Pith. sign in

REVIEW 2 major objections 5 minor 85 references

Temporal domain adaptation improves high-resolution temperature downscaling under future warming, with the largest gains when climate shift is strongest.

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 ↗

Adversarial temporal domain adaptation on LR features improves daily CONUS temperature downscaling under future warming more than statistical bias correction or non-adaptive deep models, especially late-century.

T0 review reviewed 2026-07-11 challenge →

load-bearing objection Careful empirical win for temporal DANN-style adaptation on GCM–RCM temperature downscaling; gains look real in-setup, but the per-period future-LR protocol overstates pure historical→future transfer. the 2 major comments →

arxiv 2607.05645 v1 pith:MIHIVV2A submitted 2026-07-06 cs.LG physics.ao-ph

Domain-Adaptive Climate Downscaling Under Temporal Distribution Shift

Shuochen Wang , Nishant Yadav , Auroop R. Ganguly This is my paper
classification cs.LG physics.ao-ph
keywords climate downscalingdomain adaptationtemporal distribution shiftout-of-distributiontemperaturesuper-resolutionGCM-RCMadversarial alignment
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The reading

Deep learning climate downscalers are usually trained on historical low-resolution/high-resolution pairs and then applied to future projections. Under warming the future climate is out-of-distribution, so the learned mapping can degrade. This paper treats historical climate as a labeled source domain and future low-resolution fields as an unlabeled target domain, and trains a residual super-resolution network with both supervised reconstruction loss and adversarial domain alignment. On daily temperature over the contiguous United States from three GCM–RCM pairs, the domain-adaptive model beats bilinear interpolation, classical bias-correction methods, and non-adaptive deep baselines on standard reconstruction metrics, with the biggest relative gains in the late-century period when the distribution shift is largest. Gains are especially visible over high terrain and improve spatial-temporal correlation with the high-resolution target; upper-tail temperature bias is also reduced relative to the non-adaptive network. The practical claim is that explicit temporal domain alignment can make high-resolution climate projections more reliable when the climate is non-stationary.

Core claim

Framing historical-to-future climate downscaling as unsupervised domain adaptation—supervised high-resolution reconstruction on historical pairs plus adversarial alignment of historical and future low-resolution features—yields consistently better daily CONUS temperature fields under RCP8.5 than statistical bias correction or non-adaptive deep super-resolution, and the advantage grows as the temporal distribution shift strengthens.

What carries the argument

RCAN-DA: residual channel-attention encoder plus bilinear-plus-residual reconstruction head, trained with supervised MSE on historical labeled pairs and a DANN-style gradient-reversal domain classifier that aligns unlabeled future low-resolution features so the encoder becomes less sensitive to historical-future shift.

Load-bearing premise

Unlabeled future low-resolution fields from the same model chain, aligned only in feature space, are enough to keep the historical low-to-high-resolution mapping valid under strong warming, and regional climate model output is a fair stand-in for judging real-world skill.

What would settle it

Train and evaluate the same domain-adaptive architecture on a GCM–RCM or GCM–observation pair whose future temperature distribution shift is comparable or larger; if the adaptive model no longer reduces MSE/RMSE relative to the identical non-adaptive backbone (especially late-century and over high terrain), the central claim fails.

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

Share X Bluesky LinkedIn Reddit HN

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

2 major / 5 minor

Summary. The paper studies temporal out-of-distribution shift in deep climate downscaling of daily near-surface temperature over CONUS using imperfect-model GCM–RCM pairs (CanESM2-RCA4, CanESM2-CanRCM4, EC-EARTH-RCA4). It frames historical LR–HR pairs as a labeled source domain and future LR fields as an unlabeled target domain, and proposes RCAN-DA: an RCAN residual super-resolution backbone with bilinear skip and HR auxiliaries, trained with supervised MSE on historical data plus DANN-style adversarial alignment of encoder features via a gradient-reversal domain classifier (LDA = Lsup + α Lalign). Baselines include bilinear, BCSD, QDM, CDF-t, supervised RCAN, RCAN-QDM, and RCAN-GAN. Across three future windows (2006–2040, 2041–2070, 2071–2099), RCAN-DA ranks best on MSE/RMSE/PSNR/SSIM, with larger gains under stronger warming; spatial MSE maps, elevation bins, Taylor diagnostics, and P95/quantile analyses support improved topographic skill and somewhat reduced upper-tail bias, while remaining under-dispersive.

Significance. If the gains hold under a protocol that matches operational use, the work is a clear, useful contribution: it treats historical-to-future downscaling as domain adaptation rather than pure supervised super-resolution, documents progressive temporal shift, and shows consistent ranking across three GCM–RCM chains with careful leakage control (historical chronological Optuna split; future HR unused for training/selection). Strengths include multi-dataset evaluation, spatial/elevation/Taylor/extreme diagnostics, and public code/data pointers. The main scientific value is empirical evidence that adversarial LR-feature alignment can improve non-stationary temperature downscaling skill, especially in complex terrain, within a pseudo-reality setting.

