NO. 17 · Seismic Methods

Model the Seismic Response

Type the earth in, get the seismic out. Wavelets, convolution, and the full wave equation, ending in a modeling lab where you build the model and simulate its response. The craft behind every synthetic you will ever trust.

You can build an earth model, choose the right rung of the modeling ladder for the question at hand, generate convolutional and finite-difference synthetics without fooling yourself, and read AVO gathers you made from elastic first principles.

26 competencies · 6 interactive widget challenges · 10 to 15 hours of guided study
For geoscientists who generate or consume synthetic seismic

Why model

The forward problem and the ladder

Forward modeling maps a known earth to its data, and the modeling ladder is the menu of physics you can afford.

What synthetics are for, honestly

Synthetics test algorithms, tie wells, and train networks, and the inverse crime is how they quietly lie; learn both on day one.

Wavelets

The Ricker and the wavelet zoowidget challenge

Every synthetic starts with a wavelet; the Ricker is the workhorse and the zoo tells you what else is in the stable.

Phase and sampling

Phase moves events without changing spectra, and sampling sets the frequency ceiling; both burn people who skip them.

Bandwidth and resolution

Resolution is bandwidth wearing field clothes; this is where the quarter-wavelength rule comes from and when it fails.

The convolutional model

Reflectivity and the convolution machine

Impedance to reflectivity to trace: the two-step machine that makes ninety percent of the synthetics the industry runs on.

Tuning and the wedgewidget challenge

The wedge model is the most instructive synthetic ever built; tuning is why thin beds brighten before they vanish.

A synthetic from a well log

The well tie is the convolutional model earning its keep: log to impedance to synthetic, with noise and the judgment of when convolution is enough.

2D earth models

Building 2D earth models

Faults, layers, and geology-to-impedance: the model you build here is the label your synthetic will carry forever.

Sections and training pairs

Convolving a whole section makes an image; pairing it with the fault mask you drew makes machine-learning training data, and knowing where 2D convolution lies keeps the pairs honest.

The wave equation

Deriving and discretizing

The wave equation from first principles, then onto a grid: finite differences are how the full physics becomes runnable.

Stability and numerical dispersionwidget challenge

CFL is the speed limit and numerical dispersion is the fine for ignoring grid rules; every FD run answers to both.

Absorbing boundaries and shot records

Kill the grid-edge reflections and fire a source: the shot record is the payoff, synthetic field data with known geology underneath.

What the wave equation buys

Diffractions and multiples

The two wave phenomena convolution cannot make; seeing them appear in your own FD run is where the ladder's price makes sense.

Refractions and velocity pushdown

Head waves and gas-cloud pushdown are field headaches you can rehearse in a model where you know the answer.

Convolution versus the wave equation

Same earth, two engines: the definitive side-by-side that tells you which rung of the ladder your question actually needs.

Elastic and AVO

P-waves, S-waves, mode conversion

The earth is elastic whether your model is or not; shear and converted modes are what the acoustic shortcut throws away.

Zoeppritz, Shuey, and the AVO classes

Reflection amplitude versus angle is rock physics speaking; Zoeppritz is the full sentence and Shuey the usable summary.

Synthetic AVO gatherswidget challenge

Making your own gathers, and choosing acoustic or elastic on purpose, is how AVO stops being a plot you were handed.

Anisotropy

Anisotropy and the Thomsen parameters

Real rocks are faster along the grain; epsilon, delta, and gamma are the accepted shorthand for how much.

Model building

Deterministic and stochastic builders

Layer-cake earths teach; stochastic heterogeneity tests. A modeling shop needs both dials.

Standard test models and QC

Marmousi and its siblings are the shared benchmarks of the field, and model QC is what keeps your own earths from lying to you.

Synthetic datasets

Dataset capstone: fault detectionwidget challenge

A labeled fault-detection dataset end to end: the synthetic data product that trains the networks the ML paths deploy.

Dataset capstone: CO2 time-lapse

A plume you can watch grow because you grew it: synthetic 4D for storage monitoring, with the truth attached.

The modeling lab

Capstone: the modeling labwidget challenge

Build the earth, pick the physics, run the simulation, defend the result: the lab is this whole path with your hands on it.

The fit-for-purpose card

The whole modeling ladder on one reference card: which physics, for which question, at what cost.

This page is prerendered for SEO and accessibility. With JavaScript, it hydrates into the live guided path: placement quiz, spaced practice, and interactive widget challenges.