GEOINT
Investors

Technology

Many ways to measure, one model to explain them

Different physical measurements constrain different properties of the ground. The program studies how to combine them, and how to say what each one can and cannot tell us.

What each measurement constrains

Each modality speaks to one property of the ground; none speaks to all.

  • Seismic and fibre sensingWave speed and structure
  • Controlled-source EM and magnetotelluricsElectrical resistivity
  • Gravity and gradiometryDensity
  • Magnetic surveysMagnetic susceptibility
  • BathymetrySeabed surface
  • Wells and coresGround truth
Shared model of the subsurface

Deep-water imaging challenges

Why imaging deep structures beneath deep water is hard, described qualitatively and without promises.

  1. Water depth against target depth

    The deeper the water and the target, the longer and weaker the path the waves travel, and the less the data constrain what lies at the far end.

  2. Frequency, resolution and attenuation

    High frequencies give finer detail but fade fastest in rock, so detail is traded for reach as depth grows.

  3. Salt and complex overburden

    Strongly contrasting bodies bend and scatter waves, which can hide or distort the structures below them.

  4. Velocity uncertainty

    Imaging needs a model of how fast waves travel. Its errors move and blur features and can never be removed completely.

  5. Non-uniqueness

    Different earth models can explain the same data, so we keep competing interpretations instead of a single answer.

A · Advanced seismic imaging and scientific computing

Recovering structure from recorded waves, with the physics respected and the uncertainty carried into the result.

  • OBN and OBS multi-component data

    Concept

    Seabed nodes record several components of the wavefield, which may carry richer information than one sensor type.

  • Acoustic and elastic propagation

    Concept

    Simulating how waves move through water and rock is the base of every method that follows.

  • Full-waveform inversion

    Concept

    Adjusts a model until simulated waves match recorded ones, with known risks such as cycle-skipping.

  • Reverse-time migration

    Concept

    Places reflectors by sending recorded waves backwards through a velocity model.

  • Least-squares migration

    Concept

    Treats imaging as an inverse problem to reduce artefacts caused by uneven coverage.

  • Tomography and velocity model building

    Concept

    Builds the speed model that imaging depends on, and shows how well the data constrain it.

  • Anisotropy and attenuation

    Concept

    Accounts for direction-dependent speed and for energy lost in rock; both distort images if ignored.

  • Denoising and reconstruction

    Concept

    Removes noise and fills gaps in the data while preserving the physics.

  • Physics-informed machine learning and neural operators

    Concept

    Learned components constrained by wave physics and tested on held-out geology before any use.

  • Bayesian uncertainty

    Concept

    Estimates the range of models that still fit the data instead of a single answer.

  • Evidence-linked 3D and 4D twin

    Concept

    A model of the subsurface in which every feature links back to the data and code that support it.

Wave physics and inversion

Four ideas that turn recorded waves into a model, each with its limits.

Forward modelling

Concept
Forward modellingCross-section: a source at the surface sends waves down through three layers; they reflect from an interface and return to a line of receivers.SourceReceiversReflectorLayered model
Forward modelling predicts the recording from a model of the ground. It is only as good as the model and the physics it assumes; real rock is more complex than the layers drawn here.

Migration and reverse-time imaging

Concept
Migration and reverse-time imagingCross-section: waves recorded at five receivers are sent backwards in time, and their circular fronts converge on one point that marks a reflector.Recorded wavefieldWaves sent backwardsImage point
Reverse-time migration sends recorded waves backwards through a velocity model, so reflectors appear where the fronts focus. A wrong velocity model blurs or misplaces them.

Full-waveform inversion

Concept
Full-waveform inversionFlow: a model predicts a recording, the prediction is compared with the real data, and the model is updated. The loop repeats.ModelPredictCompareUpdate
The loop adjusts the model until predicted and recorded waves agree. If the starting model is too far off, the match can lock onto the wrong wave cycle (cycle-skipping), and many models can fit equally well (non-uniqueness).

