Pith. sign in

REVIEW 4 major objections 5 minor 45 references

Adding GNSS-derived column-moisture measurements to the Aurora weather foundation model improves 6-hour precipitation forecasts, with the largest gains — up to an 8.8% rise in Equitable Threat Score — at the 99th percentile of rain intensit

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

2026-08-02 08:24 UTC pith:Y5ZLCBC4

load-bearing objection The paper is careful and the empirical pattern is consistent, but the headline GNSS attribution rests on two confounds the authors disclose yet never control for: ZWDX is an ERA5-based proxy, and the Step 2 design adds ZWD as both input and auxiliary loss, so the gain could be multi-task regularization rather than the input signal. the 4 major comments →

arxiv 2607.05658 v2 pith:Y5ZLCBC4 submitted 2026-07-06 physics.ao-ph cs.LGphysics.geo-ph

Integrating GNSS-Derived Zenith Wet Delay into a Weather Foundation Model Improves Precipitation Forecasting

Leonardo Trentini , Fanny Lehmann , Laura Crocetti , Benedikt Soja This is my paper
classification physics.ao-ph cs.LGphysics.geo-ph
keywords GNSSZenith Wet Delayprecipitation forecastingweather foundation modelAuroraextreme precipitationcolumn water vapourEquitable Threat Score
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.

This paper claims that Zenith Wet Delay (ZWD) — the delay GNSS signals accumulate from atmospheric water vapour, a direct measure of column moisture — can be added to a leading machine-learning weather foundation model (Aurora) through a simple extension of its surface encoder, and that doing so systematically improves fine-tuned 6-hour precipitation forecasts. The improvement grows with event severity: at the 99th percentile of observed precipitation, the Equitable Threat Score rises by 8.8% relative to an identically trained model without ZWD, and the predicted precipitation power spectrum becomes more realistic at synoptic and planetary scales. The gain is reproduced across training checkpoints, persists over short-to-intermediate lead times, and is larger for a smaller model, suggesting that the column-moisture signal is most valuable when the network has less capacity to reconstruct it internally. A sympathetic reader would care because heavy precipitation is precisely the forecast target where both numerical and machine-learning models are weakest, and ZWD is an all-weather, observation-derived quantity not currently used by these models.

Core claim

The paper's central claim is that GNSS-derived Zenith Wet Delay, delivered as a gridded product (ZWDX) and inserted into Aurora as a new surface variable, carries information that Aurora can exploit for precipitation forecasting. The authors fine-tune Aurora for 6-hour accumulated precipitation in two matched configurations that differ only in whether ZWD is included, and report that the ZWD model learns ZWD itself at parity with pretrained variables (R=0.998, relative MAE 1.92%), improves precipitation MAE by 0.7%, RMSE by 1.8%, MSE by 3.7%, and FSS at the 95th percentile by 1.5%, and raises ETS by +1.2%, +1.9%, +3.9%, and +8.8% at the 75th, 90th, 95th, and 99th percentiles. They further re

What carries the argument

The central object is Zenith Wet Delay (ZWD), the signal delay a GNSS satellite link accumulates from atmospheric water vapour, mapped to the zenith and separated from the hydrostatic delay; it is a direct, weather-independent measure of column water vapour. The carrying mechanism is the ZWDX gridded product (an XGBoost model trained on over 19,000 GNSS stations) whose 0.25-degree, 6-hourly ZWD fields are added to Aurora as a new surface variable through an extension of the linear surface encoder and decoder, leaving the Perceiver encoder, 3D Swin Transformer U-Net backbone, and decoder untouched. The experimental contrast is between two fine-tuned models that share initialisation, optimiser

Load-bearing premise

The load-bearing premise is that the ZWDX labels contain genuine GNSS moisture information beyond what ERA5 already supplies — the paper admits the product is XGBoost-predicted from ERA5 inputs and never tests an ERA5-only column-moisture baseline; a secondary premise is that ten checkpoints from one training run per model approximate run-to-run variability.

What would settle it

Train the 'With ZWD' and 'Without ZWD' models with the same protocol but replace ZWD with an ERA5-only integrated-water-vapour column (or with ZWDX values whose GNSS-labelled ordering has been destroyed, e.g., shuffled in time/location). If the 99th-percentile ETS gain vanishes or shrinks to noise, the reported benefit is not attributable to GNSS observations, only to adding any column-moisture variable. A second check: run the 110M model from at least 3 independent seeds and see whether the +50% ETS99 gain holds across seeds.

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

Share X Bluesky LinkedIn Reddit HN

If this is right

  • If the claim holds, weather foundation models can ingest variables absent from reanalysis and improve on their hardest task: extreme precipitation.
  • Downstream users doing short-range (6h to ~3-day) precipitation forecasting from fine-tuned foundation models should expect the largest benefit in the heavy-precipitation tail, where ETS gains reach roughly 9% at the large scale and 50% at the small scale.
  • The result implies that the moisture column, not just its level-resolved humidity profile, is a useful conditioning variable for AI precipitation models; ZWD effectively hands the model the column integral directly.
  • Because the gain grows as model capacity shrinks, smaller or regional models — the typical operational setting — may benefit disproportionately from such observational inputs.
  • The spectral improvement at synoptic/planetary scales suggests ZWD helps correct the over-smoothing of precipitation fields, not merely their intensity calibration.

