Hybrid convolutional neural network-graph attention network-gradient boosting decision tree model for seismic impedance inversion prediction

Abstract

Seismic impedance inversion is essential for reservoir characterization but remains challenging in complex geological environments due to the inherent limitations of conventional methods. This study proposes a hybrid deep learning framework integrating a convolutional neural network (CNN), a graph attention network (GAT), and a gradient boosting decision tree (GBDT) to achieve high-resolution impedance inversion. The CNN extracts local structural features from seismic waveforms, the GAT captures long-range geological dependencies through self-attention between traces, and the GBDT performs robust non-linear regression for final prediction. Extensive evaluations on synthetic and field datasets demonstrate that the method achieves a root mean square error of 285 m/s·g/cm3 on the Society of Exploration Geophysicists salt model, representing a 15.2% improvement over XGBoost and a 32.1% improvement over sparse spike inversion. The framework performs particularly well in complex regions, achieving a 22.7% error reduction at salt boundaries and a thin-bed detection rate of 92% for layers exceeding 4 m in thickness. Statistical uncertainty quantification indicates 94.2% coverage of true impedance values within 95% confidence intervals. In practical applications, the method reduces interpretation time by 40% while maintaining reservoir thickness prediction errors within ± 3 m, demonstrating strong robustness and operational value for seismic interpretation.

Keywords

Convolutional neural networks / Graph attention networks / Gradient boosting decision tree / Seismic impedance inversion / Deep learning / Geological exploration

Funding

This research was financially supported by the Scientific Research Fund of Institute of Seismology, China Earthquake Administration and National Institute of Natural Hazards, MEM, (No. IS202226322); 2025 Doctoral Special Support Program Project of Chengdu Jincheng College (NO.2025JCKY(B)0018); the Key Research Base of Humanities and Social Sciences of the Education Department of Sichuan Province, Panzhihua University, Resource based City Development Research Center Project (NO.ZYZX-YB-2404); Mahasarakham University; and the Open Fund of Sichuan Oil and Gas Development Research Center (NO.2024SY017).

References

[1] Falade AO, Amigun JO, Abiola O. Hydrocarbon prospective study using seismic inversion and rock physics in an offshore field, Niger Delta. Discov Geosci. 2024; 2(1):24. doi:10.1007/s44288-024-00030-4

[2] Zhang ZX, Gong F, Kozlovskaya E, Aladejare A. Characteristic impedance and its applications to rock and mining engineering. Rock Mech Rock Eng. 2023; 56(4): 3139-3158. doi:10.1007/s00603-023-03216-3

[3] Su Y, Cao D, Liu S, Hou Z, Feng J. Seismic impedance inversion based on deep learning with geophysical constraints. Geoenergy Sci Eng. 2023; 225:211671. doi:10.1016/j.geoen.2023.211671

[4] Lin Y. Multi-scale seismic impedance inversion based on Transformer model and deep learning. Eng Res Express. 2025; 7(1):015209. doi:10.1088/2631-8695/ada48d

[5] Wu X, Yan S, Bi Z, Zhang S, Si H. Deep learning for multidimensional seismic impedance inversion. Geophysics. 2021; 86(5):R735-R745. doi:10.1190/geo2020-0564.1

[6] Li M, Yan XS, Zhang MZ. A comprehensive review of seismic inversion based on neural networks. Earth Sci Inform. 2023; 16(4):2991-3021. doi:10.1007/s12145-023-01079-4

[7] Akingboye AS. Electrical and seismic refraction methods: Fundamental concepts, current trends, and emerging machine learning prospects. Discov Geosci. 2025; 3(1):87. doi:10.1007/s44288-025-00169-8

[8] Leite EP, Vidal AC. 3D porosity prediction from seismic inversion and neural networks. Comput Geosci. 2011; 37(8):1174-1180. doi:10.1016/j.cageo.2010.08.001

[9] Wallick BP, Giroldi L. Interpretation of full-azimuth broadband land data from Saudi Arabia and implications for improved inversion, reservoir characterization, and exploration. Interpretation. 2013; 1(2):T167-T176. doi:10.1190/INT-2013-0065.1

[10] Chen H, Innanen KA, Chen T. Estimating P- and S-wave inverse quality factors from observed seismic data using an attenuative elastic impedance. Geophysics. 2018; 83(2):R173-R187. doi:10.1190/geo2017-0183.1

