Depth Imaging and Velocity Model Building
Structural truth in depth is earned, not rendered. You estimate velocity from the data itself, compute the traveltimes migration runs on, choose an algorithm from Kirchhoff to RTM with your eyes open, and read the image for the artifacts that confess a wrong model. This is the craft of building the model and the image together.
You can read stacking velocity for what it actually measures, pick on semblance and correct the picks with residual moveout, run traveltime tomography and say where ray coverage limits it, choose a migration method for the geology and the budget, diagnose smiles, uncollapsed tails, and curved gathers as velocity errors, and drive the migrate-update loop to a model honest enough to hand to FWI.
Velocity from the data
Every migration velocity begins life as a moveout measurement on a gather; know what stacking velocity is, and what NMO stretch does to the data, before trusting either.
A velocity pick is an interpretation, not a reading; semblance is where the craft of choosing one, and correcting it with higher-order moveout, is learned.
Tomography turns residual moveout into a model update, and it is the engine the whole depth-imaging loop runs on.
Traveltime and rays
Kirchhoff migration and tomography both consume traveltime fields; the eikonal equation is the machine that makes them, classical solver or network.
Invert real traveltimes once and you will never again mistake a tomogram for the earth; ray coverage decides what the data can see and what stays invisible.
Migration
Stacking collapses the gather but leaves dips mispositioned and diffractions open; migration exists because structure lives where stacking cannot put it.
PSTM is the workhorse image most decisions are made on; know which geologies it renders honestly and where time imaging quietly bends the picture.
Kirchhoff PSDM is the first migration that takes your velocity model seriously; from here on, image quality is model quality.
Beams are the working compromise between Kirchhoff rays and full wavefields; tune one and the Fresnel zone stops being an abstraction.
When rays break down under salt you propagate the whole wavefield; one-way methods and RTM are the expensive end of the menu, and you should know exactly what the money buys.
Smiles, uncollapsed tails, and aliasing ghosts are the image telling you what went wrong; interpreters who cannot read them get misled politely.
Migrate, read the gathers, update, migrate again: the loop that builds the model and the image together is the actual job of depth imaging.
Toward the true model
Cycle skipping is what happens when the starting model is not good enough; feel the basin narrow with frequency and you will know why velocity model building exists.
FWI begins exactly where the migration-tomography loop plateaus; its low-frequency playbook tells you when a model is ready to hand over.
No one ever knows the model is right; the QC pack is how you earn the confidence to ship it anyway.