Skip to main content

Advertisement

Log in

Progressive growing generative adversarial networks using conditioning ratio for facies modeling in complex aquifers

Réseaux adversaires génératifs à croissance progressive utilisant le rapport de conditionnement pour la modélisation des faciès dans les aquifères complexes

Redes generativas antagónicas de crecimiento progresivo que utilizan la relación de condicionamiento para el modelado de facies en acuíferos complejos

利用条件比例的渐进式增长生成对抗网络进行复杂含水层相模拟

Redes adversárias generativas de crescimento progressivos utilizando razão condicionante para modelagem de faceis em aquíferos complexos

  • Paper
  • Published:
Hydrogeology Journal Aims and scope Submit manuscript

Abstract

Groundwater-flow and contaminant-transport modeling rely on methods of converting a set of field observations into geologic models that represent the subsurface structure. These geologic models also must replicate important geologic features such as connectivity. Recently, researchers have begun to use machine learning methods such as generative adversarial networks (GANs). This study focuses on a progressive growing GAN (PGGAN) to condition on measured data. Given a latent variable and an array that provides field observations, the generators of the conditioned PGGAN are tasked to produce geologically realistic images of channel aquifers that match field observations. Although largely successful, the conditioning behavior of these networks still has some issues, and how the model performs the conditioning task across its layers is not yet fully understood. To better understand this conditioning mechanism, the behavior of these networks was measured using the conditioning ratio, which is a novel metric that determines the magnitude of the influence of the conditioning data. The conditioning ratio was measured across multiple layers within the generator during training, as well as with various modifications to the network architecture. The results revealed two distinct conditioning behaviors that are based on the number of conditioning arrays injected into the generator. Results also showed that decreasing the starting resolution for the generator can slow down the learning process. Overall, the numerical experiments prove the value of measuring the conditioning ratio of layers within the generator. These approaches can be used as diagnostic tools to assist in the design of future PGGAN architectures.

Résumé

La modélisation de l'écoulement des eaux souterraines et du transport des contaminants repose sur des méthodes permettant de convertir un ensemble d'observations de terrain en modèles géologiques représentant la structure du sous-sol. Ces modèles géologiques doivent également reproduire des caractéristiques géologiques importantes telles que la connectivité. Récemment, les chercheurs ont commencé à utiliser des méthodes d'apprentissage automatique telles que les réseaux adversaires génératifs (GANs). Cette étude se concentre sur un GAN à croissance progressive (PGGAN) pour conditionner les données mesurées. Compte tenu d'une variable latente et d'un tableau qui fournit des observations de terrain, les générateurs du PGGAN conditionné ont pour tâche de produire des images géologiquement réalistes d'aquifères chenalisés qui correspondent aux observations de terrain. Bien que largement réussi, le comportement du conditionnement de ces réseaux présente encore quelques problèmes, et la façon dont le modèle effectue la tâche de conditionnement à travers ses couches n'est pas encore entièrement comprise. Pour mieux comprendre ce mécanisme de conditionnement, le comportement de ces réseaux a été mesuré à l'aide du rapport de conditionnement, une nouvelle mesure qui détermine l'ampleur de l'influence des données de conditionnement. Le taux de conditionnement a été mesuré sur plusieurs couches du générateur pendant la phase d’apprentissage, ainsi qu'avec diverses modifications de l'architecture du réseau. Les résultats ont révélé deux comportements de conditionnement distincts basés sur le nombre de réseaux de conditionnement injectés dans le générateur. Les résultats ont également montré que la diminution de la résolution de départ du générateur peut ralentir le processus d'apprentissage. Dans l'ensemble, les expériences numériques prouvent l'intérêt de mesurer le rapport de conditionnement des couches au sein du générateur. Ces approches peuvent être utilisées comme outils de diagnostic pour aider à la conception de futures architectures PGGAN.

