Extracting an attribute: from geometry to rock
Learning objectives
- Extract an RMS amplitude along a mapped horizon and choose a window
- Explain why a window that is too thin or too thick both degrade the result
- Describe how an error in the horizon propagates into a quantitative product
- Recognise why a poorer surface favours a wider extraction window
Everything so far has been geometry: where surfaces are. An attribute map is the first thing an interpretation says about the rock.
The idea is simple. Follow the mapped horizon through the volume and, at every bin, reduce the samples near it to one number. Root-mean-square amplitude in a window is the usual choice. If the reservoir's reflection strength varies with its porosity, and here it genuinely does, then that number is a porosity prediction and the map is a prospect-ranking tool.
The window is not a detail
Extract in a zero-thickness window and you read the single sample the horizon lands on. That value depends on where the surface happens to fall between two samples, so the map carries the sample rate as much as the geology. Open the window too far and you average in the reflectors above and below; the map gets smoother, more convincing, and less about your reservoir.
Between those, on this survey, the correlation with true porosity peaks near a wavelet period. You can read that peak directly off the curve beside the map, which is a measurement no interpreter can make on real data, because the true porosity everywhere is exactly what they are trying to estimate.
What a worse surface costs, and how it changes the answer
Switch the extraction to the gridded surface, built from an ordinary eight-line pick grid rather than from the truth, and the correlation drops from about 0.78 to 0.61. The window is positioned by the horizon, so every error in the map moves the window off the event. A quantitative product is never better than the interpretation underneath it.
Then look at where the peak of the curve has moved. On the true surface the best window is around 16 milliseconds. On the gridded surface it is far wider. That is not a coincidence: when the surface is uncertain, a broader window is more forgiving, because it still contains the event where the map is a little wrong. On real data you never know how wrong your surface is, so the wider window is often the safer engineering choice even though it is the worse one in principle.