Geosteering a thin carbonate with a typewell and deep resistivity

In a reservoir 8 m thick, a 0.6° error in dip moves the bit a metre every 100 m. A typewell tells the team what rock the bit is in; deep resistivity tells it how far away the boundaries are. This article works through a 1,200 m lateral in a thin carbonate, set in onshore Abu Dhabi conditions, with DrillIQ's geosteering engine: the typewell, two steering decisions, and what each kind of evidence could and could not say.

What geosteering has to work with

A geosteering team keeps the lateral in the best rock using three things. The typewell is the expected log response through the target interval, built from offset wells and plotted against stratigraphic depth. The LWD logs (gamma ray, resistivity, density) show what the bit has drilled, some metres late. And the boundary distances from a deep azimuthal resistivity tool show how far the bed boundaries are above and below the tool.

The team's working answer is the interpretation: the reservoir top along the well's path, defined by tie-points, with a dip between them and faults where the top steps. Every steering decision follows from where the interpretation puts the bit.

Building the typewell

Each offset's logs are converted from depth to true vertical thickness (TVT) below a marker, the reservoir top here, and stretched interval by interval so that the markers line up. The offsets are then averaged with weights that fall with distance from the new well's path. DrillIQ uses inverse distance squared by default, per interval:

d_i

distance from offset i, where it crosses the interval, to the new well's path in that interval

z_top

the interpreted reservoir top (TVD) at the bit's position along the section

t₀, t

the typewell's zone thickness, and the zone thickness at that position

Three vertical offsets cross the reservoir 703 m, 1,663 m and 2,515 m from the lateral. Their weights come out at 0.795, 0.142 and 0.062, and the composite zone is 7.81 m thick. The offsets also give the prognosis: a top dipping 0.3° towards the well's 060° heading. What the offsets cannot show is the structure between them.

What deep resistivity adds

A deep azimuthal resistivity tool measures at several spacings and frequencies, with antennas that see one side of the hole differently from the other. The LWD service inverts those readings into a layered model around the tool: the resistivity above and below, and the distance to each boundary. The distance is measured square to the tool axis, so it is turned into a vertical offset on the section:

In a near-horizontal well sin I is almost 1, so the tool's reading is the vertical distance. It needs resistivity contrast to work, and carbonates often give plenty: here a 110 Ω·m stylolitic seal and a 600 Ω·m dense base around a reservoir of 18–45 Ω·m. The tool in this example sees a boundary up to 5 m away, which in an 8 m zone covers most of it from the middle.

The example lateral

The example is a 6⅛″ lateral in a thin porous limestone at about 2,620 m TVD, typical of the Lower Cretaceous carbonates of onshore Abu Dhabi: layered, with dense stylolitic beds between the porous ones and very saline formation water. The well kicks off at 2,279.6 m MD, builds at 5°/30 m, and lands at 2,819.6 m MD, 3.4 m below the top. The target line sits at 45 % of the zone, 3.5 m below the top, in the best layer.

The structure along the lateral is synthetic, so it is known and every engine answer can be checked against it. The top dips 0.3° for the first 350 m, as prognosed, then 0.9° down a flank for 450 m, then steps down 3.0 m at a fault at 800 m, then dips 0.4° to TD.

Hand checks

1. Typewell weights. 1/703² : 1/1,663² : 1/2,515² = 2.02 : 0.36 : 0.16 (×10⁻⁶), which normalises to 0.795 : 0.142 : 0.062.
2. Why the best layer matters. Porosity from density on a 2.71 g/cm³ limestone matrix: (2.71 − 2.32) / (2.71 − 1.0) = 0.228. Archie, with R_w = 0.02 Ω·m for the saline water and m = n = 2: S_w = √(0.02 / (0.228² × 35)) = 0.105. The upper layer: φ = 0.170 and S_w = √(0.02 / (0.170² × 18)) = 0.196. Oil per unit of rock, φ(1 − S_w), is 0.204 against 0.137: a third less in the upper layer.
3. The drift. Holding the prognosed 0.3° on a 0.9° flank, the well climbs through the zone at tan 0.6° = 1.05 m per 100 m. From 3.6 m below the top at 350 m, it reaches the upper layer 1.6 / 0.0105 = 152 m later, at about 502 m.
4. The advisor's correction. It aims at the target line 60 m ahead: inclination = 90° − atan(Δz / 60). At decision 1 below, Δz = 2,627.66 − 2,625.86 = 1.80 m, so 88.28°. From the bit's 90.09° that is −1.81°, a dogleg of 0.91°/30 m, and 1.81 / (6/30) = 9.1 m of slide.

The lateral, decision by decision

For the first 350 m the prognosis holds and the well sits 3.4–3.7 m below the top. Then the formation starts dipping faster than the plan, and nothing in the gamma ray says so.

Decision 1, at 470 m (3,290 m MD). Against the prognosis the bit is 3.4 m below the top, close to the target line, and the gamma at the sensor reads a flat 13–14 API. The boundary tool disagrees: it puts the top 2.55 m above the sensor. Snapping a new tie-point to the tool's readings moves the top 0.97 m deeper than the prognosis, within 0.13 m of the true top, and gives a dip of 0.63° from the last tie. With that interpretation the advisor puts the bit 2.37 m below the top (truly 2.21 m) and asks for a DROP to 88.3° over 60 m, as in the hand check.

The second chart shows why. Against the prognosis, the top stays 3.4–3.8 m above the sensor the whole way to 520 m. The tool's distance starts falling soon after 350 m and is down to 2.3 m by 480 m. The team in this example waited for the gamma ray: it rose to 23 API when the bit was at 530 m, and the well spent 45 m, from 502 m to 547 m, in the upper layer before the drop brought it back. At 530 m the advisor wanted 87.7° and 11.3 m of slide, against 88.3° and 9.1 m at 470 m: waiting costs a bigger correction as well as footage.

