.G., Mattos, C.R., Hirota, M.: Unraveling divergences: disagree- ment between precipitation datasets and stations in tropical south america. Environmental Research Letters20(10), 104043 (2025)

    2025

  24. [24]

    Science of The Total Environment828, 154459 (2022)

    Zou,J.,Lu,N.,Jiang,H.,Qin,J.,Yao,L.,Xin,Y.,Su,F.:Performanceofairtem- perature from era5-land reanalysis in coastal urban agglomeration of southeast china. Science of The Total Environment828, 154459 (2022)

    2022

  25. [25]

    Scientific reports12(1), 21776 (2022)

    Xin, Y., Yang, Y., Chen, X., Yue, X., Liu, Y., Yin, C.: Evaluation of imerg and era5 precipitation products over the mongolian plateau. Scientific reports12(1), 21776 (2022)

    2022

  26. [26]

    https: //arxiv.org/abs/2601.04701

    Christensen, H.M., Barker, J., Antonio, B., Bonavita, M., Dahoui, M., Rosnay, P.: Error in ERA5 2m Temperature identified using GraphCast (2026). https: //arxiv.org/abs/2601.04701

    arXiv 2026

  27. [27]

    Geoscientific Model Development9(5), 1937–1958 (2016)

    Eyring, V., Bony, S., Meehl, G.A., Senior, C.A., Stevens, B., Stouffer, R.J., Tay- lor, K.E.:Overviewof the coupledmodel intercomparison project phase 6(cmip6) experimental design and organization. Geoscientific Model Development9(5), 1937–1958 (2016)

    1937

  28. [28]

    Allen,A.,Markou,S.,Tebbutt,W.,et al.:End-to-enddata-drivenweatherpredic- tion.Nature641,1172–1179(2025)https://doi.org/10.1038/s41586-025-08897-0

    doi:10.1038/s41586-025-08897-0 2025

  29. [29]

    Scientific data7(1), 225 (2020) 12

    Pastorello, G., Trotta, C., Canfora, E., Chu, H., Christianson, D., Cheah, Y.-W., Poindexter, C., Chen, J., Elbashandy, A., Humphrey, M.,et al.: The fluxnet2015 dataset and the oneflux processing pipeline for eddy covariance data. Scientific data7(1), 225 (2020) 12

    2020

  30. [30]

    Earth System Science Data13, 3607–3689 (2021) https://doi.org/10.5194/essd-13-3607-2021

    Delwiche, K.B., Knox, S.H., Malhotra, A., Fluet-Chouinard, E., McNicol, G., Feron, S., Ouyang, Z., Papale, D., Trotta, C., Canfora, E., Cheah, Y.-W., Chris- tianson, D., Alberto, M.C.R., Alekseychik, P., Aurela, M., Baldocchi, D., Bansal, S., Billesbach, D.P., Bohrer, G., Bracho, R., Buchmann, N., Campbell, D.I., Celis, G., Chen, J., Chen, W., Chu, H., Da...

    2021

  31. [31]

    Scientific data6(1), 74 (2019)

    Jung, M., Koirala, S., Weber, U., Ichii, K., Gans, F., Camps-Valls, G., Papale, D., Schwalm, C., Tramontana, G., Reichstein, M.: The fluxcom ensemble of global land-atmosphere energy fluxes. Scientific data6(1), 74 (2019)

    2019

  32. [32]

    Scientific Data10(1), 440 (2023)

    Nathaniel, J., Liu, J., Gentine, P.: Metaflux: Meta-learning global carbon fluxes from sparse spatiotemporal observations. Scientific Data10(1), 440 (2023)

    2023

  33. [33]

    Scientific Data 12(1), 1359 (2025)

    Yuan, Q., Wang, X., Che, T., Li, J.: Global carbon flux dataset generated by fusing remote sensing and multiple flux networks observation. Scientific Data 12(1), 1359 (2025)

    2025

  34. [34]

    Science381(6654), 164–170 (2023)

    Wong, F., Fuente-Nunez, C., Collins, J.J.: Leveraging artificial intelligence in the fight against infectious diseases. Science381(6654), 164–170 (2023)

