Permeability in Carbonates 5-8 March 2006

The following notes summarize the main points raised during the talks and breakout sessions of the Conference. Where a consensus was reached this is also included. These notes are intended to ensure that the volume of good work presented at the Conference is not lost – however since the talks were Confidential – only general and not specific issues can be mentioned.

Permeability is traditionally defined as the capacity of a rock to allow fluid to flow through it.
Some speakers argued that this definition is an oversimplification since in 100% saturated rock it only allows fluid to enter by means of a single viscosity constant. This is known to be an oversimplification in the case of gases – and some speakers argued that it could also be the case for other fluids.

Permeability is required primarily as an input into the reservoir model to allow predictions of future field performance to be made. Frequently, in order to achieve a good history match Reservoir Engineers apply substantial permeability modifiers to the geologically derived permeability field. Modifiers ranging between 0.1 and 10 are common. If significant permeability multipliers are required then much of the predictive power of the petrophysical-derived permeability is lost and the value of the efforts described below is placed in doubt.

The key issues to understand in order to create a good reservoir model are:
Identify horizontal permeability conduits “thief zones”
Vertical permeability baffles/barriers “tight zones”

If these can be predicted and inserted into the model much of the work is done – and the intermediate background permeability field is of lesser importance and spending great efforts on complex predictive techniques requires careful justification.

There has recently been significant progress on the use of X-ray tomography techniques to produce high-resolution 3D images of the internal structure of the pore space. Given this detailed view of the pore structure, it is possible to mathematically model flow through this pore space and to compute the permeability that would be expected based on it.

It is possible to rotate the image of the pore space in 3D and to understand its interconnectivity. It is commonly found that the majority of the flow occurs through a limited number of unusually well connected (but not necessarily large) pores. It is not unusual for 90% of the flow to occur through 10% of the pore network. Flow rates through the more poorly connected pores may then be extremely slow – with corresponding slow, but not necessarily low oil recovery.

It was demonstrated that some formations (notably the Arab D limestones) have a particularly simple pore size distribution and the histogram of pore size distributions shows clear and separate peaks. The preferred pore sizes have been termed “porositons”; and relate to the differing origins of the matrix that the rock is composed of. The porositons relating to large pores carry the great majority of the flow – but not necessarily the great majority of the oil in place. The porositons relating to the smaller pores (often cavities in the microfossils within the matrix) still appear to have good connectivity with the overall porosity network. These porositons are so small that they may remain water filled up to several hundred feet above the free water level, and they have a tendency to remain water wet well above this level. This means that the oil within them is easily expelled by incoming water – and these small pores frequently allow for excellent overall recovery.

Recovered core only represents a minute fraction (0.0001% or less) of the reservoir – and as one speaker pointed out — since it has been removed — it is the only rock that definitively does not contribute to flow in the reservoir. Typically, log data is available for a larger volume of the reservoir and so there is great need to utilize this abundant data to estimate permeability in uncored wells.

Poroperm plots in carbonates typically show 2-3 decades of permeability variation at any specified porosity – while the corresponding variation in sandstones is 1.5-2 decades. Locally carbonates vary so much that neighboring core plugs taken from apparently similar rock may have permeabilities that differ by a factor of 10 or even 100. Minipermeameter data shows that these very large permeability variations are present even on a sub-centimeter scale. One suggestion on overcoming this variability was to carry out experiments on whole core samples (expensive) – another (cheaper) was to accept that the variability is localized and to use averaging techniques to allow for it.

Note that massive short scale heterogeneity can average out to highly uniform effective properties on a larger scale. (Water has a uniform density – but density measurements — if they were possible to make — at a sub-atomic scale will show great heterogeneity – and even zero density at points between the atoms.)

Scatter Reduction in the Poroperm Plot
A number of methods have been proposed to reduce the scatter typically seen in Poroperm plots. They all rely on clustering the data using differing techniques and assumptions. Some popular methods that were discussed included:

Categorize using the Lucia technique
Categorize using the Flow Zone Indicator
Split into sub-layers
Categorize using rocktype

Two critical issues for determining the success of these techniques were firstly the 3D spatial location of the clusters. If the points composing the clusters are scattered randomly in reservoir space and not geologically predictable then little has been gained. Secondly, the accuracy of the predictions needs to be assessed by means of a numerical blind test – and compared against some other benchmark technique – typically a sub-layer assignment. A simple plot of predicted permeability against core permeability may lead to differing and subjective interpretations as to the quality of the prediction.

Common Prediction Techniques
Neural Networks
Fuzzy Logic
Straight line fits on poroperm plot

Several authors had used NMR data to estimate pore size distribution and then split the pores into various categories (similar to the porositons approach but not so restricted in the rocks to which it can be applied). Commonly the pores were assigned to macro, meso or micro categories. Different interconnectivities between these pores could then be assumed and consequently permeability algorithms created which took into account the pore type distribution.

