Pith. sign in

REVIEW 3 major objections 4 minor 160 references

A modified transformer-based diffusion model can both fill in and forecast hydrological time series, with weather covariates sharpening event-driven predictions.

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

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

T0 review · deepseek-v4-flash

2026-08-01 08:08 UTC pith:ZKRHM7KG

load-bearing objection A genuinely useful engineering adaptation of CSDI to hydrology, with a solid ablation study and honest limitations — but the headline claim of consistent superiority is contradicted by its own Table 16a, and the forecasting evaluation leaks future information via reanalysis covariates. the 3 major comments →

arxiv 2607.21200 v1 pith:ZKRHM7KG submitted 2026-07-23 stat.ML cs.LG

Transformer-based Diffusion models for Hydrological Time Series Probabilistic Imputation and Forecasting

Ferdinand Bhavsar (INRAE) , Lionel Benoit (INRAE) , Maxime Savatier (ANDRA) , Edith Gabriel (INRAE) This is my paper
classification stat.ML cs.LG
keywords diffusion modelsprobabilistic imputationhydrological time seriestransformerforecastingconditional score-based diffusionwater qualitymissing data
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.

Rivers are monitored at 4-hour resolution across six stations, but the sensors leave large gaps and the signals are strongly non-Gaussian and correlated across variables and sites. This paper argues that a single conditional diffusion model — built on the CSDI architecture and adapted to hydrological data — can handle both missing-value imputation and probabilistic forecasting in one network, and that it outperforms the classical and neural baselines tested on the real OPE dataset and nearly all baselines on a synthetic benchmark. The authors show that the custom model reproduces the data's daily and multi-day periodic structure and its between-station correlation pattern, and that feeding meteorological covariates into the model mainly helps forecasting, especially for rain-driven spikes such as a sudden water-level rise. If correct, this gives hydrologists one uncertainty-aware tool for gap-filling and short-term prediction, with the caveat that its forecast advantage currently assumes near-perfect knowledge of future weather.

Core claim

The central claim is that the proposed custom CSDI model — a transformer-based denoising diffusion network with added multi-scale convolutions, multiplicative-and-additive conditioning, root-mean-square normalization, and explicit Fourier encoding of day, week, and season cycles — outperforms all tested baselines on imputation and forecasting for the OPE dataset and almost all baselines on the synthetic dataset. The authors demonstrate that the model reproduces the target periodicities in simulations, reconstructs the inter-station correlation structure better than the original CSDI (though with weaker negative correlations), and, when augmented with weather covariates, successfully reproduc

What carries the argument

The central object is CSDI (Conditional Score-based Diffusion), a denoising diffusion model whose transformer backbone is trained to predict the noise added to missing time-series values while conditioning on observed values. The custom version adds: residual U-blocks with multi-scale convolutions to capture local and global temporal patterns; a conditioning mechanism with multiplicative scaling and additive offset; root-mean-square normalization for training stability; explicit Fourier encoding of periodicities (1 day, 7 days, 30 days, 90 days, 365 days); and an optional secondary input channel carrying meteorological covariates. During training, random masking simulates missingness, and th

Load-bearing premise

The headline forecasting results for the covariate-augmented model assume the model sees perfect future weather (reanalysis data) during evaluation; with real uncertain weather forecasts, the reported advantage over the no-covariate model could shrink or disappear.

What would settle it

Run the covariate-augmented model with operational numerical weather forecasts over the same forecast horizons instead of reanalysis weather; if the RMSE and CRPS advantage over the no-covariate model disappears, the forecasting benefit rests on perfect knowledge of future weather. A second check: evaluate the model on sequences longer than 15 days to see whether its periodic and low-frequency reproduction holds outside the short windows tested.

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

If this is right

  • A single diffusion model can serve both data imputation and probabilistic forecasting, removing the need for task-specific models at a monitoring network.
  • Weather covariates as conditioning can substantially improve forecasting of event-driven shifts such as rain-driven stream peaks, pointing toward a data-driven alternative to purely autoregressive hydrological models.
  • The model is well suited to imputing gaps of moderate size (around 100 synthetic timesteps) and short-term forecasting, but its advantage over classical Gaussian-process methods shrinks for longer gaps.
  • The reproduced between-station correlation structure supports multisite simulation, though negative cross-variable correlations remain underestimated, limiting some joint water-quality predictions.