    2023

  35. [35]

    Nature 638(8051), 623–635 (2025)

    Kraemer, M.U., Tsui, J.L.-H., Chang, S.Y., Lytras, S., Khurana, M.P., Van- derslott, S., Bajaj, S., Scheidwasser, N., Curran-Sebastian, J.L., Semenova, E., et al.: Artificial intelligence for modelling infectious disease epidemics. Nature 638(8051), 623–635 (2025)

    2025

  36. [36]

    In: NeurIPS Workshop on Attributing Model Behavior at Scale (2023)

    Marion, M., Üstün, A., Pozzobon, L., Wang, A., Fadaee, M., Hooker, S.: When less is more: Investigating data pruning for pretraining LLMs at scale. In: NeurIPS Workshop on Attributing Model Behavior at Scale (2023). https://openreview.net/forum?id=XUIYn3jo5T 13

    2023

  37. [37]

    In: Submitted to ACL Rolling Review - February 2025 (2025)

    Wang, Z., Wu, Z., Zhang, J., Guan, X., Jain, N., Lu, S., Gupta, S., Koshiyama, A.: Bias amplification: Large language models as increasingly biased media. In: Submitted to ACL Rolling Review - February 2025 (2025). under review. https://openreview.net/forum?id=X5IaJtOtNL

    2025

  38. [38]

    https://arxiv.org/abs/2409

    Lin, H.: Designing Domain-Specific Large Language Models: The Critical Role of Fine-Tuning in Public Opinion Simulation (2024). https://arxiv.org/abs/2409. 19308

    2024

  39. [39]

    Global Environmental Change35, 279–288 (2015) https://doi.org/ 10.1016/j.gloenvcha.2015.09.018

    Pasgaard, M., Dalsgaard, B., Maruyama, P.K., Sandel, B., Strange, N.: Geo- graphical imbalances and divides in the scientific production of climate change knowledge. Global Environmental Change35, 279–288 (2015) https://doi.org/ 10.1016/j.gloenvcha.2015.09.018

    work page doi:10.1016/j.gloenvcha.2015.09.018 2015

  40. [40]

    arXiv preprint arXiv:2402.02680 (2024)

    Manvi, R., Khanna, S., Burke, M., Lobell, D., Ermon, S.: Large language models are geographically biased. arXiv preprint arXiv:2402.02680 (2024)

    Pith/arXiv arXiv 2024

  41. [41]

    In: 2024 8th International Conference on Computational System and Information Technology for Sustainable Solutions (CSITSS), pp

    Sanu, E., Amudaa, T.K., Bhat, P., Dinesh, G., Kumar Chate, A.U., P, R.K.: Limitations of large language models. In: 2024 8th International Conference on Computational System and Information Technology for Sustainable Solutions (CSITSS), pp. 1–6 (2024). https://doi.org/10.1109/CSITSS64042.2024.10817070

    arXiv 2024

  42. [42]

    arXiv preprint arXiv:2506.16281 (2025)

    Zaidan, M.A., Motlagh, N.H., Nurmi, P., Hussein, T., Kulmala, M., Petäjä, T., Tarkoma, S.: Artificial intelligence for atmospheric sciences: A research roadmap. arXiv preprint arXiv:2506.16281 (2025)

    Pith/arXiv arXiv 2025

  43. [43]

    Computing Power Is Splitting the World Into Haves and Have-Nots

    Satariano, A., Mozur, P.: A.I. Computing Power Is Splitting the World Into Haves and Have-Nots. The New York Times. https://www.nytimes.com/interactive/ 2025/06/23/technology/ai-computing-global-divide.html Accessed 2025-11-19

    2025

  44. [44]

    Nature: Ai’s energy demand is booming — but where and by how much? Nature 639, 19–21 (2025) https://doi.org/10.1038/d41586-025-00616-z

    doi:10.1038/d41586-025-00616-z 2025

  45. [45]

    arXiv preprint arXiv:2304.03271 (2023)

    Li, P., Yang, J., Islam, M.A., Ren, S.: Making ai less “thirsty”: The hidden water footprint of ai models. arXiv preprint arXiv:2304.03271 (2023)

