Pith. sign in

REVIEW 4 major objections 5 minor 7 cited by

GraphDOP: Towards skilful data-driven medium-range weather forecasts learnt and initialised directly from observations

T0 review · 4 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read Training a graph neural network exclusively on Earth System observations, without reanalysis inputs, yields medium-range forecasts that stay skilful to day five and beat the operational physics-based system on tropical two-metre…

desk verdict A genuine first for observation-only data-driven forecasting, but the abstract overclaims 'exclusively from observations' given ERA5-based QC, and the headline IFS comparison is confounded by a five-hour initialization advantage the paper itself admits. read the letter →

arxiv 2412.15687 v1 pith:3LG6JGCC submitted 2024-12-20 physics.ao-ph cs.LG

Mihai Alexe , Eulalie Boucher , Peter Lean , Ewan Pinnington , Patrick Laloyaux , Anthony McNally , Simon Lang , Matthew Chantry
show 6 more authors
Chris Burrows Marcin Chrust Florian Pinault Ethel Villeneuve Niels Bormann Sean Healy
This is my paper · ORCID
classification physics.ao-phcs.LG
keywords graphneuralnetworkdata-drivenweatherforecastingobservation-spacelearningmedium-rangeforecastsatellitebrightnesstemperaturesdirectobservationpredictiontwo-metretemperaturelatentstaterepresentation
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper introduces GraphDOP, a forecast system that is trained and initialised exclusively from Earth System observations, with no gridded reanalysis fields used as inputs or training targets. The authors claim that a graph neural network can learn the correlations between satellite radiances and conventional in-situ measurements well enough to produce skilful global forecasts of surface and upper-air weather out to five days. In their evaluation, GraphDOP's two-metre temperature forecasts are competitive with those of the operational physics-based system, with smaller departures from verifying observations over the Tropics at day five. A sympathetic reading of the paper is that it establishes observation-only learning as a viable path toward medium-range weather prediction, even if full parity with established systems remains an open goal.

What carries the argument

The central machinery is an encoder–processor–decoder graph neural network whose encoder and decoder are built on the fly from the latitudes and longitudes of observations, projecting irregular observations onto and off a latent state of 40,320 nodes on a reduced Gaussian grid of roughly one-degree spacing. The processor is a transformer with windowed attention that advances the latent state through the forecast window, and autoregressive rollout extends the forecast to longer lead times. Because the decoder only needs observation metadata to produce a forecast, predictions can be issued at arbitrary locations and times. The training objective is a weighted mean squared error over all observation targets, $\mathcal{L}_{\mathrm{DOP}} = \tfrac{1}{T |I| |C| |O|} \sum_{t,i,c,o} w_i w_{c,i} (y_{tico}-\hat{y}_{tico})^2$, which balances per-instrument and per-channel weights so that the network learns to predict every observed quantity jointly. A quarter of satellite observations and half of the conventional observations are dropped during training to prevent overfitting.

What would settle it

Retrain GraphDOP with the reanalysis-based quality-control filter disabled and evaluate its day-five two-metre temperature departures over the Tropics against independent in-situ observations that never entered any filter; if the advantage over the physics-based system disappears or reverses, the reanalysis dependence is carrying the skill.

Watch

Extended reading notes

Core claim

The paper's central claim is that an end-to-end graph neural network, trained and initialised directly from conventional observations and level-1 satellite brightness temperatures with no physics-based reanalysis inputs, can form a coherent latent representation of the Earth system and produce skilful medium-range forecasts. Concretely, GraphDOP forecasts surface and upper-air parameters up to five days ahead; its two-metre temperature forecasts are competitive with those of the operational physics-based forecast system, and it has smaller departures from verifying observations over the Tropics at day five. The network reproduces synoptic-scale features in brightness-temperature space, including moving frontal cloud bands, jet-stream structures, sea-ice growth during a rapid Arctic freezing event, and the trajectory and intensification of Hurricane Ian. The authors acknowledge that a reanalysis-based quality-control filter is applied to some conventional observations during dataset generation, and they stress that reanalysis fields are used only for verification, never for training or initialisation.

Load-bearing premise

The load-bearing premise is that filtering conventional observations by how well they agree with a reanalysis dataset does not materially shape what the model learns; if that filter is doing the work, the system is not truly trained from observations alone and its reanalysis-verified skill could be partly circular.

Editorial extensions

If this is right

  • If the claim holds, producing a global medium-range forecast no longer requires running a large data-assimilation system: the forecast is initialised directly from the latest observation windows and can be issued within minutes of data arrival.
  • A purely observation-driven forecast model can be verified and used on any grid or location, including regions with sparse or no conventional observations, because the decoder is not tied to fixed analysis fields.
  • The reported day-five tropical two-metre temperature skill suggests that data-driven systems may challenge physics-based systems in data-sparse regions first, not in well-observed mid-latitudes.
  • Because the model ingests cloudy and surface-sensitive radiances directly, it opens a path to exploiting observation types that traditional variational assimilation must reject or approximate with complex operators.
  • The fully differentiable model can be used to compute adjoint sensitivities, giving a new tool for estimating the information content of individual observation types for forecast skill.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The strongest caveat is the paper's own: a reanalysis-based quality-control filter removes gross outliers from conventional observations before training. If that filter is what teaches the model to agree with the reanalysis, the reported advantage might partly reflect the verification target rather than independent skill; retraining without the reanalysis-dependent QC is a direct test.
  • The model's ability to forecast at arbitrary locations suggests it could be repurposed as a learned observation operator or an observation-space prior inside a hybrid data-assimilation cycle, feeding conventional analysis systems with radiance-to-state relationships.
  • The deterministic WMSE objective visibly smooths forecasts at long lead times; switching to a probabilistic or diffusion-based objective would likely sharpen features and may change the day-five comparison.
  • The headline comparison covers one winter season (December 2022 to February 2023); seasonal and interannual robustness of the tropical skill is untested, and the authors state that more work is needed on diurnal, seasonal, and regional error dependence.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. GraphDOP is an end-to-end graph-neural-network weather forecast system that is trained and initialised from conventional and satellite observations rather than from gridded reanalysis fields. The paper describes the observation dataset and quality-control choices, the encoder-processor-decoder architecture on a latent O96 mesh, the weighted MSE objective, and qualitative case studies covering IASI radiance evolution, an Arctic sea-ice freezing event, and Hurricane Ian. Quantitative verification is performed in observation space against matched conventional and satellite observations, with operational IFS as a benchmark, and in grid space against ERA5 with persistence and climatology baselines. The headline claims are that the model produces skilful forecasts up to five days and that its two-metre temperature forecasts are competitive with, and over the Tropics at day five better than, the operational IFS.