Where Pith is reading between the lines

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

  • The forecasting gains of the covariate-augmented model were measured with reanalysis weather over the forecast horizon, so they are likely an upper bound: feeding the model real numerical weather forecasts, or ensembles of them, would show how much of the advantage survives operational uncertainty.
  • The acknowledged weakness with negative correlations suggests a direct architectural test: replacing the softmax attention with a kernel that permits negative weights, such as cosFormer, could improve joint water-quality modeling without retraining the whole pipeline.
  • A hybrid deployment that delegates long gaps to a statistical baseline could leverage the diffusion model's short-gap accuracy while compensating for its long-gap degradation.
  • Because training data were corrected for sensor drift, real-time deployment on raw uncorrected sensor inputs is expected to degrade simulation quality; quantifying that degradation with a raw-data test would clarify operational limits.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. This paper proposes a modified Conditional Score-based Diffusion Model (CSDI) for probabilistic imputation and forecasting of multivariate hydrological time series. The modifications — residual U-blocks, convolutional layers, StyleGAN-style conditioning with RMSNorm, and Fourier periodic encodings — are motivated by observed failures of vanilla CSDI on the OPE dataset, and a covariate-augmented version adds meteorological covariates. The model is evaluated on a synthetic hydrology-like dataset and on a real 6-station, 13-variable water-quality/quantity dataset, against GP, original CSDI, U-Net, TSMixer, and NHits baselines, using RMSE, MAE, SMAPE, and CRPS, plus PSD and correlation-matrix diagnostics. The authors claim consistent superiority of the custom CSDI architecture, with covariate augmentation helping forecasting.

Significance. If the claims hold, this is a useful applied contribution: a single probabilistic model for both imputation and forecasting of multivariate, missing-data-heavy hydrological time series, with credible uncertainty bands and code availability. The step-by-step ablations (Appendix D) and the explicit synthetic generator (Appendix E) are valuable for reproducibility. The methodological novelty is incremental — the architecture is a composition of published components — and the real-data evaluation is limited; furthermore, the headline 'consistently outperforms all baselines' claim is internally contradicted by Table 16a. The contribution remains publishable after the claims are made metric-specific and the oracle-covariate issue is clearly labeled.

major comments (3)
  1. [§4.3, Table 16a] The claim 'across all metrics, the custom CSDI architecture (with and without covariates) consistently outperforms all baselines' is contradicted by Table 16a. At 25% missing values, the non-covariate Custom model has RMSE 0.1676±0.0049, while GP has 0.1609±0.0008; the custom model is therefore not strictly better than all baselines on this metric. Since the claim appears in Section 4.3 and is echoed elsewhere, it must be revised to a metric-specific or majority-of-metrics formulation, or the table corrected.
  2. [§3.1, §4.1, §5] The covariate-augmented model is trained and evaluated with SAFRAN reanalysis covariates over the full forecast horizon. Reanalysis is a posterior product, not an available forecast, so the forecasting comparison gives the augmented model perfect future information that covariate-free baselines do not receive. The Discussion acknowledges this ('does not fully correspond to an operational forecasting setting'), but the headline forecasting advantage in Figures 8-9 remains operational unfalsified. Please relabel these results as an oracle/perfect-information upper bound, or add an experiment with lagged or forecast-error-perturbed covariates.
  3. [§4.1, §4.3] The real-data evaluation uses a test partition of only 28 elementary windows (and 20 validation windows), each 24 timesteps long. This is small for claiming 'consistently outperforms' across all metrics and horizons. Report per-window variability, confidence intervals, or a bootstrap over windows, and ensure the claim is not stronger than the effective sample size supports.
minor comments (4)
  1. [§4.1] The text says 'See Appendix 4 for an overview of the evaluation framework' but the detailed metrics are in Appendix F; the cross-reference should be corrected.
  2. [§2.2, Eq. (8)] Equation (8) has garbled formatting: the weighting term and the norm symbols are unclear. Please re-typeset and verify the expression.
  3. [§4.3, Figure 11] The Figure 11 caption says 'simulating 15 days given 16 days', whereas the evaluation setting in §4.1 describes 24-timestep windows and horizons of 6-12 timesteps at 4-hour resolution. Clarify whether Figure 11 uses a different experiment; otherwise this is an inconsistency.
  4. [Throughout] The name 'U-NET' is used inconsistently (U-Net, U-NET). Also 'Guigues et al., ress' in the references should be 'in press'.

