pith. machine review for the scientific record. sign in

arxiv: 2605.06678 · v1 · submitted 2026-04-22 · 💻 cs.LG · q-fin.RM· stat.AP

Recognition: no theorem link

A Wasserstein GAN-based climate scenario generator for risk management and insurance: the case of soil subsidence

Antoine Heranval (BioSP) , Olivier Lopez (CREST) , Didier Ngatcha , Daniel Nkameni (CREST)
Authors on Pith no claims yet

Pith reviewed 2026-05-11 01:26 UTC · model grok-4.3

classification 💻 cs.LG q-fin.RMstat.AP
keywords Wasserstein GANclimate scenario generationsoil wetness indexdrought riskinsuranceFranceconditional GANrisk management
0
0 comments X

The pith

A conditional Wasserstein GAN called SwiGAN generates sequences of future soil wetness maps to simulate drought propagation in France through 2050.

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

The paper develops an AI method to create plausible future trajectories of the Soil Wetness Index, a measure of drought severity used in French insurance. Drought drives roughly 30 percent of indemnities under the national natural catastrophe scheme, and current regulations focus only on one-year horizons while disaster costs rise sharply. By training on historical data, the model produces spatio-temporal maps conditioned on climate scenarios for an exposed region. This gives insurers and risk managers concrete sequences they can use to test long-term strategies. The approach is presented as extendable to other climate hazards and actuarial tasks.

Core claim

SwiGAN, a conditional Wasserstein GAN, learns to generate realistic future sequences of SWI maps that reproduce drought propagation patterns up to 2050 under climate change scenarios, thereby supplying a data-driven generator of climate trajectories for use in risk management and insurance pricing.

What carries the argument

SwiGAN, a conditional Wasserstein GAN that takes climate scenario inputs and outputs sequences of SWI maps, thereby capturing the spatial and temporal evolution of drought.

If this is right

  • Insurers gain medium- to long-term scenario sets that extend beyond the one-year Solvency II horizon.
  • The generated maps allow direct quantification of how drought exposure evolves for a given region through 2050.
  • The same conditional-GAN structure can be retrained on other climatic indices to support multi-peril scenario generation.
  • Actuarial teams can feed the synthetic trajectories into economic models for pricing and capital planning.

Where Pith is reading between the lines

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

  • The generated trajectories could be combined with economic loss models to estimate future indemnity costs under different climate pathways.
  • Validation against independent regional climate models would test whether the GAN preserves physically plausible drought statistics.
  • The framework might be applied to other countries that maintain similar soil-moisture or drought indices for insurance purposes.
  • Regulators could explore using such generators to stress-test portfolios over multi-decade horizons.

Load-bearing premise

A model trained exclusively on historical SWI observations can produce future trajectories whose drought dynamics remain realistic once climate conditions depart from the training period.

What would settle it

Compare SwiGAN-generated SWI sequences for a held-out recent decade against actual satellite or ground measurements from that same decade to check whether the simulated drought patterns match observed frequency, duration, and spatial spread.

Figures

Figures reproduced from arXiv: 2605.06678 by Antoine Heranval (BioSP), Daniel Nkameni (CREST), Didier Ngatcha, Olivier Lopez (CREST).

