Q: What is the effect of TVD on the model?
A: TVD is a common data type that has the power to capture intricate subsurface features. True Vertical Depth, or TVD, is a measure of how deep a point of a well is directly from the surface of the earth. This feature can be highly predictive when used in combination with formation tops and other petrophysical data types. "The tricky part of using TVD comes down to explaining why it’s predictive, breaking out the various things that could be happening,” Dr. Brendon Hall, VP of Geoscience for Petro.ai clarifies.“ While TVD can be an effective way of training the model, we don’t just stop there.We interpret it in terms of what important geologic drivers are at work behind the scenes.”
“It’s a good baseline feature - and it typically is predictive - but you can’t explain why one TVD is better than another,” Kyle LaMotta, VP of Analytics at Petro.ai adds.“ For example, TVD can encapsulate several other variables like geographic changes. If you’re in one part of the basin, the reservoir is going to be deeper than another part of the basin. TVD is going to be changing. The question becomes is the model learning the relationship of TVD to production or is it learning the change of geography to TVD?
“The other thing that TVD can encapsulate could be lithologic changes. As you get deeper, rocks tend to be thermally more mature and have more hydrocarbons in them up to a point. So, it can also capture some of those reservoir properties.
“TVD has a high correlation with properties related to depth. But it’s hard to explain why TVD is a predictive feature and that lack of explanation makes it a data type that is less desirable than other data types that have a direct explanatory nature. The model’s only as good as the data that goes into it.
“With TVD we want to capture the spatial variation with some sort of petrophysical feature. So that could be porosity or permeability. It could be water saturation and oil saturation, hydrocarbon pore volume or the open pores in rock, what percent of those contain hydrocarbons. Those are also very predictive and much more explainable.
“This TVD graph shows a highest production at around 10,800. If you’re going to drill a well, you’re not going to drill the well to 10,800 feet just because this is the highest value. You’d want to understand, why this is the highest value and what else is going on there. It might be that this depth for a small geographic region also has a high oil saturation. You’d want to be able to explain the physical phenomenon of this depth.
“That’s why we use petrophysical features that are either measured or interpreted from well logs or seismic. It’s much easier if you’re an operator and you can say let’s go find a well that has the highest oil saturation not the well that has the highest TVD.”
“Right,” Hall agrees. "If the earth was like a layer cake model, if it was all flat layers and constant depth across those layers, then TVD would correspond to what layer you’re in. If everything’s flat, TVD would be a great measure of the lithology. Another thing that might be constant in the cake layer model is the deeper you go the higher the pore pressure gets, the higher the stresses get in general geomechanically speaking. That would mean that for a given TVD, stresses and pore pressures are going to be about the same and so that could be predictive of performance as well.
“In that idealized case, you could imagine some reasons why TVD would be a good feature and quite predictive.
“In practice though, the earth doesn’t behave like a perfect layer cake. Although in shale basins there is some layer cake construction. But you do get tilting of these layers. These layers can be dipping down in the earth or even folding back, making the relationship between depth and these other parameters that you want TVD to represent, like lithology and pore pressure, more complicated.
“That’s why we say it’s a tricky one. It’s often predictive but it’s representing some other properties in the earth that are really the drivers behind production. The depth of the well itself doesn’t really matter. It’s not in and of itself predictive.
“There’s a difference between a predictive model and a model that you can explain. You may not have all this petrophysical data available or lithology workups. We can still work with the data you do have. Typically, you know how deep the well is and we can use that as part of the modeling process. TVD can build a baseline. By adding the petrophysical features, such as oil saturation, reservoir thickness, porosity, we can actually explain why one outcome, or one model is better than the other. Accurate prediction relies on an explainable model.”