major comments (2)
  1. §6 states that a separate RCAN-DA is trained for each future validation period using that period’s unlabeled LR fields as the target domain for alignment. The abstract, introduction, and conclusion frame the result as improved robustness of models trained on historical data when applied to future projections under temporal OOD. The reported protocol tests adaptation with access to the evaluation window’s LR distribution, not pure historical→future transfer of a single frozen model. Because headline gains grow with shift (e.g., Table 2: MSE 5.008→4.387 in 2071–2099 vs RCAN), this is load-bearing for the central claim. Please either (i) add a single-model / forward-transfer experiment (historical source only, or adaptation only on near-term unlabeled LR, then evaluate later periods without re-adapting on the scored window), or (ii) restate claims and abstract to match the per-period target
  2. Discussion §8 correctly notes the pseudo-reality (RCM-as-truth) setting and intermediate-shift caveats, but does not quantify how much of the RCAN-DA advantage survives without per-period target LR or under a fixed adaptation setup. Given that the strongest claim is temporal generalization under non-stationarity, a short ablation or sensitivity table (same architecture/hyperparameters; target domain = none / early-future LR / full evaluation-period LR) would make the contribution falsifiable and proportionate to the abstract’s language.
minor comments (5)
  1. Table 1 and §6: clarify whether α and domain-classifier hyperparameters selected on the historical pseudo-target split are frozen across all three future periods or re-tuned per period.
  2. Figure 4 caption/date: specify the calendar day and GCM–RCM pair so the example is reproducible.
  3. Notation: Lalign is written as depending on θ (and ψ); make the dependence of GRL/encoder parameters explicit in §3–4 for consistency with the DANN formulation.
  4. Typos/grammar: “This study examine” (§9); “Data A vailability” (§11); occasional missing spaces in author list on the title page.
  5. Appendix Tables A3–A4 use “Full MSE/Full RMSE” while Table 2 uses MSE/RMSE—align column names.

Circularity Check

0 steps flagged

No algebraic or definitional circularity: future HR targets are held out of training and selection; reported skill is empirical, not forced by construction.

full rationale

The paper’s load-bearing claim is empirical: RCAN-DA (supervised historical LR–HR reconstruction plus DANN-style adversarial alignment of historical and future LR encoder features) improves MSE/RMSE/PSNR/SSIM on held-out future HR RCM fields relative to bilinear, BCSD, QDM, CDF-t, RCAN, RCAN-QDM, and RCAN-GAN, with larger gains under stronger temporal shift (Tables 2, A3, A4). The derivation chain is standard unsupervised domain adaptation, not a closed definition: L_DA = L_sup + α L_align (§3), with L_sup only on labeled historical pairs and L_align only on domain labels of source/target LR features via GRL (§4); future HR y^T is never used for gradients, Optuna selection, LR scheduling, or early stopping (§6). Evaluation metrics compare predictions to those held-out future HR fields, so skill is not a fitted parameter renamed as prediction, nor is any equation equivalent to its inputs by construction. Self-citations (e.g., DeepSD with overlapping co-author) are background on super-resolution downscaling and are not uniqueness theorems or load-bearing premises that force the result. Per-period training on unlabeled future LR is an experimental-design choice about operational transferability, not circularity of the reported metrics. Pseudo-reality (RCM as synthetic HR) is a validity caveat, not self-definitional circularity. No steps reduce to tautology.

Axiom & Free-Parameter Ledger

4 free parameters · 5 axioms · 1 invented entities

The central claim is empirical and rests on standard ML/climate modeling assumptions plus a few design choices (per-period unlabeled target, RCM-as-truth, adversarial feature invariance). Free parameters are ordinary training/architecture knobs selected on historical data; no new physical entity is introduced.

free parameters (4)
  • domain-loss weight α
    Balances Lsup and Lalign; selected by Optuna on a historical pseudo-target split (best 5e-4). Directly controls how strongly temporal alignment affects the encoder.
  • GRL schedule λ(p) with λ_max=1
    Hand-set adversarial schedule from DANN practice; shapes how early/late domain invariance is enforced during training.
  • RCAN architecture and training hyperparameters (blocks, channels, groups, attention reduction, residual scale, lr, weigh
    Selected via Optuna on historical validation then frozen; determine capacity and optimization of the shared backbone used by all DL variants.
  • domain-classifier hidden dimension
    Chosen as 64 among {64,128,256}; affects strength/stability of the domain signal used for alignment.
axioms (5)
  • domain assumption Historical labeled LR–HR pairs and unlabeled future LR fields share a learnable domain-invariant feature space in which the historical LR→HR map remains useful under climate change.
    Core domain-adaptation premise in §3–4; if the conditional map changes too much, alignment can cause negative transfer (acknowledged in §8).
  • domain assumption RCM high-resolution fields are an adequate synthetic ground truth (pseudo-reality) for scoring downscaling skill under future conditions.
    Stated in Discussion §8; evaluation never uses observations or reanalysis as HR targets.
  • domain assumption Imperfect GCM–RCM pairs (not perfect-model coarsened HR) are a valid and more challenging test of operational downscaling.
    Data §2 and related-work framing; biases between GCM and RCM are part of the task.
  • standard math Adversarial domain classification with gradient reversal (DANN) is a valid mechanism to reduce temporal feature shift without future HR labels.
    Uses the standard GRL/CE construction of Ganin et al.; not re-derived, treated as established transfer-learning tool.
  • ad hoc to paper Source-domain normalization statistics applied to future inputs/targets are acceptable for training and evaluation.
    §6 normalizes all periods with 1951–2005 mean/std only; a modeling choice that interacts with non-stationarity.
invented entities (1)
  • RCAN-DA temporal domain-adaptive downscaling model no independent evidence
    purpose: Combine RCAN residual reconstruction with a training-only domain classifier on LR features to improve future HR temperature maps.
    Composition of existing modules (RCAN + DANN + bilinear residual + aux elevation/land-sea), not a new physical object; no independent evidence claim beyond the reported experiments.

