ML-guided survey design

Part 8 — Compressed sensing & modern methods

Learning objectives

  • Describe ML-optimised survey design vs uniform classical design
  • Explain target-focused weighting and the resulting shot distribution
  • Quote typical cost savings (10–20% at equal target image quality)
  • List published production examples

A classical survey design fixes the shot/receiver grid by imaging rules: binning, fold, azimuth targets, roll-along efficiency. An ML-guided design optimises a different objective directly — “image quality at this specific target zone” — and lets the shot pattern fall out of the optimisation.

Ml DesignInteractive figure — enable JavaScript to interact.

The optimisation loop

Candidate shot sets are evaluated by a cheap forward model (ray-trace, wavefield extrapolation, or neural-surrogate approximation). The objective function weights target-zone imaging quality against survey cost. Gradient-free optimisers — Bayesian optimisation, genetic algorithms, reinforcement learning — explore the high-dimensional design space. The output is a non-uniform shot pattern that looks nothing like a classical grid.

Where ML wins

Target-driven exploration surveys: the client cares about drilling a specific prospect, not about a regional image. A 20% shot-count reduction concentrated on the prospect can hit the same target-zone SNR as a full uniform coverage. Where the client needs regional uniformity (reconnaissance, framework seismic), classical design still wins — ML isn’t magic, it exploits the fact that most of a regional survey’s fold is wasted on uninteresting rock.

Production examples

BP has used gradient-based WAZ-geometry optimisation since 2015. Shell used Bayesian-optimisation-selected OBN node locations in several Gulf of Mexico programmes. Academic work (Eikrem, van Leeuwen, Silvestrov et al., 2018–2022) has published RL-based designs with 10–20% cost savings at equal target-zone quality. The ML is rarely standalone — it sits inside a larger planning flow that still checks classical constraints (fold minima, azimuth envelope, crew productivity).

References

  • Vermeer, G. J. O. (2012). 3D Seismic Survey Design (2nd ed.). SEG.
  • Berkhout, A. J. (2008). Changing the mindset in seismic data acquisition. The Leading Edge, 27(7), 924–938.
  • Cordsen, A., Galbraith, M., Peirce, J. (2000). Planning Land 3-D Seismic Surveys. SEG Geophysical Developments 9.

This page is prerendered for SEO and accessibility. The interactive widgets above hydrate on JavaScript load.