Figure 1
Figure 1. Figure 1: Description of the functioning of the proposed stochastic weather generator. [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Architecture of the UNet generator. This generator takes covariates and a random [PITH_FULL_IMAGE:figures/full_fig_p011_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Architecture of the discriminator. Both the generated maps and the real maps are [PITH_FULL_IMAGE:figures/full_fig_p013_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Socio-demographic and economic characteristics of the Grand Est region of France. [PITH_FULL_IMAGE:figures/full_fig_p015_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: General architecture of the proposed forecasting methodology. [PITH_FULL_IMAGE:figures/full_fig_p018_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Values of the validation metrics in Table 2 for the final SwiGAN model. [PITH_FULL_IMAGE:figures/full_fig_p019_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: The figure shows that all trajectories generated by SwiGAN follow the trend of the real [PITH_FULL_IMAGE:figures/full_fig_p020_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Spatial distribution of the mean trajectory of SWI values generated under RCP 4.5 [PITH_FULL_IMAGE:figures/full_fig_p022_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Monthly SWI maps generated by SwiGAN under RCP 4.5. [PITH_FULL_IMAGE:figures/full_fig_p024_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Shapley values used to assess the importance of covariates in explaining the variability [PITH_FULL_IMAGE:figures/full_fig_p025_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Spatial importance of pixels in the prediction of SWI values for three selected pixels. [PITH_FULL_IMAGE:figures/full_fig_p026_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Evolution of commune eligibility for the recognition of a state of natural catastrophe [PITH_FULL_IMAGE:figures/full_fig_p029_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Future insurance damage costs due to drought-induced soil subsidence under RCP [PITH_FULL_IMAGE:figures/full_fig_p031_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Additional information on the Grand Est region of France [PITH_FULL_IMAGE:figures/full_fig_p033_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Spatial distribution of the mean trajectory of SWI values generated under RCP 8.5 [PITH_FULL_IMAGE:figures/full_fig_p034_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Monthly SWI maps generated by SwiGAN under RCP 8.5. [PITH_FULL_IMAGE:figures/full_fig_p048_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Evolution of commune eligibility for the recognition of a state of natural catastrophe [PITH_FULL_IMAGE:figures/full_fig_p049_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: Accuracy of SwiGAN in predicting commune eligibility for the recognition of a state [PITH_FULL_IMAGE:figures/full_fig_p050_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: Future insurance damage costs due to drought-induced soil subsidence under RCP [PITH_FULL_IMAGE:figures/full_fig_p050_19.png] view at source ↗
Figure 20
Figure 20. Figure 20: Evolution of mean annual temperature in the Grand Est region under the RCP 4.5 [PITH_FULL_IMAGE:figures/full_fig_p051_20.png] view at source ↗
Figure 21
Figure 21. Figure 21: Evolution of mean annual temperature in the Grand Est region under the RCP 8.5 [PITH_FULL_IMAGE:figures/full_fig_p052_21.png] view at source ↗
Figure 22
Figure 22. Figure 22: Evolution of mean annual evapotranspiration in the Grand Est region under the RCP [PITH_FULL_IMAGE:figures/full_fig_p053_22.png] view at source ↗
Figure 23
Figure 23. Figure 23: Evolution of mean annual evapotranspiration in the Grand Est region under the RCP [PITH_FULL_IMAGE:figures/full_fig_p054_23.png] view at source ↗
Figure 24
Figure 24. Figure 24: Evolution of mean annual precipitation in the Grand Est region under the RCP 4.5 [PITH_FULL_IMAGE:figures/full_fig_p055_24.png] view at source ↗
Figure 25
Figure 25. Figure 25: Evolution of mean annual precipitation in the Grand Est region under the RCP 8.5 [PITH_FULL_IMAGE:figures/full_fig_p056_25.png] view at source ↗
Figure 26
Figure 26. Figure 26: Illustration of a residual block. Noise is introduced at the output of each convolutional [PITH_FULL_IMAGE:figures/full_fig_p057_26.png] view at source ↗
read the original abstract

According to the United Nations Office for Disaster Risk Reduction (2025), the average annual cost of natural catastrophes increased from 70--80 billion USD between 1970 and 2000 to 180--200 billion USD between 2001 and 2020. Reports from organizations such as the IFOA and the WWF highlight the need for the insurance sector to adapt to this rapidly evolving context by developing medium- to long-term strategies that go beyond the one-year horizon of prudential regulations such as Solvency II. This paper introduces an artificial intelligence framework based on Conditional Generative Adversarial Networks (Conditional GANs) to generate future spatio-temporal trajectories of climatic indices. The approach focuses on the Soil Wetness Index (SWI), a key indicator used in France to assess drought severity. Drought accounts for approximately 30% of the indemnities paid under the French natural catastrophe insurance scheme. The proposed model, SwiGAN, simulates plausible drought propagation patterns up to 2050 for a region of France particularly exposed to this hazard. By generating realistic sequences of SWI maps, SwiGAN provides insights into drought dynamics under climate change scenarios and supports the design of adaptive risk management and insurance strategies. The methodology is also generalizable to other climate-related perils and actuarial applications such as economic scenario generation.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript introduces SwiGAN, a conditional Wasserstein GAN trained on historical data to generate future spatio-temporal SWI (Soil Wetness Index) trajectories up to 2050 for a drought-exposed region in France. Conditioned on climate change scenarios, the model is claimed to produce realistic drought propagation patterns that support long-term risk management and insurance strategies, with potential generalization to other climate perils.