reviewed 2026-07-11 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Domain-Adaptive Climate Downscaling Under Temporal Distribution Shift." pith.science (2026). https://pith.science/paper/MIHIVV2A

@misc{pith2026260705645,
  author       = {Pith},
  title        = {Pith review of: Domain-Adaptive Climate Downscaling Under Temporal Distribution Shift},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MIHIVV2A}},
  note         = {Machine review of arXiv:2607.05645}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Deep-learning-based climate downscaling aims to learn relationships from historical low-resolution (LR) and high-resolution (HR) climate data to generate HR climate projections. However, this setting faces a temporal out-of-distribution (OOD) challenge: models trained on historical data are commonly applied to future projections whose distributions may differ substantially from the training period. This study investigates temporal OOD shift for daily temperature downscaling over the Continental United States using paired LR-HR model simulations. We propose a temporal domain-adaptive downscaling framework that combines supervised HR reconstruction on historical data with domain alignment between historical and future climate distributions. Experiments across future validation periods show that the proposed domain-adaptive model consistently outperforms statistical and deep-learning-based bias-correction methods, with the largest gains occurring when the temporal distribution shift is strongest. Spatial analyses indicate stronger improvements over high-elevation and topographically complex regions, along with higher spatiotemporal correlation with the HR target. The extreme analysis shows that domain adaptation also reduces upper-tail temperature bias relative to the non-adaptive model. These results demonstrate that temporal domain adaptation can improve the robustness of HR climate projections under non-stationary climate conditions.

Figures

Figures reproduced from arXiv: 2607.05645 by Auroop R. Ganguly, Nishant Yadav, Shuochen Wang.

Figure 1
Figure 1. Figure 1: Temporal domain-adaptive model for climate downscaling. The LR GCM input [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Marginal distribution shift of daily near-surface air temperature over CONUS for the HR [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Spatial pattern of mean temperature change over CONUS relative to the historical training [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Downscaled daily near-surface air temperature fields. The top panel shows the HR [PITH_FULL_IMAGE:figures/full_fig_p011_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Spatial differences in MSE relative to the standard RCAN model for the CanESM2-RCA4 [PITH_FULL_IMAGE:figures/full_fig_p011_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: RMSE of downscaled daily near-surface air temperature across standardized elevation bins [PITH_FULL_IMAGE:figures/full_fig_p012_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Taylor-style diagnostics for daily near-surface air temperature over CONUS using the [PITH_FULL_IMAGE:figures/full_fig_p013_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Quantile-dependent bias in downscaled daily near-surface air temperature for the CanESM2- [PITH_FULL_IMAGE:figures/full_fig_p013_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Spatial bias in the 95th percentile (P95) of daily near-surface air temperature for the four [PITH_FULL_IMAGE:figures/full_fig_p014_9.png] view at source ↗

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

85 extracted references · 1 canonical work pages · 1 internal anchor

  1. [1]

    Overview of the coupled model intercomparison project phase 6 (cmip6) experimental design and organization.Geoscientific Model Development, 9(5): 1937–1958, 2016

    Veronika Eyring, Sandrine Bony, Gerald A Meehl, Catherine A Senior, Bjorn Stevens, Ronald J Stouffer, and Karl E Taylor. Overview of the coupled model intercomparison project phase 6 (cmip6) experimental design and organization.Geoscientific Model Development, 9(5): 1937–1958, 2016

    1937

  2. [2]

    Framing, context, and methods climate change 2021: The physical science basis

    D Chen, M Rojas, BH Samset, K Cobb, A Diongue Niang, P Edwards, S Emori, SH Faria, E Hawkins, P Hope, et al. Framing, context, and methods climate change 2021: The physical science basis. contribution of working group i to the sixth assessment report of the intergovern- mental panel on climate change ed v masson-delmotte et al.Change (Cambridge Univ. Pres...

    2021

  3. [3]

    Deep learning to represent subgrid processes in climate models.Proceedings of the national academy of sciences, 115(39):9684– 9689, 2018

    Stephan Rasp, Michael S Pritchard, and Pierre Gentine. Deep learning to represent subgrid processes in climate models.Proceedings of the national academy of sciences, 115(39):9684– 9689, 2018

    2018

  4. [4]

    Cambridge university press, 2001

    John Theodore Houghton, YDJG Ding, David J Griggs, Maria Noguer, Paul J van der Linden, Xiaosu Dai, Kathy Maskell, and Colin A Johnson.Climate change 2001: the scientific basis: contribution of Working Group I to the third assessment report of the Intergovernmental Panel on Climate Change. Cambridge university press, 2001

    2001

  5. [5]

    Robert L Wilby, Steven P Charles, Eduardo Zorita, Bertrand Timbal, Penny Whetton, and Linda O Mearns. Guidelines for use of climate scenarios developed from statistical downscaling methods.Supporting material of the Intergovernmental Panel on Climate Change, available from the DDC of IPCC TGCIA, 27:27, 2004

    2004

  6. [6]

    Approaches to the simulation of regional climate change: a review.Reviews of geophysics, 29(2):191–216, 1991

    Filippo Giorgi and Linda O Mearns. Approaches to the simulation of regional climate change: a review.Reviews of geophysics, 29(2):191–216, 1991

    1991

  7. [7]