Where Pith is reading between the lines

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

  • One testable extension the paper leaves implicit: replace ZWD with an ERA5-only integrated-water-vapour column (or with ZWDX fields whose GNSS-ordering is destroyed) and rerun the matched-pair contrast. If the 99th-percentile ETS gain vanishes, the benefit is about adding column moisture, not about GNSS observations specifically; if it survives, the GNSS labels genuinely matter.
  • If the attribution is real, feeding raw GNSS station delays directly rather than the XGBoost-smoothed product should increase the gain, especially in data-sparse regions — a prediction the paper gestures at in its discussion.
  • The same mechanism should transfer to other non-reanalysis observables (radar reflectivity, dense environmental sensor networks) given the encoder can ingest them, so the general principle is broader than ZWD.
  • The paper's checkpoint-pairing design samples within-trajectory variability, not seed-to-seed variability; a multi-seed repeat at the 110M scale would help quantify run-to-run robustness, which the authors explicitly note is prohibitively expensive at 1.3B.

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

4 major / 5 minor

Summary. The paper reports fine-tuning the Aurora weather foundation model with an additional surface variable, GNSS-derived Zenith Wet Delay (ZWD) from the ZWDX gridded product, and evaluates the effect on 6-hour accumulated precipitation forecasting. Precipitation is a new downstream task not seen in pretraining. The authors compare two models, A (with ZWD as input and auxiliary output/loss) and B (without ZWD), sharing initialization, optimizer, data, and step budget. They report that ZWD is learned at skill comparable to Aurora's pretrained variables; that including ZWD improves precipitation metrics, with gains growing with event severity (headline 8.8% ETS increase at the 99th percentile); that the precipitation power spectrum becomes closer to the target at all resolved scales; and that the benefit is larger for a smaller 110M-parameter variant. Results are evaluated on a held-out test set (April–December 2020) and are averaged over ten late-training checkpoints per model, with paired significance tests.

Significance. If the reported effect is genuinely attributable to the GNSS-derived information, the paper would be a valuable demonstration that an independent observational stream can improve a foundation-model weather forecast on a high-impact, tail-weighted target. The study has concrete strengths: public code and data links, a held-out test period, evaluation at two model scales, a checkpoint-ensemble analysis, and a generally careful discussion of limitations, including the ERA5 dependence of ZWDX and the non-conservative regridding of the precipitation target. However, the causal attribution of the precipitation gain to the GNSS input is currently underdetermined: the main comparison varies both the input and an auxiliary output/loss, and no control is provided against an ERA5-only column-moisture input. These are not mere presentation issues; they are load-bearing for the abstract's central claim.

major comments (4)
  1. [Methods, 'Experimental design'; Eq. (1)] Model A differs from Model B in two ways: A receives ZWD as an additional input and also predicts ZWD as an auxiliary output with loss weight lambda_ZWD=2, while B has neither. The text states that this contrast 'isolates the contribution of GNSS-derived ZWD,' but it cannot separate the input effect from the multi-task/auxiliary-target effect. A model that is asked to predict ZWD may develop better intermediate moisture representations even if ZWD is never used as an input. The loss-weight sweep in Supplementary Note 2 always keeps the ZWD output in A and absent in B, so it does not address this confound. A control with ZWD as output only (no input) or as input only (no auxiliary loss) is needed to support the causal claim.
  2. [Methods, 'ZWDX global gridded ZWD product'; Discussion] The manuscript acknowledges that ZWDX is an XGBoost function of ERA5 specific humidity, location, and time, trained on GNSS station labels, and that Aurora is pretrained on ERA5. Yet no control is run with an ERA5-only column-moisture field (e.g., ERA5 integrated water vapour or column-integrated specific humidity) in place of ZWDX. Without such a control, the reported precipitation gains could be obtained from any ERA5-derived moisture-column input, and the 'GNSS' attribution is not established. The Discussion's statement that the gains are a 'conservative lower bound' assumes, rather than demonstrates, that the GNSS labels contribute information beyond ERA5 humidity. An ablation replacing the ZWDX input with a purely ERA5-derived moisture column would directly address this.
  3. [Methods, 'Checkpoint ensemble and paired significance testing'] The ten checkpoints are sampled from a single training run per model at matched training steps. They are not independent observations; consecutive checkpoints along the same trajectory are autocorrelated. The paired t-test treats the ten paired differences as independent, which can underestimate the standard error and inflate the t-statistic when the differences are positively autocorrelated. The assertion that 'a paired t-test moreover remains valid under any pairing' refers to the A_i-versus-B_i pairing, not to independence across i. The reported p-values (e.g., Table 3, p=0.004 for ETS99) are therefore likely anti-conservative. A block-bootstrap or mixed-effects analysis, or at least an estimate of the effective sample size, should be reported.
  4. [Results, 'The precipitation gain costs little...' / Supp. Note 3] The three-way 110M ablation (surface-only, precipitation-only without ZWD, precipitation-plus-ZWD) is useful, but it still does not separate the input and auxiliary-output channels: the 'With ZWD' configuration adds both the ZWD input and the ZWD output/loss relative to the 'Without ZWD' configuration. The same control described in the first major comment is needed at this scale as well before the inverse-capacity-scaling claim can be cleanly interpreted as evidence about the ZWD input signal specifically.
minor comments (5)
  1. [Methods, Eq. (2)] The Log Spectral Distance is described in the text as a root-mean-square difference over wavenumbers, but the displayed equation is sqrt((log10 S_pred - log10 S_target)^2), which is just an absolute value. The averaging over wavenumbers should be made explicit in the equation.
  2. [Introduction / Abstract] The abstract states 'the first integration of GNSS-derived ZWD into Aurora' without the 'to our knowledge' qualifier used in the Introduction. Please align the wording, and note that the same applies to 'first integration of a geodetic observable into an atmospheric foundation model' in the Results/Discussion.
  3. [Results, Fig. 3 caption] The caption says the rollout scores are 'averaged over ten initialisation dates spanning the held-out test set.' Please clarify whether these ten dates are the same as the ten checkpoints used in the paired tests; if not, the relationship between the two ensembles should be stated.
  4. [Methods, 'Precipitation target and regridding'] The bilinear-versus-conservative regridding tradeoff is discussed clearly. A sentence noting that the absolute scores may depend on the regridding choice while the contrast is unaffected is already present and is helpful.
  5. [Supplementary Note 2, Table S2] At the 75th percentile, lw=10 (0.692) falls below the 'No ZWD' baseline (0.695). The text says 'drops below it,' which is correct, but the main-text sentence 'only at lambda_ZWD=10 does skill fall back to near-baseline' should be reconciled with this below-baseline value for precision.

