Pith. sign in

REVIEW 1 major objections 5 minor 54 references

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

T0 review · 1 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read Earth-observation surface embeddings improve probabilistic downscaling at unseen sites

desk verdict A well-controlled empirical paper that makes a first real contribution—using EO foundation embeddings as sub-grid descriptors for weather downscaling—with a static-embedding weakness that is real but manageable. read the letter →

arxiv 2608.12271 v1 pith:Q6RBORGJ submitted 2026-08-12 cs.LG physics.ao-ph

Pedro Sousa (1) , Will Tebbutt (2) , Sadiq Jaffer (1) , Robin Young (1) , Anil Madhavapeddy (1) , Richard E. Turner (2) ((1) Department of Computer Science , University of Cambridge , (2) Department of Engineering
show 1 more author
University of Cambridge)
This is my paper · ORCID
classification cs.LGphysics.ao-ph
keywords probabilisticdownscalingEarthobservationfoundationmodelsTESSERAembeddingsconvolutionalconditionalneuralprocessessub-gridsurfacedescriptorsCRPSoff-gridgeneralization2mtemperature
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 argues that a frozen, pre-trained Earth-observation foundation-model embedding, TESSERA, supplies exactly the sub-grid surface information that a coarse weather grid lacks. Compressing a 640-metre patch of TESSERA pixels into a 16-dimensional per-location descriptor and injecting it into a convolutional conditional neural process improves probabilistic downscaling at stations held out in both space and time. Across five climatically diverse regions, the embedding lowers CRPS by 11.5% for 2 m temperature and 6.2% for 10 m wind speed relative to a topography-only baseline, and the gain persists when the coarse input is switched from ERA5 reanalysis to Aurora forecast fields. The paper also shows that the two variables benefit differently: elevation already organises most of temperature's sub-grid structure, while the embedding supplies genuinely new, transferable surface information for wind speed. If right, this establishes long-timescale Earth-observation embeddings as a practical surface descriptor for site-specific probabilistic weather prediction, including at locations that have never hosted an instrument.

What carries the argument

The load-bearing mechanism is a two-stage compression-and-conditioning pipeline. First, a variational autoencoder compresses a roughly 640-metre by 640-metre patch of TESSERA embeddings, each a 128-dimensional self-supervised summary of a 10-metre pixel's annual surface dynamics, into a 16-dimensional latent vector per location. Second, this vector is concatenated with a three-feature topographic descriptor (elevation, elevation difference to the ERA5 orography, and multi-scale topographic position index) at the decoder of a convolutional conditional neural process, a neural process that maps a coarse grid to predictive distributions at arbitrary query points. Because the coarse grid carries no sub-grid information, the topographic and TESSERA descriptors are the only sources of local structure available to the model. The work this machinery does is to supply the land-cover, canopy, roughness, coastal, and built-environment information that elevation alone cannot express, and to force it through a bottleneck that keeps the downscaler from overfitting sparse station observations.

What would settle it

Compare CRPS at stations with documented land-cover change between 2017 and 2022: if the frozen 2017 embedding continues to deliver the same skill uplift at those stations as at unchanged stations, the claim that the gains derive from persistent surface properties would be contradicted, whereas a collapse of the uplift at changed stations would support the mechanism.

Watch

Extended reading notes

Core claim

The central claim, on the paper's own terms, is that a static land-surface embedding trained without any weather supervision carries transferable information about how a point departs from its coarse-grid atmospheric state, and that this information survives severe compression and remains useful for instantaneous, short-timescale predictions. The paper demonstrates this by conditioning a ConvCNP decoder on the embedding and measuring skill at 15% of stations never seen during training, on a test year never seen during training. The reported overall improvement is 11.5% CRPS for 2 m temperature and 6.2% for 10 m wind speed, with the gain present in all ten region-variable settings; controls show the benefit is not model capacity, degrades when descriptors are shuffled between stations, and is only partially reproduced by a 17-feature hand-crafted surface descriptor. A model-independent residual-regression probe confirms the paper's qualitative reading: for temperature, topography alone explains the persistent interpolation residual, while for wind speed the TESSERA descriptor organises it better than geography, topography, or static ERA5 surface fields. The paper also claims the embedding's advantage shows up operationally, surviving 72-hour Aurora forecast lead times and giving near-immediate wind-speed skill in a simulated deployment of a new Norwegian station network.

Load-bearing premise

The load-bearing premise is that the TESSERA embedding map from 2017 still describes the surface over the whole 2010-2020 training period and the 2022 test year; if land cover, vegetation, snow regime, or built environment changed materially at the evaluated stations, the descriptor is stale and the measured gains could be misattributed.

