The Complete Guide to Oil Material Balance Analysis

A gas reservoir gives up its secret on one plot. An oil reservoir does not. Gas has one mobile hydrocarbon phase and a compressibility that dominates everything else; oil has two phases that exchange mass across the bubble point, a gas cap that may or may not exist, and an aquifer that may or may not be connected. The material balance equation handles all of it — but only if you tell it which terms are live, and it will return a confident, precise, beautifully linear wrong answer if you get that wrong. This guide works through oil material balance from the general equation to the drive indices, with three worked examples in which the true answer is known by construction, so every claimed error can be checked rather than asserted.

Part I — FoundationsWhat the equation actually says

Material balance is bookkeeping with physics attached. Whatever volume of fluid has left the reservoir must be accounted for by something expanding to fill the space it vacated: the oil expanding as pressure falls, the solution gas coming out of it, a gas cap swelling, the rock and connate water squeezing in, or water crossing the boundary from an aquifer. Nothing else is available. Written in reservoir barrels, that statement is the whole method:

Every symbol in that line is a volume in reservoir barrels except N — stock-tank barrels of oil originally in place, written OOIP, or STOIIP where the stock-tank convention is spelled out — and m, which is dimensionless. It is deliberately not OGIP: that is the gas quantity, and the two are not interchangeable even in a reservoir that produces both. The four terms on the right are the only four ways the reservoir can give the fluid back.

R_p is the cumulative producing gas–oil ratio, G_p/N_p — not the instantaneous GOR from this month's allocation. Confusing the two is one of the most common ways a good pressure history produces a bad N, and it is silent: both are numbers in scf/STB with the same order of magnitude.

Why oil is harder than gas

For a volumetric gas reservoir the equation collapses to P/Z versus G_p, a straight line whose x-intercept is the gas in place. There is one unknown and one plot. Oil has at least three unknowns that the same pressure history has to carry: N, m, and W_e. That is the entire difficulty, and it produces a specific failure mode — the equation will happily trade one against another. Underestimate the gas cap and the fit compensates by inflating N. Miss an aquifer and the fit inflates N again. In both cases the straight line still looks like a straight line.

That is why this guide spends as much space on the diagnostics as on the fits. Every worked example below is followed by the question that matters: if the assumption behind the fit were wrong, would the plot have told me?

Part II — PVTThe two-phase formation volume factor

Above the bubble point the oil holds all its gas in solution, and as pressure falls it simply expands: B_o increases, R_s is constant. Below the bubble point gas comes out of solution. The oil left behind shrinks — B_o now falls with pressure — while the liberated gas occupies a rapidly growing volume. Using B_o alone below the bubble point therefore tells you the oil is contracting, which is true and completely misleading, because the liberated gas is still in the reservoir.

B_t^[13] is the volume that one stock-tank barrel of original oil now occupies together with all the gas that came out of it. It increases monotonically as pressure falls, on both sides of the bubble point, which is what makes E_o a well-behaved expansion term. Substituting B_o for B_t in the material balance is not a simplification; it is a different and wrong equation.

Part III — The Straight LineHavlena–Odeh

Havlena and Odeh's contribution was not a new equation but a new way of reading the old one.^[2][3] Group the expansions into a single term and the material balance is the equation of a straight line:

Plot F against E_t. If the reservoir is volumetric and m is right, the points fall on a line through the origin whose slope is N. The through-origin constraint is not a convenience; it is physics. Before any production, F and E_t are both exactly zero. A fit that returns an intercept is a fit that has been allowed to disobey a boundary condition it should be held to.

Two fits, two purposes

So report both. The through-origin slope is the answer — that is the model being claimed. The free-intercept fit is a diagnostic, and only that: if the points genuinely head for the origin, the free fit will find an intercept near zero of its own accord. What matters is not the intercept's value but whether forcing the physics changed the slope.

Worked Example A — volumetric solution-gas drive

A saturated reservoir at its bubble point, 2,500 psia, no gas cap (m = 0), no aquifer. True N = 50.000 MMSTB by construction. Seven pressure surveys with the cumulative production the material balance requires:

Notice what the production history did while this was happening. Cumulative GOR climbed from 766 to 1,067 scf/STB against an R_si of 581 — the classic solution-gas signature, gas leaving the oil faster than oil leaves the reservoir. Recovery at 1,100 psia is 11.7 MMSTB, or 23% of N, which is at the optimistic end of the 5–30% that solution-gas drive typically delivers.^[4]

Part IV — The Gas CapWhat m does to the answer

m is the ratio of gas-cap reservoir volume to oil-zone reservoir volume at initial conditions. It is not measured; it is estimated from structure, contacts and net pay, and it carries the uncertainty of all three. The material balance is more sensitive to it than most practitioners expect.

Worked Example B — gas cap drive

The same reservoir, same PVT, same pressure history, with a gas cap of m = 0.40 and the production the balance then permits. True N is again 50.000 MMSTB.