In some cases, the use of image logs was found to assist in the assignment of texture – particularly to allow an understanding of the impact of vugs.

On occasions, barite in the mud may lodge in the larger pores/vugs. This will impact the reading of the PEF curve – consequently a correlation could be found between high PEF values and macro/vuggy porosity. This correlation could be used in conjunction with conventional logs to improve the estimation of permeability.

Formation pressure testing tools can provide a permeability (strictly mobility) estimate. The volume sampled was typically greater than from a core plug, but less than from conventional open hole logs. Since mobilities derived from analysis of formation pressure testing data come from measurements based on actual flow through the rock they are potentially more reliable than information derived from other logs (except arguably for Stoneley derived data) where the prediction is only correlative and not based on actual physical flow in the formation.

The value of formation pressure testers using multiple probes to investigate the sealing capacity of low porosity beds acting as baffles to vertical flow was also stressed.

Permeability can be estimated from the Stoneley wave (attenuation). It was argued that this works best if the porosity is greater than 10% and the permeability is greater than 1md. Even if these conditions are met, the permeability estimate should be regarded as qualitative rather than quantitative because of the difficulty of estimating other inputs such as mud cake properties. The Stoneley permeability appears to respond to both fracture and matrix permeability – but cannot distinguish between them. Like formation pressure testing, but unlike most other tools the permeability prediction arguably relies on creating actual flow through the rock and so is a true measurement of the permeability (though the flow is oscillatory and does not mimic reservoir flow conditions as closely as formation pressure testing data). A major limitation of this technique is the limited vertical resolution of the present acoustic logging tools: typically 1m or 3 ft. “Super-k” layers are often much thinner and may therefore not be fully resolved, and hence possibly overlooked.. Several participants called on the logging contractors to consider designing acoustic logging tools with improved vertical resolution.

Fractures may increase the effective permeability if they are open – or impair it if they are closed and cemented.

Cemented fractures appeared to be much more common than open fractures. It was argued that the majority of reservoirs were under compression in both the X and Y directions and that consequently any open fractures would have a strong natural tendency to close. The presence of open fractures was often linked to recent tectonic activity.

Fractures appeared to be most important as permeability boosters in tight matrix with lesser impact in high permeability formations.
Ways to identify the presence of fractures included:
Micro mud losses
Bore Hole Image Logs
Sonic Logs (Chevrons)
Micro Resistivity
Caliper anomalies
Timelapse LWD: i.e LWD plus wipelog.
PLT (anomalous influx)

The need to calibrate fracture data from image logs against core data was stressed, but this needs to be weighed against the difficulty and cost of obtaining core in horizontal wells. Core recovery across fractured zones may also be limited to “rubble.”

Carbonate reservoir properties vary at nearly every scale due to complex deposition & diagenesis and representation of heterogeneity is closely related to scale. A number of techniques have been developed recently to create increasingly detailed static models that better represent complex spatial variations in rock type, porosity, permeability and fluid saturations. As a result static models have become very large and may contain up to 50 million cells; too large to simulate efficiently. More simple – coarser – dynamic models are needed to simulate fluid displacement in the reservoir, speed up history matching and assess multiple geologic realizations and development options efficiently.

Upscaling is defined as the process of constructing such relatively coarsely gridded simulation models from fine-scale geologic models, while maintaining enough precision to evaluate uncertainty in reservoir and operating conditions. Critical effects that need to be preserved in scale-up include:

  • Compartmentalization due to structure and stratigraphy.
  • Effects of vertical and horizontal heterogeneity
  • Flow dynamics and changes in saturation due to production/injection

Because of complex internal structures and types of connectivity, simple analytical methods are no longer adequate for permeability upscaling multi-million cell static models. Flow based and pressure field solution based permeability scale up methods have been developed and are preferred to generate the effective permeability tensor that preserves the effects of heterogeneity and the displacement behavior of the finely gridded complex static model. Flow Based Scale Averaging uses single-phase numerical simulation to back out directional effective permeabilities or inter-block transmissibilities of coarse grid model cells. The concept is similar to measuring permeability in the laboratory across a plug or a full-diameter core.

Pressure Field Solution (or Pressure Solver) methods are mixtures of “explicit solver” and “finite difference” methods.

Directional permeability averaging is sometimes used.

The z direction effective permeability for the coarse-grid block is approximately equal to the arithmetic average of the harmonic means of the fine-grid z direction permeabilities.

The effective permeability in the x direction for the coarse-grid block is approximately equal to the harmonic average of the arithmetic means of the fine-grid x direction permeabilities. Quality Control is an important part of model building.

It is necessary to quality control dynamic model properties and compare to the static model: utilize visual inspection of areal maps and cross sections. Perform statistical comparisons and make sure the upscaling process has preserved the effects of heterogeneity for reliable modeling of dynamic changes.

It is particularly important is preserve high permeability “Super-k” layers which may act as conduits for water or gas breakthrough and tight streaks that may act as barriers or baffles to vertical flow.

Leave a Reply