    Regional dynamical downscaling and the cordex initiative.Annual review of environment and resources, 40:467–490, 2015

    Filippo Giorgi and William J Gutowski Jr. Regional dynamical downscaling and the cordex initiative.Annual review of environment and resources, 40:467–490, 2015

    2015

  8. [8]

    Precipitation downscaling under climate change: Recent developments to bridge the gap between dynamical models and the end user.Reviews of geophysics, 48(3), 2010

    Douglas Maraun, Frederick Wetterhall, Anderson M Ireson, Richard E Chandler, Elizabeth J Kendon, Martin Widmann, Stephan Brienen, Henning W Rust, Tobias Sauter, Matthias Themeßl, et al. Precipitation downscaling under climate change: Recent developments to bridge the gap between dynamical models and the end user.Reviews of geophysics, 48(3), 2010

    2010

  9. [9]

    Methodology of the constraint condition in dynamical downscaling for regional climate evaluation: A review.Journal of Geophysical Research: Atmospheres, 125(11):e2019JD032166, 2020

    Sachiho A Adachi and Hirofumi Tomita. Methodology of the constraint condition in dynamical downscaling for regional climate evaluation: A review.Journal of Geophysical Research: Atmospheres, 125(11):e2019JD032166, 2020

    2020

  10. [10]

    The potential to narrow uncertainty in regional climate predictions.Bulletin of the American Meteorological Society, 90(8):1095–1108, 2009

    Ed Hawkins and Rowan Sutton. The potential to narrow uncertainty in regional climate predictions.Bulletin of the American Meteorological Society, 90(8):1095–1108, 2009

    2009

  11. [11]

    Downscaling general circulation model output: a review of methods and limitations.Progress in physical geography, 21(4):530–548, 1997

    Robert L Wilby and Thomas ML Wigley. Downscaling general circulation model output: a review of methods and limitations.Progress in physical geography, 21(4):530–548, 1997

    1997

  12. [12]

    Downscaling of global climate change estimates to regional scales: an application to iberian rainfall in wintertime.Journal of Climate, 6(6):1161–1171, 1993

    Hans V on Storch, Eduardo Zorita, and Ulrich Cubasch. Downscaling of global climate change estimates to regional scales: an application to iberian rainfall in wintertime.Journal of Climate, 6(6):1161–1171, 1993

    1993

  13. [13]

    The analog method as a simple statistical downscaling technique: comparison with more complicated methods.Journal of climate, 12(8):2474–2489, 1999

    Eduardo Zorita and Hans V on Storch. The analog method as a simple statistical downscaling technique: comparison with more complicated methods.Journal of climate, 12(8):2474–2489, 1999

    1999

  14. [14]

    Utility of daily vs

    Edwin P Maurer and Hugo G Hidalgo. Utility of daily vs. monthly large-scale climate data: an intercomparison of two statistical downscaling methods.Hydrology and Earth System Sciences, 12(2):551–563, 2008. 16

    2008

  15. [15]

    Bias correction of gcm precipitation by quantile mapping: how well do methods preserve changes in quantiles and extremes?Journal of climate, 28(17):6938–6959, 2015

    Alex J Cannon, Stephen R Sobie, and Trevor Q Murdock. Bias correction of gcm precipitation by quantile mapping: how well do methods preserve changes in quantiles and extremes?Journal of climate, 28(17):6938–6959, 2015

    2015

  16. [16]

    Downscaling rcm precipitation to the station scale using statistical transformations–a comparison of methods.Hydrology and Earth System Sciences, 16(9):3383–3390, 2012

    Lukas Gudmundsson, John Bjørnar Bremnes, Jan Erik Haugen, and Torill Engen-Skaugen. Downscaling rcm precipitation to the station scale using statistical transformations–a comparison of methods.Hydrology and Earth System Sciences, 16(9):3383–3390, 2012

    2012

  17. [17]

    Interrogating empirical-statistical downscaling.Climatic change, 122(4):539–554, 2014

    BC Hewitson, Joe Daron, RG Crane, MF Zermoglio, and Christopher Jack. Interrogating empirical-statistical downscaling.Climatic change, 122(4):539–554, 2014

    2014

  18. [18]

    Bias correction, quantile mapping, and downscaling: Revisiting the inflation issue.Journal of Climate, 26(6):2137–2143, 2013

    Douglas Maraun. Bias correction, quantile mapping, and downscaling: Revisiting the inflation issue.Journal of Climate, 26(6):2137–2143, 2013

    2013

  19. [19]

    A trend-preserving bias correction–the isi-mip approach.Earth System Dynamics, 4(2):219–236, 2013

    Sabrina Hempel, Katja Frieler, Lila Warszawski, Jacob Schewe, and Franziska Piontek. A trend-preserving bias correction–the isi-mip approach.Earth System Dynamics, 4(2):219–236, 2013

    2013

  20. [20]

    Deepsd: Generating high resolution climate change projections through single image super-resolution

    Thomas Vandal, Evan Kodra, Sangram Ganguly, Andrew Michaelis, Ramakrishna Nemani, and Auroop R Ganguly. Deepsd: Generating high resolution climate change projections through single image super-resolution. InProceedings of the 23rd acm sigkdd international conference on knowledge discovery and data mining, pages 1663–1672, 2017

    2017

  21. [21]