Significance. If correct, this would be a substantial step toward a new forecasting paradigm: a data-driven system that learns atmospheric dynamics and observation operators jointly from observational time series, avoiding the cost and complexity of 4D-Var while still exploiting the full observing system. The paper is commendably explicit about several limitations, including the ERA5-based quality control of training data (§2), the early-delivery IFS comparison (§5.1), and the gap between grid-space and observation-space t2m skill (§5.2). The matched-observation verification and the inclusion of persistence and climatology baselines are strengths. However, the two central claims — 'exclusively from observations' and 'competitive with IFS at day five' — are currently weakened by a known but unquantified 5-hour observation-window advantage in the IFS comparison and by a training-data quality-control dependence on ERA5. These concerns are addressable with additional experiments and more cautious claims, so the paper merits a major revision rather than rejection.

major comments (4)
  1. [Abstract and §2] The abstract claims that GraphDOP is 'trained and initialised exclusively from Earth System observations, with no physics-based (re)analysis inputs or feedbacks', but §2 states that conventional observations are filtered with 'a conservative ERA5 departure (observation minus forecast) QC check' and that this 'introduces a partial dependence of the dataset generation pipeline on the reanalysis system'. Because the conventional observations are also the targets for t2m and other headline verification variables, this is not a cosmetic caveat: if the filter preferentially removes observations that disagree with ERA5, the strict observation-only claim is weakened and the ERA5-based grid-space verification in §5.2 becomes partially circular. Please either remove or substantially soften the exclusivity claim, quantify how many conventional observations are removed by the QC check and show that the retained t2m distribution is not materially altered, or retrain without the ERA5-based filter.
  2. [§5.1 and Figures 7–9] The day-5 tropical t2m and AMSU-A channel-5 advantages over IFS are confounded by the stated ~5-hour advantage of GraphDOP relative to the IFS early-delivery analysis. Section 5.1 explains that GraphDOP uses a full 12-hour window (09z-21z or 21z-09z) while the early-delivery IFS forecasts are initialised from shorter windows (09z-16z or 21z-04z), so at a fixed valid time the two systems are not compared at equal effective lead times. For a surface variable with short memory such as t2m, an extra five hours of observations can plausibly explain part or all of the reported day-1 15% improvement and the tropical day-5 advantage. The paper provides no sensitivity test or correction for this mismatch; please add an experiment in which GraphDOP is initialised from the same truncated window as IFS early delivery, or in which the comparison is made against IFS forecasts from the late-delivery analysis, before the 'competitive with IFS' claim is made.
  3. [§5.2 and Figure 11] The grid-space verification shows a rapid growth of t2m RMSE from day 3 onward, yet the paper's headline t2m claim is based on the observation-space departures of §5.1, which are 'much lower and in line with results presented in Section 5.1'. The text says this discrepancy is 'currently being investigated', but the t2m result is the flagship quantitative claim of the paper. Please provide a quantitative explanation of the discrepancy (for example, station-elevation representativeness, diurnal sampling, or the difference between point observations and grid-cell averages), or explicitly state that the claimed IFS competitiveness applies only to station-location verification and not to gridded forecasts.
  4. [§5.1–5.2 and Figures 7–11] All quantitative verification is computed over one boreal winter (December 10, 2022 to February 28, 2023) for the observation-space comparisons and one month (January 2023) for the grid-space comparisons, with no confidence intervals or seasonal breakdown. The tropical day-5 improvements shown in Figures 7 and 9 could be specific to this season or to the particular observing-system configuration of the period. Please add uncertainty estimates, a second independent evaluation period, or a clear statement of the seasonal/regime dependence of the reported skill differences.
minor comments (5)
  1. [Title page] The title appears as 'GRAPH DOP: T OWARDS SKILFUL...' in the preprint text; please correct the spacing and capitalization.
  2. [References] The citation to Lessig (2025) as 'Manuscript in preparation' should be updated to a citable preprint if one becomes available, or removed from the reference list.
  3. [§3, Eq. (1)] Please clarify the summation convention in Equation (1): the normalization by |I| × |C| × |O| is written as a product after the sums, and it is not immediately clear how the sets I, C, and O are nested; a worked example for one instrument would help.
  4. [Appendix, Figure 15] Figure 15 is introduced as a sensitivity result from a network trained with a 3-hour decoder output interval, but this alternative model is not described elsewhere; please provide its training setup or clearly label it as an illustrative preliminary result.
  5. [§3] The training section states that 70,000 steps were run on 64 H100 GPUs, but it does not give the total compute cost or wall-clock time; adding this information would aid reproducibility and resource planning.

Circularity Check

1 steps flagged · score 4.0 of 10

Partial circularity in the 'observations-only' claim via ERA5-based training-data QC; forecast skill itself is not reduced to its inputs, and the IFS t2m comparison carries a lead-time confound.

