Full waveform inversion: Senior Geophysicist Ameur Hamdane on unlocking the next era of seismic imaging

Ellen Warren: Ameur, you have described Full Waveform Inversion as a “step-change” in resolution and accuracy. From a technical standpoint, what specific innovations in algorithms and computing have made this leap possible in recent years?

Ameur Hamdane: The step change in Full Waveform Inversion (FWI) came from two fronts: advances in physics and leaps in compute. On the physics side, we moved beyond basic least-squares fitting toward more robust misfits such as Huber/L1, phase and envelope objectives, optimal transport distances, adaptive waveform inversion (AWI), and source/receiver extensions (WRI), all of which widened the basin of attraction and reduced cycle-skipping. Angle and domain separation, along with scattering-angle filters, allowed us to emphasize the low-wavenumber content needed for velocity updates. At the same time, multi-parameter TTI/viscoacoustic kernels and approximate Hessians – using illumination compensation, diagonal preconditioning, and L-BFGS/CG with Hessian-vector products – made updates more stable and faster to converge. On the compute side, MPI+GPU with mixed precision, wavefield checkpointing, and compression improved efficiency. Combined with frequency continuation, hierarchical parameterisation, domain decomposition, and cloud bursting, these advances now enable routine delivery of 3D models once limited to research.

EW: One of FWI’s headline applications is subsalt imaging in the Gulf of Mexico. Can you explain how FWI resolves velocity uncertainties in these complex settings where conventional tomography often fails?

AH: Conventional ray-based tomography performs poorly beneath salt because ray coverage is sparse, multipathing is significant, and transmission is limited to a small set of steep trajectories. In contrast, waveform inversion exploits finite-frequency physics, where even weak transmitted, refracted, or diving energy around and through salt contributes valuable low-wavenumber updates to the background model. In practice, we adopt a multiscale strategy: starting with very low frequencies (≤3 - 4 Hz) to recover large-scale sediment velocities, incorporating water-related multiples to strengthen constraints, and progressively refining salt geometry and surrounding sediments.

Reflection FWI (RFWI) is particularly important in areas lacking diving waves, as it extracts long-wavelength information from reflected energy. For salt itself, “salt flooding” combined with level-set or topological updates enables realistic movement of salt boundaries, avoiding ineffective local adjustments. Incorporating Q compensation and TTI anisotropy is essential to prevent attenuation or anisotropy effects from being misinterpreted as velocity.

Collectively, these advances deliver sharper base-of-salt imaging, reduced depth mis-ties, and less interpretive intervention in RTM or LSRTM.

EW: Elastic FWI is gaining attention but remains computationally demanding. What breakthroughs are needed before multi-parameter elastic inversion becomes routine in industry workflows?

AH: For multi-parameter elastic FWI to become a routine industry tool, several hurdles still need to be cleared. Elastic simulations are much heavier than acoustic ones-often five to ten times more expensive-so faster, more efficient computing is critical. Advances such as GPU clusters, cloud HPC, and smarter solvers that reduce memory use will help bring turnaround times down to practical levels. On the algorithm side, we need methods that can better separate parameters and prevent cross-talk, supported by stronger regularisation techniques, including machine learning approaches. Multiples, once treated as noise, can now be harnessed to add valuable constraints and improve stability. Equally vital is embedding practical uncertainty measures, ensuring that results are not only sharper but also reliable. In the end, elastic FWI will shift from research into everyday practice only when these advances converge into workflows that are scalable, cost-effective, and straightforward to apply.

EW: Reflection FWI has been called a way to extend inversion to reservoir depths. How does it differ from conventional FWI, and what successful field applications have you seen so far?

AH: Traditional FWI depends on diving waves to capture low-wavenumber velocity updates, while reflections were long dismissed as noise in the misfit. Reflection FWI (RFWI) turns that idea around by using reflected energy to refine the background model. By separating scattering angles and penalising image-domain mismatches, it extracts transmission-like information that standard approaches leave behind. In practice, we build extended images-such as subsurface offsets or angle gathers-and force the inversion to drive spurious lags toward zero, since misfocusing signals an inaccurate velocity field. This lets us recover low-wavenumber structure even beneath complex bodies where diving energy is absent. Field applications have demonstrated the effectiveness of RFWI in settings such as Gulf of Mexico subsalt, North Sea chalk, and Brazilian presalt, provided that reflection coverage is dense and of high quality. However, RFWI is not a turnkey solution. It demands careful multiple suppression, reliable source signatures, and proper treatment of anisotropy and Q effects. When these conditions are met, RFWI significantly reduces the gap between velocity building and imaging, cutting months from the overall model-building cycle.

EW: AI is often touted as a way to accelerate FWI. In practice, where is AI already proving useful - initial model building, cycle-skipping mitigation, or workflow automation?