Significance. If the generated trajectories are shown to capture non-stationary drought dynamics beyond historical resampling, the framework could provide a practical data-driven tool for medium- to long-term scenario generation in insurance, addressing gaps in current regulations such as Solvency II. The application of conditional GANs to spatio-temporal climate indices for actuarial use is a relevant extension of existing generative modeling techniques.

major comments (2)
  1. [Abstract] Abstract: The central claim that SwiGAN 'simulates plausible drought propagation patterns' and 'provides insights into drought dynamics under climate change scenarios' lacks any quantitative validation metrics, error analysis, comparison to baselines (e.g., statistical models or physical simulations), or held-out period tests. This omission prevents assessment of whether the conditioning successfully shifts distributions to reflect future non-stationary conditions rather than interpolating historical patterns.
  2. [Methodology] Methodology (inferred from abstract description): The description of how auxiliary conditioning inputs from RCP/CMIP projections are integrated into the generator and discriminator to enforce extrapolation under climate change is absent; without explicit details on the conditioning mechanism and loss terms, it is impossible to verify that the model avoids circularity in labeling fitted outputs as future predictions.
minor comments (2)
  1. [Title/Abstract] The title refers to 'soil subsidence' while the abstract and claims center on drought and SWI; the introduction should explicitly link SWI to subsidence risk and clarify the hazard focus.
  2. [Abstract] The abstract mentions generalizability to other perils and economic scenario generation but provides no concrete examples or discussion of transferability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive and detailed comments on our manuscript. We have carefully reviewed each point and provide point-by-point responses below, along with planned revisions to strengthen the presentation of our results and methodology.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claim that SwiGAN 'simulates plausible drought propagation patterns' and 'provides insights into drought dynamics under climate change scenarios' lacks any quantitative validation metrics, error analysis, comparison to baselines (e.g., statistical models or physical simulations), or held-out period tests. This omission prevents assessment of whether the conditioning successfully shifts distributions to reflect future non-stationary conditions rather than interpolating historical patterns.

    Authors: We agree that the abstract would be strengthened by explicitly referencing quantitative validation. The full manuscript includes validation through statistical comparisons (e.g., moments, spatial correlations, and temporal persistence) of generated SWI fields against historical observations, as well as qualitative assessment of drought propagation patterns. To directly address the concern regarding non-stationarity and extrapolation, we will revise the abstract to summarize key quantitative results, including Wasserstein distances between conditioned and unconditioned distributions and performance on a held-out recent period. We will also add a comparison to a simple statistical baseline (e.g., a conditional Gaussian process) in the results section of the revised manuscript. revision: yes

  2. Referee: [Methodology] Methodology (inferred from abstract description): The description of how auxiliary conditioning inputs from RCP/CMIP projections are integrated into the generator and discriminator to enforce extrapolation under climate change is absent; without explicit details on the conditioning mechanism and loss terms, it is impossible to verify that the model avoids circularity in labeling fitted outputs as future predictions.

    Authors: We acknowledge that the conditioning mechanism requires more explicit description to avoid ambiguity. In the manuscript, conditioning is performed by embedding the RCP/CMIP scenario variables and concatenating them to the latent noise vector in the generator and to intermediate feature maps in the discriminator; the training objective augments the standard Wasserstein loss with a regression term that penalizes deviation from the provided climate covariates. To resolve the referee's concern, we will add a dedicated subsection with the precise mathematical formulation of the conditioning layers and the composite loss function, clarifying that the climate inputs are external projections independent of the generated outputs and thus prevent circularity. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper applies a standard conditional Wasserstein GAN (SwiGAN) to learn the distribution of historical SWI fields and generate new trajectories conditioned on external climate projections. Training minimizes the standard adversarial loss between generator and discriminator on observed data; generation uses separate conditioning inputs derived from RCP/CMIP scenarios. No equation or claim reduces by construction to its own fitted parameters, no self-citation supplies a uniqueness theorem that forces the result, and no ansatz is smuggled in. The central claim—that generated sequences are plausible under non-stationary conditions—rests on the model's generalization properties and external validation, which are independent of the training inputs themselves.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on standard machine-learning assumptions about the ability of GANs to extrapolate climate dynamics from historical data; no free parameters, axioms, or invented entities are explicitly introduced in the abstract.