[11] Okeugo CG, Onuoha KM, Ekwe CA, Anyiam OA, Dim CIP. Application of crossplot and prestack seismic-based impedance inversion for discrimination of lithofacies and fluid prediction in an old producing field, Eastern Niger Delta Basin. J Pet Explor Prod Technol. 2019; 9(1):97-110. doi:10.1007/s13202-018-0508-6

[12] Azevedo L, Demyanov V. Multiscale uncertainty assessment in geostatistical seismic inversion. Geophysics. 2019; 84(3):R355-R369. doi:10.1190/geo2018-0329.1

[13] Dai R, Yin C, Zaman N, Zhang F. Seismic inversion with adaptive edge-preserving smoothing preconditioning on impedance model. Geophysics. 2019; 84(1):R11-R19. doi:10.1190/geo2016-0672.1

[14] Thibodeaux B, Ramsay T, Segovia F, Hernandez L, Ibrahim M. Closed-Loop Integrated Time-Lapse Seismic Feasibility in Amberjack Field-Deepwater Offshore Gulf of Mexico. In: Paper Presented at: SPE Reservoir Characterization and Simulation Conference and Exhibition. Dayeh University, Delta, Syria. SPE-196670-MS; 2019. doi:10.2118/196670-MS

[15] Zhang J, Li J, Chen X, Li Y. Geological structure-guided hybrid MCMC and Bayesian linearized inversion methodology. J Pet Sci Eng. 2021; 199:108296. doi:10.1016/j.petrol.2020.108296

[16] Ma Q, Wang Y, Ao Y, Wang Q, Lu W. UB-Net: Improved seismic inversion based on uncertainty backpropagation. IEEE Trans Geosci Remote Sens. 2022; 60:1-11. doi:10.1109/TGRS.2022.3174911

[17] Ning C, Wu B, Wu B. Transformer and convolutional hybrid neural network for seismic impedance inversion. IEEE J Sel Top Appl Earth Obs Remote Sens. 2024; 17:4436-4449. doi:10.1109/JSTARS.2024.3358610

[18] Xiong W, Ji X, Ma Y, et al. Seismic fault detection with convolutional neural network. Geophysics. 2018; 83(5): O97-O103. doi:10.1190/geo2017-0666.1

[19] An Y, Guo J, Ye Q, et al. Deep convolutional neural network for automatic fault recognition from 3D seismic datasets. Comput Geosci. 2021; 153:104776. doi:10.1016/j.cageo.2021.104776

[20] Cao C, Wang X, Yang F, et al. Attention-driven graph convolutional neural networks for mineral prospectivity mapping. Ore Geol Rev. 2025; 106554. doi:10.1016/j.oregeorev.2025.106554

[21] Yao G, Zhang Q, Zhang H, Li Y. Non-local self-similarity guided graph attention network for DAS-VSP noise and signal separation. J Appl Geophys. 2025; 241:105835. doi:10.1016/j.jappgeo.2025.105835

[22] Zhou J, Gao Y, Lu J, Yin C, Han H. An ensemble learning algorithm for machinery fault diagnosis based on convolutional neural network and gradient boosting decision tree. J Phys Conf Ser. 2021; 2025(1): 012041. doi:10.1088/1742-6596/2025/1/012041

[23] Qian S, Peng T, Tao Z, et al. An evolutionary deep learning model based on XGBoost feature selection and Gaussian data augmentation for AQI prediction. Process Saf Environ Prot. 2024; 191:836-851. doi:10.1016/j.psep.2024.08.119

[24] Li Q, Luo Y. High-resolution Bayesian sequential impedance inversion. In: Paper Presented at: SEG International Exposition and Annual Meeting; 2020. doi:10.1093/jge/gxac035

[25] Zhao T, Chen G, Suraphee S, Phoophiwfa T, Busababodhin P. A hybrid TCN-XGBoost model for agricultural product market price forecasting. PLoS One. 2025; 20(5):e0322496. doi:10.1371/journal.pone.0322496

[26] Zhao T, Chen G, Pang C, Busababodhin P. Application and performance optimization of SLHS-TCN-XGBoost model in power demand forecasting. Comput Model Eng Sci. 2025; 143(3):2883-2917. doi:10.32604/cmes.2025.066442

[27] Bakurov I, Buzzelli M, Schettini R, Castelli M, Vanneschi L. Structural similarity index (SSIM) revisited: A data-driven approach. Expert Syst Appl. 2022; 189:116087. doi:10.1016/j.eswa.2021.116087

[28] Brunet D, Vrscay ER, Wang Z. On the mathematical properties of the structural similarity index. IEEE Trans Image Process. 2011; 21(4):1488-1499. doi:10.1109/TIP.2011.2173206