    Deepdownscale: A deep learning strategy for high-resolution weather forecast

    Eduardo Rocha Rodrigues, Igor Oliveira, Renato Cunha, and Marco Netto. Deepdownscale: A deep learning strategy for high-resolution weather forecast. In2018 IEEE 14th International Conference on e-Science (e-Science), pages 415–422. IEEE, 2018

    2018

  22. [22]

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

    2020

  23. [23]

    Climate downscaling using ynet: A deep convolutional network with skip connections and fusion

    Yumin Liu, Auroop R Ganguly, and Jennifer Dy. Climate downscaling using ynet: A deep convolutional network with skip connections and fusion. InProceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pages 3145–3153, 2020

    2020

  24. [24]

    Deep learning for daily precipitation and temperature downscaling.Water Resources Research, 57(4):e2020WR029308, 2021

    Fang Wang, Di Tian, Lisa Lowe, Latif Kalin, and John Lehrter. Deep learning for daily precipitation and temperature downscaling.Water Resources Research, 57(4):e2020WR029308, 2021

    2021

  25. [25]

    Convolutional conditional neural processes for local climate downscaling.Geoscientific Model Development Discussions, 2021:1–25, 2021

    Anna Vaughan, Will Tebbutt, J Scott Hosking, and Richard E Turner. Convolutional conditional neural processes for local climate downscaling.Geoscientific Model Development Discussions, 2021:1–25, 2021

    2021

  26. [26]

    A generative framework for probabilistic, spatiotemporally coherent downscaling of climate simulation.npj Climate and Atmospheric Science, 8(1):270, 2025

    Jonathan Schmidt, Luca Schmidt, Felix M Strnad, Nicole Ludwig, and Philipp Hennig. A generative framework for probabilistic, spatiotemporally coherent downscaling of climate simulation.npj Climate and Atmospheric Science, 8(1):270, 2025

    2025

  27. [27]

    Conditional diffusion models for downscaling and bias correction of earth system model precipitation.Geoscientific Model Development, 19(4):1791–1808, 2026

    Michael Aich, Philipp Hess, Baoxiang Pan, Sebastian Bathiany, Yu Huang, and Niklas Boers. Conditional diffusion models for downscaling and bias correction of earth system model precipitation.Geoscientific Model Development, 19(4):1791–1808, 2026

    2026

  28. [28]

    Generative diffusion-based downscaling for climate

    Robbie A Watt and Laura A Mansfield. Generative diffusion-based downscaling for climate. arXiv preprint arXiv:2404.17752, 2024

    Pith/arXiv arXiv 2024

  29. [29]

    Xiaohui Zhong, Fei Du, Lei Chen, Zhibin Wang, and Hao Li. Investigating transformer-based models for spatial downscaling and correcting biases of near-surface temperature and wind- speed forecasts.Quarterly Journal of the Royal Meteorological Society, 150(758):275–289, 2024

    2024

  30. [30]

    Enhancing regional climate downscaling through advances in machine learning.Artificial Intelligence for the Earth Systems, 3(2):230066, 2024

    Neelesh Rampal, Sanaa Hobeichi, Peter B Gibson, Jorge Baño-Medina, Gab Abramowitz, Tom Beucler, Jose González-Abad, William Chapman, Paula Harder, and José Manuel Gutiérrez. Enhancing regional climate downscaling through advances in machine learning.Artificial Intelligence for the Earth Systems, 3(2):230066, 2024. 17

    2024

  31. [31]

    Some pitfalls in statistical downscaling of future climate.Bulletin of the American Meteorological Society, 99(4):791–803, 2018

    John R Lanzante, Keith W Dixon, Mary Jo Nath, Carolyn E Whitlock, and Dennis Adams- Smith. Some pitfalls in statistical downscaling of future climate.Bulletin of the American Meteorological Society, 99(4):791–803, 2018

    2018

  32. [32]

    Machine learning in climate downscaling: A critical review of methodologies, physical consistency, and operational applications.Water, 18(2):271, 2026

    Najafi Hamed, Lagerwall Gareth Lynton, Obeysekera Jayantha, and Jason Liu. Machine learning in climate downscaling: A critical review of methodologies, physical consistency, and operational applications.Water, 18(2):271, 2026

    2026

  33. [33]

    On the suitability of deep convolutional neural networks for continental-wide downscaling of climate change projections

    Jorge Baño-Medina, Rodrigo Manzanas, and José Manuel Gutiérrez. On the suitability of deep convolutional neural networks for continental-wide downscaling of climate change projections. Climate Dynamics, 57(11):2941–2951, 2021

    2021

  34. [34]

    Alfonso Hernanz, Carlos Correa, Juan-Carlos Sánchez-Perrino, Ignacio Prieto-Rico, Esteban Rodríguez-Guisado, Marta Domínguez, and Ernesto Rodríguez-Camino. On the limitations of deep learning for statistical downscaling of climate change projections: The transferability and the extrapolation issues.Atmospheric Science Letters, 25(2):e1195, 2024

    2024

  35. [35]

    Hebatallah M Abdelmoaty, Simon Michael Papalexiou, Antonios Mamalakis, Shivam Singh, Vincenzo Coia, Melissa Hairabedian, Pascal Szeftel, and Patrick Grover. Does nonstationarity affect gan-based downscaling? insights from high-resolution wrf simulations over the canadian prairies.Artificial Intelligence for the Earth Systems, 5(2):250093, 2026

    2026

  36. [36]

    Repeatable high-resolution statistical downscaling through deep learning.Geoscientific Model Development, 15(19):7353– 7370, 2022