Resumen

El modelado del flujo de aguas subterráneas y del transporte de contaminantes se basa en métodos para convertir un conjunto de observaciones de campo en modelos geológicos que representen la estructura del subsuelo. Estos modelos también deben reproducir características geológicas importantes, como la conectividad. Recientemente, los investigadores han empezado a utilizar métodos de aprendizaje automático como las redes generativas antagónicas (GANs). Este estudio se centra en una GAN de crecimiento progresivo (PGGAN) para condicionar los datos medidos. Dada una variable latente y una matriz que proporciona observaciones de campo, los generadores del PGGAN condicionado tienen la tarea de producir imágenes geológicamente realistas de acuíferos de cauces que coincidan con las observaciones de campo. Aunque en gran medida con éxito, el comportamiento de condicionamiento de estas redes todavía tiene algunos problemas, y la forma en que el modelo realiza la tarea de condicionamiento a través de sus capas aún no se entiende completamente. Para comprender mejor este mecanismo de condicionamiento, el comportamiento de estas redes se midió utilizando el coeficiente que determina la magnitud de la influencia de los datos de condicionamiento. Se midió el coeficiente y se introdujo en varias capas del generador durante el entrenamiento, así como con diversas modificaciones en la arquitectura de la red. Los resultados revelaron dos comportamientos de condicionamiento distintos que se basan en el número de matrices que se inyectan en el generador. Los resultados también mostraron que disminuir la resolución inicial del generador puede retrasar el proceso de aprendizaje. En general, los experimentos numéricos demuestran el valor de medir la relación de condicionamiento de las capas dentro del generador. Estos enfoques pueden utilizarse como herramientas de diagnóstico para ayudar en el diseño de futuras arquitecturas PGGAN.

摘要

地下水流和污染物运移建模依赖于将一系列野外观测转换为代表地下结构的地质模型的方法。这些地质模型还必须复制重要的地质特征,如连通性。最近,研究人员开始使用机器学习方法,例如生成对抗网络(GANs)。本研究重点研究了一种渐进增长的PGGAN(Progressive Growing GAN),以对测量数据进行条件设置。在给定潜在的变量和提供现场观测数据的情况下,受条件的PGGAN的生成器的任务是产生与现场观测相匹配的通道含水层的真实地质的图像。尽管在很大程度上取得了成功,但这些网络的条件行为仍然存在一些问题,而模型如何在其层次上执行条件任务尚未完全理解。为了更好地了解这种条件机制,使用条件比率测量了这些网络的行为,该比率是一种新颖的度量方法,可确定条件数据的影响程度。在训练期间测量了生成器内多个层中的条件比率,以及对网络架构进行了各种修改。结果揭示了基于注入到生成器中的条件数组数量的两种不同的调节行为。结果还表明,降低生成器的起始分辨率可能会减慢学习过程。总体而言,数值实验证明了测量生成器内层的条件比率的价值。这些方法可以用作诊断工具,以协助设计未来的PGGAN架构。

Resumo

Modelagem de fluxo de águas subterrâneas e transporte de contaminante se baseiam em métodos de conversão de uma série de observações de campo em modelos geológicos que representam a estrutura da subsuperfície. Esses modelos geológicos também devem replicar características geológicas importantes como a conectividade. Recentemente, pesquisadores começaram a usar métodos de aprendizado de máquinas como redes adversarias generativas (RAGs). Esse estudo foca em uma RAG de crescimento progressivo (RAGCP) para o condicionamento de dados medidos. Dado uma variável latente e uma variedade que fornecem observações de campo, os geradores de RAGCP condicionados são atarefados para produzir imagens geologicamente realista de aquífero-canal que correspondessem observações de campo. Mesmo amplamente bem sucedido, o comportamento condicionante dessas redes ainda possui algumas questões, e como o modelo performa a tarefa condicionante através das camadas não foi completamente entendido. Para melhor entendimento desse mecanismo de condicionamento, o comportamento dessas redes foi medido utilizando a razão condicionante, que é uma métrica inovadora que determina a magnitude da influência dos dados condicionantes. A razão condicionante foi medida através das camadas-múltiplas com o gerador durante o treino, assim como com modificações variadas para a arquitetura das redes. Os resultados revelaram dois comportamentos condicionantes distintos que foram baseados no número de variedades condicionantes injetadas no gerador. Resultados também mostram que diminuir a resolução inicial para o gerador pode retardar o processo de aprendizado. Em geral, os experimentos numéricos provam o valor de se medir a razão condicionante das camadas com o gerador. Essas abordagens pode ser utilizados como ferramenta diagnostica para auxiliar no design de arquiteturas RAGCP futuras.