AH: The clearest benefits so far are in workflow efficiency. AI is helping automate tasks that used to take days by enabling faster data conditioning, signature stabilisation, and streamlined processing pipelines. Cloud platforms can now handle seismic updates in smaller chunks, making it possible to run lightweight inversions that deliver background velocity refreshes within the same shift. While still in early adoption, this approach has already shown impact in 4D reservoir monitoring, where timely velocity updates improve the interpretation of subtle production-related changes. Looking ahead, the same automation principles could support drilling workflows, where rapid velocity updates would be valuable for hazard prediction and decision-making.

EW: How close are we to achieving near-real-time FWI updates while drilling? What advances in cloud computing, edge devices, or algorithm design are still required?

AH: Near-real-time FWI while drilling is achievable today, but only in tightly constrained scenarios rather than at the scale of a full field. The most practical application right now is look-ahead imaging with Reverse Time Migration (RTM) or Least-Squares RTM (LSRTM) using drill-bit noise, which gives operators a view of structures just ahead of the bit. To make FWI itself work in this context, the problem has to be reduced to a 1 - 3 km window around the bit, at low frequencies, and processed on rugged GPUs at the edge. The main bottlenecks remain limited telemetry bandwidth, the very high data volumes from Distributed Acoustic Sensing (DAS), the variability of the drill-bit source, and overall latency. Progress is being made with setups that combine Ocean Bottom Node (OBN) or DAS recording, edge denoising, and robust misfit functions, but to make this routine we still need faster downhole links, better DAS compression, and standardised workflows.

EW: You’ve written about FWI’s role in CCS, geothermal, and hydrogen storage. Technically, what makes FWI particularly suited for these new applications compared to conventional seismic methods?

AH: Two things make FWI particularly well suited to CCS, geothermal, and hydrogen projects: absolute properties and repeatability. For CCS and hydrogen, you need confidence in seal integrity, reservoir connectivity, and how saturation and pressure evolve over time. FWI recovers high-fidelity, spatially continuous velocity models – and with elastic or attenuation effects, sensitivity to fluids and temperature – that tie directly to rock physics. This reduces the amount of empirical adjustment between seismic measurements and the properties engineers simulate. For monitoring, FWI’s 4D sensitivity to small background changes, especially when we invert phase or envelope content, allows us to resolve CO2 plume migration or thermal fronts with greater detectability than stack differencing alone. In geothermal projects, elastic FWI helps map fracture density and orientation through anisotropy and shear velocity, while also capturing temperature-dependent attenuation. And for hydrogen, caprock and fault characterisation are critical; high-resolution models with uncertainty bounds directly support risk management.

Importantly, these projects are often shallower than deepwater exploration, so bandwidth and aperture requirements are easier to meet with modern nodal arrays.

EW: Many researchers are exploring joint inversion with EM or gravity. How realistic is this in industry-scale projects, and what advantages could it unlock for complex reservoirs?

AH: Joint inversion combining seismic with electromagnetic (EM) or gravity data is practical in targeted pilots and is scaling where the business case is strong. Data volumes and computational burden remain high, but high-performance computing (HPC) and more mature solvers make it manageable. The payoff is a coherent, physics-consistent model: seismic resolves structure and elastic contrasts; EM adds resistivity and fluid sensitivity; gravity constrains density. Used together, these constraints reduce non-uniqueness and tighten uncertainty – particularly in complex settings such as subsalt deepwater and CCS plume monitoring – where single-physics approaches fall short.

EW: Looking 10 years ahead, do you see FWI evolving toward a “full physics” inversion (anisotropy, attenuation, multi-parameter) – and if so, what will that mean for exploration and production decision-making?

AH: In my view, the next decade will push FWI toward “full physics”: explicit anisotropy, attenuation (Q), and multi-parameter elasticity. Two factors will make this realistic: steady compute gains (massively parallel GPUs and elastic cloud scale) and stronger algorithms (parameter decoupling, Hessian-aware updates, and built-in uncertainty quantification). Just as important, deliverables will shift from velocity cubes to decision products – seal-breach probabilities, geopressure risk maps, plume-migration envelopes, and depth at TD with confidence bounds. Linking FWI outputs directly to reservoir workflows will shorten the path from prospect to development and shrink subsurface uncertainty. But this will not be push-button: success still depends on credible priors, disciplined regularisation, and rigorous QC. Focus on the physics that matter for each decision, and you get fewer surprises, faster convergence, and choices anchored in quantified uncertainty rather than argument.

To finish, FWI is not a silver bullet; it is a disciplined way to extract macro-model truth from the data you already paid to acquire. If you give it bandwidth, aperture, and an honest objective, it will pay you back in fewer surprises and better wells. The rest (AI, cloud, elastic bells and whistles) are multipliers. The core is still physics, quality control, and the courage to trust a model when it earns it.

Read the article online at: https://www.oilfieldtechnology.com/digital-oilfield/06112025/full-waveform-inversion-senior-geophysicist-ameur-hamdane-on-unlocking-the-next-era-of-seismic-imaging/