    Dánnell Quesada-Chacón, Klemens Barfus, and Christian Bernhofer. Repeatable high-resolution statistical downscaling through deep learning.Geoscientific Model Development, 15(19):7353– 7370, 2022

    2022

  37. [37]

    Regional climate model emulator based on deep learning: Concept and first evaluation of a novel hybrid downscaling approach.Climate Dynamics, 60(5):1751–1779, 2023

    Antoine Doury, Samuel Somot, Sebastien Gadat, Aurélien Ribes, and Lola Corre. Regional climate model emulator based on deep learning: Concept and first evaluation of a novel hybrid downscaling approach.Climate Dynamics, 60(5):1751–1779, 2023

    2023

  38. [38]

    Multivariate bias correction and downscaling of climate models with trend-preserving deep learning.Climate Dynamics, 62(10):9651–9672, 2024

    Fang Wang and Di Tian. Multivariate bias correction and downscaling of climate models with trend-preserving deep learning.Climate Dynamics, 62(10):9651–9672, 2024

    2024

  39. [39]

    Evaluating the transferability potential of deep learning models for climate downscaling

    Ayush Prasad, Paula Harder, Qidong Yang, Prasanna Sattegeri, Daniela Szwarcman, Campbell Watson, and David Rolnick. Evaluating the transferability potential of deep learning models for climate downscaling.arXiv preprint arXiv:2407.12517, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  40. [40]

    Deep domain confusion: Maximizing for domain invariance.arXiv preprint arXiv:1412.3474, 2014

    Eric Tzeng, Judy Hoffman, Ning Zhang, Kate Saenko, and Trevor Darrell. Deep domain confusion: Maximizing for domain invariance.arXiv preprint arXiv:1412.3474, 2014

    Pith/arXiv arXiv 2014

  41. [41]

    Deep coral: Correlation alignment for deep domain adaptation

    Baochen Sun and Kate Saenko. Deep coral: Correlation alignment for deep domain adaptation. InEuropean conference on computer vision, pages 443–450. Springer, 2016

    2016

  42. [42]

    Domain-adversarial training of neural networks

    Yaroslav Ganin, Evgeniya Ustinova, Hana Ajakan, Pascal Germain, Hugo Larochelle, François Laviolette, Mario March, and Victor Lempitsky. Domain-adversarial training of neural networks. Journal of machine learning research, 17(59):1–35, 2016

    2016

  43. [43]

    Adversarial discriminative domain adaptation

    Eric Tzeng, Judy Hoffman, Kate Saenko, and Trevor Darrell. Adversarial discriminative domain adaptation. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 7167–7176, 2017

    2017

  44. [44]

    Domain-adversarial training of self-attention-based networks for land cover classification using multi-temporal sentinel-2 satellite imagery.Remote Sensing, 13(13):2564, 2021

    Mauro Martini, Vittorio Mazzia, Aleem Khaliq, and Marcello Chiaberge. Domain-adversarial training of self-attention-based networks for land cover classification using multi-temporal sentinel-2 satellite imagery.Remote Sensing, 13(13):2564, 2021

    2021

  45. [45]

    Timematch: Unsuper- vised cross-region adaptation by temporal shift estimation.ISPRS Journal of Photogrammetry and Remote Sensing, 188:301–313, 2022

    Joachim Nyborg, Charlotte Pelletier, Sébastien Lefèvre, and Ira Assent. Timematch: Unsuper- vised cross-region adaptation by temporal shift estimation.ISPRS Journal of Photogrammetry and Remote Sensing, 188:301–313, 2022

    2022

  46. [46]

    Cross-sensor domain adaptation for high spatial resolution urban land-cover mapping: From airborne to spaceborne imagery.Remote Sensing of Environment, 277:113058, 2022

    Junjue Wang, Ailong Ma, Yanfei Zhong, Zhuo Zheng, and Liangpei Zhang. Cross-sensor domain adaptation for high spatial resolution urban land-cover mapping: From airborne to spaceborne imagery.Remote Sensing of Environment, 277:113058, 2022. 18

    2022

  47. [47]

    Emmanuel Capliez, Dino Ienco, Raffaele Gaetano, Nicolas Baghdadi, Adrien Hadj Salah, Matthieu Le Goff, and Florient Chouteau. Multisensor temporal unsupervised domain adaptation for land cover mapping with spatial pseudo-labeling and adversarial learning.IEEE Transactions on Geoscience and Remote Sensing, 61:1–16, 2023

    2023

  48. [48]

    Carbon emission limits required to satisfy future representative concentration pathways of greenhouse gases.Geophysical Research Letters, 38(5), 2011

    Vivek K Arora, JF Scinocca, GJ Boer, JR Christian, KL Denman, GM Flato, VV Kharin, WG Lee, and WJ Merryfield. Carbon emission limits required to satisfy future representative concentration pathways of greenhouse gases.Geophysical Research Letters, 38(5), 2011

    2011

  49. [49]

    SMHI, 2015

    Patrick Samuelsson, Stefan Gollvik, Marco Kupiainen, Ekaterina Kourzeneva, and Willem Jan van de Berg.The surface processes of the Rossby Centre regional atmospheric climate model (RCA4). SMHI, 2015

    2015

  50. [50]

    Coordinated global and regional climate modeling.Journal of Climate, 29(1):17–35, 2016

