You are Home   »   News   »   View Article

Using uncertainty in reservoir models

Friday, August 23, 2013

A challenge when including uncertainty in reservoir models is the amount of time it takes to build the realisations - and the solution is making the software easier and faster to use, said Tone Kråkenes, Chief Geologist of Reservoir Modelling with Roxar.

She was speaking at the Digital Energy Journal Stavanger event in April 2013, ''Improving decision making with subsurface data''.

Reservoir models often include varying amounts of uncertainty. You might be relatively sure of the position of a horizon if it has been observed in many wells- but have a high degree of uncertainty, if the position was derived from seismic data shot through a salt dome, in a region which has never been drilled.

If it was possible to display the uncertainty in the reservoir model, it would be very helpful to the people who decide where to place wells.

The biggest problem with adding uncertainty to a reservoir model is the time it takes to do it - so one of the biggest solutions might be to make the structure easier to build in the first place, Ms Kråkenes said.

The basic technique of uncertainty modelling is not so complex - instead of creating drawing one fault on your model, you draw a ''base case'' fault (where you think it is most likely to be), and add upper and lower lines around it, so you have an uncertainty envelope.

Easier structural modelling
The first step of managing uncertainty in models might be to make it easier to do structural modelling in the first place.

Structural models are 3D representations of the faults and horizons in the reservoir.

A structural model describes the extent and (composition) of oil and gas, showing the top and bottom of the reservoir, the faults, and potentially the inter reservoir layers.
Structural models can be very simple, but also extremely complex, Ms Kråkenes said.

They are built using all the data available about the reservoir, including seismic and seismic interpretations, well data, conceptual models and field analogues (data about similar fields).

You can add in available data about rock porosity, permeability and saturation, and get an understanding of fluid flow.

Structural models are a basis for reservoir models, where you model how fluid might actually flow through the reservoir. This usually involves ''coarsening'' or reducing the resolution of the model, so you don''t need so much computing power to calculate how the fluid will flow.

With a reservoir model, you can do ''history matching'', comparing the reservoir''s predicted production (made in the past) with the actual production. You can update the model when new data arrives.

You can build workflows showing how people with different roles will work with the software.

Building different structures
The structural modelling work flow is to start with input data, add in faults and horizons, build a fault model and build a horizon model, showing how all the horizons and faults in the reservoir fit together in 3D.

To mark where you think a fault is, you draw a line on the image of seismic data, and do this on a number of different slices of seismic, and then the software will make a best fit of your lines to create a structure of the fault in 3D.

The software can handle all different kinds of faults. ''Faults with complex intersections can be extremely difficult to model,'' she said.

You can use the software to build models of turbidites (geological formations caused by a kind of underwater avalanche) showing the top and the base of the channels.

Managing uncertainty
To create a fault on the software with uncertainty information, you draw the fault in the way described above, and add an uncertainty envelope around the most likely estimate indicating the furthest left or right you think the fault might be.

You do the same for all of the structure.

You then build one based case model and let the computer build a series of alternative models from the uncertainty estimates that you have given.

You can find out which input parameter has the highest effect on the estimated reservoir volume. Usually the structure itself has the highest effect, the position of the faults have less effect.

Does it work?
One audience participant from an oil major said he was sceptical about how possible it was to work with uncertainty. ''I think it''s a complex problem to actually assess uncertainty with any degree of accuracy,'' he said.

For example, errors in the velocity model for rock above the reservoir (overburden) will lead to faults being placed in the wrong place beneath them.

''I think that we shouldn''t fool ourselves by going through this big loop that we have, we should actually understand what is in it all,'' he said.

''I don''t think as an industry in general we will land in a good place assessing uncertainty especially in complex scenarios.''

Ms Kråkenes replied ''so here I agree with you.'' But she also pointed out how the computer system can work through the different possible outcomes and work out which one is most likely, based on all of the input data.

Associated Companies
» Roxar
comments powered by Disqus


To attend our free events, receive our newsletter, and receive the free colour Digital Energy Journal.


Clustering Considerations in the Machine Learning Workflow – Examples with Exploration Data
Philip Lesslar
from Precision DM


Latest Edition Jun-Jul 2020
Jun 2020

Download latest and back issues


Learn more about supporting Digital Energy Journal