Editorial extensions

If this is right

  • Downscaling at stations withheld from both space and time improves in every one of the ten region-variable settings, with all-region CRPS gains of 11.5% for temperature and 6.2% for wind.
  • The embedding's contribution is transferable across input weather fields: the uplift persists when the coarse context changes from ERA5 reanalysis to Aurora forecasts at +6, +24, and +72 hour lead times.
  • For wind speed, TESSERA behaves as surrogate surface information: in a simulated Norwegian network ramp-up, it outperforms interpolated ERA5 before any local observations exist, and the no-TESSERA baseline does not match that cold-start error even after six years of local data.
  • For temperature, the embedding acts mainly as a land-surface prior for sparse networks; a richer 17-feature hand-crafted descriptor recovers only about a quarter to a third of the embedding's CRPS gain and never matches it in any region-variable-metric comparison.
  • The approach shows that two independently trained foundation models can be composed: a coarse atmospheric state from ERA5 or Aurora plus a static surface embedding gives better local probabilistic predictions than either alone.

Reading between the lines

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

  • A natural extension the paper leaves untested is whether simpler and cheaper surface summaries suffice: its own appendix shows that sixteen fixed order statistics of the TESSERA patch recover most of the benefit of the learned VAE descriptor, so a global system might not need a trained patch encoder at all.
  • The static-embedding assumption suggests a direct test the paper names but does not run: year-matched or time-indexed TESSERA embeddings should capture vegetation, snow, water, and urban change, and would be expected to outperform the 2017 map at stations where land cover changed materially.
  • Because the paper explicitly leaves joint spatial dependence out of the predictive distribution, combining the same surface descriptor with a latent-variable or autoregressive neural process should yield coherent multi-site forecasts such as wind-power portfolio risk; this is a direct but untested consequence of the mechanism.
  • The variable split found here points to a transfer rule for other EO-conditioned downscalers: variables whose unresolved structure is governed by roughness and land cover, such as wind and surface fluxes, should benefit most from embedding descriptors, whereas strongly elevation-controlled variables like temperature will show smaller gains except where station data are sparse.
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

1 major / 5 minor

Summary. This paper proposes augmenting a ConvCNP-based probabilistic downscaling model with a learned surface descriptor derived from TESSERA Earth-observation embeddings. A VAE compresses a roughly 640 m patch of 128-dimensional TESSERA pixel embeddings into a 16-dimensional vector, which is concatenated with topographic features in the decoder. The model is trained on ERA5 coarse fields and GHCNh station observations for 2010–2020, validated on 2021, and tested on 2022, with 15% of stations held out in space. Across Europe, the United States, East Asia, Southern Africa, and Australia, the descriptor improves MAE, RMSE, and CRPS for 2 m temperature and 10 m wind speed, with overall CRPS reductions of 11.5% and 6.2%. Controls include shuffled descriptors, an extended hand-crafted descriptor, a model-independent residual analysis, Aurora forecast inputs, and a simulated Norwegian station deployment.

Significance. The experimental design is a clear strength: stations are held out in both space and time, results are averaged over three seeds with small cross-seed standard deviations, and comparisons include persistence, lapse-corrected ERA5 interpolation, a no-TESSERA ConvCNP, a shuffled-descriptor control, and an extended physical descriptor baseline. The model-independent residual probe and the Norway deployment simulation provide convergent evidence for the variable-dependent mechanism: topography organizes temperature corrections, while TESSERA supplies additional surface structure for wind speed. If the central claim holds, this is the first demonstration that frozen Earth-observation foundation embeddings can serve as transferable sub-grid descriptors for short-timescale probabilistic downscaling at previously unseen stations. The main caveat is the static 2017 embedding used for pre-2017 training data, which requires a timeliness test before the practical claims are fully established.

major comments (1)
  1. [§2.1, §3.5] The paper uses the 2017 TESSERA embedding map for all training snapshots, including 2010–2016 targets. This means that for every training pair with target time before 2017, the surface descriptor is computed from satellite imagery acquired after the target time. The stability caveat in §2.1 addresses persistence of surface properties, but it does not address the temporal direction of the training inputs. If land cover, vegetation, snow regime, or the built environment changed systematically at any station between 2010 and 2017, the model can learn a mapping that relies on future surface state. The 2022 test is not contaminated at test time, but the trained parameters may be biased by the pre-2017 future-feature pairs, and this bias could inflate or misattribute the reported gains. The same issue affects the Norway deployment experiment in §3.5, where training data from 2010–2014 are paired with 2017 descriptors. Please add a timeliness test: for example, retrain the model on data from 2017 onward only (using the same 2022 test year) and compare the TESSERA uplift, or use time-matched embeddings for a subset of years if available. This is load-bearing for the practical claim that a static descriptor improves historical and real-time downscaling.
minor comments (5)
  1. [§3.1, Table 1] The 'All regions' rows and the abstract's 'overall' improvements appear to be unweighted means over the five regions. Given that Europe contributes 5,825 stations while Australia contributes 140, the 'overall' figure could be misinterpreted as a pooled station-level improvement. Please state the aggregation explicitly and, if feasible, also report a station-weighted pooled CRPS.
  2. [§2.2, Appendix A.3] The VAE training corpus includes station-centered patches from all splits, including held-out stations. This is unsupervised and consistent with the deployment setting, but the main text should state explicitly that the VAE pretraining uses patches from held-out stations, rather than leaving this detail to Appendix A.3.
  3. [§3.4, Figure 6] The Aurora forecast experiment reports only RMSE uplift in Figure 6, although the paper's central claim is about probabilistic skill. Adding CRPS results for the Aurora leads would make the probabilistic claim directly visible in the forecast-driven setting.
  4. [§3.5, Figure 7] The sentence 'The same holds on the probes not yet deployed, which the figure omits (0.30 and 0.36 m/s)' is confusing because the right column of Figure 7 is labeled as permanently held-out Norwegian stations, not as not-yet-deployed probes. Please clarify which station set is being referred to in the main text.
  5. [Throughout] The spelling of the embedding name is inconsistent ('Tessera' vs 'TESSERA'); please use a single convention, preferably matching the official project name.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the TESSERA descriptor is an external, label-free surface representation, and the claimed gains are genuine held-out predictions.