    JF Scinocca, VV Kharin, Y Jiao, MW Qian, M Lazare, L Solheim, GM Flato, S Biner, M Desgagne, and B Dugas. Coordinated global and regional climate modeling.Journal of Climate, 29(1):17–35, 2016

    2016

  51. [51]

    Ec-earth v2

    Wilco Hazeleger, X Wang, Camiel Severijns, S ¸ Stef˘anescu, R Bintanja, Andreas Sterl, Klaus Wyser, T Semmler, S Yang, B Van den Hurk, et al. Ec-earth v2. 2: description and validation of a new seamless earth system prediction model.Climate dynamics, 39(11):2611–2629, 2012

    2012

  52. [52]

    Climate model downscaling in central asia: a dynamical and a neural network approach.Geoscientific Model Development, 18(1):161–180, 2025

    Bijan Fallah, Masoud Rostami, Emmanuele Russo, Paula Harder, Christoph Menz, Peter Hoffmann, Iulii Didovets, and Fred F Hattermann. Climate model downscaling in central asia: a dynamical and a neural network approach.Geoscientific Model Development, 18(1):161–180, 2025

    2025

  53. [53]

    The earth system grid federation: An open infrastructure for access to distributed geospatial data.Future Generation Computer Systems, 36:400–417, 2014

    Luca Cinquini, Daniel Crichton, Chris Mattmann, John Harney, Galen Shipman, Feiyi Wang, Rachana Ananthakrishnan, Neill Miller, Sebastian Denvil, Mark Morgan, et al. The earth system grid federation: An open infrastructure for access to distributed geospatial data.Future Generation Computer Systems, 36:400–417, 2014

    2014

  54. [54]

    NA-CORDEX - North American component of the Coordi- nated Regional Downscaling Experiment, 2024

    Research Application Laboratory, National Center for Atmospheric Research, University Cor- poration for Atmospheric Research. NA-CORDEX - North American component of the Coordi- nated Regional Downscaling Experiment, 2024. URL https://gdex.ucar.edu/datasets/ d316009/

    2024

  55. [55]

    Rcp 8.5—a scenario of compara- tively high greenhouse gas emissions.Climatic change, 109(1):33, 2011

    Keywan Riahi, Shilpa Rao, V olker Krey, Cheolhung Cho, Vadim Chirkov, Guenther Fischer, Georg Kindermann, Nebojsa Nakicenovic, and Peter Rafaj. Rcp 8.5—a scenario of compara- tively high greenhouse gas emissions.Climatic change, 109(1):33, 2011

    2011

  56. [56]

    Christopher Daly, Michael Halbleib, Joseph I Smith, Wayne P Gibson, Matthew K Doggett, George H Taylor, Jan Curtis, and Phillip P Pasteris. Physiographically sensitive mapping of climatological temperature and precipitation across the conterminous united states.International Journal of Climatology: a Journal of the Royal Meteorological Society, 28(15):203...

    2031

  57. [57]

    A survey on transfer learning.IEEE Transactions on knowledge and data engineering, 22(10):1345–1359, 2009

    Sinno Jialin Pan and Qiang Yang. A survey on transfer learning.IEEE Transactions on knowledge and data engineering, 22(10):1345–1359, 2009

    2009

  58. [58]

    A theory of learning from different domains.Machine learning, 79(1): 151–175, 2010

    Shai Ben-David, John Blitzer, Koby Crammer, Alex Kulesza, Fernando Pereira, and Jen- nifer Wortman Vaughan. A theory of learning from different domains.Machine learning, 79(1): 151–175, 2010

    2010

  59. [59]

    Image super- resolution using very deep residual channel attention networks

    Yulun Zhang, Kunpeng Li, Kai Li, Lichen Wang, Bineng Zhong, and Yun Fu. Image super- resolution using very deep residual channel attention networks. InProceedings of the European conference on computer vision (ECCV), pages 286–301, 2018

    2018

  60. [60]

    Photo-realistic single image super-resolution using a generative adversarial network

    Christian Ledig, Lucas Theis, Ferenc Huszár, Jose Caballero, Andrew Cunningham, Alejandro Acosta, Andrew Aitken, Alykhan Tejani, Johannes Totz, Zehan Wang, et al. Photo-realistic single image super-resolution using a generative adversarial network. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 4681–4690, 2017. 19

    2017

  61. [61]

    Enhanced deep residual networks for single image super-resolution

    Bee Lim, Sanghyun Son, Heewon Kim, Seungjun Nah, and Kyoung Mu Lee. Enhanced deep residual networks for single image super-resolution. InProceedings of the IEEE conference on computer vision and pattern recognition workshops, pages 136–144, 2017

    2017

  62. [62]

    Residual dense network for image super-resolution

    Yulun Zhang, Yapeng Tian, Yu Kong, Bineng Zhong, and Yun Fu. Residual dense network for image super-resolution. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 2472–2481, 2018

    2018

  63. [63]

    U-net: Convolutional networks for biomedical image segmentation

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image segmentation. InInternational Conference on Medical image computing and computer-assisted intervention, pages 234–241. Springer, 2015

    2015

  64. [64]

    Real-time single image and video super-resolution using an efficient sub-pixel convolutional neural network

    Wenzhe Shi, Jose Caballero, Ferenc Huszár, Johannes Totz, Andrew P Aitken, Rob Bishop, Daniel Rueckert, and Zehan Wang. Real-time single image and video super-resolution using an efficient sub-pixel convolutional neural network. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 1874–1883, 2016

    2016

  65. [65]