Decision 2, at 850 m (3,670 m MD). The tool's distance to the top falls from 3.35 m to 0.37 m between two readings a metre apart. No dip of less than a degree can do that: it is a step. The interpretation without a fault still says HOLD, 3.4 m below the top, and the advisor flags it: the tool sees the top 0.5 m above, the model says 3.4 m. The gamma at the sensor has jumped from 12 to 24 API and the resistivity has halved, 36 to 18 Ω·m: the bit is in the upper layer. With a fault at 800 m and the tie after it snapped to the tool, the throw comes out at 2.83 m (true 3.0 m), the bit is 0.46 m below the top, and the advisor asks for 86.3° over 60 m, 1.6°/30 m and 15.9 m of slide. The well came back down over the next 125 m with no dogleg over 1.8°/30 m.

Reading the typewell correlation

Each lateral reading can be placed at the stratigraphic depth the interpretation gives it and laid over the typewell. With the final interpretation the gamma ray falls on the typewell: 24 API readings in the upper 2 m, 10–15 API in the best layer. With the prognosis the same readings spread from 3.5 m down past the base, and 24 API readings sit where the typewell says 13. That spread is what the misfit alerts measure.

The same comparison has a weakness. In the best layer all four logs are flat, so the typewell has nothing to lock on to. DrillIQ's dip detector, run on the gamma sensor's position at 470 m, moved the top 1.70 m the wrong way (the true change was +1.09 m). At 530 m, with the upper layer's 24 API at the sensor, it moved it +1.70 m against a true +1.63 m, and the correlation coefficient rose from 0.22 to 0.87. Correlation needs character in the logs; the boundary tool doesn't.

Over a long window, correlation does find the structure. Run over the whole lateral against the prognosis, the auto-steer second opinion, a misfit search with a penalty on sudden changes, puts the top 4.0 m below the prognosis along the flank and 7.0 m below it just past the fault: a ramp, then a step.

The result

The well stayed in the reservoir for all 1,242 m after it entered it. Of the 1,200 m lateral, 1,029 m (86 %) was in the best layer and 172 m in the upper layer, after the two events. The average distance from the target line was 0.61 m. Had the well simply followed the plan, it would have left the top of the zone at 667 m and spent 55.5 % of the lateral in the reservoir. The final interpretation, built from tie-points snapped to the tool, is within 0.19 m of the true top everywhere more than 15 m from the fault.

Practical guidance

Use each kind of evidence where it works. Correlation needs marker beds; in a featureless layer it follows noise. Boundary distances need resistivity contrast and range; they work in clean rock and see a boundary before the bit reaches it.

Tell a dip from a fault by the shape. A dip change is a ramp: the distance to the top changes by about a metre per 100 m for each half-degree. A fault is a step: metres in one or two stands. Fitting a dip to a fault puts the error ahead of the bit.

Act on the first evidence. At 470 m the correction was 1.8° of inclination; 60 m later it was 2.3°, with 45 m in poorer rock in between.

Keep the interpretation honest. Snap tie-points to the boundary tool where it has readings, re-run the typewell correlation when a marker bed reaches the sensor, and record why each change was made.

A checklist for the rig team

• Build the typewell per interval and check the offset weights against their distances.
• Know each sensor's distance behind the bit and plot every log at its own depth.
• Compare the boundary tool's distances with the model at every survey.
• Treat a disagreement over 1.5 m as a reason to re-interpret at once.
• Don't trust a dip pick from flat logs; wait for a marker or use the boundary tool.
• When the distance to a boundary steps, add a fault rather than a steeper dip.
• Keep corrections inside the dogleg limit and re-run the advisor after each stand.

How DrillIQ handles it

DrillIQ's Geosteering module builds the typewell from any offsets, per interval, with inverse-distance, Gaussian, nearest or manual weights. The real-time section shows the logs against the typewell, tie-points, faults and thickness changes. Its dip detector suggests a dip and says when the logs can't support one, and the auto-steer second opinion searches the whole section for the best fit. The boundary-mapping capability takes the distance-to-boundary curves from the LWD service's inversion, plots them on the section, compares them with the model and offers to snap the tie-points; gamma, resistivity, density, image, cuttings chemistry and at-bit data each add their own evidence.

The steering advisor turns the interpretation into an inclination, dogleg, slide length and toolface within the dogleg limit, with every alert naming its source. Recommendations go from draft to approval to the directional driller, and the post-job report gives footage by zone and every change with the advice at the time. Everything above was worked out with the same engine.

Book a demo → See all capabilities

References

1. Archie, G.E. The Electrical Resistivity Log as an Aid in Determining Some Reservoir Characteristics. Transactions of the AIME 146, 1942.
2. Bittar, M. et al. A New Azimuthal Deep-Reading Resistivity Tool for Geosteering and Advanced Formation Evaluation. SPE 109971, 2007.
3. Li, Q., Omeragic, D. et al. New Directional Electromagnetic Tool for Proactive Geosteering and Accurate Formation Evaluation While Drilling. SPWLA 46th Annual Logging Symposium, 2005.
4. Viterbi, A.J. Error Bounds for Convolutional Codes and an Asymptotically Optimum Decoding Algorithm. IEEE Transactions on Information Theory 13(2), 1967.

About this example. The well in this article is a representative scenario set in typical onshore Abu Dhabi, UAE conditions and worked through with DrillIQ's engines. It is not a client's well, and its figures are not field results.