full rationale

The derivation is self-contained. The sub-grid descriptor z_T is produced by a frozen VAE trained only to reconstruct TESSERA patches; Appendix A.3 states that the encoder sees neither weather observations nor split labels, so the descriptor is not fitted to the downscaling target. The downscaler is trained by maximum likelihood against station observations (eq. 3) and evaluated at stations and dates withheld from training, so the reported CRPS improvements are not a fit renamed as prediction. The residual-structure analysis of Section 3.3 constructs its target from ERA5 interpolation minus station observations (eq. 15) and uses cross-validated random forests, independent of the trained ConvCNP, so it does not assume what it concludes. The Aurora forecast and Norwegian deployment experiments are direct transfer tests rather than derivations from the input. The paper's explicit caveat that the 2017 TESSERA map is used for the full 2010–2022 period (Section 2.1) is a legitimate stationarity limitation and a potential temporal-alignment concern, but it is not circular: the embedding does not encode the target weather values, so no prediction is equivalent to its input by construction. The unsupervised VAE and z-scoring also use unlabeled surface patches from all stations, which is transductive but label-free. Self-citations to TESSERA and ConvCNP are ordinary reuse of prior models; no load-bearing claim reduces to a self-cited uniqueness theorem or ansatz.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The central claim rests on the assumption that TESSERA embeddings provide a stable, transferable surface representation, and on the standard ConvCNP factorized predictive construction. The hand-chosen VAE latent dimension, patch size, and regularization weights are design choices not fitted to weather targets; they affect the descriptor but are not tuned against the held-out metric. The lapse-rate correction Gamma is a baseline fitting parameter, not part of the proposed model. No new physical entities are introduced.

free parameters (5)
  • VAE latent dimension = 16
    Hand-chosen bottleneck size; paper states it is deliberately simple and not optimized for the downscaling target (Section 2.2, Section 4).
  • TESSERA patch size = 64x64 pixels (~640 m)
    Hand-chosen local neighbourhood; not optimized for the downscaling target (Section 2.2).
  • VAE KL weight beta = 5e-4 (annealed from 0)
    Set to avoid posterior collapse; not fitted to weather data (Appendix A.3).
  • VAE gradient loss weight lambda_grad = 0.5
    Regularization to avoid flat reconstructions; hand-set (Appendix A.3).
  • Lapse-rate correction Gamma (ERA5 interpolation baseline) = fitted by least squares per region on training split
    Used to strengthen the ERA5 interpolation baseline for t2m; not part of the proposed model (Section 2.4).
assumptions (5)
  • domain assumption TESSERA 2017 embeddings remain valid representations of surface properties across the 2010-2020 training period and the 2022 test year.
    Explicitly stated in Section 2.1 and listed as a limitation in Section 4. If land cover or surface properties changed materially, the static descriptor would be stale.
  • domain assumption The factorized predictive distribution (conditional independence of target locations given context and descriptors) is an acceptable training objective.
    Explicitly acknowledged in Appendix A and Section 4 as not modelling residual spatial dependence; the paper's metrics are per-observation marginals.
  • domain assumption GHCNh station observations are treated as ground truth for near-surface temperature and wind.
    Standard practice in downscaling; station metadata elevation is used for the topographic descriptor (Appendix A.2).
  • domain assumption TESSERA patch embeddings are available at any terrestrial query location, including ungauged sites.
    Needed for the claimed applicability to newly deployed stations; stated in Section 1 and Section 2.1.
  • domain assumption The coarse ERA5/Aurora grids contain no sub-grid information; local descriptors are the only source of sub-grid structure.
    Architectural premise of the ConvCNP downscaler, stated in Section 2.3: no mapping of its encoded representation can recover location-specific structure that is absent from the grid itself.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Earth observation embeddings are effective sub-grid descriptors for probabilistic weather downscaling." pith.science (2026). https://pith.science/paper/Q6RBORGJ