axioms (1)
  • domain assumption Historical SWI data contains the statistical structure needed to generate realistic future trajectories under climate change.
    Implicit in training a generative model on past observations to produce future scenarios.

pith-pipeline@v0.9.0 · 5559 in / 1277 out tokens · 37556 ms · 2026-05-11T01:26:39.670346+00:00 · methodology

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

70 extracted references · 70 canonical work pages · 5 internal anchors

  1. [1]

    Arjovsky, S

    M. Arjovsky, S. Chintala, and L. Bottou. Wasserstein generative adversarial networks. In Proceedings of the 34th International Conference on Machine Learning, 2017. URL http://proceedings.mlr.press/v70/arjovsky17a.html. Introduces the Wasserstein distance for improved training of GANs

    work page 2017

  2. [2]

    Barthelemy, B

    S. Barthelemy, B. Bonan, M. Tomas-Burguera, G. Grandjean, S. Bernardie, J.-P. Naulin, P. Le Moigne, A. Boone, and J.-C. Calvet. Analysis of past and future droughts causing clay shrinkage in france. Hydrology and Earth System Sciences, 29 0 (11): 0 2321--2337, 2025

    work page 2025

  3. [3]

    Belhajjam, A

    R. Belhajjam, A. Chaqdid, N. Yebari, M. Seaid, and N. El Mocayd. Climate-informed flood risk mapping using a gan-based approach (exgan). Journal of Hydrology, 638: 0 131487, 2024

    work page 2024

  4. [4]

    Bianchini and F

    M. Bianchini and F. Scarselli. On the complexity of neural network classifiers: A comparison between shallow and deep architectures. IEEE transactions on neural networks and learning systems, 25 0 (8): 0 1553--1565, 2014

    work page 2014

  5. [5]

    R. Buch, S. Grimm, R. Korn, and I. Richert. Estimating the value-at-risk by temporal vae. Risks, 11 0 (5): 0 79, 2023

    work page 2023

  6. [6]

    Cai, Z.-L

    X. Cai, Z.-L. Yang, Y. Xia, M. Huang, H. Wei, L. R. Leung, and M. B. Ek. Assessment of simulated water balance from noah, noah-mp, clm, and vic over conus using the nldas test bed. Journal of Geophysical Research: Atmospheres, 119 0 (24): 0 13--751, 2014

    work page 2014

  7. [7]

    Chakraborty, D

    D. Chakraborty, D. Sharma, A. Mitra, and B. N. Goswami. Cv-gan-a generative model to simulate global climate variability. Authorea Preprints, 2025

    work page 2025

  8. [8]

    Charpentier, M

    A. Charpentier, M. James, and H. Ali. Predicting drought and subsidence risks in france. Natural Hazards and Earth System Sciences, 22 0 (7): 0 2401--2418, 2022. doi:10.5194/nhess-22-2401-2022

    work page doi:10.5194/nhess-22-2401-2022 2022

  9. [9]

    An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

    A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gelly, J. Uszkoreit, and N. Houlsby. An image is worth 16x16 words: Transformers for image recognition at scale, 2021. URL https://arxiv.org/abs/2010.11929

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  10. [10]

    J. B. Drake, P. W. Jones, and G. R. Carr Jr. Overview of the software design of the community climate system model. The International Journal of High Performance Computing Applications, 19 0 (3): 0 177--186, 2005

    work page 2005

  11. [11]

    Durall, A

    R. Durall, A. Chatzimichailidis, P. Labus, and J. Keuper. Combating mode collapse in gan training: An empirical analysis using hessian eigenvalues, 2020. URL https://arxiv.org/abs/2012.09673

    work page arXiv 2020

  12. [12]

    Ecoto and A

    G. Ecoto and A. Chambaz. Forecasting the cost of drought events in france by super learning. EGUsphere [Preprint], 2022. doi:10.5194/egusphere-2022-541

    work page doi:10.5194/egusphere-2022-541 2022

  13. [13]

    Ferchichi, M

    A. Ferchichi, M. Chihaoui, and A. Ferchichi. Spatio-temporal modeling of climate change impacts on drought forecast using generative adversarial network: A case study in africa. Expert Systems with Applications, 238: 0 122211, 2024

    work page 2024

  14. [14]

