Deep learning framework with hybrid loss function for seismic velocity inversion of natural gas hydrate reservoirs

Abstract

Natural gas hydrate (NGH) has attracted increasing attention as a promising unconventional energy resource owing to its high volumetric storage capacity, yet its development is accompanied by significant greenhouse gas risks. Therefore, accurate reservoir characterization is vital for marine resource exploration and sustainable development. Full waveform inversion (FWI) offers high-resolution imaging, yet suffers from heavy computation, sensitivity to initial models and non-uniqueness. Recent deep learning (DL) methods improve efficiency and accuracy, nevertheless, still struggle with clear boundary extraction and multi-level semantic representation. The authors propose a novel deep architecture (SC-UNeXt) designed to learn a mapping from seismic records to velocity models, which integrates a U-Net backbone, ConvNeXt residual blocks, spatial-channel squeeze-and-excitation attention and pixel shuffle up-sampling. Furthermore, a hybrid loss integrating mean squared error, multi-scale structural similarity, and perceptual discrepancy simultaneously optimizes pixel-wise accuracy, structural fidelity, and semantic consistency. Comprehensive tests on both synthetic NGH data and the 3D SEG/EAGE marine overthrust model with NGH demonstrate that SC-UNeXt outperforms FWI and advanced DL methods in boundary delineation, structural preservation, noise robustness, and computational efficiency. These results highlight SC-UNeXt as a reliable tool for high-resolution seismic characterization of NGH reservoirs, thereby supporting sustainable exploration and risk assessment of marine hydrate resources.

Keywords

Natural gas hydrate / Clean energy / Combustible ice / Unconventional energy / Seismic velocity inversion / Deep learning / Hybrid loss function / U-Net-like architecture / Full waveform inversion / Gas hydrate exploration / Risk assessment

References

[1] Aminzadeh F. 1996. 3-D salt and overthrust seismic models. In: Weimer P, Davis TL. Applications of 3-D seismic data to exploration and production. Tulsa, American Association of Petroleum Geologists, 376-391. doi: 10.1306/St42605C29.

[2] Bönke W, Alaei B, Torabi A, Oikonomou D. 2024. Data augmentation for 3D seismic fault interpretation using deep learning. Marine and Petroleum Geology, 162, 106706. doi: 10.1016/j.marpetgeo.2024.106706.

[3] Boswell R, Collett TS. 2011. Current perspectives on gas hydrate resources. Energy & Environmental Science, 4(4), 1206-1215. doi: 10.1039/C0EE00203H.

[4] Buffett BA. 2000. Clathrate hydrates. Annual Review of Earth and Planetary Sciences, 28(1), 477-507. doi: 10.1146/annurev.earth.28.1.477.

[5] Bustamante J, Fabien-Ouellet G, Duchesne MJ, Ibrahim A. 2024. Deep-learning viscoelastic seismic inversion for mapping subsea permafrost. Geophysics, 89(4), 339-353. doi: 10.1190/geo2022-0759.1.

[6] Chen JX, Chen GX, Li J, Du RS, Qi YL, Li CF, Wang NJ. 2024. Efficient seismic data denoising via deep learning with improved MCA-SCUNet. IEEE Transactions on Geoscience and Remote Sensing, 62, 1-14. doi: 10.1109/TGRS.2024.3355972.

[7] Djeffal A, Pecher IA, Singh SC, Crutchley GJ, Kaipio J. 2021. Improved imaging of gas hydrate reservoirs and their plumbing system using 2D elastic full-waveform inversion. Interpretation, 9(3), 955-968. doi: 10.1190/INT-2020-0163.1.

[8] Dou YM, Li KW, Dong MH, Xiao Y. 2024. FaultSSL: Seismic fault detection via semisupervised learning. Geophysics, 89(3), 79-91. doi: 10.1190/geo2023-0550.1.

[9] Gajanayake SM, Gamage RP, Li XS, Huppert H. 2023. Natural gas hydrates-Insights into a paradigm-shifting energy resource. Energy Reviews, 2(1), 100013. doi: 10.1016/j.enrev.2024.100089.

[10] Gao ZQ, Yang LS, Yang Z, Gao JH. 2025. Physics-constrained deep learning for attenuation compensation of nonstationary seismic data. IEEE Transactions on Geoscience and Remote Sensing, 63. doi: 10.1109/TGRS.2025.3541495.

[11] Guo ZW, Lu KX, Chen YJ. 2026. A semi-airborne transient electromagnetic data denoising method based on TOC-VMD and improved threshold. Journal of Geophysics and Engineering, 23(5), 1541-1555. doi: 10.1093/jge/gxag066.

[12] Hester KC, Brewer PG. 2009. Clathrate hydrates in nature. Annual review of marine science, 1(1), 303-327. doi: 10.1146/annurev.marine.010908.163824.