Circularity Check

0 steps flagged

No significant circularity: the paper is an empirical ML evaluation with a standard diffusion objective; fixed Fourier features and non-load-bearing self-citation do not create a circular derivation.

full rationale

This is an empirical machine-learning benchmark, not a derivation chain, so the main circularity-by-construction failure mode does not arise. The diffusion training objective (Eq. 9) and the CSDI conditional loss (Eq. 12) are standard score-matching losses; the reported imputation and forecasting numbers are obtained by training on training data and evaluating on masked held-out test windows, so no reported metric is defined in terms of the fitted parameters. The Fourier timestep encoding in Section 3.1 is explicitly a fixed, hand-selected input representation: 'we use cycles of 1 day, 7 days, 30 days, 90 days and 365 days', described as 'predefined cycle lengths'; it is not fitted to the test labels, and the synthetic-data evaluation also reproduces a 48h period that is not among the injected Fourier cycles. The only self-citation, Bhavsar et al. (2024), is used to support the generic statement that GAN training 'often suffers from stability issues'; it is not load-bearing for any central claim. The covariate-augmented forecasting setup uses SAFRAN reanalysis over the forecast horizon, and the Discussion explicitly admits that this 'does not fully correspond to an operational forecasting setting'; that is an evaluation-validity limitation, not a circular step. Similarly, Table 16a at 25% missing values reports Custom RMSE 0.1676±0.0049 versus GP RMSE 0.1609±0.0008, which contradicts the text's claim that the custom architecture 'consistently outperforms all baselines'; this is an internal inconsistency or correctness issue, not a circular argument. No prediction in the paper reduces by construction to its input, so the circularity score is 0.

Axiom & Free-Parameter Ledger

1 free parameters · 4 axioms · 0 invented entities

The paper introduces no new physical or mathematical entities. The central claim rests on hand-selected Fourier cycles and standard diffusion background, plus domain assumptions about data quality and covariate availability. No fitted parameters beyond network weights, which are not counted here.

free parameters (1)
  • Fourier cycle lengths = [1, 7, 30, 90, 365] days
    Hand-selected cycle lengths for the Fourier time encoding (Section 3.1). These encode diurnal, weekly, monthly, seasonal, and annual periodicities; the central claim depends on this design choice, though it is not fitted to the data.
axioms (4)
  • standard math Diffusion reverse process can be approximated by a noise-prediction network (Ho et al. 2020)
    Section 2.2 relies on the standard DDPM reparameterization and the objective in Eq. (9). This is a well-established mathematical framework.
  • domain assumption OPEs data quality corrections (LNE/Andra) make the corrected measurements a reliable ground truth
    Section 2.1 and the Discussion state that the data were corrected for sensor drift and malfunction. The evaluation treats the corrected data as true observations; if the corrections are biased, the results could be misleading.
  • domain assumption SAFRAN reanalysis is a valid proxy for meteorological forecasting covariates
    Section 3.1 and 4.1 use SAFRAN reanalysis as covariates for the forecast horizon. The authors acknowledge in the Discussion that operational forecasts would be uncertain, so the forecasting results implicitly assume perfect future weather knowledge.
  • domain assumption The synthetic dataset mimics the relevant hydrological dynamics of the real system
    Appendix E describes the synthetic generator. The authors use it to test the model under complete-data settings, but conclusions drawn from this synthetic process may not transfer to real hydrology.

pith-pipeline@v1.3.0-alltime-deepseek · 23959 in / 8435 out tokens · 90445 ms · 2026-08-01T08:08:56.779677+00:00 · methodology

0 comments
read the original abstract

The modeling of hydrometeorological time series with limited observations is a key challenge in the monitoring of hydro-systems and water resources, as well as for flood or drought risk assessment. Due to the high variability of the underlying processes and the sparsity of available measurements, traditional statistical approaches often struggle to accurately represent their dynamics. In this context, recent advances in deep learning offer a promising direction for improving the representation and generation of complex temporal processes sampled at several observation sites. This study investigates the application of transformer-based diffusion models to the simulation and reconstruction of hydrological time series. The proposed framework is applied to the joint modeling of water quantity and quality at six sites spread across three adjacent headwater catchments located in North-East France on a limestone plateau covered by forests and field crops. The model is calibrated and validated using available observational data, which has been quality controlled and corrected for sensor drift and malfunction through collaborative efforts by LNE metrology expertise and Andra monthly quality control over more than 15 years. Its performance is compared with several established baseline approaches commonly used for time series modeling. Quantitative evaluation metrics are employed to assess the ability of the proposed method to reproduce key temporal characteristics of the observed signals in two settings: the imputation of incomplete time series and the forecasting of upcoming hydrological conditions. Results support the effectiveness of the transformer-based approach and highlight its capacity to capture and simulate the complex patterns present in hydrological data. In particular, the results indicate that diffusion models can efficiently sample realistic time series distributions under observation settings with variable missing data for both forecasting and imputation.