Analysed with m = 0.40, the plot recovers N = 50.000 MMSTB, R² = 1.000000. The gas cap has roughly doubled recovery at the same abandonment pressure — 22.8 MMSTB against 11.7 — which is the whole reason gas caps matter.

The same data, analysed with m = 0

Now suppose the gas cap was missed. Perhaps the contact was picked from a single well, perhaps the structure map was optimistic, perhaps nobody asked. The identical production and pressure history, run through the identical equation with m = 0, gives:

This is the single most important number in the article. The reason it is possible is structural: dropping m·E_g makes E_t too small at every point by a factor that varies only slowly with pressure, and a slope fitted to systematically shrunken abscissae comes out systematically too large — while remaining almost perfectly linear. Goodness of fit cannot detect a missing term that is nearly proportional to the terms you kept.

There is a signal, but it is not R². Under m = 0 the F/E_t ratio drifts downward — 108.9 to 97.5 MMSTB across the history, a fall of 10.5%. Under the correct m it is flat at exactly 50.0. A drifting F/E_t with a beautiful R² is the fingerprint of a term that is present in the reservoir and absent from your equation.

Part V — Water DriveThe failure that costs the most

A gas cap that is missed doubles N. An aquifer that is missed can multiply it by an order of magnitude, because water influx does not merely add a term — it holds the pressure up, which shrinks every expansion term simultaneously while the withdrawal keeps growing.

Worked Example C — strong water drive

The same 50 MMSTB reservoir, now with an active aquifer. Initial pressure 3,500 psia, well above the 2,500 psia bubble point, and the aquifer keeps it there: after 7.5 MMSTB of production the pressure has fallen only to 2,600 psia. Because pressure never reaches the bubble point, no free gas is liberated and the producing GOR stays at R_si = 581 scf/STB throughout — itself a diagnostic worth noticing.

Eight and a half times. Booked as reserves, that error survives every subsequent decision: facility sizing, well count, the economics that justified the development. And the fit that produced it is visually a straight line.

The test that does work

Havlena and Odeh's own recommendation is the reliable one, and it is not the intercept. For a volumetric tank, F/E_t equals N at every point — a constant. Under water influx F carries the extra W_e while E_t does not, so the ratio stops being constant. Departure from horizontal is the signature; the direction depends on aquifer strength. A strong aquifer holds pressure up, so E_t stays small while W_e grows and the ratio climbs — the behaviour in Example C. A weak aquifer contributes water more slowly than E_t grows, and Pletcher reports the resulting Campbell-plot signature as a negative slope.^[12] Reading only “rising means water” will miss the weak-aquifer case entirely.

Estimating W_e once you believe it

Diagnosis is not quantification. Once the aquifer is admitted there are two routes. If N is known independently — from volumetrics, or from a period of history before the aquifer responded — the balance can simply be rearranged for W_e at each step, which is how the W_e column in Example C was produced. If N is not known, N and the aquifer parameters have to be fitted together, using an influx model: van Everdingen–Hurst for the unsteady-state solution,^[5] Fetkovich for a finite aquifer with a productivity-index formulation,^[6] or Carter–Tracy as a non-iterative approximation to the former.^[7]

Part VI — Drive IndicesWhich mechanism is actually producing the field

The drive indices split the withdrawal into the fraction each mechanism accounts for. They answer a different question from N — not how much oil is there, but what is pushing it — and they are the most useful single output of a material balance for anyone deciding what to do next.

Depletion, segregation, water and expansion. By construction they sum to one, and that is not a curiosity — it is the check. A set that does not sum to one does not mean the reservoir is unusual; it means an input is wrong. Treat the sum as a unit test that runs on every case.

Read the table as three different reservoirs asking for three different development plans. Case A is producing itself to death on solution gas and is a candidate for pressure maintenance before the GOR runs away. Case B has half its energy in the gas cap, which makes gas reinjection and crestal well placement the whole conversation. Case C is being produced by its aquifer, and the question is not how to add energy but how to manage water — the point at which the Chan diagnostic and conformance work take over from material balance.

Note EDI in Case C: 3.8%, against 1.5% in Case A. Rock and connate water expansion is negligible in a saturated reservoir where gas liberation dominates, and is not negligible in an undersaturated one where it is competing only with liquid expansion. Dropping E_fw is defensible below the bubble point and is a real error above it — which is exactly where the highly compressible, high-pressure reservoirs live.

Part VII — Diagnostic PlotsCampbell, Dake and the family resemblance

The Havlena–Odeh plot is one member of a family. Each member linearises the same equation under a different hypothesis, and each is therefore diagnostic for a different failure. Choosing between them is choosing what you are prepared to be wrong about.