    Pith/arXiv arXiv 2023

  46. [46]

    Cloud Providers, and Accelerators (June 20, 2025) (2025)

    Hawkins, Z., Lehdonvirta, V., Wu, B.: Ai compute sovereignty: Infrastructure control across territories, cloud providers, and accelerators. Cloud Providers, and Accelerators (June 20, 2025) (2025)

    2025

  47. [47]

    SAIS Review of International Affairs44(2), 75–92 (2024)

    Regilme, S.S.F.: Artificial intelligence colonialism: Environmental damage, la- bor exploitation, and human rights crises in the global south. SAIS Review of International Affairs44(2), 75–92 (2024)

    2024

  48. [48]

    Artificial Intelligence for the Earth 14 Systems1(3), 210002 (2022)

    Dueben, P.D., Schultz, M.G., Chantry, M., Gagne, D.J., Hall, D.M., McGovern, A.: Challenges and benchmark datasets for machine learning in the atmospheric sciences: Definition, status, and outlook. Artificial Intelligence for the Earth 14 Systems1(3), 210002 (2022)

    2022

  49. [49]

    610–623 (2021)

    Bender, E.M., Gebru, T., McMillan-Major, A., Shmitchell, S.: On the dangers of stochastic parrots: Can language models be too big? In: Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, pp. 610–623 (2021)

    2021

  50. [50]

    Science390(6769), 111–111 (2025) https://doi.org/10.1126/science.aec8146 https://www.science.org/doi/pdf/10.1126/science.aec8146

    Taalas, P.: Free global access to climate and weather data must continue. Science390(6769), 111–111 (2025) https://doi.org/10.1126/science.aec8146 https://www.science.org/doi/pdf/10.1126/science.aec8146

    work page doi:10.1126/science.aec8146 2025

  51. [51]

    Press release, United Nations (2023)

    United Nations: United Nations General Assembly Press Release: GA/EF/3587. Press release, United Nations (2023). https://press.un.org/en/2023/gaef3587. doc.htm

    2023

  52. [52]

    So- cArXiv 7e2jf, Center for Open Science

    Linsenmeier, M., Shrader, J.G.: Global inequalities in weather forecasts. So- cArXiv 7e2jf, Center for Open Science. You might wonder how this can be true when the ECMWF forecasts have so quickly improved in both the Northern and Southern Hemisphere. This is because most countries need to develop their own forecasts to get more high-resolution predictions (2023)

    2023

  53. [53]

    ESS Open Archive (2025) https://doi.org/10.22541/ essoar.174802937.77365288/v2 174802937.77365288/v2

    Olivetti, L., Messori, G.: Whose weather is it? a fairness framework for data- driven weather forecasting. ESS Open Archive (2025) https://doi.org/10.22541/ essoar.174802937.77365288/v2 174802937.77365288/v2

    arXiv 2025

  54. [54]

    Research Square (2025) https: //doi.org/10.21203/rs.3.rs-6829857/v1

    Metcalfe, D., Anders, E., Axén, H.,et al.: Gaps in tropical science arising from biased spatial patterns of sampling and citation. Research Square (2025) https: //doi.org/10.21203/rs.3.rs-6829857/v1 . Preprint (Version 1)

    work page doi:10.21203/rs.3.rs-6829857/v1 2025

  55. [55]

    Earth System Dynamics14, 81–100 (2023) https://doi.org/10.5194/esd-14-81-2023

    Vries, I.E., Sippel, S., Pendergrass, A.G., Knutti, R.: Robust global detection of forced changes in mean and extreme precipitation despite observational disagree- ment on the magnitude of change. Earth System Dynamics14, 81–100 (2023) https://doi.org/10.5194/esd-14-81-2023

    work page doi:10.5194/esd-14-81-2023 2023

  56. [56]

    Journal of Medical Internet Research25 (2023) https://doi.org/10.2196/44649

    Maddah, N., Verma, A., Almashmoum, M., Ainsworth, J.: Effectiveness of public health digital surveillance systems for infectious disease prevention and control at mass gatherings: Systematic review. Journal of Medical Internet Research25 (2023) https://doi.org/10.2196/44649

    work page doi:10.2196/44649 2023

  57. [57]