Figures

Figures reproduced from arXiv: 2607.21200 by Edith Gabriel (INRAE), Ferdinand Bhavsar (INRAE), Lionel Benoit (INRAE), Maxime Savatier (ANDRA).

Figure 1
Figure 1. Figure 1: OPE hydrological monitoring network (Marne catchment - North-East France): (a) Site [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Illustration of the diffusion process, where [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Synthetic data generated to test models. In blue we have a dynamic component, in orange a [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: The evaluation setting of our diffusion model for imputation (top) and forecasting (bottom). [PITH_FULL_IMAGE:figures/full_fig_p011_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Comparison of model performance on the synthetic dataset across missing ratios and [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Power spectral density comparison between synthetic data (blue) and the mean of 100 [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Illustration of imputation results over 24 timesteps (4 days), on one sample of the test [PITH_FULL_IMAGE:figures/full_fig_p014_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Performance comparison across missing ratios and prediction horizons on the OPE dataset. [PITH_FULL_IMAGE:figures/full_fig_p015_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Compared forecasting results for four tested models, Custom with covariates (CovAug), [PITH_FULL_IMAGE:figures/full_fig_p016_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Comparison between the Pearson covariance matrix of the training dataset (top-left) [PITH_FULL_IMAGE:figures/full_fig_p017_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Visualization of the distribution of predicted values with respect to their real value, without [PITH_FULL_IMAGE:figures/full_fig_p018_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Comparison between the original CSDI architecture and the modified architecture. The [PITH_FULL_IMAGE:figures/full_fig_p024_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Ablation study: predicted water temperature forecasts (15-day horizon, conditioned on [PITH_FULL_IMAGE:figures/full_fig_p026_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Inter-variable correlation matrices estimated, real-data (left) and our custom architecture [PITH_FULL_IMAGE:figures/full_fig_p027_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Synthetic dataset performance comparison. Best values are highlighted in green; second [PITH_FULL_IMAGE:figures/full_fig_p029_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: OPE dataset performance comparison. Best values are highlighted in green; second-best in [PITH_FULL_IMAGE:figures/full_fig_p030_16.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

160 extracted references · 8 canonical work pages

  1. [1]

    Alcaraz, J. M. L. and N. Strodthoff (2023). Diffusion-based time series imputation and forecasting with structured state space models

    2023

  2. [2]

    Allcroft, D. J. and C. A. Glasbey (2003, 09). A latent gaussian markov random-field model for spatiotemporal rainfall disaggregation. Journal of the Royal Statistical Society Series C: Applied Statistics\/ 52\/ (4), 487--498

    2003

  3. [3]

    Alornyo, S. A., W. R. McGillis, and P. J. Culligan (2026). From sensors to simulations: How ai is transforming water monitoring, management, and policy. Water\/ 18\/ (14)

    2026

  4. [5]

    Chintala, and L

    Arjovsky, M., S. Chintala, and L. Bottou (2017). Wasserstein generative adversarial networks. In International conference on machine learning , pp.\ 214--223. PMLR

    2017

  5. [6]

    Cho, and Y

    Bahdanau, D., K. Cho, and Y. Bengio (2015). Neural machine translation by jointly learning to align and translate. In Proceedings of the 3rd International Conference on Learning Representations (ICLR 2015)

    2015

  6. [7]

    Balaji, and Z

    Bansal, A., K. Balaji, and Z. Lalani (2025). Temporal encoding strategies for energy time series prediction

    2025

  7. [8]

    Janati, G

    Bedin, L., Y. Janati, G. Victorino Cardoso, J. Duchateau, R. Dubois, and E. Moulines (2025, 06). Reconstructing ecg from indirect signals: a denoising diffusion approach. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences\/ 383\/ (2299), 20240330

    2025

  8. [9]

    Desassis, F

    Bhavsar, F., N. Desassis, F. Ors, and T. Romary (2024). A stable deep adversarial learning approach for geological facies generation. Computers & Geosciences\/ 190 , 105638

    2024

  9. [10]

    Donahue, and K

    Brock, A., J. Donahue, and K. Simonyan (2019). Large scale GAN training for high fidelity natural image synthesis

    2019

  10. [11]

    Brunner, M. I., A. B\'ardossy, and R. Furrer (2019). Technical note: Stochastic simulation of streamflow time series using phase randomization. Hydrology and Earth System Sciences\/ 23\/ (8), 3175--3187

    2019

  11. [12]

    C. M., N. and N. Thangarasu (2025, 11). Enhancing dam inflow forecasting using hybrid deep learning and diffusion models: a case study on the malampuzha dam. Water Practice and Technology\/ 20\/ (11), 2276--2293