  1. other [Abstract and Section 2 (Datasets, quality control paragraph)]
    "Abstract: 'trained and initialised exclusively from Earth System observations, with no physics-based (re)analysis inputs or feedbacks.' Section 2: 'A conservative ERA5 departure (observation minus forecast) QC check helps remove gross outliers for certain conventional observation types. While it is recognised that this introduces a partial dependence of the dataset generation pipeline on the reanalysis system, it is expected that this can be removed in a future iteration of the method.'"

    The paper asserts in Section 5.2 that 'ERA5 fields are employed exclusively for verification, and were not used during the training of the network.' Section 2, however, says ERA5 departures are used to quality-control the conventional observations before they enter training. The same observation types are later used as the observation-space verification target, and the same reanalysis is the grid-space reference. The training input is therefore partly selected by the reference standard, so agreement with that standard is not fully independent. This is not a full reduction: the model never sees ERA5 fields as forecast inputs, the QC is described as conservative, and the day-5 comparisons are computed against independent observations.

full rationale

The forecast-skill derivation itself is mostly self-contained: the loss (Eq. 1) is a direct MSE against observations, training uses 2004-2021 data, and verification (Section 5.1, Figures 7-10) is computed on matched observations from Dec 2022-Feb 2023 with no gridded ERA5 fields as direct model inputs. The IFS comparison is a direct departure comparison, not an output of the training loss. The acknowledged ERA5-based QC of conventional training observations is the one place where the central 'exclusively from observations / no reanalysis' claim partially reduces to its opposite, so it is flagged as a circularity-relevant step. The separate 5-hour lead-time advantage over early-delivery IFS forecasts (Section 5.1: 'the GraphDOP forecasts have an approximately 5-hour advantage') is a real fairness confound for the headline t2m comparison, but it is a lead-time mismatch rather than a circularity. The observation-space vs grid-space t2m discrepancy (Section 5.2) is an unexplained accuracy issue, and Section 7 itself notes it 'remains to be determined whether' an observation-only system can compete with IFS. No load-bearing external uniqueness theorem or self-citation chain is invoked; AIFS citations are architectural only. Score 4 rather than 0-2 because the purity claim is partially definitional, but the central day-5 skill numbers are not forced by construction.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

No new physical entities are postulated. The latent atmospheric state on the O96 mesh is a model-internal representation, not a claim about the world. The main ledger items are empirically chosen loss weights, the 12-hour observability assumption, and the ERA5-based QC assumption, which is the most fragile entry.

free parameters (4)
  • Observation loss weights wc,i = not reported numerically
    In Eq. (1), weights wc,i for each channel and instrument are chosen empirically to balance loss contributions. They directly shape the trained forecast model.
  • Per-satellite or conventional observation weights wi = not reported numerically
    Also in Eq. (1), these can assign higher importance to selected observation targets during training; a free design choice not derived from first principles.
  • Observation dropout rates = 25% of satellite observations, 50% of conventional observations
    Random dropout during training is a hand-chosen regularization strategy described in Section 3, used to prevent overfitting.
  • Latent mesh O96 = 40320 nodes, approximately 1 degree resolution
    The latent processor grid is a modeling choice that fixes the spatial resolution of the learned state representation; not fitted to data but load-bearing for forecast detail.
assumptions (4)
  • domain assumption A 12-hour window of Level-1 and Level-2 observations contains sufficient information to infer the global atmospheric state for medium-range forecasting.
    The entire encoder-processor-decoder design and the 12-hour input windows in Section 3 rely on this identifiability assumption. If the observation network is too sparse to constrain the state, five-day skill would not be possible.
  • ad hoc to paper The ERA5-based quality control filter removes only gross outliers and does not materially bias the training distribution toward ERA5.
    Section 2 admits the QC step introduces a partial dependence on the reanalysis system. The claim that this does not undermine the observation-only nature of the model is an unproven assumption.
  • domain assumption A GNN and transformer trained with the weighted MSE objective generalizes to unseen observation layouts, new verification periods, and arbitrary target locations.
    This is the standard empirical generalization assumption in machine learning, supported here only by the short validation period (2022, January 2023) and the case studies.
  • domain assumption Observation errors are sufficiently stationary over 2004-2021 that the network can learn one set of relationships without explicitly modeling instrument drift or changing satellite constellations.
    The dataset spans satellite commissioning and decommissioning, but the model treats observation targets as exchangeable. The paper does not analyze the impact of non-stationarity on forecast skill.

how reviews work

0 comments
Cite this review

Pith. "Pith review of GraphDOP: Towards skilful data-driven medium-range weather forecasts learnt and initialised directly from observations." pith.science (2026). https://pith.science/paper/3LG6JGCC

@misc{pith2026241215687,
  author       = {Pith},
  title        = {Pith review of: GraphDOP: Towards skilful data-driven medium-range weather forecasts learnt and initialised directly from observations},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3LG6JGCC}},
  note         = {Machine review of arXiv:2412.15687}
}
read the original abstract

We introduce GraphDOP, a new data-driven, end-to-end forecast system developed at the European Centre for Medium-Range Weather Forecasts (ECMWF) that is trained and initialised exclusively from Earth System observations, with no physics-based (re)analysis inputs or feedbacks. GraphDOP learns the correlations between observed quantities - such as brightness temperatures from polar orbiters and geostationary satellites - and geophysical quantities of interest (that are measured by conventional observations), to form a coherent latent representation of Earth System state dynamics and physical processes, and is capable of producing skilful predictions of relevant weather parameters up to five days into the future.

Figures

Figures reproduced from arXiv: 2412.15687 by the authors.