[13] Kamal M, Ali A, Amin Y, Farid A, Sarosh B. 2023. Advanced data processing as a tool to enhance the vertical and horizontal resolution of bottom simulating reflector (BSR): Implications for GAS hydrate exploitation. Pure and Applied Geophysics, 180(6), 2369-2387. doi: 10.1007/s00024-023-03257-3.

[14] Korenaga J, Holbrook W, Singh S, Minshull T. 1997. Natural gas hydrates on the southeast US margin: Constraints from full waveform and travel time inversions of wide-angle seismic data. Journal of Geophysical Research: Solid Earth, 102(B7), 15345-15365. doi: 10.1029/97JB00725.

[15] Li CD, Feng PF, Jiang XH, Zhang S, Meng J, Li BC. 2024. Extensive identification of landslide boundaries using remote sensing images and deep learning method. China Geology, 7(2), 277-290. doi: 10.31035/cg2023148.

[16] Li FD, Guo ZW, Pan XP, Liu JX, Wang YY, Gao DW. 2022. Deep learning with adaptive attention for seismic velocity inversion. Remote Sensing, 14(15), 3810. doi: 10.3390/rs14153810.

[17] Li L, Zeng XB, Pan XP, Peng L, Tan YY, Liu JX. 2024. Microseismic velocity inversion based on deep learning and data augmentation. Applied Sciences, 14(5), 2194. doi: 10.3390/app14052194.

[18] Li M, Yan XS, Wu QH. 2025. A self-supervised deep learning framework for seismic facies segmentation. Expert Systems with Applications, 288. doi: 10.1016/j.eswa.2025.128290.

[19] Li P, Liu ML, Alfarraj M, Tahmasebi P, Grana D. 2024. Probabilistic physics-informed neural network for seismic petrophysical inversion. Geophysics, 89(2), 17-32. doi: 10.1190/geo2023-0214.1.

[20] Li YL, Hu QB, Wu NY, Wang HB, Sun XF, Hu GW, Sun ZW, Jiang YJ. 2023. Acoustic characterization for creep behaviors of marine sandy hydrate-bearing sediment. Scientific Reports, 13(1), 22199. doi: 10.1038/s41598-023-49523-1.

[21] Li YS, Peng C, Ran XJ, Xue LF, Chai SL. 2022. Soil geochemical prospecting prediction method based on deep convolutional neural networks-Taking Daqiao Gold Deposit in Gansu Province, China as an example. China Geology, 5(1), 71-83. doi: 10.31035/cg2021044.

[22] Lin YZ, Theiler J, Wohlberg B. 2023. Physics-guided data-driven seismic inversion: Recent progress and future opportunities in full-waveform inversion. IEEE Signal Processing Magazine, 40(1), 115-133. doi: 10.1109/MSP.2022.3217658.

[23] Liu B, Jiang P, Wang QY, Ren YX, Yang SL, Cohn AG. 2023. Physics-driven self-supervised learning system for seismic velocity inversion. Geophysics, 88(2), 145-161. doi: 10.1190/geo2021-0302.1.

[24] Liu Z, Li YL, Ye RJ, Dong L, Qi MH, Wang W, Li ZK, Ning FL, Wu NY. 2026. A hybrid unsupervised-to-supervised machine learning framework for fracture segmentation in natural gas hydrate-bearing sediments. Engineering Geology, 364, 108624. doi: 10.1016/j.enggeo.2026.108624.

[25] Liu Z, Mao HZ, Wu CY, Feichtenhofer C, Darrell T, Xie SN. 2022. A convnet for the 2020s. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 11976-11986. doi: 10.1109/CVPR52688.2022.01167.

[26] Lu HL, Seo YT, Lee JW, Moudrakovski I, Ripmeester JA, Chapman NR, Coffin RB, Gardner G, Pohlman J. 2007. Complex gas hydrate from the Cascadia margin. Nature, 445(7125), 303-306. doi: 10.1038/nature05463.

[27] Ma K, Liu JJ, Lu SQ, Huang ZH, Tian M, Deng JY, Xie Z, Qiu QJ. 2025. Improved SE-UNet network-based semantic segmentation and implied geological significance from geological map. China Geology, 8, 1-19. doi: 10.31035/cg2023146.

[28] Ning FL, Yu YB, Kjelstrup S, Vlugt TJH, Glavatskiy K. 2012. Mechanical properties of clathrate hydrates: status and perspectives. Energy & Environmental Science, 5(5), 6779-6795. doi: 10.1039/C2EE03435B.

[29] Noh K, Kim D, Byun J. 2023. Explainable deep learning for supervised seismic facies classification using intrinsic method. IEEE Transactions on Geoscience and Remote Sensing, 61, 1-11. doi: 10.1109/TGRS.2023.3236500.