Circularity Check

1 steps flagged

GNSS attribution is partly circular because ZWDX is a fitted transform of ERA5 inputs that Aurora already receives; the held-out precipitation result itself is independent.

specific steps
  1. fitted input called prediction [Methods, 'ZWDX global gridded ZWD product'; Discussion]
    "Because ZWDX is XGBoost-predicted from ERA5 inputs and station coordinates, GNSS information enters Aurora through the training-target labels rather than as raw, independent observations; the implications of this are discussed in Discussion."

    By the paper's own description, ZWDX is a fixed nonlinear transform of ERA5 humidity, location, and time, and Aurora already receives ERA5 fields. The added ZWD channel is therefore a fitted re-encoding of information already present in the model's input set, with GNSS content entering only through XGBoost's training labels. Attributing the precipitation gain to 'GNSS observations' is not tested against an ERA5-only control (e.g., a column-moisture transform trained without GNSS labels). This is partial attribution-level circularity: the precipitation target is held-out MSWEP, so the forecast comparison is not a fit to the target, but the causal GNSS reading is not established.

full rationale

The quantitative core of the paper—With-ZWD vs Without-ZWD precipitation skill on a held-out test period with matched initialization, optimizer, schedule, and data—is externally grounded and not fit to the target; no equation reduces the 6-hour precipitation score to the training objective. The main circular-adjacent element is ZWDX itself: because it is XGBoost-predicted from ERA5 inputs and station coordinates, the GNSS signal enters as training labels rather than as an independent input stream, and Aurora is pretrained on ERA5. The paper acknowledges this limitation but does not run the ERA5-only control that would separate a genuine GNSS contribution from a beneficial nonlinear re-encoding of ERA5 moisture. The Step 2 A/B comparison also changes the ZWD auxiliary output/loss together with the input; this is a confounding of attribution, but it is an experimental-design/correctness risk rather than a by-construction circularity, so it does not by itself raise the score further. Self-citations for ZWDX are from the same group, but the product is published, DOI-available, and externally falsifiable, so they are not load-bearing in the sense of a self-citation uniqueness chain.

Axiom & Free-Parameter Ledger

2 free parameters · 4 axioms · 0 invented entities

No new physical entities are introduced; ZWD is a measured atmospheric quantity and ZWDX is a statistical surrogate, not an invented entity. The central extras the reader pays for are the two loss weights and the strong assumption that ZWDX carries information beyond ERA5.