Figure 1
Figure 1. Examples of data coverage from different observation types in a 12 hour window starting at 21 UTC on [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. A summary and timeline of the observation types currently included in the training dataset. This comprises [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. A schematic representation of the GraphDOP model. In this illustration, the model receives two 12-hour [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (12 more)
Figure 4
Figure 4. Figure 4: IASI channel 921 (wavenumber 875.0 cm−1 ) brightness temperatures (K): forecasted (left), observed (middle), and difference (observed minus forecast; right). We show 12-hour samples, starting from forecast day 1 (Jan 7, 2023; top row) through to day 4 (Jan 11, 2023; bo…
Figure 5
Figure 5. Figure 5: Gridded forecasts at a lead time of 24 hours, valid on Jan 15, 2023, 12z (right) compared to the ERA5 [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: Observations of wind speed at 200 hPa used as input to the network (left) and the gridded 200 hPa wind speed [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: Normalised RMS difference between IFS (CY47R3) and GraphDOP forecast departures for 2-meter tempera [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]
Figure 8
Figure 8. Figure 8: RMS forecast departures (left panel) and mean forecast departures (right panel) with respect to AMSU-A [PITH_FULL_IMAGE:figures/full_fig_p010_8.png]
Figure 9
Figure 9. Figure 9: Normalised RMS difference between IFS (CY47R3) and GraphDOP forecast departures for AMSU-A [PITH_FULL_IMAGE:figures/full_fig_p010_9.png]
Figure 10
Figure 10. Figure 10: RMS forecast departures (left panel) and mean forecast departures (right panel) for SSMIS all-sky (DMSP [PITH_FULL_IMAGE:figures/full_fig_p011_10.png]
Figure 11
Figure 11. Figure 11: RMSE of gridded global GraphDOP forecasts, evaluated against ERA5 reanalysis, as a function of lead [PITH_FULL_IMAGE:figures/full_fig_p012_11.png]
Figure 12
Figure 12. Figure 12: AMSR-2 channel 5 (10v) brightness temperatures (K): forecasted (left) observed (middle) and difference [PITH_FULL_IMAGE:figures/full_fig_p013_12.png]
Figure 13
Figure 13. Figure 13: ERA5 reanalysis (top) and GraphDOP gridded forecasts (bottom) mean sea-level pressure (a), wind speed [PITH_FULL_IMAGE:figures/full_fig_p015_13.png]
Figure 14
Figure 14. Figure 14: Gridded five-day forecasts (Jan 20, 2023, 12z; right) compared to the ERA5 reanalysis field (middle). [PITH_FULL_IMAGE:figures/full_fig_p018_14.png]
Figure 15
Figure 15. Figure 15: SEVIRI water-vapour 6.2 channel 5 brightness temperatures (K): forecasted (left), observed (middle), and [PITH_FULL_IMAGE:figures/full_fig_p019_15.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 7 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Timestep-Conditioned Transformers for Global Weather Forecasting

    cs.LG 2026-08 conditional novelty 7.0 of 10

    A single weather transformer with inference-time timestep conditioning matches specialist models across 1 to 24 hour steps and improves rollout stability under mixed-timestep training.

  2. Global reanalysis from observations alone with machine learning

    physics.ao-ph 2026-07 conditional novelty 7.0 of 10

    Observation-only machine learning can generate multi-decade global atmospheric reanalyses with large-scale skill near ERA5 and surface errors between ERA-Interim and ERA5, in a single day of compute.

  3. Huracan: A skillful end-to-end data-driven system for ensemble data assimilation and weather prediction

    physics.ao-ph 2025-08 conditional novelty 7.0 of 10

    A fully observation-driven ensemble weather forecasting system, Huracan, reports CRPS skill comparable to ECMWF ENS on 75.4% of variable and lead-time combinations.

  4. Earth observation embeddings are effective sub-grid descriptors for probabilistic weather downscaling

    cs.LG 2026-08 conditional novelty 6.0 of 10

    A 16-dimensional descriptor compressed from TESSERA satellite embeddings improves probabilistic weather downscaling at stations never seen in training, with the largest gains for wind speed.

  5. Multimodal Spatiotemporal Atmospheric Data Assimilation with Latent Video Flow-matching

    cs.LG 2026-08 conditional novelty 6.0 of 10

    A single latent video flow-matching prior, with posterior guidance, performs super-resolution, multimodal data fusion, filtering/smoothing, and observation-to-forecast for the global atmosphere using real station obse...

  6. OCELOT: Direct Atmospheric Forecasting from Heterogeneous Earth Observations Using a Graph-Transformer Hybrid Model

    physics.ao-ph 2026-07 conditional novelty 6.0 of 10

    OCELOT forecasts weather observations directly from raw satellite and in-situ data, reaching 12-hour skill below GFS but above persistence, without any reanalysis training data.

  7. Learning from nature: insights into GraphDOP's representations of the Earth System

    physics.ao-ph 2025-08 conditional novelty 6.0 of 10

    GraphDOP, trained purely on meteorological observations, spontaneously forms unified latent representations of Earth system state, viewing effects, and synoptic dynamics.