@misc{pith2026260812271,
  author       = {Pith},
  title        = {Pith review of: Earth observation embeddings are effective sub-grid descriptors for probabilistic weather downscaling},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Q6RBORGJ}},
  note         = {Machine review of arXiv:2608.12271}
}
read the original abstract

Global weather reanalyses and forecasts resolve the evolving atmospheric state on coarse grids, but site-specific applications require predictions at arbitrary locations where near-surface conditions also depend on unresolved terrain and land-surface properties. Existing probabilistic downscalers address this gap using hand-crafted topographic descriptors. We ask instead whether Earth observation foundation models can provide transferable sub-grid surface representations for probabilistic weather downscaling. We augment a convolutional conditional neural process that downscales coarse ERA5 reanalysis fields at ~25 km resolution with a learned local surface descriptor, obtained by compressing a patch of TESSERA embeddings at 10 m resolution. Although these embeddings summarise surface conditions over annual timescales, they improve downscaling of instantaneous 2 m temperature and 10 m wind speed by encoding persistent surface properties that capture a location's departure from the coarse-grid atmospheric state. Across five climatically diverse regions, the embedding improves point and probabilistic skill at stations held out in both space and time, overall improving CRPS skill by 11.5% for 2 m temperature and 6.2% for 10 m wind speed. We further analyse how its contribution differs by variable, finding that topography explains more of temperature's sub-grid structure, while TESSERA provides additional surface information for wind speed. These improvements persist when the coarse input is changed from ERA5 to forecasts from the Aurora AI forecasting model, and when predicting at newly deployed stations with no regional history. To our knowledge, this is the first evidence that long-timescale Earth-observation embeddings can support short-timescale weather downscaling where sub-grid departures are systematically structured by persistent surface properties.

Figures

Figures reproduced from arXiv: 2608.12271 by the authors.

Figure 1
Figure 1. The five evaluation regions and their respective station networks. Black rectangles are the [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. CRPS percentage reduction of ConvCNP with [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗
Figure 3
Figure 3. Dense 0.05◦ (∼5 km) downscaling over two climatically distinct regions. Within each map, columns show (left to right) ERA5 bilinear interpolation, no-Tessera ConvCNP, and ConvCNP with Tessera, each overlaid on a high-resolution DEM. 11 [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figures from the paper (9 more)
Figure 4
Figure 4. Figure 4: Spatial structure of the change in predictive mean induced by [PITH_FULL_IMAGE:figures/full_fig_p012_4.png]
Figure 5
Figure 5. Figure 5: Cross-validated R2 of random-forest regressors predicting the persistent ERA5- interpolation residual from each descriptor space in Europe and the United States (n ≥ 100). Left: 2 m temperature. Right: 10 m wind speed. Positive values indicate transferable residual str…
Figure 6
Figure 6. Figure 6: Uplift of ConvCNP with Tessera in RMSE (% better; positive favours Tessera) over two references, by variable (rows) and region (columns). Blue: the no-Tessera ConvCNP baseline. Black: bilinear interpolation of the 0.25◦ weather field supplied to the model at that lead …
Figure 7
Figure 7. Figure 7: Simulated Norway deployment. MAE on a held-out test year as the Norwegian probe [PITH_FULL_IMAGE:figures/full_fig_p016_7.png]
Figure 8
Figure 8. Figure 8: The probe of Figure 5, widened with the extended hand-crafted descriptor. Cross [PITH_FULL_IMAGE:figures/full_fig_p032_8.png]
Figure 9
Figure 9. Figure 9: Per-station baseline error e = y −yˆb against the Tessera increment ∆ = ˆyTessera −yˆb at in-region test stations for the mapped snapshots (t2m left, wind right). Shaded quadrants denote corrections with the same sign as the baseline error, giving directional hit rates…
Figure 10
Figure 10. Figure 10: Relative CRPS skill decay with forecast lead, normalised separately for each model [PITH_FULL_IMAGE:figures/full_fig_p037_10.png]
Figure 11
Figure 11. Figure 11: Where each descriptor places Norway relative to the European training set prior to [PITH_FULL_IMAGE:figures/full_fig_p038_11.png]
Figure 12
Figure 12. Figure 12: At each point along the phased deployment, the out-of-training Norwegian stations [PITH_FULL_IMAGE:figures/full_fig_p039_12.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

54 extracted references · 28 canonical work pages

  1. [1]

    Cool surfaces and shade trees to reduce energy use and improve air quality in urban areas.Solar Energy, 70(3):295–310, 2001

    Hashem Akbari, Melvin Pomerantz, and Haider Taha. Cool surfaces and shade trees to reduce energy use and improve air quality in urban areas.Solar Energy, 70(3):295–310, 2001. doi: 10.1016/S0038-092X(00)00089-X

    work page doi:10.1016/s0038-092x(00)00089-x 2001

  2. [2]

    GraphDOP: Towards skilful data-driven medium-rangeweatherforecastslearntandinitialiseddirectlyfromobservations.arXiv preprint arXiv:2412.15687, 2024