    2025

  12. [13]

    Challu, C., K. G. Olivares, B. N. Oreshkin, F. Garza, M. Mergenthaler-Canseco, and A. Dubrawski (2022). N-hits: Neural hierarchical interpolation for time series forecasting

    2022

  13. [14]

    Chen, C., F. Wang, Z. Wang, D. Zhang, and L. Xiang (2025, June). A novel flood forecasting model based on TimeGAN for data-sparse basins. Stochastic Environmental Research and Risk Assessment\/ 39\/ (6), 2267--2280

    2025

  14. [15]

    Chen, S.-A., C.-L. Li, N. Yoder, S. O. Arik, and T. Pfister (2023). Tsmixer: An all-mlp architecture for time series forecasting

    2023

  15. [16]

    Cybenko, G. (1989). Approximation by superpositions of a sigmoidal function. Mathematics of control, signals and systems\/ 2\/ (4), 303--314

    1989

  16. [17]

    Dhariwal, P. and A. Nichol (2021). Diffusion models beat gans on image synthesis. In M. Ranzato, A. Beygelzimer, Y. Dauphin, P. Liang, and J. W. Vaughan (Eds.), Advances in Neural Information Processing Systems , Volume 34, pp.\ 8780--8794. Curran Associates, Inc

    2021

  17. [18]

    Sartran, A

    Dieleman, S., L. Sartran, A. Roshannai, N. Savinov, Y. Ganin, P. H. Richemond, A. Doucet, R. Strudel, C. Dyer, C. Durkan, C. Hawthorne, R. Leblond, W. Grathwohl, and J. Adler (2022). Continuous diffusion for categorical data. ArXiv\/ abs/2211.15089

    Pith/arXiv arXiv 2022

  18. [19]

    Fang, and C

    Feng, D., K. Fang, and C. Shen (2020, September). Enhancing streamflow forecast and extracting insights using long‐short term memory networks with data integration at continental scales. Water Resources Research\/ 56\/ (9)

    2020

  19. [20]

    Pouget-Abadie, M

    Goodfellow, I., J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, and Y. Bengio (2014). Generative adversarial nets. In Advances in neural information processing systems , pp.\ 2672--2680

    2014

  20. [21]

    Goodfellow, I. J., Y. Bengio, and A. Courville (2016). Deep Learning . Cambridge, MA, USA: MIT Press. http://www.deeplearningbook.org

    2016

  21. [22]

    Gu, S., D. Chen, J. Bao, F. Wen, B. Zhang, D. Chen, L. Yuan, and B. Guo (2022, June). Vector quantized diffusion model for text-to-image synthesis. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pp.\ 10696--10706

    2022

  22. [23]

    Guardiano, F. and M. Srivastava (1993). Multivariate geostatistics: Beyond bivariate moments. In A. Soares (Ed.), Geostatistics Troia , pp.\ 133--144. Dordrecht: Kluwer Academic Publications

    1993

  23. [24]

    Savatier, and B

    Guigues, N., M. Savatier, and B. Lalère (in press). Quantifying measurement uncertainty in real time continuous water quality monitoring: methodological approaches and implications for data interpretation. Environmental Science: Processes & Impacts\/ . In press

  24. [25]

    Lintern, J

    Guo, D., A. Lintern, J. A. Webb, D. Ryu, U. Bende-Michl, S. Liu, and A. W. Western (2020). A data-based predictive model for spatiotemporal variability in stream water quality. Hydrology and Earth System Sciences\/ 24\/ (2), 827--847

    2020

  25. [26]

    He, J., Y.-J. Chen, A. Jafari, A. Idamekorala, and G. Fox (2025). Deep learning foundation and pattern models: Challenges in hydrological time series

    2025

  26. [27]

    Zhang, S

    He, K., X. Zhang, S. Ren, and J. Sun (2015). Deep residual learning for image recognition

    2015

  27. [28]

    Lässig, S

    Herzen, J., F. Lässig, S. G. Piazzetta, T. Neuer, L. Tafti, G. Raille, T. V. Pottelbergh, M. Pasieka, A. Skrodzki, N. Huguenin, M. Dumonal, J. Kościsz, D. Bader, F. Gusset, M. Benheddi, C. Williamson, M. Kosinski, M. Petrik, and G. Grosch (2022). Darts: User-friendly modern machine learning for time series. Journal of Machine Learning Research\/ 23\/ (124), 1--6

    2022

  28. [30]

