Well correlation

The section of each well is a one-dimensional record: a sequence of rocks and log curves against depth. Correlation ties these records into a three-dimensional picture. On a correlation panel we place the well sections side by side and join equal boundaries with lines. In Russian-language practice, people distinguish general correlation from detailed correlation. General correlation matches formations, horizons and large members through the whole section. Detailed correlation works layer by layer and bed by bed in the productive interval. The result of detailed correlation is an agreed table of picks and a panel, from which we build maps and models.

The lines on a panel are not observations. They are assertions. When you join the top of a sandstone in two wells, you assert that the same sandstone lies between them: it does not pinch out (thin out to nothing), it is not cut by a fault, and it is not eroded. When you join two ash beds, you assert that they fell at the same time. These two kinds of assertion are fundamentally different, and a later section deals with them. A new well, seismic data or a response to injection can refute any of them, so we revise the correlation throughout the life of a field.

Before correlation we bring the data to a common form. Picks are taken in measured depth, but we compare sections in true vertical depths below sea level (TVDSS; depth references). For deviated wells we calculate these from the directional survey (the measured path of the well), and we compare thicknesses as true vertical thicknesses or true stratigraphic thicknesses (see the section on deviated wells). We plot all logs at one vertical scale and with the same curve scales.

The absolute values of logs in different wells cannot be compared directly. Tools of different years and contractors, hole diameters and drilling fluids (muds) shift the scale (when GR misleads). So we correlate on the pattern of the curves. For quantitative comparisons we normalise the curves. In each well we take the same reference beds: a persistent shale member and a clean sandstone or limestone. Then we convert the scale linearly so that the readings in these beds coincide with those in the reference well.

• The top and the bottom of a log run are the limits of the measurement, not of the rocks. We do not join them with neighbouring wells.
• A casing shoe, the passage of the record from open hole (the part of the well with no casing) into casing, and the splice of two runs give steps on the logs that look like geological boundaries.
• Jumps in the sonic log and washouts seen on the caliper log (a log of the hole diameter) distort the curves in individual intervals (Borehole environment: invaded zone and mudcake).
• The depths of different runs and of core are offset from one another by metres. Before correlating, we match them on the GR.

A marker bed is a thin, persistent layer or surface that we can recognise in every well and that was deposited at practically the same time over the whole area. Marker beds are the framework of correlation.

• Good markers: volcanic ash beds (a narrow peak on the GR), shales of maximum flooding surfaces (a GR maximum, often because of uranium), limestones and anhydrites that stay the same over long distances (a GR minimum, high resistivity and density), and coals of a delta plain that accumulated at the same time over a large area.
• Poor markers: channel and delta sandstones, reef buildups, coals of river floodplains and fan deposits. These are bodies of limited extent, and their position in the section changes from well to well. Shale members that look alike are poor markers too, because they are easily confused.

Lithostratigraphic correlation joins like rocks: the top of a sandstone with the top of a sandstone. Its lines are the boundaries of rock bodies, and they are diachronous: when the shoreline shifts, the same sand member is younger in one place and older in another (lithostratigraphy). Chronostratigraphic correlation joins surfaces of equal age: marker beds, biostratigraphic events, flooding surfaces (surfaces made when the sea suddenly deepened) and sequence boundaries (sequence stratigraphy). Time lines never cross. Lithological boundaries may cross them at any angle.

The difference is not just academic. In a prograding succession the shoreline moves steadily seawards. Each new sand layer is separated from the previous one by a thin shale deposited at the next flooding. “Sand-to-sand” correlation joins them into one connected layer. Correlation along flooding surfaces shows several layers that overlap only partly. Flooding shales persist over large areas and often act as barriers to crossflow. Water injected into one layer may never reach a producing well where the “sand” lies at the same level but belongs to another layer.

Biostratigraphic events in wells are the levels at which species appear and disappear (biostratigraphy). When we drill downwards, the first cuttings sample in which a species is found gives the top of its range, and this is reliable. Fragments that have caved from the borehole wall higher up the hole push the lower limit of the range downwards. So we take the lower limit from core and sidewall core (core cut from the borehole wall) (fossils in cuttings and core). Species that lived only in a particular environment appear and disappear together with the rock, not with time, so they are no use for correlation in time.

We can arrange the same wells on a panel in two ways. On a structural cross-section each well stands at its own depths below sea level. The section shows how the layers lie today: crests, troughs and faults. On a flattened cross-section (also called a stratigraphic cross-section; the operation is called flattening, or hanging) we shift all the wells vertically so that the chosen marker lies horizontal. This marker is the stratigraphic datum. If the marker was deposited horizontally, such a section removes all later deformation and shows the layers as they were when the marker was deposited.

Horizontally, we place the wells either at true distances or at equal intervals. With true distances we get a cross-section, in which shapes and slopes are visible. With equal intervals we get a correlation panel, on which it is easier to compare the sections themselves. A network of panels that intersect at wells and are drawn in perspective is a fence diagram.

• A good datum is close to the interval of interest and above it. The farther the marker, the more later events enter the panel.
• A good datum is present in all wells. It is not eroded and not cut out by a fault between them.
• A good datum was deposited horizontally. Examples are flooding shales, ash beds and persistent coals and limestones. The top of a sand body, a reef or inclined progradational layers, and an unconformity surface with relief, are not good datums.

