Ershaghi X-Plot — waterflood performance & ultimate recovery

For oil wells deep into water cut, conventional decline curve analysis no longer applies. Production rate falls quickly while cumulative oil rises slowly. Ershaghi's X-plot extracts ultimate recovery from this regime using a simple linearization of fractional flow theory — free, fast, and surprisingly defensible.

The mature waterflood problem

Picture an oil well 12 years into production. Water cut is at 88%. Daily oil rate is 60 bopd, dropping 0.3% per month. Forecasting on rate-time decline gives noisy results because most of the production is now water — small swings in water rate dominate the apparent decline.

What the engineer actually wants to know is simpler: how much more oil will this well produce before water cut hits 95%, 98%, or economic shutdown? The answer doesn't depend on time — it depends on cumulative oil and the relationship between water cut and sweep efficiency.

Ershaghi and Omoregie (1978) showed that for systems with constant fractional flow behavior, plotting log((1-WC)/WC) against cumulative oil Np yields a straight line. Extrapolating that line to economic water cut gives ultimate recoverable oil directly.

The X-plot equation

Starting from fractional flow theory in a one-dimensional displacement, water cut at a producer can be written as a function of the saturation profile and mobility ratio. For a wide range of practical waterflood conditions, this collapses to:

The slope a is always negative for water-displacing-oil systems. As N_p grows, WC approaches 1, the ratio (1-WC)/WC approaches 0, and its log goes to −∞.

Solving for ultimate oil at a target abandonment water cut:

Why log-linear works

The physical basis traces to the Buckley-Leverett fractional flow model. For waterflood in a system with reasonable layering and constant well bottomhole conditions, water cut at the producer evolves approximately as:

When relative permeability ratio scales approximately exponentially with average reservoir saturation (true for many sandstone and carbonate systems), the cumulative oil produced becomes a linear function of average saturation. Combining these relationships and taking logs yields the X-plot linearization.

The framework is empirical at the macro level — the line works whether or not the underlying physics is exactly Buckley-Leverett. Decades of field application have shown it holds for most conventional waterfloods past 30-40% water cut.

Step-by-step worked example

A 14-year-old oil well in a sandstone waterflood with the following monthly snapshots:

Step 1 — Plot log((1-WC)/WC) versus N_p.

Plotting the last two columns yields a clearly linear trend across most of the data. Early points (months 24-48) may show curvature as the displacement pattern was still developing.

Step 2 — Fit the linear portion.

Using least-squares regression on months 72-168 (where the line is stable):

Step 3 — Extrapolate to economic water cut.

For abandonment at WC = 0.95:

Interpretation: Current N_p = 3,210 Mbbl. Extrapolated ultimate at 95% water cut = 3,660 Mbbl. Remaining recoverable oil ≈ 450 Mbbl. At current rate of ~60 bopd, that translates to roughly 20 years to economic limit — though terminal decline will accelerate near abandonment.

Window selection — the practical decision

The X-plot's biggest pitfall is fitting the wrong window. Three regions need to be distinguished:

Early time (pre-breakthrough)

Before water reaches the well, WC is near zero and (1-WC)/WC is enormous. Log values are large and unstable. Always discard these points.

Transition period (first months after breakthrough)

Water cut rises rapidly but the line hasn't stabilized. Slope changes substantially as the displacement pattern matures. Discard.

Stable depletion regime

The line is straight, slope is consistent, R² > 0.95. This is the fit window. For most conventional fields, the stable regime begins at WC ≈ 30-40% and continues through WC ≈ 90%.

Late-life flattening

Near WC = 1, the line can curve upward (less negative slope) as residual oil saturation dominates. Including these points pulls the extrapolation too optimistic. Fit only the stable middle.

When auto-fit fails — engineering judgment override

Real wells often have multiple inflection points. A workover, infill drilling nearby, or change in injection support can shift the slope mid-life. An auto-fit blindly applied to all data gives an averaged slope that fits neither regime well.

The practical workflow:

1. Plot the data and inspect visually
2. Identify slope changes (typically corresponding to operational events)
3. Fit only the most recent stable segment — the regime most relevant to current behavior
4. If R² > 0.95 on the chosen window, use the extrapolation
5. If R² < 0.85, the well is not in a fittable regime — note "X-plot not currently applicable" rather than force a bad fit

Modern surveillance dashboards provide engineering-override panels that let users pick the N_p window manually when auto-fit chooses badly. This is essential — fully automatic Ershaghi tends to over-fit early-time data and under-extrapolate.