free parameters (2)
  • lambda_ZWD = 2 for 1.3B model; 0.1 for 110M model
    Loss weight for ZWD in the Step 2 objective; chosen from sweep {1,2,3,10} as a plateau optimum. Gains are stable over {1,2,3} but depend on this weighting choice.
  • lambda_precip = 1 for 1.3B model; 0.1 for 110M model
    Loss weight for MSWEP precipitation in the fine-tuning objective. Equal in Models A and B, so it cancels in the with/without-ZWD contrast, but it affects absolute skill levels.
axioms (4)
  • domain assumption ZWDX fields, trained on ERA5 humidity to match GNSS station ZWD, retain the GNSS station-resolved moisture structure that ERA5 filters out.
    Load-bearing for attributing gains to GNSS; stated in the Discussion ('gains obtained despite this residual ERA5 dependence') but not independently verified inside the paper.
  • domain assumption GNSS ZWD is a direct, all-weather measure of column moisture, linearly related to IWV via a weak temperature dependence.
    Accepted geodetic background from refs 12-14; not established by this paper.
  • ad hoc to paper Ten checkpoints sampled from one training run per model approximate run-to-run variability for paired significance testing.
    Explicitly stated in Methods as a deliberate compromise. Checkpoints are temporally autocorrelated, so paired p-values are indicative only.
  • domain assumption Bilinear regridding of MSWEP to 0.25 degrees and its biases cancel in the with/without-ZWD contrast.
    Discussed in Methods; the authors note that a conservative remapping could change absolute scores and gain magnitudes, though not the sign of the contrast.

reviewed 2026-08-02 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Integrating GNSS-Derived Zenith Wet Delay into a Weather Foundation Model Improves Precipitation Forecasting." pith.science (2026). https://pith.science/paper/Y5ZLCBC4

@misc{pith2026260705658,
  author       = {Pith},
  title        = {Pith review of: Integrating GNSS-Derived Zenith Wet Delay into a Weather Foundation Model Improves Precipitation Forecasting},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Y5ZLCBC4}},
  note         = {Machine review of arXiv:2607.05658}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Global Navigation Satellite Systems (GNSS), best known for positioning, also serve weather science, as atmospheric water vapour delays their signals. This delay, the Zenith Wet Delay (ZWD), is a direct, all-weather measure of column moisture. Although assimilated into numerical weather prediction for decades, ZWD is not yet used by leading Machine Learning Weather Models (MLWM), despite addressing a known deficiency: the underestimation of severe precipitation. Here we present the first integration of GNSS-derived ZWD into Aurora, a state-of-the-art weather foundation model. Our extended Aurora learns ZWD with skill comparable to its pretrained variables. More importantly, including ZWD systematically improves forecasts when fine-tuning for 6-hour accumulated precipitation. Gains grow with severity, reaching an 8.8% increase in Equitable Threat Score at the 99th percentile, while the precipitation power spectrum becomes more realistic at synoptic and planetary scales. GNSS observations therefore encode information that MLWM can exploit for high-impact precipitation.

Figures

Figures reproduced from arXiv: 2607.05658 by Benedikt Soja, Fanny Lehmann, Laura Crocetti, Leonardo Trentini.

Figure 1
Figure 1. Figure 1: Step 2 deterministic precipitation skill on the held-out test set. (a) Global spatial [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 1
Figure 1. Figure 1: Integrating a GNSS-derived observable into a weather foundation model. Ground-based GNSS receivers measure the tropospheric delay accumulated by satellite signals; removing the hydrostatic contribution isolates the Zenith Wet Delay (ZWD), a direct, all-weather measure of column water vapour, which the ZWDX product renders as a global 0.25◦ field. ZWD and 6-hour accumulated precipitation - both absent from … view at source ↗
Figure 2
Figure 2. Figure 2: Step 2 threshold-based and extreme-event precipitation skill with and without [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 2
Figure 2. Figure 2: ZWD reduces 6-hour precipitation RMSE across most of the globe. Global spatial distribution of the 6-hour precipitation RMSE difference for the main-analysis checkpoint (red: “With ZWD” improves upon the baseline “Without ZWD”; blue: baseline outperforms). 5 [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Step 2 spectral precipitation skill. (a) Ratio of predicted to target latitude [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figure 3
Figure 3. Figure 3: The ZWD advantage persists through the short-to-intermediate forecast range. Equitable Threat Score as a function of lead time at four climatological percentiles (75th, 90th, 95th, 99th) for Model A (“With ZWD”; solid) and Model B (“Without ZWD”; dotted), averaged over ten initialisation dates spanning the held-out test set. Forecasts are produced purely autoregressively, without rollout fine-tuning. Skill… view at source ↗
Figure 4
Figure 4. Figure 4: Where ZWD improves the forecast during high-impact events. Per-cell skill difference for extreme precipitation cells only (>P90 of the observed regional distribution), computed as the absolute “Without ZWD” error minus the absolute “With ZWD” error, so that green indicates the ZWD-enriched forecast is closer to MSWEP and purple the baseline: a South Asia monsoon, initialised 2020-08-04 12Z; b Typhoon Bavi,… view at source ↗
Figure 5
Figure 5. Figure 5: The ZWD gain concentrates in the heavy-precipitation tail of individual high-impact events. Precipitation error reduction (positive = “With ZWD” better) versus observed-intensity bin for three extreme events drawn from distinct basins and regimes, all within the held-out test set: a South Asia monsoon (initialised 2020-08-04 12Z), b Typhoon Bavi, East Asia (2020-08-24 06Z), c Hurricane Delta, Central Ameri… view at source ↗
Figure 6
Figure 6. Figure 6: Including ZWD moves the predicted precipitation spectrum closer to the target at every scale. Ratio of predicted to target latitude-weighted zonal power spectrum by wavenumber band for the main-analysis checkpoint (100% = perfect match), for Model A (“With ZWD”) and Model B (“Without ZWD”). The corresponding checkpoint-averaged Log Spectral Distances are given in Supplementary Note 5. in absolute terms: th… view at source ↗
Figure 7
Figure 7. Figure 7: Schematic of the two-step fine-tuning workflow. Starting from the pretrained Aurora checkpoint, Step 1 fine-tunes the model on the full ERA5 state augmented with ZWDX￾gridded ZWD. Step 2 trains two precipitation models - Model A (with ZWD) and Model B (baseline) - that share optimiser, learning-rate schedule, number of steps and data, and differ only in whether ZWD is included as an additional input and au… view at source ↗