    Deconvolution and checkerboard artifacts

    Augustus Odena, Vincent Dumoulin, and Chris Olah. Deconvolution and checkerboard artifacts. Distill, 1(10):e3, 2016

    2016

  66. [66]

    Accurate image super-resolution using very deep convolutional networks

    Jiwon Kim, Jung Kwon Lee, and Kyoung Mu Lee. Accurate image super-resolution using very deep convolutional networks. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 1646–1654, 2016

    2016

  67. [67]

    Deeply-recursive convolutional network for image super-resolution

    Jiwon Kim, Jung Kwon Lee, and Kyoung Mu Lee. Deeply-recursive convolutional network for image super-resolution. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 1637–1645, 2016

    2016

  68. [68]

    Spatio-temporal downscaling of climate data using convolutional and error-predicting neural networks.Frontiers in Climate, 3:656479, 2021

    Agon Serifi, Tobias Günther, and Nikolina Ban. Spatio-temporal downscaling of climate data using convolutional and error-predicting neural networks.Frontiers in Climate, 3:656479, 2021

    2021

  69. [69]

    A super-resolution framework for downscaling machine learning weather prediction toward 1-km air temperature

    Hyebin Park, Seonyoung Park, Daehyun Kang, and Jeong-Hwan Kim. A super-resolution framework for downscaling machine learning weather prediction toward 1-km air temperature. npj Climate and Atmospheric Science, 2026

    2026

  70. [70]

    Hydrologic implications of dynamical and statistical approaches to downscaling climate model outputs

    Andrew W Wood, Lai R Leung, Venkataramana Sridhar, and Dennis P Lettenmaier. Hydrologic implications of dynamical and statistical approaches to downscaling climate model outputs. Climatic change, 62(1):189–216, 2004

    2004

  71. [71]

    Probabilistic downscaling approaches: Application to wind cumulative distribution functions.Geophysical Research Letters, 36(11), 2009

    P-A Michelangeli, Matthieu Vrac, and Harilaos Loukos. Probabilistic downscaling approaches: Application to wind cumulative distribution functions.Geophysical Research Letters, 36(11), 2009

    2009

  72. [72]

    Deep learning downscaled high-resolution daily near surface meteorological datasets over east asia.Scientific Data, 10(1):890, 2023

    Hai Lin, Jianping Tang, Shuyu Wang, Shuguang Wang, and Guangtao Dong. Deep learning downscaled high-resolution daily near surface meteorological datasets over east asia.Scientific Data, 10(1):890, 2023

    2023

  73. [73]

    Patch-based image inpainting with generative adversarial net- works.arXiv preprint arXiv:1803.07422, 2018

    Ugur Demir and Gozde Unal. Patch-based image inpainting with generative adversarial net- works.arXiv preprint arXiv:1803.07422, 2018

    Pith/arXiv arXiv 2018

  74. [74]

    Generative adversarial nets.Advances in neural information processing systems, 27, 2014

    Ian J Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial nets.Advances in neural information processing systems, 27, 2014

    2014

  75. [75]

    Image-to-image translation with conditional adversarial networks

    Phillip Isola, Jun-Yan Zhu, Tinghui Zhou, and Alexei A Efros. Image-to-image translation with conditional adversarial networks. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 1125–1134, 2017

    2017

  76. [76]

    Learning to correct climate projection biases.Journal of Advances in Modeling Earth Systems, 13(10):e2021MS002509, 2021

    Baoxiang Pan, Gemma J Anderson, Andre Goncalves, Donald D Lucas, Celine JW Bonfils, Jiwoo Lee, Yang Tian, and Hsi-Yen Ma. Learning to correct climate projection biases.Journal of Advances in Modeling Earth Systems, 13(10):e2021MS002509, 2021

    2021

  77. [77]

    A reliable generative adversarial network approach for climate downscaling and weather genera- tion.Journal of Advances in Modeling Earth Systems, 17(1):e2024MS004668, 2025

    Neelesh Rampal, Peter B Gibson, Steven Sherwood, Gab Abramowitz, and Sanaa Hobeichi. A reliable generative adversarial network approach for climate downscaling and weather genera- tion.Journal of Advances in Modeling Earth Systems, 17(1):e2024MS004668, 2025. 20

    2025

  78. [78]

    Optuna: A next-generation hyperparameter optimization framework

    Takuya Akiba, Shotaro Sano, Toshihiko Yanase, Takeru Ohta, and Masanori Koyama. Optuna: A next-generation hyperparameter optimization framework. InProceedings of the 25th ACM SIGKDD international conference on knowledge discovery & data mining, pages 2623–2631, 2019

    2019

  79. [79]

    Image quality assessment: from error visibility to structural similarity.IEEE transactions on image processing, 13(4): 600–612, 2004

    Zhou Wang, Alan C Bovik, Hamid R Sheikh, and Eero P Simoncelli. Image quality assessment: from error visibility to structural similarity.IEEE transactions on image processing, 13(4): 600–612, 2004

    2004

  80. [80]

    Bias patterns and climate change signals in gcm-rcm model chains.Environmental Research Letters, 13(7): 074017, 2018

    Silje Lund Sørland, Christoph Schär, Daniel Lüthi, and Erik Kjellström. Bias patterns and climate change signals in gcm-rcm model chains.Environmental Research Letters, 13(7): 074017, 2018

    2018

Showing first 80 references.

This paper was first reviewed by grok-4.5 on July 11, 2026.