Reading the slope

The slope a carries physical meaning:

Comparing slopes across wells in the same field reveals heterogeneity. Steep slopes flag wells with poor sweep — candidates for workover or pattern realignment.

Common pitfalls

1. Using WC in percent instead of fraction. The equation requires WC as a fraction (0 to 1). Plugging in 85 instead of 0.85 ruins everything.

2. Forcing the line through early-time points. Pre-breakthrough data has WC near zero, log values explode upward. Fitting them pulls the slope wrong.

3. Ignoring workover events. Any intervention that changes well productivity (recompletion, acid, stimulation) breaks the X-plot continuity. Refit from the event date forward.

4. Trusting R² without inspection. A fit can have R² = 0.92 and still be wrong if the window includes a slope change — the fit averages two slopes into one meaningless slope. Always look at the residual pattern.

5. Assuming the line continues forever. Late-life (WC > 95%) deviates from linearity. Extrapolating to WC = 99% gives an inflated EUR. Use WC = 95% as practical upper bound, then accept that residual recovery beyond that requires a different model.

6. Treating Ershaghi as standalone evidence. Cross-check against conventional DCA, material balance, or volumetric estimates. Wide divergence between methods is a signal that one of them is wrong — usually the one closest to abandonment.

Applicability — does Ershaghi work for this well?

Before fitting an X-plot, screen the well against its drive mechanism. The X-plot is a waterflood tool: it derives from Buckley-Leverett fractional flow under immiscible displacement at near-constant voidage. When that physical setup is violated, the log-linear trend either doesn't form or forms but doesn't mean what Ershaghi intended.

Drive-mechanism screening matrix

Pre-fit checklist

Before running the X-plot, verify:

• Drive mechanism is waterflood or active aquifer — not gas drive or depletion
• Water cut > 30% — below this, fractional flow line hasn't stabilized
• Pressure support is steady — no major injection rate changes, no aquifer collapse
• No recent workover that changed completion — perforation adds, ESP changes, fracs all reset the curve
• ≥12 months of post-breakthrough data — enough points to fit a line confidently
• No active EOR or thermal recovery — these shift the fractional flow curve mid-fit

If any item fails, either skip Ershaghi entirely (use a method matched to the drive mechanism) or apply it with explicit caveats — never report the EUR as final without flagging the violated assumption.

Auto-fit screening logic

In automated tooling (GOWIS handles this) the safest screening logic is to skip Ershaghi when:

• Current water cut < 30%, OR
• Drive mechanism field = solution_gas, gas_cap, depletion, thermal, or eor, OR
• < 12 data points post-breakthrough, OR
• Workover flag is set within the analysis window, OR
• R² of the best 24-month rolling window < 0.85

When any of these trigger, fall back to Arps DCA (for oil rate) or flag the well for manual review. The principle: a confident-looking line that violates the underlying physics is more dangerous than no fit at all, because it carries authority it doesn't deserve.

Field practice: rolling reassessment

The X-plot is not a one-time calculation. Best practice is to recompute every quarter on all mature wells. As more data accumulates, the line refines. Slope drift between quarters flags wells where behavior is changing — typically due to operational events or reservoir reorganization.

For a field with hundreds of wells, this is automation territory. A surveillance dashboard can fit X-plot on every oil well nightly, flag wells with low R², and present the extrapolated N_p,ult alongside engineering-override controls.

Three takeaways

1. Window selection beats fit quality. A defensible window with R² = 0.92 beats an over-fit on all data with R² = 0.99.
2. Slope is sweep efficiency. Read the slope as a measure of how well the waterflood is working — not just as a fitting parameter.
3. Engineering override is essential. Pure auto-fit fails on wells with history. Always allow the engineer to pick the N_p window manually.

References & further reading:
Ershaghi, I., Omoregie, O. (1978). A Method for Extrapolation of Cut vs. Recovery Curves. JPT, 30(2), 203–204.
Ershaghi, I., Abdassah, D. (1984). A Prediction Technique for Immiscible Processes Using Field Performance Data. JPT, 36(4), 664–670.
Yang, Z. (2009). Analysis of Production Decline in Waterflood Reservoirs. SPE 124613.
Craig, F. F. (1971). The Reservoir Engineering Aspects of Waterflooding. SPE Monograph Vol. 3.
Willhite, G. P. (1986). Waterflooding. SPE Textbook Series Vol. 3.
Dykstra, H., Parsons, R. L. (1950). The Prediction of Oil Recovery by Waterflood. Secondary Recovery of Oil in the United States, API.