    Hochreiter, S. and J. Schmidhuber (1997, 12). Long short-term memory. Neural computation\/ 9 , 1735--80

    1997

  29. [31]

    Hu, J., P. Wang, D. Li, and S. Liu (2024). A long-term multivariate time series prediction model for dissolved oxygen. Ecological Informatics\/ 82 , 102695

    2024

  30. [32]

    Noutahi, J

    Jones, C., E. Noutahi, J. Hartford, and C. Eastwood (2026). Elucidating the design space of flow matching for cellular microscopy

    2026

  31. [33]

    Laine, and T

    Karras, T., S. Laine, and T. Aila (2019). A style-based generator architecture for generative adversarial networks. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pp.\ 4401--4410

    2019

  32. [34]

    Kawar, B., S. Zada, O. Lang, O. Tov, H. Chang, T. Dekel, I. Mosseri, and M. Irani (2023, June). Imagic: Text-based real image editing with diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pp.\ 6007--6017

    2023

  33. [35]

    Kingma, D. P. and M. Welling (2014). Auto-encoding variational bayes

    2014

  34. [36]

    Klotz, C

    Kratzert, F., D. Klotz, C. Brenner, K. Schulz, and M. Herrnegger (2018). Rainfall--runoff modelling using long short-term memory (lstm) networks. Hydrology and Earth System Sciences\/ 22\/ (11), 6005--6022

    2018

  35. [37]

    Li, C., X. Ling, Y. Xue, W. Luo, L. Zhu, F. Qin, Y. Zhou, and Y. Huang (2024). Precipitation nowcasting using diffusion transformer with causal attention. IEEE Transactions on Geoscience and Remote Sensing\/ 62 , 1--16

    2024

  36. [38]

    Renard, and J

    Mariethoz, G., P. Renard, and J. Straubhaar (2010). The direct sampling method to perform multiple-point geostatistical simulations. Water Resources Research\/ 46\/ (11)

    2010

  37. [40]

    Royer, D

    Moiroux-Arvis, L., L. Royer, D. Sarramia, G. De Sousa, A. Claude, D. Latour, E. Roussel, O. Voldoire, P. Chardon, R. Vanda \"e le, et al. (2023). Connecsens, a versatile iot platform for environment monitoring: Bring water to cloud. Sensors\/ 23\/ (6), 2896

    2023

  38. [42]

    Benoit, and D

    Obakrim, S., L. Benoit, and D. Allard (2025). A multivariate and space-time stochastic weather generator using a latent gaussian framework. Stochastic Environmental Research and Risk Assessment\/ 39\/ (9), 3677--3701

    2025

  39. [43]

    Ou, Z., C. Nai, B. Pan, Y. Zheng, C. Shen, P. Jiang, X. Liu, Q. Tang, W. Li, and M. Pan (2025, July). Probabilistic diffusion models advance extreme flood forecasting. Geophysical Research Letters\/ 52\/ (15)

    2025

  40. [44]

    Peebles, W. and S. Xie (2023). Scalable diffusion models with transformers

    2023

  41. [45]

    Zhang, C

    Qin, X., Z. Zhang, C. Huang, M. Dehghan, O. R. Zaiane, and M. Jagersand (2020). U2-net: Going deeper with nested u-structure for salient object detection. Pattern recognition\/ 106 , 107404

    2020

  42. [46]

    Qin, Z., W. Sun, H. Deng, D. Li, Y. Wei, B. Lv, J. Yan, L. Kong, and Y. Zhong (2022). cosformer: Rethinking softmax in attention

    2022

  43. [47]

    Le Moigne, Y

    Quintana-Segui, P., P. Le Moigne, Y. Durand, E. Martin, F. Habets, M. Baillon, C. Canellas, L. Franchisteguy, and S. Morel (2008). Analysis of near-surface atmospheric variables: V alidation of the SAFRAN analysis over F rance. Journal of Applied Meteorology and Climatology\/ 47\/ (1), 92--107

    2008

  44. [48]

    Saeed, M

    Rashid, M., A. Saeed, M. Khalid, A. Murtaza, and M. Waqar Saleem (2026). The transformative role of artificial intelligence in water resources engineering: A comprehensive review. Environmental Modelling & Software\/ 197 , 106857

    2026

  45. [49]

    Rattanatheerawon, W. and R. Fooprateepsiri (2025). Diffusion-driven time-series forecasting to support sustainable river ecosystems and sdg-aligned water-resource governance in thailand. Sustainability\/ 17\/ (22)

    2025

  46. [50]

