Reservoir to Seismic

Structure Plus Texture Plus Convolution

This section fuses the previous two and the convolutional model from Part 2. Start with a reservoir-scale property model: a layered sequence with a channel sand cut into it, the deterministic structure. Add the stochastic fabric that makes rock look real. That combination, expressed in acoustic impedance, is the earth model. Now run the convolutional engine down every column, one independent trace per surface location, and the impedance model becomes a synthetic seismic section.

The two panels are the forward problem laid bare. On the left is the earth you built, structure and texture together. On the right is the seismic that earth would produce. The channel sand, a low-impedance lens on the left, becomes a bright tuned reflection on the right.

The Same Two Trade-offs, at Scale

Every lesson of the course reappears here, now on a whole reservoir. Raise the wavelet frequency and the channel top and base separate, its edges sharpen, but that bandwidth is expensive and real data rarely carries it. Lower it and the reflections tune together into one loop, the resolution limit of Part 1. Add heterogeneity and the clean reflectors acquire the discontinuous, textured character of field data; remove it and they look unrealistically perfect.

Reading this pairing forwards, model to data, is synthetic seismic modelling, the subject of this whole course. Reading it backwards, data to model, is interpretation and inversion. Building the forward model well is what makes the backward reading trustworthy, which is why fit-for-purpose modelling matters so much. The next section builds a recognised community test model, a procedural mini-Marmousi, to stress the engines against known complexity.