    Mihai Alexe, Eulalie Boucher, Peter Lean, Ewan Pinnington, Patrick Laloyaux, Anthony McNally, Simon Lang, Matthew Chantry, Chris Burrows, Marcin Chrust, Florian Pinault, Ethel Villeneuve, Niels Bormann, and Sean Healy. GraphDOP: Towards skilful data-driven medium-rangeweatherforecastslearntandinitialiseddirectlyfromobservations.arXiv preprint arXiv:2412.1...

    arXiv doi:10.48550/arxiv.2412.15687 2024

  3. [3]

    Bruinsma, Tom R

    Anna Allen, Stratis Markou, Will Tebbutt, James Requeima, Wessel P. Bruinsma, Tom R. An- dersson, MichaelHerzog, Nicholas D.Lane, MatthewChantry, J. ScottHosking, andRichardE. Turner. End-to-end data-driven weather prediction.Nature, 641(8065):1172–1179, 2025. doi: 10.1038/s41586-025-08897-0

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

  4. [4]

    Deep learning for day forecasts from sparse observa- tions.arXiv preprint arXiv:2306.06079, 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 observa- tions.arXiv preprint arXiv:2306.06079, 2023. doi: 10.48550/arXiv.2306.06079

    arXiv doi:10.48550/arxiv.2306.06079 2023

  5. [5]

    Enhancing a high resolution data-driven weather prediction model with surface descriptors

    Åsmund Bakketun, Håvard Homleid Haugen, Jostein Blyverket, Thomas Nils Nipen, and Malte Müller. Enhancing a high resolution data-driven weather prediction model with surface descrip- tors.arXiv preprint arXiv:2607.02824, 2026. doi: 10.48550/arXiv.2607.02824

    work page Pith review arXiv doi:10.48550/arxiv.2607.02824 2026

  6. [6]

    Configuration and in- tercomparison of deep learning neural models for statistical downscaling.Geoscientific Model Development, 13(4):2109–2124, 2020

    Jorge Baño-Medina, Rodrigo Manzanas, and José Manuel Gutiérrez. Configuration and in- tercomparison of deep learning neural models for statistical downscaling.Geoscientific Model Development, 13(4):2109–2124, 2020. doi: 10.5194/gmd-13-2109-2020

    doi:10.5194/gmd-13-2109-2020 2020

  7. [7]

    Bruinsma, Ana Lucic, Megan Stanley, Anna Allen, Johannes Brand- stetter, Patrick Garvan, Maik Riechert, Jonathan A

    Cristian Bodnar, Wessel P. Bruinsma, Ana Lucic, Megan Stanley, Anna Allen, Johannes Brand- stetter, Patrick Garvan, Maik Riechert, Jonathan A. Weyn, Haiyu Dong, Jayesh K. Gupta, Kit Thambiratnam, Alexander T. Archibald, Chun-Chieh Wu, Elizabeth Heider, Max Welling, Richard E. Turner, and Paris Perdikaris. A foundation model for the Earth system.Nature, 64...

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

  8. [8]

    Bruinsma, Stratis Markou, James Requeima, Andrew Y

    Wessel P. Bruinsma, Stratis Markou, James Requeima, Andrew Y. K. Foong, Tom R. Anders- son, Anna Vaughan, Anthony Buonomo, J. Scott Hosking, and Richard E. Turner. Autoregres- sive conditional neural processes. In11th International Conference on Learning Representations (ICLR), 2023. doi: 10.48550/arXiv.2303.14468

    arXiv doi:10.48550/arxiv.2303.14468 2023

Show all 54 references
  1. [9]

    Neural diffusion processes

    Vincent Dutordoir, Alan Saul, Zoubin Ghahramani, and Fergus Simpson. Neural diffusion processes. InProceedings of the 40th International Conference on Machine Learning (ICML), volume 202 ofProceedings of Machine Learning Research, pages 8990–9012. PMLR, 2023. URL https://proce...

    2023

  2. [10]

    Environmental lapse rate for high-resolution land surface downscaling: An application to ERA5.Earth and Space Science, 7(5):e2019EA000984, 2020

    Emanuel Dutra, Joaquín Muñoz-Sabater, Souhail Boussetta, Takuya Komori, Shoji Hirahara, and Gianpaolo Balsamo. Environmental lapse rate for high-resolution land surface downscaling: An application to ERA5.Earth and Space Science, 7(5):e2019EA000984, 2020. doi: 10.1029/ 2019EA0...

    2020

  3. [11]

    Copernicus Digital Elevation Model: GLO-30,

    European Space Agency and European Union. Copernicus Digital Elevation Model: GLO-30,

  4. [12]

    Farr, Paul A

    Tom G. Farr, Paul A. Rosen, Edward Caro, Robert Crippen, Riley Duren, Scott Hensley, Michael Kobrick, Mimi Paller, Ernesto Rodriguez, Ladislav Roth, David Seal, Scott Shaffer, Joanne Shimada, Jeffrey Umland, Marian Werner, Michael Oskin, Douglas Burbank, and Dou- glas Alsdorf....