    Flaig and G

    S. Flaig and G. Junike. Scenario generation for market risk models using generative neural networks. Risks, 10 0 (11): 0 199, 2022

    work page 2022

  15. [15]

    Foroumandi, K

    E. Foroumandi, K. Gavahi, and H. Moradkhani. Generative adversarial network for real-time flash drought monitoring: A deep learning study. Water Resources Research, 60 0 (5): 0 e2023WR035600, 2024

    work page 2024

  16. [16]

    Glorot and Y

    X. Glorot and Y. Bengio. Understanding the difficulty of training deep feedforward neural networks. In Y. W. Teh and M. Titterington, editors, Proceedings of the Thirteenth International Conference on Artificial Intelligence and Statistics, volume 9 of Proceedings of Machine Learning Research, pages 249--256, Chia Laguna Resort, Sardinia, Italy, 13--15 Ma...

    work page 2010

  17. [17]

    Golan, C

    R. Golan, C. Jacob, and J. Denzinger. Lung nodule detection in ct images using deep convolutional neural networks. In 2016 International Joint Conference on Neural Networks (IJCNN), pages 243--250, 2016. doi:10.1109/IJCNN.2016.7727205

    work page doi:10.1109/ijcnn.2016.7727205 2016

  18. [18]

    Goodfellow, J

    I. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, and Y. Bengio. Generative adversarial networks. In Advances in Neural Information Processing Systems, volume 27, 2014. URL https://papers.nips.cc/paper/2014/hash/5ca3e9b122f61f8f06494c97b1afccf3-Abstract.html. Paper introducing the Generative Adversarial Networks (G...

    work page 2014

  19. [19]

    Gulrajani, F

    I. Gulrajani, F. Ahmed, M. Arjovsky, V. Dumoulin, and A. Courville. Improved training of wasserstein gans. In Advances in Neural Information Processing Systems, volume 30, 2017. URL https://proceedings.neurips.cc/paper/2017/hash/892c3b1c62f3a1ba0f2a64a9d0c3bb12-Abstract.html. Proposes the gradient penalty (WGAN-GP) for more stable GAN training

    work page 2017

  20. [20]

    K. He, X. Zhang, S. Ren, and J. Sun. Deep residual learning for image recognition, 2015. URL https://arxiv.org/abs/1512.03385

    work page internal anchor Pith review arXiv 2015

  21. [21]

    Heranval, O

    A. Heranval, O. Lopez, and M. Thomas. Application of machine learning methods to predict drought cost in france. European Actuarial Journal, 13 0 (2): 0 731--753, 2023

    work page 2023

  22. [22]

    J. Hu, L. Shen, S. Albanie, G. Sun, and E. Wu. Squeeze-and-excitation networks, 2019. URL https://arxiv.org/abs/1709.01507

    work page arXiv 2019

  23. [23]

    Deep networks with stochastic depth

    G. Huang, Y. Sun, Z. Liu, D. Sedra, and K. Weinberger. Deep networks with stochastic depth, 2016. URL https://arxiv.org/abs/1603.09382

    work page arXiv 2016

  24. [24]

    Planetary solvency – finding our balance with nature

    Institute and Faculty of Actuaries and University of Exeter . Planetary solvency – finding our balance with nature. Technical report, Institute and Faculty of Actuaries (IFoA), 2025. URL https://actuaries.org.uk/media/wqeftma1/planetary-solvency-finding-our-balance-with-nature.pdf. Report exploring climate risk impacts on economies and implications pour l...

    work page 2025

  25. [25]

    Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift

    S. Ioffe and C. Szegedy. Batch normalization: Accelerating deep network training by reducing internal covariate shift, 2015. URL https://arxiv.org/abs/1502.03167

    work page internal anchor Pith review arXiv 2015

  26. [26]

    Image-to-Image Translation with Conditional Adversarial Networks , journal =

    P. Isola, J.-Y. Zhu, T. Zhou, and A. A. Efros. Image-to-image translation with conditional adversarial networks, 2018. URL https://arxiv.org/abs/1611.07004

    work page arXiv 2018

  27. [27]

    Jiang, Y

    M. Jiang, Y. Liang, S. Han, K. Ma, Y. Chen, and Z. Xu. Leveraging generative adversarial networks for addressing data imbalance in financial market supervision. arXiv preprint, 2024. URL https://arxiv.org/abs/2412.15222. Méthode GAN pour corriger le déséquilibre de données en supervision des marchés financiers

    work page arXiv 2024

  28. [28]