The log of a deviated well is recorded along the hole. For correlation we first convert it to true vertical depth (TVD). Otherwise all the intervals come out stretched and the markers come out deeper than they are. But even in true vertical depths a deviated well sees thicknesses differently from a vertical one. As it crosses a layer it moves sideways, and the top of the layer is already at a different depth there.

We describe the thickness of a layer by two quantities (true thickness and vertical thickness). The true stratigraphic thickness $h$ (TST) is measured normal (at right angles) to the bedding. The true vertical thickness $h_\text{TVT}$ (TVT) is measured vertically at the given point. In geological texts they are often called simply the true thickness and the vertical thickness. In a vertical well the difference in depth between the top and the base equals the TVT. In a deviated well, the difference in true vertical depth $\Delta H$ (simply called the TVD thickness) equals the TVT only in the special case when the hole runs along the strike of the layer.

A horizontal hole runs almost along the layer and may cross the same boundaries repeatedly. In true vertical depths this looks like a repetition of the section, even though there is no fault (a fault in a well). We describe the position of such a hole by its distance from a chosen boundary, measured along the normal. This distance is the stratigraphic depth. We compare the measured log with one calculated from a model of the layer. This is how geosteering (steering the well by the rock it meets) is done (Horizontal wells).

Graphic correlation is a quantitative way to match two sections by events that date the layers. We plot the depths of the same events in the two wells: the reference well on the horizontal axis, and the well being matched on the vertical axis. The events are tops and bases of species ranges, ash beds, GR peaks of flooding surfaces, and reversals of magnetic polarity. If both wells accumulated sediment without breaks, the points fall near a straight line, the line of correlation.

• The slope of the line is the ratio of accumulation rates. A slope of 1.3 means that in the same time 1.3 times as much rock was deposited in the well being matched (or the rock there was compacted less).
• A horizontal step means that the depth in the well being matched does not change while the depth in the reference well does. Part of the section of the reference well is absent in the matched well: a hiatus with erosion, or a sharply reduced interval. A vertical step means the same for the reference well.
• A kink (a sharp bend in the line) is a change in the ratio of accumulation rates.
• Points off the line are events whose position is distorted. Examples are range bases from cuttings that caving has shifted downwards, and species that depended on the environment or arrived in the area later.

An unconformity. This is a surface where part of the rock record is missing. Beneath an erosion surface the correlation lines run up to it and end: in neighbouring wells different layers lie below the unconformity. We must not join layers across an unconformity as if they were continuous. We recognise an unconformity by the truncation of markers (they are cut off at the surface), by a sharp change of dip on the dipmeter, by the absence of biostratigraphic zones, and by weathering crusts and conglomerates at the boundary (unconformities).

Facies change and pinch-out. In a facies change the layer is still there, but the rock in it is different: a sandstone passes sideways into shale. We correlate the surfaces that bound the layer, not the sandstone, and the panel shows where the rock changes. In a pinch-out the layer thins to zero between the wells, and on the panel the lines of its top and base are brought together at a point. The wells cannot say where exactly this happens between them. Seismic data sometimes give a hint, but thin layers and the pinch-out point itself are usually below its resolution. The vertical sequences of rocks, from which we recognise the environment and the shape of bodies, and the characteristic shapes of GR curves are covered in the topic Facies, facies successions and sedimentary logs.

The last check is for realism. Correlation lines between neighbouring wells are roughly parallel or diverge gradually. They do not cross. Every change in thickness and attitude is explained by an understandable sequence of events: deposition, growth of the structure, erosion and faulting. A panel on which lines converge and diverge without any system is almost certainly wrong.

How we correlate sand bodies depends on how large they are compared with the distance between wells. If the bodies extend for kilometres, we can trace the layers from well to well, and we build the section as a “layer-cake”. Examples are shoreface and shelf layers (sands deposited near the shore), turbidite lobes (deposits of underwater sediment flows) and platform carbonates. If the bodies are comparable to the well spacing (channels and bars of river and delta systems), they abut one another with almost no gaps. But some contacts let fluid through, and others are separated by thin partings of low permeability. This is a “jigsaw”. If the bodies are much smaller than the well spacing, we cannot match them one by one at all (deterministically). Such sections are described statistically (Reservoir architecture and performance).

• The depositional environment suggests the size and shape of the bodies. We learn it from core, the vertical sequences of rocks and the shapes of the curves (Facies, facies successions and sedimentary logs).
• The net-to-gross ratio (the share of reservoir rock in the section): at a high ratio even channel sandstones merge with one another and form a connected body.
• Pressures and contacts: a common pressure gradient and a common oil–water contact show that the bodies communicate. Different gradients and contacts show separate bodies (Pressure measurement with the wireline formation tester).
• The behaviour of wells: the response of producers to injection, and interference tests (a pressure change in one well is watched in another).
• Seismic data: the shapes of bodies and their boundaries, if the bodies are thicker than its resolution.

If data are lacking, we build several versions of the correlation into the geological model and examine how the hydrocarbons in place and the production forecast depend on the choice (Geological and technical uncertainty of a project). New wells and production history then weed out the unsuccessful versions.