    Cromwell, B

    Ren, H., E. Cromwell, B. Kravitz, and X. Chen (2022). Technical note: Using long short-term memory models to fill data gaps in hydrological monitoring networks. Hydrology and Earth System Sciences\/ 26\/ (7), 1727--1743

    2022

  47. [51]

    Fischer, and T

    Ronneberger, O., P. Fischer, and T. Brox (2015). U-net: Convolutional networks for biomedical image segmentation. In Medical Image Computing and Computer-Assisted Intervention--MICCAI 2015: 18th International Conference, Munich, Germany, October 5-9, 2015, Proceedings, Part III 18 , pp.\ 234--241. Springer

    2015

  48. [52]

    Ruiz, N., Y. Li, V. Jampani, Y. Pritch, M. Rubinstein, and K. Aberman (2023, June). Dreambooth: Fine tuning text-to-image diffusion models for subject-driven generation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pp.\ 22500--22510

    2023

  49. [53]

    Saharia, C., W. Chan, H. Chang, C. Lee, J. Ho, T. Salimans, D. Fleet, and M. Norouzi (2022a). Palette: Image-to-image diffusion models. In ACM SIGGRAPH 2022 Conference Proceedings , SIGGRAPH '22, New York, NY, USA. Association for Computing Machinery

    2022

  50. [54]

    Saharia, C., W. Chan, H. Chang, C. A. Lee, J. Ho, T. Salimans, D. J. Fleet, and M. Norouzi (2022b). Palette: Image-to-image diffusion models

  51. [55]

    Shevnina, E. and A. Silaev (2019). The probabilistic hydrological marcs ^HYDRO (the markov chain system) model: its structure and core version 0.2. Geoscientific Model Development\/ 12\/ (7), 2767--2780

    2019

  52. [56]

    Sohl-Dickstein, J., E. A. Weiss, N. Maheswaranathan, and S. Ganguli (2015). Deep unsupervised learning using nonequilibrium thermodynamics

    2015

  53. [57]

    Strebelle, S. and A. G. Journel (2000). Sequential Simulation Drawing Structures from Training Images . Dissertation

    2000

  54. [58]

    Tashiro, Y., J. Song, Y. Song, and S. Ermon (2021). Csdi: Conditional score-based diffusion models for probabilistic time series imputation. In M. Ranzato, A. Beygelzimer, Y. Dauphin, P. Liang, and J. W. Vaughan (Eds.), Advances in Neural Information Processing Systems , Volume 34, pp.\ 24804--24816. Curran Associates, Inc

    2021

  55. [59]

    van Wijk , W

    Teurlincx, S., D. van Wijk , W. M. Mooij, J. J. Kuiper, I. Huttunen, R. J. Brederveld, M. Chang, J. H. Janse, B. Woodward, F. Hu, and A. B. Janssen (2019). A perspective on water quality in connected systems: modelling feedback between upstream and downstream transport and local ecological processes. Current Opinion in Environmental Sustainability\/ 40 , ...

    2019

  56. [60]

    (2016, 03)

    Tunnicliffe Wilson, G. (2016, 03). Time series analysis: Forecasting and control,5th edition, by george e. p. box, gwilym m. jenkins, gregory c. reinsel and greta m. ljung, 2015. published by john wiley and sons inc., hoboken, new jersey, pp. 712. isbn: 978-1-118-67502-1. Journal of Time Series Analysis\/ 37 , n/a--n/a

    2016

  57. [61]

    Shazeer, N

    Vaswani, A., N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, and I. Polosukhin (2023). Attention is all you need

    2023

  58. [62]

    Studer, and J

    von Freyberg, J., B. Studer, and J. W. Kirchner (2017). A lab in the field: high-frequency analysis of water quality and stable isotopes in stream water and precipitation. Hydrology and Earth System Sciences\/ 21\/ (3), 1721--1739

    2017

  59. [63]

    Zhang, S

    Wang, Q., J. Zhang, S. Song, and Z. Zhang (2014). Attentional neural network: Feature selection using cognitive feedback. In Advances in Neural Information Processing Systems (NIPS 2014)

    2014

  60. [64]

    Zhang, P

    Wang, X., H. Zhang, P. Wang, Y. Zhang, B. Wang, Z. Zhou, and Y. Wang (2023). An observed value consistent diffusion model for imputing missing values in multivariate time series. In Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , KDD '23, New York, NY, USA, pp.\ 2409–2418. Association for Computing Machinery

    2023

  61. [65]

    Kojima (2022)

    Weilisi and T. Kojima (2022). Investigation of hyperparameter setting of a long short-term memory model applied for imputation of missing discharge data of the daihachiga river. Water\/ 14\/ (2)