    E. Jung, M. Luna, and S. H. Park. Conditional gan with an attention-based generator and a 3d discriminator for 3d medical image generation. In International Conference on Medical Image Computing and Computer-Assisted Intervention, pages 318--328. Springer, 2021

    work page 2021

  29. [29]

    A style-based generator architecture for generative adversarial networks

    T. Karras, S. Laine, and T. Aila. A style-based generator architecture for generative adversarial networks, 2019. URL https://arxiv.org/abs/1812.04948

    work page arXiv 2019

  30. [30]

    Training generative adversar- ial networks with limited data,

    T. Karras, M. Aittala, J. Hellsten, S. Laine, J. Lehtinen, and T. Aila. Training generative adversarial networks with limited data, 2020 a . URL https://arxiv.org/abs/2006.06676

    work page arXiv 2020

  31. [31]

    Analyzing and improving the image quality of stylegan

    T. Karras, S. Laine, M. Aittala, J. Hellsten, J. Lehtinen, and T. Aila. Analyzing and improving the image quality of stylegan, 2020 b . URL https://arxiv.org/abs/1912.04958

    work page arXiv 2020

  32. [32]

    On convergence and stability of GANs

    N. Kodali, J. Abernethy, J. Hays, and Z. Kira. On convergence and stability of gans, 2017. URL https://arxiv.org/abs/1705.07215

    work page arXiv 2017

  33. [33]

    J. Li, Y. Luo, Z. Li, G. Xu, M. Guo, and F. Gu. Analysis of spatiotemporal variability and drivers of soil moisture in the ziwuling region. Sustainability, 17 0 (17): 0 8025, 2025

    work page 2025

  34. [34]

    H. Lin, J. Wang, F. Li, Y. Xie, C. Jiang, and L. Sun. Drought trends and the extreme drought frequency and characteristics under climate change based on spi and hi in the upper and middle reaches of the huai river basin, china. Water, 12 0 (4): 0 1100, 2020

    work page 2020

  35. [35]

    L. Lu. An empirical study of wgan and wgan-gp for enhanced image generation. Applied and Computational Engineering, 83: 0 103--109, 2024. doi:10.54254/2755-2721/83/2024GLG0066

    work page doi:10.54254/2755-2721/83/2024glg0066 2024

  36. [36]

    A Unified Approach to Interpreting Model Predictions

    S. Lundberg and S.-I. Lee. A unified approach to interpreting model predictions, 2017. URL https://arxiv.org/abs/1705.07874

    work page Pith review arXiv 2017

  37. [37]

    S. L. Markstrom, L. E. Hay, and M. P. Clark. Towards simplification of hydrologic modeling: Identification of dominant processes. Hydrology and Earth System Sciences, 20 0 (11): 0 4655--4671, 2016

    work page 2016

  38. [38]

    Mazrooei and A

    A. Mazrooei and A. Sankarasubramanian. Improving monthly streamflow forecasts through assimilation of observed streamflow for rainfall-dominated basins across the conus. Journal of Hydrology, 575: 0 704--715, 2019

    work page 2019

  39. [39]

    Spectral Normalization for Generative Adversarial Networks

    T. Miyato, T. Kataoka, M. Koyama, and Y. Yoshida. Spectral normalization for generative adversarial networks, 2018. URL https://arxiv.org/abs/1802.05957

    work page Pith review arXiv 2018

  40. [40]

    Obakrim, L

    S. Obakrim, L. Benoit, and D. Allard. A multivariate and space-time stochastic weather generator using a latent gaussian framework. Stochastic Environmental Research and Risk Assessment, 39 0 (9): 0 3677--3701, 2025

    work page 2025

  41. [41]

    Understanding the exploding gradient problem.CoRR abs/1211.5063 (2012)

    R. Pascanu, T. Mikolov, and Y. Bengio. On the difficulty of training recurrent neural networks, 2013. URL https://arxiv.org/abs/1211.5063

    work page arXiv 2013

  42. [42]

    Pedersoli, T

    M. Pedersoli, T. Lucas, C. Schmid, and J. Verbeek. Areas of attention for image captioning, 2017. URL https://arxiv.org/abs/1612.01033

    work page arXiv 2017

  43. [43]

    The effectiveness of data augmentation in image classifica- tion using deep learning,