  5. [13]

    Lisaius, Markus Immitzer, Toby Jackson, James Ball, David A

    Zhengpeng Feng, Clement Atzberger, Sadiq Jaffer, Jovana Knezevic, Silja Sormunen, Robin Young, Madeline C. Lisaius, Markus Immitzer, Toby Jackson, James Ball, David A. Coomes, Anil Madhavapeddy, Andrew Blake, and Srinivasan Keshav. TESSERA: Temporal embeddings of surface spect...

    2026

  6. [14]

    TESSERA v2: Scaling pixel-wise earth foundation models

    Zhengpeng Feng, Sadiq Jaffer, Ira Shokar, Jovana Knezevic, Mark Elvers, Clement Atzberger, Robin Young, Aneesh Naik, Niall Robinson, Andrew Blake, David Coomes, Anil Mad- havapeddy, and Srinivasan Keshav. TESSERA v2: Scaling pixel-wise earth foundation models. arXiv preprint a...

    2026 doi arXiv

  7. [15]

    Christopher A. T. Ferro, David S. Richardson, and Andreas P. Weigel. On the effect of ensemble size on the discrete and continuous ranked probability scores.Meteorological Applications, 15 (1):19–24, 2008. doi: 10.1002/met.45

    2008 doi

  8. [16]

    Andrew Y. K. Foong, Wessel P. Bruinsma, Jonathan Gordon, Yann Dubois, James Requeima, and Richard E. Turner. Meta-learning stationary stochastic process prediction with convo- lutional neural processes. InAdvances in Neural Information Processing Systems (NeurIPS), volume 33, ...

    2020

  9. [17]

    Rezende, and S

    Marta Garnelo, Dan Rosenbaum, Christopher Maddison, Tiago Ramalho, David Saxton, Mur- ray Shanahan, Yee Whye Teh, Danilo J. Rezende, and S. M. Ali Eslami. Conditional neu- ral processes. In Jennifer Dy and Andreas Krause, editors,Proceedings of the 35th Interna- tional Confere...

    2018

  10. [18]

    Tilmann Gneiting and Adrian E. Raftery. Strictly proper scoring rules, prediction, and es- timation.Journal of the American Statistical Association, 102(477):359–378, 2007. doi: 10.1198/016214506000001437

    2007 doi

  11. [19]

    Genton, and Eric Aldrich

    Tilmann Gneiting, Kristin Larson, Kenneth Westrick, Marc G. Genton, and Eric Aldrich. Calibrated probabilistic forecasting at the Stateline wind energy center: The regime-switching space–time method.Journal of the American Statistical Association, 101(475):968–979, 2006. doi: ...

    2006 doi

  12. [20]

    Bruinsma, Andrew Y

    Jonathan Gordon, Wessel P. Bruinsma, Andrew Y. K. Foong, James Requeima, Yann Dubois, and Richard E. Turner. Convolutional conditional neural processes. In8th International Conference on Learning Representations (ICLR), Addis Ababa, Ethiopia, 2020. URLhttps: //openreview.net/f...

    2020

  13. [21]

    Development of temperature patterns during clear nights.Journal of Applied Meteorology, 37(6):559–571, 1998

    Torbjörn Gustavsson, Maria Karlsson, Jörgen Bogren, and Sven Lindqvist. Development of temperature patterns during clear nights.Journal of Applied Meteorology, 37(6):559–571, 1998. 20

    1998

  14. [22]

    J. M. Gutiérrez, D. Maraun, M. Widmann, et al. An intercomparison of a large ensemble of statistical downscaling methods over Europe: Results from the VALUE perfect predictor cross-validation experiment.International Journal of Climatology, 39(9):3750–3785, 2019. doi: 10.1002/joc.5462

    2019 doi

  15. [23]

    Lucy Harris, Andrew T. T. McRae, Matthew Chantry, Peter D. Dueben, and Tim N. Palmer. A generative deep learning approach to stochastic downscaling of precipitation fore- casts.Journal of Advances in Modeling Earth Systems, 14(10):e2022MS003120, 2022. doi: 10.1029/2022MS003120

    2022 doi

  16. [24]

    The ERA5 global reanalysis.Quarterly Journal of the Royal Meteorological Society, 146(730):1999–2049, 2020

    Hans Hersbach, Bill Bell, Paul Berrisford, et al. The ERA5 global reanalysis.Quarterly Journal of the Royal Meteorological Society, 146(730):1999–2049, 2020. doi: 10.1002/qj.3803

    1999 doi

  17. [25]

    A comparative study of convolutional neural network models for wind field downscaling.Meteorological Ap- plications, 27(6):e1961, 2020

    Kevin Höhlein, Michael Kern, Timothy Hewson, and Rüdiger Westermann. A comparative study of convolutional neural network models for wind field downscaling.Meteorological Ap- plications, 27(6):e1961, 2020. doi: 10.1002/met.1961

    2020 doi

  18. [26]

    Learning dynamics in linear VAE: Posterior collapse threshold, superfluous latent space pitfalls, and speedup with KL annealing

    Yuma Ichikawa and Koji Hukushima. Learning dynamics in linear VAE: Posterior collapse threshold, superfluous latent space pitfalls, and speedup with KL annealing. InProceedings of the 27th International Conference on Artificial Intelligence and Statistics (AISTATS), volume 238...