Reference graph

Works this paper leans on

63 extracted references · 34 canonical work pages · cited by 7 Pith papers

  1. [1]

    Pathak, S

    J. Pathak, S. Subramanian, P. Harrington, S. Raja, A. Chattopadhyay, M. Mardani, T. Kurth, D. Hall, Z. Li, K. Azizzadenesheli, and P. Hassanzadeh. F our C ast N et: A global data-driven high-resolution weather model using adaptive fourier neural operators. arXiv preprint arXiv:2202.11214, Feb 22 2022

    arXiv 2022

  2. [2]

    Learning skillful medium-range global weather forecasting

    Remi Lam, Alvaro Sanchez-Gonzalez, Matthew Willson, Peter Wirnsberger, Meire Fortunato, Ferran Alet, Suman Ravuri, Timo Ewalds, Zach Eaton-Rosen, Weihua Hu, Alexander Merose, Stephan Hoyer, George Holland, Oriol Vinyals, Jacklynn Stott, Alexander Pritzel, Shakir Mohamed, and Peter Battaglia. Learning skillful medium-range global weather forecasting. Scien...

    doi:10.1126/science.adi2336 2023

  3. [3]

    K. Bi, L. Xie, H. Zhang, et al. Accurate medium-range global weather forecasting with 3 D neural networks. Nature, 619: 0 533--538, 2023. doi:10.1038/s41586-023-06185-3

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

  4. [4]

    Bruinsma, Ana Lucic, Megan Stanley, Johannes Brandstetter, Patrick Garvan, Maik Riechert, Jonathan Weyn, Haiyu Dong, Anna Vaughan, Jayesh K

    Cristian Bodnar, Wessel P. Bruinsma, Ana Lucic, Megan Stanley, Johannes Brandstetter, Patrick Garvan, Maik Riechert, Jonathan Weyn, Haiyu Dong, Anna Vaughan, Jayesh K. Gupta, Kit Tambiratnam, Alex Archibald, Elizabeth Heider, Max Welling, Richard E. Turner, and Paris Perdikaris. Aurora: A foundation model of the atmosphere, 2024. URL https://arxiv.org/abs...

    arXiv 2024

  5. [5]

    Simon Lang, Mihai Alexe, Matthew Chantry, Jesper Dramsch, Florian Pinault, Baudouin Raoult, Mariana C. A. Clare, Christian Lessig, Michael Maier-Gerber, Linus Magnusson, Zied Ben Bouallègue, Ana Prieto Nemesio, Peter D. Dueben, Andrew Brown, Florian Pappenberger, and Florence Rabier. AIFS -- ECMWF 's data-driven forecasting system. arXiv preprint arXiv:24...

    arXiv 2024

  6. [6]

    Hersbach, B

    H. Hersbach, B. Bell, P. Berrisford, et al. The ERA5 global reanalysis. QJ R Meteorol Soc, 146: 0 1999--2049, 2020. doi:10.1002/qj.3803

    doi:10.1002/qj.3803 1999

  7. [7]

    Rabier, H

    F. Rabier, H. Järvinen, E. Klinker, J.-F. Mahfouf, and A. Simmons. The ECMWF operational implementation of four-dimensional variational assimilation. I : Experimental results with simplified physics. Quarterly Journal of the Royal Meteorological Society, 126 0 (564): 0 1143--1170, 2000. doi:10.1002/qj.49712656415

    doi:10.1002/qj.49712656415

  8. [8]

    IFS Documentation CY48R1 - P art II : Data Assimilation

    ECMWF. IFS Documentation CY48R1 - P art II : Data Assimilation . Number 2. ECMWF, 06/2023 2023. doi:10.21957/a744f32e74

    work page doi:10.21957/a744f32e74 2023

Show all 63 references
  1. [9]

    20 years of 4D-Var : B etter forecasts through a better use of observations

    ECMWF . 20 years of 4D-Var : B etter forecasts through a better use of observations. https://www.ecmwf.int/en/about/media-centre/news/2017/20-years-4d-var-better-forecasts-through--better-use-observations, 2017. Accessed: Dec 9, 2024

    2017

  2. [10]

    Data driven weather forecasts trained and initialised directly from observations, 2024 a

    Anthony McNally, Christian Lessig, Peter Lean, Eulalie Boucher, Mihai Alexe, Ewan Pinnington, Matthew Chantry, Simon Lang, Chris Burrows, Marcin Chrust, Florian Pinault, Ethel Villeneuve, Niels Bormann, and Sean Healy. Data driven weather forecasts trained and initialised dire...

    2024 arXiv

  3. [11]

    Bruinsma, Tom R

    Anna Vaughan, Stratis Markou, Will Tebbutt, James Requeima, Wessel P. Bruinsma, Tom R. Andersson, Michael Herzog, Nicholas D. Lane, Matthew Chantry, J. Scott Hosking, and Richard E. Turner. Aardvark weather: end-to-end data-driven weather forecasting, 2024. URL https://arxiv.o...

    2024 arXiv

  4. [12]

    Exploring the use of machine learning weather models in data assimilation, 2024

    Xiaoxu Tian, Daniel Holdaway, and Daryl Kleist. Exploring the use of machine learning weather models in data assimilation, 2024. URL https://arxiv.org/abs/2411.14677

    2024 arXiv

  5. [13]

    Exploring the integration of a global ai model with traditional data assimilation in weather forecasting

    Hongxiong Xu, Yihong Duan, and Xiangde Xu. Exploring the integration of a global ai model with traditional data assimilation in weather forecasting. Environmental Research Letters, 19 0 (12): 0 124079, nov 2024 a . doi:10.1088/1748-9326/ad93e8

    2024 doi

  6. [14]

    Score-based data assimilation, 2023

    François Rozet and Gilles Louppe. Score-based data assimilation, 2023. URL https://arxiv.org/abs/2306.10574

    2023 arXiv

  7. [15]

    Dueben, and Torsten Hoefler

    Langwen Huang, Lukas Gianinazzi, Yuejiang Yu, Peter D. Dueben, and Torsten Hoefler. Diffda: a diffusion model for weather-scale data assimilation, 2024. URL https://arxiv.org/abs/2401.05932

    2024 arXiv

  8. [16]

    FuXi-En4DVar : An assimilation system based on machine learning weather forecasting model ensuring physical constraints

    Yonghui Li, Wei Han, Hao Li, Wansuo Duan, Lei Chen, Xiaohui Zhong, Jincheng Wang, Yongzhu Liu, and Xiuyu Sun. FuXi-En4DVar : An assimilation system based on machine learning weather forecasting model ensuring physical constraints. G eophysical R esearch L etters , 51 0 (22): 0...