    2022

  62. [66]

    Wang, and X

    Yang, C., T. Wang, and X. Yan (2023). Ddmt: Denoising diffusion mask transformer models for multivariate time series anomaly detection

    2023

  63. [67]

    Yang, D., J. Yu, H. Wang, W. Wang, C. Weng, Y. Zou, and D. Yu (2023). Diffsound: Discrete diffusion model for text-to-sound generation. IEEE/ACM Transactions on Audio, Speech, and Language Processing\/ 31 , 1720--1733

    2023

  64. [68]

    Yuan, X. and Y. Qiao (2024). Diffusion-ts: Interpretable diffusion for general time series generation

    2024

  65. [69]

    Zhang, B. and R. Sennrich (2019). Root mean square layer normalization . Red Hook, NY, USA: Curran Associates Inc

    2019

  66. [70]

    Zhang, H., J. Zhou, Y. Lu, M. Guo, P. Wang, L. Shen, and Q. Qu (2024). The emergence of reproducibility and generalizability in diffusion models

    2024

  67. [71]

    Rao, and M

    Zhang, L., A. Rao, and M. Agrawala (2023, October). Adding conditional control to text-to-image diffusion models. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , pp.\ 3836--3847

    2023

  68. [72]

    Zhu, J.-Y., T. Park, P. Isola, and A. A. Efros (2020). Unpaired image-to-image translation using cycle-consistent adversarial networks

    2020

  69. [73]

    Generative Adversarial Networks , volume =

    Goodfellow, Ian and Pouget-Abadie, Jean and Mirza, Mehdi and Xu, Bing and Warde-Farley, David and Ozair, Sherjil and Courville, Aaron and Bengio, Yoshua , year =. Generative Adversarial Networks , volume =. Advances in Neural Information Processing Systems , doi =

  70. [74]

    Denoising Diffusion Probabilistic Models , url =

    Ho, Jonathan and Jain, Ajay and Abbeel, Pieter , booktitle =. Denoising Diffusion Probabilistic Models , url =

  71. [75]

    Diffusion Models Beat GANs on Image Synthesis , url =

    Dhariwal, Prafulla and Nichol, Alexander , booktitle =. Diffusion Models Beat GANs on Image Synthesis , url =

  72. [76]

    Time Series Analysis: Forecasting and Control,5th Edition, by George E

    Tunnicliffe Wilson, Granville , year =. Time Series Analysis: Forecasting and Control,5th Edition, by George E. P. Box, Gwilym M. Jenkins, Gregory C. Reinsel and Greta M. Ljung, 2015. Published by John Wiley and Sons Inc., Hoboken, New Jersey, pp. 712. ISBN: 978-1-118-67502-1 , volume =. Journal of Time Series Analysis , doi =

    2015

  73. [77]

    Can a Training Image Be a Substitute for a Random Field Model? , volume =

    Emery, Xavier and Lantuéjoul, Christian , year =. Can a Training Image Be a Substitute for a Random Field Model? , volume =. Mathematical Geosciences , doi =

  74. [78]

    Jha and Beatrice M

    Alessandro Comunian and Sanjeev K. Jha and Beatrice M. S. Giambastiani and Gregoire Mariethoz and Bryce F. J. Kelly , title =. Mathematical Geosciences , year =. doi:10.1007/s11004-013-9505-y , url =

    work page doi:10.1007/s11004-013-9505-y

  75. [79]

    and Srivastava, M

    Guardiano, F. and Srivastava, M. , title =. Geostatistics Troia , editor =. 1993 , publisher =

    1993

  76. [80]

    , title =

    Strebelle, Sebastien and Journel, Andree G. , title =. 2000 , type =

    2000

  77. [81]

    1Introduction

    Matheron, Georges and Pawlowsky-Glahn, Vera and Serra, Jean , isbn =. 1Introduction. Matheron's Theory of Regionalised Variables. 2019 , month =. doi:10.1093/oso/9780198835660.003.0001 , url =

    arXiv 2019

  78. [82]

    J. P. Chil\`es and P. Delfiner. Geostatistics, M odeling S patial U ncertainty

  79. [83]

    Armstrong , title =

    M. Armstrong , title =. Journal of the International Association for Mathematical Geology , year =. doi:10.1007/BF01036241 , url =

    work page doi:10.1007/bf01036241

  80. [84]

    arXiv preprint arXiv:1701.04862 , year=

    Towards principled methods for training generative adversarial networks , author=. arXiv preprint arXiv:1701.04862 , year=

    Pith/arXiv arXiv

Showing first 80 references.