• F versus E_t — the Havlena–Odeh plot above. Slope N, through the origin. Diagnostic for anything missing from E_t.
• F/E_t versus cumulative production or time — the Campbell plot in its usual form. Horizontal for a volumetric tank; any departure indicates influx or a missing expansion term. It is the most sensitive member of the family because it removes the slope and plots only the residual behaviour — but the sign of the departure has to be interpreted, not assumed.^[12]
• F/E_t versus W_e/E_t — used once an influx model is assumed. Slope should be unity and intercept N; a slope far from one says the aquifer model is wrong, not that N is unusual. This is the oil analogue of the Cole plot used for gas.^[8]

Dake's treatment is the standard reference for reading these plots as a set rather than individually, and his point is worth repeating: the diagnostic value is in the disagreement between them.^[11] If every linearisation returns the same N, the model is probably right. If they disagree, the disagreement tells you which assumption is carrying the error.

Part VIII — DataWhat the method needs, and what it forgives

Pressure is the measurement that matters

Material balance is a pressure method. Production volumes are usually known to a few percent; average reservoir pressure rarely is. The number the equation wants is the volumetric average over the drainage volume, and what is available is a handful of build-up extrapolations from individual wells, each representing its own region, each with its own skin and its own incomplete build-up. In a heterogeneous or compartmentalised field, averaging them can produce a "reservoir pressure" that no part of the reservoir has ever had.

• Prefer extrapolated static pressures (p*) from build-ups long enough to reach radial flow, not flowing pressures.
• Weight by drainage volume where you can defend the weights, and say so when you cannot.
• Datum-correct every survey to a common depth before it goes anywhere near the equation.
• A single anomalous survey early in the history has enormous leverage on a through-origin fit, because E_t is small there and the ratio F/E_t is unstable. Early points deserve more scrutiny, not less.

PVT is the assumption that hides

Laboratory PVT on a properly conditioned bottom-hole sample is worth a great deal. A correlation tuned to nothing is worth much less, and the material balance cannot tell the difference — it will propagate a wrong B_t into a wrong N with complete composure. Where correlations must be used, state which ones and check the bubble point against any producing-GOR evidence available.

Pitfalls, in order of how much damage they do

• Missing aquifer. Worst case in this article: 8.7× overestimate with R² = 0.988. Test with F/E_t, not with the intercept.
• Wrong m. +97.6% with R² = 0.999. Test with F/E_t drift and with independent structural evidence.
• B_o used where B_t belongs. Silently discards the liberated solution gas below the bubble point.
• Instantaneous GOR used as R_p. R_p is cumulative G_p/N_p. The two diverge steadily once free gas is mobile.
• E_fw dropped above the bubble point. Defensible in a saturated reservoir, a real error in an undersaturated or overpressured one.
• Free-intercept fit reported as the answer. The physics fixes the intercept at zero. Use the free fit to interrogate, not to report.
• Compartmentalisation read as depletion. A tank equation applied to two tanks in poor communication returns a number that describes neither.

Part IX — ConclusionThree takeaways

1. Goodness of fit is not evidence of correctness. Both wrong answers in this article — +97.6% and +772% — came with R² above 0.98. If your defence of a material balance result is the R², you have not defended it.
2. F/E_t is the diagnostic worth running every time. Flat means the equation is complete and the level is N. Any departure means something is missing — rising for a strong aquifer, falling for an undersized E_t term or a weak one. It costs one extra column in the spreadsheet.
3. Report the drive indices, not just N. They sum to one as a check, and they answer the question the asset actually has — what is producing this field, and what happens when that runs out.

Material balance will not tell you what the field does next; that is a forecasting question and it needs different tools. What it will tell you, from nothing more than pressure and cumulative production, is how much oil the reservoir contains and which mechanism is delivering it — and it will do so before a simulation model exists, which is precisely when the answer is most needed and least available.

Frequently asked questions

What is oil material balance analysis?

It is a volumetric accounting method that estimates original oil in place (N) and identifies the drive mechanism from pressure decline and cumulative production, without requiring a reservoir simulation model. The governing relation is F = N(E_o + mE_g + E_fw) + W_e.

Why use Bt instead of Bo?

Below the bubble point B_o falls as pressure falls, because the oil is shrinking as gas leaves solution. The gas is still in the reservoir. B_t = B_o + (R_si − R_s)B_g accounts for both phases and increases monotonically, which is what the expansion term E_o requires.

How do you detect water drive on a Havlena-Odeh plot?

Not by the intercept, which can go negative on an upward-curving plot. Use F/E_t: it is constant and equal to N for a volumetric reservoir and departs from constant under water influx. In the worked example it rose 25.8% with every consecutive step increasing, while the intercept was only 5% of the largest withdrawal. A weak aquifer can produce the opposite sign, so the test is departure from horizontal, not the direction.

How wrong can a material balance be if the gas cap is missed?

In the worked example, analysing a reservoir with m = 0.40 as though m = 0 returned N = 98.8 MMSTB against a true 50.0 — a 97.6% overestimate, with R² = 0.99897. The fit quality gives no warning at all.

What do the drive indices tell you?

DDI, SDI, WDI and EDI give the fraction of reservoir voidage accounted for by depletion, gas cap segregation, water influx and rock plus connate water expansion. They sum to one by construction, so a sum that is not one indicates an input error rather than an unusual reservoir.