    2024 doi

  9. [17]

    Fuxi- DA : A generalized deep learning data assimilation framework for assimilating satellite observations, 2024 b

    Xiaoze Xu, Xiuyu Sun, Wei Han, Xiaohui Zhong, Lei Chen, and Hao Li. Fuxi- DA : A generalized deep learning data assimilation framework for assimilating satellite observations, 2024 b . URL https://arxiv.org/abs/2404.08522

    2024

  10. [18]

    F eng W u- 4DV ar: Coupling the data-driven weather forecasting model with 4D variational assimilation, 2024

    Yi Xiao, Lei Bai, Wei Xue, Kang Chen, Tao Han, and Wanli Ouyang. F eng W u- 4DV ar: Coupling the data-driven weather forecasting model with 4D variational assimilation, 2024. URL https://arxiv.org/abs/2312.12455

    2024 arXiv

  11. [19]

    David Neelin, Deliang Chen, Jie Feng, Wei Han, Libo Wu, and Yuan Qi

    Xiuyu Sun, Xiaohui Zhong, Xiaoze Xu, Yuanqing Huang, Hao Li, J. David Neelin, Deliang Chen, Jie Feng, Wei Han, Libo Wu, and Yuan Qi. FuXi W eather: A data-to-forecast machine learning system for global weather, 2024. URL https://arxiv.org/abs/2408.05472

    2024

  12. [20]

    Adaf: An artificial intelligence data assimilation framework for weather forecasting, 2024

    Yanfei Xiang, Weixin Jin, Haiyu Dong, Mingliang Bai, Zuliang Fang, Pengcheng Zhao, Hongyu Sun, Kit Thambiratnam, Qi Zhang, and Xiaomeng Huang. Adaf: An artificial intelligence data assimilation framework for weather forecasting, 2024. URL https://arxiv.org/abs/2411.16807

    2024 arXiv

  13. [21]

    Machine learning for precipitation nowcasting from radar images

    Shreya Agrawal, Luke Barrington, Carla Bromberg, John Burge, Cenk Gazen, and Jason Hickey. Machine learning for precipitation nowcasting from radar images. CoRR, abs/1912.12132, 2019. URL http://arxiv.org/abs/1912.12132

    1912 arXiv

  14. [22]

    Metnet: A neural weather model for precipitation forecasting

    Casper Kaae S nderby, Lasse Espeholt, Jonathan Heek, Mostafa Dehghani, Avital Oliver, Tim Salimans, Shreya Agrawal, Jason Hickey, and Nal Kalchbrenner. Metnet: A neural weather model for precipitation forecasting. CoRR, abs/2003.12140, 2020. URL https://arxiv.org/abs/2003.12140

    2003 arXiv

  15. [23]

    Skilful precipitation nowcasting using deep generative models of radar

    Suman Ravuri, Karel Lenc, Matthew Willson, Dmitry Kangin, Remi Lam, Piotr Mirowski, Megan Fitzsimons, Maria Athanassiadou, Sheleem Kashem, Sam Madge, Rachel Prudden, Amol Mandhane, Aidan Clark, Andrew Brock, Karen Simonyan, Raia Hadsell, Niall Robinson, Ellen Clancy, Alberto A...

    2021 doi

  16. [24]

    Deep learning for day forecasts from sparse observations, 2023

    Marcin Andrychowicz, Lasse Espeholt, Di Li, Samier Merchant, Alexander Merose, Fred Zyda, Shreya Agrawal, and Nal Kalchbrenner. Deep learning for day forecasts from sparse observations, 2023. URL https://arxiv.org/abs/2306.06079

    2023 arXiv

  17. [25]

    Jordan, and Jianmin Wang

    Yuchen Zhang, Mingsheng Long, Kaiyuan Chen, Lanxiang Xing, Ronghua Jin, Michael I. Jordan, and Jianmin Wang. Skilful nowcasting of extreme precipitation with nowcastnet. Nature, 619 0 (7970): 0 526--532, Jul 2023. ISSN 1476-4687. doi:10.1038/s41586-023-06184-4

    2023 doi

  18. [26]

    Red sky at night

    Tony McNally, Christian Lessig, Peter Lean, Matthew Chantry, Mihai Alexe, and Simon Lang. Red sky at night... P roducing weather forecasts directly from observations. ECMWF Newsletter No. 178 , 2024 b . doi:10.21957/1a8466ec2f

    2024 doi

  19. [27]

    Manuscript in preparation, 2025

    Christian Lessig. Manuscript in preparation, 2025

    2025

  20. [28]

    Dick P. Dee. Variational bias correction of radiance data in the ECMWF system. ECMWF Workshop on Assimilation of high spectral resolution sounders in NWP, 28 June - 1 July 2004, pages 97--112, 2004 2004

    2004

  21. [29]

    A tmospheric M otion V ectors: Past, present and future

    Mary Forsythe. A tmospheric M otion V ectors: Past, present and future. ECMWF Seminar on Recent development in the use of satellite observations in NWP , 2007

    2007

  22. [30]

    Fast graph representation learning with pytorch geometric

    Matthias Fey and Jan Eric Lenssen. Fast graph representation learning with pytorch geometric. CoRR, abs/1903.02428, 2019. URL http://arxiv.org/abs/1903.02428

    1903 arXiv

  23. [31]

    Forecasting global weather with graph neural networks

    Ryan Keisler. Forecasting global weather with graph neural networks. arXiv preprint arXiv:2202.07575, Feb 2022

