Reinforcement-Learning Tracking of Seismic Horizons From Sparse Expert Interpretation: A Case Study on the Achimov Clinoform Deposits of Northern West Siberia

Structural correlation of reflection horizons remains one of the most labour-intensive operations in 3D seismic interpretation, particularly within the clinoform Neocomian sections of West Siberia, where target reflections show variable dynamics, pinch-outs and are complicated by faulting. Because horizon tracking is inherently sequential, we formulate it as a Markov decision process and solve it with reinforcement learning. Observing a local seismic context and the history of previously picked samples, an agent chooses, at each step, the vertical shift of the horizon on the adjacent trace; the reward function encodes geophysical criteria – waveform similarity to the reference reflection, phase consistency, lateral smoothness and fault tolerance. The method is implemented as a two-stage scheme: a policy first correlates the horizon over a grid of reference sections with a user-defined spacing, building a reliable network of anchors, and then densifies the result into a continuous surface across the whole volume. The agent is trained in a synthetic environment produced by forward full-wavefield modelling, with imitation pretraining on reference horizon trajectories; one or two intersecting reference sections suffice as input, and additional interpretation is introduced in complex zones. Tested on Achimov deposits of a field in Northern West Siberia, the approach agrees with manual correlation within 10 ms over conformable intervals. Limitations and directions for development are discussed.

1. Abubakar A., Di H., Li Z., Maniar H., Zhao T. (2024). An artificial intelligence workflow for horizon volume generation from 3D seismic data. The Leading Edge, 43(4), pp. 235–243. https://doi.org/10.1190/tle43040235.1

2. Luo Y., Zhang G., Zhang J., Li Y., Lin Y., Li B., Liang C., Li L. (2023). Sequence-constrained multitask horizon tracking. Geophysics, 88(2), pp. IM15–IM27.

3. Mattos A.B., Civitarese D., Szwarcman D., Oliveira M., Zaytsev S., Semin D.G., Oliveira D.A.B. (2021). Enabling robust horizon picking from small training sets. IEEE Transactions on Geoscience and Remote Sensing, 59(6), pp. 5317–5324. https://doi.org/10.1109/TGRS.2020.3010124

4. Mnih V., Kavukcuoglu K., Silver D., Rusu A.A., Veness J., Bellemare M.G., Graves A., Riedmiller M., Fidjeland A.K., Ostrovski G., Petersen S., Beattie C., Sadik A., Antonoglou I., King H., Kumaran D., Wierstra D., Legg S., Hassabis D. (2015). Human-level control through deep reinforcement learning. Nature, 518(7540), pp. 529–533. https://doi.org/10.1038/nature14236

5. Naumov A.L. (1977). On the methodology for reconstructing the bottom relief of the West Siberian Early Cretaceous basin. Russian Geology and Geophysics, 18(10), pp. 38–47. (In Russ.)

6. Nezhdanov A.A., Ponomarev V.A., Turenkov N.A., Gorbunov S.A. (2000). Geology and petroleum potential of the Achimov strata of West Siberia. Moscow: Academy of Mining Sciences Publ., 247 p. (In Russ.)

7. Ronneberger O., Fischer P., Brox T. (2015). U-Net: Convolutional networks for biomedical image segmentation. Medical Image Computing and Computer-Assisted Intervention – MICCAI 2015, Lecture Notes in Computer Science, 9351, pp. 234–241. https://doi.org/10.1007/978-3-319-24574-4_28

8. Schulman J., Wolski F., Dhariwal P., Radford A., Klimov O. (2017). Proximal policy optimization algorithms. arXiv preprint arXiv:1707.06347. https://doi.org/10.48550/arXiv.1707.06347

9. Sheng S. (2023). Automatic horizon picking using multiple seismic attributes and Markov decision process. Remote Sensing, 15(3), 552. https:// doi.org/10.3390/rs15030552 Stark T.J. (2004). Relative geologic time (age) volumes – relating every seismic sample to a geologically reasonable horizon. The Leading Edge, 23(9), pp. 928–932. https://doi.org/10.1190/1.1803499

10. Su M., Qian F., Cui S., Yuan C., Cui X. (2023). Research on a 3D seismic horizon automatic-tracking method based on corrugated global diffusion. Applied Sciences, 13(10), 6155. https://doi.org/10.3390/app13106155

11. Sutton R.S., Barto A.G. (2018). Reinforcement learning: an introduction. 2nd ed. Cambridge, MA: MIT Press, 552 p.

12. Tschannen V., Delescluse M., Ettrich N., Keuper J. (2020). Extracting horizon surfaces from 3D seismic data using deep learning. Geophysics, 85(3), pp. N17–N26. https://doi.org/10.1190/geo2019-0569.1

13. Wu H., Zhang B., Lin T., Cao D., Lou Y. (2019). Semiautomated seismic horizon interpretation using the encoder–decoder convolutional neural network. Geophysics, 84(5), pp. B403–B417. https://doi.org/10.1190/geo2018-0672.1

14. Wu X., Hale D. (2015). Horizon volumes with interpreted constraints. Geophysics, 80(2), pp. IM21–IM33. https://doi.org/10.1190/geo2014-0212.1

15. Yang L., Sun S.Z. (2020). Seismic horizon tracking using a deep convolutional neural network. Journal of Petroleum Science and Engineering, 187, 106709. https://doi.org/10.1016/j.petrol.2019.106709