Joint inversion

Concept
Joint inversionFlow: seismic, electrical and gravity data all feed one shared model, whose structure they are encouraged to agree on.SeismicElectricalGravityShared model
Several data types constrain one model, and a structural coupling encourages them to agree on where boundaries lie. The coupling can force agreement that is not real if the assumed link between properties is wrong.

B · Multi-physics geophysical intelligence

Combining different physical measurements so each constrains what the others cannot.

Adding a modality is accepted only when controlled experiments show it adds information.

  • Controlled-source electromagnetics (CSEM)

    Concept

    Responds to electrical resistivity, which helps separate fluids from rock types.

  • Magnetotellurics

    Concept

    Natural electromagnetic fields probe resistivity at larger scale, with smooth, low-resolution images.

  • Gravity and gradiometry

    Concept

    Shows density contrasts at large scale; many density distributions fit the same data.

  • Magnetics

    Concept

    Constrains magnetic susceptibility, useful for volcanic and basement rocks.

  • Passive seismic and ambient noise

    Concept

    Uses natural and ambient vibrations as a low-cost source of information.

  • Distributed acoustic sensing

    Concept

    Records strain along optical fibre as waves pass; sensitivity depends on how well the fibre couples to the ground.

  • Well and petrophysical evidence

    Concept

    Wells and cores give direct but sparse truth that anchors and tests the models.

  • Joint inversion

    Concept

    Several data types constrain one shared model; a wrongly assumed link between properties can force false agreement.

  • Physics-constrained fusion

    Concept

    Combines data types under physical rules rather than by blending images.

  • Information-value survey optimisation

    Concept

    Asks which further survey would reduce uncertainty most, before it is acquired.

C · Nano, quantum and experimental sensing

A research workstream on sensor concepts that must prove themselves before they are relied on.

Evaluation criteria

  • Feasibility
  • Sensitivity
  • Operating depth
  • Technology readiness level
  • Marine limits
  • Cost

  • Nano-MEMS sensors

    Concept
    Why it matters
    Small, low-power sensors could be deployed in large numbers.
    What must be proven first
    Stable response, calibration and survival in seawater over time.

  • Graphene and nanomaterials

    Concept
    Why it matters
    New materials may give sensors with higher sensitivity or lower noise.
    What must be proven first
    That the gain survives manufacturing, packaging and the marine environment.

  • Photonic sensing and next-generation DAS

    Concept
    Why it matters
    Better optical readout could improve sensitivity along existing fibre.
    What must be proven first
    Repeatable gains over current interrogators on the same cable.

  • Diamond NV magnetometry

    Concept
    Why it matters
    Compact devices that sense magnetic fields at room temperature.
    What must be proven first
    Useful sensitivity on a moving platform in a noisy marine setting.

  • Quantum gravimetry and gradiometry

    Concept
    Why it matters
    Atom-based instruments hold the prospect of drift-free gravity measurements.
    What must be proven first
    Operation on a moving vessel and a clear gain over current gravimeters.

  • SQUID magnetometers

    Concept
    Why it matters
    They respond to extremely weak magnetic fields.
    What must be proven first
    A practical way to cool and shield them for marine use.

  • Acoustic metamaterials

    Concept
    Why it matters
    Engineered structures could shape how sensors capture sound.
    What must be proven first
    A measurable benefit over conventional designs in water.

  • Advanced transducers

    Concept
    Why it matters
    Better sources and receivers could widen the usable bandwidth.
    What must be proven first
    Bench and tank tests that show a gain without a loss of robustness.

Safe AI principles

Five rules for any learned component in the program.

  1. Physics residual checks

    Every learned output is checked against the wave equation, and the leftover misfit is reported.

  2. Held-out geology

    Models are tested on geology they never saw during training, separated by survey and setting.

  3. No structure without data support

    A feature that no measurement requires is flagged, not drawn.

  4. Human interpreter sign-off

    A qualified person reviews every result before it is used, and the decision stays with people.

  5. Explanations attached

    Each result carries the data, parameters and checks that produced it.