    L. Perez and J. Wang. The effectiveness of data augmentation in image classification using deep learning, 2017. URL https://arxiv.org/abs/1712.04621

    work page arXiv 2017

  44. [44]

    M. S. Rahman, L. Di, E. Yu, L. Lin, and Z. Yu. Remote sensing based rapid assessment of flood crop damage using novel disaster vegetation damage index (dvdi). International Journal of Disaster Risk Science, 12 0 (1): 0 90--110, 2021

    work page 2021

  45. [45]

    Ramzan, C

    F. Ramzan, C. Sartori, S. Consoli, and D. Reforgiato Recupero. Generative adversarial networks for synthetic data generation in finance: Evaluating statistical similarities and quality assessment. AI, 5 0 (2): 0 667--685, 2024. doi:10.3390/ai5020035. Application of GANs pour génération de données synthétiques financières

    work page doi:10.3390/ai5020035 2024

  46. [46]

    U-Net: Convolutional Networks for Biomedical Image Segmentation

    O. Ronneberger, P. Fischer, and T. Brox. U-net: Convolutional networks for biomedical image segmentation, 2015. URL https://arxiv.org/abs/1505.04597

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  47. [47]

    A. G. Roy, N. Navab, and C. Wachinger. Concurrent spatial and channel squeeze & excitation in fully convolutional networks, 2018. URL https://arxiv.org/abs/1803.02579

    work page arXiv 2018

  48. [48]

    Improved Techniques for Training GANs

    T. Salimans, I. Goodfellow, W. Zaremba, V. Cheung, A. Radford, and X. Chen. Improved techniques for training gans, 2016. URL https://arxiv.org/abs/1606.03498

    work page Pith review arXiv 2016

  49. [49]

    J. Sha, X. Chen, Y. Chang, M. Zhang, and X. Li. A spatial weather generator based on conditional deep convolution generative adversarial nets (cdcgan). Climate Dynamics, 62 0 (2): 0 1275--1290, 2024

    work page 2024

  50. [50]

    J. S. Shukla and R. J. Pandya. Deep learning-oriented c-gan models for vegetative drought prediction on peninsular india. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 17: 0 282--297, 2023

    work page 2023

  51. [51]

    Soubeyroux, S

    J.-M. Soubeyroux, S. Bernus, L. Corre, A. Drouin, B. Dubuisson, P. Etchevers, P. Josse, M. Kerdoncuff, R. Samacoits, and F. Tocquer. LES NOUVELLES PROJECTIONS CLIMATIQUES DE R \'E F \'E RENCE DRIAS 2020 . PhD thesis, Meteo-France, 2020

    work page 2020

  52. [52]

    Srivastava, L

    A. Srivastava, L. Valkov, C. Russell, M. U. Gutmann, and C. Sutton. Veegan: Reducing mode collapse in gans using implicit variational learning. pages 3310--3320, 2017. URL https://papers.nips.cc/paper/6923-veegan-reducing-mode-collapse-in-gans-using-implicit-variational-learning.pdf. Introduces a GAN variant (VEEGAN) designed pour atténuer le mode collapse

    work page 2017

  53. [53]

    Strachan and C

    S. Strachan and C. Daly. Testing the daily prism air temperature model on semiarid mountain slopes. Journal of Geophysical Research: Atmospheres, 122 0 (11): 0 5697--5715, 2017

    work page 2017

  54. [54]

    Thanh-Tung and T

    H. Thanh-Tung and T. Tran. On catastrophic forgetting and mode collapse in generative adversarial networks, 2020. URL https://arxiv.org/abs/1807.04015

    work page arXiv 2020

  55. [55]

    Instance Normalization: The Missing Ingredient for Fast Stylization

    D. Ulyanov, A. Vedaldi, and V. Lempitsky. Instance normalization: The missing ingredient for fast stylization, 2017. URL https://arxiv.org/abs/1607.08022

    work page Pith review arXiv 2017

  56. [56]

    Global Assessment Report on Disaster Risk Reduction 2025

    United Nations Office for Disaster Risk Reduction . Global Assessment Report on Disaster Risk Reduction 2025 . Technical report, United Nations, 2025. URL https://www.undrr.org/gar/gar2025

    work page 2025

  57. [57]