    Frontiers in Public Health.11(2023) https://doi.org/10.3389/fpubh.2023.999958

    Digital dashboards visualizing public health data: a systematic review. Frontiers in Public Health.11(2023) https://doi.org/10.3389/fpubh.2023.999958

    arXiv 2023

  58. [58]

    https:// ourworldindata.org/burden-of-disease Accessed 2024-02-01

    Roser, M., Ritchie, H., Spooner, F.: Burden of Disease (2025). https:// ourworldindata.org/burden-of-disease Accessed 2024-02-01

    2025

  59. [59]

    One Health15, 100439 (2022)

    Keshavamurthy, R., Dixon, S., Pazdernik, K.T., Charles, L.E.: Predicting infec- tious disease for biopreparedness and response: A systematic review of machine 15 learning and deep learning approaches. One Health15, 100439 (2022)

    2022

  60. [60]

    New England Journal of Medicine 388(17), 1597–1607 (2023)

    Brownstein, J.S., Rader, B., Astley, C.M., Tian, H.: Advances in artificial in- telligence for infectious-disease surveillance. New England Journal of Medicine 388(17), 1597–1607 (2023)

    2023

  61. [61]

    everyone wants to do the model work, not the data work

    Sambasivan, N., Kapania, S., Highfill, H., Akrong, D., Paritosh, P.K., Aroyo, L.: “everyone wants to do the model work, not the data work”: Data cascades in high-stakes ai. Proceedings of the 2021 CHI Conference on Human Factors in Computing Systems (2021)

    2021

  62. [62]

    Journal of Moral Philosophy 21(1-2), 62–83 (2023)

    Hellman, D.: Big data and compounding injustice. Journal of Moral Philosophy 21(1-2), 62–83 (2023)

    2023

  63. [63]

    Nature Communications16(1), 2564 (2025)

    Reichstein, M., Benson, V., Blunk, J., Camps-Valls, G., Creutzig, F., Fearn- ley, C.J., Han, B., Kornhuber, K., Rahaman, N., Schölkopf, B.,et al.: Early warning of complex climate risk with integrated artificial intelligence. Nature Communications16(1), 2564 (2025)

    2025

  64. [64]

    Technical report, Institute for Technology and Society of Rio de Janeiro (ITS Rio) (October 2025)

    Lemos, R.: Digital public infrastructure for climate: The missing backbone for climate action. Technical report, Institute for Technology and Society of Rio de Janeiro (ITS Rio) (October 2025). Report commissioned by the COP30 Presidency. https://itsrio.org

    2025

  65. [65]

    arXiv preprint arXiv:2203.15556 (2022)

    Hoffmann, J., Borgeaud, S., Mensch, A., et al.: Training compute-optimal large language models. arXiv preprint arXiv:2203.15556 (2022)

    Pith/arXiv arXiv 2022

  66. [66]

    World Bank, Washington, DC (2022)

    Eyre, J., Kull, D., Riishojgaard, L., et al.: The value of surface-based meteo- rological observation data. World Bank, Washington, DC (2022). Available at: https://openknowledge.worldbank.org/handle/10986/35178

    2022

  67. [67]

    ACM Computing Surveys57(5), 1–42 (2025)

    Zha, D., Bhat, Z.P., Lai, K.-H., Yang, F., Jiang, Z., Zhong, S., Hu, X.: Data- centric artificial intelligence: A survey. ACM Computing Surveys57(5), 1–42 (2025)

    2025

  68. [68]

    https://arxiv.org/abs/2501.19374

    Subich, C., Husain, S.Z., Separovic, L., Yang, J.: Fixing the Double Penalty in Data-Driven Weather Forecasting Through a Modified Spherical Harmonic Loss Function (2025). https://arxiv.org/abs/2501.19374

    Pith/arXiv arXiv 2025

  69. [69]

    International Journal of Health Geographics5(1), 38 (2006)

    Hanigan, I., Hall, G., Dear, K.B.: A comparison of methods for calculating pop- ulation exposure estimates of daily weather for health research. International Journal of Health Geographics5(1), 38 (2006)