    1936

  19. [27]

    Innovative short-term weather forecasting system combining data-driven and dynamic down- scaling approaches.Artificial Intelligence for the Earth Systems, 4(3), 2025

    Yubin Jin, Zhenzhong Zeng, Yuntian Chen, Yingzuo Qin, Hao Xu, and Dongxiao Zhang. Innovative short-term weather forecasting system combining data-driven and dynamic down- scaling approaches.Artificial Intelligence for the Earth Systems, 4(3), 2025. doi: 10.1175/ AIES-D-24-0125.1

    2025

  20. [28]

    Kingma and Max Welling

    Diederik P. Kingma and Max Welling. Auto-encoding variational bayes. In2nd International Conference on Learning Representations (ICLR), 2014. doi: 10.48550/arXiv.1312.6114

    2014 doi arXiv

  21. [29]

    A high-resolution canopy height model of the Earth.Nature Ecology & Evolution, 7(11):1778–1789, 2023

    Nico Lang, Walter Jetz, Konrad Schindler, and Jan Dirk Wegner. A high-resolution canopy height model of the Earth.Nature Ecology & Evolution, 7(11):1778–1789, 2023. doi: 10.1038/ s41559-023-02206-6

    2023

  22. [30]

    Decoupled weight decay regularization

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. In7th Interna- tional Conference on Learning Representations (ICLR), 2019. URLhttps://openreview. net/forum?id=Bkg6RiCqY7

    2019

  23. [31]

    Terrain tiles

    Mapzen. Terrain tiles. Registry of Open Data on AWS, 2017. URLhttps://registry. opendata.aws/terrain-tiles/. Accessed 5 August 2026

    2017

  24. [32]

    Cambridge University Press, Cambridge, 2018

    Douglas Maraun and Martin Widmann.Statistical Downscaling and Bias Correction for Cli- mate Research. Cambridge University Press, Cambridge, 2018. doi: 10.1017/9781107588783

    2018 doi

  25. [33]

    Residual corrective diffusion modeling for km- scale atmospheric downscaling.Communications Earth & Environment, 6:124, 2025

    Morteza Mardani, Noah Brenowitz, Yair Cohen, Jaideep Pathak, Chieh-Yu Chen, Cheng- Chin Liu, Arash Vahdat, Mohammad Amin Nabian, Tao Ge, Akshay Subramaniam, Karthik Kashinath, Jan Kautz, and Mike Pritchard. Residual corrective diffusion modeling for km- scale atmospheric downs...

    2025 doi

  26. [34]

    Marinescu, Susan C

    Peter J. Marinescu, Susan C. van den Heever, Leah D. Grant, Jennie Bukowski, and Itinderjot Singh. How much convective environment subgrid spatial variability is missing within atmo- spheric reanalysis data sets?Geophysical Research Letters, 51(24):e2024GL111856, 2024. doi: 10...

    2024 doi

  27. [35]

    Menne, Simon Noone, Nancy W

    Matthew J. Menne, Simon Noone, Nancy W. Casey, Robert H. Dunn, Shelley McNeill, Diana Kantor, Peter W. Thorne, Karen Orcutt, Sam Cunningham, and Nicholas Risavi. Global historical climatology network–hourly (GHCNh). NOAA National Centers for Environmental Information, Version ...

    2023

  28. [36]

    Ophelia Miralles, Daniel Steinfeld, Olivia Martius, and Anthony C. Davison. Downscaling of historical wind fields over Switzerland using generative adversarial networks.Artificial Intelli- gence for the Earth Systems, 1(4):e220018, 2022. doi: 10.1175/AIES-D-22-0018.1

    2022 doi

  29. [37]

    T. R. Oke. The energetic basis of the urban heat island.Quarterly Journal of the Royal Meteorological Society, 108(455):1–24, 1982. doi: 10.1002/qj.49710845502

    1982 doi

  30. [38]

    On the difficulty of training recurrent neural networks

    Razvan Pascanu, Tomas Mikolov, and Yoshua Bengio. On the difficulty of training recurrent neural networks. InProceedings of the 30th International Conference on Machine Learning (ICML), volume 28 ofProceedings of Machine Learning Research, pages 1310–1318. PMLR,

  31. [39]

    de Sousa, Niels H

    Laura Poggio, Luis M. de Sousa, Niels H. Batjes, Gerard B. M. Heuvelink, Bas Kempen, Eloi Ribeiro, and David Rossiter. SoilGrids 2.0: producing soil information for the globe with quantified spatial uncertainty.SOIL, 7(1):217–240, 2021. doi: 10.5194/soil-7-217-2021

    2021 doi

  32. [40]

    Gibson, Jorge Baño-Medina, Gab Abramowitz, Tom Beucler, Jose González-Abad, William Chapman, Paula Harder, and José Manuel Gutiér- rez

    Neelesh Rampal, Sanaa Hobeichi, Peter B. Gibson, Jorge Baño-Medina, Gab Abramowitz, Tom Beucler, Jose González-Abad, William Chapman, Paula Harder, and José Manuel Gutiér- rez. Enhancing regional climate downscaling through advances in machine learning.Artificial Intelligence ...