[30] Paszke A, Gross S, Massa F, Lerer A, Bradbury J, Chanan G, Killeen T, Lin ZM, Gimelshein N, Antiga L, Desmaison A, Kopf A, Yang E, DeVito Z, Raison M, Tejani A, Chilamkurthy S, Steiner B, Fang L, Bai JJ, Chintala S. 2019. Pytorch: An imperative style, high-performance deep learning library. Advances in Neural Information Processing Systems, 32. doi: 10.48550/arXiv.1912.01703.

[31] Rao SH, Li ZC, Lu HL, Deng YJ. 2025. A review of gas hydrate formation characteristics at interfaces. Fuel, 392, 134863. doi: 10.1016/j.fuel.2025.134863.

[32] Ren YX, Nie LC, Yang SL, Jiang P, Chen YK. 2021. Building complex seismic velocity models for deep learning inversion. IEEE Access, 9, 63767-63778. doi: 10.1109/ACCESS.2021.3051159.

[33] Ronneberger O, Fischer P, Brox T. 2015. U-net: Convolutional networks for biomedical image segmentation. In International Conference on Medical Image Computing and Computer-assisted Intervention, 234-241. doi: 10.1007/978-3-319-24574-4_28.

[34] Roy AG, Navab N, Wachinger C. 2018. Concurrent spatial and channel ‘squeeze & excitation’ in fully convolutional networks. In International Conference on Medical Image Computing and Computer-assisted Intervention, 421-429. doi: 10.1007/978-3-030-00928-1_48.

[35] Satyavani N, Sain K. 2015. Seismic insights into bottom simulating reflection (BSR) in the Krishna-Godavari basin, Eastern Margin of India. Marine Georesources and Geotechnology, 33(3), 191-201. doi: 10.1080/1064119X.2013.797059.

[36] Shi WZ, Caballero J, Huszár F, Totz J, Aitken AP, Bishop R, Rueckert D, Wang ZH. 2016. Real-time single image and video super-resolution using an efficient sub-pixel convolutional neural network. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 1874-1883. doi: 10.1109/CVPR.2016.207.

[37] Simonyan K, Zisserman A. 2015. Very deep convolutional networks for large-scale image recognition. International Conference on Learning Representations. doi: 10.48550/arXiv.1409.1556.

[38] Singh SC Minshull, TA. 1994. Velocity structure of a gas hydrate reflector at Ocean Drilling Program site 889 from a global seismic waveform inversion. Journal of Geophysical Research: Solid Earth, 99(B12), 24221-24233. doi: 10.1029/94JB02168.

[39] Wang BF, Li JK, Wang P, Si WP, Wang ZK. 2022. Visco-acoustic full waveform inversion using decoupled fractional Laplacian constant-Q wave equation and optimal transport-based misfit function. Exploration Geophysics, 53(2), 140-150. doi: 10.1080/08123985.2021.1901573.

[40] Wang HZ, Lin J, Dong XT, Lu SP, Li Y, Yang BJ. 2023. Seismic velocity inversion transformer. Geophysics, 88(4), R513-R533. doi: 10.1190/geo2022-0283.1.

[41] Wang JL, Jaiswal P, Haines SS, Hart PE, Wu SG. 2018. Gas hydrate quantification using full-waveform inversion of sparse ocean-bottom seismic data: A case study from Green Canyon Block 955, Gulf of Mexico. Geophysics, 83(4), B167-B181. doi: 10.1190/geo2017-0414.1.

[42] Wang P, Cui YA, Zhou L, Li JY, Pan XP, Sun Y, Liu JX. 2024a. Multi-task learning for seismic elastic parameter inversion with the lateral constraint of angle-gather difference. Petroleum Science, 21(6), 4001-4009. doi: 10.1016/j.petsci.2024.06.010.

[43] Wang P, Xu CK, Yin ZY, Song SX, Xu CJ, Dai SD. 2024b. A DEM-based generic modeling framework for hydrate-bearing sediments. Computers and Geotechnics, 171, 106287. doi: 10.1016/j.compgeo.2024.106287.

[44] Wang W,Wu NY, Wang, L., Wang LZ, Liu Z, Li YL. 2026. The interaction of gas hydrate with submarine landslides studied with cascading machine learning. Landslides, 23, 1187-1200. doi: 10.1007/s10346-025-02687-5.

[45] Wang XJ, Wu SG, Lee M, Guo YQ, Yang SX, Liang JQ. 2011. Gas hydrate saturation from acoustic impedance and resistivity logs in the Shenhu area, South China Sea. Marine and Petroleum Geology, 28(9), 1625-1633. doi: 10.1016/j.marpetgeo.2011.07.002.

[46] Wei N, Pei J, Li HT, Zhou SW, Zhao JZ, Kvamme B, Coffin RB, Zhang LH, Zhang Y, Xue J. 2024. Classification of natural gas hydrate resources: Review, application and prospect. Gas Science and Engineering, 124, 205269. doi: 10.1016/j.jgsce.2024.205269.