    Y. Wang, L. Shi, Y. Hu, X. Hu, W. Song, and L. Wang. A comprehensive study of deep learning for soil moisture prediction. Hydrology and Earth System Sciences, 28: 0 917--943, 02 2024. doi:10.5194/hess-28-917-2024

    work page doi:10.5194/hess-28-917-2024 2024

  58. [58]

    Z. Wang, P. Wang, K. Liu, P. Wang, Y. Fu, C.-T. Lu, C. C. Aggarwal, J. Pei, and Y. Zhou. A comprehensive survey on data augmentation, 2025. URL https://arxiv.org/abs/2405.09591

    work page arXiv 2025

  59. [59]

    S. Woo, J. Park, J.-Y. Lee, and I. S. Kweon. Cbam: Convolutional block attention module, 2018. URL https://arxiv.org/abs/1807.06521

    work page Pith review arXiv 2018

  60. [60]

    Tackling the insurance protection gap: Leveraging climate and nature to increase resilience

    World Wide Fund For Nature . Tackling the insurance protection gap: Leveraging climate and nature to increase resilience. Technical report, World Wide Fund For Nature (WWF), 2026. URL https://wwfint.awsassets.panda.org/downloads/wwf-insurance-protection-gap-report-.pdf. White paper examining how climate change and nature loss are widening the insurance pr...

    work page 2026

  61. [61]

    M. L. Wrzesien, T. M. Pavelsky, M. T. Durand, J. Dozier, and J. D. Lundquist. Characterizing biases in mountain snow accumulation from global data sets. Water Resources Research, 55 0 (11): 0 9873--9891, 2019

    work page 2019

  62. [62]

    M. V. Wuthrich, R. Richman, B. Avanzi, M. Lindholm, M. Maggi, M. Mayer, J. Schelldorfer, and S. Scognamiglio, 2026. Available at SSRN:https://ssrn.com/abstract=5162304 or http://dx.doi.org/10.2139/ssrn.5162304

    work page doi:10.2139/ssrn.5162304 2026

  63. [63]

    Xue and G

    S. Xue and G. Wu. A study of the dependence between soil moisture and precipitation in different ecoregions of the northern hemisphere. Hydrology and Earth System Sciences, 29 0 (20): 0 5575--5591, 2025

    work page 2025

  64. [64]

    Xueping, S

    Z. Xueping, S. M. Samuri, Q. Lin, and M. H. M. Adnan. Integrating generative adversarial networks and stacked autoencoder neural networks for enhanced credit risk assessment in fintech. International Journal of Entrepreneurship and Management Practises, 8 0 (29), 2025. doi:10.35631/IJEMP.829001. Intègre GAN et auto-encodeurs pour améliorer l’évaluation du...

    work page doi:10.35631/ijemp.829001 2025

  65. [65]

    X. Yang, X. Xie, D. L. Liu, F. Ji, and L. Wang. Spatial interpolation of daily rainfall data for local climate impact assessment over greater sydney region. Advances in Meteorology, 2015 0 (1): 0 563629, 2015

    work page 2015

  66. [66]

    S. Yun, D. Han, S. J. Oh, S. Chun, J. Choe, and Y. Yoo. Cutmix: Regularization strategy to train strong classifiers with localizable features, 2019. URL https://arxiv.org/abs/1905.04899

    work page arXiv 2019

  67. [67]

    mixup: Beyond Empirical Risk Minimization

    H. Zhang, M. Cisse, Y. N. Dauphin, and D. Lopez-Paz. mixup: Beyond empirical risk minimization, 2018. URL https://arxiv.org/abs/1710.09412

    work page internal anchor Pith review arXiv 2018

  68. [68]

    Zhang, Z

    H. Zhang, Z. Zhang, A. Odena, and H. Lee. Consistency regularization for generative adversarial networks, 2020. URL https://arxiv.org/abs/1910.12027

    work page arXiv 2020

  69. [69]

    S. Zhao, Z. Liu, J. Lin, J.-Y. Zhu, and S. Han. Differentiable augmentation for data-efficient gan training, 2020 a . URL https://arxiv.org/abs/2006.10738

    work page arXiv 2020

  70. [70]

    Z. Zhao, S. Singh, H. Lee, Z. Zhang, A. Odena, and H. Zhang. Improved consistency regularization for gans, 2020 b . URL https://arxiv.org/abs/2002.04724

    work page arXiv 2020