    2024 doi

  33. [41]

    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 Rus- sell, 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...

    2024 doi

  34. [42]

    A simple height-based correction for temperature downscaling in complex terrain.Meteorological Applications, 17(3): 329–339, 2010

    Peter Sheridan, Samantha Smith, Andrew Brown, and Simon Vosper. A simple height-based correction for temperature downscaling in complex terrain.Meteorological Applications, 17(3): 329–339, 2010. doi: 10.1002/met.177

    2010 doi

  35. [43]

    Theobald, Dylan Harrison-Atlas, William B

    David M. Theobald, Dylan Harrison-Atlas, William B. Monahan, and Christine M. Albano. Ecologically-relevant maps of landforms and physiographic diversity for climate adaptation planning.PLOS ONE, 10(12):e0143619, 2015. doi: 10.1371/journal.pone.0143619

    2015 doi

  36. [44]

    Thorarinsdottir and Tilmann Gneiting

    Thordis L. Thorarinsdottir and Tilmann Gneiting. Probabilistic forecasts of wind speed: Ensemble model output statistics by using heteroscedastic censored regression.Journal of the Royal Statistical Society: Series A (Statistics in Society), 173(2):371–388, 2010. doi: 10.1111/...

    2010 arXiv

  37. [45]

    Scott Hosking, and Richard E

    Anna Vaughan, Will Tebbutt, J. Scott Hosking, and Richard E. Turner. Convolutional condi- tional neural processes for local climate downscaling.Geoscientific Model Development, 15(1): 251–268, 2022. doi: 10.5194/gmd-15-251-2022

    2022 doi

  38. [46]

    Taylor, and John L

    Wensong Weng, Peter A. Taylor, and John L. Walmsley. Guidelines for airflow over complex terrain: Model developments.Journal of Wind Engineering and Industrial Aerodynamics, 86 (2–3):169–186, 2000. doi: 10.1016/S0167-6105(00)00009-X

    doi

  39. [47]

    Statistical downscaling of gridded wind speed data using local topography.Journal of Hydrometeorology, 18(2):335–348, 2017

    Adam Winstral, Tobias Jonas, and Nora Helbig. Statistical downscaling of gridded wind speed data using local topography.Journal of Hydrometeorology, 18(2):335–348, 2017. doi: 10.1175/ JHM-D-16-0054.1

    2017

  40. [48]

    Multi-modal graph neural networks for localized off-grid weather forecasting.arXiv preprint arXiv:2410.12938, 2024

    Qidong Yang, Jonathan Giezendanner, Daniel Salles Civitarese, Johannes Jakubik, Eric Schmitt, Anirban Chandra, Jeremy Vila, Detlef Hohl, Chris Hill, Campbell Watson, and Sher- rie Wang. Multi-modal graph neural networks for localized off-grid weather forecasting.arXiv preprint...

    doi arXiv

  41. [49]

    Dynamical global downscaling of global reanalysis

    Kei Yoshimura and Masao Kanamitsu. Dynamical global downscaling of global reanalysis. Monthly Weather Review, 136(8):2983–2998, 2008. doi: 10.1175/2008MWR2281.1

    2008 doi

  42. [50]

    ESA WorldCover 10 m 2021 v200, 2022

    Daniele Zanaga, Ruben Van De Kerchove, Dirk Daems, Wanda De Keersmaecker, Carsten Brockmann, Grit Kirches, Jan Wevers, Oliver Cartus, Maurizio Santoro, Steffen Fritz, My- roslava Lesiv, Martin Herold, Nandin-Erdene Tsendbazar, Panpan Xu, Fabrizio Ramoino, and Olivier Arino. ES...

    2021

  43. [512]

    The four blocks take the patch from64×64down to4×4, and the resulting512×4×4 feature map is flattened and projected by two parallel linear layers to the mean and log-variance of a diagonal Gaussian posterior over a16-dimensional vector. Flattening rather than global-average- p...

    2020

  44. [2007]

    doi: 10.1029/2005RG000183

    doi

  45. [2013]

    URLhttps://proceedings.mlr.press/v28/pascanu13.html

  46. [2021]

    GLO-30 global digital surface model

    URLhttps://doi.org/10.5270/ESA-c5d3d65. GLO-30 global digital surface model

    doi

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.