This is a preview of subscription content, log in via an institution to check access.

Access this article

Price excludes VAT (USA)
Tax calculation will be finalised during checkout.

Instant access to the full article PDF.

Fig. 1
Fig. 2
Algorithm 1
Algorithm 2
Fig. 3
Fig. 4
Fig. 5
Fig. 6
Fig. 7
Fig. 8
Fig. 9

Similar content being viewed by others

References

  • Arjovsky M, Chintala S, Bottou L (2017) Wasserstein GAN. arXiv preprint. arXiv:1701.07875

  • Azulay A, Weiss Y (2018) Why do deep convolutional networks generalize so poorly to small image transformations? arXiv preprint.https://arxiv.org/abs/1805.12177

  • Bao J, Li L, Davis A (2022) Variational autoencoder or generative adversarial networks? a comparison of two deep learning methods for flow and transport data assimilation. Math Geosci 54:1017–1042

  • Bao J, Li L, Redoloza F (2020) Coupling ensemble smoother and deep learning with generative adversarial networks to deal with non-gaussianity in flow and transport data assimilation. J Hydrol 590:125443

    Article  Google Scholar 

  • Chan S, Elsheikh AH (2018) Parametric generation of conditional geological realizations using generative neural networks. arXiv preprint. http://arxiv.org/abs/1807.05207

  • Deutsch CV, Journel AG (1992) Geostatistical software library and user’s guide. Oxford University Press, New York, p 119

  • Deutsch CV, Wang L (1996) Hierarchical object-based stochastic modeling of fluvial reservoirs. Math Geol 28:857–880

    Article  Google Scholar 

  • Dupont E, Zhang T, Tilke P, Liang L, Bailey W (2018) Generating realistic geology conditioned on physical measurements with generative adversarial networks. arXiv preprint. http://arxiv.org/abs/1802.03065

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

  • Gulrajani I, Ahmed F, Arjovsky M, Dumoulin V, Courville AC (2017) Improved training of Wasserstein GANs, In: Advances in neural information processing systems. MIT Press, Cambridge, MA, pp 5767–5777

  • Hauge R, Holden L, Syversveen AR (2007) Well conditioning in object models. Math Geol 39:383–398

    Article  Google Scholar 

  • Hernández-García A, König P (2018) Further advantages of data augmentation on convolutional neural networks. International Conference on Artificial Neural Networks, ICANN 2018, Rhodes, Greece, October 2018, pp 95–103

  • Heusel M, Ramsauer H, Unterthiner T, Nessler B, Hochreiter S (2017) GANs trained by a two time-scale update rule converge to a local Nash equilibrium. In: Advances in neural information processing systems. MIT Press, Cambridge, MA, pp 6626–6637

  • Holden L, Hauge R, Skare Ø, Skorstad A (1998) Modeling of fluvial reservoirs with object models. Math Geol 30:473–496

    Article  Google Scholar 

  • Honarkhah M, Caers J (2012) Direct pattern-based simulation of non-stationary geostatistical models. Math Geosci 44:651–672

    Article  Google Scholar 

  • Karras T, Aila T, Laine S, Lehtinen J (2017) Progressive growing of GANs for improved quality, stability, and variation. arXiv preprint. http://arxiv.org/abs/1710.10196

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

  • Kingma DP, Ba J (2014) Adam: a method for stochastic optimization. arXiv preprint. http://arxiv.org/abs/1412.6980

  • Lala S, Shady M, Belyaeva A, Liu M (2018) Evaluation of mode collapse in generative adversarial networks. 2018 IEEE High Performance Extreme Computing Conference. https://ieee-hpec.org/2018/2018program/index_htm_files/124.pdf. Accessed August 2023

  • Laloy E, Hérault R, Jacques D, Linde N (2018) Training-image based geostatistical inversion using a spatial generative adversarial neural network. Water Resour Res 54:381–406

    Article  Google Scholar 

  • Lochbühler T, Pirot G, Straubhaar J, Linde N (2014) Conditioning of multiple-point statistics facies simulations to tomographic images. Math Geosci 46:625–645

    Article  Google Scholar 

  • Lopez-Alvis J, Laloy E, Nguyen F, Hermans T (2020) Deep generative models in inversion: a review and development of a new approach based on a variational autoencoder. arXiv preprint. http://arxiv.org/abs/2008.12056