    2006

  70. [70]

    arXiv preprint arXiv:2510.26099 (2025) 16

    Masi, N., Balestriero, R.: Safe: A novel approach to ai weather evaluation through stratified assessments of forecasts over earth. arXiv preprint arXiv:2510.26099 (2025) 16

    arXiv 2025

  71. [71]

    Wiley Interdisciplinary Reviews: Climate Change8(6), 482 (2017)

    Bremer, S., Meisch, S.: Co-production in climate change research: reviewing dif- ferent perspectives. Wiley Interdisciplinary Reviews: Climate Change8(6), 482 (2017)

    2017

  72. [72]

    Current opinion in environmental sustainability42, 15–21 (2020)

    Turnhout, E., Metze, T., Wyborn, C., Klenk, N., Louder, E.: The politics of co-production: participation, power, and transformation. Current opinion in environmental sustainability42, 15–21 (2020)

    2020

  73. [73]

    Nature641, 1180–1187 (2025) https://doi.org/10.1038/ s41586-025-09005-y

    Bodnar, C., Bruinsma, W.P., Lucic, A.,et al.: A foundation model for the earth system. Nature641, 1180–1187 (2025) https://doi.org/10.1038/ s41586-025-09005-y

    2025

  74. [74]

    Nature637, 84–90 (2025) https://doi.org/10.1038/ s41586-024-08252-9

    Price, I., Sanchez-Gonzalez, A., Alet, F.,et al.: Probabilistic weather forecast- ing with machine learning. Nature637, 84–90 (2025) https://doi.org/10.1038/ s41586-024-08252-9

    2025

  75. [75]

    Communications Earth & Environment6, 518 (2025) https://doi.org/10.1038/ s43247-025-02502-y

    Chen, K., Han, T., Ling, F.,et al.: The operational medium-range deter- ministic weather forecasting can be extended beyond a 10-day lead time. Communications Earth & Environment6, 518 (2025) https://doi.org/10.1038/ s43247-025-02502-y

    2025

  76. [76]

    Science382, 1416–1421 (2023) https://doi.org/10.1126/science.adi2336

    Lam, R.,et al.: Learning skillful medium-range global weather forecasting. Science382, 1416–1421 (2023) https://doi.org/10.1126/science.adi2336

    doi:10.1126/science.adi2336 2023

  77. [77]

    Nature632(8027), 1060–1066 (2024)

    Kochkov, D., Yuval, J., Langmore, I., Norgaard, P., Smith, J., Mooers, G., Klöwer, M., Lottes, J., Rasp, S., Düben, P.,et al.: Neural general circulation models for weather and climate. Nature632(8027), 1060–1066 (2024)

    2024

  78. [78]

    URL https://arxiv

    Lang, S., Alexe, M., Clare, M.C., Roberts, C., Adewoyin, R., Bouallègue, Z.B., Chantry, M., Dramsch, J., Dueben, P.D., Hahner, S., et al.: Aifs-crps: Ensemble forecasting using a model trained with a loss function based on the continuous ranked probability score, 2024b. URL https://arxiv. org/abs/2412.15832

    Pith/arXiv arXiv

  79. [79]

    https://arxiv.org/abs/2406.01465

    Lang, S., Alexe, M., Chantry, M., Dramsch, J., Pinault, F., Raoult, B., Clare, M.C.A., Lessig, C., Maier-Gerber, M., Magnusson, L., Bouallègue, Z.B., Nemesio, A.P., Dueben, P.D., Brown, A., Pappenberger, F., Rabier, F.: AIFS – ECMWF’s data-driven forecasting system (2024). https://arxiv.org/abs/2406.01465

    Pith/arXiv arXiv 2024

  80. [80]

    Nature619, 533–538 (2023) https://doi.org/ 10.1038/s41586-023-06185-3

    Bi, K., Xie, L., Zhang, H.,et al.: Accurate medium-range global weather fore- casting with 3d neural networks. Nature619, 533–538 (2023) https://doi.org/ 10.1038/s41586-023-06185-3

    doi:10.1038/s41586-023-06185-3 2023

Showing first 80 references.