Review history (2 revisions) →

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

45 extracted references · 17 canonical work pages

  1. [1]

    Pathak, S

    J. Pathak, S. Subramanian, P. Harrington, S. Raja, A. Chattopadhyay, M. Mardani, T. Kurth, D. Hall, Z. Li, K. Azizzadenesheli, P. Hassanzadeh, K. Kashinath, and A. Anand- kumar. FourCastNet: A global data-driven high-resolution weather model using adaptive Fourier neural operators, 2022

    2022

  2. [2]

    K. Bi, L. Xie, H. Zhang, X. Chen, X. Gu, and Q. Tian. Accurate medium-range global weather forecasting with 3D neural networks.Nature, 619:533–538, 2023. doi: 10.1038/ s41586-023-06185-3

    2023

  3. [3]

    Ewalds, Z

    R.Lam, A.Sanchez-Gonzalez, M.Willson, P.Wirnsberger, M.Fortunato, F.Alet, S.Ravuri, T. Ewalds, Z. Eaton-Rosen, W. Hu, A. Merose, S. Hoyer, G. Holland, O. Vinyals, J. Stott, A. Pritzel, S. Mohamed, and P. Battaglia. Learning skillful medium-range global weather forecasting.Science, 382:eadi2336, 2023. doi: 10.1126/science.adi2336. 17

    doi:10.1126/science.adi2336 2023

  4. [4]

    Price, A

    I. Price, A. Sanchez-Gonzalez, F. Alet, T. R. Andersson, A. El-Kadi, D. Masters, T. Ewalds, J. Stott, S. Mohamed, P. Battaglia, R. Lam, and M. Willson. Probabilistic weather forecast- ingwithmachinelearning.Nature, 637(8044):84–90, 2025. doi: 10.1038/s41586-024-08252-9

    doi:10.1038/s41586-024-08252-9 2025

  5. [5]

    Weatherbench 2: A benchmark for the next generation of data- driven global weather models.Journal of Advances in Modeling Earth Systems, 16(6): e2023MS004019, 2024

    Stephan Rasp, Stephan Hoyer, Alexander Merose, Ian Langmore, Peter Battaglia, Tyler Russell, Alvaro Sanchez-Gonzalez, Vivian Yang, Rob Carver, Shreya Agrawal, Matthew Chantry, Zied Ben Bouallegue, Peter Dueben, Carla Bromberg, Jared Sisk, Luke Barrington, Aaron Bell, and Fei Sha. Weatherbench 2: A benchmark for the next generation of data- driven global w...

    doi:10.1029/2023ms004019 2024

  6. [6]

    Nguyen, J

    T. Nguyen, J. Brandstetter, A. Kapoor, J. K. Gupta, and A. Grover. ClimaX: A foundation model for weather and climate. InProceedings of the 40th International Conference on Machine Learning, 2023

    2023

  7. [7]

    Bodnar, W

    C. Bodnar, W. P. Bruinsma, A. Lucic, M. Stanley, A. Allen, J. Brandstetter, P. Garvan, M. Riechert, J. A. Weyn, H. Dong, J. K. Gupta, K. Thambiratnam, A. T. Archibald, C.-C. Wu, E. Heider, M. Welling, R. E. Turner, and P. Perdikaris. A foundation model for the Earth system.Nature, 641(8065):1180–1187, 2025. doi: 10.1038/s41586-025-09005-y

    doi:10.1038/s41586-025-09005-y 2025

  8. [8]

    Earth system foundation model (esfm): A unified framework for heterogeneous data integration and forecasting, 2026

    Firat Ozdemir, Yun Cheng, Salman Mohebi, Fanny Lehmann, Simon Adamov, Zhenyi Zhang, Leonardo Trentini, Dana Grund, Oliver Fuhrer, Torsten Hoefler, Siddhartha Mishra, Sebastian Schemm, Benedikt Soja, and Mathieu Salzmann. Earth system foundation model (esfm): A unified framework for heterogeneous data integration and forecasting, 2026. URL https://arxiv.or...