  • Lucic M, Kurach K, Michalski M, Gelly S, Bousquet O (2018) Are GANs created equal? a large-scale study. In: Advances in neural information processing systems. MIT Press, Cambridge, MA, pp 700–709

  • Mariethoz G, Straubhaar J, Renard P, Chugunova T, Biver P (2015) Constraining distance-based multipoint simulations to proportions and trends. Environ Modell Softw 72:184–197

    Article  Google Scholar 

  • Mariethoz G, Renard P (2010) Reconstruction of incomplete data sets or images using direct sampling. Math Geosci 42:245–268

    Article  Google Scholar 

  • Michael H, Li H, Boucher A, Sun T, Caers J, Gorelick S (2010) Combining geologic-process models and geostatistics for conditional simulation of 3-D subsurface heterogeneity. Water Resour Res 46

  • Mosser L, Dubrule O, Blunt MJ (2020) Stochastic seismic waveform inversion using generative adversarial networks as a geological prior. Math Geosci 52:53–79

    Article  Google Scholar 

  • Nesvold E, Mukerji T (2019) Geomodeling using generative adversarial networks and a database of satellite imagery of modern river deltas. In: Petroleum Geostatistics 2019, European Association of Geoscientists & Engineers, Utrecht, The Netherlands, pp 1–5

  • Salimans T, Goodfellow I, Zaremba W, Cheung V, Radford A, Chen X (2016) Improved techniques for training GANs. In: Advances in neural information processing systems (NIPS 2016), Barcelona, Dec 2016, pp 2234–2242

  • Shen C, Laloy E, Elshorbagy A, Albert A, Bales J, Chang FJ, Ganguly S, Hsu KL, Kifer D, Fang Z (2018) Hess opinions: incubating deep-learning-powered hydrologic science advances as a community. Hydrol Earth Syst Sci 22:5639–5656

    Article  Google Scholar 

  • Song S, Mukerji T, Hou J (2021a) Geological facies modeling based on progressive growing of generative adversarial networks (GANs). Comput Geosci 25:1251–1273

    Article  Google Scholar 

  • Song S, Mukerji T, Hou J (2021b) Gansim: conditional facies simulation using an improved progressive growing of generative adversarial networks (GANs). Math Geosci 53:1413–1444

    Article  Google Scholar 

  • Song S, Mukerji T, Hou J (2022a) Bridging the gap between geophysics and geology with generative adversarial networks. IEEE Trans Geosci Remote Sensing 60:1–11

    Google Scholar 

  • Song S, Mukerji T, Hou J, Zhang D, Lyu X (2022b) GANSim-3D for conditional geomodeling: theory and field application. Water Resour Res 58, e2021WR031865

  • Zahner T, Lochbühler T, Mariethoz G, Linde N (2016) Image synthesis with graph cuts: a fast model proposal mechanism in probabilistic inversion. Geophys J Int 204:1179–1190

    Article  Google Scholar 

  • Zhang T, Tilke P, Dupont E, Zhu L, Liang L, Bailey W (2019) Generating geologically realistic 3D reservoir facies models using deep learning of sedimentary architecture with generative adversarial networks. In: International Petroleum Technology Conference, OnePetro, Beijing, March 2019

Download references

Acknowledgements

The authors wish to thank the guest editor Philippe Renard and Przemyslaw Juda as well as an anonymous reviewer for their comments, which substantially helped to improve the final version of the manuscript.

Funding

This work has been supported through a grant from the National Science Foundation (OIA-1833069).

Author information

Authors and Affiliations

Authors

Corresponding author

Correspondence to Liangping Li.

Ethics declarations

Conflicts of interest

The authors have no conflicts of interest to report.

Additional information

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Published in the special issue “Geostatistics and hydrogeology”.

Rights and permissions

Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Redoloza, F., Li, L. & Davis, A. Progressive growing generative adversarial networks using conditioning ratio for facies modeling in complex aquifers. Hydrogeol J 31, 1565–1580 (2023). https://doi.org/10.1007/s10040-023-02687-6

Download citation

  • Received:

  • Accepted:

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1007/s10040-023-02687-6

Keywords

Navigation