The reservoir layering is a key element and one which the geologist has a large influence (See Slideshow, Slide 3). In general, avoid overcorrelation at the start of a project. Existing reservoir schemes may be of historical nature and may not be optimum for reservoir modelling.

Well correlation is a subject in itself, but the overall scheme should be consistent with the geological model of the area or unit being correlated. A surprising number of additional factors may need to be taken into account. Elements that show what looks similar from well to well include electric logs, cores, sidewall cores and geochemistry. Biostratigraphy shows what is the same age and should be the basis of well correlation, however it rarely has the resolution in uncored wells to give enough markers. Elements extracted from seismic are larger scale and more difficult to match to well data, but may show elements of similar properties such as thick high porosity sands, or may show elements of similar age such as unconformities or channel edges. Available dynamic data is also important: RFT pressures, PLT data give good indications of connectivity and the best flow intervals and this will also guide property modelling.

In general it is best to correlate wells according to the age of the stratigraphy, this means in some reservoirs that the lithology may vary along a particular correlation line. Having said that, try and make the picks at the boundary of flow units. If there are any interconnected bodies such as channels, these do not need to be correlated at this stage, often this can be handled by facies modelling.

What is the sedimentological interpretation?

It may be that the correlated layers from well logs will be sufficient for volumetric purposes and well planning. It is often revealing to make a simple coarse model so that the facies and property trends can be examined. For many studies and for reservoir simulation, it is usually necessary to add additional layers so that the properties assigned to individual cells behave in a realistic way dynamically. The number of layers needs to capture the changes in properties will minimal averaging of good and poor layers (See Slideshow, Slide 4).

How are the layers going to be generated?
Often a simple proportional subdivision is practical with layers all across the model

Is there significant downcutting of channels or onlap?
Onlaping cells may be more correct geologically, can be difficult to handle correctly in the static and dynamic model

Facies Modelling

Introduction

The key element of geological modelling packages is the ability to assign cells in the model to a particular facies code (See Slideshow, Slide 5). The principle is to fill 3-D space with a facies according to geological thinking honouring the facies that have been picked in the well data (See Slideshow, Slide 6). Of all the steps, this one can create the worst result if due considerations are not taken. First, it is not essential to define facies. The main purpose is to enable properties to be modelled according to the predictions away from well data. The facies need to have fairly distinct properties and or 3-D shapes, otherwise there is no reason to model them differently. In fact it can be detrimental to have too many facies as in the subsequent property modelling stage, each facies only uses property data for its own facies. This can lead to unwanted discontinuities in the properties. There is no need for a different zone to have a different facies except for convenience in naming, as each zone is usually modelled separately. If there is a clear non-net facies such as shales, then one option is to simply make a model with just net and non-net facies.

It is important to note that even if facies are not modelled, the result of the property modelling will be superior to a model that simply averages the well data. This is because the 3-D elements are taken into account and this cannot be done in simple average maps.

Facies from Cores

If cores or core photos are available then these should be used as a basis for the facies subdivision, but unless lots of wells are cored, then the facies scheme needs to be simplified so that facies can be estimated from electric well logs. The non net facies are generally of less importance than differences in high and medium permeability facies.

Does the core facies that have been described fit in with the current thinking?
The facies should not be based on small-scale features that cannot be modelled from well to well.

Is it consistently described from well to well?
Can facies be matched to electric logs?
Do core plug data match porosity calculated from electric logs?
Is the core - wireline depth match adequate?
Does the estimate of net match that of electric logs?
Net is typically greater than 1mD for oil, greater than 0.1 or 0.01 mD for gas.

Facies from logs

Facies that can be derived from logs will obviously depend on the logging suites that were acquired on various wells in the field. If core is available, it is very important to understand the relationships between core and log facies. If there are a lot of uncored wells, then facies definition will be biased more towards what can be seen in logs. While log show features down to quite small scales, of 1 foot or so, interbedded units make be averaged out when compared with core logs. This is a particular problem if there are higher permeability core plugs in the middle of a sequence that looks shaley on logs.

How many wells have a good log set?
Are there any wells with suspect logs or casing points in the middle of the reservoir?
When picking facies, beware of fluid effects such as changes in density at GOC

Is normalisation of the logs required?
A common issues is variation in GR in different hole sizes

How many facies can really be identified from the available logs?
Automatically generated electrofacies can be used as a first pass, but will normally need considerable editing according to geological thinking and to remove intervals that are too thin to be modelled.

Facies from seismic

Seismic waves respond to larger-scale features and sometimes events within the reservoir can be seen and this has proved very useful on some fields. However while features can be seen on the seismic, matching these responses to well data and supposed fluid movements in the reservoir can be enigmatic. There are many techniques, but none of them are automatic, considerable manual interpretation may be required. In some instances, these methods can be used to define broad areas of better reservoir quality or broad channel belts.