    Pith/arXiv arXiv 2026

  9. [9]

    Prithvi wxc: Foundation model for weather and climate, 2024

    Johannes Schmude, Sujit Roy, Will Trojak, Johannes Jakubik, Daniel Salles Civitarese, Shraddha Singh, Julian Kuehnert, Kumar Ankur, Aman Gupta, Christopher E Phillips, Romeo Kienzler, Daniela Szwarcman, Vishal Gaur, Rajat Shinde, Rohit Lal, Arlindo Da Silva, Jorge Luis Guevara Diaz, Anne Jones, Simon Pfreundschuh, Amy Lin, Aditi She- shadri, Udaysankar Na...

    2024

  10. [10]

    Hersbach, B

    H. Hersbach, B. Bell, P. Berrisford, S. Hirahara, A. Horányi, J. Muñoz-Sabater, J. Nicolas, C. Peubey, R. Radu, D. Schepers, A. Simmons, C. Soci, S. Abdalla, X. Abellan, G. Bal- samo, P. Bechtold, G. Biavati, J. Bidlot, M. Bonavita, G. De Chiara, P. Dahlgren, D. Dee, M. Diamantakis, R. Dragani, J. Flemming, R. Forbes, M. Fuentes, A. Geer, L. Haimberger, S...

    1999

  11. [11]

    Vaughan, S

    A. Vaughan, S. Markou, W. Tebbutt, J. Requeima, W. P. Bruinsma, T. R. Andersson, M. Herzog, N. D. Lane, M. Chantry, J. S. Hosking, and R. E. Turner. End-to-end data- driven weather prediction.Nature, 641:1172–1179, 2025. doi: 10.1038/s41586-025-08897-0

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

  12. [12]

    Bevis, S

    M. Bevis, S. Businger, T. A. Herring, R. A. Anthes, and R. H. Ware. GPS meteorology: Remote sensing of atmospheric water vapor using the global positioning system.Journal of Geophysical Research: Atmospheres, 97(D14):15787–15801, 1992. doi: 10.1029/92JD01517

    work page doi:10.1029/92jd01517 1992

  13. [13]

    Bevis, S

    M. Bevis, S. Businger, S. Chiswell, T. A. Herring, R. A. Anthes, C. Rocken, and R. H. Ware. GPS meteorology: Mapping zenith wet delays onto precipitable water.Journal of Applied Meteorology, 33(3):379–386, 1994. doi: 10.1175/1520-0450(1994)033<0379:GMMZWD>2. 0.CO;2

    doi:10.1175/1520-0450(1994)033 1994

  14. [14]

    Systematic errors in global radiosonde precipitable water data from comparisons with ground-based gps measurements.Journal of Climate, 21: 2218–2238, 05 2008

    Junhong Wang and Liangying Zhang. Systematic errors in global radiosonde precipitable water data from comparisons with ground-based gps measurements.Journal of Climate, 21: 2218–2238, 05 2008. doi: 10.1175/2007JCLI1944.1

    work page doi:10.1175/2007jcli1944.1 2008

  15. [15]

    P. Yuan, G. Blewitt, C. Kreemer, W. C. Hammond, D. Argus, X. Yin, R. Van Malderen, M. Mayer, W. Jiang, J. Awange, and H. Kutterer. An enhanced integrated water vapour dataset from more than 10000 global ground-based GPS stations in 2020.Earth System Science Data, 15(2):723–743, 2023. doi: 10.5194/essd-15-723-2023

    work page doi:10.5194/essd-15-723-2023 2020

  16. [16]

    Blewitt, W

    G. Blewitt, W. C. Hammond, and C. Kreemer. Harnessing the GPS data explosion for interdisciplinary science.Eos, 99, 2018. doi: 10.1029/2018EO104623

    doi:10.1029/2018eo104623 2018

  17. [17]

    Vedel and X.-Y

    H. Vedel and X.-Y. Huang. Impact of ground based gps data on numerical weather pre- diction.Journal of the Meteorological Society of Japan. Ser. II, 82(1B):459–472, 2004. doi: 10.2151/jmsj.2004.459

    work page doi:10.2151/jmsj.2004.459 2004

  18. [18]

    Guerova, J

    G. Guerova, J. Jones, J. Douša, G. Dick, S. de Haan, E. Pottiaux, O. Bock, R. Pacione, G. Elgered, H. Vedel, and M. Bender. Review of the state of the art and future prospects of the ground-based GNSS meteorology in Europe.Atmospheric Measurement Techniques, 9(11):5385–5406, 2016. doi: 10.5194/amt-9-5385-2016

    work page doi:10.5194/amt-9-5385-2016 2016

  19. [19]

    Jones, G

    J. Jones, G. Guerova, J. Douša, G. Dick, S. de Haan, E. Pottiaux, and R. van Malderen. Advanced GNSS Tropospheric Products for Monitoring Severe Weather Events and Climate. Springer, 2019. doi: 10.1007/978-3-030-13901-8. COST Action ES1206 Final Action Dis- semination Report

    work page doi:10.1007/978-3-030-13901-8 2019

  20. [20]

    Giannaros, V

    C. Giannaros, V. Kotroni, K. Lagouvardos, T. M. Giannaros, and C. Pikridas. Assessing the impact of GNSS ZTD data assimilation into the WRF modeling system during high-impact rainfall events over Greece.Remote Sensing, 12(3):383, 2020. doi: 10.3390/rs12030383

    work page doi:10.3390/rs12030383 2020

  21. [21]