    2022 arXiv

  24. [32]

    Mixed precision training

    Paulius Micikevicius, Sharan Narang, Jonah Alben, Gregory Diamos, Erich Elsen, David Garcia, Boris Ginsburg, Michael Houston, Oleksii Kuchaiev, Ganesh Venkatesh, and Hao Wu. Mixed precision training. arXiv preprint arXiv:1710.03740, 2018

    arXiv

  25. [33]

    N. P. Wedi. Increasing the horizontal resolution in numerical weather prediction and climate simulations: illusion or panacea? Philosophical Transactions of the Royal Society A, 372, 2014. doi:10.1098/rsta.2013.0289

    2014

  26. [34]

    The quiet revolution of numerical weather prediction

    Peter Bauer, Alan Thorpe, and Gilbert Brunet. The quiet revolution of numerical weather prediction. Nature, 525 0 (7567): 0 47–55, September 2015. ISSN 1476-4687. doi:10.1038/nature14956. URL http://dx.doi.org/10.1038/nature14956

    2015 doi

  27. [35]

    The role of satellite data in the forecasting of hurricane S andy

    Tony McNally, Massimo Bonavita, and Jean-Noël Thépaut. The role of satellite data in the forecasting of hurricane S andy. Monthly Weather Review, 142 0 (2): 0 634–646, January 2014. ISSN 1520-0493. doi:10.1175/mwr-d-13-00170.1

    2014 doi

  28. [36]

    Z. Ben Bouall\`egue, Rilwan Adewoyin, Mihai Alexe, Matthew Chantry, Mariana Clare, Jesper Dramsch, Sara Hahner, Simon Lang, Christian Lessig, Linus Magnusson, Michael Maier-Gerber, Gert Mertes, Gabriel Moldovan, Ana Prieto Nemesio, Cathal O’Brien, Florian Pinault, Baudouin Rao...

    2024

  29. [37]

    Distortion representation of forecast errors

    Ross Hoffman, Zheng Liu, Jean-Francois Louis, and Christopher Grassoti. Distortion representation of forecast errors. Monthly Weather Review, 123 0 (9): 0 2758–2770, 1995. ISSN 1520-0493. doi:10.1175/1520-0493(1995)123<2758:drofe>2.0.co;2

    1995 doi

  30. [38]

    Ebert, L

    E. Ebert, L. Wilson, A. Weigel, M. Mittermaier, P. Nurmi, P. Gill, M. Göber, S. Joslyn, B. Brown, T. Fowler, and A. Watkins. Progress and challenges in forecast verification. Meteorological Applications, 20 0 (2): 0 130--139, 2013. doi:https://doi.org/10.1002/met.1392

    2013 doi

  31. [39]

    Elucidating the design space of diffusion-based generative models

    Tero Karras, Miika Aittala, Timo Aila, and Samuli Laine. Elucidating the design space of diffusion-based generative models. arXiv preprint arXiv:2206.00364, 2022

    2022 arXiv

  32. [40]

    Data-driven ensemble forecasting with the AIFS

    Mihai Alexe, Simon Lang, Mariana Clare, Martin Leutbecher, Christopher Roberts, Linus Magnusson, Rilwan Adewoyin Matthew Chantry, Ana Prieto-Nemesio, Jesper Dramsch, Florian Pinault, and Baudouin Raoult. Data-driven ensemble forecasting with the AIFS . ECMWF Newsletter No. 181...

    2024 doi

  33. [41]

    Ensemble forecasting with the A rtificial I ntelligence F orecasting S ystem trained with a loss function based on the continuous ranked probability score

    Simon Lang, Mihai Alexe, Mariana Clare, Christopher Roberts, Rilwan Adewoyin, Zied Ben-Bouall\`egue, Matthew Chantry, Jesper Dramsch, Peter D\"ueben, Sara Hahner, Pedro Maciel, Ana Prieto-Nemesio, Cathal O'Brien, Florian Pinault, Jan Polster, Baudouin Raoult, Steffen Tietsche,...

    2024

  34. [42]

    Marco Matricardi and Anthony P. McNally. The direct assimilation of principal components of IASI spectra in the ECMWF 4D-Var . Quarterly Journal of the Royal Meteorological Society, 140 0 (679): 0 573--582, 2014. doi:https://doi.org/10.1002/qj.2156

    2014 doi

  35. [43]

    ECMWF forecast user guide

    Robert Owens and Tim Hewson. ECMWF forecast user guide. Technical report, ECMWF, Reading, 05/2018 2018

    2018

  36. [44]

    All-sky assimilation of SSMIS humidity sounding channels over land within the ECMWF system

    Fabrizio Baordo and Alan Geer. All-sky assimilation of SSMIS humidity sounding channels over land within the ECMWF system. Technical Report 38, 07/2015 2015

    2015

  37. [45]

    Assimilation of AMSU-A in all-sky conditions

    David Duncan, Niels Bormann, Alan Geer, and Peter Weston. Assimilation of AMSU-A in all-sky conditions. Technical report, ECMWF, 10/2021 2021

    2021

  38. [46]

    Evaluation of ECMWF forecasts

    Thomas Haiden, Martin Janousek, Frédéric Vitart, Maliko Tanguy, Fernando Prates, and Matthieu Chevalier. Evaluation of ECMWF forecasts. Technical Report 918, ECMWF, 09/2024 2024

    2024

  39. [47]

    Use of forecast departures in verification against observations

    Mohamed Dahoui, Gabor Radnoti, Sean Healy, Lars Isaksen, and Thomas Haiden. Use of forecast departures in verification against observations. ECMWF Newsletter 149, ECMWF, 2016

    2016

  40. [48]