[47] Weissert H. 2000. Deciphering methane’s fingerprint. Nature, 406(6794), 356-357. doi: 10.1038/35019230.

[48] Wu XM, Geng ZC, Shi YZ, Pham N, Fomel S, Caumon G. 2020. Building realistic structure models to train convolutional neural networks for seismic structural interpretation. Geophysics, 85(4), 1-48. doi: 10.1190/geo2019-0375.1.

[49] Xu Z, Hu T, Pang XQ, Wang EZ, Liu XH, Wu ZY, Chen D, Li CR, Zhang XW, Wang T. 2022. Research progress and challenges of natural gas hydrate resource evaluation in the South China Sea. Petroleum Science, 19(1), 13-25. doi: 10.1016/j.petsci.2021.12.007.

[50] Yang LQ, Fomel S, Wang SD, Chen XH, Chen W, Saad OM, Chen YK. 2023. Denoising of distributed acoustic sensing data using supervised deep learning. Geophysics, 88(1), 91-104. doi: 10.1190/geo2022-0138.1.

[51] Yang NX, Li GF, Li TH, Zhao DF, Gu WW. 2024. An improved deep dilated convolutional neural network for seismic facies interpretation. Petroleum Science, 21(3), 1569-1583. doi: 10.1016/j.petsci.2023.11.027.

[52] Yang YX, Zhang XT, Guan Q, Lin YZ. 2022. Making invisible visible: Data-driven seismic inversion with spatio-temporally constrained data augmentation. IEEE Transactions on Geoscience and Remote Sensing, 60, 1-16. doi: 10.1109/TGRS.2022.3144636.

[53] Ye JL, Qin XW, Qiu HJ, Liang QY, Dong YF, Wei JG, Lu HL, Lu JA, Shi YH, Zhong C, Xia Z. 2018. Preliminary results of environmental monitoring of the natural gas hydrate production test in the South China Sea. China Geology, 1(2), 202-209. doi: 10.31035/cg2018029.

[54] Ye JL, Qin XW, Xie WW, Lu HL, Ma BJ, Qiu HJ, Liang JQ, Lu JA, Kuang ZG, Lu C, Liang QY, Wei SP, Yu YJ, Liu CS, Li B, Shen KX, Shi HX, Lu QP, Li J, Kou BB, Song G, Li B, Zhang HE, Lu HF, Ma C, Dong YF, Bian H. 2020. The second natural gas hydrate production test in the South China Sea. China Geology, 3(2), 197-209. doi: 10.31035/cg2020043.

[55] Yin ZY, Linga P. 2019. Methane hydrates: A future clean energy resource. Chinese Journal of Chemical Engineering, 27(9), 2026-2036. doi: 10.1016/j.cjche.2019.01.005.

[56] Yu H, Wang J, Lei ZS, Kuang ZG, Deng W, Lu L. 2024. Characterization of natural gas hydrate constrained by well and seismic data in Qiongdongnan Basin. Energies, 17(21), 5411. doi: 10.3390/en17215411.

[57] Zhan LS, Kang DJ, Lu HL, Lu JA. 2022. Characterization of coexistence of gas hydrate and free gas using sonic logging data in the Shenhu Area, South China Sea. Journal of Natural Gas Science and Engineering, 101, 104540. doi: 10.1016/j.jngse.2022.104540.

[58] Zhan LS Matsushima, J, Liu HT, Lu HL. 2025. Evaluation and modeling of velocity dispersion and frequency-dependent attenuation in gas hydrate-bearing sediments. Marine and Petroleum Geology, 171, 107204. doi: 10.1016/j.marpetgeo.2024.107204.

[59] Zhang J, Sun H, Zhang G, Huang XG, Han L. 2024. Simultaneous physics and model-guided seismic inversion based on deep learning. IEEE Transactions on Geoscience and Remote Sensing, 99, 1-11. doi: 10.1109/TGRS.2024.3443970.

[60] Zhang J, Zhao XY, Chen YK, Sun H. 2023. Domain knowledge-guided data-driven prestack seismic inversion using deep learning. Geophysics, 88(2), 31-47. doi: 10.1190/geo2021-0560.1.

[61] Zhang Y, Meng DC, Zhou YF, Song LW, Dong HL. 2023. Seismic velocity inversion based on physically constrained neural networks. IEEE Transactions on Geoscience and Remote Sensing, 62, 1-17. doi: 10.1109/TGRS.2023.3339783.

[62] Zhu YH, Pang SJ, Xiao R, Zhang S, Lu ZQ. 2021. Natural gas hydrates in the Qinghai-Tibet Plateau: Characteristics, formation, and evolution. China Geology, 4(1), 17-31. doi: 10.31035/cg2021025.