    Aichinger-Rosenberger, E

    M. Aichinger-Rosenberger, E. Brockmann, L. Crocetti, B. Soja, and G. Moeller. Machine learning-based prediction of Alpine foehn events using GNSS troposphere products: first results for Altdorf, Switzerland.Atmospheric Measurement Techniques, 15(19):5821–5839,

  22. [22]

    Modelling the Troposphere with Global Navigation Satellite Systems, Meteorologi- cal Data and Machine Learning

    Laura Crocetti, Matthias Schartner, Konrad Schindler, Rochelle Schneider, and Benedikt Soja. Modelling the Troposphere with Global Navigation Satellite Systems, Meteorologi- cal Data and Machine Learning. InIGARSS 2024 - 2024 IEEE International Geoscience and Remote Sensing Symposium, pages 1689–1692, 2024. doi: 10.1109/IGARSS53475.2024. 10640441

    arXiv 2024

  23. [23]

    Aichinger-Rosenberger, M

    M. Aichinger-Rosenberger, M. Aregger, J. Kopp, and B. Soja. Detecting signatures of convective storm events in GNSS-SNR: Two case studies from summer 2021 in Switzerland. Geophysical Research Letters, 50(21):e2023GL104916, 2023. doi: 10.1029/2023GL104916

    work page doi:10.1029/2023gl104916 2021

  24. [24]

    M. Zhu, Z. Liu, and W. Hu. Observing water vapor variability during three super ty- phoon events in Hong Kong based on GPS water vapor tomographic modeling tech- nique.Journal of Geophysical Research: Atmospheres, 125(16):e2019JD032318, 2020. doi: 10.1029/2019JD032318

    work page doi:10.1029/2019jd032318 2020

  25. [25]

    Q. Li, J. Böhm, L. Yuan, and R. Weber. Global zenith wet delay modeling with surface meteorological data and machine learning.GPS Solutions, 28:57, 2024. doi: 10.1007/ s10291-023-01595-2

    2024

  26. [26]

    Zhang, Y

    H. Zhang, Y. Yao, C. Xu, W. Xu, and J. Shi. Transformer-based global zenith tropospheric delay forecasting model.Remote Sensing, 14(14):3335, 2022. doi: 10.3390/rs14143335

    work page doi:10.3390/rs14143335 2022

  27. [27]

    J. Ding, X. Mi, W. Chen, J. Chen, J. Wang, Y. Zhang, J. L. Awange, B. Soja, L. Bai, Y. Deng, and W. Tang. Forecasting of tropospheric delay using AI foundation models in support of microwave remote sensing.IEEE Transactions on Geoscience and Remote Sensing, 62:1–19, 2024. doi: 10.1109/TGRS.2024.3488727

    arXiv 2024

  28. [28]

    Olivetti and G

    L. Olivetti and G. Messori. Advances and prospects of deep learning for medium-range extreme weather forecasting.Geoscientific Model Development, 17(6):2347–2358, 2024. doi: 10.5194/gmd-17-2347-2024

    work page doi:10.5194/gmd-17-2347-2024 2024

  29. [29]

    Lavers, Adrian Simmons, Freja Vamborg, and Mark J

    David A. Lavers, Adrian Simmons, Freja Vamborg, and Mark J. Rodwell. An evaluation of era5 precipitation for climate monitoring.Quarterly Journal of the Royal Meteorological Society, 148(748):3152–3165, 2022. doi: 10.1002/qj.4351

    doi:10.1002/qj.4351 2022

  30. [30]

    O. C. Pasche, J. Wider, Z. Zhang, J. Zscheischler, and S. Engelke. Validating deep learning weather forecast models on recent high-impact extreme events.Artificial Intelligence for the Earth Systems, 4(1):e240033, 2025. doi: 10.1175/AIES-D-24-0033.1

    work page doi:10.1175/aies-d-24-0033.1 2025

  31. [31]

    Physics-based models outperform ai weather forecasts of record-breaking extremes.Science Advances, 12 (18):eaec1433, 2026

    Zhongwei Zhang, Erich Fischer, Jakob Zscheischler, and Sebastian Engelke. Physics-based models outperform ai weather forecasts of record-breaking extremes.Science Advances, 12 (18):eaec1433, 2026. doi: 10.1126/sciadv.aec1433. URLhttps://www.science.org/doi/ abs/10.1126/sciadv.aec1433

    doi:10.1126/sciadv.aec1433 2026

  32. [32]

    J.Ding, W.Chen, J.Chen, J.Wang, Y.Zhang, L.Bai, Y.Wang, X.Mi, T.Liu, andD.Weng. Spatiotemporal inhomogeneity of accuracy degradation in AI weather forecast foundation 20 models: A GNSS perspective.International Journal of Applied Earth Observation and Geoinformation, 139:104473, 2025. doi: 10.1016/j.jag.2025.104473

    arXiv 2025

  33. [33]