    Improved two-metre temperature forecasts in the 2024 upgrade

    Bruce Ingleby, Gabriele Arduini, Gianpaolo Balsamo, Souhail Boussetta, Kenta Ochi, Ewan Pinnington, and Patricia de Rosnay. Improved two-metre temperature forecasts in the 2024 upgrade. ECMWF Newsletter 178, 2024

    2024

  41. [49]

    Continuous data assimilation for global numerical weather prediction

    Peter Lean, Elias Hólm, Massimo Bonavita, Niels Bormann, Anthony McNally, and Heikki Järvinen. Continuous data assimilation for global numerical weather prediction. Quarterly Journal of the Royal Meteorological Society, 147 0 (734): 0 273--288, 2021. doi:10.1002/qj.3917

    2021 doi

  42. [50]

    Machine learning for model error inference and correction

    Massimo Bonavita and Patrick Laloyaux. Machine learning for model error inference and correction. Journal of Advances in Modeling Earth Systems, 12 0 (12), 2020. doi:https://doi.org/10.1029/2020MS002232

    2020 doi

  43. [51]

    Online model error correction with neural networks: application to the integrated forecasting system, 2024

    Alban Farchi, Marcin Chrust, Marc Bocquet, and Massimo Bonavita. Online model error correction with neural networks: application to the integrated forecasting system, 2024. URL https://arxiv.org/abs/2403.03702

    2024 arXiv

  44. [52]

    Statistical modeling of 2-m temperature and 10-m wind speed forecast errors

    Zied Ben Bouallègue, Fenwick Cooper, Matthew Chantry, Peter Düben, Peter Bechtold, and Irina Sandu. Statistical modeling of 2-m temperature and 10-m wind speed forecast errors. Monthly Weather Review, 151 0 (4): 0 897 -- 911, 2023. doi:10.1175/MWR-D-22-0107.1

    2023 doi

  45. [53]

    Developing an all-surface capability for all-sky microwave radiances

    Alan Geer, Katrin Lonitz, David Duncan, and Niels Bormann. Developing an all-surface capability for all-sky microwave radiances. ECMWF Newsletter 171, 2022

    2022

  46. [54]

    Shepherd, I

    T.G. Shepherd, I. Polichtchouk, Robin Hogan, and A.J. Simmons. Report on stratosphere task force, 06/2018 2018

    2018

  47. [55]

    Laloyaux, M

    P. Laloyaux, M. Bonavita, M. Dahoui, J. Farnan, S. Healy, E. Hólm, and S. T. K. Lang. Towards an unbiased stratospheric analysis. Quarterly Journal of the Royal Meteorological Society, 146 0 (730): 0 2392--2409, 2020. doi:https://doi.org/10.1002/qj.3798

    2020 doi

  48. [56]

    Improving Ocean Surface Temperature for NWP using All-Sky Microwave Imager Observations , 08/2024 2024

    Tracy Scanlon, Alan Geer, Niels Bormann, and Philip Browne. Improving Ocean Surface Temperature for NWP using All-Sky Microwave Imager Observations , 08/2024 2024

    2024

  49. [57]

    Weatherbench 2: A benchmark for the next generation of data-driven global weather models, 2024

    Stephan Rasp, Stephan Hoyer, Alexander Merose, Ian Langmore, Peter Battaglia, Tyler Russel, 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 Sh...

    2024 arXiv

  50. [58]

    Scale‐dependent verification of ensemble forecasts

    Thomas Jung and Martin Leutbecher. Scale‐dependent verification of ensemble forecasts. Quarterly Journal of the Royal Meteorological Society, 134 0 (633): 0 973–984, 2008. ISSN 1477-870X. doi:10.1002/qj.255

    2008 doi

  51. [59]

    Mogensen, Tim Hewson, Sarah Keeley, and Linus Magnusson

    Kristian S. Mogensen, Tim Hewson, Sarah Keeley, and Linus Magnusson. Effects of ocean coupling on weather forecasts. ECMWF Newsletter 156, ECMWF, 2018

    2018

  52. [60]

    Coupled data assimilation at ECMWF : current status, challenges and future developments

    Patricia de Rosnay, Philip Browne, Eric de Boisséson, David Fairbairn, Yoichi Hirahara, Kenta Ochi, Dinand Schepers, Peter Weston, Hao Zuo, Magdalena Alonso-Balmaseda, Gianpaolo Balsamo, Massimo Bonavita, Niels Borman, Andy Brown, Marcin Chrust, Mohamed Dahoui, Giovanna Chiara...

    2022

  53. [61]

    Exploiting interface observations in a coupled reanalysis system

    Phil Browne, Patricia de Rosnay, Tony McNally, Sebastien Massart, Noureddine Semane, Sean Healy, Katerina Anesiadou, Alan Geer, and Tracy Scanlon. Exploiting interface observations in a coupled reanalysis system. In ECMWF A nnual S eminar , 2023. URL https://ecmwfevents.com/as...

    2023

  54. [62]

    Diamond, Carl J

    Howard J. Diamond, Carl J. Schreck, Adam Allgood, Emily J. Becker, Eric S. Blake, Francis G. Bringas, Suzana J. Camargo, Lin Chen, Caio A.S. Coelho, Nicolas Fauchereau, Chris Fogarty, Stanley B. Goldenberg, Gustavo Goni, Daniel S. Harnos, Qiong He, Zeng-Zhen Hu, Philip J. Klot...

    2022

  55. [63]

    P. P. Weston, W. Bell, and J. R. Eyre. Accounting for correlated error in the assimilation of high-resolution sounder data. Quarterly Journal of the Royal Meteorological Society, 140 0 (685): 0 2420--2429, 2014. doi:https://doi.org/10.1002/qj.2306

    2014 doi

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.