    Crocetti, M

    L. Crocetti, M. Schartner, M. F. Wareyka-Glaner, K. Schindler, and B. Soja. ZWDX: a global zenith wet delay forecasting model using XGBoost.Earth, Planets and Space, 76(1): 163, 2024. doi: 10.1186/s40623-024-02104-6

    work page doi:10.1186/s40623-024-02104-6 2024

  34. [34]

    Aichinger-Rosenberger, A

    M. Aichinger-Rosenberger, A. Wolf, C. Senn, R. Hohensinn, M. F. Glaner, G. Moeller, B. Soja, and M. Rothacher. MPG-NET: A low-cost, multi-purpose GNSS co-location station network for environmental monitoring.Measurement, 216:112981, 2023. doi: 10.1016/j. measurement.2023.112981

    arXiv 2023

  35. [35]

    Marut, T

    G. Marut, T. Hadaś, J. Kapłon, E. Trzcina, and W. Rohm. Monitoring the water vapor content at high spatio-temporal resolution using a network of low-cost multi-GNSS receivers. IEEE Transactions on Geoscience and Remote Sensing, 60(1):1–14, 2022. doi: 10.1109/ TGRS.2022.3226631

    arXiv 2022

  36. [36]

    B. Soja, G. Kłopotek, Y. Pan, L. Crocetti, S. Mao, M. Awadaljeed, M. Rothacher, L. See, T. Sturn, R. Weinacker, I. McCallum, and V. Navarro. Machine learning-based exploita- tion of crowdsourced GNSS data for atmospheric studies. InIGARSS 2023 – 2023 IEEE International Geoscience and Remote Sensing Symposium, pages 1170–1173, 2023. doi: 10.1109/IGARSS5210...

    arXiv 2023

  37. [37]

    Eyring, S

    V. Eyring, S. Bony, G. A. Meehl, C. A. Senior, B. Stevens, R. J. Stouffer, and K. E. Taylor. Overview of the Coupled Model Intercomparison Project Phase 6 (CMIP6) experimental design and organization.Geoscientific Model Development, 9(5):1937–1958, 2016. doi: 10. 5194/gmd-9-1937-2016

    1937

  38. [38]

    Global, spatially explicit mod- elling of zenith wet delay with XGBoost.Journal of Geodesy, 98(4), 2024

    Laura Crocetti, Matthias Schartner, Florian Zus, Wenyuan Zhang, Gregor Moeller, Vicente Navarro, Linda See, Konrad Schindler, and Benedikt Soja. Global, spatially explicit mod- elling of zenith wet delay with XGBoost.Journal of Geodesy, 98(4), 2024. ISSN 1432-1394. doi: 10.1007/s00190-024-01829-2

    doi:10.1007/s00190-024-01829-2 2024

  39. [39]

    H. E. Beck, E. F. Wood, M. Pan, C. K. Fisher, D. G. Miralles, A. I. J. M. van Dijk, T. R. McVicar, and R. F. Adler. MSWEP V2 global 3-hourly 0.1◦ precipitation: Methodology and quantitative assessment.Bulletin of the American Meteorological Society, 100(3):473–500,

  40. [40]

    Fanny Lehmann, Firat Ozdemir, Benedikt Soja, Torsten Hoefler, Siddhartha Mishra, and Sebastian Schemm. Fine-tuning a weather foundation model with lightweight decoders for unseen physical processes.Journal of Geophysical Research: Machine Learning and Computation, 3(4):e2025JH000837, 2026. doi: 10.1029/2025JH000837

    work page doi:10.1029/2025jh000837 2026

  41. [41]

    N. M. Roberts and H. W. Lean. Scale-selective verification of rainfall accumulations from high-resolution forecasts of convective events.Monthly Weather Review, 136(1):78–97, 2008. doi: 10.1175/2007MWR2123.1. 21

    doi:10.1175/2007mwr2123.1 2008

  42. [42]

    Schaefer

    Joseph T. Schaefer. The critical success index as an indicator of warning skill.Weather and Forecasting, 5(4):570–575, 1990. doi: 10.1175/1520-0434(1990)005<0570:TCSIAA>2.0.CO; 2

    work page doi:10.1175/1520-0434(1990)005 1990

  43. [43]

    main text

    A. H. Gray and J. D. Markel. Distance measures for speech processing.IEEE Transactions on Acoustics, Speech, and Signal Processing, 24(5):380–391, 1976. doi: 10.1109/TASSP. 1976.1162849. 22 Supplementary Information Integrating GNSS-Derived Zenith Wet Delay into a Weather Foundation Model Improves Precipitation Forecasting Contents •Supplementary Note 1 -...

    arXiv 1976

  44. [2019]

    doi: 10.1175/BAMS-D-17-0138.1

    work page doi:10.1175/bams-d-17-0138.1

  45. [2022]

    doi: 10.5194/amt-15-5821-2022. 19

    work page doi:10.5194/amt-15-5821-2022 2022

This paper was first reviewed by deepseek-v4-flash on August 2, 2026.